The Oxford Handbook of Ellipsis 9780198712398, 0198712391

Table of contents :
Cover
The Oxford Handbook of ELLIPSIS
Copyright
Contents
List of Figures and Tables
FIGURES
TABLES
The Contributors
Chapter 1: Ellipais in Natural Language: Theoretical and empirical perspectives
1.1 Introduction
1.2 Structure os the Handbook
1.2.1 Part I: The theory of ellipsis
1.2.2 Part II: Ellipsis as a diagnostic tool
1.2.3 Part III: Elliptical constructions
1.2.4 Part IV: Case studies
1.3 Results and Generalizations
1.3.1 Part I: The theory of ellipsis
1.3.2 Part II: Ellipsis as a diagnostic tool
1.3.3 Part III: Elliptical constructions
1.3.4 Part IV: Case studies
1.4 Conclusion
Acknowledgements
PART I: THE THEORY OF ELLIPSIS
Chapter 2: Ellipsis: A survey of analytical approaches
2.1 Introduction: The Phenomena
2.2 Approaches to the Syntax of Ellipsis
2.2.1 Structural and non-structural approaches compared
2.3 Evidence for structure in ellipsis
2.3.1 Lower origin effects
2.3.2 Locality effects
2.3.2.1 VP-ellipsis
2.3.2.2 Fragment answers
2.3.2.3 Stripping/Bare Argument Ellipsis
2.3.2.4 Gapping
2.3.2.5 Contrast sluicing
2.3.3 The P-stranding generalization
2.3.4 Case matching
2.3.5 Complementizer deletion
2.3.6 Infinitivals: Raising vs control
2.3.7 Predicate answers
2.3.8 Agreement triggers
2.4 Evidence against structure in ellipsis
2.4.1 Absence of locality effects
2.4.1.1 Sluicing
2.4.1.2 Fragment answers
2.4.1.3 Gapping
2.4.1.4 Ellipsis in comparatives
2.4.2 Case mismatches
2.4.3 Exceptions to the P-stranding generalization
2.5 Null anaphora and 'deletion'
2.6 The identity conditions on ellipsis
2.6.1 Semantic identity and information structure
2.6.2 Syntactic identity
2.6.2.1 Voice mismatch under ellipsis
2.6.2.2 Auxiliary form matching
2.6.3 Hybrid theories
2.7 Conclusions
Acknowledgements
Chapter 3: Ellipsis in transformational grammar
3.1 Introduction
3.2 Structure
3.2.1 Arguments for the structural analysis
3.2.1.1 Case-matching effects
3.2.1.2 Agreement
3.2.1.3 Subcategorization
3.2.1.4 Preposition stranding
3.2.1.5 Idiom reconstruction
3.2.1.6 Parasitic gaps
3.2.1.7 Extraction out of VPE site
3.2.2 Deletion vs LF-copying
3.2.2.1 Arguments for the deletion analysis
3.2.2.2 Potential arguments against the deletion analysis
3.2.2.3 Explanations under the deletion analysis
3.2.3 Summary of section 3.2
3.3 Recoverability
3.3.1 Syntactic identity
3.3.2 Semantic identity
3.3.2.1 Arguments against (strict) syntactic identity
3.3.2.2 Merchant’s semantic identity
3.3.3 Summary of section 3.3
3.4 Licensing
3.4.1 Licensing heads
3.4.1.1 Agreeing heads
3.4.2 Relation between licensing heads and ellipsis sites
3.5 Conclusion
Chapter 4: Ellipsis in head-driven phrase structure grammar
4.1 Introduction
4.1.1 General properties of ellipsis
4.1.2 Head-Driven Phrase Structure Grammar
4.2 An overview of ellipsis in HPSG
4.2.1 Constructional analyses of ellipsis
4.2.1.1 Non-canonical correspondence between argument structure and valence
4.2.1.2 Non-sentential utterances: Sluicing, short answers, and gapping
4.2.2 Non-canonical correspondence between the phonology of a phrase and its daughters
4.2.3 Elliptical phenomena: Evidence from corpora
4.3 The structure of the ellipsis site
4.3.1 Connectivity effects
4.3.2 Locality effects
4.3.3 The accessibility of missing referents
4.4 Recoverability
4.4.1 Split antecedents and syntactic mismatches
4.4.2 Exophoricity
4.4.3 Incrementality
4.5 Licensing
4.6 The syntax of argument ellipsis
4.6.1 Phrase structure and valence features in HPSG
4.6.2 The Argument Realization Principle and ellipsis
4.7 HPSG Ellipsis resoution in a dialogue setting
4.7.1 A dialogical working example
4.7.2 The dialogue gameboard
4.7.3 Grammar and semantics in Type Theory with Records
4.7.4 Non-sentential utterance constructions
4.7.4.1 Propositional lexemes
4.7.4.2 Declarative fragments
4.7.4.3 Direct sluicing
4.7.4.4 Reprise sluicing and reprise fragments
4.7.4.5 Disfluency
4.7.5 A worked example
Acknowledgements
Chapter 5: Ellipsis in categorial grammar
5.1 Introductory remarks
5.2 Categorial grammar
5.2.1 Some basics of CG
5.2.2 Relevance for ellipsis
5.3 If not silent material, then what?
5.3.1 VPE
5.3.2 Fragment answers
5.4 Arguments for SLM?
5.4.1 Deep vs surface anaphora
5.4.2 Connectivity effects
5.4.3 The meaning comes for free—but does it?
5.4.4 Preposition stranding
5.5 More background: More categorial grammar
5.5.1 Wh-extraction constructions
5.5.2 Variable-free semantics of pronouns
5.6 VP-Ellipsis
5.6.1 Simple VP-ellipsis
5.6.2 Antecedent-Contained Deletion
5.6.3 Pseudogapping
5.7 Other categorial grammar accounts of ellipsis
5.8 Conclusion
Acknowledgement
Chapter 6: Ellipsis in dependency grammar
6.1 Dependency-based syntax
6.2 Existing DG accounts of ellipsis
6.3 Structure
6.4 Recoverability
6.5 Licensing
6.5.1 Catenae
6.5.2 The evidence
6.6 Conclusion
Chapter 7: Ellipsis in simpler syntax
7.1 Simpler syntax
7.2 SSH and elliptical constructions
7.3 The framework: same-except
7.4 Licensing bare argument ellipsis
7.5 Licensing other cases of ellipsis
7.5.1 Sluicing
7.5.2 Gapping
7.5.3 VP-ellipsis
7.5.4 VP-anaphora
7.5.5 Pseudogapping
7.6 Other same-except constructions, not normally considered elliptical
7.6.1 Anaphora
7.6.2 Vice versa
7.6.3 The same, except, and their paraphrases
7.7 Structure in the ellipsis site?
7.8 Conclusion
Acknowledgements
Chapter 8: Ellipsis in construction grammar
8.1 Introduction
8.2 Motivating ellipsis
8.3 Recoverability: An independently needed pointer mechanism
8.4 Licensing: Gapping and other ellipsis constructions
8.5 Cross-linguistic differences: French and English
8.6 Why positing 'underlying' structure is problematic
8.7 Conclusion
Acknowledgements
Chapter 9: Ellipsis in dynamic syntax
9.1 Languages as action and the phenomenon of ellipsis
9.1.1 The ellipsis stalemate in competing domains of analysis
9.1.2 Morphosyntactic licensing of sub-sentential talk
9.1.3 Ellipsis as completability
9.2 Dynamic syntax
9.2.1 String-content mappings
9.2.2 Formalization of tree structure and incremental tree development
9.2.3 Linking trees through term sharing
9.3 The dynamics of ellipsis
9.3.1 Context in DS: Mechanisms for recovery of content at the ellipsis site
9.3.2 Content underspecification and recoverability through copying or action replay
9.3.3 Licensing sub-sentential utterances: Morphosyntactic constraints
9.3.4 Licensing split utterances
9.3.5 Licensing the splitting of dependenciesacross turns
9.3.6 Licensing repair devices: Self-repair and corrections
9.3.7 Licensing in local processing domains: Island restrictions
9.4 Reflections: Static vs dynamic perspectives compared
Acknowledgements
Chapter 10: Ellipsis in inquisitive semantics
10.1 Introduction
10.2 Inquisitive semantics
10.2.1 What makes a semantics ‘inquisitive’?
10.2.2 Inquisitive semantics across sentence types
10.2.3 Negation and other operators
10.2.4 Questions
10.3 Sluicing is sensitive to inquisitive content
10.3.1 The need to move beyond truth conditions
10.3.2 An account based on symmetric inquisitive entailment
10.3.3 Sprouting
10.4 Conclusions: Structure, recoverability, and licensing
Chapter 11: Ellipsis and psycholinguistics
11.1 Introduction
11.1.1 Might a naive view of ellipsis processing suffice?
11.2 The ellipsis site
11.2.1 Syntactic structure in the ellipsis site
11.2.2 Empirical evidence for syntactic structurein the ellipsis site
11.2.2.1 Larger penalties of syntactic mismatch for ellipsis than anaphora
11.2.2.2 Syntactically matching remnants that violate pied-piping or clause movement conditions
11.2.2.3 Islands are obeyed inside the ellipsis site
11.2.2.4 C-command inside the ellipsis site
11.2.2.5 Indices are copied into the ellipsis site
11.2.2.6 Phonological information may be present in the ellipsis site
11.3 Identity condition
11.3.1 Recoverability
11.3.2 Syntactic structure effects
11.3.3 The relation between the grammar and the processor
11.3.4 Alternative accounts of syntactic mismatches
11.4 Licensing
11.4.1 Licensing in the grammar
11.4.2 Discourse coherence relations
11.4.3 Main assertion preference
11.4.4 Accommodation
11.4.5 Question Under Discussion (QUD)
11.5 Conclusions
Acknowledgements
Chapter 12: Ellipsis and acquisition
12.1 Introduction: What is the acquisition challenge from ellipsis?
12.1.1 General theoretical consequences: An overview of Strict Interfaces
12.1.2 The acquisition perspective on representations, licensing, and recoverability
12.1.3 Exposition path: From theory to facts
12.1.4 Acquisition theory: Inferences create a learnability challenge
12.1.5 The simplicity of mechanisms
12.1.6 The repair parallel
12.1.7 A pivotal role for inferences
12.2 Early discourse: What do children's elliptical dialogues look like?
12.2.1 Earlier evidence of non-adult ellipsis
12.2.2 First verb phrase ellipsis: Where should it be?
12.2.3 VP-anaphora
12.2.4 Early Portuguese verb phrase ellipsis
12.2.5 The movement requirement
12.3 Verb phrase ellipsis and bound variables
12.3.1 Determiner deletion
12.3.2 External reference in VPE
12.4 Antecedent-contained deletion, parallelism, and locality
12.5 Sluicing
12.5.1 Verb phrase ellipsis summary
12.6 From argument drop to the NP/DP distinction
12.6.1 Japanese argument drop
12.6.2 Strict Interface version of argument drop
12.6.3 Reproducing the VP acquisition path in VP-ellipsis?
12.6.4 Triggering full VPE
12.7 Null complement anaphora
12.8 Noun phrase ellipsis: Partitivity
12.8.1 Early evidence of partitivity
12.8.2 Licensers: Inflection, possessives, agreement
12.8.3 The movement argument
12.8.4 Semantics and partitivity
12.8.5 What defines the partitive set?
12.9 Overview: The child's discourse environment
12.9.1 Fragments
12.9.2 Question Under Discussion
12.10 Conclusion
Acknowledgements
Chapter 13: Ellipsis and discourse
13.1 Introduction
13.2 Constraints on linguistic form
13.3 Strict and sloppy readings
13.3.1 Missing readings examples
13.3.2 Hardt’s puzzle
13.4 Conclusions
Chapter 14: Ellipsis and computational linguistics
14.1 Introduction
14.2 Structure of the ellipsis site
14.3 Recoverability
14.3.1 Copying
14.3.2 Unification
14.3.2.1 Ellipsis: An abstract characterization
14.3.2.2 Generating interpretations with unification
14.3.2.3 Comparison with “Identity of Relations” accounts
14.3.2.4 Non-elliptical counterparts
14.4 Licensing
14.4.1 Identifying ellipsis occurrences
14.4.2 Generation
14.5 Other topics
14.5.1 Antecedent location
14.5.2 Dialogue
14.5.3 Text mining
14.6 Conclusions
Chapter 15: Ellipsis and prosody
15.1 Introduction
15.2 Licensing of ellipsis
15.2.1 Types of ellipses and their prosodic characteristics
15.2.2 Prosodic licensing of contrastive ellipsis
15.2.3 Prosodic licensing of givenness-marking ellipsis
15.3 Prosody and recoverability of ellipsis site
15.3.1 Phonetic mismatches
15.3.2 Prosodic disambiguation
15.3.2.1 Prosodic phrasing and disambiguation
15.3.2.2 Pitch accents and disambiguation
15.3.3 Active–passive mismatches
15.3.4 Pitch accents and the interpretation of pronouns under ellipsis
15.4 Prosody and structure in the ellipsis site
15.4.1 Extraction from ellipsis site and prosody
15.4.2 Locality restrictions and prosody
15.5 Conclusion
Acknowledgements
PART II: ELLIPSIS AS A DIAGNOSTIC TOOL
Chapter 16: Movement and islands
16.1 Introduction
16.2 Movement, islands, cyclicity
16.2.1 Islands
16.2.2 Successive cyclicity and punctuated paths
16.3 Sluiciding and islands
16.3.1 Ross’s conjecture and its consequences
16.3.2 The island amelioration generalization
16.3.3 Ceteris paribus
16.3.4 Conclusion: Sluicing and islands
16.4 Punctuated paths
16.5 Conclusion
Chapter 17: Aphasia and acquisition
17.1 The content of this chapter
17.2 VPE
17.3 The psycholinguistic relevance of VPE: Broca's aphasia
17.3.1 Broca’s aphasia as a clinical entity
17.3.2 The comprehension deficit in Broca’s aphasia: TDH vs WM/generic complexity
17.3.3 Testing VPE with patients suffering from Broca’s aphasia
17.3.4 Experiment
17.3.5 Patients’ relative success in analyzing VPE and the complexity/WM view
17.3.6 Patients’ relative success in analyzing VPE and the movement (TDH) view
17.3.7 VPE in reaction time experiments
17.4 The Psycholinguistic relevance of VPE: Acquisition
17.4.1 Spontaneous production of VPE: Parallelism is used in early childhood
17.4.2 Children comprehend VPE very early on: Experimental evidence
17.5 Where are we? What to do next?
Acknowledgements
Chapter 18: Parsing strategies
18.1 Introduction
18.2 What structure does the parser build in the ellipsis site, and how?
18.3 When does the parser recognize ellipsis?
18.4 Where does the parser find the antecedent of ellipsis?
18.5 Conclusions
Acknowledgements
Chapter 19: Codeswitching
19.1 Introduction
19.2 Theories of codeswitching
19.3 What codeswitching tells us about ellipsis
19.4 What ellipsis tells us about codeswitching
19.5 Outlook
Acknowledgements
PART III: ELLIPTICAL CONSTRUCTIONS
Chapter 20: Sluicing and its subtypes
20.1 Introduction
20.2 Many roads lead to a sluice
20.2.1 Non-isomorphic sluicing and island repair effects
20.2.2 Non-isomorphic sluicing and wh- in situ languages
20.2.3 Non-isomorphic sluicing as a default strategy
20.2.4 The balance between isomorphic and non-isomorphicsluicing
20.3 Multiple sluicing
20.3.1 Genuine multiple sluicing
20.3.1.1 Multiple focus/wh–fronting
20.3.1.2 Multiple clefting
20.3.2 Fake multiple sluicing
20.3.2.1 Single sluicing plus Heavy NP Shift
20.3.2.2 Null coordination of simple sluices
20.4 Limits of a pure deletion analysis
20.4.1 Repair effects without an evasive source
20.4.1.1 Parasitic gaps in the remnant
20.4.1.2 P-stranding effects in non-P-stranding languages
20.4.1.3 German
20.4.3 Morphological case parallelism effects
20.5 Conclusions and outlook
Acknowledgement
Chapter 21: Predicate ellipsis
21.1 What is predicate ellipsis?
21.2 English VP-Ellipsis
21.2.1 Licensing of English VPE
21.2.2 Recoverability
21.2.3 Deleted syntactic structure?
21.2.4 The size of the ellipsis site
21.3 Pseudogapping
21.3.1 Properties of pseudogapping
21.3.2 The nature of the movement operation
21.4 VPE cross-linguistically
21.5 Other predicate ellipses: Modal complement ellipsis and british english do
21.5.1 Modal Complement Ellipsis
21.5.2 British English do
21.6 Concluding remarks
Chapter 22: Nominal ellipsis
22.1 Introduction
22.2. Empty nouns vs nominal ellipsis: Some diagnostics
22.2.1 Uniformity tests
22.2.1.1 Thematic assignment
22.2.1.2 Extraction tests
22.2.1.3 Matching effects
22.2.2 More diagnostics
22.2.2.1 Antecedents
22.2.2.2 Identity effects
22.2.2.3 Lexical restrictions and productivity
22.2.3 Summary
22.3 Recoverability conditions on nominal ellipses
22.3.1 Identity in NPE
22.3.2 Recoverability and ENs
22.4 Licensing nominal ellipses
22.4.1 Some background
22.4.2 Licensing as selection
22.4.3 Government effects in nominal ellipses as epiphenomena
22.4.3.1 Selection by Num: nP ellipses
22.4.3.2 Selection by D: NumP ellipses
22.4.3.3 Selection by n: RootP ellipses
22.4.4 Summary
22.5 Conclusion
Acknowledgements
Chapter 23: Gapping and stripping
23.1 Introduction
23.2 Stripping
23.3 Gapping
23.4 Accounts
23.4.1 Small conjuncts account
23.4.1.1 Ellipsis
23.4.1.2 Across-the-board movement
23.4.2 Large conjuncts
23.5 Future directions
Chapter 24: Fragments
24.1 Introduction
24.2 Non-sententialist accounts
24.3 The sententialist approach
24.4 Arguments for non-sententialism, and sententialist replies
24.4.1 The psychological plausibility of silent syntax
24.4.2 Anti-connectivity effects
24.4.3 The problem of presupposition inheritance
24.5 Connectivity effects as evidence for sententialism, and non-sententialist responses
24.5.1 Case connectivity
24.5.2 Islandhood
24.5.3 Complementizer distribution
24.5.4 Binding in fragment answers
24.5.5 Prepositions in fragment answers
24.6 Conclusion
Acknowledgements
Chapter 25: Comparative deletion
25.1 Introduction
25.2 The syntax and semantics of comparatives
25.3 Comparative Deletion
25.3.1 The movement and deletion (matching) analysis of CD
25.3.2 The raising analysis of CD
25.3.3 Semantic theories of CD and diagnostics for structure
25.4 Ellipsis in comparatives—comparative ellipsis
25.4.1 Ellipsis in comparative and coordinate structures
25.4.1.1 Gapping
25.4.1.2 Right-Node Raising
25.4.1.3 Across-the-board movement
25.4.1.4 Other ellipsis operations in comparatives
25.4.2 Coordination vs subordination
25.5 Phrasal comparatives
25.5.1 Arguments for DA
25.5.2 Interpreting PCs
25.5.3 Arguments for RA
25.6 Conclusion
Chapter 26: Null complement anaphora
26.1 Characteristics of null complement anaphora
26.1.1 The basic data
26.1.2 The questions
26.2 Previous studies
26.2.1 Shopen (1972, 1973)
26.2.2 Hankamer and Sag (1976)
26.2.2.1 Deep vs surface anaphora
26.2.2.2 Hankamer and Sag’s (1976) tests
26.2.3 Grimshaw (1979)
26.2.4 Napoli (1983b)
26.2.5 Fillmore (1986)
26.2.6 Depiante (2000, 2001)
26.2.6.1 NCA does not allow extraction
26.2.6.2 NCA does not allow inverse scope readings
26.2.7 Huddleston and Pullum (2002)
26.2.8 Cinque (2004)
26.2.9 Haynie (2010)
26.2.10 Summary and discussion of the different analyses of NCA in the literature
26.3 Which predicates allow NCA in english?
26.4 How is NCA represented in the syntax? How is NCA interpreted?
26.4.1 Is lack of extraction a good diagnostic for lack of internal structure?
26.4.2 On the semantics of NCA
26.5 Summary of the properties of NCA
26.6 NCA cross-linguistically
26.7 Conclusions and directions for future research
Chapter 27: Conjunction reduction and right-node raising
27.1 Introduction
27.2 Coordination syntax and CR
27.2.1 Conjunction Reduction vs phrasal coordination
27.2.2 Coordination syntax: Constituency, substitutability, and coordination of likes
27.2.3 Unlike Category Coordination
27.2.4 Discontinuous coordination and stripping
27.2.5 Non-constituent conjunct strings
27.2.6 Conjunction Reduction and movement
27.2.7 Alternative: Flexible constituency
27.2.8 Other kinds of motivation for CR
27.3 Forward CR
27.3.1 Forward CR and non-coordinate ellipsis
27.3.2 Move-and-elide
27.4 RNR
27.4.1 RNR as rightward ATB movement
27.4.2 Special RNR properties
27.4.2.1 Linear properties: String-vacuous nature, Right-Edge Restriction
27.4.2.2 RNR targets
27.4.2.3 Interaction of RNR with leftward movement
27.4.2.4 Interaction of RNR with ellipsis
27.4.3 RNR as ellipsis
27.4.3.1 Form identity
27.4.3.2 A total of, same, different
27.4.3.3 Summative agreement
27.4.3.4 Backwards ellipsis vs RNR
27.4.4 RNR as Multiple Dominance
27.4.4.1 Sharing as MD
274.4.2 Non-bulk sharing
27.4.5 Non-coordinate RNR
27.4.6 Right-Node Wrapping
27.5 Conclusion
PART IV: CASE STUDIES
Chapter 28: Dutch
28.1 Introduction
28.2 Gapping and Stripping
28.2.1 Gapping: Deletion of verbal material
28.2.2 Gapping and its remnants
28.2.3 Gapping: Its distribution
28.2.4 Stripping
28.3 Predicate ellipsis (including VP-ellipsis and pseudogapping)
28.4 Conjunction Reduction
28.4.1 Forward Conjunction Reduction
28.4.2 Backward Conjunction Reduction
28.5 Sluicing and its subtypes
28.6 Fragments
28.7 Nominal ellipsis
28.8 Comaparative deletion
28.8.1 Comparative Deletion
28.8.2 Comparative Subdeletion
28.8.3 CsubD, CD, and coordination
28.8.4 Phrasal comparatives
28.9 Null complement anaphora
28.10 Ellipsis in dutch dialects
28.10.1 Clausal ellipsis in Dutch dialects
28.10.2 Nominal ellipsis in Dutch dialects
28.11 Summary
Acknowledgements
Chapter 29: Finnish sign language
29.1 Introduction
29.2 Background on finSL grammar
29.2.1 Major sign classes
29.2.2 Basic syntax
29.2.3 Constructed action
29.3 On the general role of elliptical phenomena in finSL
29.4 Elliptical phenomena involving normal phrases in finSL
29.4.1 Nominal phrase-internal ellipsis
29.4.2 Ellipsis of nominal phrases in coordinated two-clause structures
29.4.3 Other contexts licensing the ellipsis of nominal phrases in FinSL
29.5 Elliptical phenomena involving verbal predicates
29.5.1 On the elliptical phenomenon resembling VP-ellipsis
29.5.2 Gapping, stripping, and sluicing in FinSL
29.6 On other types of ellipsis in finSL
29.6.1 Answer ellipsis
29.6.2 Ellipsis of conjunctions
29.7 On the role of gesture and mime in the elliptical phenomena of finSL
29.8 Conclusion
Acknowledgements
Chapter 30: French
30.1 Predicate ellipsis and anaphora in french
30.2 Sluicing and related cases
30.3 Coordinate ellipses
30.4 Ellipses and comparatives
30.4.1 Stripping and gapping in comparatives
30.4.2 Anaphora in comparatives
30.5 Noun phrase ellipsis (NPE)
30.6 Conclusion
Chapter 31: Hungarian
31.1 Nominal ellipsis
31.1.1 Nominal ellipsis in non-possessed nominals
31.1.2 Anaphoric possessed nominals
31.2 Predicate ellipsis
31.2.1 AUX-stranding predicate ellipsis
31.2.2 V-stranding predicate ellipsis
31.2.3 Preverb-stranding ellipsis
31.2.4 Pseudogapping
31.3 Clausal ellipsis
31.3.1 Single-remnant clausal ellipsis and the theory of ellipsis licensing
31.3.2 Multiple-remnant clausal ellipsis and ellipsis repair
31.3.3 Multiple remnants in gapping
31.4 Right-node raising
31.5 Ellipsis in comparative clauses
31.6 Null complement anaphora
31.7 Conclusion
Acknowledgements
Chapter 32 Indonesian
32.1 Introduction
32.2 Sluicing
32.3 Fragment answers
32.4 Verb phrase ellipsis
32.5 Pseudograpping
32.6 Gapping
32.7 Stripping
32.8 Comparative deletion
32.9 Right-node raising
32.10 Conjunction reduction
32.11 Null complement anaphora
32.12 Nominal ellipsis
32.13 Indonesian ellipsis and the PSG
32.13.1 Sluicing
32.13.2 Fragment answers
32.13.3 Pseudogapping
32.13.4 Stripping
32.13.5 Gapping
32.13.6 Conjunction reduction
32.14 Conclusion
Acknowledgements
Chapter 33: Japanese
33.1 Introduction
33.2 Fragments
33.2.1 Stripping, sluicing, and ellipsis in comparatives as a uniform phenomenon
33.2.1.1 Stripping
33.2.1.2 Sluicing
33.2.1.3 Ellipsis in comparatives
33.2.1.4 Summary
33.2.2 Short answers
33.2.3 Summary
33.3 Right-node raising across-the-board movement, gapping, or string deletion?
33.4 Null argument construction: Ellipsis or non-ellipsis?
33.4.1 Analyses of the null argumentconstruction in Japanese
33.4.1.1 The ellipsis analysis 1: The VP ellipsis analysis
33.4.1.2 The non-ellipsis analysis
33.4.1.3 The ellipsis analysis 2: The argument ellipsis analysis
33.4.2 Summary
33.5 N'-deletion
33.6 Summary
Acnowledgements
Chapter 34: Kiswahili and shingazidja
34.1 Introduction
34.2 Background
34.2.1 Kiswahili and Shingazidja
34.2.2 Previous studies on ellipsis in Bantu languages
34.3 V-stranding VP-ellipsis
34.4 N-deletion
34.5 Right-node raising
34.6 Gapping and related phenomena
34.6.1 Gapping
34.6.2 Other cases of clausal ellipsis
34.7 Null complement anaphora
34.8 Sluicing
34.9 Fragments
34.10 Is english-style VPE possible?
34.11 Conclusion
Acknowledgements
Chapter 35: Persian
35.1 Gapping and stripping
35.1.1 Three properties of gapping
35.1.2 Towards an analysis
35.1.3 No low coordination in Persian
35.1.4 When there is just one remnant
35.2 Fragment answers
35.3 Right-node raising
35.4 Verb phrase ellipsis
35.4.1 Not just a null argument
35.4.2 Evidence for low ellipsis
35.4.3 Stranding a simple verb
35.5 Sluicing
35.5.1 Deriving sluicing in Persian
35.5.2 The properties of focus fronting
35.5.3 The island (in)sensitivity of sluicing
35.6 Noun phrase ellipsis
35.7 Summary
Acknowledgements
Chapter 36: Polish
36.1 Introduction
36.2 Sluicing
36.2.1 Multiple sluicing
36.3 Fragment answers
36.3.1 Split questions
36.4 Stripping
36.5 Omission of prepositions
36.6 Predicate ellipsis
36.7 Ohter elliptical constructions
36.7.1 Gapping
36.7.2 Right-node raising
36.7.3 Non-constituent coordination
36.7.4 Comparative deletion
36.7.5 Antecedent-contained deletion
36.7.6 Noun phrase ellipsis
36.8 Summary
Acknowledgements
Chapter 37: Russian
37.1 Introduction
37.2 Np-ellipsis in Russian
37.3 Clausal ellipsis in Russian
37.3.1 Sluicing
37.3.2 Sprouting
37.3.3 Polarity ellipsis in Russian
37.3.4 Stripping
37.3.4.1 Types of stripping in Russian
37.3.4.2 Stripping as clausal ellipsis
37.4 Predicate ellipsis
37.4.1 Gapping
37.5 Verb-stranding phenomena
37.6 Some other types of ellipsis present in russian
37.6.1 Comparative deletion
37.6.2 Right-Node Raising and fragment answers
37.6.2.1 Right-Node Raising
37.6.2.2 Fragment answers
37.7 Conclusion
Chapter 38: Varieties of english
38.1 Introduction
38.2 The licensing of VP-ellipsis
38.2.1 VP-ellipsis and auxiliary-like verbs
38.2.2 Null verbs
38.3 VP-ellipsis and non-finite auxiliaries
38.3.1 Optionality in ellipsis of non-finite auxiliaries
38.3.2 Non-finite auxiliaries and raising
38.3.3 Have-deletion
38.3.4 The anti -ing constraint
38.3.5 Additional auxiliaries in tag questions and retorts
38.3.6 Interim summary
38.4 British do
38.5 Conclusion
References
Index

Citation preview

   

ELLIPSIS

OXFORD HANDBOOKS IN LINGUISTICS Recently published

THE OXFORD HANDBOOK OF INFLECTION Edited by Matthew Baerman

THE OXFORD HANDBOOK OF HISTORICAL PHONOLOGY Edited by Patrick Honeybone and Joseph Salmons

THE OXFORD HANDBOOK OF LEXICOGRAPHY Edited by Philip Durkin

THE OXFORD HANDBOOK OF NAMES AND NAMING Edited by Carole Hough

THE OXFORD HANDBOOK OF DEVELOPMENTAL LINGUISTICS Edited by Jeffrey Lidz, William Snyder, and Joe Pater

THE OXFORD HANDBOOK OF INFORMATION STRUCTURE Edited by Caroline Féry and Shinichiro Ishihara

THE OXFORD HANDBOOK OF MODALITY AND MOOD Edited by Jan Nuyts and Johan van der Auwera

THE OXFORD HANDBOOK OF PRAGMATICS Edited by Yan Huang

THE OXFORD HANDBOOK OF UNIVERSAL GRAMMAR Edited by Ian Roberts

THE OXFORD HANDBOOK OF ERGATIVITY Edited by Jessica Coon, Diane Massam, and Lisa deMena Travis

THE OXFORD HANDBOOK OF POLYSYNTHESIS Edited by Michael Fortescue, Marianne Mithun, and Nicholas Evans

THE OXFORD HANDBOOK OF EVIDENTIALITY Edited by Alexandra Y. Aikhenvald

THE OXFORD HANDBOOK OF PERSIAN LINGUISTICS Edited by Anousha Sedighi and Pouneh Shabani-Jadidi

THE OXFORD HANDBOOK OF LYING Edited by Jörg Meibauer

THE OXFORD HANDBOOK OF TABOO WORDS AND LANGUAGE Edited by Keith Allan

THE OXFORD HANDBOOK OF MORPHOLOGICAL THEORY Edited by Jenny Audring and Francesca Masini

THE OXFORD HANDBOOK OF ELLIPSIS Edited by Jeroen van Craenenbroeck and Tanja Temmerman For a complete list of Oxford Handbooks in Linguistics please see pp. –.

   

......................................................................................................................

ELLIPSIS ...................................................................................................................... Edited by

JEROEN VAN CRAENENBROECK and

TANJA TEMMERMAN

1

3

Great Clarendon Street, Oxford,  , United Kingdom Oxford University Press is a department of the University of Oxford. It furthers the University’s objective of excellence in research, scholarship, and education by publishing worldwide. Oxford is a registered trade mark of Oxford University Press in the UK and in certain other countries © editorial matter and organization Jeroen van Craenenbroeck and Tanja Temmerman  © the chapters their several authors  The moral rights of the authors have been asserted First Edition published in  Impression:  All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, without the prior permission in writing of Oxford University Press, or as expressly permitted by law, by licence or under terms agreed with the appropriate reprographics rights organization. Enquiries concerning reproduction outside the scope of the above should be sent to the Rights Department, Oxford University Press, at the address above You must not circulate this work in any other form and you must impose this same condition on any acquirer Published in the United States of America by Oxford University Press  Madison Avenue, New York, NY , United States of America British Library Cataloguing in Publication Data Data available Library of Congress Control Number:  ISBN –––– Printed and bound by CPI Group (UK) Ltd, Croydon,   Links to third party websites are provided by Oxford in good faith and for information only. Oxford disclaims any responsibility for the materials contained in any third party website referenced in this work.

C

................................

List of figures and tables The contributors

ix xii

. Ellipsis in natural language: Theoretical and empirical perspectives



J  C  T T

PART I

THE THEORY OF ELLIPSIS

. Ellipsis: A survey of analytical approaches



J M

. Ellipsis in Transformational Grammar



H L  K F

. Ellipsis in Head-Driven Phrase Structure Grammar



J G  P M

. Ellipsis in Categorial Grammar



P J

. Ellipsis in Dependency Grammar



T O

. Ellipsis in Simpler Syntax



P W. C  R J

. Ellipsis in Construction Grammar



A E. G  F P

. Ellipsis in Dynamic Syntax



R K, E G, A E,  J H

. Ellipsis in Inquisitive Semantics S A



vi



. Ellipsis and psycholinguistics



L F

. Ellipsis and acquisition



T R

. Ellipsis and discourse



A K

. Ellipsis and computational linguistics



D H

. Ellipsis and prosody



S W

PART II ELLIPSIS AS A DIAGNOSTIC TOOL . Movement and islands



K A

. Aphasia and acquisition



Y G, I D,  L P. S

. Parsing strategies



M Y

. Codeswitching



K G́  -V  S E. R

PART III

ELLIPTICAL CONSTRUCTIONS

. Sluicing and its subtypes



L V

. Predicate ellipsis



L A  W H

. Nominal ellipsis A́  S





. Gapping and stripping

vii



K J

. Fragments



A H

. Comparative deletion



W L

. Null Complement Anaphora



M D

. Conjunction Reduction and Right-Node Raising



C W

PART IV

CASE STUDIES

. Dutch



N C  M  K

. Finnish Sign Language



T J

. French



A D

. Hungarian



A́ L́ 

. Indonesian



C F

. Japanese



T F

. Kiswahili and Shingazidja



C́   P  S M

. Persian



M T

. Polish J N



viii



. Russian



J F B  T B

. Varieties of English



G T

References Index

 

L    

........................................................................................

F IGURES .

Contrast (left) and elaboration (right)

.

Processing John upset Mary in DS

.

Unfolding structure for John upset . . .

.

Result of parsing John, who smokes, left

   

.

DS parsing context as a graph: Actions (edges) are transitions between partial trees (nodes)



.

Substitution from context at the ellipsis site of (): Pronominal anaphora (top) and VP-ellipsis (bottom)

.

Action replay from context at the ellipsis site

.

A short answer with binding restrictions

.

Incremental development of Mary’s/Bob’s context via processing words

.

Processing Chorlton? in ‘A: the doctor B: Chorlton?’

    

.

Incremental interpretation of self-repair by replaying DS actions in the Context DAG

.

Successful processing of John interviewed every student who Bill had

 

.

Ungrammaticality of () as impossibility to unify unfixed node with object of interview in second relative clause



.

Pitch extraction analysis of the VPE in (): Distribution of pitch accents and prosodic boundaries



.

Pitch extraction analysis of gapping in (): Parallel contrastive accents and prosodic boundaries



.

Pitch extraction analysis of (c): Violation of pairwise contrast in gapping



.

Pitch extraction contour of RNR in (): Licensing pitch accents and boundary tones

.

Pitch extraction contour of extraction from VPE

.

Pitch extraction contour of topicalization from VPE in a relative clause

.

Experimental stimuli

   

    

x

.

A partial taxonomy of sluicing, based on the underlying syntax of the sluice



.

FinSL verbals with the meanings (from left to right) ‘know’ (the finger pads of the open hand touch the forehead twice), ‘teach [someone in front of the signer]’ (the two hands move forward twice in the shown configuration), and ‘an oblong vehicle (e.g. a bicycle) drives forward over a mound-like location’ (the dominant hand articulates an arc-shaped movement over the stationary non-dominant hand). The verbals represent Type , Type , and Type , respectively (see also Figure .)



.

.

FinSL Type  verbal TEACH as used in the elliptical clause (). Note that the signer also employs constructed action to show the imaginary locations of the referents Video frames showing the production of the sentence ()

 

T ABLES .

Overview of Part I of the handbook: Abstract structure, recoverability, and licensing



.

Overview of Part IV of the handbook: Cross-linguistic distribution of the main ellipsis types



.

Overview of the cross-linguistic distribution of various subtypes of predicate ellipsis

.

Some previous research on the two ellipsis questions

.

A taxonomy for clarification requests (Purver )

.

Dialogue gameboard

.

Commonly discussed constructions that involve ellipsis

.

Non-elliptical versions of the attested examples of ellipsis in Table .

     

.

Less often discussed constructions that involve ellipsis, with examples and references



.

Average percentages of ‘yes’ responses broken down over condition and language from Wijnen et al. ()

.

Test conditions: VPE sentence + images

.

Control condition: Coordination + images

.

Experimental conditions

.

Patients’ demographic and clinical profiles

    

.

Number of accurate responses per condition for each patient on the VPE test



    

xi

.

Examples of spontaneous production of VPE by children

.

Test Condition: VPE + images

.

Control condition: Coordination + images

.

Experimental results

   

.

Ellipsis of A and P core arguments in the sample of  transitive clauses containing a Type  or Type  verbal predicate



T 

............................................................

Klaus Abels received his PhD from the University of Connecticut in . He is a Reader in Linguistics at University College London and co-editor of the journal Syntax. His main interests relate to movement, constraints on movement, interactions of movement types, the formal modeling of movement, and the role of movement in deriving word order typology. Lobke Aelbrecht obtained her PhD at the Catholic University of Brussels in  with a thesis entitled ‘You have the right to remain silent: The syntactic licensing of ellipsis’. Her main research interests are ellipsis, VP topicalization and VP pronominalization, and the Dutch adpositional domain. In , she published the monograph The syntactic licensing of ellipsis (John Benjamins). Scott AnderBois is Assistant Professor of Cognitive, Linguistic, and Psychological Sciences at Brown University. His primary research focus is on the ways in which utterances interact with the discourse “scoreboard,” with a particular focus on to what extent and in what ways these interactions are encoded conventionally as part of the sentence meanings, as opposed to arising from pragmatic reasoning. Specific topics of interest include apposition, discourse particles, disjunction, ellipsis, evidentiality, indefiniteness, mirativity, topics, and questions. He has explored these issues through primary fieldwork including Yucatec Maya, A’ingae, English, and Tagalog. John Frederick Bailyn is Professor of Linguistics at Stony Brook University, where he directs the Linguistics PhD program. His research interests include theoretical syntax, comparative Slavic syntax, binding, case, word order, scrambling, and ellipsis. His other interests include musical cognition and early Soviet history. He also co-directs the NY-St Petersburg Institute in Linguistics, Cognition and Culture (NYI) held every July in St Petersburg, Russia. Tatiana Bondarenko received her Bachelor and Master degrees in Linguistics from Lomonosov Moscow State University, and is currently a PhD student in Linguistics at the Massachusetts Institute of Technology. Her research interests include syntax, semantics, morphology, argument structure and event structure, ellipsis phenomena, embedded clauses (raising, control, restructuring), structures with dative and applicative arguments, aspectual systems, and fieldwork. Norbert Corver is the Chair-Professor of Dutch Linguistics in the Department of Languages, Literature and Communication of Utrecht University, and affiliated with the research institute UiL-OTS and the research group Language Structure: Variation and

 

xiii

Change. He received his PhD in Linguistics, entitled ‘The syntax of left branch extractions’, from Tilburg University in . His main research interests are located in the areas of Dutch syntax, comparative syntax, and the interaction between language (morphosyntax) and affect. Specific topics he has been working on include displacement and locality, the morphosyntax of functional categories, the syntax of adverbs, NP-ellipsis, and the morphosyntactic encoding of affect. Jeroen van Craenenbroeck is Associate Professor of Dutch Linguistics at KU Leuven, where he is also Vice-President of the Center for Research in Syntax, Semantics, and Phonology (CRISSP). He is the author of The syntax of ellipsis (OUP) and general editor of the journal Linguistic Variation (John Benjamins). His research interests include ellipsis (sluicing, swiping, spading, VP-ellipsis), expletives, verb clusters, and the left periphery of the clause. Peter W. Culicover is Distinguished University Professor at the Ohio State University. He was awarded the Humboldt Research Award in . His primary research has been in syntactic theory. He has been concerned with exploring the cognitive and computational factors that underlie the foundations of syntactic theory. Most recently he has been pursuing an evolutionary account of the origin of grammars from a constructional perspective. Anne Dagnac is Assistant Professor in French and Romance Linguistics at the University of Toulouse Jean Jaurès (France). Her research, which she conducts within CLLE (CNRS / University of Toulouse), focuses mainly on French syntax, and on the microsyntactic variation of the underdescribed French Romance dialects, in particular Picard. Marcela Depiante received her PhD from the University of Connecticut in . She is an Associate Professor in the Department of Languages at the University of Wisconsin, Eau Claire. Her research interests include comparative syntax and morphology, in particular the syntax of ellipsis, as well as the grammar of Heritage Spanish Speakers and Spanish L attriters in the US. Isabelle Deschamps is a postdoctoral research scholar in the Department of Rehabilitation at Laval University. Previously she received her bachelor’s degree in Psychology and Linguistics as well as her PhD in Communication Sciences and Disorders from McGill University. Her research interests focus on issues pertaining to phonological processes during speech perception and production. In addition, her research aims to understand the relationship between phonological processes and other cognitive functions such as verbal working memory. Arash Eshghi is a researcher in Computational Linguistics at Heriot-Watt University. He did his PhD in Psycholinguistics at Queen Mary, University of London. Ever since, he has been one of the main developers of the Dynamic Syntax computational implementation, and is currently exploring its technological applications, e.g., in building more human-like conversational systems.

xiv

 

Catherine Fortin is Associate Professor of Linguistics at Carleton College. She received her PhD in Linguistics from the University of Michigan in . Her primary research interests are the syntax and morphosyntax of Indonesian and Minangkabau, including ellipsis phenomena, clause structure, and argument structure. Lyn Frazier is a Professor in the Department of Linguistics at the University of Massachusetts Amherst. Specializing in psycholinguistics, her research spans a range of topics in syntactic processing, primarily concerning phrase structure parsing and the parsing of movement dependencies, and issues at the syntax–discourse interface. The latter include research on the role of prosody in sentence processing, ellipsis, processing of not-at-issue content, and the role of implicit and explicit Questions-Under-Discussion (QUDs) in organizing discourse. Teruhiko Fukaya is Professor in the Faculty of International Communication at Gunma Prefectural Women’s University in Japan. He is interested in the investigation of the language faculty through the studies of ellipsis phenomena, such as sluicing, stripping, and fragment answers, in Japanese and English. Kenshi Funakoshi is Lecturer in Linguistics in the Department of English at Dokkyo University. He specializes in generative syntax with an emphasis on ellipsis, verb movement in SOV languages, and Japanese. His publications include ‘On headless XP-ellipsis/movement’ (), ‘Verb-stranding verb phrase ellipsis in Japanese’ (), and ‘Backward control from possessors’ (). Jonathan Ginzburg is Professor of Linguistics at Université Paris-Diderot (Paris ). He has held appointments at the Hebrew University of Jerusalem and King’s College, London. He is one of the founders and editor-in-chief (emeritus) of the journal Dialogue and Discourse. His research interests include semantics, dialogue, language acquisition, and musical meaning. He is the author of Interrogative investigations (CSLI Publications, , with Ivan A. Sag) and The interactive stance: Meaning for conversation (Oxford University Press, ). Adele E. Goldberg is a Professor of Psychology at Princeton University where she is also affiliated with the linguistics and cognitive science programs. Goldberg’s work focuses on the psychology of language, particularly on how grammatical constructions are represented, learned, and processed. A more specific interest is in the functions of constructions and how those functions explain facts that are often assumed to be purely syntactic. Kay González-Vilbazo is Associate Professor at the University of Illinois at Chicago and Co-Director of the UIC Bilingualism Research Lab. His research focuses on linguistictheoretical aspects of bilingual phenomena, mostly the grammar of codeswitching. Current projects of his include gender agreement and concord, phase theory, wh-dependencies, ellipsis, pro-drop, case theory, the structure of PF, the phonology of codeswitching, and the theory of the bilingual lexicon. Eleni Gregoromichelaki is a Research Fellow at the Philosophy Department, King’s College London and the Cognitive Science Institute, Osnabrueck University. She works within the Dynamic Syntax and Computational Linguistics research groups, exploring analyses of

 

xv

syntactic/semantic natural language phenomena within psycholinguistically informed formalisms. Her principal research interests lie in the language–cognition interface: in particular, the formal/computational and psychological/philosophical implications of various cognitive modeling perspectives on natural language. Yosef Grodzinsky is Professor of Neurolinguistics at the Edmond and Lily Safra Center for Brain Research, the Hebrew University of Jerusalem, and a Visiting Professor at the Institute of Neuroscience and Medicine (INM-), Forschungszentrum Jülich. His research has focused on the neural basis, acquisition, and processing of syntactic and semantic knowledge. At present, his work explores the neural bases of overt and covert negation. Previously, Grodzinsky was a Professor and Tier-I Canada Research Chair of Neurolinguistics at McGill University, a Professor of Psychology at Tel Aviv University, and a Research Professor of Neurology at the Boston University School of Medicine. He is the recipient of several awards, and his research has been funded by government agencies in the US, Canada, Israel, and Germany. Alison Hall is a Lecturer in English Language at De Montfort University, Leicester, UK. She received her PhD in Pragmatics from University College London and has been a postdoctoral researcher at UCL and at Institut Jean Nicod, Paris. She has published on linguistic underdeterminacy, lexical pragmatics, and the debate between contextualism, indexicalism, and semantic minimalism. Daniel Hardt is Associate Professor in the Department of Management, Society, and Communication at Copenhagen Business School, and is Visiting Research Associate at the University of California, Santa Cruz. He holds a PhD in Computer Science from the University of Pennsylvania. His research deals with both theoretical and computational linguistics, with a particular interest in ellipsis and other matters involving semantics, anaphora, and discourse. He has published articles on ellipsis in journals such as Linguistics and Philosophy, Journal of Semantics, and Computational Linguistics. William Harwood obtained his PhD at Ghent University in  with a thesis entitled ‘Being progressive is just a phase: Dividing the functional hierarchy’. His main research interests include ellipsis, phase theory, idiomatic expressions, auxiliary verbs, aspect, verb movement, VP fronting, existential constructions, and relative clauses. Julian Hough is a Lecturer in the Cognitive Science group in the School of Electronic Engineering and Computer Science at Queen Mary, University of London (QMUL). He received his PhD at QMUL before working at Bielefeld University as a post-doc. He researches dialogue modelling and dialogue systems, with a focus on incremental processing and disfluency. Ray Jackendoff is Seth Merrin Professor Emeritus at Tufts University and a Research Associate at the MIT Department of Brain and Cognitive Sciences. He was the  recipient of the David Rumelhart Prize in Cognitive Science. His principal research goal at present is the Parallel Architecture, a theory of linguistic structure that incorporates

xvi

 

semantics, syntax, morphology, and phonology, and that integrates language with the rest of the mind. Pauline Jacobson is Professor of Cognitive, Linguistic, and Psychological Sciences at Brown University. Her research centers on formal semantics and its interaction with syntax, and she is the author of the semantics textbook Compositional semantics: An introduction to the syntax/semantics interface (Oxford University Press, ). Her research centers on the hypothesis of Direct Compositionality and on the related hypothesis that the semantics makes no use of variables. Her research program applies these hypotheses to a rich set of natural language phenomena. Tommi Jantunen holds an MA in General Linguistics from the University of Helsinki, Finland, and a PhD and a degree of Docent (Adjunct Professor) in Finnish Sign Language (FinSL) from the University of Jyväskylä (JyU), Finland. He is currently affiliated as an Academy Research Fellow at the Sign Language Centre in JyU, Department of Languages, and in his research he investigates FinSL grammar and phonetics as well as sign language technology. Kyle Johnson is Professor in the Linguistics Department at the University of Massachusetts Amherst. His research focuses on the connection between syntax and semantics. Andrew Kehler is a Professor in the Department of Linguistics at the University of California, San Diego. His primary research foci are discourse interpretation and pragmatics, studied from the perspectives of theoretical linguistics, psycholinguistics, and computational linguistics. His publications include Coherence, reference, and the theory of grammar () and numerous articles on topics such as ellipsis, discourse anaphora, and discourse coherence. Ruth Kempson FBA is Emeritus Professor of Linguistics at King’s College London and Honorary Research Associate with SOAS and the Cognitive Science research unit of QMUL, London. She is the lead developer of the Dynamic Syntax framework, and has a long-term research interest in the interface of syntax and pragmatics. Marjo van Koppen is Professor in Dutch Variation Linguistics in the Department of Languages, Literature, and Communication at Utrecht University, and senior researcher at the Meertens Institute in Amsterdam. She received her PhD, entitled ‘One probe, two goals: Aspects of agreement in Dutch dialectics’, from Leiden University in . Her main research interest is the morphosyntactic variation within Dutch dialects and the older stages of Dutch. The theoretical framework of her research is generative syntax. Howard Lasnik is Distinguished University Professor in the Department of Linguistics at the University of Maryland. He specializes in generative syntax and the formalization of syntactic theories. Among the specific topics he has worked on are phrase structure, anaphora, ellipsis, verbal morphology, case, and locality constraints on movement. His publications include eight books and over a hundred articles.

 

xvii

Winfried Lechner is Associate Professor for German Linguistics and Theoretical Linguistics at the National and Kapodistrian University of Athens, Greece. His main academic interests are located in the areas of syntax, semantics, and the interaction between these two components. He has been working on the logical syntax of scope and reconstruction, reflexivization, comparatives, ellipsis, the cross-linguistic typology of same/different, additive and scalar focus particles, Duke of York opacity, and the architecture of the grammar. Anikó Lipták is an Assistant Professor at the Leiden University Centre for Linguistics (LUCL), Leiden University. Her main field of research is comparative syntax and Hungarian, and she has published extensively on elliptical phenomena. She is currently researching issues concerning ellipsis identity, the interaction between ellipsis and morphology, and between ellipsis and intonation. Sophie Manus is an Associate Professor in Lyon, France, at the Laboratoire Dynamique du Langage, CNRS/Université Lyon . She completed her PhD in Linguistics at the Institut National des Langues et Civilisations Orientales in Paris. Since then, she has studied, taught, and directed research in Bantu languages, tone, morphology, and fieldwork methods applied to underdescribed/endangered languages, and she has more recently started working on tonal Chibchan languages spoken in Central America. Jason Merchant is the Lorna P. Straus Professor of Linguistics and Vice Provost at the University of Chicago. He is the author of The syntax of silence, the co-editor of Sluicing: Cross-linguistic explorations, and the author or co-author of more than two dozen articles on a wide variety of elliptical phenomena. Philip Miller teaches English and General Linguistics at the Université Paris Diderot (Paris ). He was recently Visiting Professor at the Universidade de São Paulo. He is the author of two monographs, Clitics and constituents in phrase structure grammar (Garland Publications, ) and Strong generative capacity (CSLI Publications, ). He has worked on clitics and on perception verbs. His current work is centered on ellipsis and anaphora, with specific interest in verbal ellipsis and verbal anaphora. Joanna Nykiel is Visiting Professor in the Department of English Language and Literature at Kyung Hee University, Seoul. Her research interests center on elliptical constructions, syntactic variation and modern quantitative methods of data analysis, the history of the English language, and, most recently, language processing. She has published articles in English Language and Linguistics, Language Variation and Change, and Lingua, among others. She is currently preparing a volume called Syntactic variation for publication in the Cambridge University Press series Key Topics in Syntax. Timothy Osborne is an Assistant Professor of Linguistics at Zhejiang University in Hangzhou, China. His research focus is on the theory of syntax, especially on the notions of constituency and dependency in syntactic analysis. Particular phenomena of syntax that he has explored are diagnostics for constituents, coordination, comparatives, ellipsis, and idiosyncratic meaning.

xviii

 

Cédric Patin completed his PhD in Linguistics at Université Paris  in . His thesis examined the tonal system of the Bantu language Shingazidja. After a postdoctorate at the Laboratoire de Linguistique Formelle (CNRS/Université Paris ), he accepted the position of Maître de conférences en phonétique et phonologie du français at the University of Lille in . His work focuses on the phonology of Bantu languages, with emphasis on the prosody–syntax interface. Florent Perek is a Lecturer in Cognitive Linguistics at the University of Birmingham in the Department of English Language and Applied Linguistics. His main research interests lie in the study of grammar from a cognitive and corpus linguistic perspective, with a particular focus on how syntactic constructions are mentally represented, how they are learned, and how they change over time. Sergio E. Ramos is a graduate student at the University of Illinois at Chicago. His main research interests lie at the intersection of bilingualism and linguistic theory. These include codeswitching, identity, and second language acquisition as well as topics of linguistic theory such as nominal ellipsis, sluicing, case, and Romance linguistics more broadly. Tom Roeper works in generative grammar on acquisition theory and experimentation and on syntactic morphology. Acquisition topics include long-distance movement, quantification, binding theory, passives, V, small clauses, aspect, ellipsis, implicatures, and theoretical work on Multiple Grammars, Strict Interfaces, Subset Theory, and African-American English dialect. In morphology he has worked on compounds, gerunds, productive affixation, and theoretically on lexical transformations, implicit arguments, and labeling theory. In addition, he has written two popular books, most recently The prism of grammar, five co-edited books, including Recursion: Complexity in cognition; co-authored an assessment test, Diagnostic evaluation of language disorders; and co-edited Language acquisition and studies in theoretical psycholinguistics. Currently he is working with collaborators on the emergence of recursive self-embedding and its theoretical implications in English, German, Dutch, Japanese, Romanian, Hungarian, and Pirahã. Andrés Saab studied Literature and Linguistics at the Universidad de Buenos Aires and at the Universidad Nacional del Comahue (Argentina). In , he defended his doctoral dissertation on the theory of ellipsis. Currently, he is Associate Researcher at the Consejo Nacional de Investigaciones Científicas y Técnicas (CONICET) and Associate Professor at the University of Buenos Aires. His main research areas are ellipsis, copy theory of movement, null subjects and, more broadly, the syntax–interface connection. Lewis P. Shapiro is an Emeritus Professor at San Diego State University. Research interests include charting the moment-by-moment unfolding of language and cognitive processing in neurologically healthy adults and those with brain damage; brain–language relations through lesion analyses and brain imaging; and the efficacy and neurological implications of treatment for adults with language disorders. Dr Shapiro’s work has been funded continuously through the US National Institutes of Health since .

 

xix

Tanja Temmerman is Assistant Professor of Dutch Linguistics at Université Saint-Louis— Bruxelles (Belgium), where she is also Lecturer in English Language and Head of the English Department. She obtained her PhD from Leiden University in  with a dissertation entitled ‘Multidominance, ellipsis, and quantifier scope’. Her principal research foci lie in (generative) syntax, issues at the syntax–phonology and syntax–semantics interfaces, Dutch dialectology, and comparative Germanic syntax. Specific topics of interest include ellipsis, the internal and external syntax of idioms, phase theory, long-distance dependencies, island effects, phrase structure, modals, and negation. Gary Thoms is Assistant Professor of Linguistics at New York University. His main research interests include ellipsis, reconstruction, dialectal and intraspeaker variation in English, Celtic syntax, polarity phenomena, predicate fronting, and the language of poetry. Maziar Toosarvandani is Associate Professor in the Department of Linguistics at the University of California, Santa Cruz. His primary research interests lie in syntax and semantics, primarily in Northern Paiute, Persian, and Zapotec. He has published in Language, Linguistic Inquiry, Natural Language and Linguistic Theory, Semantics and Pragmatics, and the International Journal of American Linguistics. Luis Vicente was a lecturer and researcher in the Department of Linguistics at the University of Potsdam, Germany. He earned a BA in English Philology from the Universidad de Deusto, Spain; a PhD in Linguistics from Leiden University, the Netherlands; and after a short lecture engagement at the University of Amsterdam, he was a postdoctoral researcher at the University of California Santa Cruz, supported by a prestigious scholarship from the Basque Government. He successfully obtained his Habilitation in July  at the University of Potsdam. He was a prolific researcher who published a large number of important and influential papers on various phenomena, mainly focusing on the interaction between syntax and semantics. He passed away on  February  at the age of  in Freiburg im Breisgau, Germany. His absence is a great loss to academia in general and linguistics in particular. He is sorely missed by us all. Chris Wilder is Professor of English Linguistics at the Norwegian University of Science and Technology in Trondheim, where he has been employed since . Prior to that he spent fifteen years as a researcher and lecturer in Germany and the USA. His research interests include English and German syntax, comparative syntax in general, and ellipsis and constituent-sharing phenomena in particular. Susanne Winkler is Professor of English Linguistics at the University of Tübingen (Germany). She has a long-standing research interest in syntactic theory, information structure, and the syntax–prosody interface. She has written extensively on the information structure of elliptical constructions and focus constructions. She is the author of Ellipsis and focus in generative grammar (Mouton de Gruyter, ) and Focus and secondary predication (Mouton de Gruyter, ), and of papers in a variety of volumes and journals. She directs a DFG-funded research project on Focus and Extraction in Complex Constructions and

xx

 

Islands (SFB ) and co-directs an interdisciplinary DFG-funded research training group (RTG ) on Ambiguity: Production and Preception. Masaya Yoshida is an Associate Professor in the Department of Linguistics at Northwestern University. His research interests are online sentence processing and syntax with special focus on the syntax and processing of ellipsis. He has published articles in a number of linguistics and psycholinguistics journals such as Linguistic Inquiry, Natural Language and Linguistic Theory, and Journal of Language, Cognition, and Neuroscience.

  ......................................................................................................................

    Theoretical and empirical perspectives ......................................................................................................................

     

. I

.................................................................................................................................. NATURAL language abounds in elliptical expressions, i.e. expressions that seem to leave certain aspects of their meaning unexpressed. Consider a random sample in (): ()

a. b. c. d. e. f.

Ed invited someone to his furniture shop, but I don’t know who. [doctor to nurse during surgery:] Scalpel! [diary entry:] Got up late again today. Philip ate more carrots than you did zucchinis. [text on a sign:] Caution: wet floor. A: Do you want an extra piece of cake? B: Do I?!

What all of these examples have in common is the fact that they feel incomplete in some sense. For instance, the intended meaning of (a) is ‘Ed invited someone to his furniture shop, but I don’t know who Ed invited to his furniture shop’, but the final portion of this sentence is missing. Similarly, even though a surgeon saying (b) only utters the noun scalpel, she conveys a directive to the effect that she be handed a scalpel. Similar observations can be made for the data in (c)–(f ). This discrepancy between what is overtly expressed and what is intended poses great challenges for theories of sound–meaning correspondence. According to the principle of compositionality usually attributed to Frege, the meaning of a complex utterance is a function of the meaning of its subparts and the way they are combined. In ellipsis, this principle appears to break down. For instance, the meaning of the surgeon’s utterance in (b) cannot simply be said to be a function of its subparts: there is meaning, but there is no corresponding sound. Given that ellipsis raises such fundamental questions about language



   

in particular and cognition in general, it should come as no surprise that it has garnered considerable linguistic and philosophical interest over the years. At the same time, the examples listed in () already suggest that ellipsis is by no means a unified phenomenon. To illustrate, while constructions such as those in (a), (d), and (f) can occur fairly freely in both written and spoken discourse, the remaining elliptical expressions are highly context- and/or register-dependent. For instance, an example like (c), in which the subject of the sentence (typically ‘I’) is left unexpressed, is only allowed in very specific registers of English, of which diary entries are a prime example. Moreover, variation in elliptical constructions is not only attested within but also across languages. Consider as a case in point the English elliptical example in ()—where the verbal predicate read War and Peace is missing from the second clause—and its close correlates from Dutch, French, and German in (a), (b), and (c), respectively. ()

Susan has read War and Peace, but Maria hasn’t.

()

a. *Susan Susan

heeft has

b. *Susan Susan

a lu La Guerre et la Paix, mais Maria n’ a pas. has read the war and the peace but Maria not has not

c. *Susan Susan

hat has

Oorlog war

Krieg war

en and

und and

Vrede peace

Frieden peace

gelezen, read

gelesen, read

maar but

aber but

Maria Maria

Maria Maria

heeft has

hat has

niet. not

nicht. not

While () is perfectly acceptable in English, attempts at a word-for-word translation of this elliptical sentence in closely related languages lead to ungrammaticality, showing how ellipsis is subject to cross-linguistic variation. It is observations such as these that have led to very detailed investigations of specific elliptical phenomena in the linguistic literature in recent years. Summing up, ellipsis is a topic that on the one hand raises general and fundamental questions about the workings of grammar and cognition, while on the other it is a veritable treasure trove of detailed and fine-grained points of inter- and intralinguistic variation. It is against this dual backdrop that the current Oxford Handbook of Ellipsis should be situated. As we make clear in the remainder of this introductory chapter, the handbook devotes attention both to fundamental theoretical questions and analyses surrounding elliptical phenomena, and to the empirical richness of this domain. This chapter is organized as follows. We first outline the general structure of the handbook and elaborate on (the rationale behind) its subdivision into four parts (section .). Next, in section ., we highlight some of the main results and generalizations that emerge from the contributions to the handbook, and in section . we conclude and provide a brief outlook on future research on ellipsis.

. S   

.................................................................................................................................. This handbook is subdivided into four parts, each of which highlights a specific aspect of the linguistic study of ellipsis. The first part (discussed in subsection ..) focuses on the theory of ellipsis, and explores the analytical approach taken towards ellipsis both in

   



various linguistic theoretical frameworks and in a number of subfields of linguistics. In the second part (subsection ..) the perspective is reversed, and ellipsis is construed not as the object but as the instrument of inquiry. The central question in the chapters of Part II is to what extent the study of ellipsis can shed new light on other research domains within linguistics. Part III (subsection ..) focuses on the traditional taxonomy of elliptical constructions known from the literature and explores the state of the art for each of them. Finally, Part IV (subsection ..) contains eleven case studies, each of which explores the elliptical inventory of a single language (or a set of closely related languages or language varieties), thus bearing witness to the empirical richness surrounding the phenomenon of ellipsis.

.. Part I: The theory of ellipsis The chapters making up the first part of the handbook can be divided into three sets. The first is the singleton consisting of the contribution by Jason Merchant entitled “Ellipsis: A survey of analytical approaches” (Chapter ). This chapter lays the groundwork for the rest of Part I: it defines the phenomena under investigation, lays out the central research questions, presents a taxonomy of approaches based on how they address those research questions, and weighs some of the evidence presented in favor of and against the various perspectives. The second set consists of Chapters –. These eight chapters discuss the analytical approach taken towards ellipsis in a specific theoretical framework. The frameworks in question are Transformational Grammar (Chapter , Howard Lasnik and Kenshi Funakoshi), Head-Driven Phrase Structure Grammar (Chapter , Jonathan Ginzburg and Philip Miller), Categorial Grammar (Chapter , Pauline Jacobson), Dependency Grammar (Chapter , Timothy Osborne), Simpler Syntax (Chapter , Peter W. Culicover and Ray Jackendoff), Construction Grammar (Chapter , Adele E. Goldberg and Florent Perek), Dynamic Syntax (Chapter , Ruth Kempson, Eleni Gregoromichelaki, Arash Eshghi, and Julian Hough), and Inquisitive Semantics (Chapter , Scott AnderBois). The third and final set of Part I is composed of Chapters –. These explore the theory of ellipsis not from the point of view of a linguistic framework, but with respect to a specific subfield or subdiscipline of linguistics. The areas covered in these five chapters are psycholinguistics (Chapter , Lyn Frazier), acquisition (Chapter , Tom Roeper), discourse (Chapter , Andrew Kehler), computational linguistics (Chapter , Daniel Hardt), and prosody (Chapter , Susanne Winkler). While ellipsis frequently surfaces as a topic of investigation in all of these frameworks and subfields of linguistics, only very rarely does this lead to an explicit comparison or evaluation of the various assumptions, arguments, and analyses. The current handbook wants to remedy this, and to this end we asked all authors of the chapters in Part I to focus on the same three theoretical issues. On the one hand, this ensures a high degree of thematic consistency across these chapters, while on the other it allows for a direct form of inter-chapter comparison. The three issues under investigation are (i) the abstract structure of the ellipsis site, (ii) recoverability/ellipsis identity, and (iii) licensing. We now briefly introduce each of these topics (see also Chapter  for further, more detailed, discussion). The conundrum regarding the sound–meaning correspondence in ellipsis raised earlier (see section .) can be paraphrased as a tension between semantics (meaning) on the one hand and phonology (sound) on the other. A central question in the study of ellipsis concerns the role of



   

syntax in this dichotomy: to what extent does an ellipsis site contain (unpronounced) syntactic structure? This is an issue of great contention in the literature: hypotheses range from the position that there is no (‘hidden’) structure whatsoever (e.g. Culicover and Jackendoff ) all the way to the other extreme, i.e. that there is full-fledged syntactic structure in an ellipsis site (e.g. Merchant ). From the latter point of view, the only difference between an elliptical sentence and a non-elliptical one is the lack of pronunciation of part of the former. Given that different theoretical frameworks and subdisciplines take radically opposing positions in this debate, the topic of “abstract structure of the ellipsis site” is ideally suited for comparison and evaluation across frameworks and subdisciplines. The second recurring topic in all chapters of Part I is recoverability, which concerns the question of how an ellipsis site gets its meaning, or more specifically, ellipsis identity, which concerns the question of the identity relation between an ellipsis site and its antecedent. Consider again the example in (a), repeated below as (). It is clear that we interpret the missing part of this sentence as ‘Ed invited to his furniture shop’ because the first half of this example contains exactly these words. Put differently, the meaning of an ellipsis site is recovered by virtue of an antecedent, with which it stands in a certain identity relation. However, the question of whether this identity relation is syntactic, semantic, morpholexical, pragmatic, etc., is far from settled, and one could even point to examples such as (b), (c), or (e) (repeated below as (a), (b), and (c), respectively) to question the very assumption that an antecedent is required in the first place. Once again, the literature on ellipsis contains many different answers to these questions, and the first part of the handbook provides a clear picture of the various arguments and positions. ()

Ed invited someone to his furniture shop, but I don’t know who.

()

a. [doctor to nurse during surgery:] Scalpel! b. [diary entry:] Got up late again today. c. [text on a sign:] Caution: wet floor.

The third central theme for the chapters of Part I is licensing, a cover term referring to restrictions on ellipsis (typically syntactic in nature) that are not related to recoverability or ellipsis identity. Consider for instance the following ungrammatical English example: ()

*John bought a blue bike and Mary bought a green.

In the second part of this sentence the noun bike is left unpronounced. Although it is abundantly clear from the context what the sentence should mean—i.e. there is no problem of recoverability—the instance of ellipsis illustrated here is ruled out in English. As shown in (), though, the Dutch analogue of () is perfectly acceptable. This shows that there are restrictions on ellipsis (within and across languages) above and beyond those related to recoverability or ellipsis identity. These restrictions typically go under the rubric of ‘licensing’ in the ellipsis literature. () Jan kocht een blauwe fiets en Marie kocht Jan bought a blue bike and Marie bought ‘John bought a blue bike and Mary bought a green one.’

een a

groene. green

   



Although the issue of licensing should be addressed in any comprehensive theory of ellipsis, only rarely does it explicitly feature in the discussion (notable exceptions are Lobeck  and Aelbrecht ). By adding licensing to the list of topics that every chapter in Part I addresses, the handbook aspires to put this theoretical notion firmly on the research agenda.

.. Part II: Ellipsis as a diagnostic tool The study of ellipsis is not only interesting in and of itself. Anyone who has ever taken an intro class in general linguistics, more specifically in constituency, knows that elliptical constructions can be used as a diagnostic tool to answer non-ellipsis-related research questions (in this simple case: the question of whether or not a string of words forms a constituent). Part II of the handbook takes precisely this reversed perspective: ellipsis is now no longer (or at least not exclusively) the object of study, but rather the means or the tool through which the study is carried out. The topics under investigation in this manner are movement and islands (Chapter , Klaus Abels), aphasia and acquisition (Chapter , Yosef Grodzinsky, Isabelle Deschamps, and Lewis P. Shapiro), parsing strategies (Chapter , Masaya Yoshida), and codeswitching (Chapter , Kay González-Vilbazo and Sergio E. Ramos). Each of these topics represents an active research area where elliptical phenomena have the potential of shedding important new light on the central research questions. For instance, certain types of ellipsis seem to bleed island effects. In Transformational Grammar, one of the central issues surrounding islands is whether these phenomena should be located at the conceptual-intentional or the articulatory-perceptual interface of the language module. The bleeding of island effects by ellipsis has been taken as an argument in favor of the latter position. Processing-based accounts, on the other hand, use island effects in ellipsis to examine the interplay between the syntactic and the discourse processor. Based on the lack of island effects in certain elliptical sentences, it is concluded that only the latter type of processor may violate islands. Similarly, the ability of aphasia patients to comprehend elliptical, i.e. incomplete, sentences can lead to a deeper understanding of the nature of their language deficit. These two examples are representative of the line of thinking that is developed in the chapters of Part II: in each case, the central research question does not concern ellipsis per se, but the study of elliptical phenomena brings us closer to answering that question.

.. Part III: Elliptical constructions The third part of the handbook is devoted to detailed studies of specific elliptical constructions. Generally speaking, ellipses appear to cluster at the clausal, predicate, and nominal level, corresponding to, respectively, clausal ellipsis, predicate ellipsis, and nominal ellipsis. Part III of the handbook starts off from this trichotomy, discussing first the most wellknown and best-investigated representative of clausal ellipsis, i.e. sluicing (Chapter , Luis Vicente), illustrated in (), followed by predicate ellipsis (Chapter , Lobke Aelbrecht and William Harwood), (), and nominal ellipsis (Chapter , Andrés Saab), ().



   

()

Tyrion Lannister: Listen to me! Sometimes, possession is an abstract concept. (Mord hits him.) When they captured me, they took my purse, but the gold is still mine! Mord: Where? (Game of Thrones, Season , Episode )

()

Frank Underwood: You’ve got this wrong, Marty. There is no ‘you’ and ‘I’ in education. (begins to backtrack) Well, I mean, yes, there’s the letters ‘U’ and ‘I’ in the word ‘education’. But ‘Education’ with a capital E, you know what I’m talking about! Marty Spinella (looking confused): I don’t think I do. (House of Cards, Season , Episode )

()

Frank Underwood: In Gaffney, we had our own brand of diplomacy. Shake with your right hand, but hold a rock in your left. (House of Cards, Season , Episode )

The remaining five chapters of this part of the handbook cover the following elliptical constructions: gapping and stripping (Chapter , Kyle Johnson), fragments (Chapter , Alison Hall), comparative deletion (Chapter , Winfried Lechner), Null Complement Anaphora (Chapter , Marcela Depiante), and Conjunction Reduction and Right-Node Raising (Chapter , Chris Wilder). Each chapter thus zooms in on one particular subtype—or very often, family of subtypes—of ellipsis, and provides a systematic and detailed overview of its basic properties and distinctive characteristics. In addition to covering the central empirical generalizations, each chapter presents a survey of the main theoretical concerns these elliptical constructions raise and discusses different analytical approaches to them. As such, this part of the handbook is meant to serve as a reference work for anyone interested in a particular subtype of ellipsis.

.. Part IV: Case studies The fourth and final part of the handbook presents case studies of ellipsis in specific languages. The languages under investigation are: Dutch (Chapter , Norbert Corver and Marjo van Koppen), Finnish Sign Language (Chapter , Tommi Jantunen), French (Chapter , Anne Dagnac), Hungarian (Chapter , Anikó Lipták), Indonesian (Chapter , Catherine Fortin), Japanese (Chapter , Teruhiko Fukaya), Kiswahili and Shingazidja (Chapter , Cédric Patin and Sophie Manus), Persian (Chapter , Maziar Toosarvandani), Polish (Chapter , Joanna Nykiel), Russian (Chapter , John Frederick Bailyn and Tatiana Bondarenko), and varieties of English (Chapter , Gary Thoms). These languages were selected on the basis of two criteria: (a) the fact that they exhibit elliptical phenomena that were previously unattested and/or that shed new light on some of the more mainstream generalizations and theories, and (to a lesser extent) (b) typological spread. All of the chapters in Part IV have a double orientation. Firstly, they inventory which of the elliptical constructions discussed in Part III are attested in the language and which aren’t, and they describe their properties. This provides typological depth to the chapters in Part III. Moreover, this offers ample opportunities for inter-chapter comparison in Part IV:

   



these language-specific studies contribute meaningfully to our understanding of the ways in which the elliptical phenomena under scrutiny manifest themselves cross-linguistically, and they open the door to formulating a more systematic cross-linguistic theory of the distribution of ellipsis types. Secondly, the chapters in Part IV focus on facts and observations that are new, previously undiscussed, and/or off the beaten track, and that therefore question or put in a different light the hypotheses and theories based on better-known languages. To name but a few concrete examples: (i) Chapter  considers the role that gesture and mime play in elliptical phenomena in Finnish Sign Language; (ii) Chapter  shows that in many Japanese ellipsis phenomena (including sluicing, fragments, and stripping), the presence or absence of a case marker on the ellipsis remnant plays a crucial role, with case-marked and non-case-marked fragments being analyzed as instances of surface and deep anaphora, respectively; and (iii) Chapter  considers in detail the fact that Indonesian permits prepositions to be omitted in certain elliptical contexts, despite preposition stranding being otherwise prohibited in the language. This is unexpected in light of the cross-linguistically robust generalization that preposition omission under ellipsis tracks preposition stranding in non-elliptical contexts (Merchant ).

. R  G

.................................................................................................................................. In this section we highlight some of the main results and generalizations that emerge from the contributions to this handbook. Needless to say, it is neither realistic nor feasible to do full justice to a + page volume in an introductory chapter, so we needed to be selective in what we present in this section. The common thread throughout the discussion is the question to what extent the structure and goals set out for the handbook (as described in section .) have yielded interesting results and generalizations. Accordingly, this section is structured parallel to the preceding one, with one subsection devoted to each part of the handbook.

.. Part I: The theory of ellipsis Given that the chapters in Part I all address the same three theoretical questions (abstract structure, recoverability/ellipsis identity, and licensing; see subsection ..), it speaks to reason to use these three questions as the structuring principle for a summarizing overview of those chapters. Table . lists for each chapter (a) whether or not it assumes that there is abstract syntactic structure in the ellipsis site, (b) which linguistic submodule is responsible for providing an elliptical expression with an interpretation, or more specifically, which module of the grammar is relevant for defining the identity relation between an ellipsis site and its antecedent (recoverability/ellipsis identity), and (c) what mechanism licenses ellipsis beyond the issue of recoverability. Before turning to the table in more detail, let us make explicit three ground rules we adhered to in creating this overview table. First, the values listed in the cells represent the approach favored by the authors in these chapters. As such, they do not necessarily represent the views of the theoretical framework or linguistic



   

Table . Overview of Part I of the handbook: Abstract structure, recoverability, and licensing Chapter

Abstract structure

 Lasnik and Funakoshi yes

Recoverability

Licensing

syntactic/semantic local structural relation with licensing head discourse semanticb QUD-based, supplemented with construction-specific restrictions discourse semantic type-shifting rulec syntactic/semantic only catenae can be elided semantic construction-specific restrictions

 Ginzburg and Miller

noa

 Jacobson  Osborne  Culicover and Jackendoff  Goldberg and Perek  Kempson, Gregoromichelaki, Eshghi, and Hough  AnderBois  Frazier

no yes no no mixedd

semantic construction-specific restrictions syntactic/semantic morphosyntactic constraints imposed by ellipsis remnants

yes yes

 Roeper

yes

 Kehler

yes

 Hardt

agnostice

 Winkler

yes

syntactic/semantic no framework-specific constraints syntactic constraints on establishing discourse coherence syntactic/semantic local structural relation with licensing head syntactic/semantic constraints on establishing discourse coherence syntactic/semantic [not directly addressed in the computational literature] syntactic/semantic/ contrastive accent on remnants and prosodic deaccenting of given material

a

An exception is made for certain varieties of left- and right-peripheral ellipsis (including RightNode Raising), where full structure is present, but partly unpronounced. b The term “discourse semantic” is meant to refer to approaches where the ellipsis antecedent is located in the discourse context. c This does not apply to fragment answers, which are licensed by being a part of a Qu(estion)– Ans(wer) unit. d This chapter adopts multiple mechanisms to derive ellipsis, only some of which generate structure inside an ellipsis site. e This chapter presents two main computational approaches, one of which assumes abstract structure, and one of which does not.

subdiscipline as a whole.1 Very often, the authors point out that even within a certain framework or discipline there is disagreement with respect to these three issues, but they then proceed to express a preference for a particular position in the debate. It is this position that is represented in Table .. Second, a “yes” in the column abstract structure is meant to cover both accounts that assume ellipsis sites contain a fully fledged syntactic This is also why we represent the chapters by their author names in Table ., rather than referring to the theoretical framework or linguistic subdiscipline that is discussed in the chapter. 1

   



representation and analyses that assume only a minimal instantiation of such structure (typically represented as a pro-form). Finally, note that the first chapter of Part I, Chapter , is missing from this table: given that it is a survey chapter that is explicitly intended to transcend individual analyses and approaches, we did not include it here. What can we learn from this table? With respect to the first question—the presence or absence of abstract syntactic structure—it is clear that this issue remains, to this day, a very contentious one. Proponents of the ‘no structure’ approach often adopt an Occam’s razorstyle position, which is backed up by cases where elliptical remnants fail to show connectivity effects with their purported syntactically fully represented source. Compare and contrast in this respect the elliptical reply in () with the non-elliptical example in () (both examples are from Ginzburg and Miller, this volume). ()

A: Who appeared to be the cause of [John and Mary]1’s problems? B: Each other1.

()

*Each other1 appeared to be the cause of [John and Mary]1’s problems.

If B’s elliptical reply in () contained (an abstract version of ) the non-elliptical utterance in (), it should be as ungrammatical as that example, quod non. Hence, adopting abstract syntactic structure in the case of () makes incorrect predictions and should be avoided. On the other hand, proponents of the structural approach also make use of connectivity effects in their argumentation, but they cite different types of data, such as the sluicing example in () (from Lasnik and Funakoshi, this volume, but orginally from Ross b). () Er will jemandem schmeicheln, aber sie wissen nicht { * wen / wem }. he want someone.DAT flatter but they know not who.ACC who.DAT ‘He wants to praise someone, but they don’t know who.’ (German) The fact that the sluiced wh-phrase necessarily bears dative case—rather than, for example, the accusative that would be assigned by the immediately governing verb wissen ‘to know’—is argued to show that there must be an unpronounced copy of the dativeassigning verb schmeicheln ‘to flatter’ inside the ellipsis site, i.e. there must be abstract syntactic structure. The contrast between ()/() on the one hand and () on the other shows that the standoff between structural and non-structural approaches to ellipsis largely boils down to which set of data should be given primacy. It seems, then, that more systematic and extensive data inventories are needed before this issue can be settled. The chapters in Part I of the handbook can be seen as a first step in that direction. The next column in Table . concerns recoverability, and more specifically which module of the grammar is relevant for defining the identity relation between an ellipsis site and its antecedent (ellipsis identity). Contrary to the previous issue, there is a greater degree of agreement regarding this second question, in that there seems to be a (near-)general consensus that there is at least a semantic component to recoverability, above and beyond any morphosyntactic or lexical requirements there might be. For the ‘no structure’ analyses, this position is a logical necessity (as is also pointed out by Merchant,



   

this volume), in that the absence of syntactic structure inside an ellipsis site makes it impossible to compare that structure to that of the antecedent. For the other accounts, though, it should be pointed out that the label “syntactic/semantic” used in Table . hides a fair amount of variation that exists between the individual accounts. While some authors (Lasnik and Funakoshi, AnderBois, Winkler) argue that recoverability intrinsically has both a syntactic and a semantic component, others claim that the recoverability mechanism can be more (or exclusively) semantic in some cases and more (or exclusively) syntactic in others, depending on the type of elliptical construction (Osborne), the stage in the acquisition process (Roeper), or the type of mechanism used to derive ellipsis (Kempson et al., Kehler, Hardt). Finally, we turn to the issue of licensing. As was pointed out in subsection .., licensing is a relative newcomer to the stage of theoretical ellipsis research. This is reflected in the final column of Table .: several frameworks and subdisciplines either make no specific claims with respect to this issue, or they lack a general, overarching licensing mechanism. Several chapters even—implicitly or explicitly—question the need for such an overarching theory, arguing instead that licensing is inherently construction-specific and hence sui generis for each individual ellipsis phenomenon (Ginzburg and Miller, Culicover and Jackendoff, Goldberg and Perek, Kempson et al.). The point is well-taken: to what extent is it possible to reduce the great diversity of elliptical constructions—recall also the sample in () in section .—to a single licensing mechanism? Another issue that emerges from the chapters in Part I is that, contrary to what is commonly proposed in the literature (see in particular Lobeck ; Aelbrecht ), licensing is not necessarily strictly syntactic in nature. Ginzburg and Miller (this volume) propose that ellipsis is licensed when it provides an answer to the Q(uestion) U(nder) D(iscussion) (see also AnderBois, this volume, though for him it is part of recoverability/ellipsis identity), while both Frazier (this volume) and Kehler (this volume) discuss constraints on establishing discourse coherence under the rubric of licensing. In short, licensing is by no means a side character in the study of ellipsis, and we expect it will grow into a full-fledged research track of its own.

.. Part II: Ellipsis as a diagnostic tool Recall the starting point of Part II of the handbook: to what extent can ellipsis be used as a tool or instrument in the study of other linguistic phenomena? Before reflecting on this part in more general terms, we first provide a brief overview of each individual chapter. Abels (this volume) discusses the interaction between ellipsis on the one hand and movement and islands on the other. As was first discovered by Ross (b), sluicing can ameliorate island effects. That is, an A′-dependency that is illicit because it crosses an island boundary can become licit when the offending structure is elided. This fact, when taken at face value, can provide valuable insight into the nature of island constraints: if simply not pronouncing an island can bleed its effect, whatever is causing that effect must be of a phonological nature (i.e. operative at the PF-interface). As Abels points out, however, once one looks more closely at the facts, they are much more complicated than they seem to be at first sight. He concludes that while ellipsis certainly has the potential of leading to new insights regarding islands (and movement, the second topic of the chapter), it cannot yet live up to that potential. Grodzinsky et al. (this volume) focus on comprehension of VP-ellipsis in

   



speakers with an incomplete language faculty: among others, patients with Broca’s aphasia. The fact that patients with Broca’s aphasia are relatively successful in comprehending sentences with VP-ellipsis suggests that Broca’s aphasia is not a general failure of working memory, i.e. a difficulty of dealing with “syntactically complex” structures, but that it is a much more specific impairment, possibly one that specifically targets movement dependencies. This in turn might have repercussions for the proper analysis of ellipsis, Grodzinsky et al. argue, in that a movement-based approach towards VP-ellipsis (see e.g. Johnson b) seems unlikely. Yoshida (this volume) uses psycholinguistic experiments involving elliptical constructions to gain more insight into the nature of the human parser. He ends up concluding that “the parsing strategies that the human parser employs achieve incremental, rapid, and grammatically detailed structure building,” which is in line with the results of sentence processing studies in other domains. Conversely, the evidence he discusses also suggests that ellipsis sites contain abstract syntactic structure, in that the parser builds the structure of ellipsis sites by copying the structure of the antecedent site. Finally, González-Vilbazo and Ramos (this volume) review the interaction between ellipsis and codeswitching. They argue, on the basis of German–Spanish elliptical codeswitching data, in favor of a constraint-free theory of codeswitching, i.e. a theory without codeswitchingspecific rules. At the same time, their data is also very informative about the theory of ellipsis itself: given that it is possible to codeswitch inside an ellipsis site, such facts can shed important new light on the nature of the identity relation between the ellipsis site and its antecedent (see subsections and .. and ..). Overlooking the four chapters that make up Part II of the handbook, it becomes clear that the dichotomy we started out with about ellipsis being the object or the instrument of inquiry is to a certain extent a false one. All four chapters make clear that there is a fruitful two-way interaction between the theory of ellipsis on the one hand, and that of whatever other phenomenon is under investigation on the other. It is only based on a specific theory of ellipsis that one can draw conclusions about other domains of grammar and vice versa. At the same time, though, we believe the approach developed in this second part is a potentially very fruitful one, and hence one that should be continued to be explored in future research. In particular, there are various other domains where we can see the study of ellipsis having a real impact, from constituency (Depiante and Hankamer ; Sailor and Thoms ), to the proper definition of the QUD (Kehler, this volume), to the workings of memory (Martin and McElree ).

.. Part III: Elliptical constructions As pointed out above (subsection ..), in structuring Part III of the handbook, we started out from a taxonomy that is quite common in the ellipsis literature. Needless to say, such an approach—like any attempt at categorization—risks not capturing phenomena that sit inbetween different categories in the taxonomy and thus fall between the cracks. Indeed, as noted by Merchant (this volume): “There are many other kinds of phenomena that go under the rubric of ellipsis as well, some better investigated than others, including argument drop, article drop, haplology, diary language and headlinese, subjectless infinitivals, copula drop, situational ellipses, small clauses, and many more; some are context-sensitive, and some are not. For various (and still incomplete) taxonomies of the missing, see Klein 



   

and Hennig : –” (and see McShane  for a similarly broad view of what qualifies as ellipsis). Interestingly, though, when taking a look at the chapters (in all four parts) of this handbook, one is struck by the fact that some of these ‘non-conventional’ types of ellipsis have found their way into one or more chapters, while others remain unmentioned. Two notable examples of the former type are argument drop (or pro-drop) and copula drop, which are discussed in several chapters of this handbook. Highly context-dependent types such as diary language or headlinese, however, have not made their way in. This might suggest that there is, after all, a natural subdivision of elliptical phenomena into meaningful subgroups (see also Merchant  for related discussion). A partially related observation is that in the chapters of this handbook, the boundaries between the various (sub)types of ellipsis are becoming increasingly blurred. For instance, when considering both the empirical descriptions and the theoretical approaches to sluicing and fragments in different chapters (and even different parts) of this handbook, one could consider not analyzing them as separate types of ellipsis, but rather as instantiations of one and the same elliptical phenomenon (as suggested, for instance, by van Craenenbroeck and Lipták , , and Temmerman ). Similarly, it becomes apparent in a number of chapters (both in Part III and in Part IV) that distinguishing between nominal ellipsis and argument drop (pro-drop)—or even between argument drop and (verb-stranding) VP-ellipsis—is not a trivial (and therefore, perhaps an unnecessary) task (cf. also Ginzburg and Miller, this volume, for similar remarks). We find such developments promising, given that they steer us away from the focus on individual constructions, a line of thinking that is reminiscent of the early construction-specific days of generative grammar (see also Johnson a: ). It seems fruitful to us to not think of specific elliptical (sub)types as construction-specific phenomena, but rather, to try to formulate empirical and theoretical generalizations that transcend the traditional taxonomy.

.. Part IV: Case studies The language-specific chapters of Part IV inventory which of the elliptical phenomena discussed in Part III are attested (and which ones are not) in the language and describe their properties. Table . presents an overview of the different languages and elliptical phenomena under scrutiny. Before discussing the contents of the table, let us first make two observations about how it came about. First off, it should be noted that this overview is crucially based on surface patterns: whether a given construction can or should be analyzed as involving ellipsis or not—a matter discussed in detail in most of the chapters of Part IV—is not taken into account here. For example, a pattern such as the Persian one in () (taken from Toosarvandani, this volume) leads to the value “yes” in the SL-column for Persian because it displays the relevant surface pattern (a constituent question reduced to its wh-phrase). Secondly, the use of gray shading indicates that these particular elliptical phenomena are not discussed in the relevant chapters. ()

Râmin ye chiz-i xarid. Hads Ramin one thing-IND buy.PAST.SG guess ‘Ramin bought something. Guess what.’

be-zan SUBJ-hit

chi. what (Persian)

   



Table . Overview of Part IV of the handbook: Cross-linguistic distribution of the main ellipsis types Language

SL

PE

G

STR

FR

NPE

CD

NCA

CRa

RNR

Dutch

yes

yes

yes

yes

yes

yes

yes

yesb

yes

yes

Englishc

yes

yes

yes

yes

yes

yes

yes

yes

yes

yes

Finnish Sign

yes

yes

yes

yes

yes

yes

yes

yes

yes

yes

yes

yes

Hungarian yes

yes

yes

yes

yes

yes

yes

yes

Indonesian yes

yes

yes

yes

yes

yes

yes

yes

Japanese

yes

yes

yes

yes

yes

yes

Kiswahili

yes

yes

yes

yes

yes

yes

Persian

yes

yes

yes

yes

yes

yes

no

Polish

yes

yes

yes

yes

yes

yes

yes

Russian

yes

yes

yes

yes

yes

yes

yes

Shingazidja yes

yes

yes

yes

yes

yes

no

yes

Language French

yes

yes

yes yes

yes

yes?d yes

yes

yes

yese

yes

yes

yes

yes

yes yes

yes

yes

yes

Abbreviations: SL = sluicing, PE = predicate ellipsis, G = gapping, STR = stripping, FR = fragments, NPE = noun phrase ellipsis, CD = comparative deletion, CR = Conjunction Reduction, RNR = Right-Node Raising a We include in CR the phenomenon sometimes referred to as non-constituent coordination, e.g. John gave a book to Mary and a car to Bill. b The set of verbs that can occur in this pattern seems to be very limited. c We have broadened the scope of Chapter 38 from “varieties of English” to English in general. Thoms (this volume) mostly focuses on predicate ellipsis, pointing out that little is known about dialectal variation in the domain of clausal or nominal ellipsis. However, the standard English facts in these domains are well-known from the literature. d Whether Indonesian exhibits RNR is somewhat controversial. Moreover, there seems to be speaker variation. e Only marginally possible, subject to speaker variation.

One thing that jumps out from this table is that (surface patterns of ) elliptical constructions are cross-linguistically widespread: most if not all of the ellipsis phenomena are attested in each of the twelve languages studied.2 While this is an interesting 2

Comparative deletion might be the odd one out: although data is lacking from four of our twelve languages, two of the remaining eight lack this elliptical construction.



   

Table . Overview of the cross-linguistic distribution of various subtypes of predicate ellipsis Language

AuxVPE

VVPE

ModVPE

PG

ACD

Dutch

noa

no

yes

noa

English

yes

no

yes

yes

yes

yesb

Finnish Sign Language French

no

no

yes

no

yes

Hungarian

yes

yes

yes

yes

yes

Indonesian

yes

no

yes

yesb

Japanese

no

yes

no

no

Kiswahili

no

yes

no

no

Persian

no

yes

no

no

Polish

yes

yes

Russian

yes

yes

Shingazidja

no

yes

no

yes

no

yes

no

no

Abbreviations: AuxVPE = auxiliary-stranding VP-ellipsis, VVPE = main-verb-stranding VPellipsis, ModVPE = modal-stranding VP-ellipsis, PG = pseudogapping, ACD = AntecedentContained Deletion a Marginally possible in comparatives. b Single example, judgments unclear.

observation in and of itself, it might also be partly illusion, caused by the fairly coarse granularity of the phenomena listed in the table (as well as the criteria used to describe the phenomena, i.e. our reliance on surface patterns). This suspicion seems to be confirmed once we increase the level of detail in our overview. Table . takes one phenomenon from our first table—the most intensively studied one, predicate ellipsis—and splits it up into five subtypes. The resulting picture is considerably more varied than could be gleaned from Table .. Predicate ellipsis is a cover term for several types of ellipsis targeting the (verbal) predicate. Different instantiations include auxiliary-stranding VP-ellipsis, modal-stranding VP-ellipsis, and main-verb-stranding VP-ellipsis. Table . clearly shows that the twelve languages studied in this part of the handbook do not all behave similarly. At first glance, the following picture seems to emerge:

   



• When a language exhibits AuxVPE, it also has ModVPE (but not vice versa).3 • When a language exhibits AuxVPE (and hence also ModVPE), it is possible that it also has VVPE (as in Hungarian), but this is by no means required (see English and Indonesian). • A language can exhibit only ModVPE (as in Dutch or French), or only VVPE (as in Persian). In addition to these three types of VP-ellipsis, there is also pseudogapping and Antecedent-Contained Deletion (ACD), two elliptical subtypes which also do not show a uniform cross-linguistical distribution. It is not straightforward to come up with a generalization that describes the (non-)occurrence of pseudogapping or ACD in a given language. For instance, the possible generalization that languages that allow AuxVPE also exhibit pseudogapping (cf. Hungarian, English, and Indonesian) is contradicted by the fact that Polish has AuxVPE, but does not exhibit pseudogapping. It could be the case, then, that AuxVPE is a necessary but not a sufficient condition for pseudogapping. Adding to the variation in Table . is the fact that several languages exhibit predicate ellipsis phenomena that seem to be sui generis, i.e. that are specific to a single language (usually because it is dependent on other material that is specific to that language). For example, Hungarian also exhibits preverb-stranding VP-ellipsis, British English has a type of predicate ellipsis that looks identical to VP-ellipsis save for the addition of a non-finite form of the verb do (known as British English do), and verb-stranding VPE in Persian is actually v-stranding VPE. The contrast between (the second column of) Table . and Table . suggests that were one to do the same exercise for, say, sluicing or nominal ellipsis, similar patterns of variation would emerge. Possible sources of variation that come to mind are (i) the different licensers of nominal ellipsis across languages (determiners, demonstratives, adjectives, possessives, etc.—recall the contrast between () and ()), (ii) the cross-linguistic distribution of different types of sluicing (sprouting, swiping, spading, multiple sluicing, etc.), (iii) the cross-linguistic distribution of polarity ellipsis, (iv) variation in the number of remnants allowed by gapping,4 and (v) variation in connectivity effects in elliptical constructions (islands, preposition stranding, case marking . . . ). In short, the empirical picture is complex and nuanced, leading to the all too familiar tension between empirical coverage and theoretical parsimony: to what extent is it feasible and/or desirable to try to construct a general, overarching theory of ellipsis? Given that this very same issue has cropped up in the previous subsections as well, it seems to us that this is one of the central questions in the linguistic study of ellipsis.

3 Note that we cannot be sure of this generalization for Polish, as Chapter  does not contain an example of ModVPE. 4 See, for instance, Patin and Manus (this volume), who show that Kiswahili allows for gapping with multiple remnants, while in the closely related Bantu language Shingazidja there is an upper limit of two gapping remnants.



   

. C

.................................................................................................................................. Taken together, the four parts of this handbook present a comprehensive, in-depth, and balanced discussion of the phenomenon of ellipsis in natural language. They devote space to well-established theories of and generalizations about ellipsis, but at the same time leave room for cutting-edge research that broadens the scope of the investigations, opening up exciting new prospects, empirically as well as theoretically. One example of the latter is the role that the notion of Question Under Discussion (QUD) plays in much current ellipsis research, as is evidenced by the fact that it shows up in a number of chapters, across different theoretical frameworks and across the different parts of the handbook. With respect to the empirical study of elliptical phenomena, it is clear that more work is needed, not only on more constructions in more languages and in more detail and depth, but also on more types of data (see, for example, the role played by corpus material in Ginzburg and Miller, this volume). In short, the work is far from done, but the future of the linguistic study of ellipsis looks very bright: it is a thriving subfield of linguistics that offers exciting prospects for new discoveries and breakthrough developments, both empirically and theoretically. It seems only fitting that we would end this chapter on an elliptical note, and so we are.

A We would like to thank the authors of the chapters, without whom there simply would be no handbook, for their interest in this project and their valuable contributions. A special mention goes out to our dear colleague and friend Luis Vicente, author of Chapter , who unfortunately passed away before he could see the completion of this handbook. We also want to express our gratitude to the following reviewers, who invested time and effort in assisting in peer-reviewing the chapters, providing excellent comments and feedback to the authors that substantially helped improve the quality of the handbook: Lobke Aelbrecht, Doug Arnold, Marc Authier, Julia Bácskai-Atkári, Matt Barros, Anne Bezuidenhout, Rajesh Bhatt, Željko Bošković, Hans Broekhuis, Ronnie Cann, Carlo Cecchetto, Simon Charlow, Rui Chaves, Ivano Ciardelli, Barbara Citko, Charles Clifton, Peter W. Culicover, Sonia Cyrino, Anne Dagnac, Marcel den Dikken, Michael Yoshitaka Erlewine, Hans-Werner Eroms, Frank van Eynde, Caroline Féry, Klaus Fischer, Lyn Frazier, Adele E. Goldberg, Lydia Grebenyova, Vera Gribanova, James Griffiths, Alison Hall, Daniel Hardt, Katharina Hartmann, William Harwood, Angeliek van Hout, Kyle Johnson, Ruth Kempson, Chris Kennedy, Katalin É. Kiss, Greg Kobele, Marjo van Koppen, Diane Lillo-Martin, Anikó Lipták, Amalia Lombart, Andrea E. Martin, Jonathan McDonald, Jeff McSwan, Jason Merchant, William O’Grady, Timothy Osborne, Victor Pan, Dan Parker, Eric Potsdam, Esther Ruigendijk, Jeffrey Runner, Craig Sailor, Mamuro Saito, Ana Lúcia Santos, Yosuke Sato, Petra Sleeman, Jennifer Spenader, Jon Sprouse, Mark Steedman, Anna Szabolcsi, Adam Szczegielniak, Daiko Takahashi, Kensuke Takita, Azita H. Taleghani, Maziar Toosarvandani, Luis Vicente, Mark de Vries, Jenneke van der Wal, Andrew Weir, Masaya Yoshida, and Jochen Zeller. Lastly, we are very grateful to Julia Steer (Linguistics Commissioning Editor), Vicki Sunter (Senior Assistant Commissioning Editor, Linguistics), Karen Morgan and Verity Rimmer (Editorial Administrators) at OUP, Manikandan Chandrasekaran (Project Manager at SPi Global), and Jess Smith (copyeditor), for their patience and assistance throughout this (long!) editing process.

  .............................................................................................................

THE THEORY OF ELLIPSIS .............................................................................................................

  ......................................................................................................................

 A survey of analytical approaches ......................................................................................................................

 

. I: T 

.................................................................................................................................. T term ellipsis has been applied to a wide range of phenomena across the centuries, from any situation in which words appear to be missing (in St Isidore’s definition), to a much narrower range of particular constructions. Ellipsis continues to be of central interest to theorists of language exactly because it represents a situation where the usual form/ meaning mappings, the algorithms, structures, rules, and constraints that in non-elliptical sentences allow us to map sounds and gestures onto their corresponding meanings, break down. In fact, in ellipsis, the usual mappings seem to be entirely absent. In ellipsis, there is meaning without form. VP-ellipsis and sluicing are two of the best-investigated instances of ellipsis and generally show remarkable similarities in the demands they make of the discourse, both usually necessitating some equivalent antecedent which is subject to some kind of parallelism. It is no exaggeration to say that debates over the nature of this parallelism have formed the core of most of the generative work on ellipsis over the last forty years. Almost all conceivable positions on the parallelism question have been explored and advanced, and these debates are important exactly because they are often used to argue for the necessity of one or another kind of linguistic representation. Most of the debate is located in the arena of semantics and abstract syntactic structures—it is clear that surface syntactic or phonological parallelism is not at stake—and as such, elliptical structures often play an important role in fundamental ontological debates in linguistics. The logic is clear: if the parallelism or identity conditions found in ellipsis resolution require reference to certain kinds of objects, then our theories of linguistic competence must countenance objects of that kind. In generative linguistics, research has focused largely on two sets of constructions.1 Central examples of the first set, drawn from English, include sluicing as in (), predicate or 1 There are many other kinds of phenomena that go under the rubric of ellipsis as well, some better investigated than others, including argument drop, article drop, haplology, diary language and headlinese, subjectless infinitivals, copula drop, situational ellipses, small clauses, and many more;





verb phrase ellipsis (VP-ellipsis) as in (), and NP-ellipsis (or N′-ellipsis, or more broadly, nominal ellipsis) as in (). Ellipsis of at least one of these kinds seems to be found in every language in which it has been looked for, though a systematic cross-linguistic theory of the distribution of ellipsis types remains to be formulated. ()

Lauren can play something, but I don’t know what.

()

Lauren can play the guitar and Mike can, too.

()

Lauren can play five instruments, and Mike can play six.

In each case, the second clause can be understood as in ()–(). ()

Lauren can play something, but I don’t know what Lauren can play.

()

Lauren can play the guitar and Mike can play the guitar, too.

()

Lauren can play five instruments, and Mike can play six instruments.

These three kinds of ellipsis are distinguished as well by the fact that distributional facts lead us to expect to find structural elements corresponding to the perceived interpretations: wh-phrases as in () require clausal sources, auxiliaries like can in () take VP complements, and determiner-like elements such as six in () require nominal complements. In other words, selectional or subcategorizational properties of particular elements require us to posit elided structures in ()–(), if we adopt the null hypothesis that these properties are uniform across ()–(). A second set of constructions in which ellipsis has been invoked include stripping (or ‘bare argument ellipsis’) in (), gapping in (), fragment answers in (), as well as a host of other cases that fall under the general rubric of ‘conjunction reduction’: ()

a. Lauren can play the guitar, {and Mike, too/and Mike as well/but not Mike}. b. Lauren can play the guitar better than Mike.

()

a. Lauren can play the guitar, and Mike the violin. b. Lauren can play the guitar better than Mike the violin.

()

Q: Who can play the guitar? A: (Not) Lauren.

All of these structures2 have been the focus of intense theoretical interest over the past four decades, and vast bibliographies can be compiled for each of the above phenomena. I can some are context-sensitive, and some are not. For various (and still incomplete) taxonomies of the missing, see Klein () and Hennig (: –.). 2 I’ve omitted pseudogapping, a construction that seems to mix properties of gapping and VP-ellipsis; see Gengel ().

:     



make no pretense of bibliographic completeness here, and refer the reader to excellent recent surveys for a more detailed treatment of the literature, especially Hartmann (), Johnson (b), Winkler and Schwabe (), Goldberg (), Winkler (), van Craenenbroeck (b), Reich (), van Craenenbroeck and Merchant (), and the introduction to Johnson (b). In what follows, I will examine some representative examples of approaches to the above and discuss their relative merits. In analyzing ellipsis, three questions have occupied much of the literature. The first is given in (), what I will call the structure question. ()

In elliptical constructions, is there syntactic structure that is unpronounced?

The answer that is given to () has far-reaching implications for the theory of grammar. If the answer is positive, we must countenance theories of grammars that permit unpronounced phrases and heads. If the answer is negative, there is the possibility that syntax may be ‘wyhiwyg’ (‘what you hear is what you get’), with no unpronounced elements. The debate on this question bears some resemblance to debates in the mid-twentieth century about the nature of abstractness and the phoneme: there are good reasons to prefer a parsimonious theory of any domain of data, but not at the expense of coverage of the facts. Some of the various strands of evidence that have been brought to bear in attempts to answer () are laid out in the next section and compared in detail in sections . and .. The second major question is what I will call the identity question: ()

What is the relationship between the understood material in ellipsis and its antecedent?

This question has generally been answered in terms of various kinds of posited identity relations: elided material (call it XPE) must be identical or parallel to or resolvable by some antecedent phrase (YPA), where the identity (or parallelism, or resolution) may be semantic or syntactic, or some mix of the two. The various approaches to the identity question are addressed in section 2.6. Putting these first two questions schematically, then, we have the following: . Is there syntax internal to the ellipsis site? . The understood material is identical to some antecedent. Is the relevant kind of identity syntactic (defined over phrase markers or syntactic derivations of some sort) or semantic (defined over semantic representations or computations of some sort)? Table . organizes a representative selection of the literature by the answers it proposes to these two questions; few if any of the works deal with all kinds of ellipsis: the table assigns them to the various categories based on the kinds of ellipsis they do discuss, though this should not impute to any of them necessarily a uniform theory of ellipsis (most deal only with VP-ellipsis in English). Further, the row labeled ‘both’ includes theories that are hybrid in various ways involving an admixture of syntactic and semantic requirements, sometimes uniformly, and sometimes varying by construction or context.





Table . Some previous research on the two ellipsis questions Is there syntax in the ellipsis site?

Is identity syntactic or semantic?

Yes

Syntactic

Semantic

Both

Sag 1976a, Williams 1977b, Fiengo and May 1994, Chung et al. 1995, Fox 2000, etc. Sag and Hankamer 1984, Merchant 2001, Aelbrecht 2010, van Craenenbroeck 2010b, etc. Kehler 2002, Chung 2013, Merchant 2013c

No

N/A (incoherent)

Keenan 1971, Dalrymple et al. 1991, Hardt 1993, Ginzburg and Sag 2001, Culicover and Jackendoff 2005, etc. N/A (incoherent)

A third major question, which so far has not attracted quite the attention the above two questions have, is the licensing question: ()

What heads or positions or structures allow for ‘ellipsis,’ and what are the locality conditions on the relation between these structures and ellipsis?

The licensing question was traditionally addressed by writing the structural description of a deletion transformation to be sensitive to whatever conditions the theorist thought relevant (for example, in Ross b, Sag a, and Hankamer ). It has been addressed in work that does not assume a transformation of deletion by Zagona (), Lobeck (), Johnson (b), Merchant (), Aelbrecht (), van Craenenbroeck (b), van Craenenbroeck and Lipták (), and Miller (b); all but the last owe a great debt to Lobeck (), whose approach is based on a kind of Empty Category Principle applied to a null pro-like element.

. A     

.................................................................................................................................. The reasons for theoretical interest in elliptical structures are obvious: in each case, the usual form–meaning correspondence appears to break down—there is meaning in ellipsis without form. In broad terms, there have been two answers to the puzzle posed by ellipsis structures: the non-structural and the structural. The non-structural approach responds by

:     



supplementing the theory of meanings, creating or exploiting devices that can generate meanings in the absence of syntactic structure. The structural approach places the burden on the syntax, and claims that the meanings are derived by (ideally all and only) the mechanisms at play in other contexts; it distinguishes itself from the non-structural approach by positing structure which is not pronounced. Within structural approaches, two main lines of investigation can be distinguished: those that posit essentially ordinary syntax, subject to some kind of ‘deletion’ to render the syntax unpronounced, and those that posit a null lexical element which is replaced or identified at some level of representation not relevant to the pronunciation (at LF or in some semantic/pragmatic component). Schematically, these various tacks can be distinguished by their answers to the following questions (see also Winkler and Schwabe  and Stainton b for more detailed taxonomies): ()

Is there unpronounced syntactic structure in ellipsis sites?

no

yes

a. Nonstructural approaches

b. Structural approaches Is there unpronounced syntactic structure in ellipsis sites throughout the entire syntactic derivation?

no

yes

i. LF-copy, null anaphora, ii. PF-‘deletion’, syntactic deletion, non-insertion of Vocabulary Items anaphora to a derivation Recent advocates of non-structural approaches to ellipsis include Ginzburg and Sag () and Culicover and Jackendoff (). Concretely, they propose that e.g. a sluicing example like that in () contains no syntactic material corresponding to the usual clausal source for wh-phrases. Instead, the wh-phrase is the sole daughter of an S node which is the complement to know: ()

John can play something, but I don’t know [S what].

The S node in this account, which has the construction type sluiced interrogative clause, is endowed with featural machinery designed to account for the observed interpretation (among other things). Culicover and Jackendoff ’s approach is similar: for them, the S node is notated ‘IL’ (for indirect licensing) and the wh-phrase is an orphan; the semantics then is constructed with a free variable F whose value is constructed from the context via ‘indirect licensing’.





()

Syntax: [S whatORPH]IL Semantics: Q[F(what)]

Among structural approaches, those that do not implement deletion on the PF side of the derivation either posit null elements in the syntax or deletion of syntactic elements. On theories with null elements, either the null element is a single, designated terminal, as in Hardt () and Lobeck (), or there are a plethora of null elements, as in Wasow () and Ludlow (). These two options assign the structures in (a,b) to examples like () as the representations that feed pronunciation. ()

a. I don’t know [CP what [IP e ]] (Spell-Out) b. I don’t know [CP what4 [IP e1 e2 e3 t4]]

The null elements are either replaced by an operation of structure copying before the structure is interpreted, yielding () (as in Shopen , Wasow , Williams b, Fiengo and May , Chung et al. , Lappin , and Fortin b), or are interpreted by nonsyntactic algorithms for anaphoric elements (as in Hardt  and Merchant a).3 ()

I don’t know [CP what4 [IP John can play t4 ]] (LF/interpreted structure)

Finally, we find the traditional generative solution to ellipsis, in which syntactic structures are subject to non-pronunciation, either as the result of some operation of deletion (which operates either in the syntax before Spell-Out or after Spell-Out in the derivation to PF, as variously in Ross b, Sag a, Hankamer , Lasnik , Saab , Baltin , Kobele a, Merchant b) or as a phonological reflex of prosodic algorithms (in the PF phonology mapping or in the phonology sensu stricto, as in Merchant , Johnson b, Aelbrecht , van Craenenbroeck b, and others). Under such approaches, a sluiced clause appears as follows, where angled brackets enclose ‘deleted’ or, more neutrally, unpronounced material:

CP

()

what1 C

John can play t1

For the most part, the differences between the various implementations seem to be fairly minor, though important for certain architectural deliberations. These differences are taken up in more detail in section .. It is also important to note that several authors have proposed revisiting and maintaining the ‘surface’ vs ‘deep’ anaphoric dichotomy of Hankamer and Sag (), allowing for the possibility that certain ‘surface’ anaphors (including relatives of VP-ellipsis and perhaps 3

Kobele () presents a system where the parser looks for memo-ized antecedents; the grammar, however, generates elided structures with an operation of deletion.

:     



VP-ellipsis itself) may be due to deletion or replacement at PF alongside ‘deep’ anaphors (null or overt), such as those found in Null Complement Anaphora or exophoric fragments (see Sag and Hankamer , Merchant , Baltin , Bentzen et al. , Miller and Pullum , and Merchant c, a).

.. Structural and non-structural approaches compared How does one decide whether some piece of syntactic structure is or isn’t there, particularly when that structure in any case does not lead to any pronounced difference? Indirectly, by necessity. Detecting and arguing for such ‘missing’ structures is analogous to searching for and determining the properties of a black hole: one can tell it’s there only by its effects on surrounding material. The logic of the hunt for elided structure is similar. If one finds effects that seem to be due to missing material, there is an argument that such structure exists. In other words, if effects are found which we would otherwise attribute to properties of structure X in similar, non-elliptical, cases, but structure X is, by hypothesis, internal to the ellipsis site, then X exists. If, on the other hand, expected properties are missing, one could conclude that structure X is absent. Structural approaches are based on what I call connectivity effects; non-structural approaches take their lead from non-connectivity effects. Connectivity effects occur when some part of the clause that contains the ellipsis shows ‘connectivity’ to some other, supposed, unpronounced part; non-connectivity is when this does not occur, despite a prior expectation that it would. In what follows, the main lines of evidence for each approach are presented.

. E    

.................................................................................................................................. There are more than thirteen sets of facts which have been used to argue for unpronounced structure in ellipsis: lower origin effects, locality effects, P-stranding effects, case-matching effects, agreement effects, the distribution of complementizers, of infinitivals, and of predicate answers, binding theoretic effects (Ott ; Ott and de Vries ), the presence of intermediate reconstruction effects in sluicing (Agüero-Bautista ), the facts of ‘spading’ (which shows evidence for an underlying cleft, as van Craenenbroeck b argues), the licensing of parasitic gaps inside ellipses (Yoshida et al. a), the existence of syntactic priming effects (Xiang et al. ), and others. In the remainder of this section, I briefly illustrate the facts from the first eight sets, and refer the reader to the literature just cited for the last five and others.

.. Lower origin effects A variety of elements can be moved out of putative ellipsis sites. However such dependencies are analyzed, they involve an origin site (or tail to a chain, or an equivalent device) inside the ellipsis site. Such moved elements include A′ elements (interrogative phrases, relative pronouns, topicalized phrases, comparative operators, etc.), A-moved elements





(subjects of passives, raising verbs, and unaccusatives), and head elements. Examples of the first are plentiful in sluicing, antecedent-contained deletions, and comparative ellipsis; examples of A-moved elements are equally easy to find (see Aelbrecht ). The following is an example of a topicalized phrase that is across-the-board extracted from an elided VP (from Wallace Stegner’s  novel Crossing to Safety): ()

But De Amicitia I could make a stab at, and could have at any time in the last thirtyfour years.

For head movement, careful work on a series of constructions has shown that certain languages with verb movement also have verb phrase ellipsis, resulting in structures in which the verb is stranded outside the VP (verb-stranding verb phrase ellipsis), or more generally, X-stranding XP-ellipsis, as dubbed by Lipták and Saab (). Representative work includes Huang (b), McCloskey (a), Cyrino and Matos (, ), Goldberg (), Gribanova (a, b), Lopes and Santos (), Thoms (), and Bennett et al. (). The clearest kind of example comes from a language that lacks prodrop, topic drop, or other processes that license null arguments of finite verbs, yet still have verb movement; the best-known such case is that of Irish (example from McCloskey a: ), which is VSO in finite clauses, with the verb moving to a position outside the verbal projection whose specifier hosts subjects: ()

Dúirt mé go gceannóinn é agus cheannaigh. said I that buy..s it and bought ‘I said that I would buy it and I did.’

As McCloskey shows, it is extraordinarily difficult to reconcile such examples with what is known of the grammar of Irish verbs and arguments without positing an ellipsis that targets a verbal projection whose head has been extracted to a position outside the ellipsis site.

.. Locality effects The evidence from locality effects is distributed across a number of domains, but all of it has the same basic form: some kind of locality constraint (typically but not exclusively island constraints) are observed to hold of elements whose putative origin site is inside the understood missing material. If any of these island constraints are due to restrictions on syntactic (broadly speaking) representations, then their presence in elliptical structures argues that those representations must be present.

... VP-ellipsis The first set of locality effects come from VP-ellipsis, where relative operators, wh-phrases, topicalized phrases, parasitic gap operators, and comparative operators all show sensitivity to islands, even when the tail of the dependency is inside an ellipsis site. The examples below are culled from and discussed in and Sag (a), Haïk (), Postal (), Kennedy and Merchant (a), Lasnik (), Merchant (, b), and Fox and Lasnik ().

:      ()



a. *I read every book you introduced me to a guy who did. b. *Abby wants to hire someone who speaks a Balkan language, but I don’t remember which (Balkan language) Ben does . c. *Abby knows five people who have dogs, but cats, she doesn’t . d. *Which film did you refuse to see because Roger was so revolted when he did after renting?4

... Fragment answers Similar effects are found in some fragment answers to implicit salient questions, as discussed in Morgan (), Merchant (a), Arregi (), Temmerman (), Griffiths and Lipták (), Weir (), and Barros et al. () (though see section ... for complications). ()

a. Will each candidate talk about taxes? b. No, about foreign policy. c. No, each candidate will talk about foreign policy.

()

a. Did each candidate2 agree on who will ask him2 about taxes (at tonight’s debate)? b. *No, about foreign policy. c. No, each candidate2 agreed on who will ask him2 about foreign policy (at tonight’s debate).

... Stripping/Bare Argument Ellipsis Examples of ‘stripping’, whose analysis appears to have much in common with that of fragment answers, show a locality effect between the correlate and the ‘bare argument’ (in Reinhart’s  term; see Depiante , Lechner , Merchant , Yoshida et al. b, and Wurmbrand ). ()

a. The man stole the car after midnight, but not the diamonds. b. *They caught the man who’d stolen the car after searching for him, but not the diamonds.

... Gapping Gapping, which is probably just a version of stripping with more than one remnant, unsurprisingly behaves like stripping in this regard as well (Coppock ; Johnson a, ; Winkler ; Toosarvandani a): ()

*Some wanted to hire the woman who worked on Greek, and others Albanian.

4 As Postal () discusses, it is important to have a parasitic gap in this example, as it forces the missing VP to host a wh-trace, yielding the observed island effect. Without the parasitic gap, this example would be acceptable, as an example of ‘vehicle change’ (see Fiengo and May ).





()

*She discussed my question which letters we wrote and he which books. (Winkler :  (b))

... Contrast sluicing Finally, note that even when a sluiced wh-phrase has an explicit correlate, we still find locality effects when the relation between the correlate and the wh-phrase is one of contrast, as originally noted in Merchant (b: ): “Sluicing with indefinite correlates repairs islands, but Sluicing with focused correlates does not.” ()

She knows a guy who has five dogs, but I don’t know how many cats. a. = b. ≠

See also Merchant (), Vicente (), Griffiths and Lipták (), and Barros et al. () (and note that accounting for the full range of facts requires something like MaxElide, as Merchant b:  and Fox and Lasnik :  n.  point out, pace Messick and Thoms ).

.. The P-stranding generalization The third major strand of evidence for structure internal to ellipsis sites comes from the distribution of preposition stranding under wh-movement out of putative ellipsis sites cross-linguistically. Both under sluicing and in fragment answers, there is a strong (if not always perfect, apparently not random or accidental) correlation between languages that allow P-stranding in non-elliptical contexts and in sluicing/fragment answers. If what regulates P-stranding cross-linguistically is some kind of morphosyntactic condition, and not due to differing semantics across languages (an assumption I know of no serious challenge to; see Abels () for insightful discussion), then the fact that this correlation holds in seemingly elliptical contexts is quite telling. Sluicing data from representative languages is given here, reproduced from Merchant (); () represents P-stranding languages (as seen in the (b) controls), while () illustrates non-P-stranding languages. ()

English a. Peter was talking with someone, but I don’t know (with) who(m). b. Who was he talking with?

()

Greek a. I the b. *Pjon who

Anna Anna

milise talked

milise talked.s

me with

me? with

kapjon, someone

alla but

dhe not

ksero I.know

*(me) with

pjon. who

:     



This parallelism is expected on structural approaches, since the grammatical constraints that govern preposition stranding will be operative in these (elliptical) structures as well. The parallelism, however, is far from perfect, and numerous empirical caveats to this generalization have been raised in the recent literature: see section ...

.. Case matching As first pointed out in Ross (b), case-matching effects found in sluicing (and fragment answers, Merchant a, and contrastive left-dislocation, Ott ) are straightforwardly accounted for if the relevant case assigners are syntactically present, though unpronounced. Ross’s particular example comes from German, where schmeicheln ‘flatter’ assigns dative, while loben ‘praise’ assigns accusative: () German a. Er will jemandem schmeicheln, aber sie wissen nicht, { *wer / he wants someone. flatter but they know not who. *wen / wem }. who. who. ‘He wants to flatter someone, but they don’t know who.’ b. Er will jemanden loben, aber sie wissen nicht, { *wer / he wants someone. praise but they know not who. wen / *wem}. who. who. ‘He wants to praise someone, but they don’t know who.’ See Barros (b) for fuller discussion of case matching that appears to hold even in certain cases where we would expect a cleft or other non-isomorphic source.

.. Complementizer deletion If fragment answers involve ellipsis (Morgan ) preceded by movement of the fragment out of an elided clause (Merchant a; Merchant et al. ; see also Thoms ), then the following pattern is accounted for, given that displaced complementizer phrases require overt complementizers. (The well-formed response variant that lacks that initial complementizer in () is not elliptical: it is an indirect answer, not a direct answer; see Merchant et al.  for discussion of this difference.) ()

What does no-one believe? #(That) I’m taller than I really am. a. No-one believes (that) I’m taller than I really am. b. *(That) I’m taller than I really am, no-one believes.





()

What are you ashamed of? (That) I ignored you. a. *I’m ashamed of that I ignored you. b. That I ignored you, I’m ashamed of.

.. Infinitivals: Raising vs control A similar distinction is found in the distribution of short answers using infinitival clauses: only control infinitivals can be clefted, and only control infinitivals can serve as fragment answers (Merchant a). ()

a. *It’s [to get asylum in Europe] that the refugees tend. b. Q: What do the refugees tend to do? A: *To get asylum in Europe.

()

a. It’s [to get asylum in Europe] that the refugees want. b. Q: What do the refugees want to do? A: To get asylum in Europe.

This is expected if the control CP can be fronted, but the raising TP cannot be; see Landau ().

.. Predicate answers Finally, predicate answers show a distribution which is somewhat puzzling if the connection between question and answer is mediated solely by some semantic/pragmatic relation, and not by syntactic structure (Hankamer ; Merchant a). ()

a. A: What did he do for his sister? B: Funded *(her). b. *He did funded her for his sister.

As Hankamer () pointed out using slightly different terms, the minimal fragment answer to a question whose semantics demand only an element of type , for example, cannot be answered with a simple verb (of type ). Instead, the minimal fragment must be a VP (of type ), despite therefore necessarily including redundant, given information (in the form of the pronoun). This pattern is expected if the short answer involves movement of a phrase to a clause-peripheral position: since English lacks long head movement (and also remnant topicalization of VPs, as Müller  discusses), only the entire VP will be available to form the answer. On non-structural approaches that basegenerate words subject only to semantic or pragmatic answerhood conditions, this restriction is mysterious. Equally mysterious for theories that impose category matching between the wh-phrase and the answer (such as Jacobson b) is well-formedness of a tensed VP

:     



funded her where such a VP is ill-formed as the complement of a tensed do, shown in (b). Overall, these facts point to the conclusion that constraints on form—mediated by structure—are active in elliptical constructions.

.. Agreement triggers Targets of agreement can be controlled by elements internal to putative ellipsis sites: this is well documented for predicate ellipses and certain kinds of nominal ellipsis (see Saab ; Merchant a, a; Saab and Lipták ) and can be seen also in the following set of data. Subject–verb agreement in English for number does not always track notional or semantic number: pluralia tantum such as nuptials trigger plural agreement on the verb, unlike its singular synonym wedding: ()

a. Beth’s wedding was in Bond Chapel, and Rachel’s wedding was in Rockefeller Chapel. b. Beth’s nuptials were in Bond Chapel, and Rachel’s nuptials were in Rockefeller Chapel.

Nominal ellipsis preserves the syntactic properties of agreement: ()

a. Beth’s wedding was in Bond Chapel, and Rachel’s was in Rockefeller Chapel. b. Beth’s nuptials were in Bond Chapel, and Rachel’s were in Rockefeller Chapel.

()

a. *Beth’s wedding was in Bond Chapel, and Rachel’s were in Rockefeller Chapel. b. *Beth’s nuptials were in Bond Chapel, and Rachel’s was in Rockefeller Chapel.

Agreement thus appears to be sensitive to unpronounced structure: the grammatical plural feature on the antecedent of the ellipsis, nuptials, must be encoded locally in the NP subject to ensure plural agreement. As Culicover and Jackendoff (:  n. ) put it, the presence of these kinds of connectivity effects would represent “impressive evidence of the reality of the invisible structure” (while reporting that they don’t find consistent island effects in cases like (b), they don’t consider the remaining facts). The conclusion pointed to by the above kinds of data would seem to be that there is (regular but unpronounced) syntactic structure inside ellipsis sites.

. E    

.................................................................................................................................. Some kinds of data, however, seem to point to the opposite conclusion: that there is no structure inside ellipsis sites (at least no structure that has the properties of its putative non-elliptical counterpart).





.. Absence of locality effects The strongest piece of evidence in favor of the non-structural approaches comes from the lack of island effects in certain ellipsis contexts, such as in many sluicing structures, in certain fragment answers, possibly in certain gapping examples, and in certain kinds of comparative ellipsis.

... Sluicing As Ross (b) famously first observed, the putative wh-extraction out of ellipsis sites in sluicing appears insensitive to islands: ()

They want to hire someone who speaks a Balkan language, but I don’t remember which.

()

Every linguist1 argued with a philosopher who took issue with one of his1 claims, but I can’t remember which one of his1 claims. (adapted from Lasnik 2001)

Though this observation holds in the first instance for cases in which the wh-phrase corresponds to an overt indefinite, Culicover and Jackendoff (:  n. ) produce one example with a merely implicit correlate which they judge acceptable (example modeled on one from Chung et al. , where the opposite judgment is reported): ()

Bob found a plumber who fixed the sink, but I’m not sure with what.

Since this relative clause is veridical, this is not a counterexample; such sluices are discussed in chapter  of Merchant (), where they are given a non-island source, with an E-type pronominal subject, equivalent here to . . . but I’m not sure with what he [=that plumber Bob found] fixed the sink. See AnderBois () for additional discussion of the properties of such sluices, and Griffiths and Lipták () for a comprehensive attempt to distinguish apparently island-sensitive from non-island-sensitive sluices, following Merchant (b), as well as Barros et al. ().

... Fragment answers Similar observations have been made for certain fragment answers (in Culicover and Jackendoff : ff.; Stainton b). ()

Is Sviatoslav pro-communist or anti-communist these days? —Pro. [*Pro, Sviatoslav is [t-communist these days.]

()

A: John met a woman who speaks French. B: And Bengali? [*And Bengali, did John meet a woman who speaks French t?]

Interpreting these data requires some care, however. First, sometimes bound prefixes can appear without their hosts, as in (). Second, the interpretation of the fragment in (B) is

:     



equivalent to the paraphrases given in (a–c) (readings which the presumably non-elliptical (d,e) can have as well, in this context), and the fragment in (B) does not appear to have the expected ‘island-violating’ reading given in rough paraphrase by (f). While this set of facts is expected on the structural approach, it is not clear how the non-structural approach rules out the interpretation in (f) for (B). ()

Sviatslav is pro-communist and Derzhinsky is anti-.

()

a. = Did John meet a woman who speaks French and Bengali? b. = Does she speak French and Bengali? c. = And does she speak Bengali (too)? d. = And what about Bengali? e. = And how about Bengali? f. ≠ And did John also meet a different woman who speaks Bengali (in addition to meeting the woman who speaks French)?

Casielles () and Stainton (b) also adduce fragment answer examples out of islands that seem quite acceptable. It is also true, as Progovac et al. (b) point out, that without a comprehensive theory of islands it may be difficult to properly assess the importance of island sensitivities (they suggest, following others, that perhaps some islands are semantic or pragmatic in nature, not syntactic); much more work is needed to ascertain the full empirical lay of the land in this domain as well. See Merchant et al. () for some experimental investigation of fragments.

... Gapping Culicover and Jackendoff (: ) also adduce one example, in (), for which they claim acceptability; to their example I add the attested examples in (). () Robin knows a lot of reasons why dogs are good pets, and Leslie, cats. ()

a. He spoke in the kind of tone a lawyer might use to address a jury, or a serious professor of history his students. (Tom McCarthy, Remainder, New York: Vintage, , p. ) b. If this narrative were a quotidian account of the history of Russia, this chapter would be a proletarian’s account of the Great October Soviet Socialist Revolution of , if a history of France, the beheading of Marie Antoinette, if a chronicle of America, the assassination of Abraham Lincoln by John Wilkes Booth. (Marisha Pessl, Special topics in calamity physics, New York: Vintage, , p. ) c. No, this was the torturous, clammy kind, when one’s pillow slowly takes on the properties of a block of wood and one’s sheets, the air of the Everglades. (Ibid., p. )

... Ellipsis in comparatives Kennedy and Merchant (a) argue that examples like (a) involve a degree phrase extracting from a left-branch (here, attributive) position within a noun phrase, structurally parallel to (b).





()

a. Brio wrote a more interesting novel than Pico did. b. *How interesting did Pico write a _ novel?

Non-structural approaches have a ready explanation for this state of affairs, if one assumes that island effects come about only in movement structures. By parity of reasoning, one could have a structural account which eschews movement in these particular structures, as Chung et al. () and Lobeck () pursue, which derives the same effect. These structures are difficult only for ‘deletion’ approaches that fall under (b.ii) and those null-structure accounts like Wasow’s and Williams’s that posit regular null structures as well.

.. Case mismatches Certain short NP answers display an unexpected case, given a simple equivalence between elliptical and non-elliptical structures. English subject questions can be answered with fragments in the accusative, where such pronouns would be ill-formed in non-elliptical sentences: ()

A: Who wants a slice of pizza? B: Me! (*Me want(s) a slice of pizza).

Such mismatches are discussed in Morgan (), Barton (, ), and Progovac et al. (a). Note that while such mismatches are not found as robustly in sluicing, as Barros (b) in particular discusses, there do seem to be a handful of problematic cases (see the list in Vicente ).

.. Exceptions to the P-stranding generalization Although the P-stranding generalization seemed to hold across a substantial set of data, Merchant () did note some apparent exceptions, such as that in () from Italian, remarking that “[i]n some cases and in some languages, it seems that speakers are willing to accept a bare wh-phrase in place of the PP, though I have not yet determined with sufficient clarity under what conditions this is possible, or whether or not this is a systematic property of a class of prepositions or languages” (Merchant : ). () a. Pietro ha parlato con qualcuno, ma non so Pietro has spoken with someone but not I.know ‘Pietro has spoken with someone, but I don’t know (with) who.’ b. *Chi ha parlato Pietro who has spoken Pietro ‘Who has Pietro spoken with?’

?(con) with

chi. who

con? with

This sub-area has been explored in more detail in recent years, with investigations of data in Serbo-Croatian by Stjepanović (, ), Brazilian Portuguese by Almeida and

:     



Yoshida (), a variety of Romance languages by Vicente () and Rodrigues et al. (), Indonesian by Fortin (b), Polish and others by Szczegelniak () and Nykiel and Sag (), and in several languages by van Craenenbroeck (a), Barros (b), and Barros et al. (). While Nykiel and Sag () take such non-connectivity effects as in () to be straightforward support for a non-structural approach to ellipsis, most of these authors attempt to find some pattern in the putative counterexamples such that these examples are subject to a different analysis (and thus not undermining the structural account). Stjepanović (, ), Vicente (), Rodrigues et al. (), van Craenenbroeck (a), for example, variously point out that there are environments (even in languages like Spanish that otherwise seem to allow P-less wh-phrases in ‘regular’ sluices) which strictly enforce the P-stranding ban—that is, where the preposition becomes obligatory, as expected on a structural account. These environments include the remnants in gapping and pseudogapping, the counterweight to pseudocleft clauses, fronted CPs, and in sluices with else-modification (only the latter illustrated here): ()

Juan ha hablado con una chica rubia, pero no sé *(con) qué chica Juan has spoken with a girl blonde but not I.know with what girl más. other ‘Juan talked to a blonde girl, but I don’t know (with) what other (kind of) girl.’

As these authors point out, it is exactly in these contexts that a copular source for the elided clause—what we may call pseudosluicing5—is unavailable, as seen in (a); for this reason they suggest (in agreement with Szczegelniak  for Polish) that the P-less ‘sluices’ in fact derive from a copular or reduced cleft-like source, as in (b) (where material in angled brackets is elided): () a. *Juan ha hablado con una chica rubia, pero no sé qué chica Juan has spoken with a girl blonde but not I.know what girl más es pro. other is it (lit. ‘Juan talked to a blonde girl, but I don’t know what other (kind of) girl it was.’) b. Juan ha hablado con una chica rubia, pero no sé cual < es pro >. Juan has spoken with a girl blonde but not I.know which is it ‘Juan talked to a blonde girl, but I don’t know which one.’ Positing cleft and copular sources for fragments and sluices has been successful in accounting for other restrictions as well, such as those found with non-intersective adjectives (from Barros b:  and Ueno : , respectively):

5

The term pseudosluicing, as originally used in Merchant () and van Craenenbroeck (b: –), covered complex copular sources that included a cleft or cleft-like clause like it is X that . . . , with independent null subjects and copulas, but we can extend the term to cover ellipsis of simple copular clauses that lack the relative-clause-like part as well, such as simple it is X, irrespective of whether null variants of it and be are available in the language.





()

#She married a heavy drinker, but I don’t know how heavy. (cf. #How heavy was he/the drinker?)

()

He gave me a tall order. #How tall? (cf. #How tall was it?)

One kind of apparent exception to the P-stranding generalization is problematic in particular for the system of Jacobson (b), in which matching is enforced between the category of the correlate and that of the remnant. But in pairs like (), the wh-phrase is a simple NP, while the fragment answer is a PP; crucially, the preposition at here is semantically vacuous. ()

Q: Which kids was Abby most angry at? A: At the ones who keep stealing her mail.

As I wrote in Merchant (), “while a movement-based deletion theory of ellipsis can make use of the usual mechanisms for handling the syntax/semantics mismatches that go under the rubric of reconstruction (here, the preposition at in the answer reconstructs), theories like that of Jacobson () . . . predict that such pairs will be ill-formed” (as Jacobson herself acknowledges). In sum, the data from preposition stranding under sluicing (and fragment answers and elsewhere) are quite complex, and have become the subject of a rich vein of work. Whether or not the data uniformly support a structural analysis, clearly the subject forms an important area of debate. For analysts who pursue non-structural approaches, mismatches (or non-connectivity) are very welcome, since such mismatches would indicate that the constraints on preposition stranding in non-elliptical dependency structures are independent of the constraints on the presence of prepositions in ellipsis. But one last point remains to be made in this respect, one which has not been made in the literature on these questions to my knowledge: if the conditions on P-stranding under non-elliptical wh-movement and those on prepositions in elliptical environments are in fact independent, what prevents us from expecting to find a language like the one below, call it ‘reverse-English’? ()

A possible language if the claim that P-stranding wh-movement is independent from P-less wh-phrases in sluicing were true: a. Who did she talk to? *To whom did she talk? b. She talked to someone, but I don’t know { *who | to whom}.

‘Reverse-English’ would be like child spoken English (and many adult varieties, too) in requiring P-stranding in non-elliptical questions, but like some strict version of German in requiring the presence of the P in sluices in which the correlate of the wh-phrase is governed by a P. If these constraints are truly independent, such a language should strike us as just as natural as the Spanish that allows P-less sluiced wh-phrases. Since the nonstructural analyses make no reference to the conditions on wh-displacement, writing constraints that impose such a requirement should be simple. In total, this kind of evidence seems to favor structural approaches to some kinds of ellipsis. If these approaches are correct, we must search elsewhere for an explanation of the

:     



sometime lack of island effects (see Merchant b, Abels , Temmerman , Bošković , Griffiths and Lipták , and Barros et al.  for some recent approaches), and other non-connectivity effects that are sometimes adduced (such as the so-called ‘vehicle change’ effects of Fiengo and May ; Merchant b). Several proposals have been advanced to account for these effects, which have been discussed extensively in the recent literature; with respect to deciding whether or not structure must be posited internal to the ellipsis site, it seems clear that it is completely unimportant which particular proposal for island and other repair effects is correct. While one can imagine many possible ways to account for repair effects, or the absence of expected grammatical sensitivities (one influential strand of thinking ties them all to properties of the syntax– phonology interface systems), it is essentially impossible to imagine an account of island effects and P-stranding that would make their presence in the elliptical structures seen here accidental or orthogonal to their presence in non-elliptical structures. At present, I see little prospect for building such a theory that would not essentially have to reimport the constraints needed for non-elliptical structure.

. N   ‘’

.................................................................................................................................. Within structural approaches, two tacks can be discerned: the null anaphora approach and ‘deletion’. The latter ranges from the traditional formulation of a deletion transformation (as in Ross b and Hankamer  among many others) to more recent proposals with implementational details more consistent with recent views of syntax and morphology. On all these approaches, the syntax of an ellipsis site is in general just the same as the syntax of its non-elliptical counterpart, but subject to some kind of operation or constraint which results in no pronounced material. In modern incarnations, the difference between an elliptical and non-elliptical XP, for example, is often cast solely in terms of the presence or absence of a feature in the structure which signals to the phonology that the phonological value of the XP is null (Merchant ; van Craenenbroeck and Lipták ; Vicente ; Ha a; Toosarvandani , ; Aelbrecht ; Corver and van Koppen , ; van Craenenbroeck b), or that Vocabulary Insertion does not take place at the level where the morphological structure is computed (Saab ; Saab and Zdrojewski ; Temmerman ; Merchant a). Such a feature—call it the E-feature for ‘ellipsis feature’—should, ideally, be the sole repository of all information about the ellipsis. That is, it should have a syntax, a semantics, and a phonology. The syntax of this feature should serve to delimit what heads or other structures can host it (the ‘licensing’ question), the semantics could be used to impose an identity condition (see section . below: the E-feature is an anaphoric device that introduces a pointer that is resolved by re-using a derivation or its output, or triggering a search for an already constructed derivation or structure—e.g., anaphora to a meaning), and the phonology would be a trigger for a rule or constraint syncopating the phrase’s phonological value (or triggering non-Insertion on all dominated terminal nodes). There are several ways to imagine implementing such a feature, and different versions are pursued by different authors. The main advantage to such accounts is that nothing more need be





said about the syntax, and all connectivity effects follow straightforwardly. For example, the origin site of the displaced wh-phrase in a sluicing example like () is inside the unpronounced sentential node: no additional, sluicing-specific mechanism need be employed to base-generate the phrase in specCP, nor to account for its selectional properties (the fact that angry idiosyncratically selects a PP headed by at): ()

Amy seemed angry, but we didn’t know [pp at who ]1 .

The [E] feature can be added to the feature matrix of the ‘licensing’ head (certain Cs for sluicing, certain Ts for VP-ellipsis, etc.), as in the following structures, or more complex relations between E and the elided material can be entertained (as in van Craenenbroeck and Lipták  and Aelbrecht ). ()

a. Someone murdered Joe, but we don’t know who.

CP

b. who1

C[E]

t1 murdered Joe ()

a. Abby didn’t see Joe, but Ben did.

b.

TP Ben T[E]

did

see Joe

The alternative is to insert an empty placeholder node in the structure, which acts like a null anaphor and which must be replaced at LF by full structure (on LF-copy approaches like Chung et al. ) or otherwise filled in or interpreted. One advantage of this family of approaches is that it assimilates the local licensing conditions on null VPs, TPs, and NPs as they appear in elliptical constructions to the more general licensing conditions on null elements; Lobeck () pursues this line, as does Johnson (b) with a different emphasis. But this advantage may be entirely illusory, since the equivalent deletion approach using a feature can capture these restrictions with arguably the same level of sophistication: we simply place the same conditions on appearance on E that we did on pro. That is, we can call E an anaphoric element subject to the same licensing and discourse constraints on appearance that adhere to pro on theories that employ pro. It is no more nor less explanatory to claim that pro is subject to the Binding Theory by virtue of a featural specification [+pronominal, anaphoric] than it is to say that E is subject to the same constraints by specifying it as [+pron., ana.]: this is just to claim that E is a discourseanaphoric element with local syntactic requirements on its appearance. Lobeck’s attempt to

:     



reduce the distribution of elliptical e entirely to that of pro failed: that analysis was forced to supplement the conditions on pro with an extra codicil specifying that e must appear in a context of ‘strong agreement’, but the definition of strong agreement does not track anything independent: English each but not every must be a ‘strong agreer’. In the end, the ‘strong agreers’ were all and only the heads that licensed ellipsis: obviously, this approach has nothing to recommend it over an equivalent listing of the heads that E can appear on (and the latter approach at least places inter- and intralinguistic variation in the class of head licensers in the most plausible domain: that of the lexicon).

. T    

.................................................................................................................................. The second major question arising in ellipsis concerns the identification of the elliptical material. That is, how does one calculate what the ‘missing’ material means? Clearly ellipsis is anaphoric, broadly speaking, and depends on its context to get its meaning: an ellipsis site apparently has no intrinsic lexical content at all. Theories that derive the attested meanings for ellipses do so on the basis of identity of the ellipsis to some antecedent, or of ‘parallelism’ or ‘resolution’ of the ellipsis, terms that are partially overlapping but generally equivalent for our purposes here. It is generally assumed that ellipsis requires an antecedent (perhaps only implicit or inferred, in the case of exophoric ellipses), on the basis of which the meaning is derived. But what kind of antecedent does ellipsis need? And what is the relation that must hold between an ellipsis and its antecedent? There are broadly three kinds of answers to these questions: ones that posit that the relation between the ellipsis and its antecedent involves a kind of identity of (or anaphora to) meaning, ones that posit a kind of identity of structure, and ones that use a bit of both. This terminology is meant to be neutral between theories that take the relation between an ellipsis site and its antecedent to be one of anaphoric ‘resolution’, recoverability, ‘parallelism’, or identity: for most purposes, these terms are picking out the same relation. Clearly what is not at stake is anything like surface identity, given examples like the following: ()

a. Jake ate the sandwich even though his friend told him not to. b. Jake ate the sandwich even though his friend told him not to eat the sandwich.

()

‘In the meantime, enjoy the ride.’ ‘I am.’ (John Updike, Terrorist, New York: Ballantine, , p. )

()

A: Pires tin tsanda mazi su? took.s the bag. with you ‘Did you take the bag with you?’ B: Yes, I did.

[Greek]





Since the bare form of the verb following to in (b) is not surface-identical to the past form of the verb in the antecedent VP, any identity relation that elided such a verb phrase in (a) based on morphological or phonological identity with its antecedent would be clearly wrong. Likewise for the imperative and the progressive participle, as the pair in () shows. Finally, this point can be seen in an even more striking way when we consider ellipsis licensed across speakers using different languages, as in (), which reports a conversation between two bilingual speakers of English and Greek; we must assume that the English ellipsis in B’s response in () is sensitive not to the overt form of the antecedent in Greek, but rather to more abstract properties not immediately obvious in the ‘surface’ form of the Greek (see Merchant b).

.. Semantic identity and information structure Although the vast majority of the generative research on ellipsis in the years from  to the mid s (e.g. Chomsky , Ross b, Hankamer and Sag , Sag a, Williams b, Hankamer , Chao , Lappin , Rooth b, Fiengo and May , Chung et al. , Lappin , and many others) worked with the assumption that the identity relation was to be stated over phrase markers (whether D-structure, deep structure, LF, or something else—often, it should be noted, faute de mieux), since the early s ever more proposals have been made that state the identity relation over semantic representations or which take the resolution of ellipsis to be essentially semantic (Dalrymple et al. , Jacobson a, Hardt , , Gardent et al. , Kempson et al. , Asher et al. , Ginzburg and Sag , Merchant , Hendriks , Hendriks and Spenader , van Craenenbroeck b, Yoshida , and many others; perhaps the earliest analysis in this vein is Keenan , with Sag and Hankamer  an important precursor as well). Sometimes the proponents of semantic approaches base their choice on the ability of these approaches to deal more directly with scopal interactions in ellipsis, and the distribution of strict and sloppy readings of pronouns (Dalrymple et al.  is one such example). But such interactions are not necessarily a direct argument for a semantic identity relation, despite first appearances. First, these effects have been dealt with in syntactic identity approaches as well, sometimes with greater empirical success (see Fox , for example). Second, taking such phenomena as arguing for or against any version of an identity condition on ellipsis is misguided. Tancredi () showed conclusively that the problem of delimiting a number of phenomena traditionally thought to belong solely to the domain of ellipsis in fact formed merely a subpart of the problem of structuring discourse coherently, in particular with respect to focus and deaccenting. Thus traditional concerns of strict vs sloppy identity, Tancredi showed, could not be addressed merely by looking at elliptical structures, but had to be approached from deaccented structures (similarly for scopal parallelism effects, the Dahl (many pronouns) puzzles, and the many clauses puzzles; see Fiengo and May ). Tancredi’s great contribution, and one whose impact is sadly often underestimated, was to show that theories of ellipsis per se did not have to deal with these

:     



phenomena at all, and that any theory of, say, the distribution of strict/sloppy readings that made reference to ellipsis was mistaken.6 Instead, the best arguments for semantic identity theories come from a large set of mismatches between the syntactic structure of the antecedent and that of the purported elided phrase. Some of these were the focus of Dalrymple () and were enumerated at greater length by Fiengo and May (), who dubbed them ‘vehicle change’ effects. Although Fiengo and May () use the term ‘vehicle change’ for about a dozen phenomena, I will illustrate only two here: pronoun/name equivalences, and polarity item/nonpolarity item equivalences. Important additional recent mismatch discoveries include Malagasy voice mismatches (Potsdam ), spading (van Craenenbroeck b) and various pseudosluice phenomena (Barros b), category switches (Fu et al. ; Johnson b; Merchant b), sprouted implicit arguments and adjuncts (Chung et al. , ; Merchant ; AnderBois ), missing expressives (Potts et al. ), as well as the other phenomena discussed in chapter  of Merchant () involving finiteness mismatches, word order, clitics, and other issues. Pronoun/name ‘vehicle change’ is illustrated by the following data (I illustrate only with names, though the problem is fully general and extends to all R-expressions); although sluicing and VP-ellipsis are licit in (), the indicated coreference between the pronoun and c-commanded name in the presumptive non-elliptical equivalents in () is ruled out. ()

a. They arrested Alex3, though he3 didn’t know why. b. They arrested Alex3, though he3 thought they wouldn’t.

()

a. *He3 didn’t know why they arrested Alex3. b. *He3 thought they wouldn’t arrest Alex3.

A similar mismatch between grammatical ellipses and their ungrammatical putative nonelliptical counterparts is found with polarity items, as noted in Sag (a: –) (and discussed in Merchant b): ()

John didn’t see anyone, but Mary did. a. . . . but Mary did see someone. b. . . . *but Mary did see anyone. c. ∃x.see(Mary, x)

()

John saw someone, but Mary didn’t. a. ≠ . . . but Mary didn’t see someone. b. . . . but Mary didn’t see anyone. c. ¬∃x.see(Mary, x)

6

And note that a semantic identity theory need not accept the claim that there is no unpronounced syntactic structure: it’s perfectly consistent to claim that while ellipsis sites have syntactic structures, the fact that they are unpronounced is due to a semantic/pragmatic requirement being satisfied. The structure question and the identity question are partially independent; see Table ..





While the semantics of names and pronouns on the one hand, and polarity and nonpolarity indefinites on the other, can reasonably be construed as equivalent (under a single assignment function g, if [ he3] g = Alex, then any proposition containing he3 evaluated with respect to g will have the same truth conditions as that proposition where Alex replaces he3; likewise for the basic semantic contributions of polarity items), but it is difficult to see how he and Alex could be syntactically equivalent.

.. Syntactic identity Despite the success of semantic theories of elliptical identity, there are several sets of data that seem to require some syntactic identity. The first set of evidence I will mention here comes from the uneven distribution of voice mismatch effects in ‘big’ vs ‘small’ ellipses, and the second from certain morphological facts; see also Chung () for an important set of facts from English and Chamorro sprouting.

... Voice mismatch under ellipsis In ‘big’/high ellipses—viz., sluicing, fragment answers, gapping, and stripping—elided material and antecedent phrase must match in voice: if the antecedent clause is in the passive, then the elided clause must also be in the passive, and likewise for the active, mutatis mutandis. This is illustrated for sluicing in () (see Merchant d for the other ellipsis types and data from additional languages). ()

Sluicing a. passive antecedent, active ellipsis: *Joe was murdered, but we don’t know who . b. active antecedent, passive ellipsis: *Someone murdered Joe, but we don’t know who by .

In contrast to big ellipses like sluicing, ‘low’ or little ellipses allow voice mismatches: the relevant ellipsis type is VP-ellipsis in English (see Merchant b for discussion of pseudogapping, which I omit here). The first attested example is from Hardt (); for further examples and discussion see Sag (a), Dalrymple et al. (), Fiengo and May (), Johnson (b), Kehler (), Arregui et al. (), Frazier (), Kim et al. (), San Pietro et al. (), and Merchant (d). ()

VP-ellipsis a. passive antecedent, active ellipsis: This problem was to have been looked into, but obviously nobody did . b. active antecedent, passive ellipsis: The janitor should remove the trash whenever it is apparent that it needs to be .

:     



The uneven distribution of these voice-matching effects does not seem to be arbitrary (in other words, it would be unexpected to find a language showing the reverse pattern of English), and can be fairly straightforwardly understood given recent proposals for the syntax of voice following Rivero () and Kratzer () which separate a Voice head from the rest of the VP. This separation allows for the differentiated targeting of nodes for ellipsis: in high ellipses (sluicing, etc.), a clausal node that necessarily includes Voice; in low ellipses (VP-ellipsis), the verbal projection that is complement to (or inside the complement of) Voice. The structure for a representative example is given in (). If the elided phrase XPE and its antecedent YPA must be identical, it’s obvious why ellipsis fails in the sluicing case in (69) (since TP deletion includes the Voice head and therefore TPA ≠ TPE) but succeeds with the articulated syntax in an ellipsis targeting a verbal projection (vP or VP) dominated by VoiceP (since the ellipsis excludes the Voice head, and so VPA = VPE). () a. *Joe was murdered (by someone), but we don’t know who.

b.

CP who1

C

< TP > t1 VoiceP

T

Voice:Active

vP murder Joe

It’s far less clear how current semantic identity proposals would handle this uneven distribution: most of them are designed to allow active/passive mismatches (such as Dalrymple et al.  and Hardt ) and consider only VP-ellipsis data. Once the sluicing data are also brought into the picture, a uniform semantic analysis becomes harder to support.

... Auxiliary form matching A second argument for syntactic identity in ellipsis comes from the exceptional behavior of be under ellipsis (Warner ; Lasnik a; Potsdam a; Roberts b; Merchant b). In general, verbs (both regular and irregular) don’t require morphological identity: ()

a. Emily played beautifully at the recital and her sister will , too. b. Emily took a break from her studies, and her sister will , too. c. Emily sang the song {because|the way} she wanted to . d. Emily went to the park because she wanted to .





But forms of be do require morphological identity: ()

a. Emily will be (beautiful) at the recital, and her sister will , too. b. *Emily was beautiful at the recital and her sister will, too. c. Emily will be elected to Congress just like her sister was. d. *Emily was elected to Congress {because|just like} she really wanted to.

Lasnik (a) accounts for this distribution by positing that forms of be are inserted into the derivation fully inflected, while other verbs get their inflection in the course of the derivation. The syntactic identity therefore is met before the inflection of most verbs, but can never be met for differing forms of be, since they differ at every level of representation. This explanation, appealing as it may be, fails to account for the fact, noted in Merchant (b), that the identical effect appears when the antecedent is in Greek (or Spanish, or German); that paper proposes that the examples are ruled out by a non-parallel representation of the binding of the tense variables on be and on the predicate, precisely as longdistance and mixed bound readings are ruled out in the Dahl puzzle cases discussed in Fiengo and May () and Fox ().

.. Hybrid theories There is a large amount of data that any theory of ellipsis needs to account for. Some of that data seem more amenable to a semantic treatment, and some to a syntactic one. For this reason, some researchers have proposed hybrid theories that incorporate both semantic and syntactic identity conditions, but impose them under differing conditions or selectively. Examples of such proposals include Kehler () (though see Frazier and Clifton  for critical discussion), Chung (, ), van Craenenbroeck (a), Chung et al. (), and Merchant (d). Some of these seek to embed a very narrow amount of syntactic sensitivity (perhaps just the case-assigning heads, or the Voice heads, or the heads that determine argument structure) into a broader semantic theory, others to make different elliptical constructions sensitive to different conditions. The proposal of Chung (), for example, includes the following conditions: ()

The E feature imposes a. e-ness: (or some other semantic equivalence or anaphoricity) [ E]] = λp : e-(p).p, where an expression ε is e- iff ε has a salient antecedent A such that [ A]] = F-clo(ε) and [ ε]] = F-clo(A), and b. Limited syntactic identity: i. Argument structure condition: If an extracted phrase is the argument of a predicate in the ellipsis site, that predicate must have an argument structure identical to that of the corresponding predicate in the antecedent clause; and ii. Case condition: If an extracted phrase is a DP, it must be Case-licensed in the ellipsis site by a head identical to the corresponding head in the antecedent clause.

:     



All of the new crop of hybrid theories promise to contribute to our growing understanding of what is meant by recoverability, and of how far the grammar can go in accommodating antecedents that fail to share local grammatical properties with their elliptical counterparts without simply letting any and every possible conceivable linguistic expression serve as a source for the computation of the anaphoric identity or resolution found in ellipses.

. C

.................................................................................................................................. Ellipsis continues to fascinate because its analysis goes directly to the heart of the main reason we study syntax: to discern the nature of the form/meaning correspondence. Theorizing in this domain requires one to tackle questions of basic ontology, and to make decisions about the nature of arguments for linguistic representations. Much work on ellipsis has taken it for granted that elliptical structures (and the way we derive their meanings) should be parallel to non-elliptical structures, and that theorizing about the two should be uniform. Indeed, this imperative underlies much work in theoretical linguistics more generally, and has been named the ‘structural uniformity’ assumption: Structural Uniformity An apparently defective or misordered structure is regular in underlying structure and becomes distorted in the course of derivation. (Culicover and Jackendoff : )

Culicover and Jackendoff () take this assumption to task and argue that in the domain of ellipsis in particular, it leads to unnecessary positing of unpronounced structures. One may respond that their own proposals, which eschew any kind of unpronounced structure at all, are an instance of a different kind of uniformity assumption: Analytical Uniformity If a certain kind of meaning or use can be made in the absence of syntactic guides to that meaning or use, then syntactic guides are never needed for computing that meaning or use.7

It is clear that, given the richness of the empirical database in ellipsis, the complexity of the analytical problems to which ellipsis gives rise, and the nature of the conclusions one can and must draw from the analysis of ellipsis, theorizing in this domain would do well to beware the ‘uniforms’ of any stripe.

A Great thanks to the excellent reviewers, and to the editors for their encouragement and patience, and to Jeroen van Craenenbroeck for many years of productive discussion of these issues.

7

Alternatively, if some device D can relate a form F and meaning M, then whenever we have M, D is being used.

  ......................................................................................................................

          ......................................................................................................................

    

. I

.................................................................................................................................. E has played a role in transformational generative grammar (TGG) since the inception of the approach in the s. In the earliest model (Chomsky ) the vast majority of descriptive work was done by syntactic transformations. Phrase structure rules created basic monoclausal structures (often represented as trees), then two kinds of transformations dealt with these trees: Generalized transformations combined separate trees into one (sometimes altering them in the process); and singulary transformations operated on single trees, both basic ones and those created via generalized transformations. These singulary transformations inserted material, rearranged material, or deleted material, this latter kind of operation—labeled ‘deformation’ by Chomsky—being responsible for ellipsis. There have been many changes over the decades in the technology of TGG but the core ideas survive: Sentences have a basic underlying structure (or set thereof) and a transformationally derived ‘surface’ structure more immediately related to phonological form. Questions about ellipsis have persisted through the various theoretical changes. ‘Recoverability’ is a long-standing theme. In classical TGG, the question concerned the precise nature of the identity requirement governing deletion, since it was known that total identity was not required. Chomsky () and Ross (b) especially contributed to this discussion. In slightly later work, it was often assumed that an ellipsis site is empty in initial structure, and then the antecedent is copied in (the inverse of a deletion transformation) to create a fully structured Logical Form (LF) representation. The recoverability question then concerns the extent to which the antecedent can be altered in the course of being copied. Williams (b) and Fiengo and May () were influential contributors to this perspective. The recoverability issue obviously intimately relates to the question of the exact nature of the ellipsis site. In the earliest work, it was simply assumed that the ellipsis site is syntactically fully represented, prior to deletion. But by the late s, two alternatives began to be

   



pursued, one involving syntactic structure (exactly how much varied), but no lexical items, in the ellipsis site, the other having no structure whatsoever (a possibility resurrected by Culicover and Jackendoff ). Ross (b) already presented strong arguments against this second possibility and at least some versions of the first. In addition to recoverability, ‘licensing’ is a requirement for ellipsis. Not just any XP can be elliptical (even in the presence of a suitable antecedent), and even an XP that can be elliptical can usually only be so in specific licensing contexts. This factor, unexplored in early work on ellipsis, became a significant research topic beginning in the s, especially under the influence of Zagona (, and much subsequent work), Lobeck (), and Saito and Murasugi (a). Licensing still remains rather mysterious. In this chapter we will examine these three themes of structure, recoverability, and licensing as they developed in transformational grammar.

. S

.................................................................................................................................. Analyses of ellipsis can be classified into two fundamental approaches: (i) ones that assume that there is full internal structure in the ellipsis site at certain levels; (ii) ones that assume that there is neither internal structure nor an empty category in the ellipsis site. Let us call (i) the structural analysis and (ii) the non-structural analysis. In order to illustrate these analyses, let us consider a sluicing example like (). ()

Ralph is going to invite somebody from Kankakee to the party, but they don’t know who. (Ross b: )

Sluicing was first explored by Ross (b), and has been discussed by van Riemsdijk (), Chao (), Lobeck (, ), Chung et al. (), Romero (), Ginzburg and Sag (), Lasnik (), Merchant (), and Culicover and Jackendoff (), among others. In the second conjunct in (), the embedded wh-question clause (the sluiced clause) is missing except for the wh-phrase who. Under the structural analysis, the missing part of the sluiced clause is fully represented at certain levels. Thus, the embedding verb know takes CP as its complement, as illustrated in (a). On the other hand, under the non-structural analysis, there is nothing in the missing part, and know directly selects the wh-phrase, as illustrated in (b). ()

a. . . . they don’t know [CP who [IP Ralph is going to invite twho from Kankaee to the party]]. b. . . . they don’t know [DP who].

In this section, we will review arguments in favor of the structural analysis over the nonstructural analysis, and then we will compare two variants of the structural analysis (i.e., the deletion analysis and the LF-copying analysis). We will see that there are arguments for the deletion analysis, and the potential problems for the deletion analysis can be solved under specific analyses of sluicing.



  

.. Arguments for the structural analysis ... Case-matching effects The first argument for the existence of syntactic structure in the ellipsis site is based on case-matching effects in German sluicing, discussed by Ross (b), who credits George Williams, and explored in great detail by Merchant (). Consider the following examples: ()

Er will jemandem schmeicheln, aber sie wissen nicht he wants someone. flatter but they know not *wen/wem. *whom./whom. ‘He wants to flatter somebody, but they don’t know who.’ (German, Ross b: )

() Er will jemanden loben, aber sie wissen nicht *wem/wen. he wants someone. praise but they know not *who./who. ‘He wants to praise someone, but they don’t know who.’ (German, Ross b: ) The verb schmeicheln ‘flatter’ assigns dative case while loben ‘praise’ assigns accusative case. () and () show that the wh-phrase in the sluiced clause (the remnant wh-phrase) must be assigned the same case that the verb in the antecedent clause assigns. This fact can easily be accounted for if there is syntactic structure in the ellipsis site in sluiced clauses and an identity requirement obtains (see section .), as the elided site then contains the same verb as the antecedent clause. On the other hand, if there is no structure in the ellipsis site, it is not obvious how to account for the matching requirement.

... Agreement Another argument, also due to Ross (b), comes from subject–verb agreement. The verb that takes a sluiced clause as its subject exhibits singular agreement irrespective of the number marking of the remnant wh-phrase itself, as illustrated by (). ()

a. He’s going to give us one old problem for the test, but which problem isn’t/*aren’t clear. b. He’s going to give us some old problems for the test, but which problems isn’t/ *aren’t clear. (Ross b: –; slightly modified)

The remnant wh-phrase in (a) is singular while that in (b) is plural. In spite of this difference, the verb must exhibit singular agreement in both clauses. Such obligatory singular agreement is a general property of a clausal subject, as shown by (). () a. b. c. d.

[That Bill left] is/*are tragic. [Why he did it] is/*are a puzzle. [Being drunk] gives/*give me a kick. (Ross b: ) [That the position will be funded and that Mary will be hired] now seems/*seem likely. (McCloskey b: , slightly modified)

   



Example () shows that a clausal subject requires singular agreement on the verb (though there are some exceptions, as discussed by McCloskey b). The similarity between () and () can be easily accounted for if there is syntactic structure in the ellipsis site.

... Subcategorization The verb wonder takes an interrogative clause as its complement while it cannot take a DP, as shown in (). () a. I wonder [CP how many men she’s inviting]. b. *I wonder [DP those old men].

(Ross b: )

Given this, Ross (b) argues that () indicates that there is syntactic structure in the ellipsis site in a sluiced clause. (Ross b: )

() She says she’s inviting some men—I wonder how many men?

In the sluiced clause in (), wonder is followed by a DP, how many men. If there is no structure in the ellipsis site and wonder takes the DP as its complement in (), the acceptability of () is a mystery. On the other hand, if there is structure in the ellipsis site, and wonder in () takes an (unpronounced) interrogative CP complement, the acceptability of () is accounted for.

... Preposition stranding Ross (b) observes that the possibility of P(reposition)-stranding in the sluiced clause correlates with that in non-ellipsis clauses. () shows that P-stranding is obligatory when the wh-phrase is the complement of with in a do-away-with construction.1 () a. Who are you going to do away [PP with twho]? b. *[PP With whom] are you going to do away tPP?

(Ross b: )

When the antecedent clause is the do-away-with construction, the remnant wh-phrase cannot appear along with the preposition with, as () shows. ()

a. Bill’s planning on doing away with one of his inlaws, but I don’t know [which]. b. *Bill’s planning on doing away with one of his inlaws, but I don’t know [with which]. (Ross b: )

Under the structural analysis, the contrast between (a) and (b) and that between (a) and (b) can be treated in the same way since the sluiced clause in (a) involves P-stranding while that in (b) does not under the structural analysis, as illustrated by the following:

1

See Ross (: –) for discussion of this phenomenon.



  

()

a. I don’t know which [IP Bill’s planning on doing away [PP with twhich]]. b. *I don’t know [PP with which] [IP Bill’s planning on doing away tPP].

Merchant () examines cross-linguistic data concerning the correlation between the possibility of P-stranding in sluicing and that in non-ellipsis contexts, and concludes that P-stranding is allowed under sluicing only if P-stranding is allowed in regular wh-movement (in non-ellipsis contexts).2,3 Merchant (: –) shows that English, Frisian, Swedish, Norwegian, Danish, and Icelandic allow P-stranding both in sluicing and in regular wh-movement while Greek, German, Dutch, Yiddish, Russian, Polish, Czech, Bulgarian, Serbo-Croatian, Slovene, Persian, Catalan, Spanish, French, and Italian do not allow it in either context. This cross-linguistic P-stranding correlation can also easily be accounted for under the structural analysis.

... Idiom reconstruction Rottman and Yoshida () provide further evidence for the structural analysis, based on idiom chunks. A part of an idiom chunk can undergo wh-movement, as shown in the following: ()

a. John pulled strings to get his position. b. Which stringsi did John pull ti to get his position? (Rottman and Yoshida : )

In (b), which strings, which is a part of the idiom chunk (pull strings), undergoes wh-movement. The sentence retains its idiomatic meaning. Rottman and Yoshida () observe that a part of an idiom chunk can be a remnant of sluicing, as shown in the following: ()

John pulled strings to get his position, but I don’t know which strings. (Rottman and Yoshida : )

The sluiced clause in () retains its idiomatic meaning. If there is no structure in the ellipsis site, it is a mystery how it yields the idiomatic meaning.4 On the other hand, under the structural analysis, the sluiced clause in () is directly analogous to (b): the mechanism via which the idiomatic meaning obtains in () is reduced to whatever mechanism induces the idiomatic meaning in ordinary wh-question sentences with overt structures like (b).

2

However, potential counterexamples to this generalization have been reported in the literature (Almeida and Yoshida ; Sato b). Although Stjepanović (), Vicente (), and Rodrigues et al. () also discuss counterexamples, they conclude that the relevant data are in fact consistent with the generalization, once alternative sources, especially cleft-like sentences, are considered. See van Craenenbroeck (a) for extensive relevant discussion. 3 Although Merchant (: ) presents the generalization as a necessary and sufficient condition, it should be just a necessary condition because a language might allow wh-movement to strand a P while ellipsis of IP is prohibited for independent reasons. 4 As a reviewer points out, this argument presupposes that idiomatic readings require the relevant parts to form a constituent at some point in the derivation.

   



... Parasitic gaps Yoshida et al. (a) provide a new argument for the structural analysis of sluicing. Their argument is based on the parasitic gap construction, which is exemplified by () (Chomsky , ; Engdahl ; Culicover and Postal ; Nunes ). ()

[Which articles]i did John file ti without reading __PG?

(Engdahl : )

This sentence involves two gaps. The first gap (ti) is called a true gap since it is in a position where extraction is normally permitted. The true gap is basically a trace left by overt A0 -movement. The second gap (__PG) is called a parasitic gap. A parasitic gap is in a position where extraction is normally not allowed (e.g., adjunct islands and subject islands). A parasitic gap needs to be licensed by a true gap in its local domain. Thus, the sentence becomes unacceptable when there is no true gap, as shown in the following: () *[Which article]i did John file this manuscript without reading __PG? The sentence also becomes unacceptable when there is a true gap but the true gap is not in a local configuration with the parasitic gap, as illustrated by the following example: ()

*Whoi did John assure ti that Mary would be surprised soon after talking to __PG? (Yoshida et al. a: )

Yoshida et al. (a) observe that a parasitic gap can appear in a sluiced clause, as shown by the following example:5 ()

The editor told me [which book]i I must review ti soon after receiving __PG, but I don’t remember [exactly how soon after receiving __PG]. (Yoshida et al. a: )

In this sentence, the antecedent clause contains a parasitic gap, which is licensed by the local true gap. What is crucial in this example is that the sluiced clause also involves a parasitic gap, which is contained in the remnant wh-phrase. Yoshida et al. (a) argue that the parasitic gap in the sluiced clause is licensed by a gap inside the syntactic structure that undergoes ellipsis. If there is no structure in the ellipsis site, it is unclear how the parasitic gap in the ellipsis clause is licensed.

5

(i)

Interestingly, the non-ellipsis counterpart of this example is unacceptable: *The editor told me [which book]i I must review ti soon after receiving __PG, but I don’t remember [exactly how soon after receiving __PG]j the editor told me which book I must review tj. (Yoshida et al. a: )

Yoshida et al. (a) argue that this indicates that sluicing can repair some island violations.



  

One might argue that the relevant parasitic gap is licensed by the true gap not in the elided structure but in the antecedent clause. However, the following example indicates that this is not the case: ()

*The editor told me [which book]i I must review ti soon after receiving__PG, but I don’t remember [exactly how soon after receiving__PG] I must review it. (Yoshida et al. a: )

() does not involve sluicing. The gap in question that is contained in the wh-phrase is followed by overt material. If it were the true gap in the antecedent clause that licenses the parasitic gap in the sluiced clause in (), () should be acceptable. To sum up, in order to license the parasitic gap in the sluiced clause in (), we need to assume that there is a true gap in the sluiced clause, which means that there is structure in the ellipsis site in sluicing.

... Extraction out of VPE site Williams (b) claimed that wh-movement out of an elided VP is generally impossible, giving the following example: ()

(*)John, who Bill saw, and who Bob did too

However, in our experience, native speakers find this example to be fine. Many parallel examples appear in the literature, almost invariably labeled grammatical. Fiengo and May () give the following: ()

I know which book Max read and which book Oscar didn’t.

()

Philby, who Angleton suspected, but who Dulles didn’t.

()

What John knows is minimal, and what he doesn’t is vast.

If there is internal structure in the elided VP, the source of the wh-phrase is straightforward, as is the existence of a trace for it to bind.

.. Deletion vs LF-copying We have seen that there is substantial evidence for the structural analysis over the nonstructural analysis. In this section, we will look at two variants of the structural analysis: the deletion analysis (Ross b; Lasnik ; Merchant ) and the LF-copying analysis (Williams b; Chao ; Lobeck , ; Fiengo and May ). Both analyses assume that there is syntactic structure in the ellipsis site. They have different views regarding at which level of derivation structure appears. Under the deletion analysis, there is structure in the elided site at the beginning of the derivation, but it is deleted at PF. On the other hand, under the LF-copying analysis, a null pronominal-like element occupies the ellipsis site at first, and this null element is replaced at LF by syntactic

   



structure, which is copied from some appropriate linguistic antecedent (see section . for what kind of linguistic antecedent is appropriate for this purpose). In section ..., we will see arguments in favor of the deletion analysis. In section ..., we first review two phenomena that are potentially problematic for the deletion analysis, and then we discuss possible analyses of them that are compatible with the deletion analysis.

... Arguments for the deletion analysis As Ross (b) and Merchant () contend, the case-matching effects that we presented in section ... as an argument for the structural analysis also tend to argue for the deletion analysis of sluicing. Under the deletion analysis, the deletion operation applies to IP after the remnant wh-phrase undergoes wh-movement.6,7 Therefore, the case that is assigned to the remnant wh-phrase is naturally maintained after the wh-phrase moves. On the other hand, under the LF-copying analysis, where the ellipsis site is empty throughout overt syntax, the remnant wh-phrase is base-generated in specCP. Thus, some extra mechanism is required to account for the case-matching effects. The P-stranding generalization also favors the deletion analysis over the LF-copying analysis, since it straightforwardly follows from the deletion (plus movement) analysis while it does not from the LF-copying (plus base generation) analysis.

... Potential arguments against the deletion analysis However, there are potential problems for the deletion analysis. In this section, we first review two of them, and then we discuss possible approaches to them that are compatible with the deletion analysis (section ...). Island violation repair by ellipsis. Islands effects are not observed (or at least are significantly lessened) under sluicing in various circumstances. For example, Ross (b) observes that sluicing is not as sensitive to Complex NP islands as regular wh-movement is, as shown in (). ()

a. *She kissed a man who bit one of my friends, but Tom doesn’t realize [which one of my friends]i she kissed [a man who bit ti]. b. ?She kissed a man who bit one of my friends, but Tom doesn’t realize which one of my friends. (Ross b: )

As Culicover and Jackendoff () argue, this is problematic for the deletion (plus movement) analysis. Under the deletion analysis, the ellipsis clause in (b) involves wh-movement out of a complex NP island, as illustrated by (). ()

6

. . . [CP which one of my friends [IP she kissed [Island a man who bit tWh ]]]

However, Kimura () proposes a deletion analysis of sluicing in which the remnant wh-phrase does not undergo movement. 7 As discussed by van Craenenbroeck and Lipták (, , ) and van Craenenbroeck (a), sluicing could be a deletion operation that applies to a CP projection rather than IP given that CP constitutes several separate projections (see Rizzi ).



  

Thus, under the deletion analysis, it might be predicted that (b) is as unacceptable as (a), since the derivation involves an island violation. On the other hand, the LF-copying analysis correctly, or nearly correctly, predicts that (b) is acceptable since the remnant wh-phrase is base-generated in specCP (i.e., wh-movement is not involved). As Ross (b), Chung et al. (), Lasnik (), and Merchant () discuss, this island-repairby-ellipsis phenomenon is widespread, applying to several other islands such as wh-islands, left-branch islands, derived position islands, and coordinate structure islands.8 We will discuss a possible explanation of island violation repair phenomena under the deletion analysis in section .... Multiple sluicing. Nishigauchi (), Richards (), and Lasnik () discuss another kind of repair by ellipsis. Unlike languages like Bulgarian and Serbo-Croatian, English does not allow multiple wh-fronting: ()

I know that in each instance one of the girls got something from one of the boys. *But they didn’t tell me [which] [from which] got something. (Lasnik : )

In the second sentence in (), the two wh-phrases (which and from which) are fronted, yielding the unacceptability. Surprisingly, the multiple wh-fronting sentence becomes acceptable once sluicing applies to it, as shown in (). () I know that in each instance one of the girls got something from one of the boys. But they didn’t tell me [which] [from which]. (Nishigauchi ; see also Bolinger ) Nishigauchi () suggests that sentences like () are similar to gapping, proposing that the first wh-phrase is in specCP while the second one is in some other position. However, Richards () points out that multiple sluicing in English exhibits different properties from gapping. For example, in gapping constructions, the gapped clause must be conjoined with the corresponding non-gapped clause, in fact IP with IP, as illustrated by the following examples: ()

a. [Mary talked about Syntactic structures] and [John about ‘Conditions on transformations.’] b. I think that [Mary talked about Syntactic structures] and [John about ‘Conditions on transformations.’] c. *[Mary talked about Syntactic structures] and I think [John about ‘Conditions on transformations.’] d. *I think that [Mary talked about Syntactic structures] and that [John about ‘Conditions on transformations.’] (Lasnik : )

8

However, Barros et al. () argue that island repair is an illusion since sluicing becomes islandsensitive once we properly control for non-island construals for ellipsis sites.

   



In these sentences, the second bracketed clauses are the gapped clauses and the first ones are the corresponding non-gapped clauses. In (a) and (b), the gapped IPs are conjoined with the corresponding non-gapped IPs, yielding the acceptable gapped clauses. On the other hand, in (c) and (d), the two IPs are not conjoined, resulting in illegitimate gapped clauses. As () shows, multiple sluicing is not subject to such a strict locality condition. ()

In each instance, [one of the students talked about one of Chomsky’s works], but I don’t know exactly [which about which]. (Lasnik : )

This indicates that multiple sluicing cannot be reduced to gapping. If a sentence like () is a genuine instance of multiple sluicing, as Merchant () and Richards () argue, this poses a potential problem for the deletion (plus movement) analysis of sluicing; it is a mystery why multiple wh-fronting is allowed in English only under sluicing. We will discuss a possible account for multiple sluicing that is compatible with the deletion analysis in section ....

... Explanations under the deletion analysis Explanations of island repair.9 Ross (b: ) takes island violation repair as evidence that “the theoretical power of [global] derivational constraints is needed in linguistic theory”; an ungrammatical sentence obtains if a syntactic island is crossed by movement, and the island appears in surface structure while a lesser degree of the ungrammaticality results if the island does not appear in surface structure as a result of a deletion transformation. Rejecting such global derivational constraints, Chomsky () presents a representational constraint to account for island violations and their repair by ellipsis. He proposes that an island is marked by * (# in Chomsky’s a presentation) when it is crossed by movement, as illustrated by the following, where a complex NP island is crossed. ()

. . . [which one of my friends]i she kissed [NP* a man who bit ti]

An output condition rules out a representation with * in surface structure, explaining the unacceptability of standard island violations. Chomsky (a) further proposes that if a later operation deletes a category containing *, the derivation is salvaged, explaining island violation repair by ellipsis.10

9

The content of this section is based on Lasnik (). Lakoff () argues that Chomsky’s (a) *-marking analysis is conceptually problematic because it requires that a new element be introduced, perhaps foreshadowing inclusiveness condition violations in modern terms. Kitahara () explicitly appeals to inclusiveness in rejecting *-marking for a different but related phenomenon, with * on a trace left behind by an overly long movement, as in Lasnik and Saito () and Chomsky (). Lasnik () suggests a technical solution to this technical problem: Everything is “born” with a √. When a violation occurs, the √ is erased. A representation with a category without a √ yields an unacceptable sentence. Also see Temmerman (), developing ideas of Merchant (b), for an argument for *-marking of traces. 10



  

Merchant () gives an empirical argument against the *-marking analysis. He points out that island violations are not always repaired by ellipsis. For example, in (), which moves out of a complex NP island, and the crossed island is deleted by VP-ellipsis. In spite of this, the sentence is unacceptable (i.e., the island violation is seemingly not repaired). ()

*They want to hire someone who speaks a Balkan language, but I don’t know whichi they do [VP want to hire [NP someone who speaks ti]].

This contrasts with sluicing on the face of it, as shown in the following: ()

They want to hire someone who speaks a Balkan language, but I don’t know whichi (Balkan language) [IP they want to hire [NP someone who speaks ti]].

Under the *-marking analysis, the contrast between VP-ellipsis and sluicing in this respect is not expected.11 Merchant () argues that all ellipsis is a deletion operation at PF and classifies islands into two types: PF islands (e.g., left-branch islands, COMP-trace effects, and derived position islands) and propositional islands (e.g., complex NP islands, adjunct islands, and sentential subject islands). PF islands are constraints that apply to PF representations. Therefore, ellipsis as a deletion operation at PF can get rid of illegitimate PF representations. Merchant () argues that the case of apparent repair of propositional islands involves a smaller subpart of the antecedent that does not contain an island. For example, () is reanalyzed as in the following: ()

They want to hire someone who speaks a Balkan language, but I don’t know whichi (Balkan language) [IP she should speak ti].

In contrast with sluicing, the case of VP-ellipsis (i.e., ()) cannot be analyzed as involving an alternative short source that does not contain an island. Thus, the contrast between () and () follows.12 Explanation of Multiple Sluicing. Lasnik () argues that multiple sluicing in English involves wh-movement of the first wh-phrase and rightward movement of the second wh-phrase, as illustrated by the following: ()

But they didn’t tell me [CP [which] [IP ti [got something tj]][from which]]

Lasnik’s () arguments for the rightward movement analysis are based on the observation that multiple sluicing in English is subject to constraints on rightward

However, Merchant (b) proposes a variant of the *-marking analysis that can accommodate this contrast between VP-ellipsis and sluicing. 12 Lasnik () argues that propositional island violations can be repaired even if an alternative short source is not available (see Barros et al.  for arguments against Lasnik’s arguments). Furthermore, Lasnik () also argues that apparent failure of repair in VP-ellipsis obtains even when Merchant’s () PF islands are involved. See Fox and Lasnik () for detailed discussion about the contrast between sluicing and VP-ellipsis in terms of island repair. 11

   



movement in general, rather than those on wh-movement. First, the second wh-phrase in multiple sluicing prefers to be a PP rather than a DP, as illustrated by the following: ()

a. ?Someone talked about something, but I can’t remember [who] [about what]. b. ?*Someone saw something, but I can’t remember [who] [what]. (Lasnik : )

The same preference for PP over DP is found in rightward movement, as shown in the following: ()

a. Some students met ti yesterday [with some professors]i. b. ?*Some students met ti yesterday [some professors]i.

(Lasnik : )

Second, it is well known that rightward movement of DP improves when the DP is made ‘heavier’, as illustrated in the following (see Ross  and Fiengo  among others): ()

a. *Mary saw yesterday [Harry]. b. Mary saw yesterday [her old friend Harry]. c. Mary saw yesterday [Harry Hetherington].

(Lasnik : )

Likewise, multiple sluicing involving DP as the second wh-phrase improves when the DP is made heavier, as shown in the following: ()

?Some linguist criticized (yesterday) some paper about sluicing, but I don’t know [which linguist] [which paper about sluicing]. (Lasnik : )

Third, rightward movement does not allow P-stranding, as illustrated by the following (see Ross  and Drummond et al. ): ()

a. *A linguist spoke about yesterday [a paper on sluicing]. b. A linguist criticized yesterday [a paper on sluicing].

(Lasnik : )

The second wh-phrase in multiple sluicing is subject to the same constraint: ()

Some linguist spoke about some paper on sluicing, but I don’t know [which linguist] [?*(about) which paper on sluicing]. (Lasnik : )

Finally, rightward movement is subject to the Right Roof Constraint (Ross ) that rightward movement out of an embedded clause is prohibited, as shown in the following: () a. Some students spoke ti yesterday [to some professors]i. b. *Some students said [that Mary will speak ti] yesterday [to some professors]i. (Lasnik : ) Likewise, the first wh-phrase and the second one in multiple sluicing cannot be separated by a clause boundary, as () shows.



  

()

*Some of the students wanted [John to go to some of the lectures], but I’m not sure [which] [to which]. (Lasnik : )

Given these similarities between rightward movement and multiple sluicing, Lasnik () concludes that multiple sluicing in English involves rightward movement of the second wh-phrase. This analysis is compatible with the deletion analysis of sluicing. Thus, multiple sluicing ceases to be problematic for the deletion analysis under Lasnik’s () account.

.. Summary of section 3.2 In this section, we reviewed arguments in favor of the structural analysis of ellipsis over the non-structural analysis, and then we compared the two variants of the structural analysis (i.e., the deletion analysis and the LF-copying analysis). We saw that there are arguments for the deletion analysis, and that potential problems for the deletion analysis (island repair and multiple sluicing) can be solved under Merchant’s () and Lasnik’s () proposals.

. R

.................................................................................................................................. Elided material must be recoverable somehow. It is generally agreed that elided material must be identical to its antecedent in order to guarantee the recoverability of ellipsis. However, the precise formulation of the identity condition of ellipsis is controversial. Some argue that elided material and its antecedent must be syntactically or formally identical (Chomsky , ; Sag a; Williams b; Kitagawa ; Fiengo and May ; Chung et al. ; Lasnik b; Fox ; Fox and Lasnik ; Chung ; Merchant a; among others) while others argue that they must be semantically or pragmatically identical (Dalrymple et al. ; Hardt c; Prüst ; Ginzburg and Sag ; Merchant ; Yoshida ; Weir , among others). In this section, we review some proposals under syntactic identity and semantic identity. First, we will present arguments for syntactic identity in section .., and then we will look at arguments against syntactic identity and for semantic identity in section ...

.. Syntactic identity The first relevant discussion of ellipsis appears in Chomsky (). He begins by considering the following examples (p. ): ()

a. b. c. d.

I saw the play and so did he. I will see the play and so will he. I have seen the play and so has he. I have been seeing the play and so has he.

   



For Chomsky, the derivation of such sentences involves a generalized transformation (conjunction), a permutation (exactly the same one used in simple yes–no questions), and a ‘deformation’ (deletion operation) eliminating the material after the first auxiliary. The first mention of an identity requirement appears here, but it is not the requirement one might expect (material to be deleted identical to an antecedent). Rather, Chomsky suggests that the auxiliaries must be identical: ()

*I will see the play and so did he.

“ . . . the auxiliary phrases must be identical in the two conjuncts . . . for the transformation . . . to apply.” That is, material outside the ellipsis site must be identical, an interesting possibility discussed briefly much later by Fox and Lasnik () and Lasnik and Park (). Simple VP-ellipsis also comes under investigation, including now classic instances such as: ()

a. b. c. d.

John will. John is. John has. John does.

Again, Chomsky notes that “these appear to be ‘truncated’ sentences, with a verb phrase ‘understood’ ” and he indicates that formal analysis leads to the same conclusion, thus “providing grounds in formal linguistic structure for this intuition.” In particular, precisely the same analysis of the English verbal system that Chomsky had motivated to account for yes–no question alternations is efficacious in describing the ellipsis patterns. Chomsky does point out an interesting problem that arises, an apparent ordering paradox. In the derivation of (c), have and the perfect morpheme en were introduced together in the Aux. So to derive the correct output, it must be that the rule separating en from have and attaching it to the following verb (called Affix Hopping in later work) must precede the deletion rule, since that affix is included in the deletion site. (And likewise for (b).) On the other hand, the tense affix in (d) survives the deletion, so in this case, deletion must precede Affix Hopping. Chomsky’s solution is to propose that there are two Affix Hopping rules, one preceding deletion, for en and ing, and another for tense, which follows deletion. Another possibility is that Affix Hopping is the right account for tense morphology, but not for aspectual morphology, the latter being ‘lexicalist’. This is not unlikely since the strong evidence for syntactic separation of tense does not obtain for other pieces of inflectional morphology. More on this possibility, suggested by Lasnik (b), later in this subsection. A syntactic recoverability condition on elided material is first explicitly articulated by Chomsky (). There are two instances. First, a “designated representative of a category” may be deleted. For instance, Chomsky hints that indefinite pronouns (‘designated representatives’ of the category NP) underlie interrogative ones and thus are deletable by the interrogative transformation. The second instance is more familiar: apart from a designated representative, “a transformation can delete an element only if . . . the structural condition that defines this transformation states that the deleted element is structurally identical to



  

another element of the transformed string” (p. ). This formal identity condition guarantees that “A deleted element is . . . always recoverable.” Chomsky () explores this recoverability condition further. He first suggests, subject to later refinement, that “an erasure operation can use the term X to delete Y just in case X and Y are identical” (p. ). Chomsky then considers the notion ‘identity’, indicating that it might be taken to mean “strict identity of feature composition” (p. ). But he then gives a pair of examples showing that this cannot be quite right: ()

John is more clever than Bill.

()

These men are more clever than Mary.

Assuming a clausal source for comparatives, () works as expected, with is clever deleted under complete identity. But, Chomsky observes in () the antecedent for deletion of is clever is are clever, which is not fully identical. The problem is even more evident in the French analogue, where the deleted adjective is not completely identical to its antecedent, differing in the feature plural, like the copula, and also in being +feminine: ()

Ces hommes sont plus intelligents que Marie.

Based on this, Chomsky makes the intriguing suggestion that features added by agreement rules do not have the same status as features that are inherent parts of lexical items: . . . in the case of Adjectives and the copula (also Verbs, which take part in similar rules) the inflectional features that are added by agreement transformations are apparently not considered in determining whether the item in question is strictly identical with some other item. (p. )

Even the plural feature of a nominal expression is not always inherent. Predicate nominals show the pattern of verbs and adjectives, as seen in ()—reasonable since in this instance as well, the feature is triggered by agreement.13 ()

I know several more successful lawyers than Bill.

Based on this, Chomsky offers two conclusions: First, features introduced by transformation into lexical formatives are not to be considered in determining when deletion is permitted; a formative, in other words, is to be regarded as a pair of sets of features, one member of the pair consisting of features that are inherent to the lexical entry or the position of lexical insertion, the second member of the pair consisting of features added by transformation. Only the first set is considered in determining legitimacy of deletion in the manner previously described. Second, what is involved in determining legitimacy of deletion is not identity but rather nondistinctness in the sense of distinctive feature theory . . . Thus consider once again the case of ‘I know several lawyers’—‘Bill is a

13

See also Merchant (a).

   



lawyer.’ The Predicate-Nominal of the latter is not singular, in the base structure; rather, it is unspecified with respect to number exactly as the nasal is unspecified with respect to point of articulation in the lexical representations of the formatives king, find, lamp, etc. Hence, it is not identical with the corresponding nominal element of ‘I know several lawyers’; it is, rather, nondistinct from it, and the example suggests that this is sufficient to permit deletion. (p. )

Finally, Chomsky argues that this analysis need not be stated or even mentioned in any specific rules of grammar, since: . . . it is, apparently, determined by a general convention regarding the form of grammar. In other words, we are tentatively proposing it for consideration as a linguistic universal, admittedly, on rather slender evidence . . . If this proposal is a correct one, then the analysis of formatives that we have suggested is a general condition on the functioning of erasure transformations. (pp. –)

In a further consideration of formal identity, Lasnik (b) presents an account of a surprising asymmetry in English VP ellipsis first noticed by Warner () and rediscovered by Colbert (). Lasnik’s account is based on a ‘hybrid’ approach to English verbal morphology, initially motivated by the divergent behavior of auxiliary and main verbs with respect to negative placement and Subject-Aux Inversion. In this hybrid approach, main verbs come to be associated with their verbal morphology (particularly their tense morphology) via Affix Hopping. Auxiliaries, on the other hand, are ‘lexicalist’, introduced into the syntactic derivation already fully inflected. The core of the ellipsis asymmetry is that finite forms of main verbs, but not of be, can apparently antecede deletion of infinitival forms: ()

John slept here, and Mary will sleep here too.

() *John was here and Mary will be here too. If, as Lasnik argued, slept is underlyingly sleep with a syntactically independent tense morpheme (as in Chomsky  and ), there is a stage in the derivation where slept and sleep are completely identical. On the other hand, if was is was throughout the derivation, there is no stage where it is identical to be, thus excluding (), if there is a strict formal identity requirement on ellipsis:14 ()

14

A form of a verb V can only be deleted under identity with the very same form.

See Rouveret () for extensive commentary on Lasnik’s (b) account, and an alternative that relies on another sort of formal identity requirement, due to Chung (). Immediately below, we will briefly discuss Chung’s proposal. Also see Potsdam (a) for a challenge to Lasnik’s (b) account, and Lasnik () for a partial response.



  

In further motivation of a formal identity requirement, Chung () observes an extremely interesting constraint on ‘sprouting’, a type of sluicing where there is no apparent antecedent for the wh-trace (see Chung et al. ): ()

She solved the puzzle, but I don’t know how.

()

He ate, but I don’t know what.

It should be noted at the outset that sprouting raises non-trivial difficulties for any identity requirement, whether syntactic or semantic, as noted by van Craenenbroeck and Merchant (). One possibility briefly suggested by van Craenenbroeck and Merchant () is that there is a null adjunct or null argument that does serve as an antecedent. We briefly return to this question in .... Chung’s discovery is that when the missing antecedent would have been a PP, the sluicing fragment must also be a PP, and not a mere DP: ()

They’re jealous, but it’s unclear *(of ) who.

()

Joe was murdered, but we don’t know *(by) who.

()

Mary was flirting, but they wouldn’t say *(with) who.

() We’re donating our car, but it’s unclear *(to) which organization. It is worthy of note that even when the preposition is predictable, as in some of the above examples, it still must show up in the sprouting construction. This is in sharp contrast with standard sluicing (in a language allowing preposition stranding, like English). In these instances, the preposition is optional (and probably even dispreferred) in the fragment: ()

They’re jealous of someone, but it’s unclear (of ) who.

()

Joe was murdered by someone, but we don’t know (by) who.

The presence of some prepositions seems entirely formally motivated, as in () but even these must show up in sprouting. Significantly, this pattern in Chung’s examples is not unique to English. It shows up, as Chung points out, in other languages with preposition stranding, such as Danish and Norwegian. Chung concludes that we must look beyond semantics and pragmatics to account for the contrasts. That is, there is a formal identity condition at work here. Merchant (a) presents another argument for a syntactic identity requirement on ellipsis. Merchant observes that sluicing doesn’t tolerate active–passive mismatches:15 () *Joe was murdered, but we don’t know who murdered Joe

15

Potsdam () argues that voice mismatches are tolerable in Malagasy.

    ()



*Someone murdered Joe, but we don’t know who by/ by whom.

On the face of it, this is rather strong evidence for a syntactic identity requirement, since actives and corresponding passives are very close to semantically parallel. But there is evidently a glitch, since VP ellipsis does seem to tolerate active–passive mismatches: ()

This problem was to have been looked into, but obviously nobody did __

()

The janitor should remove the trash whenever it is apparent that it needs to be __

Merchant’s insightful solution to this nearly paradoxical situation reaffirms syntactic identity. He proposes that the functional head responsible for active and passive voice is included in a clausal ellipsis site, but is outside of a VPE site. Thus in the case of VPE, antecedent and ellipsis site are syntactically identical (neither active nor passive), while in the case of sluicing they are not, thus accounting for the contrast between (), () on the one hand and (), () on the other.16

.. Semantic identity ... Arguments against (strict) syntactic identity Merchant (: –) provides arguments against syntactic identity. The first argument, alluded to in section .., concerns sprouting: () a. Abby was reading, but I don’t know what. b. Ben called—guess when!

(Merchant : )

In these examples, there is no overt element in the antecedent clauses that corresponds to the remnant wh-phrase (what in (a) and when in (b)). The IPs in the antecedent clauses and those in the sluiced clauses are therefore structurally different. Thus, examples like these should be unacceptable under a strict syntactic identity condition. Merchant () also argues that sluicing in Romanian poses a problem for syntactic identity. In Romanian, a D-linked wh-phase such as care ‘which’ must be accompanied by a clitic, as illustrated by the following (see Dobrovie-Sorin ; Grosu : ): ()

16

[Pe care băiat] *(l-)  which boy him ‘Which boy did you see?’

ai have.

văzut? seen (Romanian, Merchant : )

See Kertz () for an information-structure-based alternative account of the asymmetry between sluicing and VP-ellipsis in voice mismatches.



  

In this example that involves the D-linked wh-phrase pe care băiat ‘ which boy’, the clitic l- ‘him’ is required. Merchant () observes that a D-linked remnant wh-phrase in sluicing does not require a clitic in the antecedent clause, as shown in the following: () Am văzut pe un băiat dar I.have seen  a boy but ‘I saw a boy, but I don’t know which.’

nu not

ştiu [pe care]. I.know  which (Romanian, Merchant : )

There must be a clitic associated with the D-linked remnant wh-phrase in the sluiced clause. In spite of this, there is no clitic in the antecedent clause. This means that the elided IP and the antecedent IP are structurally different. Thus, () should be unacceptable under the syntactic identity condition, contrary to fact. Finally, Merchant () argues that what Fiengo and May () call vehicle change phenomena pose a problem for a strict syntactic identity condition, which incorrectly predicts that the following sentence should be unacceptable due to a Condition C violation: () They arrested Alexi, though hei thought they wouldn’t.

(Merchant : )

Under a strict syntactic identity condition, the ellipsis clause headed by though in () must be as in the following: () *Hei thought they wouldn’t [arrest Alexi].

(Merchant : )

This sentence contains an R-expression Alex that is bound by he, violating Condition C. Therefore, () should be unacceptable under a strict syntactic identity condition.17 A further argument against syntactic identity comes from a growing body of literature that provides evidence that a ‘cleft’ can be an underlying structure for ellipsis when the antecedent does not contain a cleft18 (Szczegelniak , ; Vicente ; Rodrigues et al. ; Barros and van Craenenbroeck ; see also van Craenenbroeck a).

17

The other argument against syntactic identity by Merchant () concerns English verbal morphology in ellipsis. The verbal morphology in the sluiced clause does not have to match with that in the antecedent clause, as illustrated by the following (and as already noted by Ross b): (i)

a. [Decorating for the holidays] is easy if you know how [to decorate for the holidays]. b. I remember [meeting him], but I don’t remember when [I met him]. (Merchant : –)

However, this largely ceases to be problematic under Lasnik’s (b) analysis of English verbal morphology. This is so because as we saw in section .., there is a stage in the derivation where decorating and decorate in (ia) and meeting and met in (ib) are completely identical. The deletion of to in (ia) might remain a problem. 18 The relevant literature almost invariably calls these potential ellipsis sources ‘clefts’, but that usage is in conflict with tradition. Thus, a cleft version of (69a) is really (i), which contains all the elements of the original sentence: (i)

It is Jack who is sleeping.

   



For example, Barros and van Craenenbroeck () argue that tag questions provide evidence for the possibility of clefts as an underlying source for ellipsis. As () indicates, tag clauses and assertion clauses that are followed by them are subject to a strict identity condition. () a. Jack is sleeping, {isn’t he/*isn’t it}? b. It is Jack, {*isn’t he/isn’t it}?

(Barros and van Craenenbroeck : )

Given this property of tag questions, Barros and van Craenenbroeck () argue that the ellipsis site of the fragment answer in (B) has a cleft structure rather than a structure isomorphic with the antecedent. () A: Bill met a member of the Linguistics Department. B: Yes, Ken Safir, {wasn’t it/??didn’t he}? (cf. B’: Yes, he met Ken Safir, {*wasn’t it/ didn’t he}?) (Barros and van Craenenbroeck : ) This is problematic with a strict syntactic identity condition, under which the ‘wasn’t it’ version of (B) is expected to be unacceptable.19

... Merchant’s semantic identity Given the empirical problems with syntactic identity, Merchant () argues that there is no syntactic identity requirement at all, and instead proposes a semantic identity condition on ellipsis. Informally, Merchant’s () semantic identity condition amounts to requiring that the antecedent entail the elided element, abstracting away from focused material that it contains, and the elided element also entail the antecedent, again abstracting away from

19

Another potential problem for syntactic identity concerns antecedent-contained deletion (ACD; Bouton , May , Hornstein a, and Kennedy , among others). A typical example of ACD is given in the following: (i) Dulles suspected everyone who Angleton did.

(May : )

In (i), the elided VP is contained in its antecedent VP. Thus, if ellipsis is subject to a (strict) syntactic identity requirement, (i) has the representation shown in (ii). (ii) Dulles suspected everyone who Angleton did [VP suspect everyone who Angleton did [VP suspect everyone who Angleton did [VP . . . ]]] (ii) is uninterpretable since it contains an infinitely large VP. Thus, it is a mystery under syntactic identity why (i) is interpretable. However, May’s () analysis makes ACD compatible with syntactic identity. Assuming the operation Quantifier Raising (QR) in LF, which is A0 -movement that adjoins a quantificational phrase to a dominating node (IP or VP), May () argues that (i) has the following LF representation: (iii) [IP [DP everyone whoi Angleton did [VP suspect ti]]j [IP Dulles [VP suspected tj]]] In (iii), the quantificational DP undergoes QR, adjoining to the matrix IP. As a result, the elided VP is no longer contained in its antecedent VP, deriving the well-formed LF representation. Thus, ACD becomes unproblematic for syntactic identity (i.e., at the level of LF) under May’s QR analysis.



  

focused material.20 With this condition, let us consider (a), the representation of which under the deletion analysis is given in (): () [IP Abby was reading], but I don’t know whati [IP Abby was reading ti]. There are no focused portions either in the antecedent IP or the elided IP. Given that the trace of wh-movement is translated as an existentially bound variable, the elided IP roughly means that there is something Abby was reading. The antecedent IP also entails that there is something Abby was reading. The elided IP and the antecedent IP entail each other, satisfying Merchant’s semantic identity condition. Therefore, the semantic identity condition can account for sentences like (a). Merchant’s semantic identity condition also straightforwardly accounts for the Romanian example with a D-linked wh-phrase () because the antecedent and the elided element in this example are semantically equivalent, hence entailing each other. The vehicle change phenomena can also be accommodated under Merchant’s semantic identity. Suppose that the elided VP in () contains the pronoun him rather than the R-expression Alex, as illustrated by the following: ()

They [VP arrested Alexi], though hei thought they wouldn’t [VP arrest himi].

This representation does not violate Condition C. The semantic identity condition allows ellipsis like this because the antecedent VP and the elided VP are semantically equivalent when the pronoun him refers to Alex. Yoshida () provides further evidence for semantic identity on the basis of what he calls the sluicing in adjunct (SIA) construction, exemplified by the following: ()

a. John does not love anyone without knowing who. b. John must love someone without knowing who.

(Yoshida : )

In these examples, the adjunct clauses undergo sluicing. Yoshida () argues that in the SIA construction, the antecedent for the elided IP in the adjunct clause is the matrix VP

20

The formal definitions are given below:

(i) Semantic identity condition on ellipsis (Merchant : ) A constituent α can be deleted only if α is e-. (ii) e-ness (Merchant : ; see also Schwarzschild ) An expression E counts as e- iff E has a salient antecedent A and, modulo ∃-type shifting, (a) A entails the F-closure of E, F-clo(E), and (b) E entails F-clo(A). (iii) ∃-type shifting (Merchant : , n. ) ∃-type shifting is a type-shifting operation that raises expressions to type and existentially binds unfilled arguments. (iv) F-closure (Merchant : ) The F-closure of α, written F-clo(α), is the result of replacing F-marked parts of α with ∃-bound variables of the appropriate type (modulo ∃-type shifting).

   



rather than the matrix IP. If this is correct, the SIA construction is problematic for syntactic identity because a VP and an IP are evidently different in structure. In what follows, we review Yoshida’s () argument that the antecedent in the SIA construction is the matrix VP rather than the matrix IP. The only available interpretation for the adjunct clauses headed by without in () is (). ()

. . . without knowing whoi [IP John loves ti]

This means that the elided material in () does not contain negation or must. This contrasts with ordinary sluicing. As shown in (), the only available interpretation is the one in which negation and must are interpreted in the ellipsis sites. ()

a. John isn’t inviting someone, but I don’t know who. b. You must select a color, but I can’t tell you which one.

(Yoshida : )

That the elided IPs in () do not contain negation or must means that the matrix IPs do not serve as the antecedents for the elided IPs because IPs contain negation and modal verbs. If the antecedents for the elided IPs were the matrix IPs, negation or must would be interpreted in the ellipsis site both under the syntactic identity approach and under the semantic identity approach. On the other hand, if the matrix VPs rather than the IPs serve as the antecedents, the SIA construction ceases to be problematic under the semantic identity condition. Suppose that the adjunct clauses headed by without in () are adjoined to VP,21 and that subjects are base-generated in specVP (the VP-internal subject hypothesis: Zagona ; Fukui and Speas ; Kitagawa ; Kuroda ; Koopman and Sportiche ), and that they leave copies in that position when they raise to specIP. Then, in (b), for example, the interpretation of the matrix (lower segment of) VP is ∃x. John loves x, and that of the elided IP is ∃x. he loves x. They are semantically equivalent.22 Thus, sluicing in (b) satisfies Merchant’s semantic identity. Note that this account of the SIA construction is possible only under the semantic identity approach since VP and IP can be identical only semantically, not syntactically.

Yoshida (: ) argues that the following constituency tests suggest that the adjunct clause is adjoined to VP: 21

(i)

a. I thought John loves someone without knowing who and [VP love someone without knowing who], he does tVP indeed. (VP-preposing) b. John loves someone without knowing who, and Mary does so (= [VP love someone without knowing who]), too. (Do so substitution) c. John loves someone without knowing who, and Mary does, too (= [VP love someone without knowing who]). (VP-deletion)

However, we cannot conclude from these data that the adjunct clause must adjoin to VP since they can be accounted for even if we just assume that the adjunct clause can adjoin to VP. 22 As a reviewer points out, the elided IP and the antecedent VP are not semantically equivalent if we take the tense information in the matrix clause into account. Noticing this potential problem, Yoshida () suggests that the elided IP does not contain a tense because the tense in the adjunct PP depends on the matrix tense.



  

Therefore, the SIA construction provides an argument for the semantic identity approach.

.. Summary of section 3.3 To sum up, there are arguments for syntactic identity and semantic identity. Given Lasnik’s (b) English verbal morphology argument, Chung’s () sprouting argument, and Merchant’s (a) voice mismatch argument, a semantic identity condition is not a sufficient condition on ellipsis. Furthermore, Merchant’s () sprouting, Romanian sluicing, and vehicle change arguments and Yoshida’s () SIA construction argument indicate that a syntactic identity condition should not be too strict, as Chomsky () already pointed out.

. L

.................................................................................................................................. It is not the case that ellipsis is possible whenever an identity condition is satisfied. Consider the following sluicing examples: ()

a.

One of the linguists [was going to the Leap Day party], but no one told me who [was going to the Leap Day party]. b. *[One of the linguists was going to the Leap Day party], but no one told me that [one of the linguists was going to the Leap Day party]. (Aelbrecht : )

In (a) the elided IP follows the remnant wh-phrase, while in (b) it follows the complementizer. In both examples, (both syntactic and semantic) identity is satisfied. In spite of this, (a) is acceptable while (b) is not. Therefore, there is something other than an identity condition to satisfy, in order for ellipsis to be licensed. There seems to be a near consensus that ellipsis requires the presence of a licensing head (Zagona , a, b; Chao ; Lobeck , , ; Saito and Murasugi a; Johnson b; Martin ; Merchant , a; Gergel ; Aelbrecht ): in order to be licensed, the ellipsis site must be in a local relation with a head with specific morphosyntactic properties. In this section, we review some proposals in the literature about what properties of licensing heads and what local relation between the ellipsis site and the licensing head are relevant to ellipsis licensing.

.. Licensing heads ... Agreeing heads Lobeck (, ) and Saito and Murasugi (a) argue that agreeing functional heads license ellipsis of their complements, based on Fukui and Speas’ () notion of Kase.

   



C with the feature [+Wh], I with the feature [+Tns/+Agr], and D with the feature [+Poss] or [+Plural] assign Kase to their specifier, as summarized in the following table: ()

C

I

D

Kase assigner

[+Wh] (Sluicing)

[+Tns/+Agr] (VP-ellipsis)

[+Poss], [+Plural] (NP-ellipsis)

Non-Kase assigner

that, if, for

to

the, a(n), that, this

Kase assigners are agreeing functional heads while non-Kase assigners are non-agreeing functional heads. D with [+Poss] or [+Plural] licenses ellipsis of its complement (i.e., NP), as illustrated by the following examples (this is an instance of so-called N’-ellipsis or NP-ellipsis; see Saito and Murasugi a):23 ()

a. John’s [talk] was short, but Mary’s [talk] was way too long. b. Although she might buy these [books on art history], Mary said she wouldn’t buy those [books on art history]. (Lobeck : )

Given that genitive case is assigned by D with [+Poss], NP-ellipsis in (a) is licensed by D with [+Poss]. A plural demonstrative is a D specified with [+Plural]. Thus, the D in (b) licenses NP-ellipsis. On the other hand, NP-ellipsis is not licensed by a non-agreeing functional head, as illustrated by (). ()

a. *A single protester attended the rally because the [protester] apparently felt it was important. b. *Sue toyed with the idea of buying a windsurfer, then decided she didn’t want a [windsurfer] after all. c. *Although John doesn’t like this [brand of frozen pizza], he likes that brand of frozen pizza. (Lobeck : )

Neither the, a, nor this is specified with [+Poss] or [+Plural], hence are not agreeing heads. Thus, NP-ellipsis in () is not licensed. Furthermore, the contrast between (a) and (b), repeated here as (a) and (b) respectively, can be attributed to the distinction between agreeing heads and non-agreeingheads.

A quantifier also licenses NP-ellipsis, as Lobeck (: ) observes: (i) Few [people] attended the rally because many [people] decided to watch the event on TV. Lobeck () argues that quantifiers are functional categories that head QP, taking NP as their complements, and they license ellipsis although they are not Kase assigners. This means that Kase is not a necessary condition for a licensing head. See section .... 23



  

() a. One of the linguists [was going to the Leap Day party], but no one told me who [was going to the Leap Day party]. b. *[One of the linguists was going to the Leap Day party], but no one told me that [one of the linguists was going to the Leap Day party]. (Aelbrecht : ) Given that C whose specifier is filled with a wh-phrase is specified with [+Wh] while C filled with a complementizer is not, the elided IP is licensed in (a) but not in (b). This is why the elided IP must follow a wh-phrase. An elided VP must be licensed by I with [+Tns/+Agr]. This is illustrated by the following:24 ()

a. Because she *(shouldn’t) [smoke], Mary doesn’t smoke. b. Dennis rarely plays the piano, but Susan often *(does) [play the piano]. (Lobeck : )

Given that only I that is lexically filled is specified with [+Tns/+Agr], I filled with a modal verb or with pleonastic do licenses VP-ellipsis while I that is not lexically filled does not. Non-finite I in raising constructions is not an agreeing head because it lacks [+Tns/+Agr]. Thus, it does not license VP-ellipsis, as shown in the following examples: ()

a. *I consider Pam to [VP like soccer], and I believe Rebecca to [VP like soccer] as well. b. *Bill believes Sarah to [VP be honest], and he believes Kim to [VP be honest] as well. (Martin : )

However, as Saito and Murasugi (a) and Takahashi () observe, VP-ellipsis is licensed in control infinitivals, as illustrated by the following: ()

a. Kim isn’t sure she can [VP solve the problem], but she will try to [VP solve the problem]. b. Rebecca wanted Jill to [VP join the team], so Pam persuaded her to [VP join the team]. (Martin : )

According to Martin (), control to, unlike raising to, is an agreeing functional head since it assigns null case to PRO in its specifier. Given this, the difference between raising to and control to can be reduced to the agreeing/non-agreeing distinction.

... Heads with strong features Martins () argues that functional heads with a strong feature, which triggers overt verb movement, license VP-ellipsis in Romance languages, based on her observation that there is a correlation between the possibility of enclisis in finite clauses and that of VP-ellipsis in

24

Though this specific argument is widely cited, there appear to be interfering factors. First, if we drop the shouldn’t in (81a), the resulting sentence, whether elliptical or not, is a contradiction: Because she smokes, Mary doesn’t smoke. Second, in both (81a) and (81b), deleting the VP should result in a stranded affix, triggering do-support. See the Chomsky () argument in section ...

   



Romance languages. Portuguese and Galician allow enclisis in finite clauses while Spanish, Catalan, French, and Italian do not, as shown in the following examples: () a.

()

Deste-ihe o livro? gave-him the book ‘Did you give him the book?’

(Portuguese, Martins : )

b. Décheslle gave.him

o the

livro? book

(Galician, Martins : )

a. *Diste-le gave-him

el the

libro? book

(Spanish, Martins : )

b. *Has-li have-him

donat given

c. *As-lui have-him

tu you

d. *Hai have

gli him

el the

llibre? book

donné given

le the

dato given

il the

(Catalan, Martins : ) livre? book

libro? book

(French, Martins : ) (Italian, Martins : )

Martins () observes that the cross-linguistic difference in terms of the possibility of enclisis is correlated to another cross-linguistic difference in terms of the possibility of VP-ellipsis: Languages that allow enclisis also allow VP-ellipsis while languages that do not allow enclisis do not allow VP-ellipsis. By VP-ellipsis, Martins () means V-stranding VP-ellipsis. V-stranding VP-ellipsis is VP-ellipsis that applies to a VP whose head has moved out of it (see Otani and Whitman ; Lasnik ; Potsdam a; Ngonyani ; Doron ; Cyrino and Matos ; Goldberg ; Toosarvandani ; Funakoshi , , ; Gribanova b; Abe ; Hayashi ; a number of these works show that mere object drop does not suffice to account for all the observed phenomena). When it applies to a transitive sentence, V-stranding VP-ellipsis yields a null object sentence, as illustrated by the following: ()

. . . V-I [VP . . . to ]

As () shows, VP-ellipsis is possible in Portuguese and Galician, which allow enclisis. ()

a. Si, dei [VP tdei yes gave [VP tgave ‘Yes, I did.’

lho]. him.it]

b. Si, yes

llo]. him.it]

din gave

[VP tdin [VP tgave

(Portuguese, Martins : ) (Galician, Martins : )

On the other hand, languages that do not allow enclisis do not allow VP-ellipsis, as shown in ().



  

() a. *Sí, yes

di. gave

b. *Sí, yes

he have

donat. given

(Catalan, Martins : )

c. *Oui, yes

j’ai I.have

donné. given

(French, Martins : )

d. *Sì, yes

ho have

dato. given

(Italian, Martins : )

(Spanish, Martins : )

In order to account for the correlation between enclisis and VP-ellipsis, Martins () proposes that enclisis results when a finite verb moves to a functional head (∑ in her terms; see also Laka ), and verb movement is triggered by a strong feature on the functional head. This is illustrated by the following schematic structure (Cl stands for a clitic):25 ()

. . . V-∑ . . . Cl . . . [VP tV . . . ]

The correlation follows if we further assume that VP-ellipsis must be licensed by a functional head with a strong feature.26

... Potential problems We have seen that ellipsis requires a licensing head, and a licensing head must be specified with certain features (Kase in English and a strong feature in Romance languages).27 This properly describes some facts, but not all, about ellipsis. For example, negation licenses VP-ellipsis in English, as Potsdam (a) points out (see also Baltin  and Lobeck ): () a. Ted hoped to vacation in Liberia but his agent recommended that he not [VP vacation in Liberia]. b. We think that Mary should present her case but we will ask that Bill not [VP present his case]. (Potsdam a: ) Even if we assume, following Pollock (), that sentential negation is the head of NegP, it is hard to assume that it is specified with Kase (i.e., it is an agreeing head). This means that Kase is not a necessary condition for a licensing head, as we already indicated in n. . Furthermore, a strong feature is not a sufficient condition on licensing V-stranding VPellipsis. In Danish, a V- language, verb movement is triggered by a strong feature. In spite of this, V-stranding VP-ellipsis is not allowed, as shown in the following examples:

25

Martins () assumes that a clitic adjoins to AgrS. Thoms (b) proposes a different approach to ellipsis licensing from the Lobeck–Martins approach. According to him, ellipsis is licensed by movement. Interestingly, this movement approach to ellipsis licensing makes similar predictions about V-stranding VP-ellipsis as Martins’ () approach. 27 Matos () proposes that the set of licensing heads can differ from one language to another (see also Martins : ). 26

    ()



a. *Vore øjne opfatter det ikke, men biers øjne opfatter [topfatter det]. our eyes perceive it not but bees. eyes perceive [tperceive it] (intended) ‘Our eyes don’t perceive it, but bees’ eyes do perceive it.’ b. Vore øjne opfatter det ikke, men biers gør [opfatter det]. our eyes perceive it not but bees. do [perceive it] ‘Our eyes don’t perceive it, but bees’ (eyes) do’. (Danish, Mikkelsen : )

(b) shows that ordinary (headed) VP-ellipsis is permitted in Danish. However, (a) indicates that V-stranding VP-ellipsis is not allowed.28 A more serious and general problem with these approaches is that there is no clear criterion on what counts as an agreeing head. We do not have ellipsis-independent evidence that heads with [+Wh], [+Tns], [+Poss], or [+Plural] are agreeing heads while that, if, for, and the non-control to are not. To put it more generally, it is not clear why a certain functional head with a certain feature is required to license ellipsis. Why are an agreeing functional head or ∑ with the strong feature relevant to licensing ellipsis? As far as we know, a principled explanation of this has not yet been proposed.

.. Relation between licensing heads and ellipsis sites It is generally assumed in the literature that it is not sufficient for a licensing head to have a certain feature specification in order to license ellipsis: a licensing head must be in a specific relation with the ellipsis site. It has been proposed (without much empirical evidence) that the relevant relation is government (Zagona ; Lobeck , ; Saito and Murasugi a), the head–complement relation (Merchant ), or c-command (Aelbrecht ). Aelbrecht () provides an argument that the relevant relation is not an extremely local one such as the head–complement relation. Consider the following VP-ellipsis sentence: () I hadn’t been thinking about that. Well, you should have been [thinking about that]. (Aelbrecht : ) Aelbrecht () argues that ellipsis of thinking about that in this example is licensed by the finite auxiliary should rather than the non-finite been, which is in the head–complement relation with the elided VP. That non-finite auxiliary verbs cannot license VP-ellipsis can be confirmed by the unacceptability of (a). ()

a. *I hadn’t been thinking about it, but I recall Morgan having been [thinking about it]. b. I hadn’t been thinking about it, but I recall Morgan having been thinking about it. (Aelbrecht : –)

See Funakoshi (, ) for other conditions on headless XP-ellipsis, of which V-stranding VP-ellipsis is a subtype. 28



  

The acceptability of the non-ellipsis counterpart of (a) (i.e., (b)) suggests that the unacceptability of (a) is due to (unlicensed) ellipsis. Given that non-finite auxiliary verbs cannot license VP-ellipsis, Aelbrecht () concludes that it is should rather than been that licenses VP-ellipsis in (). In (), the licensor should is not in the head–complement relation with the ellipsis site. Therefore, () indicates that a licensor does not have to be in the head–complement relation with an ellipsis site. Given this, Aelbrecht () argues that the relevant relation for ellipsis licensing is c-command (or the Agree relation, which is defined based on c-command).

. C

.................................................................................................................................. This chapter has discussed three issues related to ellipsis (i.e., the structure of the ellipsis site, recoverability, and licensing) from the point of view of transformational grammar. In section ., we reviewed arguments in favor of the structural analysis of ellipsis that mandates full internal structure in the ellipsis site at certain levels, and compared the two variants of the structural analysis (i.e., the deletion analysis and the LF-copying analysis). We concluded that there are arguments for the deletion analysis, and the potential problems for the deletion analysis (island repair and multiple sluicing) can be addressed under Merchant’s () and Lasnik’s () proposals. Section . concerned recoverability. We concluded that in order for elided material to be recoverable, a semantic identity condition must be satisfied, but that is not a sufficient condition; syntactic or formal identity must be taken into account (Lasnik’s b English verbal morphology argument, Chung’s  sprouting argument, and Merchant’s a voice mismatch argument). However, we also saw that a syntactic identity condition should not be too strict, as Chomsky () already pointed out (Merchant’s  sprouting, Romanian sluicing, and vehicle change arguments and Yoshida’s  SIA construction argument). Finally, we discussed licensing in section ., where we reviewed some proposals in the literature about what properties of licensing heads and what local relation between the ellipsis site and the licensing head are relevant to ellipsis licensing.

  ......................................................................................................................

     -                            ......................................................................................................................

    

. I

..................................................................................................................................

.. General properties of ellipsis T characterizing feature of ellipsis is that elements of semantic content are obtained in the absence of any corresponding syntactic form. The syntax thus appears to be incomplete. More specifically, the implicit semantic content is recovered from elements of the linguistic and non-linguistic context. In this sense, ellipsis is similar to anaphora, except that there is no overt anaphoric element involved. The elements present in an elliptical clause are predicates, arguments, or adjuncts of what is omitted. It is the presence of these elements that makes it possible to recognize the ellipsis. It is in principle always possible to reanalyze any elliptical phenomenon as a case of anaphora, either by hypothesizing an unpronounced pro-form in the elliptical site (zero anaphora) or by considering the licenser to be anaphoric (e.g., considering the preelliptical auxiliary to be an anaphor in verb phrase ellipsis (VPE); cf. Schachter ; Hardt ). It is an interesting point about the structure of the present handbook that there is no chapter on unrealized subjects of finite verbs, often treated in terms of ‘pro’, i.e., zero anaphora, in mainstream generative grammar. There is no a priori reason not to treat these simply as cases of ellipsis, just as there is no a priori reason not to analyze VPE as a case of anaphora. Decisions on these questions can only be made on the basis of further theoretical considerations. As is the case for overt anaphors, ellipsis can be anaphoric or exophoric. Resolution of ellipsis thus requires a theory of context, and specifically a theory of how both nonlinguistic and non-semantic material—information about the phonetic, phonological, and morphosyntactic realization of utterances—gets incorporated into context. More generally, ellipsis resolution requires a theory of conversation if one is to go beyond isolated examples and monologue. We will explain in this chapter how Head-Driven Phrase Structure Grammar (HPSG) allows one to account for context and for conversation.



  

On the other hand, most mainstream generative approaches fail to take these phenomena into account in sufficient depth. HPSG typically takes the view that there is no single mechanism underlying ‘ellipsis’. Rather, the term covers a variety of cases involving contextually based resolution. Three central cases can be distinguished. The first two of these involve constructional analyses of ellipsis (see section ..); the third involves non-canonical correspondence between the phonology of a phrase and its daughters, and is much more restricted and limited in its scope (see section ..). Case : Non-sentential utterances (NSUs), e.g., Bare Argument Ellipsis and sluicing (see sections ... and ..), which acquisition data shows are acquired over a span of years (see, e.g., Ginzburg and Kolliakou ). These are analyzed as specific constructions typically characterized by the fact that a phrase is taken to be the only constituent of a clausal structure. Case : Argument or predicate ellipsis, e.g., null subjects and null objects, VPE, nullcomplement anaphora, etc. (see sections ... and .). These are again analyzed in terms of specific elliptical constructions, typically involving a non-canonical correspondence between argument structure (ARG-ST) and syntactic subcategorization (SPR and COMPS lists), so that arguments that are semantically present can be syntactically absent. Thus, in these two cases, there is no unpronounced syntactic structure. Case : Constructions involving unpronounced syntactic structure, e.g., certain varieties of left- or right-peripheral ellipsis, in particular Right-Node Raising. These are analyzed by having the PHON (phonology) feature of the mother not simply be a concatenation of the PHON features of all of its daughters, as is normally the case. The latter analysis is rather similar in spirit to various mainstream analyses involving unpronounced syntactic structure, especially to analyses involving PF-deletion. However, HPSG studies have argued that this type of analysis is only applicable in a very limited number of cases characterized by specific properties, as discussed in section ... In the following subsection, we will present HPSG in general and in section ., we will present an overview of the analyses of elliptical phenomena in HPSG.

.. Head-Driven Phrase Structure Grammar Head-Driven Phrase Structure Grammar (HPSG) is a constraint-based linguistic theory initially proposed by Pollard and Sag () and further developed in Pollard and Sag () and numerous other papers. It is impossible to provide a detailed description of HPSG within the scope of this chapter.1 We will limit ourselves to sketching the essential elements of the framework that are crucial to understanding its approach to ellipsis. In HPSG, linguistic objects are modeled in terms of typed feature-value matrices. The value of a feature can be complex and even recursive. A language consists of a set of signs. These are abstract entities that are the locus of constraints on the interface between form, meaning, and use. A grammar is a system of constraints that conjointly define the signs of a 1 See Sag () and Müller (b) for recent synopses, and Sag et al. () for a pedagogical introduction. For further references, see the HPSG bibliography, maintained by Stefan Müller: .

  -   



given language.2 The theory uses types, organized in inheritance hierarchies, making it possible to express generalizations of varying granularity across linguistic levels.3 Because signs combine information about all levels of representation (phonology, morphology, syntax, semantics, pragmatics, discourse, and usage), HPSG is crucially non-modular. Constraints involving various levels can be easily stated. We will see that this is a powerful property of the framework for the analysis of ellipsis, as it allows one to express simultaneous semantic and syntactic constraints on ellipsis (explaining for instance the connectivity effects among non-sentential utterances) and provides means to integrate non-semantic information—information about the realization of utterances—into context. Thus, HPSG allows one to avoid two tendencies that are dominant in the literature: • ignoring linguistic complexity, as in logically-based models which abstract away from a linguistic level. • ignoring contextual complexity, as in generative accounts that abstract away from dialogue context. As we will see in more detail, HPSG initiated the constructional approach to ellipsis resolution, making central use of the notion of Question Under Discussion (QUD); cf. Ginzburg and Sag (). Furthermore, HPSG aims to be psycholinguistically plausible, specifically by providing a theory that is compatible with incremental processing.4 Finally, work in HPSG typically aims at strong empirical grounding, based on corpus research, fieldwork, acceptability experiments, etc. As in the Simpler Syntax framework (see Chapter  of this volume), no distinction is made between core and periphery. Specifically, it is unclear that phenomena relegated to the ‘periphery’ are less complex than ‘core’ phenomena, so that nothing is gained by the distinction in explaining acquisition: learning the periphery raises the same problems as learning the core. The ambition is to provide a large-scale description for a broad range of empirical data that is concise, formally precise (and thus falsifiable), and insightful. In the following section, we will introduce the two central types of analyses used to characterize ellipsis in HPSG, namely (i) constructional analyses (section ..), and (ii) non-canonical correspondence between the phonology of a phrase and its daughters (section ..). We conclude with some discussion of corpus studies of elliptical phenomena (section ..). The version of HPSG formalized in TTR (Type Theory with Records, cf. Cooper ; Ginzburg ), presented in section . refines this view. It takes a dialogue-oriented view of grammar in which a grammar provides types that characterize speech events (for detailed discussion, see Ginzburg and Poesio ). Thus, it proposes a linguistic ontology that includes both tokens (speech events, modeled as records) and types that characterize such tokens (signs, modeled as record types). This view of grammatical tokens and types plays an important role in our account of conversational interaction, and is key to understanding various elliptical resolution processes, as discussed in section .. 3 Sag et al. () and Sag () propose an integration between Berkeley-style Construction Grammar and HPSG under the name of Sign-Based Construction Grammar (SBCG). Under this view, constructions (including lexical constructions) are understood to be constraints on classes of signs and their components, which are organized into lattice-like arrays of types and subtypes. 4 In fact, recent work on phenomena such as self-repair and exclamative interjections (e.g., Ginzburg et al. b, ), following earlier work in Dynamic Syntax (e.g., Purver et al.  and Chapter  of this volume), takes a stronger line, namely that the grammar needs to be stated incrementally. See section ... for a brief discussion. 2



  

. A     HPSG

.................................................................................................................................. Our purpose in this section is to provide an overview of the types of analyses of elliptical phenomena that have been proposed in the HPSG literature. We attempt to convey the central insights behind the analyses, providing only the minimal formal details necessary so that readers unfamiliar with the framework can grasp the central intuitions and thus understand the three following sections which develop the HPSG positions on (i) the structure of the ellipsis site (section .); (ii) recoverability (section .); (iii) licensing (section .). In section ., we will provide a more technically precise account of the syntax of argument ellipsis, an area for which there has been a considerable amount of work in HPSG and for which there is a broad consensus on the analysis. Finally, in section ., we will introduce more recent developments and provide a detailed analysis of non-sentential utterances. In particular, we will introduce an alternative version of HPSG, one which allows the grammar to directly interface with dialogue context as conceived in the framework of KoS (Ginzburg ).5 As for the cases of constructions with unpronounced syntactic structure, for reasons of space, we will not be able to go into more detail than the brief introduction provided in section ...

.. Constructional analyses of ellipsis In this section we will introduce two central types of constructional analyses of ellipsis, namely cases where there is a non-canonical correspondence between argument structure and valence and cases of non-sentential utterances.6

... Non-canonical correspondence between argument structure and valence The feature structures describing lexical items in HPSG contain a feature called ARG-ST (argument structure), whose value is a list of feature structures describing the syntactic and semantic properties of the items they subcategorize for. Canonically, the ARG-ST list is the concatenation of the values of the two valence features, SPR (Specifier) and COMPS (Complements), which respectively govern the combination of the lexical item with its specifier and complements. () provides some typical examples of ARG-ST, SPR, and COMPS values. 5 KoS—a toponym, not an acronym—is a framework for describing dialogue interaction, which we will introduce in section .. 6 A further case of constructional ellipsis, namely nominal ellipsis, has not been discussed in this chapter for reasons of space. See, e.g., Nerbonne et al. (, ) and Nerbonne and Mullen () for HPSG analyses using a null head, and Branco and Costa () and Arnold and Spencer () for analyses avoiding an empty morpheme.

  -    ()

Lexical item laugh see speak will try

ARG-ST 〈NP[nom]〉 〈NP[nom],NP[acc]〉 〈NP[nom],PP[]〉 〈NP[nom],VP[base]〉 〈NP[nom],VP[]〉

SPR 〈NP[nom]〉 〈NP[nom]〉 〈NP[nom]〉 〈NP[nom]〉 〈NP[nom]〉



COMPS 〈〉 〈NP[acc]〉 〈PP[]〉 〈VP[base]〉 〈VP[]〉

The A R P (ARP; see () for a more precise formulation) states that, canonically, the value of ARG-ST for a given lexical item is the concatenation of the values of the SPR and COMPS features, as illustrated in (). Given this set-up, we can provide an analysis of the ellipsis of complements or specifiers simply by allowing non-canonical constructions in which items mentioned in the ARG-ST do not appear in the valence features (such constructions are non-canonical in the sense that they do not respect the general formulation of the ARP). For instance, we can account for verb phrase ellipsis7 by a constraint allowing auxiliaries to have alternate non-canonical lexical entries where their VP argument is specified as pronominal in the ARG-ST list and there is no corresponding VP in the COMPS list (for details, see (a) and (b)). This means that the VP meaning will be recovered anaphorically or exophorically from the context, ensuring appropriate semantic interpretation (see, e.g., Schachter ; Hardt ; Kim ), but will not be realized in the syntax, since it is not present on the COMPS list. Thus, a sentence like Kim will. receives a very simple constituent structure where the VP has only a single daughter, the V[AUX] will (see ()). We can provide an analysis for Null Complement Anaphora (NCA), illustrated in (a), along the same lines. However, it is well-known that NCA differs from VPE in that there is apparently no independently motivated subclass of verbs to which it is applicable (as shown by the classical contrast between (a) and (b)). Thus, specific idiosyncratic lexical constraints will be required for verbs like try which allow NCA, as opposed to the general constraint which applies in the case of English auxiliaries for VPE. ()

a. Kim could not open the door so Pat tried. b. *Kim could not open the door so Pat attempted.

This type of analysis can also be used for languages that allow null pronominal subjects and/or objects (sometimes called ‘pro-drop’; see, e.g., Manning and Sag ; Melnik ). In cases where they are systematically available, e.g., null objects in Brazilian Portuguese and null subjects in Spanish, a general constraint is required, similar to that suggested for English auxiliaries. Null objects, as in the Brazilian Portuguese example (), will be treated by having a pronominal NP (rather than a VP as in English VPE) as the second item of the We use the traditional name, rather than the more appropriate ‘Post-Auxiliary Ellipsis’ suggested by Sag (a: ): what is ellipted is not always a (surface) VP (in particular, it can be any predicative complement of the copula be), and it doesn’t even have to be a surface constituent. Furthermore, in examples of NCA like (a), a VP is ellipted, but the usual criteria show that this is not VPE; see Miller and Pullum (: ). Note also that we are using ‘auxiliary’ in the classical sense in English linguistics, namely a verb that has the NICE properties (cf. Huddleston ), so that copular be and some uses of the main verb have are included. 7



  

ARG-ST list, with no corresponding item on the COMPS list. Null subjects, as in the Spanish example (), are similar except that they require a pronominal NP as first item of the ARG-ST list with no corresponding item on the SPR list. This leads to constituent structures where the VP contains only the verb and the prepositional object, as in (), or the S contains only a VP daughter, as illustrated in (). Thus the constituent structures for these sentences contain no unpronounced positions in the syntax (see section .. for further detail). ()

Maria escolheu o livro e [[ Ana] [[ colocou] [[ na]] [ estante]]]. put in bookcase Maria chose the book and Ana ‘Maria chose the book and Ana put it in the bookcase.’

() Maria llegó. [[[ Estaba] was Maria arrived ‘Maria arrived. She was tired.’

[ cansada]]]. tired

In certain languages, null subjects and objects are subject to various restrictions, requiring complexification of the relevant constraints (see, e.g., Melnik  for a discussion of null subjects in Hebrew). In all of these cases the resolution of the ellipsis reduces to the resolution of the anaphoric element on the ARG-ST list. We will provide further detail of how this works for VPE in section ...

... Non-sentential utterances: Sluicing, short answers, and gapping In this section we will briefly introduce the analyses of non-sentential utterances (NSUs), more specifically  and  , which have been proposed in HPSG (see in particular Ginzburg and Sag : ch. ; Ginzburg ). We will also briefly present the analysis of  proposed by Abeillé et al. (), in which the gapped clause is treated as a cluster of NSUs. Both sluicing and short answers involve XP ‘fragments’ which receive a ‘sentential’ interpretation (e.g., as propositions and questions). Following Ginzburg and Sag (), we will analyze these as specific sentential constructions where the verbal projection is reduced to a single phrase. Despite their sentential semantics, treating NSUs as sentences in the sense of verbal projections may not be the best analysis in all cases. For certain cases, e.g., sluices, it clearly is, since it is well-known that they have a sentence-type distribution, rather than an NP or PP distribution (see, e.g., Merchant : section . and Culicover and Jackendoff : –). However, for other cases, there is cross-linguistic variation in this respect. For instance, oui (‘yes’), non (‘no’), si (‘yes’) can be embedded under the complementizer que in French (Je crois que oui/non/si. ‘I think that yes/no/yes.’), whereas this is much more difficult in English. Similarly, NP short answers can be embedded in the Brazilian Portuguese example in (). This is clearly much less natural in English, as shown by the glosses.8 8 Ginzburg and Sag () use the feature
IC to distinguish embeddable from non-embeddable clauses (as in e.g. () and ()). See van Craenenbroeck and Lipták (: section .) and Temmerman () for proposals as to which languages do and do not allow such embedded NSUs.

  -    ()

A: Quem foi pra festa? B: Acho que o who went to party think that the ‘A: Who went to the party? B: I think that João.’



João. João

As will be discussed in more detail in sections .. and ., the form and interpretation of these NSUs are determined by elements of the previous context. Gapping, illustrated in (a), has received much attention in the literature on ellipsis (see also Chapter  of this volume). Many analyses, following Ross (), invoke some form of a deletion process, which, in more recent minimalist versions, is sometimes fed by movement of the remnants to the left periphery (see among others Sag a; Merchant a; Hartmann , as well as Chaves  within an HPSG framework). Other studies have proposed a specific construction mapping a headless structure into a propositional meaning, accounting, among other properties, for the fact that the order of remnants in the gapped clause does not necessarily parallel that of their correlates in the antecedent clause, as illustrated in (b) (see among others Sag et al. ; Steedman ; Gardent ; Culicover and Jackendoff ; Mouret a; Bîlbîie ). Following this tradition, Abeillé et al. () provide a series of further arguments against deletion-based analyses (based on French and Romanian data) and propose a construction-based approach set in an HPSG framework. Specifically, they show that syntactic parallelism is less strict than is usually assumed: each remnant in gapping is required to match some subcategorization frame of the verbal predicate its correlate depends on. This is illustrated for Romanian in (c) (Abeillé et al. :  (b)), where the second remnant, tuturor copiilor, is a dative NP whereas its correlate in the antecedent, la trei dintre copii is a PP. On the other hand, discourse parallelism is required. The gapped clause is analyzed as a cluster of major fragments.9 () a. A policeman walked in at , and a fireman, at . b. A policeman walked in at , and at , a fireman. (Sag et al. : ()) c. Ion oferă mere [la trei dintre copii], iar Maria [tuturor copiilor]. Ion gives apples to three of children, and Maria all. children.. ‘Ion gives apples to three of the children, and Maria to all of them.’

.. Non-canonical correspondence between the phonology of a phrase and its daughters Right-Node Raising (RNR), also known as Right-Peripheral Ellipsis has been the subject of numerous studies (see, e.g., Culicover and Jackendoff ; Hartmann ; Johnson ). Within the HPSG literature, Crysmann (); Yatabe (); Beavers and Sag (); Chaves (, ); Abeillé et al. () have insisted that it has radically different properties from the constructions discussed in sections ... and ..., which warrant a different type of analysis. 9 For reasons of space, we have not dealt with coordinate ellipsis in this chapter, beyond what little is said here about the specific case of gapping. Among the HPSG works on the topic are: Sag et al. (); Beavers and Sag (); Mouret (, a); Chaves (); Yatabe ().



  

For reasons of space, we will not be able to discuss the HPSG analyses of RNR beyond the brief (and somewhat terse) summary given here of the most recent developments. Specifically, Chaves () has argued that many of the complexities in the vast literature on RNR can be eliminated if it is recognized that many cases of apparent RNR also have analyses either as cataphoric VPE or N’-ellipsis or as ATB Extraposition. If one restricts oneself to cases which are unambiguously RNR, it appears that RNR “(i) can target any peripheral string of words that can form an independent prosodic unit, and (ii) imposes morphological form identity” (Chaves : ).10 Chaves furthermore argues that certain cases of RNR cast doubt on the possibility of maintaining classical accounts of RNR, in which the RNRaised string undergoes movement.11 ()

a. Robin does  —or  not to play—[with a full deck]. (cf. *It is with a full deck that Robin does not play __.) b. Do you primarily work with - or with [dontists]? (cf. *Dontists, I don’t think I could work with ortho-.). c. It is possible that someone with a —and assumed that someone with an —[set of golf irons][would make this hole in one]. (Chaves : – (), (b), (a))

In (a) the putatively RNRaised string is an idiom chunk; in (b) it is a word-part; in (c) it is very difficult to argue that it could be a constituent. In all three cases, there is no reason to believe that the RNRaised string is a syntactically movable entity (as shown by the ill-formedness of the variants in parentheses). In order to account for these properties, Chaves proposes an analysis which might be considered to be close in spirit to PF-deletion. In general, in HPSG, when signs are combined, the phonology of the mother is assumed to be the concatenation of the phonology of its daughters. The construction for RNR, on the other hand, is exceptional in that it allows one to transfer only the phonology of the second of two sequences of morphophonological constituents that are identical up to morphological forms. For technical details, see Chaves (), especially pp. ff. The following properties characterize elliptical phenomena that must be analyzed in this way: (i) they require strict syntactic identity;12 (ii) they are strictly intrasentential; (iii) they are never exophoric. Morphological form identity (cf. Booij ) imposes identity beyond phonological identity, in particular the same morphemes must be used with the same senses. 11 Chaves uses small caps to signal contrastive focus, with an L+H* tone, and square brackets to identify the RNRaised string. 12 Abeillé et al. () report corpus data, among which () and (), and the results of acceptability experiments on RNR in French. These provide evidence that there can be minor mismatches in contentless material (e.g., determiners and prepositions). Note that in () it would be ungrammatical to replace des by de in what corresponds to the ellipted segment so that there is not strict identity. Similarly, in (), parvient requires the preposition à. 10

() C’est de la responsabilité de l’Eglise de venir en aide aux migrants et aux it’s of the responsibility of the-church of coming in aid to migrants and to réfugiés qu’ils aient des papiers ou qu’ils n’aient pas de papiers. refugees that-they have - papers or that-they -have not . papers ‘It is the responsibility of the Church to help migrants and refugees whether they have or they don’t have papers.’

  -   



.. Elliptical phenomena: Evidence from corpora Corpus studies are important since they provide a “lower bound” for what the grammar needs to characterize; due to sparsity issues, admittedly, some rare phenomena (or at least rare in the settings sampled by a given corpus) might be missed. Hence, corpus-based generalizations need to be made on the basis of corpora with a wide range of genres and it is important to combine this source of data gathering with an experimental program. In this section, we review some studies that form an important backdrop to the HPSG/KoS work described below. We begin with a presentation of the corpus-based classification of clarification requests developed by Purver et al. (), because it provides an example of how the extensive discussions of certain elliptical constructions in the literature (in particular sluicing and Bare Argument Ellipsis) have ignored important aspects of the relevant range of data, which are brought to light by corpus investigation. We will then mention more briefly other corpus-based work on ellipsis in HPSG. Successful coverage of the empirical phenomena presented in these papers, we believe, constitutes a significant test of adequacy for any theory of ellipsis. Purver et al. () propose a classification of the range of form/contents of clarification requests (CRs) in the British National Corpus (BNC). These can take many forms, as illustrated in Table ., all but the first three being elliptical. The examples in this table are to be understood as clarification requests in a discourse context where A has just uttered ‘Did Bo leave?’ (except for the last case, Filler, where A utters the incomplete utterance ‘Did Bo . . . ’).13 Providing explicit formal analyses of just about any of these classes is a formidable challenge for any approach where ellipsis resolution is not embedded within a theory of dialogical interaction; for detailed discussion, see Ginzburg and Cooper (); Ginzburg (), who argue that neither deletion-type accounts, nor higher-order unification-type accounts can deliver the requisite readings. Indeed, to date, we are not aware of analyses of these phenomena in existing formal grammatical frameworks apart from HPSG. We highlight just several of the most significant issues. The first point to note is that a number of these forms are ones whose sole analysis is as clarification requests—this applies to the classes Wh-Substituted Reprise and to Gap. () Ce parti ne parvient pas à surmonter ses contradictions, voire ne souhaite pas, this party  manage not to overcome its contradictions, indeed  wish not surmonter ses contradictions. overcome its contradictions ‘This party can’t manage, or even doesn’t want, to overcome its contradictions.’ They consequently suggest an analysis which allows content-less material to be asymmetrically ellipted on the left, whereas content-full peripheral material must be shared on the right. Bîlbîie () reports similar examples in English found in the Penn Tree Bank, e.g. They were also as liberal or more liberal than any other age group in the  through  surveys (WSJ). Experimental work is required to decide whether these English examples have a similar status to that of the French examples above, or whether they should be considered as errors. ‘What?’ is multiply ambiguous. It can be used to signal that the previous utterance was not understood overall (it is then classified using the tag Wot); it can also be used to clarify the inanimate filler of an argument role (one that is implicit in the case of ‘Did Bo leave?’ and makes sense only on certain senses of ‘leave’; it is then classified using the tag Reprise Sluice). 13



  

Table . A taxonomy for clarification requests (Purver ) Category name of CR:

Example of CR:

Explicit Literal reprise Wh-substituted Reprise Wot Reprise Sluices Reprise Fragments Gap Filler

B: Did you say ‘Bo’? / What do you mean ‘leave’? B: Did BO leave? / Did Bo LEAVE? B: Did WHO leave? / Did Bo WHAT? B: Eh? / What? / Pardon? B: Who? / What? / When? B: Bo? / Leave? B: Did Bo . . . ? A: Did Bo . . . B: Win?

These constitute instances of forms whose meanings cannot be analyzed in interaction-free grammar. A second point relates to cross-turn parallelism. Ginzburg and Cooper (); Ginzburg () argue in detail that reprise fragments have two main classes of uses, one to request confirmation about the content of a previous sub-utterance; the other to find out about the intended content of a previous sub-utterance. Both uses have strong parallelism requirements, which require semantic and structural information to be projected across multiple turns into context. Thus, the former requires identity of morphosyntactic category between source and target, as illustrated in (a) and (b); the latter requires segmental identity between source and target, as exemplified in (c). Parallelism of the latter kind seems needed also for the Gap class of CRs: ()

a. A: Did she hit him? B: #He? / Him? b. A: Was she biking? B: Biking? / Cycling? / #Biked? c. A: Did Bo leave? B: Bo? [Intended content reading: Who are you referring to? or Who do you mean?] Alternative reprise: B: Max? [Lacks intended content reading; can only mean: Are you referring to Max?]

More generally, Fernández and Ginzburg (); Schlangen () characterize the range of NSU types in the BNC, showing that they can be reliably classified into a small number of categories, revolving around the commonality in semantic resolution process.14 Subsequently, detailed semantic and syntactic accounts of these classes have been developed in Schlangen (), Fernández (), and Ginzburg (), some of which we sketch in section .. These classifications have been extended cross-linguistically, with minor modifications: French (Guida ), Chinese (Wong and Ginzburg ), Spanish (Marchena ), and to other genres (see, e.g., Filtopoulos ). Focusing on one subclass of NSUs, which has attracted much attention, namely sluicing, the large generative literature on the topic (see e.g. Ross b; Chung et al. ; Merchant 14

Fernández et al. () develop a machine-learning based algorithm for this classification which achieves F-scores of approximately  percent.

  -   



, and many chapters in this volume) has, with very few exceptions, ignored the fact that bare wh-phrases are systematically ambiguous and concentrated almost exclusively on what Ginzburg and Sag () dubbed direct uses of sluicing. Fernández et al. () propose the existence of a four-way ambiguity, an ambiguity they demonstrate to be reliably coded by human subjects: • Direct The utterer of the sluice understands the antecedent of the sluice without difficulty. The sluice queries for additional information that was explicitly or implicitly quantified away in the previous utterance.15 () a. A: Can I have some toast please? B: Which sort? [BNC, KCH, –] b. Caroline: I’m leaving this school. Lyne: When? [BNC, KP, ] • Reprise The utterer of the sluice cannot understand some aspect of the previous utterance which the previous speaker assumed as presupposed: () a. Geoffrey: What a useless fairy he was. Susan: Who? [BNC, KCT, ] b. Pat: You might find something in there actually. Carole: Where? [BNC, KBH, ] • Repetition The sluice is used to ask for repetition of the previous utterance as a whole. ()

June: Only wanted a couple weeks. Ada: What? June: Only wanted a couple weeks. [BNC, KB, ]

• Wh-anaphor The utterer of the sluice cedes the turn to the previous speaker, who has indicated his wish to answer a (possibly embedded) wh-question s/he has just uttered: ()

A: We’re gonna find poison apple and I know where that one is. B: Where? [BNC, KD, –]

‘Repetition what’ can be produced after any utterance, regardless of its content, and is indeed the highest-frequency type of sluice in the BNC; wh-anaphor uses are significantly more selective and complex in that they both presuppose that the previous utterance embeds a question and involve phonological parallelism between the fragment and the

15 In other words, further elaboration of the referents associated with a previously occurring quantifier NP or with elements which license such a content via inference (e.g., ‘Bo ate.’ implies ‘There is something Bo ate.’ and ‘There is a time t such that Bo ate at t.’).



  

wh-phrase in the prior utterance. We discuss direct and reprise sluicing in ... and ..., whereas the other two classes are analyzed in Ginzburg (). Beyond the cases of clarification requests and NSUs just discussed, other work in HPSG has been based on detailed corpus investigations, e.g. Bîlbîie () on gapping in Romanian and French, Abeillé et al. () on gapping in Romance, Abeillé et al. () on peripheral ellipsis in French, Miller and Pullum () on exophoric VPE, Miller () on pseudogapping.16

. T     

.................................................................................................................................. Overall, the default assumption in HPSG is that there are no empty morphemes or unpronounced syntactic structures. This is a consequence of a general principle of economy (Occam’s razor, cf. Miller ). See Arnold and Spencer () for recent explicit argumentation against empty morphemes and Müller (a: –) for extensive discussion of HPSG analyses and a comparison with other grammatical frameworks. For elliptical constructions this means that, in the absence of strong evidence to the contrary position, there should be no syntactic structure at the ellipsis site, as mentioned above for argument ellipsis and NSUs in section ... Similarly in long-distance filler-gap constructions, it is assumed that there is no empty syntactic node at the foot of the construction (i.e. there is no “trace”; cf. Sag and Fodor ).17 On the other hand, HPSG does not make the assumption that all forms of ellipsis should necessarily be treated by a single formal mechanism, so that unpronounced structure can be assumed if warranted in a given case, as was argued for RNR in section ... With respect to unpronounced syntactic structure, HPSG is in clear disagreement with much mainstream generative syntax. Numerous publications have argued in detail that such unpronounced structure is necessary to provide a motivated explanation for various constraints on ellipsis. It is thus important in the context of this chapter to review the central types of evidence that have been brought to bear on this question and to present the counterarguments that have been advanced on the HPSG side (in line with numerous other studies, see e.g. Chapters , , ,  of this volume). We will discuss connectivity, locality effects, and the accessiblity of missing referents in turn. 16 Nielsen (); Bos and Spenader () on VPE and Beecher () on sluicing also show how corpus studies allow one to bring to light new theoretically important phenomena. Though these studies are not set in an HPSG framework, their assumptions are largely compatible. 17 Though not assuming empty morphemes is the default position in HPSG, various analyses involving such morphemes have been proposed. Specifically, the analysis of filler-gap constructions in Pollard and Sag () did involve a trace, and, as mentioned in n. , several analyses of nominal ellipsis have involved hypothesizing an empty nominal head. Similarly, one could analyze VPE or nominal argument ellipsis as involving an empty VP or NP position in the syntax, rather than no position at all: instead of having a situation where an element present in the argument structure is missing on one of the valence lists (as proposed in section ...), it is possible to include the element in the valence list, while specifying that it will have an empty phonological realization.

  -   



.. Connectivity effects Since Ross’s and Morgan’s seminal papers on sluicing and short answers (Ross b; Morgan ) it has been known that fragment answers and sluices exhibit connectivity effects, i.e., fragments typically appear in the form that would be theirs in the corresponding understood full clause. Various types of connectivity can be distinguished. In languages with morphological case, NP fragments typically appear in the case that they would have in the understood complete sentence, as in the classical examples () from Ross (b). Similarly, PP fragments appear with the preposition that they would have, as in (). () a. Er will jemandem schmeicheln, aber sie wissen nicht wem/#wen. he wants someone. flatter, but they know not who./#who. ‘He wants to flatter someone, but they don’t know whom.’ b. Er will jemanden loben, aber sie wissen nicht wen/#wem. he wants someone. praise, but they know not who./#who. ‘He wants to praise someone, but they don’t know whom.’ ()

a. I gave it away. I do not know to whom. (COCA) [cf. #with whom.] b. He seemed to be competing with someone, though we never knew with whom. (COCA) [cf. #to whom]

Along the same lines, the form of pronoun fragments is frequently the one you would find in the corresponding complete sentence and thus appears to be explained by the usual application of the binding theory: ()

Who did Pat see in the mirror? Herself/#Her. [cf. Pat saw herself/*her in the mirror.]

The apparent generalization is that fragments appear in a form that would be appropriate in the corresponding understood full sentence. On the face of it, this might seem to be a convincing argument for hidden syntactic structure, since the presence of such structure would explain why the fragment has the form it has, rather than some other. However, a closer look at the data shows the limits of the putative generalization and makes such an analysis less plausible than it initially seems. First, there are numerous cases where fragments appear in a form that would be impossible in a complete sentence. Specifically, in English, only pronouns have a nominative/accusative distinction. Possessive case, on the other hand, appears both with pronouns and with NPs (realized as an NP final ’s). Possessives exhibit typical connectivity effects as in (a). However, in all cases where a possessive is not expected, it is the accusative pronouns that appear in elliptical utterances, not only as expected by connectivity in (b), but also when a nominative would be necessary in a full clause, as in (c), and when there is no obvious full clausal structure to be recovered, as in (d). ()

a. Whose book is that? Kim’s/#Kim/Mine/#Me/#I. [cf. That is Kim’s/#Kim/mine/#me/#I] b. Who did Sandy see? Kim/Me/#I/#Mine. [cf. Sandy saw Kim/me/*I/*mine]



   c. Who saw Sandy? Kim/Me/#I/#Mine. [cf. Kim/I/*Me/#Mine saw Sandy] d. [Choosing players for a ball game, captain gestures towards prospective players]: Me, him, her, . . . [cf. #I, #he, #she, . . . ]

Similarly, for binding effects, it turns out that pronouns often appear in forms that would not be licensed in a full clause. () a. A: Who will punish Bill? B: He himselfi/Himselfi/#Hei/#Himi (#He himselfi/*Himselfi will punish Billi if he fails). (Ginzburg and Sag :  (a)) b. A: What caused the computer to break down? B: A power surge? A: Perhaps, but the most intriguing answer is: [the computer itself]i/itselfi/#it. (#The computer itselfi/*Itselfi caused the computeri to break down.) c. A: Who appeared to be the cause of John and Mary’s problems? B: Each other. (*Each other appeared to be the cause of John and Mary’s problems.) It thus appears that although connectivity is a common situation, there are various cases where fragments could not be reintegrated as such into a well-formed complete clause. Because of the non-modularity of the sign, HPSG has no problem in accounting for connectivity effects without reference to hidden structure. In section .., we sketch an account of parallelism in NSUs originating in Ginzburg and Sag (), which amplifies QUDs to include a restricted amount of non-semantic information. Syntactic parallelism is obtained by imposing matching conditions between a fragment and a prior utterance. Thus, prior syntactic structure is not used in constructing the content of a fragmentary utterance; it functions merely in establishing the appropriateness of the fragment in the given context (hence the # judgment on the unacceptable like (), (), (), etc., which we claim are inappropriate in the discourse context provided, rather than ungrammatical). In embedding the antecedent information within QUD, Ginzburg and Sag (: ) make strong predictions about the extent of categorial parallelism. Specifically, only properties of the maximal element of QUD (MAX-QUD) can trigger syntactic or phonological parallelism.18 On the other hand, examples of non-connectivity such as () and () raise problems for any analysis that wishes to explain the cases where connectivity holds in terms of unpronounced syntactic structure. They require various ad hoc assumptions to account for the absence of the expected connectivity effects. For detailed exemplification, see e.g. Ginzburg (, ); Stainton (b); Nykiel (). Furthermore, because deletion or reconstruction-based analyses usually do not offer an explicit theory of discourse context,

18

What of the putative cross-linguistic generalization known as P-stranding relevant both to sluicing and short answers? For an excellent discussion of the empirical and theoretical status of this generalization see Jacobson (b). Building on earlier work by inter alia Sag and Nykiel (), Jacobson points to the significant empirical unclarity the generalization faces. Jacobson sketches a potential account of certain facts related to the generalization available to non-deletion-based accounts such as Ginzburg and Sag () and her own Categorial Grammar approach. She further points out challenges faced by both deletion-based and surfacist accounts in providing a complete account of the data.

  -   



even in those cases where connectivity does hold, they almost invariably fail to offer an explicit account of how the reconstructed syntactic material is accessed in the context and introduced into syntactic structure. This is true even in cases where the antecedent is anaphorically retrieved, but is even more problematic in exophoric cases.19 It should also be noted that cases where connectivity holds (as it typically does crosslinguistically with case morphology, for instance) raise the symmetrically inverse challenge for analyses whose architecture enforces the assumption that the context used for interpreting fragments is purely semantic e.g. higher-order unification accounts (Pulman ) and Categorial Grammar (see, e.g., Jacobson’s contribution in this volume). Jacobson (b) opts for an account that closely resembles the one proposed by Ginzburg and Sag () for short answers, by postulating a QuAns construction which enables casematching phenomena to be captured and offers much insightful discussion of issues relating to connectivity.

.. Locality effects Locality effects have provided another classical motivation for unpronounced syntactic structure. Specifically, it has been argued that various cases of ill-formedness in elliptical constructions can be explained in terms of island violations involving such structure. Consider the following examples. Sentence (a) is claimed to be ungrammatical because it exhibits the same island violation in unpronounced structure as its non-elliptical counterpart (b), an analysis which can only be expressed if there is unpronounced structure (cf. the similar examples provided by Chung et al. : ). ()

a. *They want to hire someone who speaks a Balkan language, but I don’t remember which they do [want to hire someone who speaks]. (Merchant :  ()) b. *They want to hire someone who speaks a Balkan language, but I don’t remember which they want to hire someone who speaks. (Merchant :  (b))

However, these facts raise a number of problems. First, as is known since Ross (b), sluicing appears to absolve island violations, as illustrated by the contrast between () and (). ()

They want to hire someone who speaks a Balkan language, but I don’t remember which. (Merchant :  (a))

19 Consider an example like (a), where the second speaker’s answer Mine is assumed to have a structure like That is mine. The second speaker is assumed to be able to ellipt here through recovery of the contents of the first speaker’s question. However, the question of exactly how this content is recovered is not addressed, presumably being considered to be a problem of performance that is irrelevant to a theory of competence. But this makes it impossible to address the question of constraints on the recovery mechanism, which we have just suggested are linked to the accessibility of MAX-QUD. The question becomes even more difficult in a case like (d) where it is unclear how the speaker is expected to recover a specific syntactic structure from the exophoric context.



  

Second, as shown in (), there are cases where what would be putative island violations in unpronounced structure do not lead to ungrammaticality (though some speakers may find acceptability slightly reduced, their status is not at all comparable to that of (a)):20 ()

a. He managed to find someone who speaks a Romance language, but a Germanic language, he didn’t [manage to find someone who speaks t]. b. He was able to find a bakery where they make good baguette, but croissants, he couldn’t [find a bakery where they make good t].

In sum, the factors that constrain the interaction of wh-movement and VPE are at present not completely understood and it is hard to see how a strong argument for unpronounced structure can be made from these facts. Similar arguments have been made, for instance, for Bare Argument Ellipsis (see, e.g., Merchant a) and gapping (see, e.g., Coppock ), purporting to show that they are affected by similar island violations in unpronounced structure. However, once again, it has been argued that similar examples which should exhibit the same island violations can also be perfectly acceptable, e.g. by Culicover and Jackendoff () as well as Chapter . Thus, overall, the evidence appears to be mixed and does not at present allow one to draw strong conclusions either in favor of or against unpronounced structure. Our expectation is that a better understanding of the discourse conditions on the elliptical constructions in question and their interaction with wh-movement might lead to a more satisfactory explanation of the data.

.. The accessibility of missing referents To conclude this section, we briefly present one final argument against hidden syntactic structure which can be derived from Arnold and Borsley (), whose central point is to argue against the orphan view of non-restrictive relative clauses. Specifically, they observe that fragment answers do not allow non-restrictive relative clauses connected with part of the “missing” material, whereas anaphoric it can recover the same referents despite the fact that they are not expressed. This is illustrated in (). () a. A: Who owns a dog? B: Kim, which is regrettable. 20

As pointed out by Fox and Lasnik (), there are cases where VPE interacts with wh-movement triggering unacceptability, despite the absence of island violations in the putative unpronounced structure. For example, neither the non-elliptical variant, nor the sluice in (a) raise a problem, but the VPE variant in (b) is ungrammatical: () a.

They heard a lecture about a Balkan language, but I don’t know which Balkan language (they heard a lecture about). b. *They heard a lecture about a Balkan language, but I don’t know which Balkan language they did. (Fox and Lasnik :  (), ())

See Merchant (b) for a proposal aiming to account for the ill-formedness of (b) in terms of the Max-Elide constraint which “Roughly put, [ . . . ] states that if ellipsis applies in a structure with a wh-trace, ellipsis should target the largest constituent possible” (p. ).

  -   



b. A: Who owns a dog? B: *Kim, which is a dachshund. c. A: Who owns a dog? B: Kim, and it’s a dachshund. (Arnold and Borsley : –) As shown in (a), a non-restrictive relative clause can have as its antecedent the whole proposition expressed by the fragment, but it cannot have an understood but syntactically unrealized NP as its antecedent, as (b) shows. In this, which differs from the pronoun it which can access such a referent, as evidenced in (c). They argue that this shows that non-restrictive relatives are syntactically integrated and require a syntactically represented antecedent, whereas anaphoric it simply requires that its antecedent be accessible in conceptual structure. Assuming the presence of unpronounced syntactic structure in such cases (something like: Kim owns a dog, which is a dachshund. for (b)) makes it much more difficult to distinguish the cases in a relevant way, since the unpronounced referent is syntactically accessible. On the other hand, as shown by Arnold and Borsley (), the analysis for fragments proposed by Ginzburg and Sag () (presented below in section .) immediately predicts the data.

. R

.................................................................................................................................. The idea that ellipsis requires some form of recoverability is inherent to the concept itself. The current technical debate on recoverability originates with Chomsky, who proposed a “convention to guarantee recoverability of deletion” which allows (among other cases) deletion of elements that are “otherwise represented in the sentence in a fixed position” (: –). This is the source of the position that frames recoverability as a requirement for the presence of a syntactically identical antecedent, allowing recovery of the ellipted material. Following Sag and Hankamer (), HPSG studies have assumed that recoverability is governed by discourse semantics, rather than syntax: the central condition is the availability of an appropriate antecedent in the discourse context (which combines both linguistic and non-linguistic information). Providing formal conditions on recoverability therefore requires an articulated theory of context, and in particular of conversational context, which is sketched in Ginzburg and Sag () and elaborated in detail in Ginzburg (). Of course, as discussed in the preceding section, HPSG also provides means for accounting for well-known syntactic constraints on the relationship between the elliptical material and its antecedent-trigger,21 though these are argued to be less numerous than is usually thought. Specifically, it has been argued that many of the classical cases where acceptability is reduced in the absence of a syntactically identical antecedent can in fact receive an alternative explanation, namely that the reduced acceptability is due to violations of 21 We take the term ‘antecedent-trigger’ from Cornish (), who uses it to designate the segment of text allowing one to construct the antecedent (he uses ‘antecedent’ to refer to the discourse-model representation making interpretation of the anaphor or ellipsis possible).



  

general discourse conditions (cf. Kehler ; Kertz ) or of construction-specific discourse conditions (cf. Miller and Hemforth a; Miller and Pullum ). We will discuss in turn: (i) split antecedents and syntactic mismatches; (ii) exophoric uses; and (iii) incrementality.

.. Split antecedents and syntactic mismatches A first class of cases that argue against the idea that recoverability should be based on syntactic identity are those where the previous linguistic context provides no identical antecedent-trigger (we ignore here cases of morphological mismatch known not to affect grammaticality). These include the well-known cases of split antecedents and those where there is some form of syntactic mismatch between the antecedent-trigger and the ellipted material. We will illustrate these in turn. VPE with split antecedents, first discussed by Webber (: –), is illustrated in (), where they hadn’t is understood as meaning ‘Willem hadn’t stayed in Wyoming and become a ranch hand himself and Jude hadn’t wound up in prison, or in a hospital, or dead, or worse.’: ()

But then again, he[=Willem] would think, what about his life—and about Jude’s life, too—wasn’t it a miracle? He should have stayed in Wyoming, he should have been a ranch hand himself. Jude should have wound up where? In prison, or in a hospital, or dead, or worse. But they hadn’t. (Hanya Yanagihara, A Little Life, p. )

In fact, split antecedents are a special case of a more general situation where the antecedent is recoverable from the contents of the previous discourse without actually being present in it in any appropriate form. Consider the following examples: () a. Caleb is out running, as he does every morning. (Hanya Yanagihara, A Little Life, p. ) b. Malcolm had built them a set of bookcases, which had partitioned their squish of a living room into such a meager sliver that when you were sitting on the sofa and stretched your legs out, you stretched them into the bookcase itself. But he [=Jude] had wanted the shelves, and Willem had said he could. (Hanya Yanagihara, A Little Life, p. ) c. ‘Thanks for the ride,’ Dicey said. ‘Any time,’ Jeff answered. ‘[ . . . ] I was hoping to meet your sister who sings.’ ‘We’re in a hurry for dinner tomorrow,’ Sammy [=Dicey’s brother] said. ‘Because it’s Thanksgiving. [ . . . ]’ ‘Maybe another day,’ Dicey said. ‘You mean that?’ he [=Jeff ] asked. [ . . . ] ‘Sure,’ she said. ‘Because I will,’ he [=Jeff ] warned her. ‘Good,’ she answered, puzzled and amused. (Cynthia Voigt, Dicey’s Song, p. ) In (a), the antecedent ‘goes running’ is inferrable from the previous sentence. Similarly, in (b), the antecedent ‘have the shelves’ is inferrable from ‘wanted the shelves’. As for (c),

  -   



the previous context makes it clear that the elliptical clause is intended to mean something like ‘I will come back to meet your sister who sings’. Again, this content appears nowhere as such in the conversational context; but it can be inferred from the dialogue as a whole.22 Similarly, in the second instance of sluicing in (), the intended antecedent is recovered by inference from the discourse context. ()

She always knew exactly who was in and who was missing; and if you were missing, Madame probably knew why, and with whom. (Neil Bartlett, Ready to Catch Him Should He Fall) [= knew why you were missing and with whom you had left, rather than with whom you were missing]

Cases of syntactic mismatch between the antecedent and the elliptical sentence (e.g., voice mismatches, category mismatches) have been extensively discussed in the literature, especially for the cases of VPE and sluicing (see, among many others, Hardt ; Merchant , d; Kehler ; Kertz ). For sluicing, the great majority of the literature has agreed that mismatches are impossible. For VPE, the situation is much more complex because of the fact that there is a wide range of different acceptabilities for different cases of VPE with mismatched antecedents, ranging from the perfectly acceptable to the very unacceptable. This has led to two different positions. Either one assumes that the syntax disallows mismatches, in which case acceptable cases of mismatch are analyzed as ungrammatical but repaired. This position has been defended by Lyn Frazier and her colleagues in a series of studies (see, e.g., Arregui et al. ; Grant et al. ). The other is to assume that mismatches are always grammatical, and that unacceptable instances arise from the violation of independent discourse constraints (e.g., Kehler ; Kertz ).23 Within this debate, HPSG studies have followed the lead of Kehler and Kertz, but have focused on developing construction-specific discourse constraints. For VPE, Miller (a), Miller and Hemforth (a), and Miller and Pullum () distinguish two subtypes of non-comparative VPE, which they call Auxiliary Choice and Subject Choice respectively, and propose the following constraints.

Type : Auxiliary Choice VPE F : The subject of the antecedent is identical with the subject of the VPE construction and the auxiliary is (at least weakly) stressed, signaling a new choice of tense, aspect, modality, or (in the most overwhelmingly frequent case) polarity. D : A choice between the members of a jointly exhaustive set of alternative situations must be highly salient in the discourse context, and the point of the utterance containing the VPE is limited to selecting one member of that set.

22

Elbourne () proposes an analysis of split antecedents according to which the antecedent of the ellipted VP is constructed from the two VPs present in the discourse. Though his proposals would work for classical split antecedents such as (), it is unclear how they could deal with the examples in (). 23 Kim et al. () develop an alternate proposal suggesting that the reduced acceptability of mismatches can be attributed to processing difficulties. Miller and Hemforth (a) point out that this is compatible with the proposals of Kehler and Kertz, and suggest that discourse factors and processing factors can cumulatively reduce acceptability.



  

Type : Subject Choice F : The subject of the antecedent is distinct from the subject of the VPE construction, and stressed if it is a pronoun. D : A particular property must be highly salient in the discourse context, and the point of the utterance containing the VPE must be limited to identifying something or someone possessing that property.24 Importantly, they point out that when these discourse constraints are satisfied, VP anaphors like do it are dispreferred, even with mismatched antecedents. This is illustrated in (a) for an Aux-Choice case. (b) provides a Subj-Choice case. ()

a. A—Does Bo eat meat? B—She does. [vs #She does it.] b. A—Bo eats meat. B—Kim does too. [vs #Kim does it too.]

For VPE with nominal antecedents, Miller and Hemforth (a) show how these discourse constraints can explain the unacceptability of most cases. Specifically, for the case of Aux-Choice VPE, nouns are typically incapable of making an alternative salient. However, they point out the existence of a small class of ‘polar nouns’ that can be interpreted as concealed polar questions, as is the case for survival in (a), which can be paraphrased by an indirect polar interrogative as shown in (b). ()

a. Mubarak’s survival is impossible to predict and, even if he does [survive], his plan to make his son his heir apparent is now in serious jeopardy. [COCA: CBS Evening News] b. Whether or not Mubarak will survive is impossible to predict and even if he does [survive], his plan [ . . . ]

Miller and Hemforth show that when a polar noun appears in a discourse context that makes its interpretation as a concealed polar question salient, as is the case in (a), it can serve as a highly acceptable antecedent for VPE. These observations are corroborated through a series of acceptability experiments. On the other hand, NPs are incapable of making open propositions salient, so that Subj-Choice VPE is never acceptable with a nominal antecedent. Similarly, though no detailed study has been conducted, we expect that the discourse conditions on sluicing (namely that it requires that a quantified proposition be at issue, cf. n.  and section ...) can account for the fact that voice mismatches are unacceptable. On the other hand, examples such as (), due to Beecher (), illustrate the possibility of sluicing with a nominal antecedent. In both cases, there exists an entailment of the requisite quantified proposition and, despite the lack of a clausal (verbal) antecedent, the sluice seems acceptable, as expected from an account based on the above discourse condition:

24 Note that asking a polar question and asserting a proposition are central ways of making a polar alternative salient (see section ..), whereas asking a wh-question is a central means of establishing the salience of a property. Although we will not attempt to formalize these discourse constraints in this chapter, we note that this could be done quite straightforwardly in terms of (Maximal)-Question Under Discussion, using the KoS framework developed in section ..

  -    ()



a. We’re on to the semi-finals, though I don’t know who against. (Beecher :  (a)) b. The only thing I can come up with is contamination but I do not know what from. (Beecher :  (a))

.. Exophoricity Let us now consider exophoric uses. In these, there simply is no linguistic antecedent at all. Examples of exophoric VPE are provided in (). ()

a. [Entering a construction site, A hands a helmet to B] Do I have to? [compare: #Do I have to do it?] (Miller and Pullum :  (a)) b. When the rain began on Thursday, I simply had the kids throw on slickers and use plastic grocery bags to cover backpacks. I rode with them to school, for solidarity, but when we pulled up, an upperclassman looked at them, then me, and said halfaccusingly, ‘How could you?’ Our family experiment had suddenly gone from cool and hip to strange and compulsive. [compare: #How could you do it?] (COCA, Miller and Pullum :  ())

Miller and Pullum () argue that exophoric VPE is in fact freely available syntactically (contrary to Hankamer () who claims that the rare cases of exophoric VPE are idioms) but that it is relatively rare because the discourse conditions on VPE are hard to satisfy through extralinguistic context. Specifically, for Aux-Choice VPE, they argue that, in general, it is difficult for non-linguistic context to make an alternative salient. The most usual means for doing so are polar questions and assertions (see section .), but these typically cannot be performed non-linguistically. Thus, only less frequent strategies for establishing an alternative as the QUD can give rise to exophoric VPE. Directives, for example, can make an alternative salient (whether or not to comply), whether they are performed linguistically or deictically, as in (a). Similarly uttering a reproach, as in (b), forces accommodation of the alternative between doing or not doing the incriminated behavior, satisfying the discourse conditions. Crucially, they point out that when the discourse conditions on VPE are satisfied, do it is less felicitous, as shown in the variants of (). On the other hand, they suggest that it is almost impossible for non-linguistic context to make an open proposition salient, so that exophoric Subj-Choice cases are almost impossible. Similarly, though exophoric sluicing is rare in corpora, it is clearly attested, as shown by examples like (). ()

a. She had reached the head of the line. Her eye caught a name on the list, and she made a snap decision. When the ticket seller said ‘Where to?’ she answered, ‘Gmintagad.’ (COCA) b. ‘Bonsoir,’ he said. ‘How may I help you?’ [ . . . ] Sophie [ . . . ] simply laid the gold key on the counter in front of the man. The man glanced down and immediately stood straighter. ‘Of course. Your elevator is at the end of the hall. I will alert someone that you are on your way.’ Sophie nodded and took her key back. ‘Which floor?’ The man gave her an odd look. ‘Your key instructs the elevator which floor.’ She smiled. ‘Ah, yes.’ (COCA)



   c. In Treatment, But In Which Language? [title of article, New York Times, March , ; intended interpretation: in (psychotherapeutical) treatment, but in which language should the treatment be taking place?]

Once again, we assume that the relative rarity of exophoric sluices is due to the discourse conditions on the use of the construction: sluicing requires that a quantified proposition be at issue (see discussion of example () and Ginzburg : section .).25 This is difficult to construct non-linguistically, in a way parallel to what is proposed for VPE. Another class of exophoric ellipsis is declarative fragments tied to various interactional genres, as in (): () a. (Buying a train ticket): Client: A return to Newcastle please. (= I want a return . . . , please give me a return . . . , . . . ) b. (Buying in a boulangerie): Baker: Two euros, forty cents. (= This costs two euros, forty cents. . . . , please pay me two euros, forty cents. . . . , . . . ) Finally, we note here a class of non-sentential utterances, for which exophoric resolution is close to being the rule. This involves the little-discussed phenomenon of exclamative sluices (Ginzburg ), exemplified in (): ()

a. It makes people ‘easy to control and easy to handle,’ he said, ‘but, God forbid, at what a cost!’ (COCA) b. What a mess / relief / surprise / goal / jerk / view!

Example (a) illustrates a case of an anaphoric exclamative sluice that is parallel to typical interrogative sluices. As for examples like (b), they are by far the most frequent and are typically exophoric. This shows that any account of sluicing based on the long-standing assumption that wh-binding requires a linguistic antecedent is not viable. It is important to emphasize the real difficulty these exophoric cases pose for analyses that claim that recoverability must be stated in terms of any kind of syntactic identity (including LF-identity). It is completely unclear how such analyses can make sense of exophoric uses at all, barring the assumption that a speaker constructs a specific syntactic representation of the relevant extralinguistic properties, which could then be used as an ‘antecedent’. However, this idea defeats the basic purpose of the idea of ‘recoverability’ of deletions, as initially proposed by Chomsky, since it then becomes unclear why there are any constraints at all on exophoric uses.26

In section .., we analyze the case of sluices of the form ‘Who else?’, which are compatible with a referential antecedent ‘X V’ed’. Nonetheless, their acceptability clearly requires the issue ‘Somebody else apart from X V’ed’ to be worthy of discussion. 26 Merchant () has the most developed discussion of exophoric uses from an unpronounced structure perspective that we are aware of. He proposes that examples of NSUs like (a), (b), and () should be analyzed using the concept of scripts (in the sense of Schank and Abelson ). It is not clear, however, whether an example like (c) is amenable to such an analysis. And it is even less clear how scripts could account for the wide range of exophoric VPE cases reported in Miller and Pullum () and illustrated here in (). 25

  -   



.. Incrementality An additional argument against viewing a construction such as sluicing as S-deletion (or similar) is that it is possible as soon as its nominal antecedent has been introduced, as in (); it need not occur as part of a completed sentential construction. This can occur either in monological examples such as (a) or in dialogical cases such as (b) and (c). Once one adopts an incremental view of semantic processing, accommodating such data becomes straightforward, as we will show in section ..., based on Ginzburg et al. (). ()

a. Someone, {John? / I can’t figure who}, has taken the kitchen scissors. b. A: [vehemently announcing to B] I’m really annoyed. B: Ok. A: Some sleazy type. B: {John? / Who?} A: Not sure, has taken the kitchen scissors. c. A: Somewhere. B: Where? A: Not far from here. B: Aha. A: Someone is enjoying the cheese I forgot on the train.

. L

.................................................................................................................................. As has been mentioned above and as will be developed in much more detail in section ., the HPSG approach to ellipsis resolution makes central use of the notion of Question Under Discussion (QUD), which is augmented to include restricted amounts of non-semantic information. In a sense then, HPSG assumes that QUD is the central factor in the licensing of ellipsis. This makes sense under a more general view of anaphora resolution involving some notion of accessibility or givenness (see, e.g., Ariel ; Gundel et al. ), as it is known that, in general, phonologically weaker anaphors require more accessible antecedents. Since ellipsis can be seen as the phonologically weakest form of anaphor, it requires antecedents that are most accessible, namely, antecedents involved in the most salient QUD at any moment in the discourse. This feature of the model thus makes strong general predictions about the syntax, phonology, semantics, and pragmatics of elliptical constructions, namely that only properties of the most salient QUD can be relevant in determining their form and interpretation. Beyond this, HPSG does not assume that there is a single mechanism underlying ellipsis. Rather, elliptical constructions each have their own specific licensing conditions. To the extent that generalizations can be made, they can be captured through the hierarchy of types. For example, it is known that different subtypes of VPE have subtly different properties (see, e.g., Miller a; Kertz ). The general properties of VPE can be attributed to the construction-type VPE, whereas the various subconstructions can be specified for their specific properties, while inheriting the general properties from the more general construction. Similarly, it would be possible to represent the generalizations available for different types of argument ellipsis (see section ... and .), including VPE, through the type hierarchy, and more generally to associate the general properties of ellipsis involving QUD to an overarching ellipsis-type, though to our knowledge this has not yet been attempted. In sum, the basic methodological assumption of HPSG is that one should not move too quickly in assuming that properties of constructions are generalizable. The complexity of the varieties of subconstructions must first be established in as much detail as possible before meaningful generalizations can be made.



  

Throughout this chapter, the discussion of the various elliptical constructions highlights their specific licensing properties. We have distinguished three central cases: (i) certain constructions license non-sentential constituents as complete utterances (e.g. Bare Argument Ellipsis, sluicing, gapping); (ii) certain constructions license non-canonical correspondence between ARG-ST and valence features (e.g., null subjects and objects, VPE, NCA); (iii) certain constructions license a non-canonical correspondence between the PHON of a mother and its daughters (e.g., peripheral ellipsis, RNR). More specific constructions involve more specific licensing conditions, e.g. VPE is licensed by the presence of an auxiliary in English, whereas sluicing requires a wh-phrase.

. T    

.................................................................................................................................. In this section we provide a more explicit discussion of the constructional analysis of argument ellipsis which was sketched in section .... As mentioned there, the crucial insight underlying the analysis is that there is a non-canonical correspondence between the argument structure of a head and its valence. More specifically, the value of the ARG-ST feature is not simply the concatenation of the values of the valence features SPR and COMPS.

.. Phrase structure and valence features in HPSG In HPSG, the combination of heads with their sisters is governed by various rules, two of which are crucially relevant in the present context. The H-C R combines the head (H in (a)) with its complements (  to n in (a)). The resulting phrase inherits the HEAD value of its head-daughter, but crucially, not the value of the COMPS feature whose specification has been satisfied, so that it has an empty list as the value for COMPS. Similarly, the H-S R combines the head with its specifier (  in (b)), satisfying the valence requirement. ()

a. Head-Complement Rule phrase

→ H word 1 , ..., n

VAL

VAL

COMPS

COMPS

1 , ..., n

b. Head-Specifier Rule phrase VAL

→ 1 SPR

H

SPR

1

VAL COMPS

This is illustrated in tree (). In order to make these rules and the tree fully clear, we need to introduce the way S S works in HPSG. It is indicated by boxed numbers. Putting the same boxed number in several places in a feature structure or rule, indicates that the corresponding values are token identical. Thus, in tree (), the  expresses the fact that

  -   



the feature structure which is the sole element in the value of the COMPS list of saw is token identical to the feature structure describing the NP Pat. This means that it would be entirely redundant to replace ‘COMPS〈  〉’ by ‘COMPS 〈  NP〉’ (of course, the tree would have expressed exactly the same property if we had had the latter above saw and only the former above Pat). Similarly, in (a), the shared boxed numbers indicate that the feature structures describing the items on the COMPS list and those describing the corresponding daughters are token identical.27 ()

S SPR COMPS

1 NP

Kim

1

VP SPR COMPS

1

V SPR COMPS ARG-ST

2 NP

2 1,2 Pat

saw

Similarly, tree () provides an example with the auxiliary will. ()

S SPR COMPS

1 NP

Kim

VP SPR COMPS

V[aux] SPR COMPS ARG-ST will

1 2 1,2

1

2 VP[base ] SPR COMPS

V[base ] SPR COMPS ARG-ST see

27

1

1

3 NP

3 1,3 Pat

Signs in HPSG can be lexical or phrasal. In the latter case, formally speaking, the constituent structure is modeled within the feature structures using the features MOTHER, HEAD-DTR (HeadDaughter), and NON-HEAD-DTRS (Non-Head-Daughters). Thus tree representations of the type given in () are an informal simplification enhancing readability.



  

.. The Argument Realization Principle and ellipsis Given this set-up, we can provide an analysis of the ellipsis of complements or specifiers simply by allowing non-canonical constructions for heads, in which items mentioned in the ARG-ST do not appear in the valence features (such constructions are non-canonical in the sense that they do not respect the general formulation of the A R P provided in ()). For instance, we can account for verb phrase ellipsis by constraint (a) which allows for auxiliaries with a non-canonical relation between the COMPS and ARG-ST lists. Constraint (a) will be inherited by will (since it is an auxiliary), providing it with the alternate COMPS list given in (b), and thus giving rise to a structure like () (see, e.g., Kim ).28 ()

a. aux:

SPR

1

COMPS

b. will:

ARG-ST

1 ,VP[pro]

SPR

1 NP[nom]

COMPS ARG-ST

1 , VP[pro, base]

S SPR COMPS

()

1 NP

Kim

VP SPR COMPS

V[aux] SPR COMPS ARG-ST

1

1 1 , 2 VP[pro,base]

will

28

Note that the properties [nom] for the NP subject and [base] for the VP are respectively inherited from the basic lexical constraints for auxiliaries in general and for modal auxiliaries in particular. Note also that some recent versions of HPSG would not have the non-branching VP which we have used for simplicity in () (a similar comment can be made for the non-branching S node in ()).

  -   



Since (a) applies to all auxiliaries, we correctly express the classical generalization that VPE is systematically available behind auxiliaries. Specifying that the VP argument of preelliptical auxiliaries is pronominal essentially captures the intuition of Schachter () and means that it will be interpreted through general principles for resolving anaphoric and/or exophoric dependencies (cf. e.g. Hardt ).29 In the light of this example, we can state the A R P as follows: ()

Argument Realization Principle

word ⇒ VALENCE

ARG-ST

SPR

1

COMPS

2

1 ⊕ 2 ⊕ list (non-canonical)

In (), ⊕ is the list concatenation operator. For the purposes of this chapter, noncanonical items on the ARG-ST list will always correspond to null pronominals.30 We can now more precisely indicate the analysis for null pronominal subjects and objects, as in the Spanish and Brazilian Portuguese examples () and (), repeated here. () Maria escolheu o livro e [[ Ana] [[ colocou] [[ na] in Maria chose the book and Ana put ‘Maria chose the book and Ana put it in the bookcase.’ () Maria llegó. [[[ Estaba] Was Maria arrived ‘Maria arrived. She was tired.’

[ estante]]]. bookcase

[ cansada]]]. tired

The constraint in (a) allows a pronominal subject argument to appear on the ARG-ST list but not on the SPR list, giving rise to () as the analysis for (). (b), on the other hand, allows for a pronominal object argument that does not appear on the COMPS list, and provides () with an analysis similar to that given for will in ().31 29 Further work on VPE in HPSG includes Grover et al. () on strict/sloppy ambiguity; Lappin () and Gregory and Lappin () on antecedent-contained deletion; Egg and Erk () and Arnold and Borsley () on auxiliary-stranding relative clauses; and Lee (). For an early HPSG analysis of pseudogapping as a case of VPE where the pre-elliptical auxiliary selects a complement, see Miller (). 30 Other non-canonical items include extracted elements and certain types of clitics; see, e.g., Miller and Sag (). 31 Brazilian Portuguese allows Ana colocou ele na estante, with the overt strong object pronoun ele, as a variant of (). These two sentences will receive exactly the same analysis with respect to syntax and semantics, except that the pronominal NP will appear both on the ARG-ST and on the COMPS list in the latter case, causing the object pronoun to be overtly realized. The intended anaphoric link to the object of the previous sentence in the context provided is mediated through the ARG-ST list, which is the same in both cases, so that anaphoric resolution works similarly. Note however that the details of the discourse pragmatics will be different, so that constraint (b) will also have to impose additional properties at that level.



  

()

a. V:

b. V:

SPR COMPS

b

ARG-ST

NP[non-can, ppro, nom] ⊕ b

SPR

a

COMPS

b

ARG-ST

a ⊕ NP[non-can, ppro, acc]

⊕ b

In (a),  is the COMPS list which is concatenated to a list consisting of a non-canonical nominative personal pronoun NP in the ARG-ST. In (b), this non-canonical argument is missing in the COMPS list (list  may of course be empty). ()

S SPR COMPS

VP SPR COMPS V SPR COMPS ARG-ST

2 AP[fem, 3sg] 2 NP[non-canon, ppro, nom, fem, 3sg], 2 estaba

cansada

. HPSG      

.................................................................................................................................. Dialogue is the primary setting for language, and elliptical language, where content is resolved largely based on context, is particularly characteristic of this setting. In this section we offer an explicit model of dialogue context and show how this can be used in the analysis of various elliptical constructions. In so doing we also introduce an alternative version of HPSG, using the formalism of Type Theory with Records. As we argue in .., this version

  -   



of HPSG, HPSGTTR—or at least its defining characteristics—is needed for an analysis that strives to deal with the challenges of spoken language.32

.. A dialogical working example We will illustrate how HPSG analyzes ellipsis constructions with reference to the constructed example in (). A poses a wh-question. This triggers clarification interaction concerning the intended reference of the NP ‘Bo’ via a reprise fragment (see section ...); B’s question in () is answered via the short answer in () (see section ...); the original question posed in (), three turns earlier, is then the antecedent for the VPE utterance in (); via inference, () triggers a (direct) sluice in () (see section ...). B responds with the short answer in (). This, in turn, triggers the VPE query in (), which exemplifies a self-correction utterance (see section ...). This gets answered by the polar particle ‘no’ (see section ...). () A(): Who visited Bo? B(): Bo? A(): My cousin. B(): Jack did. A(): Who else? (#Whom else?) B(): No one. A(): Didn’t Bill, I mean Mary? B(): No. () exemplifies various claims we made earlier: • The role of inference in ellipsis resolution (e.g., in () resolution is driven by B’s inability to resolve the reference of ‘Bo’ in ()—there is no salient linguistic content around to provide the content ‘Who do you mean “Bo”?’ for this NSU; similarly, in () the resolution emerges from the possibility that Jack does not constitute the sole answer to ()). • The need for resolution to operate across turns (virtually all cases), in some cases multiple turns (e.g., ()). • The need to project structural (categorial and phonological) information into context in order to capture cross-turn parallelism of varying degrees (e.g., segmental identity in (), categorial identity in ()). • The interleaving of self- and other-repair in ellipsis resolution (e.g., () and ()). The account we propose is based on two main components. First, independently motivated processes of dialogue context dynamics—these are described in section .., with a brief sketch of the presupposed semantic framework in section ... Second, constructional/lexical specification that can interface directly with dialogue context that contains both linguistic and non-linguistic information. These are sketched in section ... The two components are combined in an analysis of the above dialogue in section .., which concludes the chapter. 32

More generally, Ginzburg and Poesio () propose constraints that any formalism striving to deal with spoken language needs to satisfy.



  

.. The dialogue gameboard Our account of dialogue context follows that developed within KoS (cf. n.  and Ginzburg ; Larsson ; Ginzburg and Cooper ; Fernandéz ; Purver ; Ginzburg and Fernandez ; Ginzburg ).33 KoS is a theory that combines an approach to semantics inspired by situation semantics and dynamic semantics with a view of interaction influenced by Conversational Analysis. On the approach developed in KoS, there is actually no single context—rather, analysis is formulated at a level of information states, one per conversational participant. Each information state consists of two ‘parts’, a private part and the dialogue gameboard, inspired by Lewis (), that represents information that arises from public interactions. For recent psycholinguistic evidence supporting this partition see, for instance, Brown-Schmidt et al. (). The structure of the dialogue gameboard (DGB) is given in Table .. The Spkr and Addr fields allow one to track turn ownership; Facts represents conversationally shared assumptions; VisualSit represents the dialogue participant’s view of the visual situation and attended entities; Pending represents moves that are in the process of being grounded and Moves represents moves that have been grounded; QUD tracks the questions currently under discussion, though not simply questions qua semantic objects, but pairs of entities which we call InfoStrucs: a question and an antecedent sub-utterance; motivation for this view of QUD is given below.

Table . Dialogue gameboard Dialogue Gameboard Component

Type

keeps track of

Spkr Addr utt-time

Individual Individual Time

Turn ownership

Facts

Set(propositions)

Shared assumptions

VisualSit

Situation

Visual scene

Moves

List(Locutionary propositions)

Grounded utterances

QUD

Partially ordered set(⟨question, FEC⟩)

Live issues

Pending

List(Locutionary propositions)

Ungrounded utterances

In terms of the DGB, the model of context in Sag and Hankamer () had some version of Facts (a mental model as in Johnson-Laird  or situational model like Zwaan and Radvansky ) and of Moves (a ‘propositional textbase’; see Fletcher ). The DGB, then, goes beyond such a view of context in at least two significant ways which derive 33

For an alternative view on how to integrate pragmatic information from context in HPSG, see Bertomeu and Kordoni ().

  -   



from the fact that it is intended as a contextual resource for dialogical interaction, in contrast to earlier models of context, intended to process text or monologue. The two fundamental innovations relate to QUD and Pending. Questions get introduced into QUD by a number of processes. These include: querying (asking q makes the question q QUD-maximal); assertion (asserting p makes the (polar) question p? QUD-maximal); accommodation triggered by clarification interaction (e.g., if A’s sub-utterance u is difficult to resolve or involves an error, the issue ‘what did A mean by u?’ can become QUD-maximal); accommodation triggered by interaction in a conversational genre (e.g., in a customer/client interaction, the issue ‘what does the client require?’ can become QUD-maximal). Adopting the assumption that (structural) parallelism typically exhibits a similar time course to the salience of the relevant entity of QUD, we can capture such effects by viewing QUD as tracking not simply questions qua semantic objects, but pairs of entities: a question and an antecedent sub-utterance. This latter entity provides a partial specification of the focal (sub-)utterance, and hence it is dubbed the Focus Establishing Constituent (FEC) (cf. parallel element in higher-order unification-based approaches to ellipsis resolution (e.g. Gardent and Kohlhase ); Vallduví () relates the FEC with a notion needed to capture contrast. Thus, the FEC in the QUD associated with a wh-query will be the wh-phrase utterance, the FEC in the QUD emerging from a quantificational utterance will be the NP utterance, whereas the FEC in a QUD accommodated in a clarification context will be the sub-utterance under clarification. Pending is the contextual resource corresponding to utterances that are still in progress or under clarification. One of the key structuring aspects of conversational interaction is the ubiquitous metacommunicative interaction between the participants observable via periodic verbal and gestural backchannels, and occasionally via clarification questions of various kinds (e.g., ‘What did the speaker mean by “ . . . ” ’). This cycle of grounding (Clark ) and clarification interaction (Ginzburg and Cooper ) is also present for a given speech participant monitoring her own speech, overt evidence for which are various disfluencies such as hesitations (‘the next thing to say is problematic’) and self-corrections (‘the recent sub-utterance needs fixing’). Ginzburg () offers detailed arguments on this issue, including considerations of the phonological/syntactic parallelism exhibited between CRs (clarification requests; cf. ..) and their antecedents and the existence of CRs whose function is to request repetition of (parts of ) an utterance; see (). Taken together with the obvious need for Pending to include values for the contextual parameters specified by the utterance type, Ginzburg concludes that the type of Pending combines tokens of the utterance, its parts, and of the constituents of the content with the utterance type associated with the utterance. An entity that fits this specification is the locutionary proposition defined by the utterance: in the immediate aftermath of a speech event u, Pending gets updated with a record whose two components are u and Tu, a grammatical type for classifying u that emerges during the process of parsing u. Locutionary propositions are instances of Austinian propositions; see Barwise and Etchemendy (). The original Austinian conception was that s is a situation deictically indicated by a speaker making an assertion whose truth involves s being of type T. A locutionary proposition specializes this notion to the case of a speech event u and a grammatical type Tu, in other words, an entity such as the sign in the sense of HPSG. The relationship between u and Tu—describable in terms of the proposition:   sit ¼ u pu ¼ sittyp ¼ Tu



  

—can be utilized in providing an analysis of grounding/clarification-interaction conditions of an utterance: ()

a. Grounding: pu is true: the utterance type fully classifies the utterance token. b. Clarification interaction: pu is false, either because Tu is weak (e.g., incomplete word recognition) or because u is incompletely specified (e.g., incomplete contextual resolution—problems with reference resolution or sense disambiguation).

.. Grammar and semantics in Type Theory with Records In this section, we introduce the basic semantic and grammatical notions we use in the rest of this section. KoS is formulated within the framework of Type Theory with Records (TTR) (Cooper ; Cooper and Ginzburg ), a model-theoretic descendant of Martin-Löf Type Theory (Ranta ) and of situation semantics (Barwise and Perry ; Ginzburg and Sag ). TTR enables one to develop a semantic ontology, including entities such as events, propositions, and questions. With the same means TTR enables the construction of a grammatical ontology consisting of utterance types and tokens and of an interactional domain in which agents utilize utterances to talk about the semantic universe. Ginzburg (: ch. ) provides detailed argumentation for what makes TTR advantageous for a dialogically oriented grammar and, in particular, for the advantages of analyzing signs in terms of record types rather than typed feature structures, as done in standard HPSG (henceforth HPSGTFS) and other ‘unification-based’ frameworks. We summarize the arguments here: • Object level types and tokens: HPSGTTR provides access to both types and tokens at the object level, specifically here to utterance tokens (or speech events) and utterance types (also known as signs). This is crucial in formulating grounding/clarification-interaction conditions of an utterance, as in (), which constitutes the basic dynamic process in utterance processing. It is also a crucial ingredient in providing semantics for repair constructions (see, e.g., sections ... and ...) and for quotation (Bonami and Godard ; Ginzburg and Cooper ), which are anaphoric to utterance tokens and involve inference processes that make references to utterance types. HPSGTFS specifies a grammar in terms of types but has no corresponding means of directly referring to tokens that constitute speech events.34 34

Ginzburg () reviews the history of HPSG’s view of formalizing grammars on this score which were, with one exception, formulated either in terms of tokens or, subsequently, types. Pollard and Sag () suggested thinking about feature structures as ‘partial descriptions of signs (or sign tokens) and other linguistic objects which occur as part of signs’ (p. ). However, this was violently repudiated by Pollard and Sag (): ‘One thing that [language] certainly does not consist of is individual linguistic events or utterance tokens, for knowledge of these is not what is shared among the members of a linguistic community’ (p. ). Paul King, on the other hand, in his formalizations of HPSG (see. e.g., King , ), does view a grammar as a characterization of the class of well-formed utterance tokens. In King , he identifies a token as a pair 〈u, I〉 of an entity u and an interpretation I (in a technical sense King develops). From this, he constructs types as equivalence classes of indiscernable tokens. Whether King’s theory could

  -   



• Direct access to semantic entities: HPSGTTR directly provides semantic entities, whereas HPSGTFS simulates them. Thus, whereas HPSGTTR can effect variable binding, function definition, and abstraction, in HPSGTFS these notions are merely coded up.35 To take a concrete example: as we will shortly see, HPSGTTR represents the contextual parameters of a meaning by means of record types. A token of such a type—a record— represents a specific context for an utterance. In HPSGTFS there are ways of representing contextual parameters but there is no explicit way of modeling the fact that these need to be instantiated in context. For current purposes, the key notions of TTR are the notion of a judgment and the notion of a record. • Typing judgments A typing judgment a:T classifies an object a as being of type T. • Records A record is a set of fields assigning entities to labels of the form (a); a concrete instance is exemplified in (b).36 Records are used here to model events and states, including utterances, and dialogue gameboards. ()

a. l1 = val1 l2 = val2 … ln = valn b. temp = –28 e-time = 2AM, Feb17, 2011 e-loc = Nome

• Record Types: A record type is simply a record where each field represents a judgment rather than an assignment, as in (a). Record types are used to model utterance types (HPSG signs, as in (b)), as components of semantic entities such as propositions and questions, and to express rules of conversational interaction.37

serve to underpin linguistic description of utterances used in repair or quotation is not straightforward to discern since it is not clear that his formalism allows for types as first-class entities. 35 As Penn (: ) puts it (in discussing a related set of issues), ‘At this point, feature structures are not being used as a formal device to represent knowledge, but as a formal device to represent data structures that encode formal devices to represent knowledge’. 36 Assignment of an entity to a label is noted by ‘=’. 37 In (b) we make use of the notation [x=k: T] which is based on the notion of a manifest field (Coquand et al. ). This is a shorthand for a judgment x : Tk where Tk is the singleton subtype of T whose only witness is k. For instance, ‘ = V[+fin] : syncat’ means cat : syncat[+fin], so in such a case  is restricted to the subtype of the type syncat whose sole witness is V[+fin].



  

()

a.

l1 : T1 l2 : T2 … ln : Tn

b.

PHON : is georges here CAT = V[+fin] : syncat CONSTITS = is, georges, here, is georges here : set(sign) spkr : IND addr : IND utt-time : TIME c1 : address(spkr,addr,utt-time) DGB-PARAMS : s : SIT l : LOC g : IND c3 : Named(g, georges) sit = s cont = Ask(spkr,addr, ?

): IllocProp sit-type = In(l,g)

The basic relationship between records and record types is that a record r is of type RT if each value in r assigned to a given label li satisfies the typing constraints imposed by RT on li. To take a simple example, (a) classifies the situation in Nome (see (b)) at a certain point in time, assuming the conditions in (b) hold. ()

a. temp : Ind e-time : Time e-loc : Loc b. –28 : Ind; 3:45AM, Feb 17, 2011 : Time; Nome : Loc;

To take a more linguistic example, a conversational state r will be a record such that the conditions in () hold—it is of the type dialogue gameboards must satisfy; in other words, r should have the make-up in (b) and the constraints in (c) need to be met:

  -   



() a. r : DGBType

b. r =

spkr = A addr = B utt-time = t c-utt = putt(A,B,t)

: DGBType

FACTS = cg Moves = 〈m1,…,mk〉 QUD = Q Pending = 〈p1,…,pk〉 c. A: IND, B: IND, t: TIME, putt(A,B,t) : addressing(A,B,t); cg: Set(Proposition); 〈m1,…,mk〉 : list(locutionaryProposition); Q : poset(InfStruc); 〈p1,…,pk〉 : list(locutionaryProposition) The final two notions we need are propositions and questions. We have already mentioned the notion of an Austinian proposition. Truth conditions for Austinian propositions are defined in ():

()

A proposition p =

sit

=

s0

sit-type

=

ST0

is true iff s0 : ST0

Extensive motivation for the view of questions as propositional abstracts has been provided in Ginzburg () and Ginzburg and Sag ()—TTR contributes to this by providing an improved notion of simultaneous, restricted abstraction: A (basic, non-compound) question is a function from records into propositions. Here we assume a minor refinement of this view proposed by Ginzburg et al. (a).38 This involves introducing a notion of Austinian questions, defined as records containing a record and a function into record types, the latter associated with the label ‘abstr(act)’. The role of wh-words on this view is to specify the domains of these functions; in the case of polar questions there is no restriction, hence the function component of such a question is a constant function. () exemplifies this for a unary ‘who’ question and a polar question. For notational simplicity, we will notate questions in what follows just in terms of the associated abstract.

38 This refinement is motivated in part by the need to enable conjunction and disjunction to be defined for questions; it also enables an account of the interaction between questions and adjectives, as in ‘difficult/quick/philosophical question’.



()

   a. Who =

x1 : Ind c1 : person(x1)

b. Whether = Rec c. ‘Who runs’ ↦ sit = r1 abstr = λr:Who( c : run(r.x1) )

d. ‘Whether Bo runs’ ↦ sit = r1 abstr = λr:Whether( c : run(b) ) We combine all elements introduced in this section in a sketch of how clarification interaction can be specified. In principle, one could have a theory of clarification interaction based on generating all available CRs an utterance could give rise to. But in practice, there are simply too many to be associated in a ‘precompiled’ form with a given utterance type. Instead, CRs can be specified by means of a uniform class of conversational rules, dubbed Clarification Context Update Rules (CCURs) in Ginzburg (). Each CCUR specifies an accommodated MaxQUD (the maximal element of QUD) built up from a sub-utterance u1 of the target utterance, the maximal element of Pending (MaxPending). Common to all CCURs is a license to follow up MaxPending with an utterance which is co-propositional with MaxQUD.39 () is a simplified formulation of one CCUR, parameter identification, which allows B to raise the issue about A’s sub-utterance u0: what did A mean by u0? (‘pre’ expresses the preconditions and ‘effects’ specifies the effect of the rule; ‘MaxPending.sit.constits’ expresses a path): () Parameter identification

pre

Spkr : Ind : MaxPending : LocProp u0 ∈ MaxPending.sit.constits q = λxMean(A,u0,x) MaxQUD = fec =

effects

:

sit = u0

: InfoStruc

sit-type = Tu0 LatestMove : LocProp c1: CoProp(LatestMove.cont,MaxQUD.q)

39

Two utterances u0 and u1 are co-propositional iff the questions q0 and q1 they contribute to QUD are co-propositional. Co-propositionality for two questions q0 and q1 means that their range is not disjoint, i.e., there exists a record r such that q0(r) = q1(r).

  -   



.. Non-sentential utterance constructions The detailed theory of context sketched in previous sections enables the development of a grammar of the various types of sentential fragments discussed earlier. The basic strategy adopted in KoS/HPSGTTR to analyze NSUs is to specify construction types where the combinatorial operations integrate the denotata of the fragments with elements of the DGB. We sketch how this can be done with several such construction types; a detailed account of a wide variety of sentential fragments analyzed in such terms can be found in Fernández () and Ginzburg ().

... Propositional lexemes yes In its most straightforward use, ‘yes’ affirms the proposition p, where p? is the maximal element in QUD, either because the polar question p? was posed or as a side effect of an assertion of p. Hence, the lexical entry in (). Categorically ‘yes’ is classified here as an adverbial 40 that can only occur as a root utterance (cf. n. ), whereas semantically its content is the proposition component of a polar question:41 ()

phon : yes cat.head = adv[+IC] : syncat dgb-params.max-qud : PolQuestion cont = max-qud([]) : Prop

no ‘No’ has a number of distinct uses, including an exophoric use (on which see Cooper and Ginzburg , where a treatment of negation in TTR is discussed) and a ‘pragmatic denial’ use, for which see Tian and Ginzburg (). Here we provide a lexical entry that captures the basic fact about the most prototypical use of ‘no’ in dialogue—its content is always a negative proposition. This can arise either by negating a proposition, when MaxQUD is a positive polar question (e.g., A: Did Bo leave? B: No (= Bo did NOT leave)), or by affirming it when MaxQUD is a negative polar question (e.g., A: Bo didn’t leave? B: No (= Bo did not leave)): ()

phon : no cat.head = adv[+IC] : syncat dgb-params.max-qud : PolQuestion cont : NegProp c1: (cont = maxqud([])) ∨ (maxqud([]) : ̸ NegProp ∧ cont = ¬ maxqud([]))

40 The rationale behind this proposal is that it makes ‘yes’ resemble its counterparts in other languages (see discussion in ...), as well as other adverbials in English (‘probably’, ‘possibly’) which have a use as a propositional lexeme; for English, one might make other decisions on this score, e.g., classifying ‘yes’ as an interjection. 41 Recall that a polar question p? is, in the semantics we propose, a constant function mapping any record r to the proposition p. Hence, in particular applying p? to the empty record [] yields p.



  

... Declarative fragments Consider the declarative fragment (a): () a. B: Four croissants. b. Context: A: What did you buy in the bakery? Content: I bought four croissants in the bakery. c. Context: [A: smiles at B, who has become the next customer to be served at the bakery.] Content: I would like to buy four croissants. d. Context: A: Dad bought four crescents. Content: You mean that Dad bought four croissants. e. Declarative-fragment clause: Cont = DGB.MaxQUD(u-frag.cont) : Prop B’s utterance in (a) can receive a variety of contents, depending on the context in which it is uttered and the intonation contour it receives: it can be interpreted as a short answer, as in (b); it can be interpreted without any prior utterance, as in (c), though in such a case various paraphrases are possible, depending on the conversational genre;42 it can also be interpreted as the (‘metalinguistic’) correction in (d). The different mechanisms underlying these resolutions can be uniformly described by the schema in (e). This indicates that the content of the construction type declarativefragment clause arises by predicating the propositional function constituted by the maximal element of QUD of the content of the bare fragment utterance, a generalization of a rule proposed already in Hausser and Zaefferer (). The particular content exhibited in (b) could arise because the issue ‘What did you buy in the bakery?’ is MaxQUD as a result of A’s query; (c) arises given that the issue ‘What does the current customer want to buy?’ is a characteristic issue of the BakeryShopping genre (as it is of many related genres); the content in (d) could arise if B decided not to ground A’s utterance, but using the parameter identification conversational rule (discussed in section ...) to initiate repair interaction, accommodates the issue ‘What did you mean by utterance “four crescents”?’ as MaxQUD. We have also emphasized that different NSU constructions exhibit morphosyntactic and/or phonological parallelism with their antecedents. In other words, not only the combinatorial semantics of NSU constructions integrates information from the DGB, but this is also potentially true of the morphosyntactic and phonological specifications of such constructions. Here we utilize this to deal with connectivity: we specify that the fragment has to match the categorial specification of the FEC. In light of this, we can write a specification of decl-frag-cl as in (). Categorially the construction is sentential, as discussed in section ...; it has one DGB parameter—i.e. contextual parameter—the maximal element of QUD, whereas as just mentioned, its content arises by functional application of MaxQUD to the entity denoted by the fragment:

42

See Wittgenstein (); Clark () for discussion and exemplification.

  -    ()



cat = υ : syncat dgb-params.max-qud :

q : UnaryWhQuestion fec : LocProp

cont = max-qud.q(hd-dtr.cont.x) : Prop

hd-dtr : cat = max-qud.fec.cat : Syncat cont : x: IND

... Direct sluicing Direct sluices are analyzed in a similar fashion. In Ginzburg and Sag () declarativefragment clauses and sluices are unified as subtypes of a phrasal type dubbed hd-frag-ph. This construction builds a verbally headed phrase from an NP or PP under the constraint that the context’s FEC is categorially identical to and coindexed with it.43 Three aspects distinguish a direct sluice, specified by the clause type sluice-int-cl from a declarative-fragment clause: the sluice is constructed from a wh-phrase, the sluice clause denotes a question, and, arguably, it has a distinct contextual background. What is the contextual background of a direct sluice? Ginzburg and Sag () and Ginzburg () argued that this involved the QUD-maximality of a quantified polar question (cf. n.  and surrounding text). This seems to be accurate in most cases. There are, however, several cases where this is less persuasively so. An alternative, suggested by Ivan Sag (p.c.), is to assume that it is the corresponding wh-question. This, however, requires a somewhat more ‘inferential’ view of QUD since in many cases there will not have been a prior utterance of a wh-interrogative. Nonetheless, since it is clear that QUD needs to be updated by some restricted inferences in a number of cases, as we have already discussed, this is not a qualitative modification. Indeed, as will become clear from the discussion below, from a compositional semantics point of view, it makes little difference. Assuming the view of wh-phrase content sketched in section .. and remaining agnostic for the moment as to the QUD-specification of a sluice’s context, we can explicate how the content of a sluice arises straightforwardly, using exactly the content of the wh-phrase and MaxQUD. As summarized in (a), the sluice denotes a question (i.e., a function from records into propositions) whose domain is the type denoted by the wh-phrase and whose range is that given by MaxQUD’s proposition where the wh-phrase’s 43

Of course, in languages where connectivity properties are not as strict as in English, this constraint can be weakened. See Sag and Nykiel () and Nykiel () for an example of how this can be done for sluicing in Polish. Similar analyses are possible for languages like French or Portuguese where a wh-NP can have a PP correlate (cf. n. ). See also Kim () on sluicing in Korean.



  

variable is substituted for that associated with the FEC (‘↝’ signifies ‘reduces to by b-conversion’). ()

a. sluice-int-cl.cont = (whP.rest)MaxQUD.q.prop[fec.x ↦whP.x] b. A: A student left. B: Who? c. ⟦Who?⟧ =

z : Ind x : Ind (r :

)

c2 : student(z)

(z ↦ x)

c1 : person(x) c3 : leave(z) ↝ x : Ind x : Ind (r :

) c2 : student(x) c1 : person(x) c3 : leave(x)

We note that this view extends to two tricky cases for deletion-based accounts. First, exophoric cases like (). In such cases awareness of the genre is what leads QUD to be updated (see Ginzburg  for details). The essential idea of that proposal is that a given genre can be characterized, in part, by a partially ordered set of questions, discussion of which constitutes its defining subject matter. At appropriate points these questions can be accommodated into QUD without being uttered overtly.44 The way this account applies to declarative fragments was explained in the discussion of (c). This is what licenses exophoric sluices as in (), illustrating the taxi-cab genre. ()

a. Where does customer want to travel to? What is the price of the trip? b. Driver: [Picking up the passenger] Where to? [On arrival] Thirty euros.

Second, incremental cases like (a), already discussed in section ... Assuming an incremental view of semantic processing (Purver et al. ; Rieser and Schlangen ; Ginzburg et al. b), the sluice in (a) is predicted to mean, immediately after it is uttered, ‘Who is that person (that has some as yet uninstantiated property)’, whereas ‘John?’ means ‘Is it John (that has some as yet uninstantiated property)?’. The incremental view merely needs the (incremental context) assumption that QUD can get updated word by word (rather than at turn boundaries), roughly as in (b):45

44

More precisely, what gets accommodated are InfoStrucs, in other words question utterance pairs that therefore include also the specification of an FEC. For some discussion as to how this relates to case specification, see Ginzburg (: section .). 45 For a detailed account of this, see Ginzburg et al. ().

  -    ()



a. Someone—John?/Any idea who?—took the kettle and didn’t return it. b. Content: ∃x,P[Person(x) ∧ P(x)]; QUD: ?∃x,P[Person(x) ∧ P(x)]

... Reprise sluicing and reprise fragments A key component of our analyses of elliptical reprise constructions are the questions accommodated into QUD as a consequence of the clarification interaction process triggered by incomplete understanding of the previous utterance, as described in section ... With that in place, the existing grammar can then directly yield the requisite readings for reprise sluicing and for the confirmation readings of reprise fragments. Assume the utterance to be clarified is (a). B uses the CCUR parameter identification to build a context as in (b): ()

a. A: Did Bo leave? B: Bo? (= Did you mean Bo?) b. - = λxMean(A,u0,x); FEC = A’s utterance ‘Bo’

Given this, the analysis of the construction is illustrated in (): the construction decl-frag-cl builds the proposition Mean(A,u2,b); the construction polarization builds a polar question from this: ()

S polarization CONT = ?hd-dtr.cont = ?Mean(A,u2,b) : Questn

S decl-frag-cl

maxqud =

q = λxMean(A.u2.x) : Questn fec = p2 : LocProp

cont : [x : Ind] hd-dtr : cat = fec.cat : syncat cont = maxqud.q(hd-dtr.cont.x)

NP BO Intended content readings of RFs involve a complex mix of a prima facie non-transparent semantics and phonological parallelism. Independently of intended content readings, we need to capture the utterance anaphoricity of ‘quotative’ utterances such as ():



  

()

a. A: Bo is coming. B: Who do you mean ‘Bo’? b. D: I have a Geordie accident. J: ‘accident’ that’s funny.

We assume the existence of a grammatical constraint allowing reference to a sub-utterance under phonological parallelism. () exemplifies one way of formulating such a constraint: the phon value is type identical with the phon value of an utterance identified with the FEC, whereas the content is stipulated to be the utterance event associated with the focusestablishing constituent:46 ()

utt-anaph-ph = phontype = max-qud.fec.sit-type.phon : Type phon : phontype cat : syncat max-qud : info-struc cont = max-qud.fec.sit : Rec

With this in hand, we turn back to consider the issue of how intended content RFs arise grammatically. It is worth emphasizing that there is no way to bring about the desired content using decl-frag-cl, the short-answer/reprise sluice phrasal type we have been appealing to above, regardless of whether we analyze the NP fragment as denoting its standard conventional content or alternatively as denoting an anaphoric element to the phonologically identical to-be-clarified sub-utterance. This is a prototypical instance of appeal to constructional meaning—a complex content that cannot be plausibly constructed using ‘standard combinatorial operations’ (function application, unification, etc.) from its constituents. Thus, one way of accommodating intended content RF is to posit a new phrasal type, qud-anaph-int-cl. This will encapsulate the two idiosyncratic facets of such utterances, namely the -/ identity and the hd-dtr being an utt-anaph-ph: ()

qud-anaph-int-cl = MAX-QUD : InfoStruc cont = max-qud.q:Question hd-dtr: utt-anaph-ph

Given this, we can offer the following analysis of (): ()

A: Is Bo here? B: Bo?

B lacks referent for ‘Bo’; uses parameter identification to update MaxQUD accordingly:

46

Postulating such an ambiguity goes back to Frege and Quine, who suggested that phrases within quotative operators denoted the string itself. () makes one simplifying assumption: identifying the phon value of the focus-establishing constituent with that of the utterance anaphoric phrase. In practice, this should only be the segmental phonological value.

  -    ()



spkr = B addr = A sit = w0’

pending =

sit-type = IGH

maxqud =

q = λxMean(A,u2,x) : Question

: InfoStruc

fec = u2 : LocProp Using qud-anaph-int-cl yields: ()

S qud-anaph-int-cl maxqud =

q = λxMean(A,u2,x):Question

: InfoStruc

fec = p2 : LocProp CONT = maxqud.q S utt-anaph-ph bo = maxqud.fec.sit-type.phon : Type phon : bou BO

... Disfluency Disfluencies are viewed as a performance phenomenon in most formal grammatical treatments.47 Ginzburg et al. (b) provide extensive argumentation for the need to integrate disfluencies in the competence grammar, demonstrating that they participate in semantic and pragmatic processes like anaphora, implicature, and discourse marker content, as well as being subject to cross-linguistic variation and also exhibiting some universals. Ginzburg et al. (b) develop their account of disfluencies in KoS by extending the account mentioned in previous sections of the coherence and realization of clarification requests: as the utterance unfolds incrementally there potentially arise questions about what has happened so far (e.g., what did the speaker mean with sub-utterance u1?) or what is still to come (e.g., what word does the speaker mean to utter after sub-utterance u2?). These can be accommodated into the context if either uncertainty about the correctness of a sub-utterance arises or the speaker has planning or realizational problems. Thus, the monitoring and update/clarification cycle is modified to happen at the end of each word utterance event, and in case of the need for repair, a repair question gets accommodated into 47

Though not by psycholinguists: see, e.g., Levelt (); Clark and Fox Tree ().



  

QUD. In this way, the coherence of ‘spelled-out’ self-repairs such as (a) can be explained, but also of elliptical self-repairs such as (b). The latter is analyzed as a declarative-fragment clause (see section ...), where the antecedent question is the accommodated ‘what did A mean by “earphones”?’ with the utterance ‘earphones’ being the FEC: ()

a. Take that book in I mean from the shelf. b. Have you seen Mark’s erm earphones? Headphones. (BNC, KP –)

.. A worked example As exemplification of what we have done in this section, we return to the example we introduced earlier, (), where the numbering henceforth relates to the sequence of utterances in that example. Our illustration involves a sequence of dialogue gameboards that arise sequentially via updates triggered by conversational rules. We will abuse notation somewhat and notate by ‘:=’ cases where a field gets assigned a new value. Another point to note is that in some cases we consider the dialogue from A’s point of view (in which case we have fields such as ‘A.LatestMove’); in other cases from B’s point of view (in which case we have fields like ‘B.QUD’). This is an inevitable consequence of the fact that in dialogue there is not always a context that is identical for the conversational participants. () A: Who visited Bo? Initial query: posed by A; updates his DGB accordingly with the question expressed by the utterance in ().

A.LatestMove := Ask(A,B,q0) A.QUD :=

q = λxVisit(x,b) fec = Who

() B: Bo? A’s utterance cannot be grounded by B, hence it remains in Pending. Here A’s utterance () is assigned to the label ‘sit’, whereas the sign associated with the form ‘Who visited Bo?’, which we denote as ‘WVB’ is assigned to the label ‘sit-type’. B accommodates a clarification question (‘Who did A mean by the utterance “Bo”?’), thereby licensing a clarification question expressed as a reprise (the construction type qud-anaph-int-cl ).

B.Pending =

sit = (1) sit-type = WVB

B.QUD =

q = λxMean(A,u0,x) fec = Bo

() A: My cousin. B’s question receives an answer in the form of a short answer, licensed by the QUD above (the construction type decl-frag-cl); ‘Bo is A’s cousin’ is added to FACTS and the initial question is now MaxQUD for B as well:

B.LatestMove := Assert(A,B,p1) p1 = Cousins(A,b) B.QUD := p1?,

q = λxVisit(x,b) fec = Who

B.FACTS = Cousins(A,b)

B.QUD :=

q= λxVisit(x,b) fec =Who

() B: Jack did. B can now ground the utterance () with the referent she was provided in (): the question from () has become MaxQUD. This licenses an utterance using the SubjChoice VPE construction (cf. section .., λxVisit(x,b) is the salient property). B asserts p2, hence p2? becomes MaxQUD.

A.DGB.LatestMove := Assert(B,A,p2) p2 = Visit(j,b)

A.QUD := p2?,

q= λxVisit(x,b) fec =Who

() A: Who else? A deduces from B’s utterance the question pp3?=whether someone other than Jack visited   q¼p3 ? Bo. The info-struc licenses the sluice (the construction type slu-int-cl): fec¼Jack

FACTS := Cousins(A,b), Visit(j,b)

A.QUD :=

q=p3 ? fec=Jack

A.LatestMove := Ask(A,B,q2) A.QUD :=

q=q2 fec=Who else

,

q=p3? fec=Jack



  

() B: No one. A’s question receives an answer in the form of a short answer, licensed by the QUD above:

B.DGB.LatestMove := Assert(B,A,p4) p4 = ¬∃x, x≠j, Visit(x,b) B.QUD := p4?,

q=q2 fec=Who else

,

q=p3? fec=Jack

FACTS := Cousins(A,b), Visit(j,b) () a. A: Didn’t Bill . . . A does not accept B’s assertion and using the question q2 to license VPE.48 In midutterance (i.e. with Pending consisting of the initial segment ‘Didn’t Bill’) A realizes that ‘Bill’ was not what she meant to utter. She accommodates the issue ‘Who did A mean to utter with “Bill”?’ as MaxQUD:

A.Pending :=

sit = (7a) sit-type = Didn’t B

A.QUD :=

q=λxMean(A,‘Bill’,x)

, p4?,

fec=Bill

q=q2

,

q=p3? fec=Jack

fec=Who else

b. … I mean Mary? A self-corrects, addressing the issue of ‘Who did A mean to utter with ‘Bill’?’, which leads to her intended utterance—a polar question about Mary: A.LatestMove := Assert(A,B,p5) p5 = Mean(A,‘Bill’,m) A.QUD := p5?,

q=λxMean(A,‘Bill’,x) fec=Bill

48

,

q=q2

,

q=p3?

fec=Who else fec=Jack

Why is q2 MaxQUD? In KoS QUD is not taken to be a stack, but rather a partially ordered set. Ginzburg (), motivated in part by multi-party dialogue, proposed that when a question q is pushed onto QUD it doesn’t subsume all existing questions in QUD, but rather only those on which q does not depend. Hence, wh-questions do not get outranked by polar questions which are their subquestions.

  -   



A.LatestMove := Ask(A,B,p6?) p6 = ¬Visit(m,b) FACTS := Mean(A,‘Bill’,m), Cousins(A,b), Visit(j,b)

A.QUD := p6?, q=q2,fec=Who else , q=p3?,fec=Jack () B: No. B refutes A’s hypothesis, using ‘No’, which reaffirms MaxQUD: the negative proposition p6:

B.LatestMove := Assert(B,A,p6) p6 = ¬Visit(m,b) FACTS := Mean(A,‘Bill’,m), Cousins(A,b), Visit(j,b) A.QUD := p6?, q=q2,fec=Who else , q=p3?,fec=Jack

A This piece is dedicated to the memory of our friend, teacher, and collaborator, Ivan Sag. Ivan was originally intended as a co-author of this piece, but died before it was drafted. His inspiration and many ideas are evident, though the chapter is much the poorer due to his absence. We would like to thank Doug Arnold, Bob Borsley, Rui Chaves, Jong-Bok Kim, Jason Merchant, François Mouret, Stefan Müller, Enric Vallduví, and Tom Wasow, as well as two anonymous reviewers for comments on the first version. We are also very grateful for the help and extreme patience of the editors, Jeroen van Craenenbroeck and Tanja Temmerman, whom we kept waiting endlessly. We acknowledge the support of the French Investissements d’Avenir-Labex EFL program (ANR--LABX-) and the Disfluency, Exclamations, and Laughter in Dialogue (DUEL) project within the projets francoallemand en sciences humaines et sociales funded by the ANR and the DFG.

  ......................................................................................................................

    ......................................................................................................................

 

. I 

.................................................................................................................................. T chapter focuses on ellipsis within the general framework of Categorial Grammar (CG), whose leading ideas and their implications for ellipsis are elucidated in section .. We focus primarily on VP-ellipsis (VPE) and fragment answers as two case studies. In a nutshell, the version of CG endorsed here would make it surprising to find deletion or silencing of linguistic material under identity with some other linguistic expression. This raises several questions. First: what is the structure of the ‘ellipsis site’? The answer here is straightforward: there is no structure; see . for more explicit discussion. This in turn raises the question of how the relevant meaning is computed—how is it that material is understood when nothing is there? This is addressed in . (see also .. and ..) (the answer to this question is somewhat different for fragment answers than it is for VPE). Finally, this raises the question of what ‘licenses’ these kinds of ellipsis. To put that differently, how is it that the syntax specifies that the relevant material is a well-formed expression? Section . details this with respect to VPE (including Antecedent-Contained Deletion and pseudogapping). For the case of fragment answers, the relevant question can be framed in a slightly different way. That they are well-formed is unsurprising; they are expressions of various categories (they simply are not expressions of category S). But a slightly different although related question arises: What allows particular expressions to serve as ‘answers’ to questions? This is discussed in ...

. C G

.................................................................................................................................. While there are several versions of CG, part of the interest of CG for the analysis of ellipsis stems from one of these versions embodying a rather strong hypothesis about the syntax/ semantics interface. This is the hypothesis of Direct Compositionality (hereafter, DC) articulated in, among others, Montague (). DC rejects the notion that the syntax and

   



semantics work separately and rejects the idea that the syntax computes a representation which is ‘sent’ to the semantics for interpretation. Rather, the syntax is a system of rules/ principles which prove expressions well-formed (often proving larger expressions wellformed on the basis of smaller ones), and the semantics works in tandem, supplying a model-theoretic interpretation of each expression that is ‘built’ (i.e., proven well-formed) in the syntax. This has some interesting consequences. First, there is no use of an intermediate level of representation such as Logical Form (LF) mediating between the actual (pronounced) sentence and its interpretation. Second, each local expression that is well-formed according to the syntax has a meaning. And since the semantics works compositionally, its meaning cannot depend on the presence of other material in the larger sentential or discourse context in which it is embedded.1 We will return to the consequences of this for the analysis of ellipsis.

.. Some basics of CG First, we develop a rudimentary CG syntax and semantics. Thus every linguistic expression (a word, sentence, or anything in-between) can be seen as a triple of sound, category, and meaning. We use the notation [α] to represent the sound of an expression α (although I use orthography for convenience), and [[α]] to represent the meaning of α. Note that by ‘meaning’ I mean some model-theoretic object, not a string of symbols, although I will use such symbols as a way to name the model-theoretic objects. Following standard assumptions in much formal semantics (including non-CG theories), we take the basic building blocks of ‘meanings’ to be a set of individuals (notated e), a set of worlds (w), a set of times (i), and the two-member set {,} which is the set of truth values (t). We take each linguistic expression to have as its semantic value something from one of these sets, or some function built from these sets (including functions which have other functions as their domain and/or range). Let mean the set of functions from a set a to a set b. (Hence, is the set of functions whose domain is the set of individuals and whose range is the set of truth values.) For the most part I will simplify in this chapter and ignore intensions (i.e., the role of worlds and times). Having made that simplification, the value of a VP like chased the rat is a function in . We thus say that the semantic type of a VP is . Under this conception, the grammar is a system of rules each of which takes one or more expression(s) (i.e., triples) as input and yields a triple as output. The interest in a CG syntax lies in its conception of syntactic categories. As is the case for the semantics, we assume a set of basic syntactic categories; for present purposes let these include S, NP, N, and AP. In addition, we recursively define additional categories as follows: If A is a category and B is a category then A/RB is a category and A/LB is a category.2 The intuition behind these 1 This is oversimplified in that there certainly are expressions whose semantic value depends on the context of utterance—indexicals being the obvious case. A compositional semantics can accommodate these by taking the semantic value of every expression to be a function from speech contexts to something else (hence the value of I live in Providence is a function from contexts to propositions). We will ignore the case of indexicals (and related expressions) for now; see n.  for discussion about their relevance to the account of ellipsis. 2 Following much of the tradition within CG, I use NP rather than DP for the category of things like the disobedient husky and Samantha. The reader preferring DP can make the substitution as needed.





complex categories is that an expression of category A/RB says ‘give me an expression of category B to my right and I’ll return an expression of category A’, and similarly an expression of category A/LB takes a B to its left to yield a new string of category A. Thus, the syntactic categories directly encode the distribution of an expression. Note also that ‘VP’ is not in the list of primitives; this is because it can be recast as S/LNP. I will, however, often use the term ‘VP’ for convenience. We introduce additional categories later. The advantage of adopting this notion of syntactic categories is that it allows for the combinatory principles/rules to be stated in very general terms, and it allows for a transparent mapping between the syntax and the semantics. Thus, assume that all expressions of the same category have the same type of meaning. Let any expression of category A have as its meaning some member of the set a, and let any expression of category B have as its meaning some member of b. Then an expression of category A/B (whether it be a ‘left slash’ or a ‘right slash’) has as its meaning a function from b to a, i.e., it is a function in . For example, let all expressions of category NP denote individuals, and all expressions of category S denote truth values. Then S/NP (which is a ‘VP’) is of type —it denotes (relative to a world) some function from individuals to truth values. Given that, we can state the syntax and the semantics (extensionally) with two very general rule schemas: ()

a. If α is an expression of the form and β is an expression of the form then there is an expression γ of the form . b. If α is an expression of the form and β is an expression of the form then there is an expression γ of the form .3

There is also no reason not to expect unary rules—i.e., rules taking a single triple as input to return a triple as output. The rules that have gone under the rubric of ‘type-shifting’ rules in the literature are a special case of these. We will see some such rules later. To illustrate this mini-grammar, take a simple sentence like Tabby chased the rat. First we need some lexical entries. Assume that proper nouns like Tabby are listed in the lexicon as being of category NP (and each denotes an individual). Ignoring intensions again, take a basic intransitive verb such as howled to be listed in the lexicon as being of category S/LNP with a meaning of type (thus it characterizes a set of individuals), while a transitive verb such as chased is of category (S/LNP)/RNP and of type . Rat is listed as N and presumably the type of expressions of category N is also (i.e., also a function which characterizes a set of individuals), and let the be listed as NP/RN where its meaning is of type and, more specifically, it maps a set to its unique or unique contextually salient member. Incidentally, the directional slashes should not be stipulated item by item as they are entirely predictable; see Jacobson () for conventions allowing lexical items to be listed in an underspecified form with rules supplying the directional features. Given this, the two rule schemas in () allow the to combine with rat to give the rat of category NP 3

Recall that we are ignoring intensions here; the true meaning (intension) of expressions is actually a function from worlds and times to other things, and so the rules in () are actually slightly more complex. For treatments of intensionality, see any introductory formal semantics textbook such as Dowty, Wall, and Peters (), Heim and Kratzer (), or Jacobson ().

   



(and it will denote an individual); chased to combine with the rat to give the expression chased the rat of category S/LNP whose extension is some set, and that in turn to combine with Tabby to the left to give the S Tabby chased the rat (whose extension will be  or ).

.. Relevance for ellipsis What is the relevance of all this for ellipsis? In a nutshell, if direct compositionality is correct, we would be surprised to find rules which suppress the phonology of some material under identity (either formal or semantic) with other linguistic material. To see why, take the case of VP-ellipsis. (I use the term ‘verb phrase ellipsis’ in a theory-neutral way, without of course endorsing a view in which linguistic material is deleted or silenced. Moreover, ‘verb phrase’ is too narrow, since adjective phrase complements of be can also be missing.) Suppose we try to think of the ‘silencing’ as a unary rule taking some triple as input and returning as output a triple with the same category and meaning, but where the phonology is empty. We can begin to formulate this as in (): ()

Let α be an expression as follows: . Then there is an expression β of the form provided that there is another expression γ in the discourse context such that possibility (a): the form of γ = the form of α or possibility (b): [[γ]] = [[α]].

The two possibilities listed in () are intended to instantiate two different theories of ellipsis under identity: (a) is one in which formal identity is required while (b) requires instead semantic identity. And the rule as stated in () is too crude in that obviously not any VP can have its phonology suppressed, but that part can be fixed by appropriate use of features. Hence, there is nothing wrong with the idea of a unary rule that changes the phonology, even suppressing it altogether—a rule that deleted or silenced a constant (such as, perhaps, that) is perfectly compatible with the framework here. But the problem with () for the kind of locality imposed by direct compositionality comes from the ‘provided that’ part: being identical (in some sense) to other material in the discourse context is not a local property of the input expression.4

4 This is not to say that there is no way such a rule could conceivably be stated. As noted in n. , there are indexicals whose value depends on the speech context, and so one might take VPE as a kind of indexical. But we will not pursue that possibility here as this is both complex and does not resemble other phenomena whose value varies with the speech context. The value of speech-context-dependent expressions usually depends on properties like who is the speaker, who is the hearer, what material is salient in the context, etc., and is generally not dependent on the presence or absence of (or successive) utterances. Note, for example, that deaccenting of ‘old information’ (or, lack of accent on this) does not require overt mention of that information; it can be inferred from other facts about the discourse context or even just be salient in virtue of facts about the world. Of course, the claim that VPE is similar (in picking up a salient property) raises the question as to why VPE is so much ‘fussier’ than deaccenting (see, e.g., Rooth a). But there is a reasonable answer: in the case of deaccenting the processor does not need to do the same work since the deaccented material is actually said. This is not the case for ellipsis constructions.





Yet if there is no actual linguistic material (either at some level, or simply unpronounced) in the position of an ‘ellipsis site’, two questions arise. First, what licenses material which many theories would take to be syntactically incomplete? In the case of VPE, for example, auxiliaries usually select for VPs, so what allows for a VP to be absent? In the case of fragment answers, what allows certain subsentential expressions to stand alone and be construed as answers to questions? The answer to this will be addressed as we present an analysis of each of the domains discussed here. The second question is: how is it that extra material which is not actually present is understood in the context of a fuller discourse? In theories that do posit silent linguistic material (hereafter, SLM) the usual assumption is that the relevant material is understood in the way it is because it is in some sense actually present in the linguistic string, albeit in silent form. We return to challenge this received wisdom in section ... Nonetheless, for now one might wonder how a theory without SLM can account for the understanding of ‘ellipsis constructions’. Thus the remainder of this chapter centers primarily on VPE as a case study, and also briefly touches on fragment answers. Our first task (.) is to sketch an answer to the second question above: how, in each of these cases, a listener understands extra material which does not correspond to any actual linguistic string. Our next task (.) is to briefly answer some of the many arguments that have been given for the SLM point of view. Space precludes a thorough discussion, but we will show that large classes of these arguments are very theory-dependent and/or are in some cases based on questionable assumptions or logic. Section . develops some more tools of CG which are used in . to provide a more detailed analysis of VPE (as well as of ACD and pseudogapping) within a CG framework. Obviously the brief sketch of fragment answers combined with a more detailed analysis of VPE (including ACD and pseudogapping) does not exhaust the landscape of phenomena that have gone under the rubric of ellipsis, but hopefully provides a representative sample of how a CG can easily handle these.

. I   ,  ?

..................................................................................................................................

.. VPE I will suggest that VPE (and many other ellipsis constructions) is basically no different than the case of ordinary free pronouns. The obvious apparent differences (e.g., the so-called deep vs surface anaphora distinction) stem from the nature of the object that these pick up. To clarify, take a typical VPE case as in (): ()

Bode will win a gold medal, and Lindsay will too.

Without additional supporting discourse context, the interpretation of the second sentence is that Lindsay will also win a gold medal. Traditional SLM accounts posit that this is because the representation of the full second clause is something like Lindsay will win a gold medal too, where the strikethrough material represents material that is either deleted (but is present at the level that semantic interpretation cares about) or is present but with null phonology. But I will argue there is no silent material. Will is allowed (by mechanisms to be

   



spelled out in ..) to occur with no complement, and—following (roughly) Hardt () and many since—we assume that the understanding is supplied by the discourse context. Put informally (again the formal details are spelled out in ..), one can think of the actual meaning of () to include a free variable P over properties (i.e., functions of type ); so that the second clause in () is the proposition that Lindsay will also have the P property, and where the value of P is supplied by context. (Note that more precisely P is of type —i.e. a time-dependent set, since [[will]] maps a time-dependent set into another time-dependent set but we continue to ignore worlds and times.) Crucially, I am not claiming that VPE sometimes involves a contextually supplied property and other times involves some kind of deletion/silencing under identity. Rather, the claim is that VPE always involves picking up a contextually salient property. But it just happens that very often the relevant property is made salient by having been named—hence the illusion of an ‘identical antecedent’. Here the first clause contains a VP whose meaning is [[win a gold medal]] and hence this property is obviously very salient. In other words, there never is a real ‘antecedent’; the terms ‘matching’, ‘identity’, etc. are all inappropriate as there is no formal or grammatical relation between the understood complement of the auxiliary and the VPs whose meaning happens to make the relevant (i.e., understood) property salient. Of course this view immediately runs afoul of much common wisdom beginning especially with Hankamer and Sag () which has shown that in general it is extremely difficult (at best) to supply a contextually salient property unless it is explicitly named. Let me put this objection aside for the moment and return to it in section ...

.. Fragment answers Many ellipsis constructions can be handled similarly (an obvious example would be Null Complement Anaphora). But I don’t wish to claim that everything that has gone under the rubric of ellipsis is a matter of an anaphoric process picking up contextually salient material. Fragment answers to questions, for example, have a tighter linguistic connection to the question. For example, it is well known that in languages with rich case marking on NPs, the case of a fragment NP answer must match the case on the wh-word in the question (Morgan , Merchant a, and many others). But this too can be accounted for—as can the understanding of a fragment answer in a discourse—without SLM. Take a question–answer pair like (): ()

Q: Who did Claribel invite? A: Bozo.

Obviously the answer here is understood as conveying the proposition that Claribel invited Bozo. But this follows under the view of questions and answers put forth in Groenendijk and Stokhof () and also (in slightly different terms) in Ginzburg and Sag () whereby the grammar contains question–answer pairs as an actual linguistic unit (call it a Qu-Ans). There is, after all, no reason to think that the largest unit the grammar has anything to say about must be a sentence, and if there are larger units then surely a question–answer pair is a good candidate for a full-blown linguistic unit. Thus, we can say that a Qu-Ans has a syntax and a semantics. The syntax is such that the pair constitutes





a Qu-Ans only if the category on the fragment ‘answer’ matches the category of the whword. Moreover (as pointed out in both of the works above), the case-matching facts follow as long as one makes the simple assumption that this matching requirement includes case marking. The semantics relies on the assumption that the meaning of the question in () is a function of type : it is the function λx[Claribel invite x], although obviously it also contains some kind of illocutionary force marker distinguishing this from an expression like was invited by Claribel. The semantics of Qu-Ans specifies that a proposition is derived by applying the meaning of the question to the meaning of the answer. This means that Bozo in the answer in () is nothing more than a bare NP whose meaning by itself is just the individual b. However, in conjunction with the question, the full Qu-Ans gives the proposition that Claribel invited Bozo. This is an interesting case in that the propositional information is therefore shared between the questioner and the answerer. Jacobson (b) details a number of advantages of this over SLM accounts. While space precludes detailed discussion of this we mention one here. Consider the question–answer pair in (): ()

Q. Which mathematics professor came to the party? A. Jill.

The short answer Jill commits the responder to the ‘presupposition’ that Jill is a mathematics professor. This follows immediately under the Qu-Ans analysis: the meaning of the wh-expression contributes to the meaning of the question as a whole. Hence the semantics of the question is the function λxx E[ math prof]] [x came to the party]—i.e., it is a partial function defined only for mathematics professors. Thus it can combine with the referent of Jill only if that person is a mathematics professor. No such commitment is made by a responder using a ‘long reply’ (which is not technically an ‘answer’ in the sense of the Qu-Ans analysis); thus the long reply Jill came to the party does not commit the responder to this view that Jill is a mathematics professor. (In fact, it is most natural in contexts where the responder is not sure about this, and most natural if preceded by Well, and uttered with the ‘fall–rise’ intonation discussed in Ward and Hirschberg .) Jacobson (b) details that the SLM view has no ready account of this difference between the short answer and the long ‘reply’. While space precludes demonstrating this here, we can certainly note that the Qu-Ans analysis accounts for the case connectivity between the wh-word and the answer, accounts for the meaning of the entire Qu-Ans construction without SLM, and accounts for the presupposition facts in (). As to what allows particular expressions to stand alone as answers (the ‘licensing’ question): under this analysis they must match the category of the wh-word. Hence it would be those expressions whose category can be fronted wh-expressions in questions.

. A  SLM?

.................................................................................................................................. Readers familiar with the many arguments for SLM will wonder how they can be addressed under the view here. Obviously space precludes discussing more than but a fraction of these, but a few of direct relevance to the points here are worth commenting on (for further discussion, see, among others, Jacobson b).

   



.. Deep vs surface anaphora The first obvious objection to the view in .. with respect to VPE stems from the seminal observations of Hankamer and Sag (): quite often it is impossible to retrieve the relevant property just from the discourse context, even when the context is loaded to make a property salient. It is this, perhaps more than anything else, which has driven the idea for decades that the grammar requires an overt antecedent for VPE and this then leads to the questions of the nature of the identity that must hold between that antecedent and the ‘silenced’ VP. The VPE case contrasts vividly with the case of ordinary personal pronouns which of course can refer to entities never having been named (see, e.g., Lasnik ). But it has also been known for decades that it is not true that VPE must have an overt linguistic antecedent: there are many examples of VPE picking up a property which has not been named. Such examples come in two varieties. In the first, there is no remotely plausible linguistic antecedent. The literature contains many such examples: see, e.g., Dalrymple, Sheiber, and Pereira (), Merchant (a), and Miller and Pullum () for a variety of attested naturally occurring examples. The second type are those in which there is linguistic material which can plausibly be seen as some sort of ‘antecedent’ but where—if overt material had been in the ellipsis site—it would not match that ‘antecedent’ either in form or in meaning. Examples include () (Webber ) (where the understanding is something like neither of themi will go the place that hei wants to go), and the wellknown voice mismatch examples such as the one in () (Hardt ): ()

Bill wants to go to Paris and Tom wants to go to Rome, but neither of them will.

()

That problem needed to be looked into, so Gorbachev did.

Notice that mismatch in () is not just a matter of mere formal mismatch. I assume that a passive VP such as (be) looked into does not have the same meaning as an active VP such as look into the problem, hence the so-called antecedent and the VP that would have been there had there been pronounced material match in neither meaning nor form. And then there are of course a number of cases involving things like ‘vehicle change’ (Fiengo and May ), and Negative Polarity Item mismatches (see Sag a and many since) which surely involve at least formal mismatch; whether or not these involve a semantic mismatch depends on particular analyses. At any rate, though, there are plenty of cases running counter to the Hankamer and Sag generalization. And so, as long as there are any cases whatsoever of an elided VP without an ‘antecedent’ to match, then any theory needs an account of why this is not always possible. However one answers this question, it is surely better to have a single account of VPE that can be supplemented with a theory of why sometimes the relevant properties are difficult to access without prior linguistic mention than it is to posit two entirely separate mechanisms (as seems to be forced under the SLM under identity view) and still need to supplement that with an account of why it is that the mechanism at work when there is no actual antecedent applies only in limited circumstances. The hope of the account here, then, is that VPE is always a case of ‘deep anaphora’—it just picks up a contextually salient function of type —but that, unlike individuals, such





functions are fragile objects. They are not easy to access without heavy contextual support, and they therefore like to be made salient by having been recently named. Hence the illusion of there being an ‘antecedent’ just comes from the fact that very often the property is made salient by being named—it is the meaning of a nearby VP. Such functions, unlike individuals, also decay in salience quickly: it is not enough to have the relevant property be named by overt linguistic material; it must also have been named recently (see Sag and Hankamer ). Under this view, then, the question still arises as to why sometimes we can infer the property from the context, sometimes it helps to have linguistic material that does not overtly name the property but from which we can infer the relevant property (e.g., in () the first sentence makes highly salient the function mapping an individual x to the property of going to the place x wants to visit), while at other times the property needs to be explicitly named (as in the classic examples in Hankamer and Sag ). And I do not have any definitive answers to give here. But this reframes the question to be one about processing and how entities (including properties) become sufficiently salient in a discourse. In any case, it is worth stressing again that as long as there are any good cases of a ‘missing’ VP without an overt linguistic antecedent—and such cases surely do exist—then any theory needs to explain why this is limited. So no extra burden of explanation is placed on the theory here. There is one final objection one might raise to the claim that VPE is really an instance of a ‘deep anaphoric process’ and requires no overt linguistic antecedent. If the illusion that VPE does require an overt antecedent is just due to the fact that properties (i.e., functions of type ) are fragile and not easy to access, then why do we see the contrasts between do it and the case of a missing complement of an auxiliary (including the auxiliary do) that played such a prominent role in Hankamer and Sag? That is, why is (a) perfect while (b) is at best quite degraded in the same context? ()

Context: You and I both know that Sally has been trying to summit Mt Everest for years, and we finally see a picture of her on Facebook at the summit. I say: a. ?*I can’t believe it, but I guess she finally did! b. I can’t believe it, but I guess she finally did it!

But there is no reason to think that the referent for it in this or other discourses is the same kind of object as the meaning that needs to be picked up in (a) (for relevant discussion, see Miller and Pullum ). Arguably, the vast amount of work assuming that there is a moral to be drawn from the contrast between do (with no complement) and do it is based on an accident of the homophony between the two different do’s. Auxiliary do and main verb do are different and have different meanings, and so the type of objects that their complements would pick out are also different. For example, we know that main verb do only occurs with agentive predicates, and find contrasts like (): ()

a. Bill knows French, and Tom does too. b. *Bill knows French, and Tom does it too.

Presumably, it is an anaphor over events (or something similar), and in particular over events with agents. The function [[know French]] which in (b) is made salient by the first conjunct can be picked up as the ‘missing’ property complement of auxiliary do, but not as

   



the complement of main verb do (and so it cannot refer to that function). Thus, there is no reason to think that the way in which events can be made salient is the same as the way in which the relevant functions can be.

.. Connectivity effects Aside from the fact that silencing under identity would be surprising under DC, there is another way in which the DC hypothesis bears on the analysis of ellipsis. Thus, many of the arguments for the presence of SLM are based on the following reasoning: Premise: There is some phenomenon P which can be stated only with respect to large chunks of representation. Fact: Smaller expressions (such as fragment answers) show the effects of phenomenon P. Conclusion: Therefore, the small expression must hiddenly be surrounded by more material. To use fragment answers as illustration, a perfect example of this comes from ‘bound’ pronouns. Thus, it is often assumed that these must be c-commanded by a binder (perhaps at LF) and that this would be the only way for the semantics to put together the relevant reading. Yet we find question–answer pairs like the following: ()

Q: Who does every third grade boyi call (on the first day of school)? A: Hisi mother.

(The indices are included here just to indicate the intended reading, and are not assumed to be part of the actual grammatical apparatus.) If indeed the pronoun his must be c-commanded by an instance of every third grade boy in order to get the right interpretation, then the answer here must hiddenly be a fuller sentence such as Every third grade boy calls his mother where the strikethrough indicates silent material. But analyses of any phenomenon that rely on constraints on chunks of representation (such as a c-command constraint, Principle A, Principle B, etc.) in any case could not be correct under DC, as the syntactic and semantic rules do not have access to such chunks of representation. Trees are just convenient representations for the linguists and not something that the grammar ‘sees’ and it cannot state such constraints. So according to the DC worldview, the premise in arguments of this sort cannot be correct. Of course, saying that constraints (such as a c-command constraint on binding) could not be stated under DC would be of little consolation if there were no alternative ways to account for the phenomena in question. But indeed there are. Thus the kind of variable-free semantics developed in Jacobson () automatically accounts for the question–answer pair in () (note that variable-free semantics is motivated by entirely independent facts). A necessarily very brief sketch is as follows. First, the question is a functional question: it asks for the identity of a function f of type such that every third grade boy is an x who is in the call-relation to f(x) (see Groendijk and Stokhof  and Engdal ). That this question asks for a function is not novel in the variable-free approach, but what is novel here is that no extra apparatus is needed in order to get that reading. Second, the answer his mother automatically denotes a function of this type: it is the function mapping each (male) individual onto his mother. As such, it is appropriate as an answer to a functional question. Further details for a related case can be found in Jacobson () and for this case in Jacobson (b). We give this here as just one example, but the moral is that arguments for SLM based on connectivity effects are generally irrelevant under the DC worldview.





.. The meaning comes for free—but does it? While not always made explicit, one observation which seems to implicitly underlie part of the motivation for the SLM view is that the natural understanding of B’s utterance in a discourse like in () is automatic if this utterance really contains an instance of the silent material win a medal too: ()

Speaker A: Lindsay will win a gold medal. Speaker B: Yeah, Bode will too.

It certainly seems as if Bode will too  the same as Bode will win a gold medal too, and this is automatic if B’s utterance really hiddenly is just that. But does B’s utterance really mean [[Bode will win a gold medal too]]? Actually, we have no way of knowing that—the only empirical fact is that in the context of a discourse like () most speakers (A or some other observer C) would assume that this is what B intends to convey. Notice that the proposition conveyed by Bode will too is not stable and depends entirely on the discourse context (which is unlike other literal contributions to the truth conditions of a sentence which are stable). In fact, a bit of reflection reveals that the claim that B’s utterance hiddenly contains silent material provides no more of an explanation for the empirical facts about a listener’s understanding than does the account here. After all, the listener has no direct access to the silent material, and must infer what it is. And this in turn raises interesting questions about how processing really works under the SLM account. Do we wish to say that the listener, having heard the string [win a gold medal], supplies this phonological material and then decodes the string [Bode will win a gold medal] exactly as if this were what s/he had heard? This is highly unlikely for two reasons. First, since the VP [win a gold medal] has already been processed (i.e., assigned a meaning), it would be an odd processor that had to go back and reprocess it. Second, as known since at least as early as Sag and Hankamer (), the situation with indexical elements shows that what has to be supplied by the processor is a meaning and not a form. For example, consider the discourse in (): ()

Speaker A: Sally won’t vote for my mother. Speaker B: Too bad, but Jack will.

The understanding of the ‘missing’ complement in B’s utterance is the property of voting for A’s mother, not B’s mother, and so it can’t possibly be that what is supplied is the string [vote for my mother] which is then interpreted. And it is not just a matter of a simple mechanism switching first- and second-person pronouns. Thus consider (): ()

Speaker A to B: Who is going to nominate your mother? Speakers C and D are eavesdropping, and C turns to D and says: I know. Sarah will.

Here the ‘missing’ material cannot be nominate your mother or nominate my mother. It must be a meaning which is the value of the VP in A’s utterance in the relevant discourse context (i.e., property of nominating B’s mother). No simple recipe can be given to predict

   



these without consulting the meaning (for more extensive discussion in the context of fragment answers, see Jacobson b). So the listener must supply some meaning as the missing complement. Given that, the way in which ‘missing’ VP meanings are understood in a discourse context provides no evidence for the existence of SLM. For assuming that the listener directly ‘picks up’ the relevant property (and that there is no SLM) works every bit as well to account for the empirical facts about understanding in a discourse. Of course, this does not mean that there could not be other reasons to believe in SLM, but the fact that these elliptical constructions appear to mean something more in a discourse context is not one of these reasons.

.. Preposition stranding Finally, a brief look at the so-called preposition-stranding (P-stranding) generalization (see, e.g., Merchant ) is in order as it has played such a large role in the literature. Since this generalization has received considerable discussion elsewhere (including the question of how robust it is) we will not review the evidence for and against P-stranding as usually framed (see Merchant this volume for relevant discussion, as well as the references cited there, and see Jacobson a for a possible way to capture this generalization—should it survive—without SLM). It is, however, worth pointing out a relevant set of facts discussed in Jacobson (a) as these have played less of a role in discussions about the P-stranding generalization; these center on apparent mismatches (in questions and answers) in English. Consider first the question–answer pair in (): ()

a. Into which hole did Alice fall? b. The one belonging to the Mad Hatter.

Note that here (and in Jacobson a) we crucially use a meaningful preposition. These appear at first glance problematic for the Qu-Ans analysis shown above, but a solution to this under Qu-Ans is sketched in Jacobson (a) and we simply refer the reader to that discussion. On the other hand, it is not clear how SLM can account for the mismatch here. Under formal identity the trace in (b) is crucially an NP trace while in the answer it is a PP. Nor is there semantic identity between the question portion of () (Alice fall [PP t]) and the apparent silent material Alice fell into t, since the semantic contribution of into is part of the meaning of the silenced part of the answer, but is not part of the meaning of the relevant portion of the question. (One might think that a copy theory of movement could help for this case, since under such a view into remains in the trace position in the question. But the copy theory of movement work wreaks havoc with the identity of questions and answers in many other cases, as the interested reader can verify.) Note incidentally that a semantic (rather than syntactic) identity condition could work for those cases like () but where the preposition is meaningless. But crucially () shows that these mismatches are allowed even with meaningful prepositions. While this particular case appears to actually give evidence for Qu-Ans over SLM, the situation unfortunately is more complex. For neither theory can in any obvious way handle the reverse case shown in (): ()

a. Which hole did Alice fall into? b. Into the one belonging to the Mad Hatter.





Interestingly, Merchant (this volume), also references this type of case from Jacobson (a). His discussion, however, deals only with cases analogous to () but which involve a meaningless preposition (as in be angry at; Merchant acknowledges that the fact that at here is meaningless is crucial for his account). But (as noted above) Jacobson (a) points out that meaningful prepositions behave the same way, and that the goodness of the pairs in () therefore presents a problem for both SLM and Qu-Ans. In sum, the P-stranding facts are sufficiently complex—not only cross-linguistically but even the simple case of matching/mismatching in P-stranding languages like English—that at this point there is no reason to believe that SLM handles the full range of facts better than does the analysis here.

. M : M C G

.................................................................................................................................. Let us move away from ellipsis per se to introduce some apparatus within Categorial Grammar which allows for a more explicit treatment of VPE, including AntecedentContained Deletion (ACD) and ultimately also pseudogapping.

.. Wh-extraction constructions We begin with a proposal of Partee and Rooth () (see also Montague ) which we embed into a theory with a CG syntax. Let any ordinary NP which denotes an individual map to a generalized quantifier (in particular, it denotes the set of sets containing that individual) with a corresponding syntactic category change. This is a unary rule; it takes a single triple as input and returns a new triple as output: ()

Let α be an expression of the form . Then there is an expression β of the form . Call this lift.

This can be stated more generally: the directional slashes on the output need not be stipulated in the rule and any expression which is normally an argument to some other expression may ‘lift’ to a function taking the latter as argument. Such conventions are spelled out in Jacobson , but due to space considerations we content ourselves here with (). This means that for an ordinary ‘NP’ like Tabby which denotes an individual (call that individual t) there is a homophonous expression of category S/R(S/LNP) with meaning λP[P(t)]—that meaning is a function characterizing the set of all sets having t as a member. Next, Steedman (), Dowty (), and others proposed that wh-extraction and related constructions could be accounted for by adding to the grammar additional binary combinatory schemas allowing function composition (in both the syntax and the semantics). I illustrate this informally as I will propose a slight revision momentarily. The idea is that, for example, an expression of category A/RB could take an expression of category B/RC to its right to give an A/RC, and the associated semantics is function composition. The case of ‘left slashes’ is parallel: A/LB can combine with a B/LC to its left to give an expression of category A/LC.

   



With this, consider a relative clause such as the italicized portion in (): ()

the rat (that) Tabby chased

Chased is of category (S/LNP)/RNP. Tabby can lift as above to give S/R(S/LNP) and these two functions compose to give the expression Tabby chased of category S/RNP and whose meaning is the function which characterizes the set that Tabby chased. (We ignore the step combining this with that; nothing hinges on that detail.) Allowing the grammar to compose expressions like Tabby chased as well-formed, meaningful expressions has benefits besides just for the treatment of wh-constructions. For example, it provides a ready account of so-called RightNode Raising sentences such as Tabby chased and Orinda caught that pesky little mouse that had stolen the cheese. Here Tabby chased and Orinda caught are both of category S/RNP and have meanings of type ; they conjoin as they have the same category (and type of meaning), and then the conjoined expression (also of category S/RNP) takes the mouse-NP to its right. Note too that the final meaning of (that) Tabby chased in () is the same as is assumed in most other treatments of relative clauses; it is a set of individuals. But the way in which we arrive at this meaning is different. A fairly common view in movement-based theories is that a (here silent) wh moves and leaves a trace, so that the material following the rat is which (that) Tabby chased t. The t translates as a variable but this variable is lambda-abstracted over (see, for example, Heim and Kratzer ) so that at the end of the day the material pronounced [Tabby chased] has the same meaning as it does on the CG account above. The difference is that the CG account makes no use of a variable and no trace; the transitive verb chase directly combines with (lifted) Tabby. This will be crucial in the analysis of ACD in ... I will, however, make one modification to the above. Rather than introduce a binary function composition rule, we instead posit a unary rule which, when combined with function application (as in the schemas in ()), is equivalent to function composition. This rule is often referred to in the CG literature as ‘Geach’ (or Division). Hence, assume a rule (call it g) allowing an expression of category A/RB to map to one (A/RC)/R(B/RC); note that this can then combine with an expression of category B/RC to give A/RC. The situation is parallel for left slashes: A/LB can map to (A/LC)/L(B/LC). The semantics of g is such that given a function f in , then g(f) is a function in such that g(f ) = λX[λcC[f(X(c))]]. The reader can verify that g(h)(f ) = h o f. The derivation of (that) Tabby chased is exactly as above, except that rather than lifted Tabby directly functioncomposing with chased, it first maps to an expression of category (S/RNP)/R((S/LNP)/RNP). To get the intuition of this category: the normal lifted category wants a VP (to its right) to give S. The ‘Geached’ category wants an incomplete VP to its right—i.e, it wants a transitive verb to its right, and it returns an S that is still missing an NP to its right. Thus Tabby can now take chased as argument, yielding S/RNP just as in the function composition derivation sketched above. The result here too is that the meaning of Tabby chased is the function characterizing the set of individuals chased by Tabby.

.. Variable-free semantics of pronouns As noted above, this is a variable-free treatment of relative clauses. Jacobson () extends this to pronouns, and since we are claiming that VP-ellipsis involves some kind of anaphora it





is relevant to develop that treatment of pronouns here. Thus in the variable-free semantics of Jacobson () there is no use of variables as the meaning of pronouns (or any other expression) and no use of indices in the syntax. Rather, an expression such as he left is a function from individuals to propositions (rather than, as is standard, an assignmentdependent proposition) and his mother is a function from individuals to individuals (i.e., it is the-mother-of function). Quite central to the program is the treatment of ‘binding’ in something like Every man loves his mother but we can skip that here as it plays no role in the remarks below. Consider, then, the meaning of a pronoun. In a standard view with indices and variables, it is an individual relative to an assignment function (where every assignment function is a way to assign values to the variables). Here, there are no variables, no indices, and no assignment functions. Thus the hypothesis is that any expression containing a pronoun ‘unbound’ within that expression—including a pronoun itself—denotes a function from individuals to something else. More generally, we introduce an additional set of categories: If A is a category and B is a category then AB is a category. Like A/B, any expression of this category denotes some function in . The ‘slash’ and the superscript thus have a different syntax, but they have the same semantic effect. Let the lexical category of a pronoun be NPNP and let its meaning be the identity function on individuals—thus it is of type . The next question is how a pronoun (or something containing a pronoun) can be the subject of lost in a simple sentence like He lost (where the pronoun remains free) or in a more complex case like Every boyi fears that hei lost (where he lost is embedded and he is ultimately ‘bound’). Again we skip the mechanism for bound readings (it is effected by a unary rule operating on the category and meaning of fears), but the composition of he lost is straightforward. Simply extend the g rule given above so that something of category A/B can map not only to (A/C)/(B/C) (call that g-slash) but also to AC/BC (call that g-sup) with the same semantics as given above for g-slash. Intuitively, then, this allows something to combine with a pronoun (or any expression containing a pronoun unbound within it) and it ‘passes up’ the information that there is a pronoun. The consequence of this is that lost of category S/LNP can map to SNP/LNPNP and can combine with he. The result is he lost of category SNP with meaning (ignoring the contribution of gender here) the same as [[lost]]—a function from individuals to the proposition that that individual lost. So, a free pronoun is not represented in this system as a ‘free variable’ and He lost is not a proposition but rather a function from individuals to propositions. The listener, however, derives a proposition from this by applying this to some contextually salient individual. While it might seem odd to think of [[He lost]] as not a proposition but actually a function of type , it is important to realize that under a system with variables and assignment functions its semantic value is not a proposition either. It is an assignment-dependent proposition—technically, a function from assignment functions (ways to assign values to the variable, or indices) to a proposition, and the listener derives a proposition only by picking some contextually salient assignment. Here the listener needs to supply a contextually salient individual, rather than a contextually salient assignment. Before returning to VPE, we make one further generalization. We have seen that g can introduce either a ‘slash’ or a superscript to a would-be argument of a function, and pass that up to the result category. Let us generalize this so that not only can expressions of

   



category A/B input the g rule but so can AB (Jacobson a shows that this immediately accounts for the fact that pronouns have paycheck pronoun meanings). We won’t formalize this here—the idea simply is that an expression of category A/B can map to one of category (A/C)/(B/C) or one of category AC/BC, and an expression of category AB can map (B ) to (AC) or to (A/C)(B/C). All four of these have the same semantics: the semantics of g shown above. C

. VP-

..................................................................................................................................

.. Simple VP-ellipsis With this set of tools, we now make more formal the notion that VPE is a case of deep anaphora, where the ‘missing’ complement of an auxiliary is understood as a contextually salient property. To illustrate, take the auxiliary will listed in the lexicon with syntactic category VP/RVP—call this the lexical item will. (Since will takes only VPs with bare stems, assume that there is some additional feature on the argument VP. Moreover, auxiliaries form a distinct class in terms of their syntax, so presumably they all have a feature [AUX].) It is of course oversimplified to talk about the meaning of will while ignoring times, since [[will]] takes a time-dependent set and returns a time-dependent set (hence its type is ). But we nonetheless continue to extensionalize for convenience, and so assign [[will]] the simplified type . Then assume a unary rule whereby auxiliaries—i.e., items of the category VP/VP with the feature [AUX]—can map to another item with the same phonology and the same meaning but with category VPVP. In this sense, they are like other ‘proforms’ (although, with the possible exception of do, they do not denote identity functions). This then predicts the existence of a second item—call it will of category VPVP—which has the same meaning (and phonology) as will. In other words, what licenses auxiliaries to appear without overt complements is simply the fact that there is a unary rule mapping them to VPVP. Consider then a sentence like Bode will. Here we have the item will, whose category is VPVP. Bode can lift to be S/R(S/LNP), which (using our convenient abbreviation) is the same as S/RVP. Then by g-sup this maps to SVP/VPVP which takes will as argument. At the end of the day, then, Bode will is of category SVP, and the semantics is such that its meaning is λP[[[will]](P)(b)] or, slightly more informally, λP[Bode will P]. And this is supplied by the listener to a contextually salient function of type . In the case of someone hearing speaker B’s utterance in (), that function is the [[win a gold medal]] function which is made salient in virtue of having been named by speaker A.

.. Antecedent-Contained Deletion What about other cases of VPE such as Antecedent-Contained Deletion (ACD) discussed first in Bouton  and illustrated in ()? ()

Sarah will read every newspaper that Katie will.





It is generally assumed that this is just an instance of VPE—i.e., that whatever mechanism one adopts for a case like () should automatically extend to this case. Indeed, I assume that that is correct. The question, then, is whether the CG mechanisms sketched above accomplish this. Happily, the answer is yes; the existence of cases like () follows immediately. Moreover, ACD has been taken since at least as early as Sag (a) to necessitate a level of LF at which quantified objects are raised from their pronounced position. But—as shown in Cormack (), Evans () (for a different but related case), Jacobson (a,b, ,  etc.)—the CG mechanisms above mean that no LF (and no rule of Quantifier Raising) is needed. To illustrate, we need first to consider the parallel case without ellipsis: ()

Sarah will read every newspaper that Katie will read.

Standard wisdom is that the relative clause can be put together only if the material following will is a full VP (of the form read t). From this it follows under SLM that the ‘missing’ or silent material in the ACD case () must also be read t. We know that VPE at least certainly likes to have overt linguistic material to supply the relevant meaning (or form), but in () there is no material that it can supply unless the object NP is pulled out by QR giving a representation read t. (We have, of course, disputed the claim that there must be an overt antecedent. But since it is indeed difficult to get VPE out of context without this, we agree that ACD cases can be used as a diagnostic for the form and/or meaning of the ‘antecedent’—i.e., of the linguistic material that makes salient the meaning that is supplied.) But notice that under the CG account of relative clauses in general, no full VP is needed in the parallel non-ellipsis case in (). Rather, will is mapped by g-sl to a version which directly takes a transitive verb (rather than a VP) as argument. And so () (the ACD case) can also be seen as a matter of a missing transitive verb (phrase)—we call this TVP ellipsis.5 What is supplied is just the two-place relation [[read]] which is made salient by being the lexical meaning of the verb in the first conjunct. To detail how this works, the key point is that will can map not only to a ‘proform’ over VP meanings but also one over TVP meanings. Thus recall that it maps to VPVP. But because we have defined the g rule in full generality, that item in turn can map to a (VP/NP)(VP/NP)— intuitively a proform over transitive verbs. I will not spell out the full semantics (the interested reader can do that), but the ultimate meaning is λR[λx[[[will]] (R(x))]. And the superscript feature (and argument slot that corresponds to this in the semantics) is passed up, such that the ultimate meaning of () is (informally) λR[Sarah will read every x such that x is a newspaper and Katie will R(x)]. Thus, this is a function from two-place relations to propositions, and picks up the contextually salient [[read]] relation—made highly salient by being the lexical meaning of read in the first clause. 5 We use the term ‘transitive verb phrase ellipsis’ because it can pick up complex transitive verb meanings. For example, in (i) the two-place relation [[said that Suzy should read]] can be picked up as the missing complement of did. Note that the CG conventions allow said that Suzy should read in the first clause to combine together to give a meaningful expression of type .

(i)

The teacher said that Suzy should read every book that the principal also did.

For discussion, see Cormack () and Jacobson (a), who also show that the generalization in Sag (a) to the effect that these require a de re interpretation for the object also follows.

   



Since the case of TVP ellipsis follows here directly from the mechanisms needed for VPE in general, we should assure ourselves that it—like ordinary VPE—can pick up the missing meaning across sentences and can also occur with no antecedent. Example () (from Evans ) shows the first point, and () shows the second: () Bagels, I like. Donuts, I don’t. ()

Bill has just taken a tray of cookies out of the oven. They are still hot; a cool tray of cookies lies next to them. Seeing his daughter about to reach for the hot ones he says (pointing to the cool ones): Uh uh. These you may. But those you can’t, not until they have cooled down.

Finally, we should mention that Bouton () discussed another type of case of ACD exemplified in (): ()

Sally spoke to every boy who wanted her to.

In Jacobson (a, , and ) it is shown that the conventions here combined with the account of pronominal binding developed there immediately extend to this case too. Moreover, cases like () which were discussed later under the rubric of ‘rebinding’ (Merchant ) are also shown in Jacobson (a) to follow from the variable-free account; interested readers can consult Jacobson (a) for details: ()

Sally spoke to every boy. But only Tommy wanted her to.

.. Pseudogapping Interestingly, pseudogapping also follows immediately from a generalization of the conventions discussed here. Note first that the lift rule in () was formulated to only lift NPs to be of category S/(S/NP). But it can (and should) be stated more generally, with many useful consequences. For example, Dowty () shows that generalizing this allows for various coordination cases like Captain Jack served scallops yesterday and steamers today. Part of Dowty’s analysis involves lifting an object NP (e.g., scallops) over transitive verbs—thus this NP maps to VP/L(VP/RNP). (That then combines with yesterday whose category is VP/LVP by the conventions above generalized to the case of left as well as right slashes.) Again the particular directions of slashes on the output category of lift can follow from general conventions; see Jacobson . Now take a pseudogapping case like: ()

Captain Jack won’t serve steamers, but The Boathouse will scallops.

Let scallops lift as above to be VP/L(VP/RNP). But this can then also map by g sup in such a way as to introduce a VP/RNP superscript feature. To make reading this simpler, we abbreviate VP/RNP as TV. Hence lifted scallops maps to VPTV/LTVTV. The intuition here is that this category says ‘give me an anaphor over TVs to my left (such as will) and I’ll yield a VP with an unbound TV meaning within it’.





Thus nothing new needs to be said to in order to license the possibility of a ‘missing’ TV as is found in pseudogapping. Ultimately all of this comes from the unary rule mapping will to a VPVP—all of the rest follows simply from the existence of the generalized g rule. We would thus be unhappy if pseudogapping did not exist. Of course, many researchers might take the fact that pseudogapping comes ‘for free’ to be a disadvantage of the approach here, for it is well known that there are heavy constraints on pseudogapping and it does not show all the same behaviors as does VPE (or ACD). But arguably this does not show that the grammar itself should have a separate category of ‘pseudogapping’. Miller () suggests that these constraints all derive from independent principles of information structure and/or discourse and are not constraints imposed by the grammar.

. O C G   

.................................................................................................................................. We have concentrated here on one version of CG and, within this, one account of VPE and fragment answers. There are, however, other accounts of ellipsis within CG and accounts of additional constructions to which we can only give a few pointers here. First, Szabolcsi ()—also within a variable-free version of CG—proposes that VPE can be treated as a kind of ‘binding’ of a missing VP. One difficulty with treating all cases of VPE this way is that it cannot handle VPE across sentences and speakers (although one might try to pursue a dynamic approach to binding to extend to such cases), nor can it handle the cases of VPE without overt antecedents. Charlow (), however, points out that this does not mean there are no instances of ‘bound’ VPs, and uses the variable-free apparatus of Jacobson () to extend to VPE in such a way that VPE is similar to pronouns: the grammar makes available both ‘bound’ cases and cases where a contextually salient property is supplied (as in the account here). He argues further that focus provides evidence for having both mechanisms. We have said nothing here about gapping; see Steedman () for one approach within CG, and Kubota and Levine (, ) provide a somewhat different account within a different variety of CG. See also Kubota and Levine () for an analysis of pseudogapping quite different from the one presented here. Of course, this is by no means an exhaustive list of ellipsis work in CG, but hopefully gives the reader pointers to additional work of relevance.

. C

.................................................................................................................................. As noted in the introduction, an important aspect of CG for a theory of ellipsis comes from the hypothesis of direct compositionality, which would make it surprising to find silencing of linguistic material under identity with other overt material in the discourse context. But we have seen that no SLM is necessary; VPE and fragment answers both have rather natural analyses without this. Moreover, the traditional arguments for SLM are not strong: some in

   



fact rely on a non-DC worldview so become irrelevant under DC. Finally, we have seen that given some simple CG apparatus (in particular, a generalized g rule) it is easy to account for not only VPE but also ACD and pseudogapping as special cases. The ACD case is particularly interesting, in that under this view there is also no reason to believe that ACD necessitates a level of Logical Form; its existence and interpretation are entirely compatible with Direct Compositionality.

A This research was supported in part by NSF Grant BCS.

  ......................................................................................................................

    ......................................................................................................................

 

. D- 

.................................................................................................................................. D grammars (DGs) assume that hierarchical relationships between words are immediate; these relationships are not mediated by the intermediate groupings associated with phrase structure grammars (PSGs ≈ constituency grammars). Dependencies are directed; they exist between a head word and its dependents. A brief illustration of a dependency tree is provided next, and the corresponding phrase structure tree is included for comparison: ()

V N D

A P

A

N a. The hierarchical relationships between words are direct. S NP D

A

VP N

PP P

V

A

N

b. The hierarchical relationships between words are indirect.1

The phrase structure tree (1b) is flatter than many modern phrase structure grammars would assume; it allows n-ary branching, as opposed to just binary branching. This ‘flat’ analysis is given here in the interest of promoting the comparison. Dependency hierarchies are relatively flat by their very nature. 1

   



A main difference between the dependency tree (a) and the phrase structure tree (b) concerns the presence/absence of the phrasal nodes (NP, PP, VP, S). Dependency trees lack these purely phrasal nodes;2 the hierarchical relationships between syntactic units are therefore direct. The distinction between dependency- and constituency-based approaches to natural language syntax illustrated with trees (a) and (b) is discussed in numerous places (e.g. Matthews : –, Fischer : ch. .., and Osborne ).

. E DG   

.................................................................................................................................. When one considers ellipsis through the lens of the theories of syntax that assume dependencies more or less along the lines of (a), one learns that many established DGs have not been so concerned with ellipsis. There are, though, a number of DGs that have at least touched on ellipsis. Some of them will now be briefly considered. Lucien Tesnière, the French linguist who is widely acknowledged as having produced the first comprehensive dependency-based theory of syntax, addressed ellipsis in his main work Elements of Structural Syntax (/) mostly only in connection with his subtheory of transfer (Fr. translation). He was hence interested in lexicalized forms of ellipsis, rather than in the productive types of ellipsis that more recent theories of syntax acknowledge and explore. However, Tesnière’s analysis of coordination includes two examples that allow one to assume that he was willing to posit the existence of null material in (at least some) cases of ellipsis. The relevant examples involve what would be viewed as stripping in our modern times (Stemmas  and ): ()

Alfred aime les roses, non les épines. Alfred loves the roses, not the thorns aime et aime Alfred

les roses les épines

non

(Adapted slightly from Tesnière 1959: Stemma 271) Tesnière’s Stemma in this case posits an empty node corresponding to a second occurrence of aime ‘love’. Interestingly, however, Tesnière did not posit a null node in related cases involving gapping—see his Stemma . The tendency to assume null nodes in cases of ellipsis is also present in Weber (). Weber’s approach to coordination, and to syntax more generally, is based closely on Tesnière’s work. Weber does something that Tesnière did not do, though: he posits empty nodes to accommodate cases of gapping (although Weber does not use the term gapping), as the following example from German demonstrates: The qualification ‘purely’ points to the fact that every node in a dependency tree can be viewed as both lexical and phrasal. That is, every node marks the presence of a word as well as the phrase that that word heads; the nodes in DG structures are hence playing two roles at the same time. In contrast, these roles are played by distinct sets of nodes in PSGs. 2

 ()

 Berthold liebt seine Arbeit, und Bernhard liebt seine Freundin. Berthold loves his work, and Bernhard loves his girlfriend. liebt – und – [-] Berthold – und – Bernhard

Arbeit Freundin seine

seine

(Adapted from Weber 1999: 74) The [-] marks a null instance of the verb liebt ‘loves’. For Weber this null element is necessary to maintain an account that in this case sees coordination as necessarily coordinating clauses. Heringer (: , –, –) also uses null elements to address aspects of ellipsis. His understanding of these null elements is in terms of deletions (Ger. Tilgungen). He relies heavily on deletions to accommodate aspects of coordination. For instance, he assumes deletions in cases of gapping: () G. arbeitete vor allem mit Chlor, und seine Kollegen arbeiteten mit Benzol. G. works above all with chlorine, and his colleagues work with benzene. (Heringer : ) And he assumes deletions in cases of VP-ellipsis, e.g.: () Keiner will eigentlich das Chlor mit Benzol zusammenmixen, nobody wants actually the chlorine with benzene mix aber alle müssen das Chlor mit Benzol zusammenmixen. but all must the chlorine with benzene mix (Heringer : ) ‘Nobody actually wants to mix the chlorine with benzene, but everyone must mix the chlorine with benzene.’ Heringer’s discussion of ellipsis is like other DG accounts, though. These accounts merely touch on aspects of ellipsis, without using established terminology. There is no systematic effort to investigate one or more of the various types of ellipsis and to determine their distributions. In his dependency-based syntax of German, Hans-Werner Eroms (: ff.) addresses ellipsis also mainly just with respect to certain instances of coordination. He assumes the presence of empty nodes in order to accommodate certain readings, e.g.: ()

Elefanten große schwarze

und

Elefanten kleine weiße

große schwarze Elefanten und kleine weiße Elefanten large black elefants and small white elefants (Adapted from Eroms 2001: 472)

   



According to Eroms, the null node in this case is necessary to accommodate the only available reading, the one involving two groups of elephants. It is, namely, impossible for an elephant to be both big and black and small and white. In his prolific works on the dependency-based Meaning–Text Theory (MTT), Igor Mel’čuk has hardly addressed ellipsis.3 He produces a brief mention of the necessity to posit an empty node in the case of gapping: ()

went Alan

and to

went Leo

to

Paris

Singapore

Alan went to Paris, and Leo went to Singapore. (Adapted from Mel’čuk 2003: 219) Mel’čuk comments that the empty node is necessary in such cases in order to account for the fact that in many languages the elided verb imposes lexical and inflectional choices on its dependents. The dashed line marks an anaphoric relation across the elided verb and its antecedent. Hudson (: –, ) provides a bit more discussion of ellipsis in his framework Word Grammar. He produces empirical observations that support an approach to ellipsis in terms of “unrealized words.” One of the main empirical considerations he produces is the same point that Mel’čuk makes, and it is in fact the point that appears time and again as justification for the existence of empty structure in many cases of ellipsis: in languages that have morphological case, NPs are marked for a certain case despite the absence of an overt case assigner. This point is illustrated with an example of sluicing from German: ()

Sie ist jemandem begegnet, aber she is someone-dat run.into, but sagt sie

nicht wem begegnetg sie

ist

sie sagt nicht wem sie begegnet ist. she says not whom-dat she run.into is ‘She ran into someone, but she isn’t saying who.’ This example involves an interrogative proform in an embedded clause. The analysis follows the approach developed by Groß and Osborne (), which assumes rising. The wh-element wem ‘whom’ has risen to become the root of the embedded interrogative clause. The dashed 3

This statement is based on personal communication with Mel’čuk.





dependency edge and g subscript indicate the presence of rising.4 The notion of rising is not to be understood literally, but rather it is merely a convenient metaphor for denoting certain constellations, in particular those that contain (what would otherwise be) discontinuities.5 The particular analysis of fronting shown in () is secondary to the point at hand. This point is that one can account for the dative form of the interrogative pronoun wem by assuming that there is a covert instance of begegnet present as shown. The verb begegnen ‘run into’ is unlike most transitive verbs in German in that it assigns dative case instead of accusative. Without the covert presence of this case assigner, it would be more difficult to account for the appearance of the dative wem. Plewnia (, ) has worked directly on ellipsis using a DG approach. In his  dissertation he posits “ghost structures” to accommodate ellipsis. These structures appear in the third dimension on a plane behind the surface plane. In his  article, in contrast, he shows the presence of null structure directly in the dependency tree on the surface plane. He explores a type of ellipsis in German that omits one or more verbs from the right edge of a clause. He takes the following example from Kurzeck’s novel Vorabend ‘Evening Before’ (: ): () Ob du einen Beitrag für den Almanach whether you a contribution for the almanac schreiben kannst, sagt Sibylle am Abend zu mir. write can says Sibylle at evening to me ‘Whether you (can write) a contribution for the almanac, said Sibyl to me in the evening.’

Ob kannst schreiben

du Beitrag

für Almanach

einen den Ob du einen Beitrag

für den Almanach schreiben kannst

The verb chain schreiben kannst ‘can write’ can be interpreted as elided from the right edge of the embedded clause. Plewnia produces a number of similar examples, all taken from Kurzeck’s novel. These examples allow one to interpret the elided material as verbal in nature. Interestingly, Plewnia does not give the ellipsis mechanism involved a name. The term right-edge ellipsis (REE) might be appropriate. 4 The dashed dependency edge identifies a head (here wem) that is not the governor of (one of) its dependent(s), and the g subscript marks the governor of the risen catena. In the case of (8), begegnet is the governor of the risen wem. 5 Rising is the means by which Groß and Osborne () and Osborne () address discontinuities (long-distance dependencies). ‘Rising’ denotes a constellation in which a given word takes on another word as its head that is not its governor. In DG parlance, it is the means by which projectivity violations can be avoided in surface syntax. There is unfortunately not room here to introduce and illustrate rising more extensively, so the reader is referred to the articles cited. The reason an example involving rising is given here is that sluicing is a common type of ellipsis that involves a risen wh-expression. More instances of sluicing appear in section ..

   



This brief survey of a few previous approaches to ellipsis in the DG tradition has established two points. The first is that ellipsis has not been systematically investigated in many established DG frameworks. Detailed accounts of various ellipsis mechanisms (e.g., gapping, stripping, VP-ellipsis, sluicing, pseudogapping, answer fragments, left-edge ellipsis, etc.) are rare in the DG tradition, if they exist at all. The second point is that when DGs do address ellipsis phenomena, there is a strong tendency to posit the existence of empty structure, i.e. null nodes, to accommodate central aspects of ellipsis. This empty structure is sometimes shown in dependency trees as empty nodes, and at other times it is given in strings in terms of deletions. This being said, there has in fact been a recent development in the DG tradition that can provide a solid foundation for the exploration and analysis of ellipsis in general. This development is the catena unit. The catena was proposed by O’Grady () as the basis for his analysis of the syntax of idioms, although O’Grady used the term chain instead of catena. The concept has since been seized upon by others and extended to other phenomena of syntax, including ellipsis (see Osborne  and Osborne et al. , ). The claim is that ellipsis mechanisms are eliding catenae, whereby these catenae often fail to qualify as constituents. This point is developed in some detail in section .. Before turning to the catena, however, the discussion considers the nature of the elided material, i.e. the structure of ellipsis sites, and the relationship that can or must obtain between an ellipsis site and its antecedent or postcedent, i.e., recoverability.

. S

.................................................................................................................................. The strong tendency for DGs to assume null nodes in cases of ellipsis is easy to understand, since without the covert nodes, ellipsis would often result in dependency hierarchies that are disconnected. The following two competing analyses of an instance of gapping illustrate this point about connectedness: ()

Tom arrived yesterday, and arrived Susan

today with children her

a. Susan arrived today with her children.6 Susan

today with children her

b. Susan 6

today with her children.

The arrow dependency edges in (10a) identify adjuncts (as opposed to arguments or complements). The practice of using arrows to identify adjuncts has precedent in the DG literature, although the exact nature of the arrows employed and the exact nature of what they signify varies considerably. See Tesnière (1959: ch. 21); Engel (1994: 43ff.); Eroms (2000: 217ff.), and Mel’čuk (2003: 193).





Example (a) posits an empty node corresponding to the elided verb arrived, whereas example (b) assumes that no such empty node is present. By positing the presence of the empty node, the analysis in (a) can maintain the widespread assumption that dependency hierarchies are collections of connected nodes (words), which in most DGs are assumed to be trees. The difficulty with the analysis in (b) should be apparent in this regard, since without the null node corresponding to the elided verb arrived, the three subtrees in the gapped clause (Susan, today, and with her children) end up disconnected. Many syntactic structures involving ellipsis would become a seemingly arbitrary mish-mash of disconnected parts, rendering a coherent analysis of ellipsis phenomena more difficult.7 While there is thus strong motivation for DGs to assume that ellipsis involves null structure, the actual status of this empty structure can be debated, as stated in section .. The approach to ellipsis pursued in this contribution assumes that empty structure is present in the form of null nodes, but these null nodes often do not have the same syntactic status as the corresponding overt words. In other words, the empty nodes enjoy a unique status. The motivation for granting them a unique status can be seen in the following example of an answer fragment: ()

did I

did it

Who did it? a. – *I did it.

Me vs

it

b. – Me did it.

The object form of the pronoun me is acceptable as an answer fragment, whereas the subject form of the pronoun I is not acceptable. When ellipsis has not occurred, in contrast, the acceptability judgments are reversed: I did it vs *Me did it. Such data show that if answer fragments are construed as involving null nodes, these null nodes must have a special status; they cannot have the same syntactic status as the corresponding overt words. A second observation demonstrating that the null nodes of (at least some) ellipsis mechanisms must enjoy a unique status occurs with negation. Many ellipsis mechanisms are incapable of including clausal negation in the elided material, e.g.: () a. b. c. d.

7

*You did not call me, and I did not call you.8 *Sam did not call you, and Jill did not call you, too. *She has not tried because he has not tried. *He does not address you more than he does not address me.

– Failed gapping – Failed stripping – Failed VP-ellipsis – Failed pseudogapping

Note that this point about connectedness is not valid for PSGs, since they can and often do posit empty structure independently of whether ellipsis is deemed to have occurred. This is particularly true of the X-bar schema. Often a given head word appears without a specifier or complement, which means the specifier and complement positions are unoccupied. And when movement occurs, the moved unit vacates a syntactic position, leaving that position unoccupied in a sense. 8 Sentence (12a) is actually acceptable, but on a different reading from the one indicated: ‘It is not the case that I called you and you called me’. This reading is such that the negation scopes over the entire sentence. The availability of this other reading is consistent with the claim here, namely that ellipsis cannot be eliding the negation. A similar comment applies to example (12b).

   



The indicated reading is not available each time because the ellipsis mechanism at hand cannot elide the negation not.9 If the null nodes of ellipsis had the same syntactic status as their corresponding overt counterparts, there would be no reason to expect these acceptability judgments. Apparently, no null node of gapping, stripping, VP-ellipsis, or pseudogapping can be the equivalent of clausal negation.10 The two points just produced demonstrate that the empty structure of at least some ellipsis mechanisms (answer fragments, gapping, stripping) cannot be construed as having the same syntactic status as the corresponding overt words. Given this state of affairs, an approach to ellipsis is warranted that takes the middle ground. This middle ground views ellipsis as involving empty structure in the form of null nodes, but these null nodes enjoy a special status; they are not syntactically equivalent to the corresponding overt words.

. R

.................................................................................................................................. The question of recoverability concerns how an ellipsis site gets its meaning. Many occurrences of ellipsis are such that the meaning of the ellipsis site is retrieved from overt material in the linguistic context. The elided material often stands in an identity relationship with an antecedent or postcedent. Recoverability concerns the nature of this relationship; is it syntactic, semantic, or pragmatic? Since the DG tradition has for the most part not devoted much effort to exploring ellipsis phenomena, as discussed in section ., the answer to this question is not established. However, the stance adopted in this contribution points to the fact that ellipsis mechanisms are a heterogeneous bunch; the one ellipsis mechanism can behave very much unlike the next with respect to recoverability. This point will now be demonstrated by considering a few central traits of three ellipsis mechanisms: gapping/stripping, VP-ellipsis, and left-edge ellipsis (LEE). Dependency-based tree structures are employed for illustration. Gapping and stripping are quite unlike other ellipsis mechanisms in a number of ways. Above all, they occur in the non-initial conjuncts of coordinate structures only, whereby there is always an overt antecedent in the initial conjunct that supplies the meaning of the gapped/stripped material. In other words, there is a clearly defined syntactic relationship that obtains between the ellipsis and its antecedent. For one, there must be an antecedent; postcedents are not possible, and the ellipsis usually matches this antecedent quite closely, e.g.:

9 In contrast to gapping, stripping, VP-ellipsis, and pseudogapping, other ellipsis mechanisms (sluicing, answer fragments, null complement anaphora) can easily elide a negation, e.g. He did not do it, but he won’t say why he did not do it. 10 The qualification “clausal” is necessary. An editor points out that a negation that is confined to a VP can be elided, e.g. I have deliberately not spoken with Joe, and you have deliberately not spoken with Joe, too.

 ()

 Did

or he offend

did you offend him

you Did he offend you, or

did you offend him?

The antecedent did . . . offend matches exactly the elided material. Some minor variation is possible, for instance concerning number and or person morphology, e.g., He likes Susan, and you like Susan, too (likes vs like). In general, though, the identity relationship across ellipsis and antecedent is clearly defined in syntactic terms. The relationship between ellipsis and antecedent (or postcedent) is less stringent for VP-ellipsis. The elided VP usually has an antecedent, but postcedents are also possible. Furthermore, ellipsis and antecedent in cases of VP-ellipsis need not appear in the conjuncts of a single coordinate structure, but rather the one can appear in a clause that is superordinate to the clause containing the other. These points are illustrated with the following examples: ()

is Larry

also trying hard because is Phil

trying hard

a. Larry is also trying

hard because Phil is trying hard. is

Because

Larry is Phil

also tryingg hard

trying hard

b. Because Phil is trying hard, Larry is also trying hard. c. Because Phil is trying hard, Larry is also trying hard. In example (a) and (c), the ellipsis has an antecedent, whereas in example (b), it has a postcedent, and in example (c), the ellipsis appears in the matrix clause at the same time that its antecedent is located in the subordinate clause. Interestingly, however, there is in fact a solid syntactic restriction on recoverability in cases of VP-ellipsis. The material that provides the content of the ellipsis (i.e., the antecedent or

   



postcedent) must precede or govern the ellipsis. In other words, if the ellipsis has a postcedent, this postcedent must govern the ellipsis. If it does not, VP-ellipsis is impossible, e.g.: ()

is Larry

trying hard because is Phil

trying hard

d. *Larry is trying hard because Phil is trying hard. This instance of VP-ellipsis fails because the elided material governs its postcedent, that is, the elided trying hard (indirectly) governs the overt trying hard. Based on this observation, there is in fact a solid syntactic restriction on the recoverability relationship that obtains between the ellipsis and antecedent/postcedent of VP-ellipsis. Be that as it may, the recoverability relationship for VP-ellipsis is indeed much more flexible than for gapping and stripping. As long as the VP-ellipsis has an antecedent, this antecedent can appear at a considerable distance to the ellipsis; ellipsis and antecedent can even appear in separate sentences that are separated by a good amount of other linguistic material, e.g.: ()

I have to work. It’s an unfortunate situation. I also think that Bill does have to work, too.

And it is even possible for the ellipsis to lack a linguistic antecedent altogether. The antecedent can be supplied by the situational context, e.g.: () Situational context A boy picks up a rock and gets ready to throw it into a pool of water. His mother barks out the command: Don’t do that! (do that = throw the rock) And there is yet a further aspect of VP-ellipsis indicating that the recoverability relationship between ellipsis and antecedent/postcedent is flexible. The ellipsis and antecedent can at times be dissimilar in syntactic category. The widely discussed mismatches in voice (active vs passive) illustrate this flexibility: ()

The dishes should have been scrubbed, but we didn’t scrub the dishes.

The antecedent is passive (the dishes . . . been scrubbed), whereas the elided material is active (scrub the dishes). This sort of flexibility across ellipsis and antecedent is impossible in cases of gapping and stripping, e.g., *The dishes were scrubbed by Tom, and Bill scrubbed the dishes, too.





Moving to a third type of ellipsis, i.e., left-edge ellipsis (LEE), it is quite unlike both gapping/stripping and VP-ellipsis in the ways under discussion. LEE occurs in relaxed registers (everyday conversations, emails, text messages, diary entries, etc.).11 It elides material from the left edge of an utterance. The elided material is of a sort that can be easily inferred from the situational context. First- and second-person subject pronouns as well as auxiliary verbs are thus typical candidates for omission in terms of LEE, e.g.: () a. b. c. d. e. f.

Do you Understand? Do You understand? It is Time for some traffic problems in Ft. Lee. Does anyone Want to play ball? Does Anyone want to play ball? We Arrived at midnight. We Went straight to bed.

It should be apparent that the question of recoverability receives a much different answer when the focus is on LEE. A linguistic antecedent is typically not present at all, but rather the elided material is inferred from a conventionalized context. In a context where something is being explained, the question Understand? is immediately understood despite its fragmentary nature; the elided material Do you is easily inferred. The same is true of diary entries that elide first-person pronouns. Diary entries are typically about the author’s daily experiences, whereby the first person, i.e. the author, is front and center. In such a conventionalized context, it is not necessary to continually repeat the first-person pronoun I or we. What these observations mean for recoverability in cases of LEE is that it is pragmatic in nature. Situational context is responsible for recovering the elided material, not the presence of a linguistic antecedent or postcedent. This section has considered the question of recoverability using DG tree structures for illustration. The main message has been that ellipsis mechanisms are a heterogeneous bunch. The one ellipsis mechanism can be quite unlike the next in terms of the recovery of the elided material. Gapping and stripping are very stringent in this area. They require the presence of a linguistic antecedent that immediately precedes the elided material and matches it rather closely in syntactic form. VP-ellipsis is more flexible than gapping and stripping regarding recoverability, although it too places a concrete syntactic restriction on the relationship between ellipsis and antecedent/postcedent. LEE, in contrast, is altogether different regarding recoverability. The recovery of elided material in cases of LEE occurs through a conventionalized context, meaning that recovery is pragmatic in nature, not syntactic. In sum, then, there is no one answer to the question of recoverability when addressing ellipsis in general. There are, rather, a number of distinct answers that vary according to the ellipsis mechanism at hand.

Aspects of the LEE mechanism are addressed in at least five places (Napoli ; Wilder ; Kay ; Merchant a; Fitzpatrick ). The term itself, i.e. left-edge ellipsis, is borrowed from Wilder (: ). These accounts of LEE vary significantly. Kay (), for instance, is interested only in LEE in cases of tag questions (e.g. It is Difficult, isn’t it?) and Fitzpatrick () is concerned only with the ellipsis of an auxiliary verb (e.g. Is Anyone interested in dessert?). 11

   



. L

.................................................................................................................................. The area in which DG can make a major contribution to the theory of ellipsis in general concerns licensing. Licensing is understood here as the syntactic constraints that restrict the material that the various ellipsis mechanisms can elide. Consider the following instances of ellipsis in this respect: () Would you want to say that to me, or would I want to say that to you? () Someone played a prank on someone, but we don’t know who played a prank on whom.

– Pseudogapping

– Sluicing

Example () is an instance of pseudogapping, and example () an instance of sluicing. The elided material in each case—want to say that in () and played a prank in ()—does not qualify as a constituent in most theories of syntax. This situation presents a challenge, since it requires that the theory of ellipsis somehow account for the fact that many ellipsis mechanisms appear to be eliding non-constituent units. Many theories of syntax take the constituent to be the basic unit of syntactic analysis, which means that when they are confronted with data like ()–(), they choose to augment the theory of syntax in one way or another to account for the fact that many ellipsis mechanisms appear to be eliding non-constituent units. Perhaps the favored approach in this respect is to assume movement first, followed by ellipsis second (e.g. Sag a; Jayaseelan ; Johnson b, ; Merchant , a, to name just a few). The remnant—to you in () and on whom in ()—first vacates an encompassing constituent so that the encompassing constituent can then be elided. In this manner, a constituent-based theory of ellipsis can be maintained. There is a significant problem with this sort of approach, however. There is little empirical evidence supporting the assumption that movement of the remnant(s) occurs (and the evidence that is produced is contrived). In the cases of () and (), there is little empirical support for the stance that to you in () and on whom in () are each first moved out of the relevant encompassing constituent so that this constituent can then be elided. In fact, the movement-first-ellipsis-second approach in general is ad hoc. It bears witness to a flawed interpretation of syntactic structure. The basic unit of syntactic analysis is not the constituent, but rather the catena. As stated above in the introduction, the catena unit was introduced by O’Grady (), and it has been developed much further in a series of recent articles (Osborne , , ; Osborne et al. , ; Osborne and Groß ; Groß and Osborne ). One of the main empirical insights backing the catena unit is the observation that the elided material of ellipsis mechanisms are catenae. The elided strings want to say that in () and played a prank in () qualify as non-constituent catenae. The following sections first introduce the catena unit and then demonstrate that ellipsis mechanisms of all sorts are eliding catenae. By acknowledging the role of catenae, a major step in the direction of a coherent and comprehensive theory of the licensing of ellipsis is accomplished.





.. Catenae The catena unit is defined as follows: Catena A word or a combination of words that is continuous in the dominance (vertical) dimension.12 The catena stands in contrast to the string, which is defined as follows: String A word or a combination of words that is continuous in the precedence (horizontal) dimension. The catena unit is an inclusive unit of syntactic analysis, much more inclusive than the constituent of phrase structure grammars. In other words, many more word combinations in most sentences qualify as catenae than as constituents. The catena unit is illustrated with the following tree structure. The capital letters serve to abbreviate the words: ()

contains C sentence B This A

catenae E numerous D

This sentence contains numerous catenae. Each individual word counts as a catena by definition, and each combination of words is a catena if the words are continuous in the vertical dimension: that is, if they are linked together by dependencies. All the distinct catenae in () are listed next: A, B, C, D, E, AB, BC, CE, DE, ABC, BCE, CDE, ABCE, BCDE, and ABCDE. There are hence fifteen distinct catenae in (). Compare this number with the total number of distinct word combinations in the sentence,  (=5 ).13 This means that there are  (= ) distinct word combinations in () that fail to qualify as catenae, e.g. AC, AE, BD, ABD, ABDE, etc. As the number of words present in the sentence at hand increases, the discrepancy between the number of distinct catena and non-catena combinations grows. Hence, while the catena is an inclusive unit of syntactic analysis, it is not overly inclusive. In most sentences, the number of non-catena word combinations far outnumbers the number of catena combinations. The claim put forth has been that the catena unit is inclusive enough to acknowledge those word combinations that are pertinent for the exploration of syntactic phenomena and at the same time, exclusive enough to exclude the word combinations that are not relevant for the exploration of syntactic phenomena.

12

The catena unit is defined over DG trees more formally as follows:

Catena (formal definition) Given a dependency tree T, a catena is a set S of nodes in T such that there is one and only one member of S that is not immediately dominated by any other member of S. 13 The total number of distinct word combinations is calculated using the formula 2n 1, where n = the number of words/nodes present.

   



The potential of the catena unit for the theory of ellipsis becomes apparent when one begins to examine the word combinations that typical ellipsis mechanisms elide. The claim put forth is that ellipsis elides catenae; ellipsis mechanisms of all sorts are eliding catenae. This claim is expressed concretely here as the Catena Condition: Catena Condition An instance of ellipsis elides a catena. The following subsection presents evidence supporting this claim. Examples of numerous types of ellipsis are considered.

.. The evidence This section produces evidence from gapping, VP-ellipsis, pseudogapping, sluicing (including multiple sluicing), and answer fragments supporting the claim that ellipsis mechanisms are eliding catenae. The examples discussed are consistent insofar as when ellipsis is possible, the elided material has the status of a catena, whereas attempts at ellipsis where the elided material is not a catena are usually bad. A type of exception to the Catena Condition is also acknowledged. Sentence () is a well-known example from Ross (: ); it is expanded here to illustrate the Catena Condition:

want

()

I

to try to begin to write novel a

I

want to try to begin to write a novel, and

a. Mary wants to try to begin to write a play. b. Mary wants to try to begin to write a play. c. Mary wants to try to begin to write a play. d. Mary wants to try to begin to write a play. e. *Mary wants to try to begin to write a play. f. *Mary wants to try to begin to write a play. g. *Mary wants to try to begin to write a play. h. *Mary wants to try to begin to write a play. i. *Mary wants to try to begin to draft a play.





Examples (a–d) are all at least somewhat acceptable instances of gapping, and in each of them, the elided material qualifies as a catena. Examples (e–i), in contrast, are all unacceptable on the readings shown. They are unacceptable because the elided material is not a catena. For instance, the elided word combination wants . . . to begin to write is not a catena in (e), nor is wants . . . a play in (i). The same is true of (f–h); in each of these examples, the elided word combination shown fails to form a catena. Note that examples (e–h) are actually acceptable instances of gapping if an alternative reading is assumed, one in which just the verb wants has been elided. The point here, however, is that the readings shown in (e–i) are not available. The elided material of VP-ellipsis is, trivially, always a catena, e.g.:

admits

()

He

– VP-ellipsis

mistakes before his

will she

admit mistakes her

a. He admits his mistakes before she will admit her mistakes. b. *He admits his mistakes before she will admit her mistakes. VP-ellipsis usually elides complete non-finite VPs (here admit her mistakes). On a dependency-based analysis of syntactic structure, non-finite VPs are complete subtrees, and complete subtrees are, trivially, catenae. As complete subtrees, non-finite VPs are a particular subtype of catena, namely ones that include all the words that the root non-finite verb dominates. Example (b) is bad precisely because the elided words admit her do not qualify as a complete subtree—they do not even qualify as a catena. The next examples involve pseudogapping: ()

He will want you to stay more than will

he

want

– Pseudogapping me to stay

a. he will want me to stay. b. *she will want you to stay. c. *she will want you to stay. Pseudogapping is like VP-ellipsis in that the gap is introduced by an auxiliary verb (here will), and it is like gapping in that there is (at least) one remnant present (here me). Note that the elided words in (a) are not a string; want . . . to stay is not a string. Despite this

   



fact, it is a catena because want immediately dominates to and to immediately dominates stay. Examples (b) and (c) are bad in part because the elided material shown—will . . . you in (b) and will . . . you stay in (c)—does not form catenae, and they are also bad insofar as they do not actually qualify as attempts at pseudogapping, since the gap of pseudogapping must be introduced by an overt auxiliary verb. Sluicing occurs in matrix and embedded interrogative clauses that are introduced by wh-expressions. In most cases, everything is elided from the sluiced clause except the wh-expression. In certain cases, however, more than just a/the wh-expression can appear in the sluiced clause. These cases are of particular interest here since they provide strong evidence supporting the catena, e.g.: ()

They are laughing at someone, but I don’t know at

whom

are they

– Sluicing laughingg

a. at whom they are laughing. b. *at whom they are laughing. who are they

– Sluicing laughing atg

c. who they are laughing at. d. who they are laughing at. e. *who they are laughing at. These trees show particular aspects of the approach to discontinuities put forth by Groß and Osborne ()—see examples () and (b), and nn.  and . These aspects are of secondary importance to the point at hand, this point being the fact that the elided material of sluicing qualifies as a catena. In the acceptable instances of sluicing (a), (c), and (d), the elided words they are laughing (at) form a catena, whereas in the unacceptable attempts at sluicing in (b) and (e), the elided words they . . . laughing do not qualify as a catena. Observe the alternative word orders across (a) and (c–d). Despite the presence of the overt preposition in (c), the elided material they are laughing still qualifies as a catena. Examples (b) and (e) are bad because the presence of are means that the elided words they . . . laughing fail to qualify as catenae. Instances of so-called multiple sluicing (Merchant ) provide further support for catenae. Multiple sluicing occurs when two (or perhaps more) wh-expressions are present in the sluiced clause. In such cases, only one of them can be fronted; the other remains in situ, e.g.:





()

Jack lived with different people at various times in college, but I don’t know when

livedg

– Multiple sluicing with

he

whom a. when he lived with whom. with whom

livedg he

– Multiple sluicing when

b. with whom he lived when. who lived he

– Multiple sluicing withg when

c. ?who he lived with

when.

Which of these variants is preferred will likely depend on the register in which the sentence is uttered and the speaker’s choice. Regardless of which variant actually occurs, the elided words he lived form a catena in each case. Answer fragments provide particularly strong support for the role of catenae in ellipsis. A given answer fragment is usually a complete subtree (a phrasal constituent in phrase structure grammars), which means the elided material, which corresponds to the rest of the sentence, is necessarily a catena, e.g.: ()

tried friend

to steal

Sam’s

music Bill’s a. b. c. d.

Whose friend tried to steal Bill’s music? – Sam’s Who tried to steal Bill’s music? – Sam’s What did Sam’s friend try to do? – Sam’s Whose music did Sam’s friend – Sam’s try to steal? e. What did Sam’s friend try to steal? – Sam’s 14

friend friend friend friend

tried tried tried tried

to to to to

steal steal Steal steal

Bill’s Bill’s Bill’s Bill’s

music.14 music. music. music.

friend tried to steal Bill’s music.

Examples (27a) and (27d) actually involve two ellipsis mechanisms, N-ellipsis (also known as NP-ellipsis or NPE) as well as the answer fragment mechanism. N-ellipsis is responsible for eliding friend

   



In each of (a–e), the answer fragment qualifies as a complete subtree, i.e. a constituent, which means that the elided material is necessarily a non-constituent catena. This is so even when the elided material is not a string, as in example (d) where the overt Bill’s interrupts the elided material in the linear dimension. Answer fragments in which the elided material does not qualify as catena are bad, e.g.: ()

tried friend Sam’s

to steal music Bill’s

f. Who that Sam knows tried to steal Bill’s music?

– *Sam’s friend tried to steal Bill’s music.

. What happened con- – *Sam’s friend tried to steal Bill’s music. g. cerning Sam’s friend and stealing Bill’s music? h. What did Sam’s friend try to do concerning Bill’s music?

– *Sam’s friend tried to steal Bill’s music.

i. What of Bill’s did Sam’s friend try to steal?

– *Sam’s friend tried to steal Bill’s music.

Each of these answer fragments is bad because the elided material is not a catena. Observe that the questions, which are attempting to elicit the answer fragments indicated, are themselves awkward. This awkwardness is an additional indication that the syntax of answer fragments is sensitive to catenae. Forming questions that might elicit non-catena fragment answers is itself difficult to do, if not impossible. While the discussion so far has revealed that many ellipsis mechanisms that have been widely acknowledged and explored in the literature on ellipsis are consistent with a catenabased approach to syntax, there are narrow exceptions involving answer fragments that must be conceded. Certain types of questions are capable of focusing an individual word that does not qualify as a complete subtree of dependency structures (because that word dominates other words), and when this occurs, the answer fragment that is produced can fail to qualify as a DG constituent, which means, in turn, that the elided material no longer qualifies as a catena,15 in (27a) and music in (27d). Despite the presence of N-ellipsis in these two examples, the other elided material, i.e. the other material not elided by N-ellipsis, still counts as a catena, as examples (27b) and (27e) demonstrate. 15

Observe that answer fragments can easily fail to qualify as DG constituents but still be acceptable. This occurs when there are two or more remnants present in the fragment, as in, for example, sentence (26a). In such cases, though, the elided material is a catena.

 ()

 Are you going to BUY or RENT your skis? am

I

going to Buy them

– I am going to Buy them. e.g.: The or-question can focus specific words (here buy and rent), whereby these words are not complete subtrees. In such cases, the answer fragment that is produced also need not be a complete subtree, which means the elided material is not a catena. A similar type of exception to the Catena Condition involves the interrogative words who and what. When they appear in an echo question, these words can also focus an individual word that does not qualify as a complete subtree of dependency structures—e.g. You saw Bill’s what? – I saw his Unicorn—which means the elided material is, again, not a catena. The data examined have demonstrated that excepting narrow cases involving answer fragments elicited by or-questions and certain echo questions, the Catena Condition is a necessary condition on the elided material of ellipsis mechanisms in English. The data have not, however, demonstrated that the Catena Condition is sufficient for predicting where and when ellipsis can occur. Indeed, no claim has been made to this effect. In other words, while it appears as though the elided material of ellipsis mechanisms should be a catena (in most cases), being a catena is not sufficient to license ellipsis. There are many further aspects of ellipsis beyond the Catena Condition that influence whether a given attempt at ellipsis succeeds or fails. For instance, gapping, stripping, and pseudogapping require a remnant to stand in contrast to the parallel constituent in the antecedent clause, and there is a condition on many ellipsis mechanisms that requires a preposition to be present if its complement is present (gapping, stripping, pseudogapping, sluicing, LEE).

. C

.................................................................................................................................. To conclude this contribution, some of the main ideas expressed above are enumerated here in abbreviated form: . Many established DG frameworks have not devoted much effort to explore ellipsis. . When DGs do address ellipsis, there is a strong tendency to assume the presence of null material/nodes. . The null nodes of ellipsis enjoy a special status; they do not have the same syntactic status as the corresponding overt words. . Aspects of recoverability vary significantly from one ellipsis mechanism to the next. Some ellipsis mechanisms place clear syntactic restrictions on recoverability, whereas for others, recoverability is determined pragmatically. . The catena unit serves as the basis for a comprehensive theory of licensing in a DG theory of ellipsis.

   



. Typical ellipsis mechanisms are eliding catenae, whereby attempts at ellipsis where the elided material does not form a catena are usually bad. . The catena condition on the elided material of ellipsis is a necessary condition, but not a sufficient one. Many other aspects of ellipsis play a role in determining which material a given ellipsis mechanism can and cannot elide. With the importance of the catena established, the door is now open to comprehensive DG accounts of the various ellipsis mechanisms. The restrictions placed on elided catenae and on the remnants of ellipsis can be identified and explored in a principled manner, without recourse to unmotivated stipulations such as movement-first-ellipsis-second.

  ......................................................................................................................

    ......................................................................................................................

 .    

. S S

.................................................................................................................................. S Syntax (henceforth SS) seeks to specify the role of syntax in linguistic theory in such a way as to achieve maximum overall explanatory power. The Simpler Syntax Hypothesis is formulated as follows in Culicover and Jackendoff (): Simpler Syntax Hypothesis (SSH): The most explanatory syntactic theory is one that imputes the minimum syntactic structure necessary to mediate between phonology and meaning. SSH calls for the rigorous application of Occam’s razor to syntactic analysis. It hypothesizes that syntax per se is severely restricted in the sorts of phenomena it can account for, especially in comparison to more conventional Chomskyan syntactic theories (Chomsky , , b, , ).1 SS adopts the Parallel Architecture perspective of Jackendoff (), in which phonological, syntactic, and semantic structures are subject to their own well-formedness conditions, and are related to one another through correspondence (or interface) rules. Individual words are treated as small-scale interface rules that map pieces of phonology to pieces of syntax and pieces of meaning. SS also countenances meaningful constructions (in the sense of Construction Grammar), defined in terms of similar correspondences, but at a phrasal level. Hence phrasal syntactic structures correspond to larger pieces of meaning, but not necessarily in one-to-one Fregean fashion, since aspects of phrasal meaning can be specified by particular constructions. For instance, the meaning of the comparative correlative construction (the more . . . , the more) is to some extent a property of the entire construction and cannot be localized in any of its words (Culicover b). We exclude the Minimalist Program (Chomsky b), which appears in some respects to converge on SS. 1

   



SS assumes no abstract syntactic structure (i.e., no functional heads, no massive binary branching, etc.), no movement, no invisible constituents (except Aʹ trace), and no (or limited) UG island constraints. We do not assume the stipulations of GB theory and Principles and Parameters theory such as the Theta Criterion, the Projection Principle, the Extended Projection Principle, and so on. To the extent that SS is on the right track, the phenomena that fall under these various devices must find explanation outside of syntactic theory proper, either in semantic well-formedness conditions, in the interface between syntax and semantics, or in processing complexity.

. SSH   

.................................................................................................................................. To see the consequences of SSH with respect to elliptical constructions, consider bare argument ellipsis (BAE), illustrated in B’s reply to A in example (). ()

A: Ozzie says that Harriet’s been drinking. B: Yeah, scotch.

B’s reply can convey the same meaning as sentence (), thus going beyond the meanings of yeah and scotch. ()

B: Yeah, Harriet’s been drinking scotch.

There are two basic approaches to explaining how the meaning in () is conveyed. One is that (B) is a complete sentence, with covert syntactic structure matched to the portion of (A) that gets the interpretation ‘Harriet’s been drinking x’; and scotch is substituted for x. The other is that the syntax of (B) is simply Yeah, scotch, and the meaning in () is computed by matching scotch with the unexpressed argument of drinking in the interpretation of (A). Both approaches have a long and distinguished history, and are well-represented in the chapters of this volume. Other things being equal, SS leads us to prefer the second analysis, because it does not posit invisible syntactic structure. If an account in terms of invisible syntactic structure and one in terms of interpretive rules were empirically equivalent, it could be argued that this preference is simply an aesthetic one. However, they are not empirically equivalent: the two accounts diverge in explanatory power with respect to the full range of phenomena. Specifically, the relation between the elliptical utterance and its antecedent depends not on syntactic identity, but rather on the interpretation of the antecedent. We illustrate with four cases. First, there is no syntactic difference among A’s utterances in () and (), but the interpretation of the antecedent is clearly different. () a. A: Ozzie fantasizes that Harriet’s been drinking. B: Yeah, scotch. [‘Ozzie fantasizes that Harriet’s been drinking scotch’, not ‘Harriet’s been drinking scotch’] b. A: Ozzie doubts that Harriet’s been drinking. B: #Yeah, scotch. [no plausible interpretation]



  

These differences cannot be explained by an approach to ellipsis that depends only on syntactic structure. And if such an account sorts out the interpretations of Yeah, scotch in various contexts by relying on additional rules of pragmatic inference, we would argue that the latter sort of rule, whatever it may be, is sufficient to account for the facts. Second, as is well-known, the putative hidden syntactic forms for many examples of ellipsis either are ungrammatical (Bi and Bi) or diverge wildly from the form of the antecedent (Bii and Bii). ()

A: John met a guy who speaks a very unusual language. B: Which language? i. *Which language did John meet a guy who speaks? ii. Which language does the guy who John met speak?

()

A: Would you like a drink? B: Yeah, how about scotch. i. *Yeah, how about would you like scotch. ii. Yeah, how about you giving me scotch.

Third, the antecedent can extend over more than one sentence, so the ellipsis cannot straightforwardly be derived from a covert syntactic clause. ()

It seems we stood and talked like this before. We looked at each other in the same way then. But I can’t remember where or when. (Rodgers and Hart )

Fourth, as seen in (), the antecedent can be a non-linguistic situation in the environment, in which case there is no linguistic context at all to motivate deletion of elliptic syntactic material. ()

The claim of SSH, then, is that rules of interpretation and inference are required to account for the full range of ellipsis phenomena. By Occam’s razor (and SSH), if the crucial cases require rules of interpretation and inference, and if these rules are sufficient for the uncontroversial cases, then there is no need to posit invisible syntactic structure as well. The rest of this chapter sketches a very general approach to elliptical constructions within an interpretive framework, developed in more detail in Culicover and Jackendoff

   



(). In anticipation of the analysis, we summarize our stance on the three major issues in ellipsis sketched in Chapter : • Structure: Following the SSH, we assume that for most types of ellipsis, the ellipsis site itself has no hidden structure. For instance, the syntactic structure of Yeah, scotch in () is essentially just that. • Recoverability: As we have seen in ()–(), the interpretation of the elliptical expression is based in part on the interpretation of its antecedent, with syntax often playing a secondary role. What we take to be most novel in our approach is that the interpretation invokes a domain-general cognitive relation which we call -, to be introduced in section .. • Licensing: We take a constructional approach to the licensing of elliptical expressions. Licensing is not a matter of specifying what can be deleted or what structure can be empty. Rather, each elliptical construction (BAE, VPE, gapping, etc.) specifies a particular type of syntactic fragment that can serve as an elliptical expression, along with a specification of how it is to be interpreted in terms of the - relation. Sections . and . show how the - relation is expressed in syntactic structure in a variety of elliptical constructions (including Bare Argument Ellipsis), and section . shows how it also can be expressed by syntactic constructions not normally thought of as elliptical. Section . discusses some phenomena that have been adduced as evidence for hidden syntactic structure, and section . summarizes.

. T : -

.................................................................................................................................. The - relation is a domain-general cognitive mechanism. It spontaneously evaluates similar objects, extracting the properties that they share (the ) and those that they do not (). Figure . illustrates two basic cases, pointed out by William James (), which we call contrast and elaboration. In the left-hand pair, the two wugs are immediately seen as (nearly) the ,  for their contrasting head decorations, which stand out as a result. In the right-hand pair, they are seen as (nearly) the ,  for the presence of a head decoration in the second one. This same relation can be experienced not only with objects, but also with actions, musical motives, and even food.2 Our hypothesis is that this - relation appears in natural language semantics as well as in all these non-linguistic domains, and it can be expressed in language in a variety of ways, one of which is with the words same and except and their synonyms (see section .).3 When two juxtaposed constituents are sufficiently similar, their interpretations are taken to be the same, except for the part that is overtly different. One of the two juxtaposed A reader suggests the following locution as an explication of the same-except relation: “judged to be overall similar in virtue of identity of all attributes other than the points of contrast.” 3 To see how ubiquitous such expressions are, readers are invited to count the number of times in this chapter that we use locutions such as X is the same as Y, except for Z. 2



  

Contrast

Elaboration

 . Contrast (left) and elaboration (right)

expressions (usually the second) may be elliptical, in which case its interpretation is the  as the other expression—the antecedent— for the part that contrasts with the antecedent. For example, in (), B’s response is taken to express an assertion that is the same as A’s assertion Harriet’s been drinking, but with the elaboration scotch. That is, scotch plays a role parallel to the head decoration on the rightmost wug in Figure .. An immediate virtue of this analysis is that the semantics associated with ellipsis is not specifically linguistic: it follows from a general property of human cognition. Before turning to ellipsis in more detail, it is useful to examine the marking of contrast by stress. Consider (). Contrastive accent is marked in caps. ()

FRED likes fish, and SUE likes fish.

The two expressions Fred likes fish and Sue likes fish are understood as the ,  for the pair Fred/Sue. In effect, SUE likes fish means exactly the  thing as Fred likes fish,  that SUE likes fish is about SUE. Note that Fred and Sue are syntactically parallel. Notice also that the contrastive accent marks the elements that are interpreted as participating in the  part of the relation. Sentences like () violate this condition. ()

a. *FRED likes fish, and Sue likes FISH. b. *FRED likes fish, and FRED likes soup.

Here the phonologically contrasted expressions do not support a semantic contrast. That is, Sue likes FISH does not mean the same as Fred likes fish  for fish, and FRED likes soup does not mean the same as Fred likes fish  for Fred. To keep track of the relationships that hold between the two expressions, we employ the notational device of a tableau, as in ().4 4

In an early incarnation of this work, we attempted to implement a formalism along the lines of lambda abstraction, which in some respects does the sort of work we want. However, we eventually set this formalization aside for two reasons. First, it rapidly became unwieldy, to the point of obscuring the overarching generalizations we were trying to express. Second, - is a domain-general

   



() FRED likes fish, and SUE likes fish.   SAME

[FRED LIKES FISH] [SUE LIKES FISH]

EXCEPT FRED

SUE

More generally, the form of a tableau is shown in (). 



E

E

EXCEPT P

P

() SAME

E and E are the Entities judged to be the same, e.g. the wugs in Figure .; P and P are the respective Parts of E and E that contrast, e.g. the head decorations on the left-hand pair of wugs in Figure .. In the alternative configuration, elaboration, P is null and P adds a new part, as in the right-hand pair of Figure .. In order to establish a - relation, the two entities to be related must be identified. In the case of Figure ., the items in question are adjacent. In the linguistic case, the constituents expressing the two entities in question are typically adjacent or nearby. For instance, in (), the two clauses are the entities that are . We use the informal term find for the computational operation that identifies the two entities to be related as . A second operation, which we informally call align, involves finding the relevant features of the entities and the correspondences between them—picking out visual parallelisms or, in the linguistic case, identifying parallel thematic roles and the like. Finally, in order to compute , there must be an operation of identify differences that picks out the aligned parts that are different. For instance, in (), Fred and Sue are aligned, and different from each other. The most typical environment for linguistic juxtaposition is coordination, as in (). Coordination offers about the simplest parallelism of structure, making it straightforward to find entities to compare. However, many other syntactic environments also count as ‘juxtaposed’, such as a main and a subordinate clause (a) and relative clauses within the subject and object of the same verb (b). ()

a. If FRED likes fish, then SUE must like fish. b. People who WHISPER resent people who SHOUT.

relation, and it is hard to see how to apply lambda abstraction to non-linguistic phenomena such as wugs, melodies, and soups. We therefore have settled on the relatively informal tableau notation, which we find more revealing; it provides a conceptual basis from which more precise formalisms can be developed. Similarly, we initially tried to lay out our semantic representations in terms of a more standard predicate–argument sort of notation, of the sort in Jackendoff (, ). Again, we found that insisting on detailed formalization got in the way of the basic insights, and so we have represented the semantics very informally.



  

. L   

.................................................................................................................................. Returning to ellipsis: Again, the basic idea behind our account is that ellipsis constructions express a - relation, in which the antecedent of ellipsis expresses E, and the elliptical expression itself expresses P—the part of E that is different from E. The rest of E is unexpressed, but it can be inferred from its ness to E. The interpretation of the ellipsis is recovered by matching the structure and interpretation of the elliptical expression with that of the antecedent, along the lines summarized in the preceding section. We illustrate the approach by applying it to Bare Argument Ellipsis (BAE), as in (), elaborating the account presented in section .. ()

a. A: I hear Ozzie’s drinking vodka again. B: No, scotch. b. A: I hear Ozzie’s drinking again. B: Yeah, scotch.

In order to formalize the BAE construction, we have to decide on its syntactic structure. Culicover and Jackendoff () argue that BAE is not a clause in the usual sense, because it does not embed freely (*Harriet realizes (that) scotch, etc.). For convenience, we assign it to the category Utterance (of which S is a particular type), dominating a phrase of any syntactic category (XP).5 Simpler Syntax posits that the ellipsis site for BAE has no further syntactic structure aside from the bare XP. Under this account of the syntax, the BAE construction can be characterized as (), an interface rule correlating syntax and semantics. The coindex  matches the syntactic Utterance to the E cell in the semantic tableau, and the coindex  matches the XP constituent to the P cell. The underlining in the E and P cells indicates that these pieces of the meaning are to be found in the antecedent. ()

Bare Argument Ellipsis Syntax: [Utterance XP2 ]1 Semantics:   SAME

E

SITUATION1

EXCEPT

Ø/P

P2

5 Not all non-sentential Utterances are elliptical. For instance, words like hello, ouch, and abracadabra can stand on their own and do not embed. Other non-sentential Utterance types have argument structure, such as How about XP?, What about XP, and P with NP! (e.g. Off with his head!).

   



() can be thought of as a meaningful construction of English in the sense of Construction Grammar: it licenses a connection between a particular syntactic structure and a particular semantic interpretation, going beyond the meanings of the words in the syntax. Let us see how this construction derives the interpretation of scotch in (aB). First, the coindexation in () is applied to derive (). ()

Phonology/Syntax: [Utterance [NP scotch]2 ]1 Semantics:   SAME

E

SITUATION1

EXCEPT

Ø/P

SCOTCH2

Next, E and P must be found and aligned. In this case they are found in the preceding sentence, but in other cases they might instead come from a non-linguistic situation. The result is (). () Phonology/Syntax: [Utterance [NP scotch]2 ]1 Semantics: 



SAME

OZZIE DRINK VODKA SITUATION1

EXCEPT

VODKA

SCOTCH2

It remains to specify E. Informally, we need to subtract P (VODKA) from E and add P in its place. We can notate this operation as the equation (a), and the resulting tableau as (b). ()

a. E≈OZZIE DRINK VODKA (
VODKAþ SCOTCH)¼ OZZIE DRINK SCOTCH b. Phonology/Syntax: [Utterance [NP scotch]2 ]1 Semantics:  SAME



OZZIE DRINK VODKA [OZZIE DRINK SCOTCH]1

EXCEPT VODKA

SCOTCH2

This derivation illustrates the ‘contrast’ case of -, in that scotch contrasts with vodka. The other pattern, elaboration, appears in (b): the elliptical expression supplies an argument (or adjunct) that does not correspond to anything in the antecedent. In this case P is null; hence in deriving E, P is simply added without any subtract step.6

Culicover and Jackendoff () make a further distinction between “elaboration” and “specification” which we will not pursue here. 6



  

Every part of these derivations has counterparts in other theories. In a syntactic theory of ellipsis, the syntactic antecedent must also be located and aligned with the elliptical expression. More importantly, our informal process of subtracting and adding can be instantiated more formally in terms of lambda abstraction of P and application of the resulting property to P. Subtracting P is equivalent to lambda abstraction, while adding P is equivalent to applying the lambda abstract to P. The same basic mechanism has also been used in standard accounts of the semantics of focus and ellipsis (e.g., Merchant ; Culicover and Jackendoff ). It is based in turn on the earlier approaches of Kraak (), Chomsky (), Jackendoff (), and Akmajian () and developed formally by e.g. Dalrymple et al. (), Rooth (b), and Lappin (). In fact, all approaches to focus and ellipsis that we are aware of assume procedures along the lines of find, identify differences, align, subtract, and add; they differ only in terms of the particular level of representation that the procedures apply to: syntax, LF, or semantics. An important semantic constraint on BAE follows from the character of -. Consider the minimal pair in (). ()

A: Ozzie drank the scotch in five minutes. B: a. Yeah, in the kitchen. b. *No, in the kitchen.

The difference in acceptability between B’s two replies follows from the discourse relations, which are crucial to the interpretation. As before, B’s yeah in (Ba) indicates that the following utterance is understood as assenting to and elaborating A’s statement. On the other hand, B’s no in (Bb) marks the utterance as a contrast, and - stipulates that the contrast must be found in P. But no part of A’s statement can serve as P, since the time period in five minutes does not bear the same semantic relation to E as the location in the kitchen does to E. Hence there is a failure of align, and the intended reading of (Bb) is ill-formed. Notice, though, that the two adjuncts are syntactically parallel; it is only in the semantics that alignment fails. This result speaks in favor of a semantically based account of ellipsis. Notice also that the same effect occurs with contrast signaled by words like but and contrastive stress, as our account predicts. () *Ozzie drank the scotch in  MINUTES, but Harriet drank it in the KITCHEN. The E cell of a tableau for BAE specifies that the Utterance has the semantics of a Situation (i.e., a State or Event), even though the syntax specifies only one of its constituents, namely XP. Since illocutionary force can be assigned to expressions of Situations, BAE may act semantically as an interrogative or an imperative, as well as a declarative. (Similar arguments appear in Stainton b.) () a. A: I’ll pour you some vodka. B: No, scotch! [Imperative] b. A: Hey, look! Ozzie’s drinking something! B: Scotch? [Interrogative]

   



. L    

.................................................................................................................................. In this section we show how the - mechanism licenses and renders recoverable a range of familiar types of ellipsis discussed in the literature: sluicing, gapping, VP-ellipsis, and pseudogapping. Sluicing and gapping are like BAE in that they involve the interpretation of one or more fragments, while VP-ellipsis and pseudogapping involve a bit more complication. We also mention VP-anaphora which is strictly speaking not ellipsis, but appears to employ the same interpretive mechanisms.

.. Sluicing Sluicing involves a bare wh-phrase in a position that licenses an indirect question. ()

a. Ozzie drank the scotch in five minutes, but I can’t tell you where. b. A: Ozzie’s drinking again. B: Yeah, but I don’t know what. c. Abby speaks the same language that some guy in this class speaks, but I’m not sure who.

(a,b) are instances of elaboration, like (b); these cases are called ‘sprouting’ in the sluicing literature (Chung et al. ). (c) is an instance of contrast, where who contrasts with some guy in this class; this case is referred to as ‘matching’ in the sluicing literature. Sluicing is a wh-interrogative counterpart to BAE. We assume as before that the ellipsis site per se has no hidden structure. An important clue to the nature of sluicing is the fact that any type of phrase that can occur in sluicing can also occur in BAE (e.g., parallel to (b), A: Ozzie’s drinking again. B: What?). Hence an account of sluicing should be parallel to that of BAE insofar as possible. There are, however, two differences. First, the single phrase in sluicing must be a wh-phrase. Second, this phrase counts syntactically as a clause rather than as an Utterance: it appears in positions characteristic of clauses, such as extraposed position (a); and, like a clause, it conditions singular agreement even when the wh-phrase is plural (b). ()

a. We were supposed to do some problems for tomorrow, but it isn’t clear which problems. [cf. * . . . but it isn’t clear the answers] b. We were supposed to do some problems for tomorrow, but which problems isn’t/ *aren’t clear. [cf. . . . but the answers aren’t/*isn’t clear]

The structure of the sluicing construction is given in (), licensing an interpretation of a sentence consisting only of a wh-phrase. () Sluicing Syntax: [S wh-phrase2 ]1 Semantics:   SAME

E

EXCEPT Ø/P

SITUATION1 P2



  

The Syntax part of () captures the clausal nature of the sluiced phrase by stipulating that it is an S that contains (only) a wh-phrase. This is where sluicing differs from BAE. Aside from this syntactic specification, the two constructions are identical and should follow from the same principles. Given that there are BAE examples such as (), (), and () that have no conceivable syntactic source, we therefore conclude that sluicing cannot be accounted for in syntactic terms, without losing the fundamental parallelism between it and BAE. Derivational accounts of sluicing assume either that this syntactic structure is derived by deleting the rest of the S (e.g. Ross b), or that the wh-phrase is the visible part of a complete clausal structure (e.g. Merchant ). Both approaches appear to be at least as stipulative as the constructional one proposed here, since both have to say what is deleted/invisible, and what can be left behind. The present account, however, is not focused on describing what can be left out. Rather, it describes in positive terms the surface configuration and what it means. (See section . for more discussion of these accounts.)

.. Gapping Culicover and Jackendoff (, ) analyze gapping as essentially double BAE:  with two contrasting sites, as in (a).7 As in the case of BAE and sluicing, the ellipsis site has no hidden structure. Gapping works the same way as full sentences with double contrastive stress such as (b), except that the non-contrastive parts are left out. ()

a. FRED likes FISH, and SUE, SOUP. b. FRED likes FISH, and SUE likes SOUP.

In (a), Fred and Sue constitute one contrast, and fish and soup another, as in the tableau (), which has two EXCEPT rows. () Phonology: Sue2 soup3 Syntax: [Utterance NP2 NP3 ]1 Semantics:   SAME

FRED LIKE FISH SITUATION1

EXCEPT

FRED

SUE2

EXCEPT

FISH

SOUP3

The general form of the gapping construction appears in (). Given that only the contrast case is possible, there is no option for null P and Q, as there was in BAE and sluicing.

7

The idea of gapping as multiple contrast goes back at least to Pesetsky ().

   



() Gapping Syntax: [S XP2 YP3 ]1 Semantics:   SAME

E

SITUATION1

EXCEPT P

P2

EXCEPT Q

Q3

After finding the antecedents for E, P, and Q, this yields () for the structure of Sue soup in (a). As in previous cases, Sue soup is understood as ‘Sue likes soup’ by virtue of ‘solving for E’ in (), using the inferential steps subtract and add. ()

E ≈ FRED LIKE FISH (– FRED + SUE) (– FISH + SOUP) = SUE LIKE SOUP

We note that gapping is syntactically more constrained than prosodic contrast and BAE. For example, in a subordinate clause it is possible to have contrast, but not gapping, as shown by (a,b); and it is not possible to have gapping in the typical BAE position as an isolated Utterance (c,d). () a. Contrast in subordinate clause: FRED likes FISH, and I think that SUE likes SOUP. b. Gapping into subordinate clause: *FRED likes FISH, and I think that SUE, SOUP. c. BAE as isolated Utterance: A: Who likes fish? B: Fred. d. Gapping as isolated Utterance: A: Who likes what? B: *Sue, soup. These contrasts can be thought of as stemming from a difference in where find looks for an antecedent. In gapping, it looks within a quite constrained range: preceding conjuncts and a very small number of other environments. This difference could be a stipulation of individual elliptical constructions, or it could follow from independent principles. We have no evidence one way or the other. We note however that any theory of ellipsis has to make the same distinctions.

.. VP-ellipsis VP-ellipsis (VPE) is the best-studied variety of ellipsis. The literature on VPE offers many options for the syntax, including VP-deletion under syntactic identity (Sag ), a pro-VP (Hardt ; Lobeck ), a fully structured but phonologically empty VP (Wasow , ), and no VP (Culicover and Jackendoff ). Because VPE allows non-linguistic antecedents (as in our cartoon in ()), we can immediately rule out deletion under identity.



  

For convenience, we adopt the pro-VP option here, represented in our notation as a phonologically null VP (though we recognize that other things being equal, the SSH would favor a solution with no VP at all, thereby minimizing syntactic structure). A first approximation of the rule is (). () VP-ellipsis Phonology: Ø1 Syntax: VP1 Semantics:  SAME

E

 SITUATION1

That is, the interpretation of the pro-VP is the  as that of the antecedent; contrasting  constituents may present, but need not be, as in A: Mary called. B: Yes, she did. Example () illustrates the construction in a case where the subjects contrast and the auxiliaries contrast in polarity, hence a double contrast. () Phonology/Syntax: Joe didn’t sneeze, but Bill did [VP Ø]1. Semantics:   SAME

NEG JOE SNEEZE

SITUATION1

EXCEPT

JOE

BILL

EXCEPT

NEG

POS

E ≈ NEG JOE SNEEZE (– JOE + BILL) (– NEG + POS) = POS BILL SNEEZE To flesh out this analysis, of course, many questions need to be resolved about how VPE in various syntactic environments is mapped into tableaux, and about the degree to which align depends on syntactic configuration. The specifics of the analysis ultimately depend on a number of assumptions about the syntax of VPE, control, raising, and expletive subjects. To pursue all these issues in detail would take us far beyond the scope of the present treatment. Some of them are discussed in Culicover and Jackendoff (: ch. ). Here we mention just two such issues. First, a longstanding problem is “vehicle change,” where the interpretation of the missing VP cannot be derived by copying over the syntax of the antecedent. () presents six typical cases. ()

a. b. c. d. e. f.

John has always enjoyed running, but I’ve never wanted to (*running). John likes those pictures of himself, but I don’t (*like those pictures of himself). John doesn’t have any bananas, but Bill does (*have any bananas). These issues should be discussed further, but we won’t (*be discussed further). We should discuss these issues, but they probably won’t be (*discuss). Mubarak’s survival is impossible to predict and, even if he does [‘survive’], his plan to make his son his heir apparent is now in serious jeopardy. (COCA: CBS Evening News. Cited from Miller and Hemforth a.)

   



Such examples are problematic for a theory of VPE that requires syntactic identity. Within the present approach, the question is how much syntactic parallelism is required in order to establish semantic alignment; these cases show that alignment cannot always require syntactic identity. We leave this question open. Another puzzle for VPE is the “missing antecedents” problem posed by Grinder and Postal (). ()

John doesn’t have a gorilla, but Bill does, and it smells.

The antecedent of it cannot be the overt NP a gorilla, because if but Bill does is omitted, () is illformed. Rather, it clearly refers to the gorilla Bill has, not to the gorilla John doesn’t have. But this antecedent is not expressed in (). Grinder and Postal () use this observation to argue that the underlying form of Bill does is Bill has a gorilla, which then undergoes deletion of the VP. In the present approach, Bill’s gorilla emerges in the semantics: ()

Phonology/Syntax: Bill2 [VP does]1 Semantics: 



SAME

NEG JOHN HAVE GORILLA

SITUATION1

EXCEPT

JOHN

BILL2

EXCEPT

NEG

POS

E ≈ NEG JOHN HAVE GORILLA (– JOHN + BILL) (– NEG + POS) = POS BILL HAVE GORILLA When we “solve” for E, we get the meaning ‘Bill has a gorilla’, in which the gorilla is asserted to exist. And this gorilla, derived through the inferential process of “solving for” the ellipsis, is what it refers to. Setting these syntactic issues aside, the main point we have tried to establish in this section is that the semantics of VPE can be represented in terms of -. We believe that the difficulties the construction poses are primarily with how its syntax maps into the tableau.

.. VP-anaphora It is instructive to compare our account of VPE with VP-anaphora (do it or do that), as in (). There is one difference in the semantics: VPE can denote any Situation—an Event or a State—whereas do it has to denote an Action on the part of the subject. () (You took a nap yesterday?) Phonology/Syntax: I2 [did that [last week]3]1. Semantics: 



SAME

YOU TAKE NAP YESTERDAY [I ACT LAST WEEK]1

EXCEPT

YOU

I2

EXCEPT

YESTERDAY

LAST WEEK3



  

Note that in this construction, E can be an entity in the visual environment rather than in a previous sentence. (The notation {} indicates meaning that is derived from the nonlinguistic context.) ()

[Addressee is spreading peanut butter with a knife:] Phonology/Syntax: Can [you do that using [a spoon]2]1? Semantics: SAME {YOU SPREAD PEANUT BUTTER [YOU ACT WITH SPOON]1 WITH KNIFE} EXCEPT

{KNIFE}

SPOON2

We thus take do it-anaphora to be an expression of . () Do it-anaphora Phonology: do4 it5 Syntax: [VP V4 NP5]1 Semantics:   SAME

E

X ACT1

Informally, () says that do it means ‘X performs an action that is the same as E’, where X is the external argument of ACT (either an overt or a controlled subject), and E is either expressed by an antecedent VP or observed in the non-linguistic context. Comparing () with the VPE construction (), we see that they are essentially the same, except for two critical differences. First, () is stated in terms of a VP with the phonological form do it and corresponding syntactic structure, while VPE is stated in terms of a phonologically and syntactically empty VP. Second, do it-anaphora is restricted to an ACT interpretation, while VPE is not.

.. Pseudogapping Consider finally pseudogapping, illustrated in ().8 ()

That may not bother you, but it does me.

In this construction, the main verb is absent, but the auxiliary is overt. In this respect, pseudogapping is similar to VPE (Hoeksema : ). The difference is that pseudogapping includes an argument of the missing verb. Culicover and Jackendoff (: ) observe, following Levin (), Hoeksema (), and Merchant (a), among others, that pseudogapping is somewhat less restrictive than gapping. For instance, pseudogapping is marginally possible when the antecedent is in a preceding subordinate clause (a,b), while gapping is not (c,d). 8

For a comprehensive derivational treatment of pseudogapping, see Lasnik (b).

   



() Pseudogapping: a. ?Whenever/Because Robin speaks French, Leslie does German. b. ?Leslie will not speak German, because Robin does French. Gapping: c. *Whenever/Because Robin speaks French, Leslie, German. d. *Leslie will not speak German, because Robin, French. Pseudogapping typically has a stranded NP complement, but it also allows a stranded PP (a) or S (b) complement. Moreover, the antecedent of the stranded complement need not be VP-final, as seen in (c). ()

a. Sandy will smile at Terry, and Terry will [‘smile’] at Sandy. b. Sandy claimed that it would rain, and Terry did [‘claim’] that it would snow. c. Sandy put a case of beer in the fridge, and Terry did [‘put’] a can of Coke [‘in the fridge’].

Examples such as these suggest a formulation of pseudogapping that incorporates properties of VPE and gapping. () offers two possible formulations. They differ in whether the remnant XP is embedded under an otherwise empty VP (a), or whether it is simply dominated by S, concatenated after the Aux (b). (SSH favors (b), because it has less structure, but we have no empirical evidence to decide between them.) () Pseudogapping Syntax: a. [S NP Aux [VP XP2]]1 or b. [S NP Aux XP2]1 Semantics:   SAME

E

EXCEPT P

SITUATION1 P2

(c) will then have an interpretation equivalent to ‘Sandy put a case of beer in the fridge and Terry did the same, except that it was a can of Coke’. Pseudogapping is most felicitous when there is a further contrast in the sentence, either the subject, the auxiliary, or both. This could be stipulated in rule () by adding further  rows, or it might follow from, say, principles of information structure. We leave the question open. The main point is that the semantics follows the same pattern as all the other elliptical constructions; the differences lie in the specification of the syntax that can be linked with this semantics.9

9

There is no question that this description is largely ad hoc, and we have omitted many interesting details. But under any account, it is necessary to stipulate that it is just the verb and perhaps a complement that are pseudogapped, and not a larger structure. For instance, pseudogapping may not apply to a sequence of verbs, leaving just the inflected AUX. (i) a. *Terry will refuse to speak French, and Sandy will [‘refuse to speak’] German. b. *Terry will be speaking French, and Sandy will [‘be speaking’] German. c. *Terry would have spoken French and Sandy would [‘have spoken’] German.



  

. O - ,    

.................................................................................................................................. Our account of ellipsis rests on the semantics of the - relation, and on the operations find, align, identify differences, subtract, and add that are invoked in filling out a tableau. As it turns out, these operations are invoked in the interpretation of non-ellipsis constructions as well, which lends them more motivation than just their utility in an account of ellipsis. This section briefly presents three cases where there is no possibility of positing an invisible constituent.

.. Anaphora Consider the interpretation of a definite pronoun such as he, she, or it. Johni called, and hei was angry.

()

The pronoun expresses complete identity10—there is no . The standard coindexation notation in () is the usual way of showing this  relation. In our tableau notation, the lexical entry of the pronoun can be notated as in ().

An account that assumes movement of the stranded XP and deletion of the remaining VP (e.g. Lasnik b) cannot delete too broadly or strand non-complements, or it will end up deriving ungrammatical sentences like (ii). (ii)

a. *Terry will claim that I speak French, and Sandy will [‘claim that I speak’] German. b. *You probably just feel relieved, but I do [‘feel’] jubilant. (Lasnik b: ) c. *Rona sounded annoyed, and Sue did [‘sound’] frustrated. (Lasnik b: )

However, for corpus evidence that pseudogapping is freer than is typically claimed, see Miller . A still more complex construction is comparative ellipsis, which has cases resembling many other types of ellipsis, e.g.: (iii) a. b. c. d. e.

Terry has more beans than Sandy. Terry has more beans than Sandy, cookies. Terry has more beans than Sandy does. Terry has more beans than Sandy does cookies. Terry is more stingy than Sandy is generous.

[like BAE] [like Gapping] [like VPE] [like Pseudogapping] [sui generis]

While the semantics in terms of - is fairly clear, the mapping to syntax, and in particular the variety of available ellipses, prove far less so. 10 We are abstracting away here from exactly what counts as ‘identity’, given the complex issues surrounding the distinction between discourse anaphora and binding of pronouns, strict and sloppy identity, etc. In particular, discourse pronouns such as that in () convey token identity, while in sloppy identity, a definite pronoun conveys type identity.

   



() he Phonology: he1 Syntax: NP..1 Semantics:   SAME

E

[DEF.SG.MALE]1

This stipulates that the reference of the pronoun must be in the  relation with its antecedent. To derive the reading of the pronoun in (), the underlined E says that he has to find an antecedent. When the antecedent John is found, its meaning is plugged into the E cell to yield (). () Phonology/Syntax: [NP he]1 Semantics:   SAME

JOHN [DEF.SG.MALE]1

The standard binding principles A, B, C (or their counterparts in other theories) can be interpreted as constraints on find. A definite pronoun usually denotes token-identity. This distinguishes it from the indefinite pronoun one, which normally expresses identity of sense, i.e. ‘same type’, as in (a). However, one with a definite determiner, e.g. (b), can signify token-identity to a non-linguistic antecedent. ()

a. See that wug? I used to have one. b. I want that one. [accompanied by pointing]

Parallel to he, one has the lexical representation in (). () One-anaphora Phonology: one2 Syntax: [NP . . . N2 . . . ]1 Semantics:   SAME

E

[COUNT.SG]1

The NP headed by one corresponds to a count individual in the semantics that serves as E in the tableau. (The feature COUNT is necessary to rule out mass antecedents, as in *John drank wine, and Bill drank one too.) As in the case of definite pronouns, the antecedent E is supplied by find.



  

Unlike definite pronouns, one allows -constituents—differences among tokens of the same type. One way to express  is as a modifier of one, as in (). ()

a. Pat ordered a pizza with olives, but Sam wanted one with pepperoni. b. Pat ordered a small onion pizza. We really needed a large one with mushrooms.

In such cases, one sets up a - relation, which initiates find, align, and identify differences. The contrasting modifiers result in  rows being added to the tableau, producing a tableau like () for (b). 

() SAME



SMALL ONION PIZZA LARGE PIZZA WITH MUSHROOMS

EXCEPT SMALL

LARGE

EXCEPT ONION

MUSHROOM

Notice that the contrasting modifiers require contrastive stress, as discussed in section .. Notice also that align connects the prenominal modifier onion with the postnominal modifier with mushrooms. This shows that align relies on semantic rather than (or in addition to) syntactic parallelism. (See Culicover and Jackendoff  for more details.) We claim, then, that definite and indefinite anaphora also fall within the ambit of - relations. This reflects back on the ellipsis constructions discussed in sections . and .: there should be a uniform account of all of them. Not since the earliest days of generative grammar have definite pronouns been generally considered to be the product of syntactic deletion.11 Thus we conclude that ellipsis constructions should not be either.

.. Vice versa Next consider another non-ellipsis construction, vice versa (Fraser ; McCawley ; Kay ; Yamauchi ). ()

Kim likes Pat, and vice versa.

Vice versa requires an antecedent clause. Within the antecedent, there are two constituents whose places are exchanged in the interpretation of vice versa (hence the synonymous the other way round). The general case looks like (). Syntactically, it differs from everything we have seen so far, in that all the  constituents are anaphoric and are filled in from the antecedents. Yet the semantics is a variation on the same theme.

11

A notable exception is Elbourne (, ).

   



() Vice versa Phonology: [vice versa]1 Syntax: Utterance1 Semantics:   SAME

E

SITUATION1

EXCEPT

P

Q

EXCEPT

Q

P

The semantics of () says that the meaning of vice versa is the same as that of its antecedent E, except that two parts of the meaning of the antecedent are substituted for one another. Consider how it applies to (). Find locates the antecedent of vice versa: Kim likes Pat. Align and identify differences match KIM with PAT and PAT with KIM. Solving for E, by subtracting KIM and PAT from KIM LIKE PAT and adding them back in according to the alignment, yields the interpretation ‘Pat likes Kim’. () E ≈ KIM LIKE PAT (– KIM + PAT) (– PAT + KIM) = PAT LIKE KIM ()

Phonology/Syntax: (Kim likes Pat, and) vice versa1 Semantics:   SAME

KIM LIKE PAT [PAT LIKE KIM]1

EXCEPT

KIM

PAT

EXCEPT

PAT

KIM

Note the similarity of () to gapping, for instance in (), which has the same interpretation as (). ()

Kim likes Pat, and Pat, Kim.

The difference is that in gapping, P and Q (here the Utterance Pat, Kim) are overt, whereas in vice versa they are identified as copies of the contrasting constituents in the antecedent clause. Again, not since the earliest days of generative grammar has anyone considered vice versa to be transformationally derived from a partial syntactic copy of its antecedent. The fact that it fits naturally into the - analysis is, we think, important evidence for our approach.

.. The same, except, and their paraphrases Finally, consider the lexical items the same, except, and their variants, as mentioned in section ..

 ()

   a. This wug is the same as/identical to/similar to/like that wug, i. except for/aside from/apart from the crest on its head instead of a curl. ii. only this one has a crest on its head instead of a curl. b. Gapping works the same way as full sentences with double contrastive stress, except that it leaves the non-contrastive parts out.

A tableau for (a) is shown in (). 

() SAME



THAT WUG THIS WUG

EXCEPT THAT

THIS

EXCEPT CURL

CREST

Informally (see Culicover and Jackendoff  for more detail), same and its variants set up a  row in a tableau, in which the subject (in (a), this wug) occupies the E cell and the entity to which it is being compared (that wug) occupies the E cell. The complements of except and its variants (this and crest) supply the P cell(s); and if present, the complements of instead (here, that and curl) supply the P cell(s). There are numerous syntactic variants. The same can act like a symmetrical predicate, with both E and E in its subject (a); the same and except can have plural arguments (b); they can compare three or more entities (c) (resulting in multiple columns in the tableau); and the same and instead can be anaphoric, using find to locate their antecedent (d,e). ()

a. Gapping and vice versa are the same, except that in gapping P and Q are overt. b. In Figure ., the two wugs are immediately seen as the same, except for/aside from their contrasting head decorations. c. All of these rules are the same except for the way they map semantics to syntax. d. () maps the complements of the same and instead into E and P in the tableau in (). This sentence works the same way, except that E and P are found in the previous sentence instead. e. The cops don’t need you and man they expect the same. [Bob Dylan] [the same is actually interpreted as vice versa.]

Again, nobody to our knowledge has proposed a syntactic account of these constructions that accounts for their interpretation. Yet they share their semantics with all varieties of ellipsis. We take the possibility of capturing these generalizations as strong evidence for our approach.

. S    ?

.................................................................................................................................. We have argued that the interpretation of ellipsis constructions does not require invisible syntactic structure: interface rules of a type independently needed for the constructions in

   



sections . and . will suffice. This means that ellipsis constructions such as BAE and sluicing are syntactically ‘WYSIWYG’, albeit with different syntactic properties and mappings into semantics. Likewise, our analysis of VPE posits an empty VP with no internal structure, which facilitates statement of the interpretation (and we do not rule out an account lacking a VP altogether). This said, we must still address arguments in the literature for invisible structure in ellipsis, based not on interpretation, but on syntactic distribution.12 These arguments date back to Ross (b), who argues against a semantically based theory of sluicing; and they have been extended more recently by e.g. Merchant (, a). The arguments are all based on the same principle: the syntactic form of an elliptical construction is subject to those constraints that would have been placed on it, if it had been in a full sentence whose syntax was a copy of the antecedent. Moreover, inasmuch as these constraints are sentence-level constraints, they must be imposed before deleting the elided material. The derivation thus involves movement of a copy of the visible material to a peripheral position in the structure, and deletion of the entire remnant under identity with an antecedent. For instance, consider (), the simplest example of so-called connectivity effects. () A: Johni likes someone a lot. B: i. Yeah, himselfi. ii. *Yeah, himi. The proper form of B’s BAE response is determined by the fact that it has to be bound to John. But reflexive pronouns are normally bound intrasententially, not across discourse. So the only way to accomplish this binding, it is argued, is to provide an underlying form for B’s reply that contains an antecedent for the reflexive, e.g. himselfi John likes ti a lot, where the material struck out is either deleted in syntax or not pronounced in PF. A similar argument involves case marking, where a BAE fragment carries the case demanded by the verb of which it is understood as the complement. (), one of Ross’s original examples, illustrates this phenomenon in German. () a. A: Wem folgt Hans? whoDAT follows Hans ‘Who is Hans following?’ B: Dem Lehrer. theDAT teacher ‘The teacher.’ b. A: Wen sucht Hans? whoACC seeks Hans? ‘Who is Hans looking for?’ B: Den Lehrer. theACC teacher ‘The teacher.’ 12

Here we are recapitulating points made in Culicover and Jackendoff (, ).



  

Merchant (a) adduces similar examples in Korean, Hebrew, Greek, Russian, and Urdu. Again, the argument is that the case marking of the overt material must be governed by an invisible or deleted verb. Similar examples in English involve the prepositions governed by verbs, as in () (from Culicover and Jackendoff : ). () a. A: I hear Harriet has been flirting again. B: i. Yeah, with Ozzie. ii. *Yeah, Ozzie. b. A: John is very proud. B: Yeah, of/*in his stamp collection. [cf. proud of/*in NP] c. A: John has a lot of pride. B: Yeah, in/*of his stamp collection. [cf. pride in/*of NP] The verb flirt requires the ‘flirtee’ to be marked by the preposition with. This requirement is apparently responsible for the necessity of the preposition in B’s response in (a). The very close paraphrases (b,c) push the point home further: they differ only in that proud requires its complement to use the preposition of, while pride idiosyncratically requires in. The replies, using BAE, conform to these syntactic requirements, just as if the whole sentence were there. If the antecedent includes the relevant preposition, though, the ellipted expression does not need it. () a. A: Harriet is flirting with someone. B: Yeah, (with) Ozzie. b. A: John is very proud of something. B: Yeah, (of) his stamp collection. c. A: John has a lot of pride in something. B: Yeah, (in) his stamp collection. In contrast, German requires the preposition even when it is present in the antecedent. () A: Mit wem hast du gesprochen? with whoDAT have you spoken B: Mit Hans./*Hans. with Hans Merchant (b) presents evidence that this difference correlates cross-linguistically with the ability to strand prepositions: languages that can strand prepositions behave like English, and those that cannot strand behave like German. Thus his explanation of () is that B’s reply is derived from a full sentence that has undergone topicalization, which in German must pied-pipe the preposition. Culicover and Jackendoff () propose an alternative approach to this problem called Indirect Licensing (IL). The general idea is that an item related to an antecedent can be

   



syntactically licensed, not by the sentence it is itself in, but via its semantic role in relation to the antecedent. IL can be seen in constructions like topicalization and pseudocleft: for example, in (). ()

a. Himselfi, Johni likes ti. b. The one whoi Johnj likes ti is himselfj.

Consider first (a). Topicalization is standardly analyzed as ‘movement’, where himselfi is licensed before movement takes place. Alternatively, though, the licensing of himselfi can be explained without movement if we can implement the following: (i) Johni binds the direct object argument, (ii) the direct object argument is a gap, (iii) the topicalized constituent is identified with the gap, (iv) therefore Johni indirectly binds himselfi, (v) therefore the reflexive is indirectly licensed. (This is a variant of what has been called ‘reconstruction’.) Now consider (b). There is no plausible movement account in this case. But himselfi can be licensed by the same mechanism envisioned for topicalization: (i) Johni binds the direct object argument, (ii) the direct object argument is a gap, (iii) whoi is identified with the gap, (iv) himselfi is identified with whoi across the copula, (v) therefore Johni indirectly binds himselfi. In both cases, the critical step that permits IL is the identification of the reflexive with a particular constituent of the sentence, here the gap in the direct object. From this it is a minor step to extend IL to the interpretation of ellipsis. On this view, IL permits a phrase XP to be syntactically licensed indirectly through a - relation. The conditions on such licensing are (i) there is an antecedent sentence S that can serve as E; (ii) XP aligns with the interpretation of some part YP of S, so that XP serves as P and YP serves as P; and (iii) the rest of E is implicit.13 In such a configuration, IL says that the syntactic properties of XP are those appropriate for its semantic role in E. To be more specific, we consider three cases of IL, which differ in how XP (= P) is related to P. The first case is Contrast, in which P is expressed by YP, a constituent of S, as in (). ()

A: John drinks vodka.

B: No, scotch.

Here IL says that XP, in this case scotch, must have the syntactic features appropriate to the position of YP(= P), here vodka. This case also includes the German examples in (): dem Lehrer is P, so by align it receives the semantic role of the individual being followed, wem (= P). That role is expressed as the object of folgt. Since folgt licenses dative case on its object, Indirect Licensing says that dative case is licensed for dem Lehrer as well. Binding connectivity () works in much the same way. Himself in (Bi) plays the role of P, and someone, the direct object of like, plays the role of P. Someone is locally c-commanded by the subject John, so IL says that himself also counts as locally c-commanded by John, and hence can be coreferential with John. Similarly, the pronoun

Culicover and Jackendoff () used the terminology ‘antecedent’ for E, ‘indirectly licensed’ (IL) for E, ‘target’ for P, and ‘orphan’ for P. 13



  

him in (Bii) counts as locally c-commanded by John, and therefore cannot be coreferential with John. The second case of IL is where P is an implicit argument: that is, it is present in the semantics but not in the syntax, as in (). ()

A: John drinks.

B: Yeah, scotch.

Here John drinks has an implicit object argument INDEF, not present in the syntax, that serves as P. B’s response scotch (= P) aligns in the semantics with INDEF. IL therefore says that scotch acquires syntactic properties appropriate to this role—that is, those of an overt direct object of drink. This case also includes the examples in (). Intransitive flirt in (a.A) has an implicit ‘flirtee’ (= P), and this role aligns with OZZIE (= P). IL therefore says that Ozzie must satisfy the subcategorization requirements of the verb flirt. When flirt has an overt ‘flirtee’, it requires the governed preposition with, and therefore we must have with Ozzie and not simply *Ozzie. The third case of IL is elaboration, where P is semantically as well as syntactically null, for instance in (). ()

A: John drinks scotch.

B: Yeah, every day.

Because every day (=P) is a time expression, A’s statement (=E) has no counterpart, syntactically or semantically, that can serve as P. Hence A’s statement imposes no syntactic licensing conditions on every day—just semantic conditions such as compatibility with tense and aspect. Therefore, we predict that elaboration is possible when XP is interpreted as an adjunct to S: the syntactic properties of adjuncts, unlike those of arguments, are determined entirely by their semantic role. Based on the evidence in Culicover and Jackendoff (: chs  and ), we think this is a correct result.

. C

.................................................................................................................................. We have argued that there are two types of ellipsis, neither of which involves hidden syntactic structure. One type, exemplified by BAE, interprets a fragment as being the  as that of an antecedent,  for the part of the interpretation associated with the fragment. A second type, exemplified by VPE, assigns an interpretation to a sentence with a pro-VP in much the same way: the meaning of the sentence is the  as that of the antecedent,  for the part of the interpretation associated with the overt material. This analysis fits well with the constructional, interpretive framework called for by the Parallel Architecture and Simpler Syntax, in that the syntactic structure of elliptical utterances is as minimal as possible. Each elliptical construction licenses its own mapping between the - semantics and a stipulated syntactic configuration. These mappings not only license the syntactic structure but also enable the hearer to recover the full intended semantic content.

   



We have shown in addition that this choice of framework is not merely aesthetic: it allows us to express empirical generalizations not available in other approaches. The interpretation of both types of ellipsis can be seen as elaborations of the interpretation of anaphora, and also as sharing semantics with non-elliptical constructions such as contrastive stress and with lexical items such as the same, except, and instead. Moreover, this semantic structure is not proprietary to language: it is responsible for judgments in every modality of perception. We take these to be important and unexpected generalizations both within the language faculty and in human cognition as a whole. We recognize that at this point many aspects of our analysis are quite informal. In particular, the tableau notation, the inferential processes find, align, identify differences, add, and subtract, and the mechanism of Indirect Licensing need to be fleshed out in considerably greater detail. However, we have observed that on the one hand these correspond to parallel notions in other theories of ellipsis, and on the other hand, they need to be sufficiently general to apply to non-linguistic as well as linguistic situations. We take this as a challenge for the field.

A Most of this chapter is adapted from Culicover and Jackendoff (), which analyzes in more detail each of the cases discussed here. The wug image is used with permission of Jean Berko Gleason. We are grateful to Neil Cohn for the wug variants in Figure . and the cartoon in ().

  ......................................................................................................................

          ......................................................................................................................

 .    

. I

.................................................................................................................................. E constructions are formal patterns in which certain syntactic structure that is typically expressed is omitted. Some of the most commonly discussed ellipsis constructions along with an attested example of each are provided in Table .. All examples in quotes within and following Table . come from the Corpus of Contemporary American English (COCA) (Davis ). Every language balances the need to be expressive with the need to be sufficiently easy to produce. These two major functional principles (described by Goldberg :  as Maximize Expressive Power and Maximize Economy) give rise to different networks of learned constructions in different languages via general processes of grammaticalization or constructionization (Paul ; Bybee et al. ; Hopper and Traugott ; Fried ; Traugott ). This chapter emphasizes the shared communicative motivation of ellipsis constructions that leads to cross-linguistic similarities and certain predictable functional constraints (section .), while we also emphasize the fact that ellipsis is licensed by a system of motivated constructions—i.e., learned pairings of form and function. Specific constructions readily capture a range of restrictions on form and function, including those related to semantics, discourse context, register, genre, and dialect. Constructions’ specific licensing properties, as well as generalizations across constructions, are captured within a network of constructions. On this view, our knowledge of language is a learned system of constructions that are strongly motivated by communicative concerns. Constructionist approaches avoid positing ‘underlying’ levels of syntactic representation or invisible/inaudible structure; instead, semantic recoverability is accounted for by an independently needed psychological ‘pointer’ function (section .). An explicit account of the English  construction and a discussion of several other constructions that have received less attention in the literature are offered in section .. A brief comparison of French with English in section . makes clear that there exist cross-linguistic differences

   



Table . Commonly discussed constructions that involve ellipsis Verb phrase ellipsis Sluicing Gapping Stripping Comparatives

‘French kids eat spinach and ours can too.’ ‘He said that I was “different.” He didn’t say how.’ A: ‘You made me what I am today.’ B: ‘And you me.’ ‘George Greenwell was a patriot but not a fool.’ ‘His front teeth seemed to protrude more than Henry remembered.’

Source: Corpus of Contemporary American English (COCA), Davis 

even in closely related languages. Finally, some standard arguments against the sort of semantic, surface-based proposal suggested here are addressed in section ..

. M 

.................................................................................................................................. This section reviews some general commonalities among ellipsis constructions, before delving into more detailed analyses. In a very general way, the existence of elliptical constructions is clearly motivated by our need to express our messages economically (Paul ; Grice ; Hankamer and Sag ). When part of an intended interpretation is recoverable from context, there is no need for it to be overtly specified (Shannon ; Piantadosi et al. ). Thus ellipsis constructions likely exist in every language. In fact, while speakers often have the option of redundantly expressing material that could be elided, other times, non-elided counterparts sound quite odd, and ellipsis is required. The non-elliptical counterparts of expressions in Table . are given in Table ., and while the first three sound fairly acceptable with appropriate intonation, the last two are much less felicitous than their elliptical counterparts (indicated by ‘#’), as they sound quite robotic. Many commonly discussed ellipsis constructions involve a semantic relationship that Culicover and Jackendoff () describe as “-.” That is, what is conveyed by ellipsis constructions is generally a semantic proposition that is the same as one that has been uttered or is otherwise recoverable, except that it differs in some key respect. Culicover and Jackendoff note that the - relationship is independently needed to account for lexical phrases like the same/identical/similar/alike . . . except/aside from. In order to highlight what is distinct while taking for granted what is the same, it is natural, indeed iconic, to assert only what is distinct.1 Ellipsis constructions require that the omitted information be recoverable, either on the basis of an overt clause or phrase (Chomsky , ; Katz and Postal ; Hankamer and Sag ), or from the non-linguistic context (Dalrymple et al. ; Culicover and Jackendoff ). That is, in order to note differences, as implied by the - function, whatever is the  must be recoverable. Whether the recoverability is based on an overt string or whether non-linguistic context can potentially supply the information depends on the particular ellipsis construction involved. Certain expressions would be impossible to interpret without reference to something uttered in the context (see further discussion below; also 1

Another available option that can be used to emphasize what is distinct is the use of contrastive stress (see acceptable examples in Table .).



  

Table . Non-elliptical versions of the attested examples of ellipsis in Table . French kids eat spinach and  kids can eat spinach . He said that I was ‘different.’ He didn’t say  I was different. A: You made me what I am today. B: And  made me what I am today. # George Greenwell was a patriot but George Greenwell was not a fool. # His front teeth seemed to protrude more than Henry remembered his front teeth protrude.2

Chomsky ; Hankamer and Sag ; Murphy ). For example, a pair of noun phrases (e.g., you, me as in the third example of Table .) is hard to interpret unless it is licensed by a gapping construction, which provides the missing semantic relation. The - function implicitly assumes a psychological  mechanism to some overtly expressed linguistic material, or to some relation that is recoverable from the non-linguistic context (see also Tanenhaus and Carlson ; Culicover and Jackendoff ; Martin and McElree ; Abeillé, Bîlbîie, and Mouret ). This psychological  mechanism is discussed in section ..

. R: A    

.................................................................................................................................. Just as the - semantic function is part of the meaning of many words and phrases that do not involve ellipsis, a psychological function that ‘points’ to previous linguistic material is likewise required by many words and phrases, such as those underlined in ()–(): ()

‘Peggy McMartin Buckey, , and her son Raymond, , spent two years and five years, respectively, in jail before their acquittal on  criminal counts.’

() ‘She longed for a cold shower and a soft bed. Not necessarily in that order.’ ()

‘They could see that we were making efforts to accommodate them and vice versa.’

()

‘Never let another man see you apply lip balm. Ditto hand cream.’

()

‘First of all, it’s not worth getting angry about. Secondly,  percent of the people have insurance.’

2 An anonymous reviewer suggests that this example could be partially addressed if we were to assume that a constituent, his front teeth to protrude, exists before a “raising” operation on the “subject” of that phrase, and that it is this phrase that is copied and deleted by ellipsis. But remember does not allow a VP with to (*Henry remembered his front teeth to protrude), so positing this as an “underlying” form that is then raised and deleted would require that ungrammatical forms are base generated, counter to prevailing assumptions.

   



The only way for the underlined terms in the examples above to be interpreted is for the listener to understand them to refer to (or presuppose) some overtly specified linguistic material. Thus these terms ‘point’ to another word or phrase, much as anaphoric pronouns do (see also Asher and Lascarides ). In examples ()–(), respectively and vice versa point to an overt linguistic string with a particular word order.3 On the other hand, secondly presupposes only that there was some ‘first’ statement, without referencing (pointing) to the form of the first statement (as does on the other hand itself). It is a benefit of positing a general pointing function that it allows some constructions to point to a quite specific overt linguistic string, while others only require that a semantic entity or proposition be evoked. Debates about the nature of ellipsis typically center around whether detailed inaudible syntactic structure of some sort is actually present at the ellipsis site, and this issue is discussed further in section .. For now, note that the lexical items in ()–() do not lend themselves to such an analysis. Surely no inaudible words or phrases are in any sense ‘within’ these terms. Occam’s razor would suggest that whatever means is used to account for the examples in ()–() should be used to account for standard ellipsis constructions. The same reasoning also applies to the many other examples that do not lend themselves to a derivational account (see section .). In fact, psycholinguistic evidence from Martin and McElree () supports the idea that ellipsis is interpreted via a pointer mechanism instead of by appeal to inaudible syntactic structure at the ellipsis site (see also Tanenhaus and Carlson ; Culicover and Jackendoff ; Martin and McElree ). Specifically, in a task that required participants to determine whether a sentence made sense or not, Martin and McElree () found that increasing the length or complexity of the phrase required for interpretation did not result in an increase in the time it took for comprehension (see also Tanenhaus and Carlson ; Frazier and Clifton ), while increasing the distance between the antecedent and the site of ellipsis did reduce accuracy, an indication that a memory of the original phrase or its semantic interpretation is retrieved, but is not reconstructed syntactically. For example, it took no longer to decide that () made sense when compared with (); nor did it take longer to decide that () did not make sense than it did to make the same determination for (). Importantly, the first clause has to be processed in order to correctly recognize whether the sentence makes sense or not. Sensical: ()

The history professor understood Rome’s swift and brutal destruction of Carthage, but the principal knew the overworked students attending summer session did not ___. [more complex antecedent]

3

An anonymous reviewer points out that respective is used to evoke a semantic pairing that is not explicit in the linguistic material in the following example: a. The bacteria were classed in their respective genera. Interestingly, this does not seem to be possible for respectively: a.’ ?? The bacteria were respectively classed in their genera.

 ()

   The history professor understood Roman mythology, but the principal was displeased to learn that the overworked students attending summer session did not___. [less complex antecedent]

Nonsensical: ()

The history professor understood Rome’s swift and brutal destruction of Carthage, but the principal knew the overly worn books used in summer session did not___. [more complex antecedent]

()

The history professor understood Roman mythology, but the principal was displeased to learn that the overly worn books used in summer session did not___. [less complex antecedent]

This finding suggests that the elided phrase is not created anew at the ellipsis site, undermining a copy and deletion mechanism. Martin and McElree () instead suggest that ellipsis involves a pointer to structures in memory, the perspective we endorse here. Phillips and Parker () issue a word of caution on Martin and McElree’s interpretation of their results because the task did not require that the full antecedent be semantically retrieved. While this is true, claims that the antecedent’s structure is copied generally assume that the structure is copied automatically as the sentence is processed, not merely that it can potentially be, if relevant. The lack of complexity effect found by Martin and McElree is then only dismissable under a copying mechanism account, if participants were not actually processing the sentences as they normally would. Xiang, Grove, and Merchant () report that a double-object antecedent followed by either VP-ellipsis or a repeated double-object construction primes the production of a double-object expression, when compared to a double-object construction followed by an intransitive clause.4 But the authors acknowledge that the finding is compatible with the idea that ellipsis leads to the activation of the antecedent phrase in memory rather than the construction of syntax at the ellipsis site: “It is important to note that our goal here is to examine whether syntactic structures are accessed or activated at the ellipsis site, not the narrower question of whether the parser incrementally builds such structures at the ellipsis site” (: , emphasis added). The pointer mechanism predicts reactivation of a previously mentioned phrase if the phrase is required for interpretation; thus the finding is consistent with the present proposal. To summarize, the psychological pointer mechanism can account for the interpretation of elliptical expressions without the need for inaudible syntactic structure at the site of ellipsis. That is, the pointer mechanism that is required for certain cases (e.g., ditto) can be readily extended for cases of ellipsis as well. The pointer mechanism is also consistent with evidence from psycholinguistic processing. In the following section we outline how ellipsis is licensed by constructions by detailing the English gapping construction.

4

The assumptions behind the design are not entirely clear to us, since it would normally be expected that the double-object construction should be primed in all three cases, assuming structural priming lasts beyond one intervening clause (Bock and Griffin , although see Bernolet, Hartsuiker, and Collina ).

   



. L: G    

.................................................................................................................................. Table . listed several well-known examples of elliptical constructions. As already observed, all such constructions are motivated in a general way by communicative concerns of efficiency. Yet at the same time, each individual elliptical construction has a distinguishable function in discourse and a corresponding different form (see also Miller a, ). For example, gapping () and pseudogapping (a,b) in English are similar in terms of the amount and type of elided material. In both constructions, the overtly expressed remnant contains two constituents that act as arguments or adjuncts to an unexpressed predicate recovered from an antecedent. In gapping, only the arguments or adjuncts are overtly expressed (), while the remnant of pseudogapping contains a tensed auxiliary as well (a,b). ()

Gapping ‘The more I touched her and she me [= and she touched me], the more I was reminded of Basya.’

() Pseudogapping a. ‘Zenobia likes the sales staff as little as she does me.’ [= as she likes me] b. ‘He’s cuddling it like he would you.’ [= as he would cuddle you] As Hoeksema () has observed, gapped and pseudogapped fragments bear different semantic relations to their antecedents. In a corpus of naturally occurring examples, he finds that the vast majority of instances of gapping ( percent) occur in coordination with their antecedent (as in ()). Moreover, the gapping construction requires contrastive stress (Culicover and Jackendoff : ). On the other hand, most tokens of pseudogapping ( percent) occur as the second part of a comparative construction (as in (a,b)), and the overt tensed auxiliary often provides the contrastive information (see also Miller  who estimates that as many as  percent of instances of pseudogapping are comparative). To see that gapping and pseudogapping constructions are not generally interchangeable, note that () and () become unacceptable if pseudogapping is used instead of gapping (), or vice versa (): ()

Pseudogapping ??5 The more I touched her and she did me, the more I was reminded of Basya.

() Gapping a. ?? Zenobia likes the sales staff as little as she me. b. ?? He’s cuddling it like he you. In order to understand how elliptical constructions can be represented, we provide the example of the English gapping construction (generalized to include “argument cluster We use ‘??’ instead of the more traditional ‘*’ in recognition of the fact that judgments are gradient and affected by many factors including context and intonation. 5



  

conjunction,” described below). A constructionist account allows us to specify that the construction involves two conjoined semantic propositions. The first proposition is expressed as a regular clause, while the second proposition is expressed formally only by exactly two filler phrases that designate arguments or adjuncts that contrast in meaning with two in the first clause. We can represent this as follows: ()

G (+ argument cluster conjunction) construction Register: formal Form: overtly expressed: [P(X, Y, Z*)], [ [X’, Y’]] Function: P(X, Y, Z*) P(X’focus, Y’focus, Z’*) X’ ≠ X; Y’ ≠ Y ; Z’ ≈ Z Determine second use of P using P function to a recently uttered simple or compound verb including tense, aspect, and voice. X, Y, Z: arguments or adjuncts. Underlining is used to indicate form as opposed to interpretation. Boldface indicates lexical stress (here, on X’ and Y’). Constituents are indicated by brackets. *:  or more.

The representation in () is best unpacked by considering an example, as in (): ()

A: You made me what I am today. B: And , .

As captured by the representation in (), gapping is restricted to formal registers, which predicts that it occurs much more often in written than spoken language (Tao and Meyer ; Hurford ). The content of the example in () makes clear that the context is, in fact, a formal or respectful one. While the words you and me are repeated in the gapped phrase in (), they are recognized to necessarily contrast with the intended referents in the first sentence; this non-identity requirement is captured in () as a requirement on the construction’s function. The understood predicate in () is made, here a simple verb in the past tense. The complement, what I am today stands in a ‘sloppy identity’ relationship in the first and second clause, since ‘I’ refers to person A in the first clause and person B in the second clause. The ‘≈’ allows for such sloppy identity. The correspondences required by the gapping construction in () for the example in () are made explicit below:6 P = made X: you (referring to person B) Y: me (referring to person A) Z: what I(person A) am today X’: you (referring to person A) Y’: me (referring to person B) Z’: what I(person B) am today 6

See Abeillé et al. () for relevant discussion and formalization of the parallel gapping constructions in French and Romanian.

   



The predicate P in our representation of the gapping construction in () specifies either an active or a passive construal. That is, active and passive predicates serve distinct functions in terms of information structure, since the actor is topical in an active transitive sentence, while the undergoer is topical in a passive sentence. Thus, the representation in () predicts that voice mismatches are not possible in the gapping construction and this prediction is borne out. That is, it is impossible to interpret the elided phrase as passive if the first predicate is in active voice (a), or vice versa (b). Here and below, following convention, when we wish to make an elided phrase explicit, it is represented by a crossed-out phrase, although, as already argued, we do not intend that the crossed-out phrase literally exists at the ellipsis site (see also section .). ()

a. ?? She ate ice cream, and string beans were eaten by him. b. ?? The duck was struck by a car, and a truck struck the goose.

The representation in () is appropriately general in allowing for cases that are not traditionally considered gapping such as that in (): ()

We visited [Jan on Monday] and [Y, on T]. (Beavers and Sag : (d))

Such cases have been referred to as “argument cluster conjunction,” and assumed to be distinct from gapping since if the verb were expressed in the second clause it would not intervene between the two constituents (Beavers and Sag ). But what is important to the gapping construction in () is the interpretation of the omitted verb, not its position relative to the two expressed arguments. Thus () is naturally accounted for by the representation in (), where P = visited; X: Jan; and X’= Y; Y: on Monday; and Y’ = on T. In this way, the emphasis on surface structure in constructionist approaches leads to a more general formulation than is possible from the derivational account (see also Goldberg ). We do not label the grammatical category of the [X’, Y’] phrase in () because it is not an instance of any familiar category. Instead of coining a new category label, or stretching an otherwise familiar category to include a unique type of constituent, constructionist approaches allow certain pairings of form and function to be restricted to particular constructions (Croft ; Culicover and Jackendoff : ). In the case of (), two argument or adjunct phrases correspond to a full propositional meaning. While the semantic interpretation involved in the gapping construction involves an identical interpretation of a predicate, which itself must necessarily be previously expressed, other types of ellipsis are more flexible. For example, the requirement that the elided predicate involves identical voice is weakened in the case of VP-ellipsis in limited contexts that involve contrastive topics (Kertz ) and/or cause–effect interpretations (Kehler ), as in (). ()

The problem was to have been looked into, but obviously nobody did look into the problem. (Kehler )

In an empirical judgment study, Kertz finds that examples like () are of intermediate acceptability when compared with VP-ellipses that involve matching voice, and those that



  

involve mismatches of voice in different discourse contexts. Sentence fragments (Morgan ; Stainton b) fall at the other extreme in that they generally require no overt linguistic antecedent at all as long as the intended interpretation is recoverable (Culicover and Jackendoff : –). Constructionist approaches readily allow for subtle differences of this kind on individual constructions. Individual elliptical constructions can be characterized by syntactic, semantic, discourse, and register properties, or they may underspecify aspects of these dimensions. Attention to these properties serves to undermine the idea that the constructions are simply shorter variants of full-fledged sentence patterns, or that they should all be accounted for in the same way. It instead supports recognizing them as constructions in their own right. Constructionist approaches do not stipulate a distinction between a ‘core’ part of grammar and some sort of ‘residue’ or ‘periphery’. Instead, we aim to account for all form and function correspondences, as is needed for any theory to be descriptively adequate. Thus, elliptical expressions that are restricted in terms of genre or which are not fully productive in terms of lexical options are also treated as ellipsis constructions, as Occam’s razor dictates that they should be, in the sense that they are captured by direct pairings of form and meaning in which key aspects of their semantics are unexpressed. For instance, the examples provided in ()–() are all lacking a main verb and yet each qualifies as a full utterance. (These might be characterized as instances of “deep anaphora” according to Hankamer and Sag .) Each is discussed in turn below. ()

Elise, Casey.

()

Down with Materialism and Up with Nature. That was Byron’s motto.

()

Yes we can! (Obama slogan )

()

‘Well, I never!’ I exclaimed.

()

WHITE CHRISTMAS? Right, as if.

The discourse functions of these ellipsis constructions go well beyond simple recovery of some previously mentioned content; in fact, they do not require the pointer function to reference any linguistic material at all. The example () (Elise, Casey) can be uttered without a linguistic antecedent as a means of introducing Casey to Elise. This use of paired proper names or descriptions, uttered with a pause and typically a gesture from the first person to the second, is conventionally interpreted to mean, ‘Elise, this is Casey.’ This simple construction is represented in (): ()

Form: [NP, NP], where NP is vocative and NP is a proper name or a definite description Function: Introduction: NP, this is NP

Several other phrases express a strong emotional response of one sort or another. For example, the Down with construction exemplified in () and () expresses the speaker’s strong disapproval of whatever is named by the noun phrase.

    ()



‘I gave up and lifted the bowl to my lips. It’s the new me. Down with etiquette.’

This conventional construction involves no linguistic antecedent and in fact has no nonelliptical counterpart, although it is elliptical in the sense of not providing an overt main verb. Notice that its interpretation requires a constructional analysis since its meaning is not compositionally derived from the words themselves. In fact, the same phrase, down with , has an almost opposite meaning in colloquial English if it follows a subject and copula (). Note, too, that adding a negation as in () does not solve the nonequivalence with (). ()

I am down with etiquette. (implies that the speaker approves of etiquette)

()

I am not down with etiquette. (implies something much milder than Down with etiquette.)

Another expression which has become conventional is Obama’s campaign slogan, Yes we can, which is an idiomatic instance of VP-ellipsis, but which requires no antecedent and conveys that ‘we’ can accomplish some contextually evoked agenda. The elliptical (Well), I never is a conventional phrase that only exists in a formal register in certain (stereotypically female) dialects. It implies that the speaker is appalled at some event that is contextually recoverable but not stated linguistically (e.g. (a)). It is infelicitous in contexts that do not lend themselves to expressions of outrage (b): ()

a. Well, I never! b. # Well, I never!

As if is another conventional phrase that is limited to certain (younger) dialects. It implies sarcasm indicating that the speaker views the elided, hypothetical event to be highly unlikely as is clear in (): () ‘ “You can ditch me in the crowd of tourists.” She laughed. As if. He would probably need a fire hose to get her off of him.’ In order for any of these constructions to be used with their conventional interpretations, speakers must recognize that they require special interpretations that do not follow from any general principles of composition or deletion. Thus each of these constructions represents a conventional pairing of interpretation with surface form: a construction. A few other specialized constructions involving ellipsis that have been discussed elsewhere are provided in Table .. If one wishes to account for all of the nuances of speakers’ knowledge of language, it is an inescapable conclusion that multiple constructions are needed. The following section demonstrates the fact that these constructions, while highly motivated (section .), differ in their specifics cross-linguistically.



  

Table . Less often discussed constructions that involve ellipsis, with examples and references Name of construction

Example

Let alone construction (Fillmore, Kay, and O’Connor 1988) Mad Magazine construction (Akmajian 1984; Lambrecht 1990) Coffee construction (Stainton 2006b; Heine 2011)

It’s no way to run a hotel, let alone a democracy. Him, a presidential candidate? Coffee? Tea? Biscuit?

. C- : F  E

.................................................................................................................................. We expect elliptical constructions to exist in every language, since they are motivated by general communicative pressures. The general motivation also ensures that such constructions are typically not difficult to comprehend, even upon initial encounter. At the same time, constructionist approaches predict that constructions vary in their specifics crosslinguistically (Croft ; Haspelmath ; Evans and Levinson ; Boas ), and ellipsis constructions are no exception. Thus, speakers need to learn exactly the nuances of how individual ellipsis constructions are conventionalized in each particular language. For example, at first blush, French contains several very similar elliptical constructions as English, e.g., gapping in (), sluicing in (), and not-stripping in () (examples from the Frantext corpus).7 () Elle conduisait la voiture et toi la moto. she drove the car and you the motorbike. ()

Elle sait qu’elle se marie dans quinze jours mais elle ne sait pas she knows that-she  marries in fifteen days but she  knows not avec qui. with whom ‘She knows she’s getting married in two weeks but she doesn’t know with whom.’

() Ton oncle Daniel your uncle Daniel

a le droit has the right

de jurer, to swear

mais pas toi ! but not you!

Cf. . The corpus examples were taken from contemporary texts (i.e., written after 1989). An English gloss is provided for all French examples in this section. A translation is also given if it differs from the gloss. 7

   



Yet, there are striking differences between the two languages with respect to the possibility of VP-ellipsis (Busquets and Denis ; Dagnac ; Authier , ; Abeillé et al. ). In particular, only a very limited number of verbs can be used in French VP-ellipsis, such as the modal pouvoir ‘can’ () and a few other modal-like verbs. Neither avoir ‘have’ (), nor être ‘be’ in the simple perfect (‘passé composé’) (), nor the passive auxiliary, nor copular verbs are allowed, while the corresponding English equivalents are fully acceptable. ()

Charles a piloté cet avion, mais François Charles has piloted this plane but François ‘Charles piloted this plane, but François couldn’t.’

n’a pas -has not

pu. could

() ??Charles a traversé l’Atlantique mais François n’a pas. Charles has crossed the-Atlantic but François -has not. ‘Charles crossed the Atlantic, but François didn’t.’ ()

??Charles est venu à la cérémonie, mais François n’est pas. Charles is come to the ceremony but François -is not. ‘Charles came to the ceremony, but François didn’t.’

Moreover, VP-ellipsis in French requires that the subject of the remnant be coreferential with the subject of the antecedent when the ellipsis site is within a relative clause (a,b), a subordinate clause, or in a comparative construction (Dagnac ). English has no such constraint as is evident by the fact that either translation in (b) is acceptable: () a. Charlesi pilote Charles flies b. ??Charles pilote Charles flies

tous les avions all the planes

qu’ili peut. that-he can

tous les avions all the planes

que François peut. that François can

Thus the comparable elliptical construction is much more restricted in French. French has a different means of expressing the function that VP-ellipsis commonly serves in English. In coordinating contexts, French has a specific type of stripping construction, in which a particle appears in the fragment. The particle varies depending on the polarity of both the antecedent and the fragment: aussi ‘too’ (positive/positive: ()), pas ‘not’ or non ‘no’ (positive/negative: ()), oui or si ‘yes’ (negative/positive: ()), and non plus ‘neither/not either’ (negative/negative: ()). () Charles a traversé l’Atlantique et François aussi. Charles has crossed the-Atlantic and François too ‘Charles crossed the Atlantic, and François did too.’ () Charles a traversé l’Atlantique mais pas François / mais François non. Charles has crossed the-Atlantic but not François / but François no ‘Charles crossed the Atlantic, but François did not.’



  

() Charles n’a pas traversé l’Atlantique mais Charles -has not crossed the-Atlantic but ‘Charles did not cross the Atlantic, but François did.’

François François

() Charles n’a pas traversé l’Atlantique et François Charles -has not crossed the-Atlantic and François ‘Charles did not cross the Atlantic, and François did not either.’

oui / si. yes / yes

non no

plus. more

The direct translations in English are (marginally) acceptable in the case of a positive antecedent (a,b), and then only in main clauses, contrary to the French counterpart (a, b). The translations involving a negative antecedent sound quite odd in English (a,b). ()

a. Charles crossed the Atlantic, and François too. b. Charles crossed the Atlantic, but not François / ?but François no.

() a. ?Charles crossed the Atlantic, and François said that he too / not him. b. Charles a traversé l’ Atlantique et François a dit que lui Charles has crossed the-Atlantic and François has said that him aussi / lui non. too / him no ‘Charles crossed the Atlantic, and François said that he did too / he did not.’ ()

a. Charles did not cross the Atlantic, and ?François neither / ?François not either / ?not François either. b. ?Charles did not cross the Atlantic, but François yes.

The French stripping constructions are available for all verbs and in all tenses, but unlike VP-ellipsis, due to the lack of a finite verb, they are unable to convey a distinction in tense or modality from the antecedent. In sum, even languages that have been in close contact for hundreds of years differ in the specifics of their ellipsis constructions. Constructionist approaches anticipate such differences, and can readily capture them within each language’s system of constructions.

. W  ‘’   

.................................................................................................................................. As was mentioned earlier, the constructionist approach to ellipsis proposes that while the requisite semantic structure must be recoverable (sections . and .), there is no ‘underlying’ syntactic structure at the ellipsis site, either unpronounced or deleted. In this section we review the facts that have led many researchers to assume that such underlying syntactic structure is needed, and counter that positing such structure raises more questions than it resolves. Ross (), and many others since, have observed certain intriguing “connectivity effects” between an expressed remnant and an antecedent word or phrase (see Merchant 

   



for a review). This is what has led many researchers to assume the existence of an identical, albeit unpronounced word or phrase at the site of the ellipsis. For example, case marking is sometimes determined by the antecedent clause, as illustrated in examples () and (). () A: Wem hilft der Lehrer? whoDAT helps the teacher ‘Who does the teacher help?’ B: Der Lehrer hilft Dem Schüler. the teacher helps theDAT pupil. ()

(Culicover and Jackendoff : )

A: Who does Sarahi like best? B: Sarahi likes herself i best.

Accounts that assume there is unpronounced syntactic structure predict such connectivity effects (e.g., Ross ; Merchant a). In fact, such accounts predict that overt form must be identical with what it would be if there were no ellipsis, because the assumption is that there exists an underlying level of representation in which there is no ellipsis. Yet there exist many examples in which connectivity effects do not hold. For example, in (b) the accusative me is preferable as a response to the question in (a), and yet it is clearly unacceptable in the non-elliptical (b’): ()

a: Who wants some ice cream? b: Me wants some ice cream! (? I!) b’: ??Me wants some ice cream.

This type of example would seem to provide evidence against deletion in favor of direct interpretation based on surface form and semantic recoverability (Barton ). In an effort to defend the deletion account, Merchant (a) suggests that the example in (b) should actually be based on a left-dislocation construction as in (), where the entire clause following the pronoun me is deleted. ()

Me, I want some ice cream.

But an analysis in terms of left-dislocation does not work for all relevant cases. For example, most English speakers prefer accusative pronouns in comparative ellipsis ((a) vs (b)), even though nominative case is required in the counterpart involving VP-ellipsis (b). Unlike the example in (), (a) cannot readily be analyzed as involving left-dislocation, because the dislocated version is unacceptable (): ()

a. I deserve it more than him. b. %I deserve it more than he.

()

a. ??I deserve it more than him does. b. I deserve it more than he does.

 ()

   ??I deserve it more than him, he deserves it.

Likewise, the attested gapping example in (a) involves accusative me, whereas the nonelliptical version involves nominative case (b). ()

a. ‘So you don’t have to trust me or me, you.’ b. So you don’t have to trust me or I don’t have to trust you.

What happens if we assume that the second clause involves left-dislocation, as in ()? ()

So you don’t have to trust me, or me, I don’t have to trust you.

Notice that the elliptical example (a) is most naturally interpreted conjunctively: ‘you don’t have to trust me and I don’t have to trust you.’ That is, the negative (don’t) is naturally interpreted as having wide scope over the disjunction (or) (via De Morgan’s Law). On the other hand, the non-elliptical versions (b) and () do not allow the wide-scope reading; they are instead interpreted to mean: ‘either you don’t have to trust me or I don’t have to trust you’ (Oehrle ; Johnson ). Instead, the wide-scope interpretation that is natural in (a) requires an explicit indication of the intended wide scope as in (), if there is no ellipsis: ()

It’s not the case that you have to trust me or I have to trust you.

Fiengo and May () recognize certain such cases where connectivity effects are lacking and attribute them to a process of “vehicle change” in which the proposed deleted structure is not identical with an overtly expressed form. This idea is taken a step further in recent work by Barros et al. (: ) who allow the elided material to be wholly distinct from any overt linguistic antecedent. That is, the unpronounced structure in () is assumed instead of that in (): ()

A: Ben left the party because someone wouldn’t dance with him? B: Yeah BETH it was.

()

A: Ben left the party because someone wouldn’t dance with him? B: Yeah Ben left the party because BETH wouldn’t dance with him. (Barros et al. : )

Barros et al. () make this move in an effort to avoid the apparent violations of island constraints that expressions like that in () would entail if the remnant were assumed to “move out” of the unpronounced structure. But of course the non-identity of deleted structure with an overt antecedent undermines the original argument in favor of deletion, which was based on the claim that the omitted material was necessarily identical with an overtly expressed form. As Merchant (: ) puts it, “structural approaches [to ellipsis] are based on connectivity effects” (emphasis added). Culicover and Jackendoff () provide still other examples in which some linguistic material appears to serve as an antecedent for the elliptical meaning, but the antecedent spans more than one sentence (), or is discontinuous ().

    ()

I know someone introduced you to me. I also know it was before last year. I just don’t know who introduced you to me or when before last year.

()

Pat invited Sam to the dance and Chris, invited Tad to the dance.



The fact that connectivity effects sometimes exist suggests that they serve to facilitate the correct identification of the relationship between overt and omitted material. That is, case marking often indicates the semantic role of an argument in an event, so the connectivity effects that exist are likely motivated by their tendency to facilitate comprehension. Alternatively, since certain constructions point to or evoke a specific verbal predicate, the overtly expressed arguments associated with that predicate may simply be primed to bear the case marking they normally bear. To summarize, connectivity effects in ellipsis constructions cut both ways. Sometimes the elided phrase aligns with the non-elided paraphrase (, ), but other times it does not (b, a, a, , , ). The examples in which connectivity effects are not in evidence argue against unpronounced syntactic structure in favor of an account that assigns interpretation directly to surface form. That is, one cannot both endorse “vehicle change” and assume “connectivity facts” hold, while maintaining a single rule for ellipsis.

. C

.................................................................................................................................. Because ellipsis constructions are motivated by general communicative goals to express our messages efficiently, they are expected to recur across languages. Communicative demands insure that elliptical utterances which require an interpretation that is not recoverable in context will not be felicitous. In addition, constraints associated with individual constructions bear directly on acceptability judgments. This short overview of ellipsis in terms of constructions has emphasized their motivated form, and a range of formal and functional properties. The proposal outlined here has much in common with other surface-based approaches to ellipsis, as these are also essentially constructionist in nature (Murphy ; Fillmore, Kay, and O’Connor ; Ginzburg and Sag ; Culicover and Jackendoff ; Osborne a; Miller a, ; Sag and Nykiel ; Osborne and GroB ). A psychological  mechanism is endorsed as a means by which particular overtly expressed linguistic material is evoked by ellipsis (see also Tanenhaus and Carlson ; Culicover and Jackendoff ; Martin and McElree ). This semantic pointer mechanism is independently required for non-elliptical expressions such as ditto and respectively, elliptical idiomatic expressions without linguistic antecedents (well, I never!), fragments (Ok, Tomorrow.), and examples that cannot be reconstructed by copying syntax from an antecedent (you don’t have to trust me, nor me (*don’t have to trust) you. Moreover, the non-derivational, semantic proposal is consistent with current psycholinguistic evidence, while syntactic copy and deletion accounts are not. Many elliptical expressions have quite distinctive interpretations and restrictions on use. Some ellipsis constructions (e.g., gapping) require a linguistically expressed antecedent;



  

others (e.g., sentence fragments) can occur as long as the intended meaning is recoverable in context. We offer an account of gapping that is both general, in that it can include cases of traditional “argument cluster coordination,” and restrictive in that it includes a constraint on register. In a comparison between English and French, French VP-ellipsis can be seen as much more restricted than English. We conclude that a constructionist approach that emphasizes semantics and surface structure, and which allows for distinctions within and across languages, is preferable to universalist copy and deletion proposals. Each construction posited makes predictions that are then testable against new data. A single general rule is not explanatory because every theory must ultimately account for the subtle differences among constructions that exist within and across languages. Thus invisible formal structure does not license the form, interpretation, and distribution of ellipsis, constructions do.

A We are grateful to three anonymous reviewers as well as to Jeroen van Craenenbroeck and Tanja Temmerman for very helpful comments on an earlier draft of this chapter. We also wish to thank the Einstein Foundation of Berlin for generous funding to AEG.

  ......................................................................................................................

    ......................................................................................................................

 ,  ,  ,   

. L       

.................................................................................................................................. F the perspective that takes language as a form of action, ellipsis is (part of) the phenomenon where use of verbal actions by an interlocutor is deemed less efficient for coordination in interaction than relying on the context, either of one’s own recent verbal actions (see both utterances in ()), somebody else’s verbal actions ((B), (B), and (B)), or of the physical environment (): ()

A: I’m sorry I upset you. I didn’t think I had. B: You didn’t. Somebody did but I can’t tell you who.

()

A: If he needs to interrupt, . . . B: then I’m sure he will.

()

A: Is he going to address . . . B: the European aspect of the problem?

()

Father (said to son apparently about to dive off a cliff): I wouldn’t if I were you.

This view of ellipsis, as a phenomenon seamlessly integrated in and enabling the joint action of conversational participants, has been obscured by theoretical linguistic approaches that abstract linguistic phenomena away from interaction and environmental resources and view natural language (NL) instead as a code, to be analysed in the model of formal languages with a specification of sentential syntactic organization and parallel mapping to propositional semantics as the explanatory basis. In such an approach, linguistic strings like those in () are analysed as (a) containing syntactically determined



, , ,  

ellipsis sites, in the sense that the interlocutor’s presumed silence is attributed some underlying structure, which (b) needs to be ‘licensed’ by special linguistic devices so that (c) a complete proposition including the meaning of the ‘missing’ part is recovered on the basis of identity with some other linguistically derived structure. Assumptions (a) and (b) immediately exclude () and () because here there is no obvious ellipsis site, nor can there be predefined linguistic rules which determine the interactive meaning of these utterances as continuations, interruptions, or explicit commentary on (aspects of) another’s talk. Assumption (c) immediately excludes the explanation of phenomena like the one in () as part of the model because here there is no linguistically derived content to rely on in deriving interpretation. As a result, such accounts of ellipsis restrict their attention to strings like the ones in the individual utterances in () where, it is claimed, the specifically linguistic knowledge that people exhibit in interpreting them circumscribes them as a distinct phenomenon. We argue here that this delimitation of the phenomenon, and its commitment only to explaining a subset of the data, show that the standard methodology of abstract sentencedescriptive grammars impedes proper understanding of what a unified account of ellipsis really shows: namely, language users’ know-how to employ and derive significance from multiple resources as they engage moment-to-moment with each other. For this reason, we argue, the standard competence–performance distinction that underlies the standard methodology has to be replaced by a different perspective in which the coordinated and context-dependent actions of conversational participants are central to the explanation. In this chapter, we sketch an alternative view of ‘syntax’ as an ability to act jointly, hence displaying properties in common with those that have been assumed up to now as characteristic of semantics, pragmatics, and processing models. From this perspective, we will show how the presumed ‘syntactic’ licensing of an ‘ellipsis site’ as in () can be accounted for without the assumption of an autonomous level of syntactic analysis. Under the same terms, we also aim to show how recoverability of ellipsis interpretations, including those without linguistic antecedents such as (), can be achieved on an incremental dynamic basis without assuming the necessity of sentence–proposition mappings. We do this through modelling the various phenomena within the architecture of Dynamic Syntax (DS, Kempson et al. ), an inherently incremental and action-oriented formalism for modelling interaction. We introduce DS in §. and then show how it captures the various interactive ellipsis phenomena in §. after we discuss the challenges for other standard accounts in the next subsections.

.. The ellipsis stalemate in competing domains of analysis Syntactic accounts of ellipsis rely on models of linguistic ability that postulate distinct types of declarative propositional knowledge as necessarily underpinning human linguistic performance. Having adopted this perspective, they then accordingly partition the phenomena by attributing to some instances causes distinct from those attributed to other intuitively similar cases. This is because such models, by ab initio definition of their remit, exclude in principle certain cases from consideration (see, e.g., Stabler ). For example, the cause of the surface appearance of the strings in (), so-called VP-ellipsis and sluicing, is assumed to be underlying syntactic structure that has remained unpronounced: since Ross

   



() and Sag (a), the ellipsis site is analysed as projecting complex sentential structure which is deleted at a late stage of the derivation under conditions of identity with that of an antecedent clause. However, even for this now narrowly circumscribed domain of phenomena, problems arise for such an account, which is symptomatic of the illegitimacy of this fractionating process. Firstly, the particular interpretations naturally recovered for various such ellipsis sites turn out to appear very diverse, which, due to compositionality considerations, then leads to constant differentiation of categories of ellipsis types (e.g. for VP-ellipsis: endophoric, exophoric, null-complement anaphora, tautosentential, discourseanaphoric, etc., all terms and phenomena illustrated in Miller and Pullum ; for various differentiated semantic categories of sluicing, see, e.g., Fernández et al. ). Secondly, it has been shown that identity of surface structure is not required for the licensing and recoverability of meaning of ellipsis sites (see, e.g., Hardt ). And, lastly, metatheoretically, the account lacks parsimony, since antecedents which present a single intuitive reading have to be assigned ambiguous underlying structures to allow for the multiple possible ellipsis types. From an empirical point of view, this makes such accounts psychologically unrealistic, since a left-to-right incremental version of them must invoke ambiguity to account for the subsequent interpretation of the ellipsis site. A particularly notorious such problem is the strict vs sloppy ambiguity (Sag a), where, for a single fixed interpretation of the antecedent clause, the ellipsis site allows for two interpretations, e.g., in () his can be taken to denote either John as ‘Bill checked John’s mistakes’ (the strict interpretation), or Bill as in ‘Bill checked his own mistakes’ (the sloppy interpretation): ()

John checked his mistakes, and so did Bill.

To avoid this problem, a competing semantic account (Dalrymple et al. ) presumes that ellipsis sites are syntactically simple with no hidden structure. The burden of explanation for the strict/sloppy ambiguity is then shifted to an abstraction over semantic content of a clausal antecedent, which involves either abstracting out only the subject content (the strict interpretation), or abstracting out all occurrences of the term occupying that subject position (the sloppy interpretation). Notwithstanding their differences, syntactic and semantic accounts share the property of being sentence-based: the elliptical string is taken to correspond to a sentence which is in some sense incomplete, with interpretation recovered from a clausal antecedent. Both thus confront the difficulty of characterizing sub-sentential utterances which provide no evidence of having a sentential antecedent and yet are able to achieve illocutionary acts, which Stainton (b) argues require a pragmatic inference account over and above whatever semantic or syntactic accounts of ellipsis might achieve: ()

Chef (to assistant): Flour! [command]

Moreover, there are sub-sentential utterances where no identical, linguistically provided antecedent is available on which the recoverability of some particular interpretation can be based: ()

A: Billy left. B: Billy? [‘Who is Billy?’] [clarification]

(Ginzburg )



, , ,  

.. Morphosyntactic licensing of sub-sentential talk The difficulty confronting both semantic and pragmatic accounts of ellipsis when applied to accounts of sub-sentential utterances, is that the morphosyntactic forms of such utterances remain the same whether they appear as sentential constituents or not (Merchant a; Ginzburg ; Gregoromichelaki ). For example, they display specific case requirements: () I Maria to egrapse to grama? Oxi, ego / *emena. the Maria it write. the letter no I/ *me ‘Did Maria write the letter? No, I did/*Me did.’

[Greek]

Contra Stainton (b), in such case-rich languages, the form has to match whatever is taken as the required interpretation, even in the absence of any sententially, or even linguistically, derived antecedent (Gregoromichelaki , ). For example, in a context as in (), den Arzt needs to be in accusative case:1 () [Context: A and B enter a room and see a woman lying on the floor] A to B: Schnell, den Arzt /*der Arzt ‘Quick, the doctor /*the doctor’ [command]

[German]

A further problem indicating the need to integrate formal constraints with interpretational ones, is that ellipsis is argued to be subject to island constraints (Ross ), until very recently, widely taken to be diagnostic of a syntactic phenomenon (see Merchant, this volume). For example, the so-called Complex NP Constraint is exhibited in so-called Antecedent-Contained Deletion/Antecedent-Contained Ellipsis constructions. In such cases, a dependency needs to be licensed within the restrictor of a quantifying element provided by the adjunction of a relative clause. A further elaboration of the restrictor provided by a second relative clause prevents such a dependency: ()

John had interviewed every politician [who Bill had interviewed].

()

*John interviewed a journalist [who Mary turned away everyone [who Bill had interviewed]].

The challenge of defining accounts of sufficient richness to express such phenomena has been addressed in an impressively detailed manner by Ginzburg () and colleagues, 1

Note that the claim here is not that a sentential structure or linguistically-based logical form needs to be recovered in order to interpret the case-marked NP (e.g. a verb like rufen in German); instead, as argued later, what is advocated here is that such morphological constraints restrict directly the role of referents within conceptualizations of events/situations (Gregoromichelaki ; Hough ), the latter conceived as non-linguistically mediated, geometric-topological structures (as in, e.g., Gärdenfors ), or perceptually grounded type classifications (see, e.g., Larsson ). There is no necessary one-to-one mapping from such conceptual structures to the words of a particular language.

   



within a constructional version of HPSG formulated in the Type Theory with Records (TTR) representational framework (HPSGTTR, Ginzburg ; Ginzburg and Miller, this volume). In this model, sub-sentential strings, as in (), are taken to express paraphrases of what a whole sentence formulation would denote; and so the sub-sentential elements are mapped directly onto the various semantic/pragmatic functions they are taken to accomplish. However, this attempt to analyse ‘fragments’ as commensurate to whole sentences means that the HPSGTTR framework retains conservative assumptions vis-à-vis syntax: fragments are modelled as ‘constructions’ of sentential type, involving sui generis syntactic mappings of sub-sentential constituents directly to ‘sentences’ paired with necessarily (quasi-)propositional interpretations. For example, one of the grammatically derived interpretations of the string in () will be ‘A asks B who A meant by the use of “Billy” ’. In consequence, each such sub-sentential string is assigned multiple such pairings and is by definition characterized to be as spuriously ambiguous as the constructions it is involved in. This means that the problematic aspects of the syntactic accounts, namely ambiguity instead of general mechanisms, is not sidestepped—on the contrary, it is wholly embraced as is natural in a constructional approach. However, as argued in Gregoromichelaki (), the interpretational options for such fragments are open-ended, so a constructional approach faces the problem of ignoring the radical flexibility and adaptability of linguistic resources. Additionally, there remains the empirical problem that morphological/syntactic restrictions on ‘fragments’ pervade any occurrence of subsentential utterances whether these depend on linguistic or wholly non-linguistic contexts (see () and ()):2 ()

[Context: A is contemplating the space under the mirror while rearranging the furniture and B brings her a chair] A to B: tin karekla tis mamas? / *i karekla tis mamas? [Greek] ‘theACC chairACC of mum’s? / *theNOM chairNOM of mum’s?’ (Ise treli? ) (‘Are you crazy?’) [clarification] (Gregoromichelaki )

The defined HPSGTTR constructions apply only to linguistically provided partial antecedents which are clarifiable, so the rules specified by the grammar will not apply to the intuitively similar clarification cases such as these. An account that aims to capture global syntactic generalizations while, at the same time, accounting for licensing idiosyncrasies without an independent level of syntactic representation over strings is the general framework of Minimalist Grammars (MG, Stabler ; Kobele ). The MG formalism, applied to elliptical phenomena in Kobele

2 Again the claim here is not that Accusative vs Nominative choice in Greek indicates the presence of linguistically articulated but hidden structure; on the contrary, as in (8)–(9), morphological case, like any other linguistic constraint, restricts the conceptualization of the ongoing or envisaged event/situation, in this instance, the roles of the entities referred to; see also n. 1. Our general claim in this chapter is that exactly this function of linguistic elements is the reason why linguistic structures routinely do not have to be completed up to a sentential level in order to be interpretable.



, , ,  

b, , requires, firstly, the relation between sentence string and semantics to be mediated by extensive transformations. This is because a copying account of ellipsis recoverability under identity is sustainable only on the basis of a decompositional mapping of surface appearance to semantic interpretation. This mapping is effected through various grammatical operations, like merge, move, and indeed e(llipsis), applying over fine-grained linguistic categories (‘types’) of expression, as constrained by global syntactic constraints such as the Shortest Move Constraint (SMC). This account presents a significant improvement on standard syntactic analyses, since an ellipsis site is considered, as in semantic accounts, syntactically atomic, i.e. no underlying syntactic structure is derived. Instead, ellipsis sites are generated due to particularized grammatical operations mapping empty or fragmentary strings to fine-grained syntactic categories. Because both syntactic categories and operations are so fine-grained, this account allows ellipsis sites to behave like pro-forms (as in Hardt ) whose antecedents are syntactically restricted to be particular ‘derivational contexts’, i.e. syntactic derivations that behave as functions on the potentially overt linguistic elements that are mismatched between the antecedent and the ellipsis site. Given that this account incorporates global syntactic restrictions like the SMC, it can deal with formal licensing constraints like the island restrictions of (). Additionally, the fact that it makes use of ‘derivational contexts’ as antecedents, instead of syntactic structure, means that it can deal with cases of mismatched antecedents, as in (), without having to postulate destructive structural operations or unnecessary ambiguity in the antecedent itself. However, this model still maintains standard assumptions regarding syntax, in that its explanations rely on assigning psychological reality to a view of syntax that is independent and distinct from other cognitive abilities. Syntactic categories, operations, and global constraints, like the SMC, are assumed to be specifically linguistic and not attributable to general cognitive constraints (see also Stabler ). This has consequences: for example, in order to fully account for phenomena like (), the grammatical characterization provided is not adequate. Given that MG is a competence model, it has nothing to say on how antecedents are defined. In order for antecedents to be provided, MG has to be integrated in a performance model, a parser/ generator (Kim et al. ; Kobele b). This is because the ‘derivational contexts’ invoked by the assumed pro-forms simply represent, as far as the grammar is concerned, decontextualized descriptions of the combinatorial licensing of linguistic elements. So, from our point of view which takes the grammar–processor distinction as unsustainable and an artifact of the view of language as a code, the MG account (a) introduces theoretical redundancy because syntactic restrictions are not modelled on the basis of general cognitive processes directly; (b) in the case of ellipsis, this has the consequence that such phenomena are analysed as involving various cognitively arbitrary category specifications or being explainable on the basis of otherwise general processing explanations; and (c) as a consequence, because the MG grammatical account is confined to linguistic structures, it is unable to account for intuitively analogous non-linguistic antecedents such as () and (), since, even embedded in a performance model, there is no linguistic derivation in the context to be invoked as the antecedent grammatical operation. Finally, along with all the accounts already examined, given that the syntactic derivations of MGs license whole sentential strings paired with propositional interpretations, it will be unable to account for dialogue phenomena like (), as we will now argue.

   



.. Ellipsis as completability Accounts of ellipsis assuming a sententialist methodology struggle to address the challenge of the dynamics of conversational dialogue. The problem posed by conversational exchanges is that people commonly and purposefully do not utter full sentences. Conversations are replete with fragmentary forms which build incrementally on what has gone before, in which participants effortlessly switch between speaker and hearer roles, and commonly no-one in the conversation will have anticipated its overall content in advance: ()

A: B: C: B: C: A:

We’re going to . . . Burbage to see Auntie Ann with the dogs? if you look after them in the garden unless it rains.

If one attempted to preserve the assumption that all fragments are independent incomplete sentences, one might consider data like () as at least five elliptical sentences each to be mapped onto representations of propositional content; but it is far from obvious what source to provide to any such posited underpinning sentential structures when the overall structure cannot be anticipated. Even the illocutionary flavour of the subparts may shift as participants are more or less authoritative, or in disagreement: ()

A: What this shows is that B: you are completely wrong.

An account in terms of sentential ellipsis of the above misses the significance of B’s action in () as an interruption and the ones in () as continuations. In any case, such an ellipsis account cannot be sustained since this split utterance phenomenon can take place at any point in an emergent structure (Gregoromichelaki et al. ; Howes et al. ), and affects every syntactic/semantic dependency: preposition– complement (), reflexive–antecedent (), quantifier–pronoun binding, negative polarity, quantifier–tense (), even head–complement dependencies (): ()

Joe: We were having an automobile discussion . . . Henry: discussing the psychological motives for Mel: drag racing in the streets. (Sacks )

()

[Context: smoke coming from kitchen, Bob emerging] Mary: Did you burn . . . Bob: myself? No fortunately not.

()

A: Has every player handed in . . . B: his registration form? A: or any other documents?

 ()

, , ,   A: B: A: B:

Will you . . . resign? I think not unless Bill, uh, forces you to? Yeah.

Therefore, no account that relies on specialized constructions or syntactic types, instead of general mechanisms, can provide an adequate explanation for this phenomenon. Instead, sub-sentential strings need to be licensed by the grammar incrementally and without the requirement that a whole sentence needs to be licensed eventually (as in Kobele b and others). This is because such sub-sentential strings are not only interpreted incrementally but also can be used perfectly naturally to perform speech acts that rely on their very incompleteness (Gregoromichelaki ): ()

A: And you’re leaving at . . . B:  o’clock.

But even incremental syntactic accounts (e.g. Poesio and Rieser , or a conceivable extension of Kobele b, ) cannot deal with such data if they maintain an independent level of syntactic analysis since, as () shows, splicing together the two strings will yield ungrammatical results. Given the rapidly accumulating experimental evidence that parsing and production are both incremental and interdependent in achieving action and representational coordination during interaction (Pickering and Garrod ), accounts of sub-sentential utterances as somehow incomplete cannot be sustained. Firstly, as we saw in () and () for adult interaction, sub-sentential talk underpinned by the context dependency of human interaction can achieve any function that sentential and propositional units perform. Secondly, such utterances form the basis of language acquisition. Children at the one-utterance phase can perform perfectly adequate verbal actions employing sub-sentential strings relying on the adult caregiver to provide what they lack at that stage, namely, the requisite conceptual breakdown of a holistic situational representation so that new linguistically expressible concepts can be acquired: ()

A (pointing to an empty mooring site on the canal): Daddy B: Yes, you were here with Daddy yesterday, clearing out the boat. That’s right, dear.

Moreover, the split utterance data cannot be excluded from the core remit of the grammar as performance disfluencies since they also constitute effective verbal actions for the language-acquiring child to coordinate with caregivers: ()

Carer: Old McDonald had a farm . . . On that farm he had a Child: cow.

()

[addressing class of new nursery intake] Teacher: Your name is Child: Mary Teacher: And your name is Child: Tommy.

   



Instead, in our view, the characterization of sub-sentential data needs to make reference to the dynamics of real-time processing expressed in a vocabulary that allows contribution from environmental resources and non-linguistic cognitive domains at a sub-sentential level. Such a model, in our view a grammar, needs to: • include a set of mechanisms for inducing recovery of content from both linguistic and non-linguistic sources; • model word-by-word incremental, semantic licensing; • model the context as an evolving record of all inputs, including words, structures, and procedures that induce such incrementally developing content; • define shared mechanisms for parsing and production. The model we sketch below, Dynamic Syntax (DS), meets these criteria. In DS, the pervasive use of sub-sentential elements across dialogue turns and participants will not only be wholly predicted, but will be seen to serve a key purpose in communication by narrowing down the otherwise mushrooming alternative structural and interpretative analyses through the ongoing interactive coordination of participants (see Gargett et al. ; Eshghi et al. ). We will argue that this perspective not only illuminates the sharing of linguistic actions in context, as seen in ()–(), but also provides a processing approach which reveals a novel and surprising meta-theoretical generalization: namely, the lack of need for conceptual distinction between syntactic vs semantic vs pragmatic procedures in the integration of verbal actions with coordinated joint action. In all three domains of traditional linguistic inquiry the mechanism of jointly developing initially underspecified information operates uniformly. Just as pronominal anaphora construal has been shown not to be definable as confined within sentential, speaker-turn, or linguistic boundaries, ellipsis needs to be modelled in the same way, as an anaphoric mechanism both relying on and sustaining participant coordination and context-dependent action. This is achieved in DS by modelling both pronominal anaphora and ellipsis through the initial projection of temporarily underspecified contents imposing their resolution as a goal for all conversational participants and thus subsequently resolved by exploitation of either environmental or linguistic resources. Furthermore, we argue, in contrast to any other formalism, this perspective allows us to extend the parallelism beyond ellipsis and pronominal anaphora, to the explication of the function of syntactic mechanisms. Syntactic dependencies introduced by fragments, morphosyntactic constraints introduced by e.g. case morphemes, but also the mechanism underpinning long-distance dependency, will be modelled as introducing initially underspecified structural projections with goals imposed on either participant in a conversation to resolve them either through context or by offering an appropriate subsequent verbal action.

. D S

.................................................................................................................................. Dynamic Syntax (DS) is a grammar formalism based on the psycholinguistically inspired action-based modelling of NL string–interpretation mappings in context. NL syntax is reconceptualized to be no longer its own level of representation, but, instead, a set of



, , ,  

licensing actions for inducing or linearizing semantic content, incrementally, predictively, and on a word-by-word basis. Parsing and production are interdefined as coordinated activities operating in tandem, each having access to the same set of construction steps. Context in DS is modelled as including an incrementally evolving record of all the actions employed and their outcomes, enabling recoverability of any of these for reuse. The interaction achieved in split utterances emerges as an immediate consequence, since both parsing and production employ predictive actions which make available to each interlocutor, at each sub-sentential step, the role of parser or generator.

.. String-content mappings To model this action-directed perspective, DS is founded on a dynamic modal logic that defines the transitions among states taken to constitute the current context of processing at each point in a parse/generation task (see Kempson et al.  for formal details). The accessibility relations among these states are defined through actions which license goaldriven, time-linear transitions from state to state. Such states can be taken to model the total context of each processing step, linguistic and non-linguistic, so that the whole system licenses mappings from context to context. In DS, particular conceptualizations of eventualities are taken to involve actions of building (parsing) or linearizing (production) a semantic tree whose nodes incrementally come to reflect the content of some utterance. Keeping in mind that processing can start at any point, if we take the case of aiming to achieve a propositional structure, the first step is a one-node tree that merely states the goal to be achieved, namely, to derive a formula of propositional type (Figure ., above +), indicated by the requirement ?TyðtÞ, the query symbol ? indicating this is a goal as yet unrealized. This single node tree is then incrementally enriched as word-by-word processing proceeds, eventually leading to a complete tree of propositional type with no goals left outstanding (Figure 9.1, below +), a result achieved through combining grammatical actions with content derived from context. DS trees are invariably binary, reflecting functor–argument structure (argument node on the left branch, functor node on the right), with a pointer ◊ identifying the node under development). Each node is annotated, not with words, but with terms of a logical language (e.g. Mary 0 ), and their type (e.g. Ty(e)), these terms here being subterms of the ?Ty(t), ⇓ Upset′(Mary′)(John′)(S),Ty(t), S, Ty(es)

Upset′(Mary′)(John′),Ty(es → t) John′, Ty(e)

Upset′(Mary′), Ty(e → (es → t)) Mary′, Ty(e)

Upset′, Ty(e → (e → (es → t)))

 . Processing John upset Mary in DS

   



resulting propositional representation at the root node. The representation includes an event/situation argument S of type es, enabling tense/aspect construal (we suppress details here; see Gregoromichelaki ; Cann ).3 Since event/situation arguments are first-class citizens of the combinatorial structure, predicate assignments reflect this: Tyðe ! ðes ! tÞÞ the intransitive-verb type for combining with a subject first and then the situation argument; and Tyðe ! ðe ! ðes ! tÞÞÞ, the transitive type.

.. Formalization of tree structure and incremental tree development In order to talk explicitly about how such structures are constructed incrementally, trees are formally defined, so that their incremental construction can be achieved through actions that induce the requisite tree development. DS adopts a logic with modalities indicating not only what is currently true about the tree but also what is to be true at future developments of the tree (Blackburn and Meyer-Viol ). For example, the daughter relation is defined as h↓i: h↓ia holds at a node if a holds at its daughter (with variants h↓0 i and h↓1 i for argument and functor daughters respectively). There is also the inverse mother relation h↑ia with argument/functor variants, h↑0 i, h↑1 i. Thus a DS grammar includes a set of actions which are procedures for building/ linearizing trees: ‘syntax’ is just the subset of actions which unfold tree structure under the guidance of linguistic elements. Syntax is formulated as the employment of packages (macros) comprising simple atomic actions—make(X) for creating new nodes, go(X) for moving the pointer, put(Y) for annotating nodes, where X and Y are tree relations and node annotations (labels) respectively. Computational actions are language-general macros not needing a lexical trigger: these introduce tree relations; move the pointer ◊ around some partial tree under construction (e.g. , which moves the pointer out of a node once its type requirement is satisfied); remove satisfied requirements (); and perform function-application (-) operations compiling information up a tree on a bottom-up basis once all daughter decorations are suitably provided. Primary amongst these is the construction of ‘unfixed nodes’ (*-): such actions are defined using the Kleene star operator * annotating a so-called ‘unfixed’ node with the specification h↑ iTnðaÞ, which indicates that this node is related to some node a along a sequence of zero or more mother relations to it. Lexical actions are macros triggered by individual words of the particular NL whose actions also perform the same tree or string development along with introducing semantic content on the nodes of the tree.

3 In this exegesis we ignore quantification, merely noting that NPs (DPs) in DS project actions that lead to the construction of arbitrary names denoting a witness of type e, with scope variation expressed as anticipatory statements of term dependency incrementally projected during the construction process (Kempson et al. ; Cann et al. ). Proper names will be shown here in the abbreviated form John0 , Bill0 , etc. Tense/aspect/modality specification, defined as sortally restricted event terms, are indicated by Si,j, . . . (Gregoromichelaki ; Cann ).



, , ,  

Partial trees: imposing constraints on update. The underpinning dynamic which drives NL processing is the gradual resolution of initial underspecifications. There are three basic types, each with an attendant requirement for update: () content underspecification, expressed with insertion of a typed placeholding metavariable as the content that is contributed by anaphoric elements: U : e, U : e ! ðes ! tÞ; such content is accompanied with the requirement ?∃xFoðxÞ indicating the constraint that the metavariable needs to be replaced by a proper value; () type underspecification, either expressed through requirements ?Ty(e), ?Tyðe ! ðes ! tÞ etc., or by assigning an underspecified type Ty(U) which will then also be accompanied by the requirement ?∃xTyðxÞ; () tree-relation underspecification (‘unfixed nodes’), expressed as h↑ iTnðaÞ with requirement ?∃xTnðxÞ. Such underspecifications all allow modal constraints on their resolution, giving rise to a variety of locality restrictions. For example, anaphoric expressions can impose locality constraints associated with limits on recoverability of their antecedent. Locality within a given predicate domain is definable as the tree traversal of an unbroken functor path: arguments local to a given predicate meet the characterization h↑0 ih↑1 iTnðaÞ (i.e. there is a path up one argument relation plus a possibly empty sequence of function-path relations from the current node to the dominating Tn(a) node). Reflexive anaphors are characterized as projecting a restriction on putative antecedents that they be ‘co-argument’ terms, i.e. holding along the relation h↑0 ih↑1 ih↓0 iTnðaÞ. Conversely, pronominals exclude as antecedent any formula at a node standing in such a local relation; and wh-pronouns contribute a metavariable (WH), defined as awaiting substitution by a term, the answer, in some future utterance, a yet further long-distance dependency effect (see Gregoromichelaki 2006, 2013 for a reformulation of Binding Theory in DS terms). Unfixed tree relations vary as to whether or not their resolution is subject to such tight locality restrictions. An unfixed node within some tree is identified by the annotation h↑ iTnðaÞ; ?∃xTnðxÞ which defines it as unfixed but within a tree whose tree root is Tn (a). There is also a more specific localized variant defined as h↑0 ih↑1 iTnðaÞ, i.e. the tree root now encompasses a local predicate–argument sub-tree. Morphosyntactic specifications also involve locally statable constraints. Accusative case marking, for example, is definable as imposing a requirement on a node of the form ?h↑0 iTyðe ! ðes ! tÞÞ, dictating that in the outcome, it must be immediately dominated by a predicate node. In every case, what drives processing is the set of requirements lexically or computationally imposed: such requirements operate as predicted goals for either participant in the joint task to accomplish that constraining of future developments. Of these, it is underspecification of hierarchical position within a tree that lies at the core of the account, underpinning many discontinuities, and expressible through the construction of unfixed nodes. Figure . shows how the building of a locally unfixed node (-*-) can feed lexical actions, so that in combination, the word order properties of each particular NL are derived. In some languages this underpins scrambling. In English, this mechanism, in combination with the lexical action of the verb, induces the imposed SVO (subject–verb– object) ordering: the DP before the verb is taken to annotate a locally unfixed node; its

    ⟶ Local-*-Adj

?Ty(t),

⟶ John

?Ty(t),Tn(0)

?Ty(t),Tn(0)

?Ty(e), ⟨↑0⟩⟨↑1∗⟩Tn(0), ?∃x.Tn(x), ⟶ Completion



Johnʹ : e, ⟨↑0⟩⟨↑1∗⟩Tn(0), ?∃x.Tn(x),

?Ty(t), Tn(0),

Johnʹ : e, ⟨↑0⟩⟨↑1∗⟩Tn(0), ?∃x.Tn(x), ⟶ upset

?Ty(t) Johnʹ : e, ⟨↑0⟩⟨↑1∗⟩Tn(0), ?∃x.Tn(x)

Spast

?Ty(es → t)

?Ty(es → (es → t))

?Ty(e)

?Ty(e), ⟶ Merge

Upsetʹ Ty(e → (e → (es → t)))

?Ty(t)

?Ty(es → t)

Spast

Ty(e), Johnʹ : e

?Ty(e → (es → t))

?Ty(e),

Upsetʹ Ty(e → (e → (es → t)))

 . Unfolding structure for John upset . . .

position as logical subject (or not) is determined by the action sequence projected by the following verb, which among other actions includes such specification. The left-peripheral nature of long-distance dependency is expressed through the building of an unfixed tree relation without such a single-clause locality restriction, which involves a computational action defined in the presence of a node requiring completion to a propositional type TnðaÞ; ?TyðtÞ with a precondition that the node is not already developed in any other way, relative to which the action defined licenses the construction of an unfixed node annotated



, , ,  

as h↑ iTnðaÞ,Ty(X),?∃xTnðxÞ, i.e. with a resolution of this underspecified tree relation having to be at some possibly later point in the construction of the tree being developed from Tn(a) (this is the final step of 4 in Figure .); hence the potentially ‘longdistance’ discontinuity effect. In all tree developments, the closing stages invariably involve modalized functionapplication (beta-reduction) steps which compile up the content of all non-terminal nodes to finally yield a complete tree with no requirements outstanding (see Figure .).

.. Linking trees through term sharing In addition to the incremental projection of predicate–argument structures, a freely available computational action licenses paired ‘ed’ trees, which are in effect correlated through a conventionalized anaphoric device that results in a shared term appearing in both such trees. Formally, a transition from some node in the main tree onto the top node of a new tree is licensed, this new tree-beginning having imposed on its development a requirement that it include the term appearing on the node from which the transition was initiated. Relative clauses are the paradigm case: the relative pronoun triggers actions which reflect the LINK transition and some initial construction within the newly emergent tree of an unfixed node. The WH-metavariable then functions in a fixed anaphoric way to provide the required copy from the main tree, thus creating the resultant shared term in the two trees (see Figure ., which displays the outcome of processing the string John, who smokes, left).5 Such ed trees provide opportunities mid-sentence for NL processing to shift temporarily to a distinct structure for purposes of elaboration, expansion, explanation, etc. of terms in the main structure. And this can happen either within a single individual’s utterance, giving structures like relative clauses, Hanging-Topic Left-Dislocation, clausal and phrasal adjuncts, etc., or across speakers where the effects include clarifications, confirmations, continuations, etc. (see e.g. Gargett et al. ). This is because LINK-

LINKed Tree ⟨L–1⟩Tn(n), Smokeʹ(Johnʹ)(Sj), Ty(t) Sj

Matrix Tree Leaveʹ(Johnʹ)(Si) ∧ Smokeʹ (Johnʹ)(Sj), Ty(t)

Smokeʹ (Johnʹ), Ty(es → t) Smokeʹ, John’ , Ty(e) Ty(e → (es → t))

Leaveʹ (Johnʹ) Ty(es → t)

Si

Leaveʹ, Tn(n), John’ Ty(e) Ty(e → (es → t))

 . Result of parsing John, who smokes, left

4 5

This use of the term Merge is a unification step, not to be confused with its use in Minimalism. The arrow linking the two trees depicts the so-called link relation and hL1i reads as ‘linked to’.

   



adjunction allows indefinite iteration of adjunct extensions of arbitrary type, a common way of extending utterance exchanges: ()

A: B: A: B: A:

Sue is on her way back from . . . Paris with the dogs? and the parakeet. All in the car. Even Susie, the cat.

The fact that ed trees temporarily shift processing to an independent adjunct tree to elaborate a term on the main tree provides the basis for a requirement of anaphoricity from the one structure to the other. It also explains why this form of tree relation is what is needed for modelling sluicing (see e.g. Merchant ), amongst other structures which may be sensitive to island constraints as we shall see in detail in §.. for antecedentcontained ellipsis. In such cases, for example, where relative clauses constitute the linked ‘islands’, the relative pronoun provides the copy that links the newly introduced tree to the main tree. This copy process creating the shared term is  island-constrained with respect to its antecedent. Nonetheless, since that copied term has to occur on an unfixed node as the new tree is being introduced, the resolution of its position within that new tree is domain-sensitive: it must occur within the construction of that very tree. In this sense, island constraints in DS emerge from the need to shift processing to another domain along with needing to complete this structure before resuming processing of the one temporarily interrupted.

. T D  E

.................................................................................................................................. With the formal underpinnings of its semantic tree construction process to hand, we now describe the DS model of ellipsis. We show how the incremental notion of DS context can account for the varying types of elliptical forms found in interaction.

.. Context in DS: Mechanisms for recovery of content at the ellipsis site The concept of context appropriate for a processing-oriented account is substantially richer than is expressible in either model-theoretic accounts or semantically blind syntactic accounts. For this reason, context in DS is conceived as a dynamic, multi-modally induced record of (a) words; (b) conceptual content notated as tree structures as described above; and (c) the sequence of actions in building the emergent strings and trees (Cann et al. ; Purver et al. ; Eshghi et al. ). Since a DS grammar, through the imposition of predicted goals, models the recovery from memory of the range of options available for the next processing steps, this set of options is attributed psychological reality so it also constitutes part of the model of context. To model context time linearly, we follow Sato (a), Purver et al. (), and Eshghi et al. () in defining this range of options as a



, , ,   ‘john’ local *-adj

‘upset’ john

T1

T4

thin

comp

upset T7

T6

T12

link-adj T0

link-adj *-adj

T2

abort john

T3

T5

thin

T9

abort

T11

abort

comp T10

 . DS parsing context as a graph: Actions (edges) are transitions between partial trees (nodes)

Directed Acyclic Graph (Context DAG) as in Figure ., where each node represents the current (partial) tree and each edge in the graph records the action taken. The entire DAG represents the search space for parsing/generation, which will also include any conceptual content derived from context. In Figure ., the top path from T to T is the chosen one given the lexical input, and corresponds to the derivation outlined in Figure .. The context of a partial tree is then the path back to the root of this graph; and actions, as well as conceptual content notated in tree form, are recoverable through a mechanism of backwards search for reiteration/reuse in creating new construals. As a result, there are three basic mechanisms by which an ellipsis site, being an underspecified element awaiting resolution, can exploit the Context DAG for content recovery: (a) Reuse of content (semantic-formulae) from some (partial) tree on the Context DAG, which can include recoverability of information direct from the utterance scenario, yielding indexical construals. (b) Reuse of sequences of actions from the DAG (sequences of DAG edges). (c) Direct reuse of structure, i.e. extension of some (partial) tree in context.

.. Content underspecification and recoverability through copying or action replay Forms of the ellipsis site. We take first the familiar cases of English VP-ellipsis. Under DS, these involve an initial content underspecification projected at the ellipsis site, with optional triggers for content recovery in the form of strings such as do so, too, and bare auxiliaries. Auxiliaries are taken, like pronouns, to project a temporary content placeholder of predicate type (e.g. U : e ! ðes ! tÞ) with an accompanying requirement ?∃x:FoðxÞ which triggers the context search. Recovering predicate content. In such cases, two types of information are recoverable from context: (a) a formula content copied from some tree; (b) a sequence of actions to be reiterated at the ellipsis site.

    Tree as Context:

Tree under Construction:

?Ty(t)

Upset′(Mary)(John′)(Spres) Spres

Upset′(Mary)(John′) John′



Bill′

Upset′(Mary′) Mary′

?Ty(es → t)

Spres

?Ty(e → (es → t)) Ty(e), U, Annoy′ ?∃xFo(x),

Upset′

substitution Tree as Context:

Tree under Construction:

Upset′(Mary)(John′)(Spres) Spres

?Ty(t)

John′

Upset′(Mary′) Mary′

?Ty(es → t)

Spres

Upset′(Mary)(John′)

Upset′

Bill′

Ty(e → (es → t)), U, ?∃xFo(x),

substitution

 . Substitution from context at the ellipsis site of (): Pronominal anaphora (top) and VP-ellipsis (bottom)

The two mechanisms fetching such information to the new processing task through backwards search are implemented by the pair of computational actions Substitution and Regeneration.6 Figure . displays the identical substitution process for (a) a pronominal and (b) an elliptical expression as in (): ()

A: John upsets Mary. B: Bill annoys her. / B: Bill does too

Figure . then displays the regeneration process at an ellipsis site for cases like B’s answer in (): ()

A: Who upset himself? B: John did.

The new content in B’s answer is derived through rebinding of the variable projected by the anaphor to the newly introduced subject. This is achieved by searching the DAG and then reusing the actions involved in constructing the question to flesh out the ellipsis site in the answer, as triggered by the metavariable placeholder projected from the auxiliary. The processing of the question in () involves parsing of the subject who, constructing a predicate as indicated by the verb, and constructing its object argument; these actions, 6

For formal definitions, see Kempson et al. ().



, , ,  

Tree under Construction

Context

Who upset himself? John did

Tree:

?Ty(t)

Upsetʹ(WH)(WH) (Spast)

SPAST

Upsetʹ (WH)(WH)

WH

?Ty(es → t)

Spast

Johnʹ upset, himself, tree-completion, WH-substitution

Ty(e → t), U,

Upsetʹ(WH) Upsetʹ(Johnʹ)(Johnʹ)(Spast) WH

Upsetʹ Spast

ACTIONS:

Upsetʹ(Johnʹ)(Johnʹ)

Johnʹ

Upsetʹ(Johnʹ)

{. . . , upset-macro, himself-macro, tree-completion macros}

Johnʹ Upsetʹ action retrieval: regeneration

 . Action replay from context at the ellipsis site

having been stored as a sequence in context, can be reused in the next stages. But with John’ now annotating the subject node, these actions will now yield a rebinding of the object argument to the new local subject. The effect achieved is the same as the higher-order unification account of Dalrymple et al. () but without anything beyond what has already been used for the processing of the previous linguistic input and, consequently, without any need to assign some distinct type of expression to the elliptical element did or the subject John. All that has to be assumed is that the metavariable U contributed by the anaphoric did can be updated by an action sequence taken from the DAG context. This duality of mechanisms for content recovery provides a way of capturing all strict/ sloppy ambiguities observed in several forms of ellipsis. Either content specifications or sequence-of-action specifications can be searched for and reiterated, the first preserving some previous construal, the other preserving only the pattern of construal: ()

A: Bill will help his students B: John will too but less willingly.

   



This gives precisely the right basis for the ambiguity without having to invoke ambiguity of the antecedent. In (), a strict construal λx.Help0 (Students0 (of – Bill0 ))(x), ‘help Bill’s students’, is carried over as the fixed predicate content to be predicated of the subject John0 . The sloppy interpretation involves the sequence of actions associated with processing help + his + students in the first conjunct reapplied to the new subject John0 . This is not a mechanism identified specifically for ellipsis: the same type of analysis applies to the DS analysis of quotation (Gregoromichelaki ) and two subcases of pronominal anaphora too, i.e. coreferentiality and so-called lazy pronouns. Coreferential construals constitute a replication of some antecedently constructed content of individual type e: ()

A: John came into the room. B: He looked very sick.

Lazy construals involve recovery of actions to be rerun at the site indicated by the pronoun to yield some requisite term that is distinct from that picked out by the antecedent. In (), them is resolved by rerunning actions used to process his keys in the previous clause: ()

A: John always keeps his keys in the bowl. B: Michael just dumps them down anywhere when he comes in.

Under this view, neither ellipsis sites nor pronominals deviate in any way from the usual processing mechanisms DS defines: like all other morphosyntactic and lexical specifications, such elements impose predictive goals constraining the subsequent actions of the interlocutors as these unfold in the DAG model of context; general processing mechanisms like backwards search, Substitution, and Regeneration can then be employed, by either interlocutor, to achieve appropriate expansions of such potentially radically underspecified content in ways that reflect their own purposes. Since the DAG unfolds and constrains these future processing options on a word-by-word basis, employing and resolving underspecification can be achieved either through the exchange of propositional contents, as seen above, or through the exchange of sub-sentential elements that exploit verbal or non-verbal actions for becoming integrated in the functioning of the participants’ joint actions.

.. Licensing sub-sentential utterances: Morphosyntactic constraints Use of sub-sentential utterances in dialogue shows how particular morphosyntactic forms impose goals on interlocutors that curtail the processing paths compatible with further processing or backwards search. For example, as we saw earlier, so-called wh-pronouns project a specialized metavariable requiring instantiation externally to the tree under construction, this in question–answer pairs being the answer to that question. On the other hand, the metavariable provided by reflexives is restricted to obtaining a locally provided antecedent. These two types of constraint can combine leading to interpretations that are very finely constrained: short answers to a wh-question (as in ()) need to be interpreted within the structure which that question projects, while the local substitution imposed by a



, , ,   Context Tree:

becomes

Upset′(WH)(John′)(Spast)

Spast

Upset′(WH)(John′)(Spast)

Upset′(WH)(John′)

John′

Tree under Construction:

(Spast)

Upset′(WH)(John′)

John′

Upset′(WH)

SUBSTITUTION WH

Upset′

Upset′(WH) WH, Ty(e),

Upset′

 . A short answer with binding restrictions

reflexive as shown in Figure . will lead to a cascading effect affecting all occurrences of the WH-metavariable, thus yielding an appropriate interpretation of the answer. ()

Q: Who did John upset? Ans: Himself.

Sub-sentential utterances may also bear morphological features, which are expressed in DS via imposition of predictive goals to interlocutors to pursue processing paths that conform to some particular specification of the conceptual content being developed. Like indexical pronominals, imposing, for example, gender restrictions in English, accusative case marking, as in ()–(), introduces the requirement: ?h↑0 iTyðe ! ðes ! tÞÞ, a restriction that a node so annotated be immediately dominated by a node of predicate type. Since backwards search through the DAG will not fetch the appropriate content in this case, the imposition of such morphological constraints has the effect of imposing particular conceptualizations of the utterance situation becoming newly available as processing options in the DAG. Since the DS-DAG is not restricted to representing specifically linguistically derived content, such morphologically triggered context enrichment does not presuppose the mediation of any syntactic derivations. Nevertheless, unlike other frameworks (e.g. Ginzburg 2012), any DS rules applicable to linguistically derived content will be able to apply unchanged to the subsequent development of such contextually derived content (as in e.g. (12)).

.. Licensing split utterances Given that processing in DS is modelled as incremental and predictive, the licensing of subsentential utterances is achieved word-by-word without requiring necessary expansion to a propositional structure for a string to be interpreted or produced. So sub-propositional strings can be employed to perform speech acts as naturally as propositional ones, as seen in () and (). The DS explanation of the licensing of split utterances without having to assume inference of propositional intentions relies on the DS modelling of the tight coupling of parsing and production (‘mirroring’). Parsing in DS incorporates aspects of

   



production, i.e., the parser does not passively expect the scanning of input to produce structure; instead, the Context DAG is expanded at each word step with the generation of predictions (goals) in order to accommodate what will be encountered next. On the other hand, production in DS employs exactly the same mechanism and DAG as parsing, with the only difference that, instead of scanning input, every step of articulation is licensed by a richer tree, a so-called goal tree. The goal tree is a tree at least one step forward from the tree-under-construction, and a potential next processing step through the DAG is licensed by the tree-under-construction subsuming the goal tree. Subsumption means that the partial tree that is being developed must be extendible into that goal tree by following the licensed actions of the system (Purver and Kempson ). Put simply, whereas parsers have to follow what the speaker offers them, speakers have to have at least some partial idea of what they are going to be communicating at the next step. So given this incrementality and predictivity of processing, first, there is nothing to prevent speakers initially having only a partial structure to convey, i.e. the goal tree may be a  tree, perhaps only one step ahead from what is being voiced. Secondly, since the DS-grammar implements a notion of predictivity, if, at some stage during sub-sentential processing, an interlocutor has the ability to satisfy the projected goals via their own resources, e.g. via lexical access or by extending the current tree with a ed tree, it is perfectly sanctioned by the grammar for them to take over and continue extending the partial tree through DAG paths perhaps not foreseen by the original speaker (see e.g. () and ()). Moreover, since DS does not assume the mediation of any connected syntactic derivation in such cases, it is able to deal even with cases where, as we saw in (), repeated modified here as (), split utterances can take forms which would be ungrammatical under standard assumptions (*Did you burn myself?): ()

Mary: Did you burn Bob: myself? No.

Figure . displays the partial tree induced by processing Mary’s utterance Did you burn, which involves a substitution of the metavariable projected by you with the term standing for the current addressee, Bob. At this point, Bob can complete the utterance with the reflexive. This is because a reflexive, by definition, just copies a formula from a local co-argument node onto the current node, just in case that formula satisfies the person/ number conditions of the expression, in this case, that it designates the  speaker. Since sub-sentential/sub-propositional elements can perform various discourse functions (‘speech acts’), as seen in e.g. (), the DS analysis of split utterances applies equally to cases like (reprise) sluicing, reprise clarifications, corrections, etc. without assuming that such sub-sentential constituents need to be sententially/propositionally expanded (Kempson et al. ; Gargett et al. , ; Gregoromichelaki et al. ; Eshghi et al. ). For example, cases of elliptical clarification interaction, mid-sentence, as in (), only involve locally attached link structures, with their discourse function modelled only as effects on the DAG (see e.g. Eshghi et al. ): ()

A: I had to go back to the hospital for a follow-up appointment. The doctor B: Chorlton? A: Mhm. He said I had a shadow on my lungs.



, , ,   Mary: Did you burn ?Ty(t), Q ?Ty(es → t)

Si

?Ty(e), Ty(e), U, ?∃xFo(x), Bob’

?Ty(e → (es → t)) Ty(e → (e → (es → t))), Burn′

?Ty(e),

Bob: myself? ?Ty(t), Q ?Ty(es → t)

Si

?Ty(e → (es → t))

Ty(e), Bob’ Ty(e), Bob’ ,

Ty(e → (e → (es → t))), Burn′

 . Incremental development of Mary’s/Bob’s context via processing words

Ty(e), Chorlton′(U)

?Ty(t)

The Doctor Ty(e), (Doctor′(U) ∧ Chorlton′(U))

 . Processing Chorlton? in ‘A: the doctor B: Chorlton?’

Such clarifications are construed relative to whatever constituent immediately precedes them, as in (), where what is pertinent to the sub-sentential interruption is the immediately preceding DP the doctor. So such utterances are analysable in exactly the same terms as split utterances more generally, where the fragment is taken to directly extend its antecedent structure in context, here just the content derived from the doctor through the independently motivated link adjunction. Figure . shows the result of processing (generating for B, parsing for A) of the utterance Chorlton.7

7

Simplifying here, with definites taken to contribute anaphoric content needing resolution, the node marked with the box, ‘the doctor’, shows the incorporation of additional information identifying the individual sought to replace the metavariable U, namely, that he is a doctor and named Chorlton, this being the result of the conjunction of the formula annotating the root node of the linked tree, with that of that node—the operation associated standardly with evaluation of linked trees (see §..).

   



.. Licensing the splitting of dependencies across turns Having given a unified account of linguistic processing as mappings from context to context, making redundant extra machinery of idiosyncratic constraints applying to structure over strings, it becomes evident that the phenomena ‘anaphora’ and ‘ellipsis’ are really part of the wider phenomenon of incremental licensing of NL strings and interpretations. We therefore expect them to display parallel properties, and this is in fact what we find. We first note the parallel potential between pronouns and ellipsis for anticipatory forward-looking resolution (‘cataphora’). In English, only some pronouns license this localized anticipatory function, with the clausal sequence following the predicate used to provide the value for the anticipatory expletive pronoun (for details, see Cann et al. ).8 Given our modelling of split utterances in terms of DAG options constraining the next processing actions of both interlocutors, as we would expect, the requirement of resolution of such an anticipatory device can be taken over by another party to formulate according to their own context or to serve their own purposes (see also ()): ()

A: Even if you could . . . B: I wouldn’t help you, that’s right.

()

A: It is obvious . . . B: that you are wrong.

Further, as we show above (e.g. ()), there is the parallelism between pronominal anaphora and VP-ellipsis in that they are both resolvable, indexically, from the utterance situation. So ellipsis, like pronominal anaphora, is resolvable either through backwards search recovering content or actions from the Context DAG, or by just imposing a constraint for its resolution and restricting the available DAG paths for both interlocutors to those providing content for its resolution, or, indexically, by introducing in the DAG aspects of the utterance situation. Now, a generalization missed by other frameworks (e.g. Kobele b) but revealed through this modelling is that this threefold parallelism extends to ‘syntactic’ mechanisms, for example, those underpinning long-distance dependencies. First, in parallel to cases ()–(), the method of initially building an unfixed node and resolving its position later is the standard mechanism used in DS for long-distance dependencies. But the structure that finally resolves the structural underdetermination of such unfixed nodes can equally be provided by backwards DAG search and reiterating actions from context. For example, Mary in () is uttered in a context where actions are available for repeated application: ()

8

A: Sue, John upset. B: Mary too.

Many languages display anticipatory as well as anaphoric uses, commonly called Pronoun Doubling (Gregoromichelaki ).



, , ,  

Here, triggered by the presence of too, Mary too is interpreted as ‘John upset Mary’ via a sequence of actions in which Mary’ is taken to annotate an unfixed node, later unified to become the internal argument of Upset’, exactly following actions used in interpreting the preceding string. And given that DS parsing and production mirror each other, resolution goals imposed on the DAG by one interlocutor both constrain and are resolvable by the other: ()

A: The books, I’m told that we needn’t insure. B: The Assyrian horse . . . A: we definitely must insure.

Completing the parallelism is the resolution of an unfixed node indexically. Of these, striking instances are children’s one-word utterances in early language acquisition, where they rely on their interlocutor to provide some open structure into which their presented fragment can be incorporated, as in () repeated here as (): ()

A (pointing to an empty mooring site on the canal): Daddy B: Yes, you were here with Daddy yesterday, clearing out the boat. That’s right dear.

Thus, from this perspective, use of fragments in split utterances (see e.g. ()–()) isn’t merely an optional economy measure, it is an essential vehicle for coordination, which is crucially exploited during both language acquisition and in day-to-day human interaction. Moreover, this prediction of a parallel dynamics between pronominal anaphora, ellipsis, and long-distance dependency resolution, inexpressible in other frameworks, provides meta-theoretical foundations supporting not merely the dynamic perspective on ‘syntax’ proposed, but also the potential it provides for context-grounded interaction.

.. Licensing repair devices: Self-repair and corrections But even further than this, a unified account of linguistic processing as mappings from context to context, aiming to model action coordination during interaction, can accommodate, in the same terms, the phenomena that arguably overtly promote coordination, namely, repair (see e.g. Schegloff ), which has traditionally been rendered a performance phenomenon unworthy of interest for theoretical linguistics. For example, self-repair, a pervasive phenomenon in dialogue (Shriberg ), is naturally accounted for given the DS notion of the DAG context and content recoverability employing backwards search. The same backwards-search mechanism described above for various other ellipsis cases applies again in repair cases; however, these are triggered by the current processing state being abandoned and processing resuming again from appropriate prior points in the context (see Hough  and Hough and Purver  for details). This strategy models common types of self-repair phenomena such as short stuttered repeats (I, I go to Paris) and substitutions (John likes, uh, loves Mary), but it can also deal with more complicated licensing phenomena where, as predicted, ellipsis interacts

    ‘Paris’

‘to’ < to > S0

‘on’

‘Monday’

‘sorry’

< London > S1

S5

repair



‘London’

< on > < Monday > S6 S7

regeneration < Paris >

S2

S3

S4

< on > < Monday >  . Incremental interpretation of self-repair by replaying DS actions in the Context DAG

with the repairing of structure. In the following example, ellipsis reconstruction must operate across an interruption point + in a repair:9 ()

Peter went to [Paris on Monday, + {sorry,} London]

This is a case where the content of London repairs the content derived from Paris on Monday. However, under one interpretation, the speaker continues to describe the event as one occurring ‘on Monday’, and this requires elliptical resolution after abandoning the content of Paris and replacing it with the content of London. To model this, the action sequences triggered by on Monday must be replayed after the substitution of one argument content for another. Schematically, the incremental steps leading to this can be seen in Figure ., where the operation of Regeneration applies after processing has resumed from the point just before the DAG options including Paris are abandoned (characterized as state S in the diagram), finally arriving at the revised interpretation at S. This same backtracking with Regeneration mechanism naturally extends to othercorrections, where the repair is not of one’s own utterance but another’s. It is the fact that actions are first-class citizens in the DS context that allows this straightforward integration of parsing and generation to enable the direct reflection of incremental dialogue-level coordination across interlocutors.10 Nevertheless, as we will see immediately below, such an action-based framework remains capable of expressing even presumed arbitrary formal ‘syntactic’ constraints as those applying to restrict the licensing of various forms of ellipsis, for example, antecedentcontained ellipsis as in ().

9

The bracketing and symbolization follows Shriberg (), where the square-bracketed material to the right of the interruption point + repairs the material to the left of it. 10 For a more detailed account of contextual updates as a result of self-repairs, corrections, clarification interaction and their responses, and other forms of feedback in dialogue, see Eshghi et al. ().



, , ,  

.. Licensing in local processing domains: Island restrictions DS takes the grammar to be a model of the joint processing actions of speakers and hearers. From this perspective, standard syntactic constraints are reconceptualized as constraints on processing DAG options, expressed in the same vocabulary, and, as we saw, expected to interact with phenomena like pronominal anaphora and ellipsis. We’ve already seen that standard, presumed sui generis linguistic constraints, like island restrictions as displayed by, for example, relative clauses, can be accounted for through the assumption of selfcontained processing domains (see §..). Processing a relative clause, as we saw in §.., is taken to involve temporary shifting to a newly introduced tree structure in order to provide the context deemed necessary for proceeding with the processing of the main tree. For this reason, while processing goes on in the new domain, the fixing of an unfixed node initiated in the main tree has to await shifting of the processor back in the main tree (see Kempson et al. ; Cann et al. ). With this perspective on island constraints, the restrictions on antecedent-contained ellipsis emerge unproblematically: ()

John interviewed every student [who Bill already had].

()

*John interviewed every student [ who Bill ignored a teacher [who already had] ].

Simplifying for reasons of space here, in these cases, the DP is minimally made up of a determiner (every), a nominal (student), and a relative pronoun (who) initiating a relative clause which contains the ellipsis site (had). This relative clause is expected to provide an added restrictor to the variable bound inside the epsilon-term t,x,Student0 (x) which is the content contributed by the DP (see Figure .).11 The DP word sequence has to be processed under the usual principles governing the processing of relative clauses. First, processing the determiner phrase every student involves constructing an abstract λy.t,y,Student0 (y) that will result in binding a variable x introduced by the noun student. Second, a ed tree is constructed from the node occupied by x with the requirement to include this variable as one of the arguments of this new ed tree in order to furnish it with further restrictions. Because relative pronouns in English appear as ‘left-dislocated’ elements, an unfixed node is introduced initiating the building of this linked tree in order to process the relative pronoun, who. Processing the relative pronoun, through anaphoric means, duly annotates this unfixed node with a second copy of the variable x. It is then the underspecified domination relation associated with the unfixed node, (h↑*iTn(a)), which determines the locality of that initially unfixed node as being within the domain of a single tree. As a result, this then precludes the possibility that this unfixed node could be unified as the argument of a further ed tree as, by definition, there are no dominance relations holding across linked structures (Figure .): DS employs the epsilon-calculus (Hilbert and Bernays ) with arbitrary names (epsilon-terms) as the content derived from quantified DPs. 11

   



?Ty(t) ?Ty(es → t)

Spast

?Ty(e → (es → t))

John′

?Ty(e) x

Interview′

Tn(a), ?Ty(t) ⟨↑*⟩Tn(a), Spast x

?Ty(es → t)

λy.[τ, y, Student(y)]

V, Ty(e → (es → t))

Bill′

 . Successful processing of John interviewed every student who Bill had ?Ty(t) ?Ty(es → t)

Spast John′

Tn(a), ?Ty(t)

?Ty(e → (es → t)) ?Ty(e) x

Interview′

λy.[τ, y, Student(y)]

(↑*)Tn(a), Spast x

?Ty(es → t) ?Ty(e → (es → t))

Bill′

Ty(e)

Tn(b), ?Ty(t) y

?Ty(es → t)

Spast y

Ignored′

λz[ , z, Teacher′z]

V, Ty(e → (es → t)) ?Ty(e)

Interview′

 . Ungrammaticality of () as impossibility to unify unfixed node with object of interview in second relative clause

Now, in coming to resolve the metavariable V which the ellipsis site had has contributed inside the relative clause, a sequence of actions from the context has to be retrieved that will result in a sub-tree of Ty(e!(es! t)). But now the choice of which sequence to select is constrained: the selected sequence that will resolve the ellipsis has to conform to the already mentioned independent restriction on unfixed nodes imposed on the partial tree already constructed from the relative pronoun. Hence the variable x can only appear in the local tree and cannot cross further to another ed one. This explains the island sensitivity yielding ungrammaticality in () where this constraint cannot be satisfied (Figure .).



, , ,  

Notice the significance of this result. In other frameworks, island constraints would be articulated within the component of syntax, independent of any interpretation/processing actions, hence not expected to interact with ellipsis/anaphora construal. In DS, however, with syntax defined in terms of update of the DAG content, such restrictions, being constraints on local processing domains, are directly predicted also to constrain ellipsis. This is because ellipsis is also modelled as a process of contextual development, therefore required to conform to any restrictions applying independently to such processes.

. R: S    

.................................................................................................................................. Stepping back from the details, we can see how the present account of ellipsis compares to standard accounts. Like earlier accounts, the present one also addresses the issues of the nature of the ellipsis site itself, the recovery of its content, and licensing limits on that recoverability. However, the main explanatory tool for addressing standard cases of ellipsis here does not provide descriptions of decontextualized syntax–semantics mappings (derivations, e.g. Kobele ). Instead, DS models directly the parser/generator actions involved in processing the skeletal information projected by some ellipsis sites. Such actions, as in routine anaphora resolution cases, include the ability to introduce, retrieve, and reuse contextual resources. Replacing a competence model embodying propositional knowledge of sentence–proposition mappings, we take grammars instead to model the know-how of interpreting and producing language in context. As such, the grammar in general, and syntax in particular, do not involve sui generis devices or representations; instead, domain-general mechanisms like underspecification and update are the core notions. As in the general domains of perception and action, the conceptualizations afforded by linguistic elements proceed incrementally and predictively, and since participants’ actions in an interaction form each other’s context, actions and conceptualizations can interleave in jointly achieving coordination across sentence boundaries and across turns. From such a perspective, participants do not need to formulate hypotheses as to each other’s intentions, since the goal-directed subpersonal mechanisms of the grammar are adequate for enabling low-level coordination without having to explicitly derive propositional descriptions of each other’s actions.

A This chapter is grounded in work done on the ESRC project The Dynamics of Conversational Dialogue (DynDial) (ESRC-RES---), in collaboration with Patrick G. T. Healey, Matthew Purver, Ronnie Cann, Christine Howes, and Wilfried Meyer-Viol, who have each played a central role in the project. We are extremely grateful to them for their input and to many others who have contributed to the way Dynamic Syntax has evolved, in particular, Stergios Chatzikyriakidis, Miriam Bouzouita, Jieun Kiaer, and others such as Graham White and Dov Gabbay, who have supported this work in more general ways. Eshghi’s research in this paper was supported by the EPSRC, under grant number EP/MX/ (BABBLE project) and Hough’s by the DFG-funded Cluster of Excellence, Cognitive Interaction Technology ‘CITEC’ (EXC ) at Bielefeld University, and the DFG-funded DUEL project (grant SCHL /-).

  ......................................................................................................................

    ......................................................................................................................

 

. I

.................................................................................................................................. R years have seen a proliferation of approaches to ellipsis which make crucial reference to the semantic interpretations of ellipsis sites and their antecedents in various ways. At the same time, recent decades have witnessed a sea change within the field of semantics, with many researchers treating sentence meanings not in terms of mere truth conditions, but rather in terms of a broader notion of Context Change Potential (CCP) or Information Exchange Potential. It seems natural, then, to ask the question of how this broader notion of semantic content can be brought to bear in the analysis of ellipsis. In this chapter, we engage this question by focusing on one particular branch of semantic theories with this broader conception of meaning: inquisitive semantics (Groenendijk ; Groenendijk and Roelofsen ; AnderBois a inter alia; see Ciardelli et al.  for a recent overview). Inquisitive semantics holds that sentence meanings for both declarative and interrogative sentences consist of (or at least determine) sets of alternative propositions. For the study of ellipsis, then, the hypothesis is that semantic conditions on certain ellipsis processes will (or at least may) make reference to this broader, alternative-rich notion of semantic content rather than to mere truth conditions. The outline for this chapter is as follows: §. introduces inquisitive semantics; §. briefly presents the most fully fleshed-out account of an ellipsis process using inquisitive semantics: AnderBois (); §. concludes by addressing the questions of S, R, and L from the perspective of the account in §. and discusses how they might differ under other possible ways of incorporating inquisitive semantics into a theory of ellipsis.





. I 

..................................................................................................................................

.. What makes a semantics ‘inquisitive’? The core intuition behind inquisitive semantics is that the meaning/CCP of sentences not only includes truth-conditional information, but also includes the issue(s) that it raises, i.e., its inquisitive content. This has been long recognized, of course, for the CCPs of interrogative sentences. Inquisitive semantics extends this idea to capture the intuition that assertions, especially those containing widest scope disjunctions and indefinites, also raise issues in discourse. For example, () introduces two alternatives—‘that it will rain’, ‘that it will snow’—and thereby makes this issue salient in the output context in a way that truthconditionally equivalent sentences such as ‘it will precipitate’ do not (assuming that rain and snow are the only forms of precipitation). ()

(Either) it will rain or it will snow.

Inquisitive semantics therefore builds on a number of recent works on disjunction and indefinites (e.g., Kratzer and Shimoyama ; Simons ; Alonso-Ovalle ) in what has been called ‘Hamblin’ or ‘Alternative’ semantics. These works hold that indefinites and disjunctions introduce alternatives into semantic composition. While the name ‘inquisitive semantics’ refers to a family of related semantic/pragmatic theories (see, e.g., Groenendijk  for discussion), there are two fairly consistent ways in which inquisitive semantics differs from Hamblin semantics. First, the two differ in the ways in which alternative-rich meanings are composed and what their formal properties are. These properties have no clear importance for the study of ellipsis, so we refer the reader to Ciardelli et al. () and references therein for detailed discussion. Second, alternative-evoking (or lack thereof) is treated as an aspect of the top-level meaning of assertions and questions alike and therefore of their contribution to discourse as well.1

.. Inquisitive semantics across sentence types We turn now to give a concrete version of inquisitive semantics for both questions and assertions. Given our present purposes, the presentation here is necessarily informal (see Groenendijk and Roelofsen  and AnderBois a for more detailed formal presentations). The core formal shift in inquisitive semantics is to treat sentence meanings not as sets of possible worlds (i.e., propositions), as is done classically, but rather as sets of sets of possible worlds (i.e., sets of propositions). This move itself has a precedent in Hamblin’s () semantics for interrogatives, but differs in that expressions other than questions will make 1

It should be noted that while this was a driving motivation in many early works in inquisitive semantics (e.g., Groenendijk ; Groenendijk and Roelofsen ; Ciardelli ), some more recent works have used the moniker ‘inquisitive semantics’, yet lack this second property (e.g., Farkas and Roelofsen, forthcoming, propose that declarative clauses include a closure operator eliminating this possibility).

   



use of these richer meanings. In particular, we assume that disjunctions, indefinites, and other forms of existential quantification also contribute alternatives into semantic composition. Following Groenendijk and Roelofsen (), we will call a sentence inquisitive if its interpretation contains more than one alternative. The idea, then, is that not only interrogatives, but also declarative sentences may be inquisitive in this sense. Furthermore, if covert existential quantification is also taken to be inquisitive (as we suggest in §..), the inquisitivity of declaratives will be quite regular. Taking () as an example, we illustrate this idea in informal set notation in (a) and graphically in the diagram (b). Diagrams like (b) provide a pictorial representation of the interpretation of the formula in a toy model. We assume a toy model containing four possible worlds (w00, w01, w10, w11), represented visually by the four named circles. The names here correspond to the truth values of two propositions, p and q. In our current example, then, p = ‘that it will rain’ and q = ‘that it will snow’. The boxes, then, represent alternatives in the interpretation of the sentence. () Classical disjunction: [[p ∨ q]]cl a. { ‘that it will precipitate’ }

Inquisitive disjunction: [[p ∨ q]]inq { ‘that it will rain’, ‘that it will snow’ }

b. 11

10

01

00

← Precipitation

Snow→

11

10

01

00

← Rain

The classical semantics for disjunction is not inquisitive since it produces only a single ‘alternative’, the proposition that it will precipitate. In contrast, an inquisitive semantics for disjunction produces the alternative-rich interpretation with distinct alternatives for the different forms of precipitation mentioned in the sentence. While these two interpretations differ in their inquisitive content, they contain the same informative content, i.e., determine the same truth conditions. It is the same set of possible worlds which appear in some alternative or other in the interpretation of the two formulas, just structured differently. Inquisitive semantics therefore allows us to distinguish truth-conditionally equivalent formulas on the basis of the alternatives they evoke. To continue with our toy example, then, we assign a sentence with a disjunction like () the alternative-rich interpretation on the right of (), while assigning () the different (yet truth-conditionally equivalent) semantic representation on the left of (). ()

It will precipitate.

Extending such a semantics to indefinites and other forms of existential quantification is fairly straightforward at an intuitive level.2 Whereas a disjunction specifies alternatives one 2

Formally, it has been argued that this extension is in fact somewhat more fraught in the case of models with non-finite domains. See Ciardelli () for detailed discussion.





by one, an indefinite produces a set containing one alternative per individual in the restrictor set. For example, a sentence like () will receive an interpretation with n alternatives, one per each of the n individuals in the interpretation of ‘student’.3 ()

Prof. Ramírez met with a student.

()

{ ‘that R met with Al’, ‘that R met with Bella’, ‘that R met with Chad’, . . . }

In this section, we have introduced the basic conception of sentence meaning in inquisitive semantics with a focus on the two main alternative-evoking elements: disjunction and existential quantifiers such as indefinites. Simple sentences containing these elements make salient a set of alternatives and simultaneously contribute the information that the world of evaluation lies within some alternative(s) in this set. In §.., we provide a semantics for questions which makes the parallel with indefinites and disjunction explicit, while also capturing the difference between the two classes. Before doing so, however, we introduce a class of operators which interact with the inquisitive component of formulas to which they apply.

.. Negation and other operators Thus far, we have given an informal introduction to an inquisitive semantics for disjunction and existential quantification. This semantics holds that sentences containing these elements make salient in the discourse a set of alternatives in a way that truth-conditionally equivalent sentences may not. While this intuition seems fairly clear for the simple sentences we have looked at thus far, it turns out only to hold for sentences which contain wide-scope disjunctions and existential quantifiers. For example, a disjunction within the scope of negation, as in (a), seems to be no more inquisitive than its non-disjunctive counterpart, (b). ()

a. It’s not the case that it rained or snowed. b. It’s not the case that it precipitated.

Moreover, this behavior in fact follows from the way negation is naturally defined in inquisitive semantics (see Roelofsen  and references therein for detailed discussion of its mathematical foundations). Since sentence meanings are sets of alternatives, negation rejects each of these alternatives as in ((), middle), returning the maximal alternative which does not overlap with any of these. One important consequence of this is that, while it preserves truth conditions, double negation is no longer semantically vacuous since it eliminates the alternative-rich structure of the formula to which it applies, ((), right).

3

While in principle they are equally applicable, pictorial representations like (b) become unhelpful for larger sets of alternatives.

    ()

⟦∃x.φ(x)⟧

⟦¬∃x.φ(x)⟧



⟦¬¬∃x.φ(x)⟧

11

10

11

10

11

10

01

00

01

00

01

00

It follows, then, that not just any sentence containing an inquisitive element will be inquisitive, but rather those where the inquisitive element takes widest scope. We have illustrated this here for negation, but other operators may also have this property, such as the C operator found in appositive relative clauses. This fact is parallel to the observation (e.g., Chung et al. ; Romero ) that inner antecedents for sluicing must take wide scope (indeed, licensing of sluicing is often used as a diagnostic for scope of indefinites).4

.. Questions While inquisitive semantics assigns a more question-like semantics to sentences with widescope disjunctions and existential quantifiers, we still need to distinguish these latter elements from questions. The basic approach in inquisitive semantics—at least in matrix clauses—is to differentiate the two in terms of their informative potential. For ease of exposition, we will simply use the term ‘informativity’, though in all cases the relevant notion is possible informativity rather then actual (see AnderBois a for detailed discussion). The inquisitive semantics for disjunction in the right side of () not only introduces a set of alternatives, it also includes the truth-conditional information that some alternative or other holds. That is to say that a declarative with a wide-scope disjunction rules out the possibility that none of the alternatives holds. Intuitively, and perhaps definitionally so, questions are not possibly informative in this way. The literature on inquisitive semantics has seen two ways to cash out this insight, which we can call absolute and relative uninformativity. We can illustate these two approaches for the wh-question in (), recalling that we keep the pictures to two positive alternatives. The absolute uninformativity approach is exemplified by Groenendijk and Roelofsen (), who propose a Q(uestion) operator which adds in the ‘no-one’ alternative, as in the left picture of (). The second option, due to AnderBois (a), is to claim that questions have an existential presupposition and that the alternative set of the

4

While other scope-taking elements such as conjunction and universal quantification are defined by most authors in ways that allow their alternatives to be ‘passed up’ the composition, one could alternatively define them in ways that do not have this effect. The choice ultimately depends on how one wishes to handle scope-taking more generally and we therefore set aside this concern here.





question is uninformative only relative to this presupposition. We can indicate this pictorially by shading out the worlds presupposed not to be live options (just world  in this case): ()

Who told John about the party?

()

Two different inquisitive semantics for questions:

11

10

11

10

01

00

01

00

[Absolute uninformativity]

[Relative uninformativity]

Here, we will follow AnderBois () in adopting this latter option. Ultimately, however, the decision between these two approaches is an empirical one, resting largely on the longstanding question of whether questions contribute existential presuppositions. Beyond the unresolved nature of this question for English, it is of course possible that both options are needed across languages or across question constructions within a single language (e.g., argument vs adjunct wh-questions, wh- vs alternative questions).

. S     

.................................................................................................................................. Having introduced inquisitive semantics, we turn now to apply it to the analysis of ellipsis and, in particular, AnderBois’s (, ) account of both merger and sprouting subtypes of sluicing. Finally, §. will conclude by exploring other possible ways of incorporating the core insights of inquisitive semantics into a theory of ellipsis, drawing on related theories such as dynamic semantics and QUD approaches to discourse.

.. The need to move beyond truth conditions As a theory of the semantic content of questions and assertions, inquisitive semantics is only of direct relevance to ellipsis to the extent that semantics itself is (or pragmatics which is sensitive to semantics). In principle, it would be consistent for inquisitive semantics to provide an appropriate theory of semantic content, yet for ellipsis to be resolved in a purely syntactic (or LF-syntactic) way (as in, e.g., Sag a; Chung et al. ). However, there is a large body of work across many different frameworks arguing that semantics/pragmatics

   



do play a crucial role in ellipsis phenomena (Sag and Hankamer ; Hardt ; Ginzburg and Sag ; Merchant ; Culicover and Jackendoff ; Chung et al. , among many others). Assuming that this is right in some way, inquisitive semantics naturally raises the question of whether this condition will be sensitive not only to truth-conditional information, but also to the inquisitive aspect of semantic content. There are several different kinds of data which have been argued to support the need for the semantic condition on sluicing to be sensitive to inquisitive content. Except where noted, we focus on data from English, though there is no reason to expect the observations we make to not be more general. In this section, we focus on data from the subtype of sluicing which Chung et al. () dub ‘merger’, i.e., cases where there is an overt ‘inner antecedent’ in the A(ntecedent) clause corresponding to the wh-phrase in the E(lided)clause. In §.., we turn to cases of Chung et al.’s () ‘sprouting’, i.e., cases where there is no overt inner antecedent, with the wh-phrase instead corresponding to an implicit argument or adjunct in the A-clause. Perhaps the most fundamental observation supporting the relevance of inquisitive semantics for sluicing is the role played by inquisitive elements as inner antecedents. We see this clearly in the contrasts between the felicitous sluices in (a), whose antecedents are inquisitive sentences, and the infelicitious ones in (b), whose antecedents are noninquisitive. Notice that in both cases the corresponding full-clausal versions are felicitous. ()

a.

She said she had spoken to {someone/a student/John or Bill}, but Harry didn’t know who.

b. #She said she had spoken to {everybody/most students/the student/him/John and Bill}, but Harry didn’t know who. In addition to this basic observation, it is well-known that these elements must take wide scope in order to serve as inner antecedents (e.g., Chung et al. ; Romero ; Barker ). As we will see in a moment, this generalization follows straightforwardly from an inquisitive semantic approach to sluicing. In fact, from the inquisitive semantic perspective, the ability of disjunctions and indefinites to serve as inner antecedents is simply another manifestation of the interrogative-indefinite-disjunction affinity (e.g., Haspelmath ; Bhat ; Haida ). While the privileged role of disjunctions and indefinites is of course quite suggestive, this alone leaves somewhat open the possibility that it is some other aspect of these expressions which is crucial. For example, focusing primarily on indefinites, Chung et al. () argue that it is the logical form of these elements (i.e., the fact that they contribute a variable in the Heimian view, Heim ) which is crucial. Merchant () argues that it is the truth conditions following existential closure of the A- and E-clauses which is relevant. One kind of data which are problematic for the former view at least (as Chung et al.  note) are disjunctive inner antecedents where the disjunction is not of arguments, as was the case in (a), but of entire clauses, as in (). ()

a. (Either) Freddie is baking a cake again or something is on fire, but I can’t tell which (one). (AnderBois ) b. Russ is in the back or Ali is working alone, but I can’t tell which (one). (AnderBois )





As for the latter view, there are several cases of expressions which are truth-conditionally equivalent to overt widest-scope indefinites, yet do not license sluicing (i.e., cannot serve as an inner antecedent). Such cases are unexpected if truth conditions are all that the semantic condition cares about, but entirely expected from the view of inquisitive semantics since they plausibly have different inquisitive content. Perhaps the most straightforward case where truth conditions alone prove inadequate are examples like () where we see that indefinites and NPIs with double negation and negative quantifiers with single negation (in Standard American English) fail to license sluicing. The would-be A-clauses have counterparts with no negation (e.g., ‘Someone left.’ for (b)) which are true in the same circumstances but differ in their ability to license sluicing. ()

a. #It’s not the case that Bill didn’t bring a dish, but I don’t know which (one). (AnderBois ) b. A: It’s not the case that no one left. B: #Who? c. #It’s not true that Bill didn’t talk to anyone, and Jane just asked me who.

The second case discussed by AnderBois () are indefinites that occur inside appositive (non-restrictive) relative clauses. In contrast to restrictive relative clauses, the content of appositive relative clauses is generally thought of as having sentence-level scope (or perhaps as being in some sense ‘scopeless’), but having no other truth-conditional impact. Therefore, a truth-conditional account predicts that inquisitive elements inside of appositive relative clauses ought to readily serve as inner antecedents for sluicing. As seen by the infelicity of (), based on examples in AnderBois (), this prediction is not borne out. ()

a. #The valiant knight, who defeated {a masked enemy/someone} in a duel, still wonders who. b. #Joe, who once killed {a man/someone} in cold blood, doesn’t even remember who.

More recently, this generalization has been investigated experimentally by Collins et al. () who argue that such examples can be improved by (i) using a wh-phrase of the form which + NP in the E-clause rather than a bare wh-word like who, and (ii) making the issue raised by the E-clause (or a related one) salient in the discourse preceding the target sentence. Given the space limitations of the present work, we will leave a detailed discussion of these issues to future work.5 However, it is worth noting that both of these manipulations are ones which plausibly raise the salience of the E-issue in the 5

We would also note that it is not entirely clear how to interpret felicity judgments for these data. Reading these sentences, it is relatively easy to figure out what such examples were supposed to have meant after the fact. However, it still seems somewhat unlikely in our opinion that speakers in fact produce such sentences frequently and possible that they have difficulty processing them when encountering them in natural speech. Therefore, it seems quite possible to imagine an analysis of this gradient pattern of judgments in which the grammar of appositives and sluicing does not generate such sentences, but rich context and other ‘repair mechanisms’ allow speakers to figure out what they are to

   



ambient discourse and therefore in our view should not be seen as evidence that the indefinite inside the appositive can in fact serve as the inner antecedent, so much as casting doubt on the logically prior question of whether a linguistic antecedent is strictly necessary in the first place (i.e., whether sluicing is an instance of ‘surface’ or ‘deep’ anaphora in the terms of Hankamer and Sag ), as the authors point out. At the same time, the existence of contrasts like () make clear that inquisitive elements at least can play a privileged role in licensing sluicing, one which is unexpected for Ginzburg and Sag (), as discussed in §.. As an anonymous reviewer notes, there is one aspect of Merchant’s () account that one might expect could help capture such data: the requirement that the antecedent be salient. While salience is often not fleshed out in much detail, it seems a priori plausible that double negation and apposition reduce the salience of the antecedent and that this is the reason they impede sluicing. However, we find that other elements typically thought to be sensitive to salience such as pronouns and even VP-ellipsis can readily find an antecedent in these environments as in (). Therefore, it seems that the effects we are seeing cannot be straightforwardly attributed to salience of the sort relevant for anaphoric processes more generally. ()

a. b.

It is not true that John didn’t bring an umbrella. It was purple and it stood in the hallway. (Krahmer and Muskens ) John, who helps people if they want him to, kisses them even if they don’t want him to help them. (AnderBois et al. )

A third case where truth-conditional equivalence proves insufficient to license sluicing comprises certain cases of bare noun incorporation, such as those discussed in detail by Collins () for Samoan. Collins observes that, unlike other truth-conditionally equivalent indefinite-like expressions in the language, incorporated bare NPs do not license sluicing (nor do they license pronominal anaphora) as seen in (). Collins () pursues an analysis in the closely related framework of dynamic semantics, though clearly this could potentially be treated as a difference in inquisitivity in the current context. ()

a. Sā fa‘apāgotā-foamea  arrest-thief ‘The chief thief-arrested.’

le the

ta‘ita‘i leader

b. #‘o ai? who Intended: ‘Who?’ Thus far, we have seen several cases in which sentences with identical truth conditions to those provided by overt indefinites are unable to license sluicing. We turn now to one further kind of support for the relevance of ‘issues’ to sluicing: the fine-grained patterns have meant when encountered in experimental settings. In any case, further corpus and experimental work is needed in this area.





of variation across different types of nouns and wh-words investigated by Barros () (see also Dayal and Schwarzschild  and Barker  for related observations). Barros observes that the felicity of sluicing varies depending on complex interactions between the nominal content of the inner antecedent and the properties of the wh-remnant.6 For nouns in the A-clause, Barros claims that the felicity of such sentences depends on whether the noun is a ‘basic-level’ noun or not (in the sense of Brown , Cruse , and others). Basic-level nouns are nouns that have a privileged status tied to their encoding a ‘neutral’ level of specificity (i.e., presumably for non-linguistic reasons). For example, out of the blue it would sound more natural to talk about my ‘cat’ than my ‘mammal’ even though clearly both are equally truthful descriptors for a cat.7 ()

a. #Jack has a profiterole, but Fred doesn’t know what (exactly). b. #Jack ordered an éclair, but Fred doesn’t know what (exactly). c. #Jack got a cat, but Fred doesn’t know what (exactly).

()

a. Jack had a drink, but Sally can’t recall what (exactly). b. Jack ordered food, but Sally doesn’t know what (exactly). c. Jack ordered an appetizer, but Sally can’t recall what (exactly).

Beyond variations based on the specificity of the noun itself, Barros argues that the animate wh-word who differs from the inanimate what in allowing more ‘specific’ nouns in the inner antecedent: ()

Sally met with a {scientist/geologist/seismologist}, but I don’t know who.

Glossing over important details, the basic idea Barros pursues is that wh-words lexically specify a particular level of specificity, that is, they specify an issue whose alternatives have a particular level of granularity. For sluicing to be felicitous, then, the descriptive content of the noun or other inner antecedent material must not be more specific than the level specified by the wh-word. Who and what differ in the level of specificity they specify, leading to the asymmetry seen above. Barker () makes similar observations and proposes the generalization in (): () The Answer Ban: the antecedent clause must not resolve, or even partially resolve, the issue raised by the sluiced interrogative. We return to these observations in more detail in a moment once we have presented AnderBois’s () account. For now, let us remark that these observations fit naturally in a view where semantic content is alternative-rich.

6

Barros () also notes a further interaction with the presence or absence of exactly in the E-clause which we will set aside here. 7 See Barros () and references therein for further discussion and independent linguistic diagnostics of these categories.

   



.. An account based on symmetric inquisitive entailment Inquisitive semantics is a theory of semantic content, and as such can be implemented within a variety of different theoretical approaches to ellipsis. In this section, we present the most worked-out inquisitive semantic account of an ellipsis process, AnderBois’s () account of sluicing (AnderBois ,  present earlier versions of more or less the same approach). AnderBois’s () approach builds on the approach of Merchant (), but incorporates the inquisitive semantic conception of semantic content. Given this, we very briefly review the major features of Merchant’s () account. Merchant’s () theory of ellipsis assumes, along with many other authors, that pronounced material in the E-clause (i.e., the wh-remnant), arises from a fully articulated clausal version, as in (). From this starting point, an additional mechanism specifies the non-pronunciation of the redundant material in this clause, indicated in strikethrough, an approach commonly known as PF-deletion. ()

[John talked to someone]A, but I don’t know [whoi John talked to ti]E.

For Merchant, this PF-deletion operation is subject to the condition in (), which ensures that there is a semantically identical antecedent salient in the surrounding context.8 The existential type-shifting portion of the definition existentially quantifies over missing arguments in order to be able to apply the definition to ellipsis processes which operate over parts of clauses (e.g., existentially quantifying over the subject in order to compute entailment between verb phrases for VP-ellipsis). ()

Merchant’s () e-Givenness condition: An IP α can be deleted only if α is e-G.

()

F-closure: The F-closure of α, written F-clo(α), is the result of replacing F-marked parts of α with ∃-bound variables of the appropriate type.9

8

This condition itself builds on the one proposed by Schwarzschild () for deaccenting. It is worth noting therefore that, as observed by AnderBois (), deaccenting appears to be sensitive only to truthconditions rather than inquisitive content as well. Consider, for example, the contrast between () and the following (underlining indicates deaccenting): (i) It’s not the case that Bill didn’t donate a book to the library, but I don’t know which book he donated. 9 The focus-closure (F-clo) part of the definition is needed primarily to handle two kinds of sluices which we will not discuss here: ones where the wh-phrase contains else, as in (i), and so-called ‘contrast’ sluices like (ii), both examples from Merchant ().

(i) Abby called BenF an idiot, but I don’t know who else. (ii) She has five catsF, but I don’t know how many dogsF. We set aside these cases here, while not denying their importance.

 ()

 e-Givenness: An expression E counts as e-G iff E has a salient antecedent A and, modulo existential type-shifting, a. A entails F-clo(E), and b. E entails F-clo(A).

AnderBois () departs from this basic set-up in two ways, only the latter of which is directly related to inquisitive semantics. First, following Chung () and many subsequent works, the work adopts a ‘hybrid’ approach which supplements the semantic condition with a lexical identity condition in (). The numeration is the minimalist term for the list (technically a multiset) of lexical items that comprise the sentence. ()

No New Morphemes: Every lexical item in the numeration of the sluice that ends up (only) in the elided IP must be identical to an item in the numeration of the antecedent CP.

This lexical identity condition is primarily relevant for handling cases of sprouting (see §..), but is aimed at capturing the infelicity/ungrammaticality of examples like (), where the A- and E-clauses clearly have the same truth conditions, yet sluicing is not possible. While it is not impossible to imagine that certain such cases can be captured semantically (e.g., AnderBois  attempts such an account for (a)), we set aside this possibility here. It should be noted, however, that even accounts of ellipsis which do not posit silent linguistic material (e.g., accounts of ‘fragment answers’ by Ginzburg and Sag  and Jacobson b) similarly treat such data as arising from a minimal (morpho)syntactic condition regarding the syntactic category of the E-remnant. ()

a. #[The cake was eaten]A but we don’t know [who ate the cake]E. b. #[John is jealous]A, but I don’t know [who John is jealous of]E.

The second difference from Merchant () is that the symmetric entailment condition analogous to () is formulated with respect to inquisitive semantic contents, rather than just truth conditions. Groenendijk and Roelofsen () give the formal definition for entailment in (). To unpack this definition a bit, recall first that the interpretations of any given formula will be sets of alternatives. Given this, to see if ’ entails c, we check to see if each alternative in [[’]] is a subalternative of some alternative or other in [[c]]. Thinking in terms of the pictorial representations above, then, entailment checks to see if each box in the interpretation of ’ fits inside some box or other in the interpretation of c. Symmetric entailment then, means that the interpretations of two formulas have the same alternatives. ()

Entailment: ’ entails c iff 8α ∈ [[’]] is such that ∃β ∈ [[c]] such that α  β

One important thing to note about this definition is that it does not take into account the presuppositions of either formula, just the alternatives in the proposed output state.10 10 The idea that such a notion of entailment is relevant for natural language has been independently proposed in the literature on NPI-licensing (von Fintel  et seq.), where it has been dubbed ‘Strawson entailment’.

   



As we saw in §.., the existential presupposition is the only thing distinguishing the interpretation of an interrogative and a corresponding declarative with a wide-scope indefinite. This therefore allows for the condition on sluicing to be formulated as follows: ()

Symmetric Entailment Condition on sluicing: Given a structure

CPE C+Q

,

IPE

IPE can be elided only if there is some salient antecedent CPA such that: a. CPE entails CPA, and b. CPA entails CPE. This condition essentially adapts Merchant’s e-Givenness condition to the inquisitive setting (we set aside issues related to F-closure; see n. ). Given the deep semantic connection between interrogatives and indefinites/disjunctions in inquisitive semantics, however, it can be stated over the entire clause (CP), rather than just the deleted portion, IP. Since the E-clauses in sluicing are necessarily interrogative, the inquisitive entailment condition in () derives what AnderBois () calls the ‘inner antecedent generalization’ in (). ()

Inner antecedent generalization: An expression α can serve as an inner antecedent for sluicing only if α makes an inquisitive contribution.

Beyond this, as argued by Barros (), Barker’s ‘answer ban’, (), follows as a particular case of (). If the A-clause resolves (or partially resolves) the issue raised by the E-clause, the E-clause by definition does not entail the A-clause and therefore fails the condition in (). Accounting for all of the data of this sort does require a more fine-grained semantics for whwords than we will give here, so we again refer the reader to Barros () for further details.11 With this in place, we now show how the account tackles basic cases of sluicing as well as the infelicitous cases discussed in §... First, let’s consider a basic case where the inner antecedent is an overt indefinite ‘someone’. The inquisitive interpretation of the A-clause someone left will be a set of alternatives of the form ‘x left’, (, left). The interpretation of the E-clause, who left, consists of the same set of alternatives (, right). While the E-clause additionally includes a (non-inquisitive) existential presupposition, the entailment condition in () ignores this, and so, the symmetric entailment condition is met and () is correctly predicted to be felicitous. We can visually verify that () is met by looking at the 11 Barker (), it should be noted, does briefly address Barros’s () claim that the answer ban follows from the inquisitive entailment condition, but expresses skepticism for two reasons. First, Barker apparently was familiar only with AnderBois (), which only analyzed cases of merger, whereas AnderBois () and () also address sprouting (see §..). Second, Barker notes that Ciardelli et al. () propose an inquisitive-semantics-based account of epistemic possibility modals like might which do not license sluicing:

(i) #John might leave, but I don’t know which. However, while it is true that Ciardelli et al.’s () account of might is couched in a version of inquisitive semantics, their approach does not actually claim that ‘John might leave’ has the same semantics as ‘John will leave or John won’t leave’, and in fact does not claim that ‘John might leave’ is inquisitive at all (but instead defines a new category: ‘attentive’). Therefore, even taking this analysis of might at face value, we do not predict (i) to be well-formed.





diagrams in () and seeing that the alternatives on the two sides are the same. Disjunction behaves the same, differing only in that the specific nature of the antecedent obliges the use of the D-linked which in the E-clause (see Dayal and Schwarzschild ; Barros  for discussion of the conditions on which). ()

[Someone left]A, but I don’t know [who left]E.

()

a. ()A ⇝ ∃x.leave0 (x) b. ()E ⇝ ∃x.leave0 (x) (Presupposes: !∃x.leave0 (x))

()

⇔

[[()A]]

[[()E]]

11

10

11

10

01

00

01

00

Turning to the infelicitous cases, we look first at double negation. In (), we saw that double negation in inquisitive semantics preserves truth conditions, as one would expect, yet eliminates the inquisitive content of the formula to which it applies. The result is that an attempted sluice in () is interpreted as in the picture in (). Applying the inquisitive entailment condition, we find that the E-clause does entail the A-clause since each of the alternatives of the form ‘that x left’ is a subalternative of some alternative or other in the interpretation of the A-clause (namely, the single alternative ‘that someone or other left’). In the reverse direction, however, the entailment does not hold. Given the single alternative in the A-clause, we cannot find any super-alternative in the E-clause. Since the symmetric entailment condition fails, the account correctly predicts that sluicing will not be possible in this case. ()

#[It’s not the case that no one left]A, but I don’t know [who left]E.

()

⇏

[[A]]

[[E]]

11

10

11

10

01

00

01

00

   



Finally, let’s look at the case of the indefinite inside an appositive relative clause. A number of recent works have argued in one form or another that appositive relative clauses have a special discourse status of one sort or another (e.g., Potts ; Amaral et al. ; Simons et al. ; AnderBois et al. ). One aspect of this special status is that appositives represent purely informational updates which do not interact with the Questions Under Discussion (QUDs) in any direct way. AnderBois () proposes that in order to capture this aspect of their meaning, appositives ought to be treated as lacking the alternative-rich structure inquisitive semantics assumes for at-issue assertions, and instead be assigned a single classical proposition of type st. Compositionally, this is achieved through the C operator in (): ()

[[C(’)]] = {w | there is some α ∈ [[’]] s.t. w ∈ α }

Since inquisitive meanings are captured as non-singleton sets of alternatives, the C operator ensures that the formula to which it applies—as it enters the discourse record— will not be inquisitive regardless of its internal composition. As in the case of double negation, then, appositives deliver truth-conditionally equivalent interpretations, yet lack the alternative-rich meanings needed to meet the symmetric entailment condition. Summing up, we have seen in this section that an account of sluicing based on symmetric entailment defined over inquisitive semantic interpretations captures both the data which motivated Merchant’s () semantic approach (and indeed previous approaches dating back at least to Sag and Hankamer ) as well as a number of other sets of data, including several cases where sluicing fails despite truth-conditional equivalence.

.. Sprouting Having examined cases of ‘merger’ sluicing with overt indefinite or disjunctive inner antecedents, we turn now to cases of so-called ‘sprouting’ where the wh-phrase has no overt correlate in the A-clause. In some cases, of course, there is good reason to believe that, despite the lack of an overt inner antecedent, there is nonetheless an implicit argument present in the A-clause. For example, a large body of literature dating back to Fillmore () holds that apparently intransitive uses of verbs like eat in (a) include an existential/ indefinite implicit argument (see AnderBois b for recent discussion of sluicing and the typology of implicit arguments). Beyond this, there are cases like (b) where there is clearly an existential entailment, and arguably also (contextually restricted) existential quantification depending on one’s semantics for tense. ()

a. [Fred ate]A, but I don’t know [what Fred ate]E. b. [Fred baked a cake]A, but I don’t know [when Fred baked a cake]E.

Such data, therefore, have given rise to the claim that in a certain sense, there is no sprouting, but rather that there is always an implicit argument either syntactically (e.g., Fortin b, ) or semantically (Merchant ). While this approach is potentially viable for the data in (), Chung () points out that there are other cases where even an existential entailment





(let alone a true indefinite) is clearly not present, (). People can finish projects on their own, and Seth can arrive by car, bike, helicopter, etc. Such cases, therefore, present somewhat the opposite puzzle to what we have seen in §.. for merger. Here, the A-clause is not even truth-conditionally equivalent to the E-clause, yet sluicing is possible. () a. [He finished the project]A, but we don’t know [with whose help he finished the project]E. (Chung ) b. [Seth arrived]A, but I don’t know [on which bus Seth arrived]E. c. A: [Fred learned French]A. B: [For who did Fred learn French]E? For the examples in (), there is a clear intuition that these instances involve some kind of accommodation. For (a), this accommodation is pretty easy given the nature of projects. For (b), the ease of accommodation seemingly depends on various kinds of world knowledge: Is Seth someone who is likely to take the bus? Are there multiple buses which he could have taken? etc. While B’s question in (c) sounds fairly odd out of the blue, it sounds quite natural in the admittedly unusual context where it is known that Fred only learns languages to impress foreign visitors. Unconstrained, however, accommodation runs the risk of overgenerating and predicting that sprouting should be possible quite generally. However, there are many cases, as in (), where sprouting remains infelicitous even though it would seem quite plausible given world knowledge. () a. #Agnes wondered how John could eat, but it’s not clear what. b. A: The fact that Seth arrived was surprising. B: #On which bus?

(Chung et al. )

These sorts of restrictions on sprouting were first noted by Chung et al. (), who attribute them specifically to the presence of syntactic islands, claiming that in contrast to the well-known island insensitivity of sluicing more generally, sprouting is sensitive to islands. However, subsequent work by Romero () and Merchant () argues on the basis of pairs like () that this difference is not limited to islands (since the non-elliptical control in (b) is grammatical), and is therefore best captured by appealing to the independently observable narrow scope of implicit existential quantification. For example, whereas an indefinite and clausemate negation ordinarily give rise to a scope ambiguity, an existential implicit argument like the one in ‘Sally didn’t eat’ unambiguously takes narrow scope relative to negation (i.e., ‘Sally didn’t eat’ does not have a reading paraphraseable with ‘There is a thing/meal x such that John didn’t eat x’). () a. *Ramon is glad that Sally ate, but I don’t remember which dish. b. I don’t remember which dish he is glad that Sally ate.

(Romero )

For cases like these with an implicit argument, then, the merger account can be extended straightforwardly. Felicitous sprouting as in () is possible because the A-clause has an indefinite implicit argument and therefore has an interpretation which is inquisitive. In examples like (a) and (a), there is still an indefinite implicit argument, but one which cannot take widest scope. Therefore, the whole sentence’s interpretation is non-inquisitive and sluicing is correctly predicted to be infelicitous.

   



What, then, about examples like () in which we have seen following Chung () that no implicit argument or even existential entailment is present? Building on the above intuition that at least some cases of sprouting involve accommodation of some sort, AnderBois () proposes an account which is partially semantic and partially pragmatic. On the semantic side, the proposal extends the inquisitivity that we have thus far associated with overt indefinites and disjunctions to existential quantification quite generally, including covert quantification over neo-Davidsonian eventuality arguments. The meaning of a simple sentence like ‘John left’ not only includes the information that there is some event or other which is a leaving event and of which John is the agent, but also the issue of which event it is that satisfies these requirements. On the pragmatic side, then, the account claims that sluicing is felicitous to the extent that the alternatives in the interrogative E-clause covary with that of the A-clause, an accommodation process AnderBois dubs ‘issuebridging’. The rest of this section spells out both parts of this proposal a bit more, though we refer the reader to AnderBois () for further details. A central notion in inquisitive semantics is the idea that the kind of indeterminacy we find in indefinites and disjunctions is intimately related (and in some cases compositionally related) to the inquisitivity we find in questions. In both cases, a set of alternatives is made salient, leaving the issue of which alternative(s) in fact hold as at least a safe potential topic for future conversation. For example, the sentence in (a) introduces a set of alternatives in (c), makes salient the issue of which of these in fact hold, and conveys the information that at least one does. ()

a. Someone left. b. ∃x.leave0 (x)

John left Maribel left c.

Alexis left Ignacio left ...

Here, we extend this idea beyond overt indefinites to existential quantification over covert arguments, in this case, the neo-Davidsonian event(uality) argument. As is clear in (), the proposed semantics for the covert existential is formally entirely parallel to what we have seen in (). A sentence like (a) introduces a set of alternatives, (c), makes salient the issue of which of these alternatives in fact hold, and conveys the information that at least one does. ()

a. Seth arrived. b. ∃e.arriveʹ(e) ∧ Agent(S, e) e1 is an event of Seth arriving e2 is an event of Seth arriving c.

e3 is an event of Seth arriving e4 is an event of Seth arriving ...





The issue it makes salient is, however, a somewhat odd one, paraphraseable as ‘Which event is an event of Seth leaving?’. The apparent oddity of this issue, however, is due not to anything about the inquisitive quantification itself, but rather the ontological status of events in the first place (as discussed, for example, by Parsons ). Although it is standard to take events to be things in the actual world in more or less the same way that individuals are, it is far less intuitive to do so for events. With this semantics in place, we turn now to the pragmatic part of the story. For concreteness, we will work with the example in (b), repeated as (). The semantics above holds that the A-clause makes salient a quite fine-grained issue of the form ‘which event?’. The E-clause, on the other hand, makes salient a more coarse-grained issue about some aspect of the event in question, in this case its manner. The inquisitive entailment condition is therefore not met, given the difference in the granularity of these two issues.12 ()

[Seth arrived]A, but I don’t know [on which bus Seth arrived]E.

While these two issues are not identical, the claim is that they are sufficiently similar that the E-clause can be accommodated. AnderBois () calls this accommodation process ‘issuebridging’, on analogy with bridging definite descriptions like that in (). The existence of a driver is not simply accommodated directly, but rather by virtue of a salient relationship with something whose existence and discourse salience are already established, a bus. ()

A bus went by. The driver had on sunglasses.

Rather than bridging to an individual, however, indirect sprouting involves bridging to an issue introduced in the A-clause. Just as a driver is typically an aspect of a bus, times, locations, manners, etc. are typically aspects of events (see also Barros  for a related approach which works directly with these categories rather than events). Concretely, then, the prediction is that sprouting (and indeed sluicing more generally) should be subject to the condition in (). One important feature of the account to note is that it relies crucially on the presence of inquisitive material in the A-clause, and thus avoids overgenerating and allowing examples like () since the event quantification in question does not take wide scope.13 ()

Covariation condition: Sprouting is felicitous to the extent that the context allows for the inference that the alternatives in the A-clause covary with the alternatives in the E-clause.

This section has shown two ways to extend the inquisitive semantic account of sluicing to sprouting—one for cases where an indefinite implicit argument is present, and one for cases where no such argument is found. Both accounts rely on independently known scopal

12

In this particular example, there is a second apparent lack of symmetric entailment: the fact that the E-clause in this example has a presupposition not found in the A-clause. However, this concern is already taken care of given that we have adopted a notion of Strawson entailment. 13 The account raises a number of issues related to scope-taking which we cannot address here for lack of space. See Charlow () for recent discussion of scope-taking and sluicing in a closely related semantics.

   



properties of implicit existential quantification to help constrain the account, thus deriving the asymmetries between merger and sprouting first discussed by Chung et al. ().

. C: S, ,  

.................................................................................................................................. In this chapter, we have informally introduced inquisitive semantics, reviewed the most fleshed-out inquisitive semantic account of ellipsis to date—AnderBois’s () work on sluicing—and explored various kinds of data consistent with this view. We conclude here by considering how three major questions in the theory of ellipsis—S, R, and L—are answered under this account as well as briefly considering other potential ways of incorporating inquisitive semantics into the theory of ellipsis. By S, we mean the question of what syntactic structure, if any, is found within the ellipsis site itself. R refers to the way in which the ellipsis site’s interpretation is arrived at. L covers any additional constraints or conditions on ellipsis that are not clearly part of the latter two categories. Since inquisitive semantics is a theory of semantic content rather than of the interfaces between semantics and syntax or phonology, it in principle need not impose any requirements on the theory of ellipsis. That said, for inquisitive semantics to play a role in accounting for a given ellipsis process, the condition on R must be at least partially semantic in nature. Inquisitive semantics locates alternatives in the interpretation itself, rather than in the LF, and so even an LF-syntactic approach to recoverability such as Chung et al. () will not suffice. A complete theory of ellipsis must of course address the other major questions as well, which we do presently both for the main account described here, AnderBois (), as well as briefly discussing how these answers might change under other potential ways of incorporating inquisitive semantics into a more comprehensive theory of ellipsis. Since the account of sluicing in §. builds off of Merchant (), it addresses these major issues in largely similar ways. On the question of S, both accounts posit silent linguistic material with ellipsis consisting of PF-deletion. Typically, we assume that the deleted material is full interrogative clauses, though nothing in the analysis rules out other underlying structures, such as clefts, provided that they satisfy the relevant identity conditions (see, e.g., Barros  for a closely related but non-inquisitive approach making use of this option). On the question of R, the primary condition is a semantic one: symmetric entailment between A- and E-clauses with entailment crucially being defined over inquisitive semantic representations rather than just truth conditions. Beyond this, we have departed slightly from Merchant (), along with Chung () and others in supplementing this semantic condition with a minimal lexico-syntactic one to handle certain issues that arise in sprouting. Ultimately, then, the approach to R in §. is a hybrid one, in line with recent works in a variety of otherwise quite different approaches (e.g., Ginzburg and Sag ; Chung ; Jacobson b). Finally, for L, it does not seem that inquisitive semantics imposes any particular constraints on possible accounts. Given the close parallels with Merchant (), we refer





the reader to chapter  in that book (and Lobeck’s  work cited therein) for discussion of this issue. One place where inquisitive semantics does help shape the range of answers to the L question (or perhaps R) is that it provides a semantic account to certain kinds of cases that one might have thought were due to syntactic or other form-based constraints (e.g., the case of double negation in ()). However, we again stress that the way the account in §. answers the question of S is not intrinsic to inquisitive semantics per se, and that there are in principle many different frameworks for understanding ellipsis in which inquisitive semantics could be incorporated. Given the tight connection between inquisitive semantic issues and Questions Under Discussion (QUDs), one obvious candidate would be to build on Ginzburg and Sag’s () QUD-based approach. In a nutshell, their approach is a structure-free one which, beyond a minimal condition on form referring to a salient utterance (SAL-UTT), fills in the interpretation of the wh-phrase anaphorically from the maximal QUD (MAX-QUD). On a classical semantics for indefinites and disjunctions, however, this approach offers no clear way to explain the privileged role that these elements play in sluicing (in fact, Ginzburg and Sag  briefly argue against this claim on p. ). Adding inquisitive semantics to this picture, however, these elements conventionally make salient a possible QUD, thus explaining their privileged role. While such an account is in many ways an attractive one, further work on the various linguistic and non-linguistic ways in which QUDs arise in discourse is needed to make it viable. For example, () is a case where a clear QUD is established contextually, it would seem, and yet sluicing appears to be quite bad. ()

Scenario: I see the silhouette of someone is knocking on my office door, but can’t see the person’s face. #Who? // #I wonder who. // #Do you know who? // Who is it? // Who’s there?

Another alternative would be to seize upon the deep parallels between inquisitive semantics and dynamic semantics and draw upon work that treats ellipsis as discourse reference of a special sort, such as Hardt () (possibly supplemented with a lexico-syntactic condition of some sort). One challenge for such an approach, however, is that individual discourse reference does not exhibit the same interactions with double negation and appositives as does sluicing, as discussed by AnderBois (). Nonetheless, such an approach is in principle possible and, again, would give quite different answers to the question of S at least and possibly R and L as well. To summarize, inquisitive semantics proposes that the context change potential of sentences containing disjunctions, indefinites, and other existential quantification is ‘alternative-rich’. We expect therefore that we might find ramifications of this richer notion of semantic content in various areas of the grammar. We hope to have shown that this is so for ellipsis and to have shown one way of modeling such effects by focusing primarily on a particular ellipsis process: sluicing. Given the central role of interrogatives, indefinites, and disjunction in sluicing, these issues are naturally most salient here. However, it should be clear that once we adopt alternative-rich sentence meanings, the question arises of whether other ellipsis processes may similarly require reference to inquisitive semantic representations in some form.

  ......................................................................................................................

   ......................................................................................................................

 

. I

.................................................................................................................................. T chapter begins by looking at a range of ellipsis structures and asking if there are any empirical problems for a naive view of ellipsis processing—one where a structural ‘shell’ is copied from the antecedent clause and then remnants (overt constituents in the ellipsis clause) are lined up with the most similar constituents in the antecedent clause (section ..). The chapter then turns to the question of whether psycholinguistic evidence supports the claim that there is syntactic structure in the ellipsis site (section .), what psycholinguistic evidence says about recoverability and the nature of the relation between an antecedent and the ellipsis site (section .), and licensing (section .). Section . concludes. Before embarking on substantive discussion, let’s consider what psycholinguistic studies can tell us about the grammar of ellipsis. One traditional approach to linguistics grants intuitions a privileged status, as if linguistic intuitions directly reflect grammatical knowledge and only grammatical knowledge. A grammar, on this view, is just a systematic approach to characterizing linguistic intuitions. On this view, intuitions really are intuitions of grammaticality, perhaps with the caveat that some unprocessable sentences may appear to be ungrammatical (‘bad’) even though they have grammatical structures. An alternative view is that linguistic judgments are judgments of goodness and, like other judgments humans make, they are influenced by a variety of factors. These include the grammar, processing complexity, social factors, and the like. Empirical evidence favors this view. Marginal sentences are more likely to be rejected if the judge is in front of a mirror, because mirrors make people self-conscious, and self-conscious people are more conservative (Carroll et al. ; Nagata ). Complex sentences are rated as less good than less complex sentences, even when they are uncontroversially grammatical. The fact that processing influences acceptability judgments is perhaps brought home by two examples: ungrammatical sentences are rated more acceptable if they have a temporarily grammatical analysis than if they do not even temporarily permit a grammatical analysis (Fanselow and Frisch ); and fully grammatical sentences containing a licensed polarity item are rated less acceptable if they contain a temporary analysis on which the polarity item is unlicensed





than if they do not (Frazier and Clifton ). Such facts and many others are difficult to reconcile with the view that intuitions or acceptability ratings reflect only the grammar. On this second view, which will be assumed throughout the present chapter, the notion of ‘grammatical’ or ‘ungrammatical’ is a theoretical notion. Linguistic intuitions are judgments about the goodness of a sentence. It is up to the theorist to determine whether utterances that are judged bad do not conform to the grammar in some respect or whether they are difficult to process, for example. Examples that are judged good may be grammatical or they may just be something a native speaker might say or hear, i.e., an utterance that conforms to most of the constraints of the grammar, is transparent in terms of the message–form mapping, and thus might sound good in certain circumstances despite being ungrammatical. So intuitions of acceptability do not wear their interpretation on their sleeve. It is the overall theory of language (grammar plus processor) that must account for data—whether the data be intuitions/judgments or processing time data. This view makes processing studies germane even for those who might be interested only in grammar: without an empirically motivated theory of processing, it will be unclear whether it is the grammar or the processor that should explain some fact or some contrast in the acceptability status of sentences. It is less controversial, perhaps obvious, that psycholinguistic data, in the form of experimentally collected acceptability judgments, may contribute to linguistics by showing that some linguistic generalization reliably holds across a range of items and for a group of speakers. It may be less obvious, but I suspect it is true, that typically it is only experimentally collected acceptability judgments that permit interactions to be tested. For example, if one were to claim that d-linked phrases are not subject to island constraints then one would be claiming that the difference between d-linked and non-d-linked interrogatives is larger in the context of an extraction from an island than in the case of an extraction not out of an island (and not just that d-linking helps extractions out of islands). But this sort of ‘interaction’ judgment, where the size of the difference between the judged acceptability of one pair of sentences is predicted to be larger than the size of the difference between another pair of sentences, is not easy to make with confidence intuitively, though it is precisely the relevant comparison. Of course assessing this sort of interaction is exactly what experiments with factorial designs can do. In sum, psycholinguistic evidence about ellipsis and processing ellipsis can guide us to better theories of the grammar of ellipsis. Experimentally collected data can shore up the empirical basis for acceptability judgments, especially those that in reality involve predicted interactions, and data about processing ellipsis can offer a more nuanced view of ellipsis in part by helping to determine what should be accounted for in the grammar and what in the processor. Keeping open the possibility that the processor may bear some of the burden of accounting for acceptability judgments will become particularly important in .. in the discussion of recycling or repair theories based on the proposal that inputs that are likely syntactic blends may be repaired in comprehension.

.. Might a naive view of ellipsis processing suffice? Many types of ellipsis exist. Carlson () investigated replacives (), stripping (), and comparatives () and found strong parallelism effects for all of them. In what follows, these

  



studies will be reviewed and then used to evaluate a simple approach to ellipsis processing where the abstract syntactic structure of the antecedent clause is copied to create a syntactic ‘shell’ into which remnants are mapped, with the similarity between remnants and constituents in the antecedent clause used by the processor to carry out the mapping. A copied ‘shell’ might also apply in the processing of conjoined clauses in general, including those without ellipsis, explaining the processing advantage (priming) observed in processing structurally parallel conjoined clauses (e.g., Frazier et al. ). As we will see, this simple approach to processing ellipsis and conjunction would account for the bulk of the studies to be reviewed in this section, but it falls down dramatically in accounting for one robust effect. Carlson () reported the results of written studies in which participants were given a sentence and two paraphrases. They were asked to indicate which paraphrase corresponded to their interpretation of the sentence. The choice of an associate for the remnant (the phrase with which the remnant contrasts) was influenced by the lexical parallelism between remnant and associate in terms of the type of the DP, e.g., whether it was a name or a definite description. For example, participants chose Maude, the matrix subject as associate  percent of the time when the correlate was Marjorie in (a), but only chose Maude as the associate  percent of the time when the remnant was a fireman in (b). In ()–(), the percentages are the percent subject response. () a. Maude called a policeman for help, not Marjorie. (Replacive, Subject parallelism) % b. Maude called a policeman for help, not a fireman. (Replacive, Object parallelism) % () a. Maude called a policeman for help, but not Marjorie. (Stripping, Subject parallelism) % b. Maude called a policeman for help, but not a fireman. (Stripping, Object parallelism) % () a. Tasha called him more often than Sonya. (Comparatives, Subject parallelism) % b. Tasha called Bella more often than Sonya. (Comparatives, Neutral parallelism) % c. He called Tasha more often than Sonya. (Comparatives, Object parallelism) % For replacives (), stripping (), and comparatives (), a strong lexical parallelism effect was found: readers chose a phrase as an associate more often when it was the same type of DP as the remnant than when it was not. Carlson further showed that placing a prominent pitch accent on a potential associate in auditory interpretation studies also increased the probability of choosing that phrase as the associate. For example, accenting Maude in () increased the probability of choosing Maude as associate relative to a condition where Maude did not receive a prominent pitch accent.1 1

See Biezma () for studies of replacives and stripping in Spanish showing that the mere presence of an overt subject in pro-drop languages serves as a focusing device and thereby increases the probability that an overt subject will serve as an associate. See Grant () for the processing of comparatives in English though the work is not focused on comparative deletion per se.





Turning to gapping (a),2 Carlson () conducted a written judgment study where participants read sentences ambiguous between a gapping analysis and a conjoined VP analysis. Participants were asked to read each sentence, choose the best paraphrase for it, rate the difficulty of the sentence, and finally indicate whether it could have any other meaning. Testing examples like those in (), she found that a gapping analysis (Josh visited the office during the vacation and Sarah visited the office during the week) was not chosen often unless the potential subject remnant in the gapped clause was lexically parallel only to the subject of the antecedent clause.3 Even in this case (a), only a third of the responses indicated a gapped as opposed to a conjoined VP interpretation. ()

a. Josh visited the office during the vacation and Sarah during the week. b. Josh visited Marjorie during the vacation and Sarah during the week. c. Theresa introduced a new policeman to the mayor and Susan to the councilmen. % Gapping % Conjoined VP interpretation (a) % % (b) % % (c) % %

In short, the gapping analysis was disfavored strongly, though promoted to some extent by similarity between the subject of the antecedent clause and the potential subject remnant. In addition to the interpretation data presented under (), Carlson also found that gapped structures were rated as being more ambiguous and more difficult than their ungapped alternatives. As with replacives, stripping, and comparatives, placement of a prominent pitch accent on the subject of the antecedent clause and the potential subject remnant also promoted the gapping analysis in auditory interpretation studies. Carlson () placed a prominent pitch accent on the subject or object of the antecedent clause in sentences like (). ()

a. BOB insulted the guests during dinner and SAM during the dance. b. Bob insulted the GUESTS during dinner and SAM during the dance.

She found significantly more gapping analyses ( percent) in (a) than in (b), where only  percent of the responses indicated a gapping analysis. But even a prosody favoring the gapping analysis, (a), resulted in a majority of non-gapping analyses. Imagine that perceivers copied the syntactic structure of the first conjunct of a conjoined clause sentence, creating a predicted clausal template or shell, and then lined up the remnant(s) of the second conjunct with corresponding lexically parallel or focused constituents in the first conjunct, as illustrated in (). 2 Gapping is a plausible instance of ellipsis though it has received various analyses in the linguistic literature. In particular, see Johnson () for an analysis based on low coordination and across-theboard movement. 3 This pattern of favoring a particular analysis primarily only when one potential antecedent is accented and the other is not accords well with other findings, such as those reported in Carlson et al. (a). It also fits with current views characterizing memory as a cue-based direct-access system, where antecedents are activated based on their degree of match with retrieval cues (Lewis, Vasishth, and Van Dyke ), rather than as a (backwards) search mechanism.

   ()



[Josh [took [Susan [to the dentist]]]] and [Fred] [to the doctor]. [

[

[

[

]]]]

In effect the copied structure would permit constituents in the ellipsis clause to be mapped to empty counterparts of the corresponding constituents in the antecedent clause, e.g., with to the doctor mapped into the empty constituent under to the dentist. This would permit very simple essentially predictive processing of the ellipsis clause. The existence of strong parallelism effects would be explained. Assuming that a remnant and its associate contrast and thus should be accented, the effect of prominent pitch accents would follow naturally. Assuming that clausal conjuncts always give rise to syntactic shells, a clause containing ellipsis would be predicted to be easier to process than a corresponding unelided clause: fewer constituents need to be parsed. Clausal conjunction would be predicted, correctly, to show strong parallelism effects, i.e., an advantage when the second conjunct is lexically and structurally parallel to the first conjunct (Frazier et al. ). In terms of processing, the problem with this view is that it makes the wrong empirical predictions. Consider (c) Theresa introduced a new policeman to the mayor and Susan to the councilmen. Susan is more similar to Theresa than to other arguments and to the councilmen is most similar to to the mayor. So the processor should map the constituents in the ellipsis/conjoined clause to the positions beneath the most similar constituents, resulting in the gapping analysis (Susan introduced a new policeman to the councilman) of (c). But the gapping analysis was chosen only  percent of the time. The strong dispreference for gapping should not exist on this naive view of ellipsis. Thus, Carlson’s evidence seems to disconfirm the predictions of this simple and in some ways attractive approach to ellipsis. In (a) then, it must be something like the structural economy of the conjoined VP analysis that results in the preference for the ungapped structural analysis (see discussion of the results of Carlson, Dickey, and Kennedy  in Chapter  for further support for the assumption that structural economy is at issue). What does account for the results presented in this section? Carlson () argued that the effects that she observed suggest that the existence of similarities/parallelism between corresponding constituents of two clauses breeds a desire for further similarities so that constituents that are similar to each other preferentially are assigned similar phraseexternal syntax and constituents in corresponding syntactic positions preferentially have similar properties to each other phrase-internally. In short, parallelism breeds parallelism. Before turning to a discussion of structure in the ellipsis site, it should be noted that the types of ellipsis discussed so far, gapping (b), stripping (b), and comparatives (b), do not easily embed: ()

a. Mary took donations to the library and Jane to the church. b. *Mary took donations to the library and Bill thinks Jane to the church.

()

a. Maude called a policeman for help, but not Marjorie. b. *Maude called a policeman for help, but not Bill thinks Marjorie/but Bill thinks not Marjorie.

 ()

 a. Tasha called him more often than Sonya. b. *Tasha called him more often than Bill thinks Sonya.

By contrast, the types of ellipsis that have been most studied, such as VP ellipsis () or sluicing (), readily embed: ()

a. I left and Jane did too. b. I left and Bill thinks Jane did too.

()

a. Pablo knew someone called but he didn’t say who. b. Pablo knew someone called but Bill thinks he didn’t say who.

Frazier and Clifton () suggested that the two types of ellipsis, those that readily embed and those that do not, may be processed differently. They argued that a cost-free copying operation is available for the latter, allowing an antecedent to be copied into an elided constituent. As Carlson argued, the similarity of the remnant and its associate seems to be critical in the processing of ellipsis that doesn’t embed, such as the examples discussed in this section. By contrast, it will be argued in . that information structure notions are critical in processing the types that do freely embed, such as VP ellipsis, e.g., Sam stumbled but Bill didn’t, and sluicing, e.g., Someone left but I don’t know who. Further, it seems to matter for the latter types of ellipsis whether the elided constituent is part of asserted material (e.g., in a conjoined predicate) or is presupposed (in temporal or comparative clauses) (see discussion of Moulton ,  in section ..). To what extent these differences between types of ellipsis, those that embed and those that don’t, are deep, or even will hold up descriptively, is not evident at present.

. T  

..................................................................................................................................

.. Syntactic structure in the ellipsis site Theories of ellipsis differ in whether they claim that there is syntactic structure in the ellipsis site. One possibility is that an ellipsis site has no internal structure (Culicover and Jackendoff ), on a par with many accounts of pronouns. Another possibility is that the ellipsis site contains structure that is fully articulated but not pronounced (Merchant ). Psycholinguistic evidence relevant to this debate is discussed in section ... Examples will be given of various types of evidence or arguments for the existence of syntactic structure in the ellipsis site. There is a class of findings showing that syntactic mismatch effects are stronger, or temporally different, for ellipsis than for anaphora (section ...), and a class where syntactic matching of antecedent and remnant does not suffice because the derivation of the remnant would violate a constraint on movement such as pied-piping in German, or clause fronting in English (section ...). Violations of various types of island constraints and c-command violations are discussed (sections ...–...). We

  



conclude the section with evidence that a syntactic index is copied from an antecedent even when the copied index produces a semantic anomaly (section ...), and suggestive evidence that even phonological information copied from the antecedent may be available in the ellipsis site (...). In addition to the comprehension-based arguments and evidence that will be presented,4 investigators have also argued for the existence of syntactic structure in the ellipsis site in production studies: structure in the elided constituent primes the structure used to describe a picture (see Xiang, Grove, and Merchant , Hogue , and also Cai, Pickering, and Sturt , for the failure to find priming from “VP ellipsis” in Mandarin Chinese).

.. Empirical evidence for syntactic structure in the ellipsis site ... Larger penalties of syntactic mismatch for ellipsis than anaphora Hankamer and Sag () contrasted what they called “deep anaphora” and “surface anaphora.” Deep anaphors take a conceptual antecedent; surface anaphors require a linguistic antecedent, which must be syntactically parallel to the ellipsis site. Tanenhaus and Carlson () tested this theory in a series of “makes sense” judgment experiments. Participants were asked to read a context sentence, labeled S in (), read a second sentence (S), and judge whether S made sense in context. In one study, they manipulated voice (passive vs active) and anaphor type: VP ellipsis (a surface “anaphor”) and do it, a deep anaphor, as illustrated in (). (See also Ward, Sproat, and McKoon .) ()

S: Someone had to take out the garbage. The garbage had to be taken out.

S: But Bill refused to do it. But Bill refused to.

(Active) (Passive) +parallel parallel (active S) (passive S) (Deep Anaphor)  %  % (‘Surface Anaphor’)  %  %

What they found in this experiment and several other closely related studies was that “makes sense” judgments (percentage yes) reflected the critical interaction, with the lowest acceptability for surface anaphors in the non-parallel condition. However, response times showed a penalty for non-parallelism for both surface and deep anaphors. Tanenhaus and Carlson explained the response time advantage for parallel structures in deep anaphora in terms of the ease of retrieving an antecedent when it is parallel. Even if a deep anaphor doesn’t require a linguistic antecedent, identifying a conceptual antecedent may be easiest when a linguistic antecedent is readily identifiable.

4

Evidence from Antecedent-Contained Deletion has also been presented based on quantifier raising and the ease of ACD under conditions where raising has already applied for independent reasons (Hackl et al. ). Unfortunately, the reading-time results do not hold up; the acceptability judgments, which do, have been attributed to an alternative source (Gibson et al. ).





Since the time of Tanenhaus and Carlson’s pioneering work, it has generally been assumed that ellipsis is not the same as anaphora. There is considerable further work comparing ellipsis and anaphora and finding stronger syntactic mismatch penalties for ellipsis; see Ward, Sproat, and McKoon , for example. Roberts, Matsuo, and Duffield () used eye-tracking during reading, and found different effects of syntactic mismatch for VP ellipsis than VP anaphora and also a different time course for the two types of dependencies. Though the existence of syntactic mismatch penalties for ellipsis is relatively uncontroversial, not everyone agrees that they should be attributed to the grammar. See discussion in .. (and discussion of Dickey and Bunger  in Chapter  of the present volume).

... Syntactically matching remnants that violate pied-piping or clause movement conditions One way to test for the existence of syntactic structure in the ellipsis site is to examine whether conditions on movement apply within the ellipsis site. Merchant (a) argued that fragment answers to questions involve movement of the answer/new information to a focus projection high in the syntactic tree, and ellipsis of the remainder of the tree. The analysis predicts that (Bb) is ungrammatical as an answer to (A). It is ungrammatical for precisely the same reason that (b) is ungrammatical: movement of a clause requires the clause to have an overt complementizer. ()

Speaker A: Speaker B:

()

a. That he had lied, John denied. b. *He had lied, John denied.

What did John deny? a. That he had lied. b. *He had lied.

The analysis explains why only constituents are acceptable fragment answers. It is because only constituents move. Similar to the explanation for the contrast in (), the account also predicts that in a language where preposition stranding is not allowed, DP answers corresponding to the complement to a preposition will not be allowed (though various potential counterexamples have been noted, e.g., in Polish—see Nykiel ). Merchant et al. () confirmed the prediction experimentally. German is a language that does not permit preposition stranding. Hence, it is predicted that (B) should be a fully acceptable fragment answer, but (B) should not (since the DP would need to move to a focus projection, stranding a preposition.) In a written German acceptability rating study using a seven-point scale, testing forty participants at the University of Potsdam, the prediction was confirmed. Fragment answers like (B) received a mean rating of . whereas the DP fragment answers like (B) received a significantly lower rating of .. (Note that it is the difference in ratings that can be interpreted in this sort of study. The absolute numeric value means little apart from the relation of the experimental sentences to the filler sentences.)

   ()

Speaker A: Speaker B:

() Speaker A: Speaker B:

Haben sie mit dem MANN have they with the man Nein, mit der FRAU. no with the woman

gesprochen? spoken

Haben sie mit dem have they with the ??Nein, der FRAU. no the woman.

gesprochen? spoken

MANN man



The result is striking because both answers are constituents, both provide the information requested, and both syntactically match the question. Merchant et al. () also tested the prediction that a clausal fragment answer requires a complementizer. In a written acceptability rating study in English using a scale of  . . .  (=good), the predicted grammatical form received a rating of ., which was significantly higher than the rating for the no-complementizer counterpart (which received a mean of .). ()

Speaker A: Speaker B:

What did Kylie concede? That she took the keys. She took the keys.

In an eye movement recording study of items with the structure of (), a reading time penalty for no-complementizer fragment answers was found in first-pass data and number of fixations on the moved clause, as well as in later measures (Clifton and Frazier, in progress).5 This evidence indicates that penalties for ellipsis involving illicit movement to the focus phrase show a processing cost online, and not only in deliberative judgments or offline intuitions.

5

Examples from Clifton and Frazier (in progress) appear below, along with the data from a number of fixations in the penultimate region. Slashes indicate analysis regions but were not visible to the participants. (i) a. Ellipsis +that: Speaker A: What did Meg infer about school recitals?/ Speaker B: That/ she should go to school recitals/ whenever she can./ b. Ellipsis that: Speaker A: What did Meg infer about school recitals? Speaker B: She should go to school recitals whenever she can. c. No ellipsis +that: Speaker A: What did Meg infer about school recitals? Speaker B: She inferred that she should go to school recitals whenever she can. d. No ellipsis that: Speaker A: What did Meg infer about school recitals? Speaker B: She inferred she should go to school recitals whenever she can. Number of fixations in penultimate region: ellipsis no ellipsis no that . . that . . (interaction t=.)





... Islands are obeyed inside the ellipsis site Another way to test for syntactic structure in the ellipsis site is to check whether islands are obeyed inside elided constituents. Frazier and Clifton () reported several written acceptability judgment studies of island violations in sluicing structures including relative-clause islands, subject islands, and adjunct islands (d). They found significant penalties, drops in acceptability, for island violations inside an elided constituent, in a speeded acceptability judgment task. Interestingly, the island violation imposed a significantly greater cost when sprouting (Chung, Ladusaw, and McCloskey ) was required, i.e., in examples like John ate but I don’t know what, where no copied structure is available to house the variable that the interrogative must bind.6 Frazier and Clifton () tested adjunct island violations (d) in a visual “got it” task, where participants read the sentence and then indicate whether it is or is not a good sentence of their language. Crucially, there was a significant drop in acceptability both for an island violation in an overt constituent ((b) vs (a)) and in an elided constituent ((d) vs (c)). However, the size of the island violation penalty ((d) vs (c)) was smaller when the violation occurred in an elided constituent compared to when it occurred in an overt constituent ((b) vs (a)). (The sentences in (e,f) just show that the baseline sentences are comparable when there is no extraction.) ()

a. What lecture was Sally impressed with? . Ok b. What lecture was Sally impressed after? . Ok c. Sally was impressed with some lecture, but I don’t remember what. . Ok d. Sally was impressed after some lecture, but I don’t remember what. . Ok e. Sally was impressed with some lecture. . Ok f. Sally was impressed after some lecture. . Ok

They accounted for the fact that the adjunct island violation penalty was smaller inside an elided constituent than in a pronounced constituent in terms of the effect of focus on attention allocation during sentence processing. Other things equal, more attention is allocated to focused material than to unfocused material (e.g., Cutler and Fodor ), and by hypothesis more to overt material than to covert material. See also Sprouse and Hornstein () for discussion.

6

Frazier and Clifton () explained the high cost of sprouting island violations in terms of the failure of repair: the syntactic processor cannot make the repair because the syntactic processor does not violate islands, and the discourse processor cannot make the repair in sprouting examples because it cannot build syntactic structure.

  



... C-command inside the ellipsis site If syntactic structure exists inside an elided constituent, then syntactic relations such as c-command should be defined for dependencies inside elided constituents. Yoshida, Dickey, and Sturt () exploited the gender mismatch effect to investigate whether there are articulated c-command relations inside the ellipsis site in sluicing. Consider (a). If the sentence ends after himself, then the only analysis of the sentence is one with sluicing, where an interrogative clause has been elided. On the sluicing analysis, the reflexive, e.g., himself in (a) in the interrogative constituent (which story about himself), will be bound by grandfather, and no gender mismatch will result. However, in (b) there will be a gender mismatch when himself is bound by grandmother. Consequently, if readers compute the sluiced analysis, there should be a boggle at himself in (b). The sentences in (c,d) are not even temporarily open to a sluicing analysis, so they serve as controls. Reading times in (d) should be no longer than in (c) since a sluicing analysis is not possible in either of the two sentences. In a self-paced reading study, reading times at himself and the following word confirmed the prediction that there would be a gender mismatch effect only in (b). ()

Ellipsis temporarily possible if sentence ends after himself: a. Jane’s grandfather told some stories but we couldn’t remember which story about himself from the party his brother was so very impressed with. b. Justin’s grandmother told some stories but we couldn’t remember which story about himself from the party his brother was so very impressed with. Ellipsis not possible (with which): c. Jane’s grandfather told some stories but we couldn’t remember with which story about himself from the party his brother was so very impressed. d. Justin’s grandmother told some stories but we couldn’t remember with which story about himself from the party his brother was so very impressed.

The results suggest that binding of the reflexive takes place inside the ellipsis clause. This in turn suggests that c-command relations are defined in the ellipsis site given that typically only c-commanding antecedents are considered for reflexives (Dillon et al. ). That the c-command relations in the elided constituent are assessed so quickly, in circumstances where there is no need to posit ellipsis at all, suggests that the parser automatically computes syntactic relations for material that is copied from the antecedent.

... Indices are copied into the ellipsis site Investigating the processing of elided constituents whose antecedents contain syntactic indices could be informative with respect to different hypotheses about the grammatical structure in the ellipsis site. Semantic priming has been used to study the processing of the strict interpretation of sentences like (a), where the fireman defended the policeman. Participants listen to a sentence and at some point during the sentence a letter string, the target for a word/non-word decision, appears on a computer. On the critical “yes” (word) decisions, a participant’s response times are faster if the word is semantically related to





what the participant is processing than if it is semantically unrelated. Shapiro and Hestvik () investigated whether there would be priming for words related to the first clause subject in the ellipsis clause, at the site of the elided verb phrase, in examples like (a). The question is whether hearers compute the strict interpretation, where the first clause subject binds the reflexive, online. The question is of particular interest because it is assumed that it is the sloppy interpretation (the fireman defended the fireman) that is preferred intuitively. Their results suggested that priming from the strict interpretation does exist. Following up this result, Shapiro et al. () used priming to investigate whether the strict (the fireman defends the policeman) interpretation is computed even in sentences containing a predicate like perjure, which is not felicitous when applied to a different person, i.e., the fireman cannot perjure the policeman. ()

a. The policeman defended himself and the fireman did too, according to someone who was there. b. The policeman perjured himself and the fireman did too, according to someone who was there.

The results indicated that priming from the first clause subject did take place at the ellipsis site even in (b). The investigators concluded that, when copying the antecedent VP into the ellipsis site, the index on the reflexive is copied too, and it is processing this index that leads to reactivation of the first clause subject (though see Koeneman, Baauw, and Wijnen  for discussion). In short, if the ellipsis site contains copied syntactic structure, the result makes sense. But without the assumption that syntactic structure exists in the ellipsis site, this result is puzzling. If listeners only copied the semantic property (perjure-self) from the antecedent into the ellipsis site, as expected on a purely semantic theory of ellipsis, then no priming of words related to policeman should have been observed.

... Phonological information may be present in the ellipsis site Few studies have tried to investigate whether phonological structure is activated in an elided constituent. However, Snider and Runner () is an exception. They investigated the types of information activated at the ellipsis site, comparing VP ellipsis (a) and VP anaphora (b). ()

a. The security guard opened the lock, and the night watchman did too. b. The security guard opened the lock, and the night watchman did it too.

They conjectured that if ellipsis involves full syntactic structure at the ellipsis site then words should be present, and they should activate their phonological properties. In a visual world study, they showed that semantic associates of the antecedent object (key) were available for both VP ellipsis and VP anaphora. But at the point at the ellipsis site/pronoun, effects of phonological neighbors (log) of the antecedent object were activated only for VP ellipsis, but not for VP anaphora. Although any single study of this sort probably cannot

  



show compellingly that syntactic structure is present at the ellipsis site, various types of evidence are now accumulating which converge on this conclusion.

. I 

..................................................................................................................................

.. Recoverability To recover the interpretation of an elided constituent there must be an identity relation of some sort with an antecedent. Theories of ellipsis differ on whether it is identity at some level of syntactic structure or semantic identity. It is to this topic that we turn in section ... However, given that certain morphosyntactic features may differ between the antecedent and the ellipsis site, e.g., the value for gender in The boy hurt himself and the girl did too, the term ‘matching’ (identity up to differences in morphological features such as gender or number) will be used in the present chapter rather than ‘identity’. In section .. we take up a proposal based on repair of the input that may restore syntactic matching of antecedent and ellipsis site under particular circumstances. Section .. takes up accounts that do not require syntactic matching in the grammar, but nevertheless favor matching for processing reasons.

.. Syntactic structure effects From a psycholinguistic perspective, many of the findings discussed in section . suggest the existence of a syntactic matching constraint. The reason is that comprehenders (listeners and readers) generally could only identify structure in the ellipsis site by copying it from the antecedent. This copying will result in syntactic matching of antecedent and ellipsis.7 If the relation between antecedent and ellipsis site were one of semantic matching in a purely semantic theory of ellipsis, it would be puzzling how the comprehender managed to intuit the structure in the ellipsis site. Indeed, it would be puzzling why the comprehender would even bother to spell out the structure in the ellipsis site if an interpretation had already been arrived at through the semantic matching constraint and the grammar did not require syntactic structure to be present in the ellipsis site. If the grammatically imposed matching relation were semantic, the question also arises of why identity/matching effects are stronger for ellipsis than for anaphora, why semantically matching DPs (for German fragment answers) are less acceptable than PPs in some contexts (those prohibiting preposition stranding), why gender mismatch effects should appear in (b) but not (d), and why priming of the first clause subject should appear at the gap site in the second clause when the corresponding (strict) interpretation would be semantically impossible as in (b). The hypothesis that syntactic identity is grammatically required also offers a natural account for effects implicating the phonological content of the empty phrase on the assumption that what is copied includes words. It is possible that a 7

Copying of the antecedent and structure sharing may be notational variants, though see Martin and McElree (, ), for an alternative view.





theory embracing a semantic identity/matching constraint might be devised, e.g., one based on activating all properties of the antecedent, including its phonology, when the semantic antecedent is retrieved. However, the challenge would then be to say why the same thing didn’t happen when the antecedent for a pronoun is retrieved. Turning to a hybrid grammatical account, claiming that syntactic structure exists in the ellipsis site but that the identity relation is semantic (Merchant ), in principle one might imagine that the processor builds syntactic structure in the elided constituent not by copying it from the antecedent, but from the bottom up, syntactic inference by syntactic inference (assuming each syntactic phrase postulated must be licensed by the grammar). But this incorrectly predicts a general preference for small ellipsis sites and it predicts, counterfactually, that large elided constituents should take longer to process than small ones because more structure must be built (see Frazier and Clifton  and Martin and McElree  for evidence that large elided constituents are no more costly than small ones).8

.. The relation between the grammar and the processor As noted in the introduction, the observation that it is the grammar together with the processor that account for actual data, including acceptability judgments, is important because it means that intuitions need not be explained by the grammar alone. Arregui et al. () proposed a recycling approach to ellipsis, where a syntactically mismatching antecedent is repaired at LF. In later developments of the approach, it is shown that likely syntactic blends are repaired in general, not just in the case of ellipsis (Frazier ). On this approach, it is the processor which accounts for the relative acceptability of some easy-torepair sentence tokens. Arregui et al. () argued that mismatch ellipsis is an instance of acceptable ungrammaticality. In effect, a sentence that may have been produced as a syntactic blend (a speech error where elements from two different formulations of a message are mixed together) may be repaired in comprehension. In a speeded acceptability judgment study, Arregui et al. () tested passive antecedent-active ellipsis (a) mismatches and active antecedent-passive ellipsis (b) mismatches. They predicted that the passive–active mismatches would be more acceptable than active–passive mismatches because passives are misremembered as actives more often than actives are misremembered as passives (Mehler ). The prediction was confirmed in a self-paced reading study. ()

a. The dessert was praised by the customer after the critic did already. b. The customer praised the dessert after the appetizer was already.

8 See Paape () for evidence that in examples where the antecedent contains a garden path there is no parsing difficulty at the ellipsis site. If the parser built structure in the ellipsis site by reparsing the antecedent’s word string, difficulty would have been expected. Note that one early study did find elided constituents with large antecedents to be more difficult to process than ones with small antecedents (Murphy ). It’s not entirely clear why, though perhaps the use of small clause antecedents is relevant.

  



They proposed that a speaker might utter a passive antecedent clause and then at the ellipsis clause forget the voice of the antecedent clause and utter an active ellipsis clause even if she is using a grammar requiring syntactic matching. Listeners might find such mismatches relatively acceptable because they sound familiar, in the sense of corresponding to an expected output of the human language system. Further, they are easy to repair. Indeed, Arregui et al. also present data that the number of operations implied by a repair correlates inversely with acceptability (see Kim et al.  for further evidence). On this account syntactically mismatching ellipsis examples are ungrammatical but certain tokens of mismatch ellipsis that are motivated by the principles of the production system are nevertheless relatively acceptable (if they are also easy to repair and yield a plausible interpretation).9 The repair hypothesis predicts that other structures where the speaker has two common ways of syntactically formulating the same message should also give rise to relatively acceptable syntactic mismatches, and to interpretations that would result from the repaired structure. For example, a speaker predicating two predicates of the same individual might express the message using conjoined VPs or using conjoined clauses. This might result in VP ellipsis even when the antecedent contained conjoined clauses: a speaker might utter a conjoined clause antecedent, forget that it was conjoined clauses rather than conjoined VPs, and then utter an elided VP. The comprehender might repair the input and assign it a plausible interpretation. Frazier and Clifton () performed an interpretation study of sentences like those in ().  percent of the responses indicated that readers computed the conjoined VP (George travels in winter and stays home in summer) interpretation of both the conjoined VP and the conjoined clause examples, as if the comprehender repaired the conjoined clause input to a conjoined VP, i.e., to the antecedent structure that the speaker might have chosen but didn’t actually choose. ()

a. Ian travels in winter and stays home in summer. George does too. b. Ian travels in winter and he stays home in summer. George does too.

The speech error repair account was developed on the assumption that the grammar requires syntactic matching of antecedent and ellipsis site. Thus, mismatch ellipsis on this account is always technically ungrammatical, but sometimes it is acceptable. In particular, as already noted, mismatch ellipsis was argued to be acceptable when an input could be the result of a likely production error, one that is easy to repair, and interpreting the repaired input results in a plausible interpretation. Frazier () argues that the assumption that repaired utterances may sometimes be relatively acceptable is needed in general and not just for ellipsis. She argues that there are two interpretive systems for natural language: one that is compositional, type-based, and delivers both plausible and implausible interpretations; and one that makes reference to the grammar and the processor, is token-based, and delivers only plausible interpretations. (See Frazier and Clifton  for further evidence of repairing syntactic blends outside the domain of ellipsis.) 9

One issue is why voice mismatch seems less acceptable in sluicing than in VP ellipsis. The difference probably lies in the stronger need for parallelism when variables must be identified as is always the case in sluicing, where the interrogative must bind a variable. This predicts that voice mismatches should also be worse in VP ellipsis when the elided VP contains a variable, as in ACD cases.





.. Alternative accounts of syntactic mismatches Proponents of ellipsis theories that do not require syntactic matching (Kim et al. ; Miller and Hemforth b), or do not require it in certain discourse (Kehler ) or information structure (Kertz ) circumstances, do not think that experimental demonstrations of a reliable acceptability penalty for syntactic mismatches are problematic for their theories, even though mismatches are predicted on their theories to be fully grammatical. They point out that there might be another explanation for the mismatch penalty—some explanation other than that the grammar requires syntactic matching. For example, Kertz () compared “contrastive topic” sentences (a,b) and auxiliary focus sentences (c,d), expecting a syntactic mismatch effect only with “contrastive topic” sentences. She found a syntactic mismatch penalty even for auxiliary focus sentences ((d) relative to (c)), where it was not predicted. She noted that sentences without a syntactically matching antecedent may have a small LF mismatch penalty (though she does not elaborate on what kind of penalty this is supposed to be). ()

a. b. c. d.

Venomous snakes are easy to identify, and poisonous plants are as well. It’s easy to identify venomous snakes, and poisonous plants are as well. It’s easy to identify venomous snakes, and most experienced hikers can. Venomous snakes are easy to identify, and most experienced hikers can.

Miller and Hemforth (b) compared verbal (b) and nominal (a) antecedents for VP ellipsis. They found a mismatch penalty in (a) not expected on their account where the grammar requires only semantic matching of antecedent and ellipsis site. But they note that the matching examples might be easier to process because a matching antecedent (b) might be easier to identify and retrieve than a syntactically mismatching antecedent, e.g., in (a).10 () a. It is impossible to predict Andrew’s participation in the chess tournament. He is sure to win if he does/does it. b. It is impossible to predict whether Andrew will participate in the chess tournament. He is sure to win if he does/does it. The idea that syntactic matching may facilitate processing or antecedent retrieval seems perfectly reasonable, and it makes a prediction. The effect of syntactic matching should be greater in more complicated sentences than in easy sentences where retrieval of the antecedent should not be difficult. Although this prediction has not been tested experimentally, my own intuitions do not support it. Indeed, they suggest the reverse. It is precisely in short simple unadorned sentences like () that a violation of syntactic matching seems worst despite the fact that there is little processing complexity, ambiguity, or memory load. 10 Miller and Hemforth (b) also manipulated properties of the context, not illustrated here, because it is not of relevance to the present discussion. In the end they propose a theory with a semantic identity requirement. They suggest that Occam’s razor favors this account over one containing repair operations. However, as discussed in section ., if we assume that the theory of language must include both a grammar and a theory of processing, then given the independent evidence that repair operations are needed, e.g., for garden paths, the Occam’s razor argument no longer holds.

   ()



John was slapped and Mary did.

In any case, it is reassuring that the distinct approaches make distinct predictions. In sum, a theory with a semantic identity/matching requirement in the grammar and a memory retrieval account of the (as far as I know) always present syntactic mismatch penalty predicts a larger mismatch penalty in more complex sentences. A theory with a syntactic matching requirement predicts a syntactic mismatch penalty should always be present, even in very simple easy-to-process sentences.

. L

..................................................................................................................................

.. Licensing in the grammar Ellipsis cannot occur just anywhere but instead is restricted. This is usually discussed in terms of licensing conditions (e.g., Lobeck ). The ellipsis site must be licensed by a “governor” that permits its complement to be empty (see also Merchant , for example). In an example like (), the complement of denied could not be elided because there is no licensor for an empty CP (and deny is not a verb that permits null complement anaphora). Consequently, the ellipsis is not licensed and the sentence is ungrammatical. ()

*Kyle said Angelika left already but Lisa denied [that Angelika left already].

In the psycholinguistic literature, this type of licensing has not been investigated at all, perhaps because it is difficult to compare the processing of grammatical/licensed and ungrammatical/unlicensed ellipses. Other types of licensing are related to discourse and information structure. Discourse coherence relations will be discussed in .., the main assertion hypothesis in .., and the role of accommodating presuppositions and its interaction with ellipsis in ...

.. Discourse coherence relations Kehler () proposed an interesting theory of discourse coherence relations in which each clause of a discourse is connected by a discourse coherence relation such as a causal relation or a “resemblance” relation involving contrast (John ate cauliflower. Bill ate peas) or similarity (Amanda ate snails. Joe ate them too). An ellipsis clause connected by a resemblance (similarity, contrast) coherence relation requires syntactic matching because sub-clausal constituents (John, Bill, cauliflower, peas) must be identified in order to identify the similarities/contrasts that underlie the resemblance coherence relation. In contrast, an ellipsis clause connected by a causal coherence relation does not require syntactic matching on this theory because a causal coherence relation may be established based on propositional information alone, without consulting sub-clausal constituents. See Chapter  (this volume) for an enlightening discussion and development of this line of inquiry. The results of psycholinguistic studies of ellipsis and coherence relations





(e.g., Frazier and Clifton ) have often been less than compelling to some proponents of the theory, because they have not taken into account the observation, clearly supported in Kehler (), that connectives may establish more than one discourse coherence relation. To ensure that participants understood the experimental sentences with the intended discourse coherence relation, Clifton () tested examples like those in () in a written acceptability judgment task. In () there is a question appropriate for either a resemblance (c,d) or causal (a,b) construal of the answer. The answer contained verb phrase ellipsis with either a matching (b,d) or mismatching (a,c) antecedent. (The word too was present only in the second of two experiments, which were otherwise identical.) () a. Why did Lee get involved in checking out the problem, do you think? The problem was looked into by Kim so Lee did (too). b. Why did Lee get involved in checking out the problem, do you think? Kim looked into the problem so Lee did (too). c. Who else besides Lee got involved in checking out the problem, do you think? The problem was looked into by Kim just like Lee did. d. Who else besides Lee got involved in checking out the problem, do you think? Kim looked into the problem just like Lee did. In both experiments, Clifton found a penalty for syntactic mismatches, but one of equal size statistically for resemblance (c) and for causal (a) relations. This suggests that a causal coherence relation cannot license syntactically mismatching ellipsis. Finding the same size cost for syntactic mismatches in resemblance coherence relation and causal coherence relation examples cannot be due to ambiguity in the coherence relation in Clifton’s experimental materials.

.. Main assertion preference Focus is important in choosing an antecedent, as discussed in connection with Carlson () for studies of replacives, stripping, and comparatives. Studies of sluicing show that focused antecedents are also preferred in the case of sluicing (Frazier and Clifton ). In addition to focus, other information structure notions are also important, such as the notion of “main assertion”—typically the content expressed by main clauses, not relative clauses or subordinate clauses.11 Frazier and Clifton () argued that across sentence boundaries there is a tendency for an utterance to relate to the main assertion of the preceding utterance. Consider () where the elided VP in the final clause is ambiguous between an interpretation with the embedded VP taken as antecedent and an interpretation based on the matrix VP as antecedent. In written comprehension studies, the VP ellipsis in (a), the baseline, was interpreted with the matrix VP antecedent  percent of 11

The notion of main assertion is related to the notion of Question Under Discussion (QUD), assuming the QUD is generally introduced by material in a prominent position in the tree. Elided material preferentially addresses the QUD (Grant et al. ); the QUD is preferentially derived from alternatives introduced by material in the main assertion (Clifton and Frazier ).

  



the time. By contrast, in (b) where the only difference is the presence of a sentence boundary, there were significantly more ( percent) matrix VP antecedents. ()

a. John said Fred went to Europe and Mary did too. b. John said Fred went to Europe. Mary did too.

In () one might be concerned that the main-assertion tendency is really just the absence of the usually observed syntactic low attachment preference. In a self-paced reading study of the sentences in (), however, the preferred interpretation, as revealed by the answers to questions, was for the matrix antecedent roughly  percent ( percent in (a),  percent in (b)) independent of the order of main and subordinate clauses. In () there is no attachment issue, but the main assertion antecedent was favored nonetheless. ()

a. Mary laughed after she made a joke about the supervisor. Then Tina did too. b. After Mary laughed she made a joke about the supervisor. Then Tina did too.

Similarly, Frazier and Clifton () report results from a self-paced reading study of potentially epistemic clauses like I think in (a), which are open to an analysis where the matrix clause only contributes information about the speaker’s degree of certainty about the assertion expressed in the complement clause. Since (b) is not open to such an interpretation, the prediction of the main assertion principle is that there should be fewer matrix antecedents in (a) than in (b). This prediction was confirmed by the answers to questions about the elided VP, e.g., What does Sam do?, with  percent matrix VP responses in (a) and significantly more ( percent) in (b). (As usual, it is the difference between conditions that is of interest to us rather than the baseline level of interpretations of each type.) ()

a. I think Mary smokes. Sam does too. b. The teacher thinks Mary smokes. Sam does too.

A control study tested verbs like announce, which are not open to an epistemic interpretation, and found that the effect was not due to the presence of a first- vs third-person subject per se. One might wonder whether the main assertion principle holds only of ellipsis. Presumably it does not. In a study where participants were given a sentence followed by two words and then had to decide whether it was true that both words appeared in the sentence or not (a “two-word probe study”), Walker, Gough, and Wall () tested items like () and found the fastest times for probe words both drawn from the main clause (scouts killed in ()). In this task participants are simply understanding a sentence and then checking whether the probe words occurred in the sentence. So presumably what the results indicate is that the main assertion of a sentence is what’s most salient after the sentence has been understood. ()

The scouts the Indians saw killed a buffalo. saw killed Scouts . . Indians . .





The results thus suggest that the preferred antecedents for ellipsis are reflecting general properties of memory and information structure, not something specifically tied to ellipsis.

.. Accommodation Moulton () proposed a principle of minimizing accommodation. He focused on the information status of different constituents, as illustrated in (). ()

The Canadian team accepted the silver medal when they arrived at the stadium. Asserted: The Canadian team accepted the silver medal. Presupposed: The Canadian team arrived at the stadium.

In a written interpretation study, he found that in interpreting an elided VP in a coordinated main clause, there was a preference for large antecedents. However, when the elided VP was in a structure that generally conveys presupposed information (b,c), then there was a preference for small antecedents. ()

Coordinated clauses (asserted) versus temporal (after, before, when) and relative clauses a. Eddy quietly sang the song and Fred did too. Coordinated main clause % Large b. Eddy quietly sang the song after Fred did. Temporal adverbial % c. Eddy quietly sang the song that Fred did. Relative clause (ACD) % What did Fred do? i. quietly sing a song ii. sing a song

Large antecedent Small antecedent

In sum, the choice of antecedent is sensitive to whether the elided constituent corresponds to asserted or presupposed content. In presupposed contexts, the processor minimizes accommodation. See Moulton () for studies ruling out alternative syntactic accounts of the results, and Moulton () for additional studies, interpreted in terms of the more general hypothesis that comprehenders prefer to minimize embedded content, specifically content reported in a restrictor clause.

.. Question Under Discussion (QUD) The idea that discourse is organized by a series of questions has received considerable attention in semantics (Beaver and Clark ; Roberts ), and the notion has been applied to ellipsis (Ginzburg ). Grant, Frazier, and Clifton () proposed that the presence of alternatives that might be taken as the Question Under Discussion (QUD) facilitates mismatch ellipsis if the elided constituent comments on those alternatives/the QUD (see Clifton and Frazier , for the role of alternatives and QUD in processing in general). In their study, it was the presence of a modal that introduced alternatives (needed in (b)), presumably introducing the

  



implicit question of whether the state of affairs described (the information being released) was actual or not. The modal increased the acceptability of mismatches in written acceptability judgment studies. ()

a. This information was released / but Gorbachev didn’t / even though / lives were clearly at stake. b. This information needed to be released / but Gorbachev didn’t / even though / lives were clearly at stake.

In an eye movement recording study, it also resulted in faster reading times in (b) than (a) in the region after the ellipsis site (even though in ()).12 (The slashes in () indicate analysis regions, but the slashes were not visible to participants.) The results suggest that minor syntactic deviance can be tolerated better when the offending form addresses the QUD. In Miller and Hemforth’s (b) study of polar nouns described in connection with (), they also manipulated whether the context introduced an alternative or not (polar alternatives, e.g., It is impossible to predict whether Andrew will participate/Andrew’s participation in the tournament. He is sure to win if he does/does it.). In addition to a penalty for nominal antecedents compared to verbal antecedents for verb phrase ellipsis, they also observed an interaction whereby having an alternative in context facilitated verb phrase ellipsis more than verb phrase anaphora. One might take this as further indication that once alternatives are under discussion they facilitate ellipsis that comments on those alternatives. Miller and Hemforth’s own conclusion was that there is a special type of verb phrases ellipsis, aux focus ellipsis, which is limited to circumstances where alternatives involving tense, aspect, mood, or polarity are salient, and the entire point of the utterance containing ellipsis is to pick one member from that set. In the end they opt for a semantic theory of recoverability because they think it is simplest.13

. C

.................................................................................................................................. Many different types of psycholinguistic arguments and data support the hypothesis that there is syntactic structure in an ellipsis site. Stronger penalties are observed for structural mismatch between an ellipsis clause and its antecedent clause than between a pronouncontaining clause and the clause containing its antecedent. Gender mismatch effects show that c-command relations in the elided constituent exist and are computed rapidly and 12 Specifically, in the region following the ellipsis site, first fixation durations ( ms vs  ms) and gaze durations ( ms vs  ms) were significantly longer in the non-implicature (a-) forms than in the implicature (b-) forms. In the final region of the sentence, ‘go past’ times showed the same pattern ( ms vs  ms, all p < .). 13 While Miller and Hemforth’s theory might be simple and adequate when considering only polar nouns, they do not really explain why non-polar nouns are not as good as polar nouns: if the grammar does not require anything except semantic matching, non-polar nouns should also be acceptable antecedents.





automatically even when ellipsis is not necessary. Effects of island violations have been established in elided constituents. Indices from an antecedent are copied even when they are incompatible with the semantic constraints of the predicate. Suggestive evidence indicates that phonological content may be copied from an antecedent into an ellipsis clause but not into a clause containing anaphora. Recovering meaning for an ellipsis clause is dependent on there being an identity or matching relation with an antecedent clause. It was argued that this relation is one of syntactic matching. In the absence of mental telepathy between speaker and hearer, the very existence of syntactic structure in the ellipsis site is at least suggestive evidence for a syntactic relation since it is by copying, or sharing, structure with the antecedent clause that the processor could obtain syntactic structure.14 Although in principle the processor might build syntactic structure in the ellipsis site bottom-up, inference by inference, guided by meaning or by properties of the antecedent or by re-parsing the antecedent word string, any of these possibilities would be difficult to reconcile with the fact that large ellipsis sites are often preferred to smaller ones. Further, there is no cost for large elided constituents relative to small ones (Frazier and Clifton , ; Martin and McElree ). If structure were built bottom-up, this would fly in the face of the parser’s very pervasive structural economy preferences. In sum, given syntactic structure in the ellipsis site, it is probable that it comes into being by copying the syntax of the antecedent clause. One central reason that theorists prefer a semantic identity requirement is their belief in Occam’s razor: a semantic identity requirement in the grammar appears to be simpler than a syntactic matching requirement plus a theory of repair (see discussion in Miller and Hemforth b). However, the theory of sentence processing must independently contain a theory of (garden path) repair and an account of the repair of syntactic blends not involving ellipsis (Frazier ; Frazier and Clifton ). Once Occam’s razor is applied to the entire theory of language, both the grammar and the theory of processing, the claimed simplicity of a grammar with semantic identity vanishes. It was emphasized that it is the grammar together with a theory of processing that makes predictions about actual data such as linguistic intuitions/judgments, reading times, and the like. When comprehenders encounter an ungrammatical sentence, they do not assign an asterisk and stop processing. Rather they assign a best-fit analysis and interpretation. Under certain circumstances such utterances may be acceptable: when the utterance sounds natural (expected given the principles and constraints of the human production system), the input is easy to repair, and the interpretation is plausible, especially if it addresses a tempting Question Under Discussion. The assumption that “acceptable but technically ungrammatical” utterances may be rated as being acceptable relative to other ungrammatical sentences is needed independently (Frazier  and references therein). In the case of ellipsis, it offers a potential explanation for why certain tokens of ellipsis sentences with a syntactic mismatch between antecedent and elided constituent may sound quite good even if the grammar requires syntactic matching of antecedent and ellipsis site.

14 The listener might posit structure in the ellipsis site by guessing randomly and then try to match it to the antecedent, and then guess again, etc. But this sort of Analysis-by-Synthesis model is inefficient at best and has largely been abandoned.

  



Discourse factors are also important in ellipsis. In particular, across sentence boundaries, other things being equal, listeners and readers prefer antecedents that are part of the main assertion. Preferred ellipsis sites in constituents with asserted content tend to be as big as possible. By contrast, preferred ellipsis sites in constituents with presupposed content appear to be as small as possible, presumably to minimize accommodation.

A I am grateful to the editors, two anonymous reviewers, Chuck Clifton, and Tim Hunter for helpful comments on the manuscript.

  ......................................................................................................................

   ......................................................................................................................

 

. I: W      ?

.................................................................................................................................. H does a child acquire the special properties of ellipsis which vary from language to language? (See Johnson  for background.) How does she know that (a) is impossible in English with articles but possible in German (b) with case marking (ein = nominative)? Furthermore, she must know that demonstratives with noun agreement are fine in English as shown in (c), where this and that are singular (not these, those), marking a deleted noun as singular: ()

Here are some cookies. a. *I want a__[=cookie] but not the__[=cookie]. b. Ich will ein__[=Kuchen], aber nicht das__[=Kuchen]. c. I want this__[=cookie] but not that__[=cookie].

Likewise, English allows verb phrase ellipsis (VPE) but other languages do not. Languages that allow VPE often do not allow a hanging to—a language-particular licenser—but English does. ()

a. You sang, then I did__[=sing]. b. You sang, but I didn’t want to__[=sing].

It will become clear (a) that subtle, language-particular features like agreement license ellipsis, (b) that the child must have one (or more) operations to recover what is missing, and (c) that different degrees of precision are possible for the representation that is created. Sometimes gender, case, and number can be ignored, suggesting a more abstract or incomplete representation, but at times, some but not all adult speakers will experience jarring ungrammaticality if gender (her arm) is altered (to his arm): ()

Mary1 broke her1 arm, and so did John2. ‘Mary1 broke her1 arm, and John2 broke her1/his2 arm.’

  



Such sentences indicate unresolved ragged edges for ellipsis which continue to perplex linguistic theory but which are present in the child’s everyday linguistic environment. There is an immediate acquisition consequence here. Where the input data is especially complex and decisive cases are rare, then abstract principles must play an organizing role. Only via narrow and sharp UG principles could a child possibly carry out the induction of where ellipsis is possible/impossible and what meaning must be reconstructed as. In general, we know that our grammar can lead to amazingly exact forms of ellipsis over complex phrases. ()

John wore a bright red, speckled HAND-made Indian bracelet, and so did Mary__. (Intended meaning) ‘Mary also wore a bright red, speckled, HAND-made Indian bracelet.’

When does a child know a full reconstruction with focus is necessary? This is one of many experiments waiting to be done. What will a -year-old child say if we ask, ‘what did Mary wear’? Will it be, ‘a bright red, speckled HAND-made bracelet’ with identical phonology and focus intonation, showing that a full phonetic representation has been recovered? Or, will it be just the head ‘bracelet’? Thus far, no one has done this experiment.1 The entire question of the role of topicalization and focal contrast in ellipsis acquisition remains in early stages.2

.. General theoretical consequences: An overview of Strict Interfaces We begin with an extensive introduction to orient linguistic theory to the acquisition challenge. As we proceed, we provide small previews of the data which we then navigate in fairly microscopic detail. Our exposition aims to provide an overview of what has been observed about ellipsis for major structures in acquisition while proposing a theoretical approach.3 The core idea is that the child should begin with some universal invariant Strict Interfaces. They function as a teleological goal of acquisition; it is what a child looks for and seeks to impose. The abstract ideal of a Strict Interface is as follows:4

1

At least to my knowledge it has not been done. See Renato () who demonstrates some sensitivity to the pragmatics of topic and focus for double-object constructions among -year-olds, but resistance to ellipsis, indicating that the computation of topic focus and ellipsis is not initially complete. 3 We will not address comparatives or gapping (see Syrett  and Bryant , respectively, for interesting work which points toward an important role for semantics). 4 A major presupposition here is that interfaces must be biologically stipulated, as they are elsewhere. The links between the heart and lungs (i.e., interfaces) are just as innate as the heart and lungs, or the link between separate three-dimensional representations for eyes and hands needed in eye–hand coordination. The view that interface links are innate stands in contrast to arguments that cognitive third-factor features of language (Chomsky ) mean that much of grammar is not innate. See Roeper () for discussion of how the grammatical interface with a cognitive notion of  is modified at the interface with grammar. 2

 ()

 Strict Interface Hypothesis: The acquisition device assumes that there should be an isomorphic connection between LF-semantics, syntax, and phonology.

An intuitive example of a Strict Interface is the notion of a negative imperative: ‘don’t!’. This is understood by children below  years although it is heavily elliptical (and it is hard to be sure where general inference enters). A speech act, in this case an imperative, links (a) syntax: a missing subject and empty verb (you, verb, e.g. push), (b) semantics: illocutionary force relation, (c) phonology: focus intonation, (d) pragmatics: an open inferential variable referring to possible action in a context. Language-specific morphology (an imperative inflection) is possible in some languages, but the core connections are arguably universal for imperatives, and critical to early communication. Although animal communicative powers remain largely unknown, it is plausible to argue that Strict Interfaces are species-specific. Strict Interfaces can, we propose, allow the child to reject grammars that fail to have a tight link between inherently different representational levels. It stands in contrast to what we call Open Interfaces. ()

Open Interfaces: A representation that allows inference to offer many interpretations.

We consider the notion of Strict Interfaces to be implicitly entailed in proposals by Chomsky (, , ) that interfaces are a crucial dimension of UG. Open Interfaces link naturally to claims about pronominal representations for ellipsis, which have been widespread for over twenty-five years (Hardt ; Merchant , this volume; van Craenenbroeck ), and which we shall explore. Our goal is to show that the Strict/Open contrast is a natural part of UG that will predict the acquisition path and represent an available default structure for adults. The UG acquisition principle moves the child from a minimal representation (simple pro) with inferences to a maximal one (full syntactic and phonological representation) under the following principle: ()

Minimization Goal: Minimize pragmatic inference and maximize the information determined by grammar.

This leads to our general acquisition path claim: ()

Children will step by step replace a pronominal representation with an explicit syntactic representation and a fixed link to semantic representations.

We argue that such diverse features of grammar in a language are a necessary and valuable property of UG. Such a theory should both account for intuitional data and capture the acquisition path.

  



.. The acquisition perspective on representations, licensing, and recoverability Representations. UG allows diverse and incompatible representations for elided material which operate with different primitives. If the simplest representation can be an empty pro, while the most complex representation contains phonological, syntactic, and semantic information, then the acquisition path should exhibit the simpler forms first. It is essential to observe that alternative projections of VPE (simple pro and full PF) are not straightforwardly statable as a single rule with options. Where options exist, for instance in parametric choices, they are of the same type (+ or – subject, VO or OV). Therefore, since they are separate representations in a language (pro or full PF), but not collapsible, then, in effect, the language contains Multiple Grammars in miniature. We argue that both options are necessary to make acquisition succeed.5 Those options may in turn be reflections of deep typological differences, like the contrast between “hot” languages that allow contextual reference and “cool” languages that do not (Huang ) or NP and DP languages (Bošković , ). That is, some languages project bare nouns and a simple NP where others project a full DP. It should not be a surprise that the acquisition path is guided by access to Multiple Grammars rooted in different typologies. Licensing. Exactly what features must be present to license ellipsis? Merchant (a, this volume) argues that a specific E(llipsis)-feature must be present on a head. For the child, the input data should be immediately informative about what carries the E-feature given two UG assumptions: (a) licensing requires a head from certain maximal projections, such as TP, and (b) it may presuppose a moved head under some theories (e.g., Thoms b) but one which is UG-required. The choice of head can be very clear from the input. For instance, tense (e.g., ‘did__’), or ‘to__’. Early child uses of ‘I did’ suggest that licensing is quickly identified from the elliptical input (see more examples below). Where no UG-determined head is available, then the next step is to use verb subcategorization as the only alternative the input offers to know what is elided. Input evidence again should straightforwardly license Null Complement Anaphora (NCA), as in John liked to sing, Bill wouldn’t try => (to sing). Other verbs fail: *Bill wouldn’t like. Notably, NCA is linked to specific lexical items, not syntactic categories, which is a sign of limited generality in the adult grammar. Recoverability. It is clear that full PF recovery is ultimately necessary, which can be accomplished by traditional concepts of copy/delete. Our approach suggests that at earlier stages only minimal syntactic structure will be copied, just heads, and later full maximal projections. Finally, semantic abstractions of antecedent VPs at LF are added. While this acquisition theory is intuitive, there remain a great many empirical gaps, and occasions where children fall back on earlier (pronominal) representations.

5 There is now a substantial literature on Multiple Grammars (see Roeper , ; Yang ; Amaral and Roeper ) and comments and replies in the same volume (Rankin , ; Roeper , and others).





.. Exposition path: From theory to facts This chapter is organized as follows. We first look at the relevant acquisition logic that demands a heavily constrained theory of interfaces. We carry this out with an eye to the persistent role of general inference. Then we refine the concept of Strict Interfaces, the role of simplicity, and parallels with repair strategies in parsing. We propose an abstract version of the acquisition path it predicts. In section ., we first survey early evidence of non-adult ellipsis, disorders, and the first evidence for ellipsis. We argue that early stages are inferentially based. Then we turn to evidence about VPE acquisition in early Portuguese and then, in section ., to the role of a Strict Interface, based on Kennedy’s () Semantic Identity Hypothesis (SIH). ()

Semantic Identity Strict Interface: A recovered syntactic representation generates an LF form with variables that allow sloppy readings.

In sections . and . we turn to variants such as antecedent-contained deletion and sluicing which show that wh-movement may be involved in ellipsis. In section ., we turn to argument drop in Japanese and English and in section ., to NCA. Throughout, we argue for a pronominal representation to capture children’s behavior that deviates from adult grammar. Then we sketch a Strict Interface for noun phrase ellipsis (NPE) (e.g., ‘John has a big cake, and Bill has a small__’) in section .. Here, an LF representation allows definite reference to be imposed. ()

Strict DP Interface: LF for empty objects entails definiteness.

Then, we develop a Strict Interface based upon an LF that expresses a partitive relation, such that a sentence like ‘I want five’ entails the presence of a larger set of which ‘five’ is a subset. ()

Strict Partitive Interpretation: Empty NP is linked to an LF where a set/subset relation is projected.

These serve as examples of what should ultimately be a rich array of interface connections that significantly simplify the acquisition process and explain default resort options in adult grammar. Finally, in section ., we consider the potentially interfering and obscuring role of fragments and shifted Questions Under Discussion (QUD), which a bias toward Strict Interfaces should enable the child to accommodate. We argue that a pronominal alternative is a constant acquisition variant throughout these diverse structures. In sum, the very obscurity of ellipsis in conversation demands a theory of Strict Interfaces to explain how children arrive at highly restricted interpretations for highly complex and often rare structures.

  



.. Acquisition theory: Inferences create a learnability challenge A strong alternative to systematic interfaces has been proposed: Culicover and Jackendoff () claim that ellipsis can be completely governed by extra-grammatical general inference that can be coupled with any form of contextual parallel alignment—an ability not limited to language—to motivate ellipsis (assumed by many psychologists as well): Our approach thereby unifies the semantics of all of these phenomena under a common account that is based on a domain-general cognitive principle. (Culicover and Jackendoff : )6

In general, we know that minimal conversations often provide no basis for discovering a clear verbal antecedent for ellipsis, requiring intricate inferences. Consider the following conversation: ()

Dialogue at a newsstand: A: ‘New York Times.’ B: ‘A dollar fifty.’ A: ‘Ok.’ B: ‘Thanks.’

As we shall see, much dialogue with children works this way too. Such inferences could be and often are sufficient to substitute for actual reconstruction, as Culicover and Jackendoff (and also others) argue. However, their approach overgenerates possible interpretations. Our approach seeks to account for how possible and plausible inferred interpretations are excluded. If inferences can be sufficient, we have a learnability challenge: Why use intricate reconstruction if general cognitive inference suffices? The Culicover/Jackendoff approach reduces the “learnability problem” to virtually nothing since it is presumed to be inborn and obviously advanced in the behavior of any -year-old. They utilize cognitive parallelism to establish what an antecedent for ellipsis must be. They comment, “we also achieve a gain by eliminating the machinery of deletion under identity,” and ask further, “This might be considered a wash: Why prefer a new untested piece of machinery to the old more or less reliable one?” (Culicover and Jackendoff : ). This view has intuitive appeal to many and so it needs to be addressed.

.. The simplicity of mechanisms Culicover and Jackendoff ’s approach overlooks a fundamental concept of simplification. The answer to their question is: After it is acquired, a complex system of reconstruction can 6 They add “We compare our approach with accounts of ellipsis based on syntactic copying or deletion, showing that although both approaches have their difficulties, the challenges to the copying/ deletion approach are more severe” (Culicover and Jackendoff : ).





simplify mental computation. That is, it becomes easier to systematically reconstruct missing information syntactically or phonologically with an automatically rich representation than to undertake a complex guess based on context where a huge number of factors introduce a huge number of possible inferences. For instance, the precision of reconstructing a contrastive element inside a compound ‘HAND-made’ in () does not easily follow from general inference, but must follow from total reconstruction. To use an analogy, a very complex object like a computer can be designed for usesimplicity such as Amazon’s one-click ordering system. Likewise, once constructed, the grammar is easier to use than inference. The critical claim is that difficulty of construction—following the acquisition path—is compensated for by ultimate simplicity. To make such a model work, however, the system being built must know where it is going (just like a computer programmer has a programming aim for user-friendly machines). We argue that the child knows that Strict Interfaces should be identifiable in the language input.7 Once the grammar of reconstruction is established, it operates with such efficiency that it is always preferable to the available repetition of the VP. Looking ahead, even -year-olds, who begin with full repetitions (e.g., saying ‘I did eat’ after hearing ‘Did you eat?’) easily move to ‘I did’, ‘I didn’t’, ‘I can’, ‘me, too’ (Renato ). Over  examples of the fairly rare word ‘either’ with ellipsis are found with -year-olds on CHILDES (MacWhinney ). ()

(;) Adult: Child:

Do the animals stand up? No, these not either.

Therefore, prima facie evidence can immediately project missing syntactic material (these [animals] do not [stand up]) with the sophisticated semantics of negative either, although it is still not fully grammatical. Easy use of such kinds of ellipsis suggest that a sophisticated syntax/semantics Strict Interface is what the child is seeking. Nonetheless, a much more refined analysis is necessary before we can grasp an acquisition path for more unusual elliptical constructions.

.. The repair parallel In a parallel argument, Frazier and Clifton () argue that the parser repairs representations in the process of reconstruction which must likewise presuppose and prefer the same Strict Interfaces. For instance, ellipsis can ignore (active–passive) voice mismatches ‘hotdogs were eaten, but I didn’t want to’, inflectional variation ‘John is playing baseball, but 7

The idea that UG has an ideal or perfect form of interfaces which constrain acquisition marks a shift in acquisition history. Early theories always assumed that ‘performance’ obscured the child’s grammar. Most of the work has shown that there are few or no performance constraints that children obey. An example of early advanced knowledge is in Roeper (), who found that -year-old children grasped the control difference between ‘show the dog jumping’ and ‘show the dog the jumping (where a non-dog can jump)’, which entailed a tight connection between the syntax of determiners and the semantics of control. The experiment indicated that interface aspects of determiners must be represented very early: the semantic constraint on control in verb–noun conversion is signaled to -year-olds by the determiner ‘the’. They know that there can be control in ‘John likes singing songs’, but not in ‘John likes the singing’.

  



Bill won’t’, and some kinds of referential flexibility. Speakers, along a gradient, consider these worse than PF-perfect forms of reconstruction, as Frazier and Clifton () have shown in their theory of ellipsis repair. Therefore, we assume, in concert with the theory of repair, that Strict Interfaces must be presupposed (see Merchant b).

.. A pivotal role for inferences Nevertheless, rather than reject general inference, we need to incorporate it as part of the system that produces a plausible acquisition path. Again, we claim that at Stage  a child begins with general cognitive inference applied well beyond language, but which step by step replaces inference with grammatical reconstruction. This goal of minimizing pragmatic inference is built into UG. So grammars will vary in where they allow full reconstruction or continue to use pro. Two distinct traditions—pronominal representation and total reconstruction—have sought to characterize approaches to ellipsis without either proving completely adequate. Acquisition logic offers a different approach. Both systems are needed to account for how the child projects first representations and then refines them. First we need to look closely at how they contrast. It is obvious that a pronoun will work in an example like () where an inference or a reconstruction can arrive at ‘sang’. ()

a. Mary sang, then I did___ [=sang]. b. Mary sang, then I did that [=sang].

But note that a pronominal form, substituting either for the whole VP or a larger or smaller part, often allows other inferences as well. ()

a. Mary hated washing dishes, and so I did___[=hated washing dishes], too. b. Mary hated washing dishes, and so I did that [= hated washing dishes or washed dishes], too.

In (b) that refers to ‘hated washing dishes’ or it can just refer to ‘washing dishes’. Thus, a simpler grammatical representation introduces a plethora of inferential options. These can only be resolved by applying further inferences to select which is right, adding to overall mental computation. Experiments below show that children are susceptible to just such misinterpretations. In more complex environments, Merchant (this volume) and van Craenenbroeck () have argued that just such a pronominal form reappears for adults. Consider this case from Hardt () and Schwartz () cited by van Craenenbroeck: ()

I’ll help you if you [want me to x1]. I’ll kiss you even if you don’t [want me to x2] = even if you don’t want me to help you; or = even if you don’t want me to kiss you.





He observes (: ) “ . . . the second VPE-site (x2) takes as its antecedent the bracketed [VP in [VP]] in the antecedent clause” [= a variable inside a variable].” Merchant suggests that the second site should be filled with just do that, and Craenenbroeck argues that “the second ellipsis site is pronominal in nature.” Göksun et al. () provide a suggestive acquisition experiment where children were given both: John wanted to make pancakes, but Bill doesn’t want to__ or Bill doesn’t want some. In both instances they chose a picture of pancakes rather than an action of cooking, suggesting a missing pro-N rather than a VP, which is compatible with the argument and experiments below. The pro option should not be a random alternative, but a predictable option. In a nutshell, it means that if there is a complex case, involving discourse or quantification, with linked VPEs, then UG may allow reversion to a simple pro at the second ellipsis site with new inferential demands. If the child’s grammar is less advanced—say, in knowledge of quantification—the reversion to pro could happen in other grammatical environments. While the seemingly intractable nature of these theoretical conflicts has led to proposals that there should be hybrid theories (see Merchant this volume and van Craenenbroeck  for a summary), we argue that what seems like a hybrid reflects a layering of systematic representational options so that the term “hybrid” no longer applies. Instead, the availability of these options allows an acquisition micro-path that is just coming into view.8 We propose a rough path through these options. ()

a. Pure inference: A word is like an object in a visual array b. Fragments: Inference, then Fragment = syntactic topicalization (see section ..) c. Anaphoric reference: Ellipsis = parallel V or N in previous discourse (equivalent to anaphoric pronouns) d. Full syntactic reference: Full DPs and VPs reconstructed e. Full syntactic and semantic reference: LF projections with variables allowing systematic sloppy readings

These proposals remain programmatic until a magnified vision of the acquisition path can be articulated, but they provide the basis for a research program that looks carefully at each form of ellipsis.

For instance, it could be the case that sloppy identity was not available for (i) but first appears with quantificational and contextual support as in (ii). 8

(i) John saw his mother, and then Ben did___. (ii) John saw his mother, and then every boy did___.

  



. E : W  ’    ?

.................................................................................................................................. Every conversation demands that we add unspoken information. What do we (or a child) think when we hear ‘oh dear’ or just ‘oh’ or ‘uh-oh’? ‘Uh-oh’ is often a child’s first word. They are different from ‘um-hm’ (assent) or ‘un-unh’ (disagree) or just ‘mhm’ (understood), all of which require a verbal context (generally). A child’s life is full of expressive utterances like these. By their fourth year, children control dialogues where both nominal and verbal ellipsis are carried along across lengthy exchanges. Here is a typical dialogue (), with a -year-old (Wijnen et al. ): ()

CHILD: I drink it all up. CHI: give me some more. CHI: a lot. MOTHER: I don’t see any more. CHI: yes, you do. MOT: want a little milk? MOT: want some? CHI: (a)n(d) shake it all up. CHI: a bigger one? MOT: mmhm.

It has noun phrase ellipsis (NPE), ‘more’, and verb phrase ellipsis (VPE), ‘yes you do’, which reproducing the dialogue must have NPE inside it (i.e., ‘yes you do see more [N]’). The ‘it’ is never revealed, and even ‘mmhm’ requires reconstruction of elided information. The -year-old has mastered a great deal beyond what a -year-old knows.

.. Earlier evidence of non-adult ellipsis Very early child moments, when the child seems to assume ellipsis, often fail to communicate. ()

FATHER: do you want them [cheerios] with milk? CHILD: no. CHI: on it. CHI: on. FAT: what on it? CHI: on. CHI: milk in it. FAT: what do you want # Naomi? (MacWhinney )





The child implicitly assumes a broad inference that gets to ‘I want milk on it’, while the adult needs syntactic precision and never figures it out. A look at other failed dialogues reveal children trying to use ellipsis inadequately or inappropriately, which give the reader a sense of frustrating real child dialogues (Jensen and Thornton ). () a. (;) CLINIC (CLI): what else shall we buy at the store? CHILD (CHI): tomato store. CLI: where shall we put sock? CHI: on. CLI: what is the bird doing? CHI: on leaf. b. -year-old CLI: what do you do in an airplane? CHI: up in the air. c. -year-old CLI: do you want to let the people out now? CHI: come a bus. [= here comes a bus for the people to get into] CLI: what are you going to do with these? CHI: up here. CLI: where did you take the people in the truck? CHI: go doctors. people dead. ()

older child: -year-old CLI: what’ll we use these spoons for? CHI: eat.

There is no doubt that some representation from the previous sentence is presupposed, but its grammatical characterization is obscure. Often a verb is presupposed ‘(put) on leaf ’, ‘(put) up here’, or ‘(go) up in the air’, or silently projected in a way that adult grammar does not allow.

.. First verb phrase ellipsis: Where should it be? We look first for early stages where UG principles can be directly reflected. We predict: Where ellipsis primarily reflects a straightforward UG principle, then it should be triggered immediately for very young children, with minimal evidence. In fact, young children perceive and use VPE in both Portuguese and English (see section ..), even though VPE is not universal. Although VPE is language-particular, a dialogue provides dramatic evidence. When someone says ‘I did’, both adult and child see immediately that something missing must be filled in. Given the UG principle that the immediately preceding head must license ellipsis,

  



acquisition is instant with no delay or detour, allowing some form of reconstruction of the missing complement. In addition, as Aelbrecht () shows, some non-adjacent cases must be accommodated, which we can predict will be more challenging. The adjacency claim does not guarantee, as we shall see, that what is reconstructed is entirely adult; it may still contain an invisible pronoun. A cursory survey of ten children between ; and ; years for their use of ‘did’ reveals that in every instance but one, an earlier tensed sentence occurs before the child’s VPE with ‘did’. ()

MOT: gon (t)a take the seed out? CHI: yeah. (child takes seed out) . . . I did.

() MOT: dolly-’has done a boff? CHI: yeah. MOT: she didn’t. CHI: Andy did.

(MacWhinney )

While I did by an adult provides sufficient information to know that an empty element is present, it does not reveal how richly structured the missing information is. That is, we satisfy the licensing requirement, but the representation is not automatically adult. It could be a general pro projection with little structure, as our model suggests. Jensen and Thornton () show that twenty children (;–;yrs) control these structures (‘I did’) early but there is an interesting dip in their use of VPE as follows. At first we have wh-questions like below. () Adult: what else shall we buy at the store? Nina: tomato store. The simple answer ‘tomato’, all that a wh-question requires, arises only when children have sluicing sentences like ‘I want to eat, but I don’t know what’, which exhibits A-bar whmovement and ellipsis (see section . on sluicing). This correlation provides a hint that when children first use VP-ellipsis, a representation like a simple pro could be present, but when sluicing arises with wh-movement, it triggers a richer projection into the elided position. This grammar shift for the child can explain the dip in ‘I did’ before sluicing arises: children will briefly avoid systematic ellipsis when they are unsure of what should be projected, pro or a full VP.

.. VP-anaphora In contrast to I did, an expression like I did it occurs as explicit VP-anaphora (including so-called deep anaphora) in a child’s experience. It is found in early files (a). At the same age we find cases where an explicit pronoun it is linked to a contextual equivalent—an action (b):





() a. (;) CHI: cow won’t go Mama. MOT: oh. [action: the child is trying to put the cow in the barn] CHI: hey, I did it. b. (;) CHI: I did it. MOT: what did you do? CHI: put my shoes on. (MacWhinney ) The child knows that if no previous VP is available, then the pronoun must be overt, and verb phrase anaphora is required. Although such a study should be done more carefully, the evidence points to the notion that VPE occurs only when a tensed IP is present, which is immediately triggered through child/adult conversation. By contrast, so-called deep anaphora calling for an explicit pronoun ‘did it’ triggers direct reference to context.9

.. Early Portuguese verb phrase ellipsis In her careful work on early Portuguese, Santos (b) finds  percent ( out of  utterances) must be elliptical examples, indicated by the correct tense marking on a single word, echoing the English facts, thus evidence clearly compatible with the view that Tense is a licensing head.10 () a. Tomás (;.) ADT: olha # são legos? look are Lego ‘Look, are these Lego pieces?’ TOM: são. are ‘(=Yes.)’ b. Inês (;.) MAE: estás lhe a dar colo? are -  give lap ‘Are you putting him in your lap?’ INI: (es)tou. am ‘(=Yes.)’ Reasoning across English and Portuguese, the licensing principle involving functional categories for ellipsis is apparently available before wh-movement. This is what we would 9

See also Duffield and Matsuo () for experimental evidence that children control this distinction with older children. 10 Other examples in Santos’ collection are ambiguous, raising the possibility that non-Tense elements could license ellision (see also Thornton ).

  



naturally expect if ellipsis is at the core of conversation, but wh-movement is acquired later because of language-specific phenomena.

.. The movement requirement Thoms’ (b) analysis involves an important extra step and a strong constraint (arguably too strong):11 ()

a. Non-moved elements will not license ellipsis.

Thus if an element, like an auxiliary (could in (b)), moves to a tense domain, then the TP can carry an E-feature which now allows ellipsis: () b. [John [T couldi [vP ti sing]]] and [you [T couldj [vP __]], too]. Failure to move blocks ellipsis because no functional category licenser occurs. Progressives in English do not move, staying in the lexical vP, and therefore do not allow ellipsis. It is difficult to prove absence, but no one as far as I know has reported or seen cases like (b) in the CHILDES database: () a. John is being helped, and Bill is __ , too. b. *John is being helped, and Bill is being __, too. It is raised to a T-position, but the progressive being is projected without movement under an Aspect node. Since being does not license ellipsis, we correctly predict that we find being only with a predicate in acquisition data: () a. being on swing. b. going being careful.

(;.) (;.) (MacWhinney )

Out of  examples on CHILDES, none ends with being and ellipsis, and likewise out of roughly  examples of having none occurs with ellipsis. Evidence of such subtlety and precision obtained from very young children should be seen as signal support for Merchant’s theory of E-features and Thoms’ (b) theory of movement (also Lobeck ). Nevertheless, Aelbrecht and Harwood () discuss evidence that the sub-features on the E-feature can vary from language to language, making the fact of early recognition more remarkable, since it means that the child must be sensitive to specific input as well as UG consequences.

11

K. Johnson (p.c.) points out that NPE lexical exceptions are not obviously moved (‘I don’t want any’), or, as J. van Craenenbroeck (p.c.) observes, obligatory ‘to’ is not obviously moved.





. V     

.................................................................................................................................. If we now ask about the recoverability of meaning when binding is involved, it calls for a more elaborate syntax/semantics interface, and the role of Strict Interfaces become critical. We utilize the formulation by Kennedy (), the Strict Identity Hypothesis (SIH), where LF is shorthand for logical meaning: ()

Strict Interface = Strict Identity Hypothesis Deletion of VP1 is licensed at PF (phonetic form) only if there is some VP2 in discourse such that VP1’s LF (Logical Form) is identical to VP2’s LF.

We claim this is the Strict Interface that the child seeks; he knows an LF correlate should be projected. It is commonly represented as abstract copied syntactic categories together with linked lambda abstractions (which allows subjects to be treated as abstract extracted variables (x), so that subjects can be switched in the ellipsis site [Oscar => Bert] in ()) if they are each equal to the variable x as Foley et al. () demonstrate in the following experiment.12 Consider, for example, (), a scenario where the child could either select the banana from Oscar’s plate for Oscar to eat, or select the banana from someone else’s plate. The child could retain that object for Bert to eat or could select another banana. ()

a. Oscar bites his banana and Bert does, too. b. Oscar [VP λx(x bites his banana)] and Bert does [VP [V e ] [NP [N e ]]] too.

They provide sharp evidence for children’s representation of bound variables with a series of experiments systematically testing pronominal, definite, indefinite, and full NP variations with children from  to  years. ()

Oscari bites his apple and Bertj does too. i = subj, j = second subject, k = rd person external

()

Sloppy interpretation for () O bites O’s apple and B bites B’s apple. (ii jj)

()

Strict interpretations for () a. O bites O’s apple and B bites O’s apple. (ii ji) b. O bites B’s apple and B bites B’s apple. (ij jj) c. O bites E’s apple and B bites E’s apple. (ik jk)

12

Foley et al. () have a similar view that lambda abstraction is favored in order to preserve a form of economy.

   ()

Ungrammatical interpretations for () a. *O bites O’s apple and B bites E’s apple. (ii jk) b. *O bites B’s apple and B bites O’s apple. (ij ji) c. *O bites B’s apple and B bites E’s apple. (ij jk) d. *O bites E’s apple and B bites O’s apple. (ik ji) e. *O bites E’s apple and B bites B’s apple. (ik jj)

()

Response rate of sloppy reading a.  years (N = ) % b.  years (N = ) % c.  years (N = ) % d.  years (N = ) %



It is clear that the sloppy readings are the preferred option from very early, and all the other options in () are excluded by  years.13 Still one third of the -year-olds do not exhibit sloppy identity and an acquisition path may be involved.14 In what follows, we propose that Strict Interfaces are at the root of the explanation.

.. Determiner deletion Abdulkarim and Roeper () provide further sharp evidence that favors the presence of a binding variable.15 Consider these results, which are distinctly non-adult, from an act-out experiment by Abdulkarim and Roeper (): ()

John pulled John’s truck and so did Fred. ‘John1 pulled his1 truck and Fred2 pulled his2 truck.’ (= John λx.[x pulledj truckh] and Fred λx.[x tj th]) (cf. Adult: ‘John1 pulled his1 truck and Fred2 pulled his1/*2 truck.’)

Thirty-five Arabic- and thirteen English-speaking - to -year-old children essentially drop the explicit possessive (i.e., John’s truck) and impose a bound variable reading (i.e., x pulls x’s truck (Fred pulls Fred’s truck)); that is,  percent of the Arabic children and  percent of the English children. ()

Arabic version: Fawaaz rafa? dgeib Mansour we Mohammad ba?ad. Fawaaz lift.. jeep Mansour and Mohammad too. ‘Fawaazi1 lifted Mansour2’s jeep, and Mohammed3 lifted his3 jeep, too.’

13 Work by Syrett and Lidz () as well as Thornton and Wexler () supports this view, as well as Kiguchi and Thornton () and others cited therein. 14 See related phenomena in Bever () and Roeper et al.’s () demonstration that children prefer “distributivity”. 15 The status of all the sloppy-identity results relates to some degree to further evidence of quantifier spreading. See Philip () for a summary, and Roeper et al. ().





Like English-speaking children, Arabic-speaking children tend to ignore an explicit antecedent (i.e., Mansour’s jeep) and instead reconstruct the VP as having a bound variable reading (i.e., Mohammed’s jeep). This is consistent with the notion that children prefer binding variable identity even over a simple PF copy with a strict interpretation of a filled possessive. The fact that they will impose the bound variable reading by altering the representation antecedent ‘Bill’s truck’ to ‘Fred1 pulls his1 truck’ fits the claim that they are seeking a Strict Interface like the SIH whenever there is possible contextual support. Why do adults not allow this default as well? The children seem to be able to delete a possessive determiner where the adult English speaker requires it. The answer is not obvious, but if the parametric choice of DP over NP is not fully realized yet, then it is possible that the child retains the NP option with no determiner which can fulfill the Strict Interface for a bound reading, and this motivates determiner deletion in comprehension.

.. External reference in VPE The matter is not quite settled, though. Vasić, Avrutin, and Ruigendijk (a, b) gave children sentences like (), where the grandfather scratched “a third external person like the uncle’s dog,” quite at odds with adult English, but not Japanese (as we shall see). ()

Scene: boy scratches his own dog, grandfather scratches his own dog, and grandfather then scratches the uncle’s dog. ‘The boy scratches his dog and the grandfather (does), too.’

When given a choice between grandfather or uncle,  percent of Dutch -year-olds took the alternate contextually available reference ‘uncle’ rather than the binding variable reading ‘grandfather’; when given a choice of ‘boy’ or ‘uncle’,  percent took the alternate reference ‘uncle’. This interpretation is also compatible with the view that the child reconstructs does that with a pronoun that allows broader inferences. ()

The boy scratched his dog and the grandfather did that, too.

Acquisition of more complex structures with antecedent-contained deletion points in the same direction.

. A- , ,  

.................................................................................................................................. Our central claim that VPE manifests a Strict Interface between syntax and semantics can be further complicated by a prior operation: quantifier raising. Sentences like (a) entail an infinite regress if reconstruction is carried out:

   ()



a. John read every book that Bill read . . . b. John read every book that Bill read [every book that Bill read [every book that Bill read . . .

The solution is to raise the quantifier beforehand and leave only the verb to be copied.16 ()

[every book that Bill did [read t] [ John [read t]]

It involves LF extraction to a higher node prior to VPE. In that sense, it is a complex interface between syntax and LF, and therefore we can expect a more complex acquisition path. In an elegant experiment Syrett and Lidz () demonstrate that -year-olds carry out both quantifier raising (QR) and VPE, supporting again the UG-guided nature of the process. ()

a. Miss Red jumped over every frog that Miss Black did [jump over every frog that . . . ]. b. Subset of frog = {Frog , Frog , Frog , Frog }

The correct answer is that Miss Red jumped Frog , ,  but not Frog , therefore Miss Black jumped over the same subset of frogs. Syrett and Lidz show that QR must be an available UG option because children apply it at the age of  years, despite the rarity of antecedent-contained deletion (and other QR environments). In particular, they showed that in () children allowed QR to raise both to an intermediate position with embedded scope ‘every book that Goofy read’ or to matrix scope ‘every book that Scooby wanted Goofy to read’. This is important supportive evidence for the presence of QR, but to ascertain that the children have the adult grammar, they must also exclude what adults exclude. The sentences in Experiment  were ambiguous, such as: ()

Clifford wanted Goofy to read every book that Scooby did __ [= did read/did want Goofy to read].

However, the sentences in their next experiment are typically reported to be unambiguous. This lack of ambiguity is classically taken as evidence that QR is clause-bounded. ()

Clifford said that Goofy read every book that Scooby did __ [= that Scooby read/6¼ Scooby said that Goofy read].

16 Sugisaki (a, b) with intriguing but limited evidence suggests that swiping is also available just over  years, which is possibly connected to pied-piping or P-stranding, again suggesting very subtle knowledge, very early:

Aran (;) (i) (ii) (iii) (iv)

What in? Who for? Who from? What with?





Notably  percent of children allowed exactly the wide-scope QR over tensed complements too, which adults reject (b). ()

a. [= Scooby read every book that Goofy read.] b. [= Scooby said Goofy read every book that Scooby did.]

As Syrett and Lidz point out, this violates subset theory; learnability is maintained only when you expand the set of readings from evidence—the absence of negative evidence makes it impossible to restrict the set of possible readings (see Fox ). They make a variety of suggestions about parsing limits or discourse pressure to explain why children do this. In contrast, we claim that the structure of UG itself should accommodate this variation: Strict Interfaces, once fully applied, will eliminate these default readings. If again, we claim that the child uses ‘and Scooby did that too’, then a wide range of interpretations become possible, including ‘Scooby said that Goofy read . . . ’. The idea that general inference is invoked is supported by the fact that the meaning reconstructed is more, not less, complex than the adult grammar delivers, and it incorporates as much of context as possible, that is, the larger context of saying, not just the narrower context of reading. Supporting this approach is the fact that adults do precisely the same in a significant minority of cases, as we predict under the default pro option:  percent. Moreover, they do something similar in the original experiment. Syrett and Lidz added a further control to see if a simpler form of coordination would allow pure VPE, which is in the spirit of recognizing conjunction as a default alongside the Strict LF we argue is another default, which would translate (a) into (). ()

Miss Red jumped over every frog and Miss Black did.

Their adult controls seem to have made such a substitution. They comment as follows: () “Adults occasionally interpreted every frog as referring to every frog in the entire context, not every contextually appropriate frog.” This result follows from a larger theory of defaults where conjunction or parataxis plays a central role and which is supported in work for over forty years in acquisition, parsing, and fieldwork (see work on recursion in Amaral, Maia, Nevins, and Roeper  and others). The parallelism would restore the full VP without the operation of raising. The imposition of a default form of conjunction is another example of how principles of grammar continue to govern behavior that is sometimes seen as “performance errors”.

. S

.................................................................................................................................. Sluicing entails both wh-movement and ellipsis. We noted Jensen and Thornton’s () claim that wh-answers were correctly minimal (‘what did you eat?’ ! ‘tomato’) only when children showed knowledge of sluicing which involves both wh-movement and ellipsis.

  



In fact, early instances of sluicing include those from - to -year-olds where there are hundreds of ‘know what’ examples. () GAI: what did they look like? CHI: um # I can’t tell ya . . . I don’t know what.

(MacWhinney )

However, other wh-words such as why and when are also observed. () a. (;) CHI: an(d) let-’us turn around. (action: CHI moves one side of playhouse closer to the other) CHI: you know why? b. (;) CHI: I just left it in one day. FAT: when? CHI: I just don’t know when. c. MOT: want to put it off? CHI: no. CHI: can’t you work this thing? CHI: I know why. (MacWhinney ) The case presented in (c) involves more because the child must convert his question into an elided declarative ‘I know why you can work this thing’, a topic to which we return. Jensen and Thornton claim that simple VPE I did (pro) is available very early but then shifts when sluicing arrives. That is, sluicing entails a full reconstruction of the complement. When it becomes available, then I did pro is shifted to a full VP: I did [VP]. These hints from the naturalistic data are buttressed by experimental evidence from Lindenbergh, van Hout, and Hollebrandse (). They demonstrate that -year-old Dutch children comprehend sluicing in a contrastive pragmatic environment. They contrasted pictures where a hidden person painted a guitar and another painted a flower, and yet another was holding a flower, then asked: ()

Someone is painting a flower, but I can’t see who ___.

The accuracy was around  percent: the children clearly understood sluicing (only the person painting the flower is intended) and, moreover, they generally produced it as well in an elicited production task. These subtle interactions indicate strongly that children, while they can project a broad pronominal, proceed not only to an embedded IP but to a very precise CP representation linked to a sluiced complement.17 17 Like Thornton and Jensen’s () claim that wh-movement in sluicing is linked to VPE, Foley et al. () argue that both sloppy and strict readings involve long-distance wh-extraction which accounts for the fact that children acquire strict and sloppy readings at the same time.





.. Verb phrase ellipsis summary In sum, child external references in elided VPs can go beyond what adults allow, which suggests that the apparent binding variable reading as well could have a different pronominal origin that does not require the parallelism of a binding variable at LF entailed in Kennedy’s () SIH, but is just a pronominal choice. How does the child ever get rid of this non-parallel external reference reading? The pronominal path involves an Open Interface, supplemented by inference. Our answer is that a Strict Interface when recognized will overrule the Open Interface and impose full parallelism for VPE and therefore must be biased against any possible non-parallelism that an Open Interface entails.

. F     NP/DP 

.................................................................................................................................. If the argument for a pronominal instead of VP works to explain departures from adult VPE, then we should expect it to be present when just an object, which is less than a VP, is elided. While some grammars allow missing objects which can be filled in by context, in English we have what seems to be a process of intransitivization that allows an indefinite reading, where the object is an unprojected argument that is automatically unspecified, as in ‘John wants to eat’ = eat something. A non-existent empty object could be where the acquisition path starts, but not necessarily. While Saito () argues that there is complete argument drop which effectively converts the sentence into an intransitive, Hoji () argues that some form of pro is available for Asian languages. Huang () provides a typological backdrop. He argues that there are “hot” and “cool” languages where the former give priority to contextual reference, hence favoring the projection of a pro that allows definite reference. That is, in many “hot” languages you can say (a), while in English one must say (b): ()

a. Can you give me— [⇒ contextually supplied]. b. Can you give me that.

Languages like English block an elided object, while it would be pragmatically superfluous in Chinese. Pérez-Leroux et al. () argue at length for the presence of an object pro in early grammars that leads to definite reference. They use a wide range of evidence under a principle of Default Transitivity. A sharp example comes from Pérez-Leroux et al. (). ()

Look the doggy is chewing the book, so Emily is not reading. Is that right?18

18

Sample picture from Pérez-Leroux et al. ().

  



If the child uses an object pro, then it should refer to discourse or context and book is the clear, relevant, salient topic.19 The answer would be “no she’s not”, because she is not reading the book, but if an intransitive or an unspecified interpretation is projected, then the object can be anything in general, so the answer should be “yes she is”, because she is reading the newspaper. Thirty percent of English children, in various experiments, answer “no” suggesting object pro. Pérez-Leroux et al. claim the principle of Default Transitivity leads to object pro.20,21

.. Japanese argument drop The pro option leads to something deeper: a parametric contrast between NP and DP grammars explored by Bošković (, , and references therein). Languages that allow empty objects allow empty determiners. A fundamental issue is: a child must decide if determiners will be present or absent. Is it an NP or a DP grammar which separates English and Japanese? Matsuo () tested argument drop with both English and Japanese children using sentences like (), which children were asked to confirm or deny.22 Fourteen English children between ;–; years and nineteen Japanese ;–; years heard this scenario. () Scenario: Bear finds blue fish and tiger finds pink fish. Prompt: ‘The bear found a blue fish and the tiger did, too.’

19

L. Amaral (p.c.) reports that this kind of misunderstanding is very common in L English speakers whose L is Spanish or Portuguese. 20 We will not address here the complex interaction with clitics and the large literature on object clitic omission. 21 See Roeper (, ), Yang (), and Amaral and Roeper (), among others. 22 See Sugisaki (forthcoming) for further evidence of argument drop in early acquisition and a discussion of the complex agreement facts that are a putative part of an ellipsis parameter.





Japanese adults, Japanese children ( percent), and a significant subset of English children ( percent) say “yes”. In Japanese, the main verb is repeated and the grammar fits the Pérez-Leroux et al. account of a null object pro. ()

The bear found a blue fish and the tiger found pro, too.

This produces the indirect potential for apparent VP-anaphora if pro has an open interpretation. The Japanese children used an open pro substantially more than the English children which would be compatible with the view that the Japanese children reconstructed just the noun ‘fish’, which is possible in Japanese but not English. The  percent English response remains significant and suggests that they could still have access to that grammar type as well, but it is not the primary grammar. It is another indication of the presence of Multiple Grammars. In another context where the object was totally different, Matsuo found smaller but significant groups in both languages ( percent English,  percent Japanese) said “yes” in this scenario. () Scene: Bear eats asparagus and tiger eats carrots. Question: ‘The bear ate the asparagus and the tiger did___, too. Is that true?’ This result is again not surprising if the children actually generate an empty pro (did eat pro = eat something), which can be interpreted through context. However, adults do not do this in Japanese, preferring the bound-variable reading. If they choose full VPE identity, the transitive reading is required. The fact that there is a minority of child responses in both languages for non-binding variable interpretation suggests that both transitive and intransitive grammatical options could be available to children.

.. Strict Interface version of argument drop We have examined the Japanese option under the notion that a pro is projected. Could there be a Strict Interface at LF as well? Bošković () argues that argument deletion is directly linked to formation of a semantic at LF which in turn permits sloppy identity. This is presented as an alternative to universal pro in this position. We will just sketch how acquisition connects to this evolving perspective. Bošković shows that languages with no DP permit argument drop which in turn allows sloppy readings. He argues that it is directly linked to an LF-moved NP with the semantic properties of converted into at LF. In his words (p. ): “Simplifying somewhat, in the syntax itself argumental traditional NPs are then of type e in DP languages and of type in NP languages. D turns NPs of type to in DP languages; while in NP languages this is accomplished via type-shifting.” Then he claims: argument ellipsis affects elements of type .

  



To which he adds the further claim that: only elements of type can be copied in LF.

In other words, an automatic LF operation does for NP languages what DP does explicitly in the syntax for DP languages.23 While the purpose of his claim is to capture sloppy readings, we suggest that it could extend naturally to capture the definiteness property entailed in contextually supplied objects. We will not explore the details here but simply assume that a parallel argument for LF interpretation leads to the capacity for reference to contextually salient objects if a child has an NP-grammar available. Therefore, children, if they are in an NP-phase or retain that alternative, could spontaneously project an LF even though there is no triggering evidence from adults in their own language. This would then justify saying “no” in PérezLeroux et al.’s () experiment because the children have projected an LF which permits the definite reference. This is then the exact same kind of Strict Interface correlation we found in full VPellipsis, which entails an LF representation under Kennedy’s Strict Identity Hypothesis, and which we argued could be spontaneously applied because it is a UG goal.

.. Reproducing the VP acquisition path in VP-ellipsis? Why would English children who themselves use articles, other determiners, possessives, and adjectives not reconstruct them in VPE? An intriguing hypothesis is as follows: ()

Children reproduce their initial acquisition path for VP in generating possible ellipsis structures with VP.

Children all go through an early stage where maximal projections are absent in production and they use just heads [V [N]] instead of [VP [NP]]. They first analyze “want cookie” as [V [N]] not as [VP [NP]]. That option would lead again to the following. ()

Bear found a blue fish and tiger did, too.

The VP ‘found a blue fish’ can be analyzed as [V [N]] = [find fish] at the ellipsis site if it fits the context. Thus, they might begin VPE reconstruction with minimal representations as well, in effect reproducing the initial acquisition path where first they use V but not VP and N but not NP or DP. Moreover, the bare object representation continues to be needed in adult English as an input to compound formation: make cookie => [[cookie-maker]]; a cookie cannot be incorporated *[[a cookie]-maker]] (see Roeper and Siegel ; Harley ). Therefore, it should not be surprising that children would retain the non-Maximal Projection grammar as a possible default which applies in VPE contexts. 23 Bošković () adds: “It should be noted that Tomioka () (i.e. his property pro) is an important predecessor of the analysis argued for here, which situates the gist of Tomioka’s proposal within a broader perspective.”





.. Triggering full VPE The full DP form of VPE would have to emerge. What forces it? It could be because, as we have suggested, the acquisition device seeks a Strict Interface which would be a fully parallel VP in the ellipsis site: faithful reconstruction of the VP would entail the full DP form. Again, this ideal Strict Interface is playing a role in learnability by leading to the rejection of forms that fail to honor parallelism. Nevertheless, the child needs an experience which forces the revision of grammar. Suppose the child hears this question/answer between adults. ()

(Bill chose a blue sweater) John chose a red sweater, and Bill did__, too. Is that true?

If the adult says “no”, then this would be evidence that simply reconstructing sweater would be actually wrong, since he should have said “yes” if that were even an option in his grammar. Now the Japanese NP option is blocked. Note again that this approach is independent of statistical frequency and depends on the dialogue experience—more likely than one might imagine—as children naturally pay attention to the focus of adult conversations. Such triggers should be carefully imagined for all complex grammatical decisions.

. N  

.................................................................................................................................. A host of exceptions to strict reconstruction dominate the current literature (van Craenenbroeck ). Many of them are far too arcane to be obvious triggers a child would use to change their grammar. Others, however, as some argue, remain central to grammar and therefore could be the basis upon which the child—or the adult—organizes their entire grammar. Consider NCA. It involves contrasts like (): ()

John didn’t want [to wash dishes]VP1, a. *but Mary wanted/decided/was required. b. but Mary volunteered, consented, refused, tried, offered pro1.

(a)—which a child would never hear—lacks the licenser to but (b) works without it, by projecting pro into the ellipsis site and coindexing with VP1. Therefore, no isomorphic reconstruction site occurs and no precise functional category licenser is necessary.24 How 24 Weir (p.c.) pointed out that it is possible to say (i) but we note that it does not work with all verbs as shown in (ii).

(i) John can eat whatever he wants. (ii) *John will be able to eat whatever he hopes. cf. John will be able to eat whatever he hopes to We note that (i) with whatever only works with verbs that independently allow an object: wants food/ *hopes food.

  



does the child keep NCA limited to particular lexical items? The child must not decide from a few examples that a licenser like ‘to’ is never necessary because that would generate (a). The early evidence on ‘try/want’ (see below, this section), but no other verbs, suggests that they do in fact avoid a broader generalization to all verbs when they hear NCA. We can capture this pattern if we assume that UG guides the child to limit the default options to specific lexical items wherever possible. In this case, the option of not reconstructing an ellipsis site in full should make it a first preference for a child because it looks representationally much simpler. We noted above that children never allowed ellipsis with ‘being’, but we do find null complement anaphora with ‘trying’ and we find hundreds of cases among -year-olds of all the options ‘try’, ‘try it’, ‘try to’, and ‘start’. This suggests that they honor lexical restrictions from the start: ()

a. b. c. d. e. f. g. h.

but they was trying. them was trying. (;.) let’s try again. (;.) my try it again. (;.) just try. (;.) I try not to. (;.) try that here. (;.) yep I’m starting. (;.) let’s start. and then start up here. (;.) (MacWhinney )

If -year-old children have both VPE with strict reconstruction and the anaphoric option, they must be defined in relation to each other in order to prevent overgeneration. Why should we have a difference between ‘refuse’ and ‘attempt’ (i.e. ‘he refused’ vs *‘he attempted’ only ‘he attempted to’)? Note that L speakers easily say incorrect things like ‘do you want?’. It is notable however that we simply do not find stranded want in the database for children. This leads to the conclusion that NCA seems to depend on specific verb-by-verb triggering to which the child is sensitive, if not the L speaker. This is not surprising: anything that is lexically specific can allow the presence of another grammartype within the language, just as pro-drop is restricted in English to a few verbs like ‘seem’, ‘look like’ but not ‘happen’, or ‘matter’. Again, if we find an evolution in grammar from pro to more precise reconstruction, we can maintain the view that both options are independently in UG, not written into a single form of VPE. That would lead to this conclusion: NCA is a potentially lexically restricted reflection in the adult grammar of an earlier stage in acquisition which has roots in another typology where to is not necessary.

. N  : P

.................................................................................................................................. Our approach will gain force if we can show that NPE also has a pronominal alternate and a Strict Interface at LF. If correct, our first hypothesis should be that NPE may exhibit a syntactically simple pro at the first stage. Whether one projects pro or a copied/deleted





trace-NP, we can ask: is there an LF goal which children seek to impose on ellipsis as a Strict Interface? A number of scholars have argued that children’s NPE is linked to partitivity (see Sleeman and Hulk  and references therein) which again may be linked to an LF projection.

.. Early evidence of partitivity What is the path for NPE? Among the first dozen words of a child is often ‘more’. While it is hard to prove that there is an invisible N [more [N]] rather than a broadly construed Expressive about an Event, an [N] could be what one would expect if the child pursued a direct relation between invisible information (i.e., pro) and context. If the child projects empty [N], how rich is the representation in the ellipsis site? It is not just anything for adults. For instance, from discourse it must be the highest NP. Take the following sequence. ()

You drank some juice next to the cake. Want more?

This refers only to juice, not cake, because only the highest NP1 is relevant in [NP1 juice [PP next to [NP2 cake]]]. An anti-pragmatic experiment could test this, but has not been done: ()

There are some nails next to the cake. Want to eat some?

This question asks only if you want to eat ‘nails’ not ‘cake’, while cake is the pragmatically natural alternative. Only when a child can show this knowledge do we have clear evidence of syntactic reconstruction that the child obeys despite an anti-pragmatic context.

.. Licensers: Inflection, possessives, agreement The second question is what licenses NP ellipsis. In English, we find comparatives, numerals, and possessors will license it, and arguably each of those concepts—degree phrase, numeral phrase, possessive phrase—can introduce a functional node which carries a licensing capacity. ()

a. I want more— [NP]. b. I have one— [NP]. c. I like John’s— [NP].

Sugisaki () has done a study noting that children never say *‘five ones’, which suggests that from the outset ‘one’ itself is in a numeral phrase, not a possible lexical item in an adjective phrase, which she might think if she hears the one hat. If ‘one’ were an adjective, we would predict that other adjectives could appear, but they do not for either adults or children. ()

John has a toy. *I have a blue __ [=toy].

  



Under Merchant’s () approach, each licenser must carry an E-feature on a functional category that specifically licenses a deletion operation. In fact, the possessive licenser seems to be extremely early. Jensen and Thornton note possessive ellipsis from a -year-old: () (;) Mother: Whose hat is that? Nina: Mrs Wood’s___ [=hat]. A more interesting case is African-American English (AAE). Conner () has shown that AAE children, whose exposure to ’s is more limited, consistently project ’s in ellipsis environments while it is absent otherwise. That is, children even say recursive things like (a), or even phrasal cases like (b) which involve phrasal possessives. () a. They Mama room. (= their Mama’s room) b. The boy in the back Bible. (= the boy in the back’s bible) (Green’s personal data, p.c.) Nevertheless, the ’s suddenly appears before ellipsis. As Green () and Conner () show, AAE children do not say sentences like (a) but (b): ()

a. *This book is Carl— [=book]. b. This book is Carl’s— [=book].

Thus, AAE, despite the rarity of possessive ’s, illustrates the requirement of a specific licenser, which has a Merchant-style E-feature, following Conner (). This is perhaps the sharpest evidence that a particular morpheme can carry licensing capacity.25 In languages where adjectives carry inflection, we find that ellipsis occurs easily: () a. German Hans hat ein rotes Auto und Fritz Hans has a red. car and Fritz ‘Hans has a red car and Fritz has a blue (car).’

hat has

ein a

blaues___. blue.

b. English *John has a red car and Fritz has a blue___. Here the child must note the presence of the inflection as a licenser, which is unavailable in English (see Hubert ). Now an interesting learnability conundrum arises: at early ages children in inflectional languages appear to be insensitive to them. Thus, a German child could say sentences like (), lacking inflection -es, which must occur inside an NP marking gender (neuter) for ein blaues Auto. The child putatively hears: ()

‘blau Auto’ [not ‘blaues Auto’] blue car

Aelbrecht () (see van Craenenbroeck  for review) claims that the licenser could be nonadjacent with language-specific features. This view significantly complicates the acquisition problem. 25





If an adult said () but the child does not hear the inflection, then it could think it had evidence that the adjective by itself would license ellipsis: ()

Du hast ein rotes Auto und ich habe ein blaues ___. you have a red. car and I have a blue. ‘You have a red car and I have a blue___ [=car].’

If UG dictates that a licenser must be present, the child must assume its presence until it is phonologically able to identify it. In fact, in Dutch uninflected adjectives do not allow ellipsis in the adult language (Sleeman and Hulk ), supporting the licensing argument.

.. The movement argument Thoms (b) claims that ellipsis also requires some form of movement, not just the presence of recognizable inflection. If true, then the child could know from UG that a bare adjective would not license ellipsis, but only one with overt evidence of movement. This view calls for designing the grammar to have movement of an adjective to a higher  node where the inflection resides. However, the fact that there is a corollary acquisition argument—the absence of inflectionless ellipsis errors in AAE— should be seen as striking evidence on behalf of Merchant’s argument for a licenser as well as compatible with Thoms’ proposal that movement to a higher node carrying 0 is present. We have shown that when unmoved inflection is present, children do indeed fail to elide *John is being bad, but I’m not being. However, careful work in Dutch and French by Sleeman and Hulk (), where they summarize much work beyond their own, shows early stages where exactly the licensing features of agreement are not found in Dutch (and French). () a. Een. ‘One.’

(Daan ;)

b. Oh, grote. ‘Oh, a big one.’ c. Groene. ‘The green one.’

(Matthijs ;) (Matthijs ;)

d. Moeten ze een nieuwe must they a new ‘They have to buy a new one.’

kopen. buy (Sarah ;)

They comment: “In child L Dutch, noun ellipsis is apparently not licensed by inflection (neither is it in the adult language), because adjectival agreement (and correct gender assignment) is not target-consistent before the age of  (Blom et al. ). The schwa and common gender are the default options until that age.” While it may seem that agreement could not be present here, it is still not unreasonable to suppose that it is still satisfied formally by a schwa, given the cross-linguistic comparison with English.

  



.. Semantics and partitivity Sleeman and Hulk (), following Valois et al. () and Algueró et al. (), argue with sharp early evidence that toddlers’ single words entail a partitive reading. The speaker and the hearer share a common-ground assumption that a set exists—or minimally a focal contrast. Words like ‘other’ directly imply such contrast. Their data show that French children use ellipsis to communicate a partitive meaning at ; years, and Dutch children at ; years (p. ): () a. Anaïs (;) French En veux encore e.  want more ‘I want more.’ b. Abel (;) Dutch Hebben we er nog eentje e? have we  still one ‘Do we have another one?’ c. Nathan (;) French Moi je veux deux me I want two ‘I want to have two.’ d. Sarah (;) Dutch Ik wil ook een I want also one ‘I also want one.’

moi. me

e [= one].

They list a range of words that are typical, such as encore ‘more’, nog ‘still’, ook ‘also’, and many others. How should we define the semantic dimension? Gagnon (b) offers this definition of partitivity based upon Barker (): ()

Partitivity: [of] = λx.λy. y < x [= there is an x and a y such that y is less than x]

In effect, then there is a presupposed set x of which y is a subset (see n.  for discussion of other options and formalism). For our purposes, we will avoid questions about whether LF, information structure, or semantic arguments are involved and simply propose that the child, when empty structure inside the NP arises, will assume a Strict Interface with some restrictive semantic representation.26 26

Gagnon (b) argues for a narrower contrast between NPE and partitive ellipsis which ultimately argues that a null pronoun is present equal to ‘of them’. There are many efforts to provide formal definitions of ‘partitivity’, some interestingly related to whether a numeral can apply to a singular noun, which we will not attempt to choose among. The





() Strict Interface: Quantifier with ellipsis yields partitive reading. This Interface allows the child to immediately project the ellipsis/partitive connection. Gagnon argues that the ellipsis involved is a lexically marked argument of the quantifier which explains its absence for some quantifiers like ‘all’, ‘the’, ‘a’, and ‘every’. English evidence supporting this account can be seen in the following cursory exploration of the contrast between ‘want more’ and ‘want all’. We found over  examples of ‘want more’ at the - and -year-old stage. Here is a group from one child Sarah, including assertions and questions (MacWhinney ): ()

a. b. c. d.

You want more? (;.) I want more. (;.) I want more. (;.) Want more? (;.)

We found zero examples of ‘want all’ with ellipsis with any child, but always with a filled NP-argument and often an overt form of partitive. ()

a. I want all them. (;.) b. Do we want all of them? (;.) c. Me do not want all that money. (;.)

An interesting case arises where NPE occurs inside the argument. ()

I want all of these__ NP. (;.)

This data, once again, occurs remarkably early and with remarkably precise languagespecific constraints.27

formalisms involved raise many obscure acquisition questions unless one argues, as we do, that the child is able to comprehend them as an inseparable cognitive unit whose parts are not subject to independent steps of acquisition. The topic requires extensive separate discussion. For our purposes, we assume that some formulation is immediately available to the child. Another case where the partitive properties are more explicit comes from Ionin, Matushansky, and Ruys (): a. PART(x) = def fZ : Z is a partition of xg b. A partition of an aggregate x is a set of aggregates Z such that the joint (sum) of all the elements in Z is equal to x (W Z = x) and for any two elements, w and v, in Z, the meet of those two elements is empty (w^v = / ). 27 Where inflectional licensing is available, such as the agreement marker -e in German, the ellipsis is licensed again for child and adult.

(i) Ich mag alle. I like all+e ‘I like all.’

  



.. What defines the partitive set? The notion that partitive/ellipsis connection involves a subset/set relation leads to a new question: what determines the relevant set in context? Wijnen et al. () asked specifically about discourse reconstruction when the context offers multiple possible sets. Three kinds of pictures were presented with the following scenario and two sorts of sentences.28 () ‘Here’s a playground. It’s great to do all kinds of funny things when you’re out in the playground, like swinging, making a sand castle or climbing on the monkey bars. There are some kids playing in the sand box.’ ()

a. Are two__ upside down? b. Are there two__ upside down?

A: Two kids inside the sand box are upside down

B: Two kids outside the sand box are upside down

C: Two adults outside the sand box are upside down

28

Experiment pictures from Wijnen et al. ().





Table . Average percentages of ‘yes’ responses broken down over condition and language from Wijnen et al. ()  kids (outside) Dutch English

.% .%

 kids (inside) % .%

 adults .% .%

Reconstruction requires both a noun and an adjoined PP. There are three possibilities: ()

a. Reconstruct nothing: Two adults b. Reconstruct only N: Two [kids]N c. Reconstruct N and PP: Two [[kids]N [in the sandbox]PP]

The yes/no questions in () were asked of each picture for children between  and  years in English and Dutch.29 The results showed an interesting divergence from adult grammar. A few children gave “yes,” “are two __ upside down” for the third picture, but mostly correct “no”. A significant number ( percent Dutch,  percent English) allowed (b) to be “yes”. In effect, as they argue, the noun kids is obligatorily reconstructed, but not the adjunct PP in the sandbox. Children by  years are clearly able to reconstruct ellipsis (N = kids, 6¼ adults), against the suggestive pragmatics of the picture (adults who are upside down). See Table . for detailed comparisons. Both Dutch and English children allow reconstruction of two kids dropping the adjunct as a restrictor (see Wijnen and Kaan  for a semantic representation that goes in the direction of the approach here). Some younger children, but far fewer, allow two adults which shows context overruling syntactic reconstruction. Therefore, most of the children meet the requirement of phonological and syntactic reconstruction. How about partitivity? Given a selection of options, half of the Dutch children and two thirds of the English children are able to reconstruct the PP as well, allowing them to reject the contextually offered sets. That is, it is not a partitive reading from the set of children, but a subset of the set of children in the sandbox. Nonetheless, a significant number seem to be able to adjust the question at issue so that they can limit the restrictor of the reconstructed noun. A smaller number can adjust the obligatory noun as well: They interpret ‘two__’ as ‘two [of [set of people in context]]’. Here, the set of people include adults, who are visually upside down, so they become an acceptable partitive subset of the whole array. Thus, we have evidence of the primacy of phonological/syntactic reconstruction, and the unsurprising residual capacity to effectively adjust the at-issue content to match the

29

The experiment was repeated by Obdeijn () with the same results and with van Weelden’s () eyetracking study (which produced some increase in contextual reference).

  



context. One can almost see children struggling to obtain a perfect match between all dimensions of the Interface.30

. O: T ’  

.................................................................................................................................. The child’s verbal landscape remains daunting: it includes interpretive dimensions that significantly magnify the learnability challenge. First general inference keeps a robust presence and is needed to filter out irrelevant aspects of context. Children’s earliest utterances like ‘uh, oh’, ‘bye-bye’, ‘hi’—just like adults’—have complex meanings but show no signs of systematic grammar, so they must lean on inference. How far does it reach? Our earlier examples of tomato store or up in the air show that inference is at work where more structured recovery is available to adults. Now consider cases like ‘me, too’ that look like there must be an elided VP or clause. They are found with no surrounding discourse from the ages of ; among many children. Consider adult and child uses of elliptical expressions like ‘me, too’: ()

Scene: Adult carrying coffee pot. Utterance: Me, too [=I would like a cup too].

Very young children do the same: ()

Scene: Adult lifts older child. Utterance: Me, too. (;)

It is noteworthy that there are no reports that a child simply says ‘too’, which with inferences might be sufficient. They have already registered the fact that ‘me, too’ is used by adults with missing information. A glance at twenty-five early uses of ‘me, too’ finds that most of them at ; years or  years on CHILDES are like these (MacWhinney ): () a. Zeke (;) MOT: ok. (action: turns nesting cup over. resituates nesting cup that Zeke put on stack.) CHI: me, too. (action: adds nesting cup to tower.) b. Ruth (;) MOT: your baby. happy birthday dear . . . who is it this time? little white baby. CHI: me, too. There is a further variant which promotes such adjustments: sentences with ‘there’ (or er in Dutch). In Dutch er is obligatory, unlike English, but we will not explore the matter here. 30





c. Ruth (;) MOT: did you just give it Anna then, did you? MOT: oh right. MOT: just put those down there. CHI: me, too. MOT: okay. Here the whole exchange seems to be highly dependent upon obscure inferences about actions, not systematic reconstruction of verbal antecedents. ()

Eve (;) FAT: I’ll be right back. (action: goes to leave.) FAT: you close the door # please. FAT: no # I’m going out to look at the car. you stay right here. CHI: me, too. [= go out to the car, overruling the imperative ‘stay here’.]

Note that ‘me, too’ reaches back to the sentence [go out]. One point is again of critical importance here: if all these cases were syntactically generated, then the prediction should be that they require actual clear reconstruction from full clause antecedents, which would lead us to expect them to be much later. Moreover, if were they introduced through the abstract Kennedy-type formalism, and if that formalism cannot appear until its ingredient structures (i.e., VP, operator, etc.) are in the grammar with full clause antecedents, they should again be late. We are led to conclude that they must involve a simple representation, like pro for the missing clause, and substantial inference.

.. Fragments To fill out the child’s challenge, we must now reckon with recent theoretical insights that show that these fragments are not interpreted only by inference, as is clear from work on fragments by Merchant () and a comprehensive theory by Weir (). Weir uses the example of a child trying to do something and the parent says ‘both hands!’ which entails an elided imperative clause ‘(you must use) both hands’. Children do not balk at such utterances, but regularly seem to understand them, just as -year-olds grasp ‘don’t’. Do children show any knowledge of grammar with fragments? Jensen and Thornton () provide this example of an embedded fragment from a -year-old: ()

(;) Experimenter: What do sheep eat? Georgia: I think just real little food. (= I think [they eat] just real little food.)

Weir, following Barbiers () and Temmerman (), shows that embedded fragments only occur with bridge verbs which include ‘think’ but not ‘know’. Therefore, we predict that we will never find sentences like *‘I know just a little real food (= ‘I know that they eat

  



just a little real food’)’. Thus, even early fragments ultimately show the same kinds of UG constraints found elsewhere. We will not explore the details, but this theoretical work shows that fragments submit to complex analyses in terms of focus, movement, an E-feature, and clausal ellipsis which are not identical cross-linguistically.31 For instance, Merchant () shows subtle languageparticular effects. He shows that bare nouns may be grammatical only in some environments. While in English one can answer as (aB), in German, where preposition-stranding is not allowed, the preposition must be supplied, as in (bB). () a. A: Who did you speak with___? B: John. b. A: Mit wem hast Du gesprochen___? with who. have you spoken B: Mit Hans / *Hans. ‘With Hans.’ In other circumstances, not yet perfectly understood, English also demands the preposition (aB), just like German (bB), but not (bB): () a. A: How did she sweep the room? B: With a broom. / *Broom. b. A. When will you play chess? B. Noon [= at noon => deleted Prep]. It is not, however, inevitable that children immediately realize these adult uses of fragments. One form of language disorder is a failure to eliminate inference in favor of structure by some children. In fact, disordered children often answer with fragments which are communicatively adequate, but perceived by adults as grammatically imperfect. Consider what many disordered children do in answering (aA): they answer exactly: ‘broom’ (Roeper and Seymour , Seymour et al. ). ‘Broom’ communicates adequately—with an easy inference about instrumentality—but ‘with a broom’ is required from adults to honor the case requirements of full reconstruction (i.e. ‘she swept the room with a broom’). Why do adults allow Prep-deletion with a temporal PP noon, but reject manner Prep-deletion of with for ‘broom’? In fact, E. C. Klein () observes that it is a frequent L error as well, which suggests that it is not clear what kind of disorder it is.

.. Question Under Discussion A second factor can dramatically tilt discourse in a new direction: Questions Under Discussion (QUD). We (and a child) must always determine what the chatter is about, in other words, what question is at issue. Surprisingly, often adults shift the focus. Consider this commonplace interaction: 31

Merchant and Weir argue that clausal ellipsis is coupled with a moved focus item which satisfies the E-feature.





()

CHI: can I watch TV? MOT: you have to go to bed.

The child must infer an answer to the actual question, ‘no’, and see how the assertion carries this implication. Often the shift is oblique, which requires the inference ‘yes, but’. ()

CHI: can I watch TV? MOT: you have to eat lunch first.

Can children handle such shifts? Two of our examples already illustrate the capacity. The child himself performs such a shift in () (Adult ‘can’t you work this thing?’ followed by Child ‘I know why’). The child has shifted from a yes/no question to the reconstruction of an elided indirect question which presupposes the truth of the prior question: ‘I know why you can’t work this thing’. In the ‘me, too’ conversations (), the child ignores the opposite command ‘you stay here’ to a response that presupposes an elided VP from an earlier sentence: ‘I’m going outside’ to which the ‘me, too’ refers. These conversational gymnastics, coupled with the emotional shifts they entail, are impressive evidence that the child is manipulating a complex interface that depends upon both grammatical and social factors. They are far beyond ‘logical’ induction that a computer can mimic. Thus, we might say that we have here—though not fully resolved—a candidate for a Strict Interface which is present from the earliest cases: determine the Question Under Discussion before you reply. One might want to attribute this to the child from the outset and many adult expressions toward children presuppose just this kind of ability to determine what is at issue.

. C

.................................................................................................................................. We have argued that children are able from the outset to license ellipsis, recover missing structure from discourse and context, and generate both an open pro and a full PF representation. The capacity seems to be immediately triggered because we find ‘I did’ from the earliest verbal interactions, though not with perfect adult results. Each of these capacities must be sensitive to language-particular licensers, varying principles of recoverability, and the abstractness of the representations achieved in cases where mismatches are allowed. We have addressed the special ellipsis properties of: VPE with antecedent-contained deletion, sluicing, argument drop, and then NCA and NPE with partitives. We have labored to embed these observations in an updated version of what has been called the Primary Linguistic Data (Chomsky )—the adult input from which children must derive these structures. They involve not only the entire syntactic structure of discourse, but properties of Focus, QUD, and constraints on fragments. An important conclusion follows: To cope with an obscure verbal and pragmatic environment the child must utilize UG Strict Interfaces as goals to isolate what is linguistically and pragmatically relevant. Strict Interfaces substitute for an evaluation metric as a method to exclude grammars that allow non-adult interpretations.

  



We have argued that (a) Kennedy’s () Strict Identity Hypothesis serves as an LF interface behind sloppy identity, (b) a semantic notion of definiteness entails an invisible LF-projection for NPE, and (c) a semantic representation of partitivity is immediately available to children. These goals are realized in steps—not yet perfectly visible—which move the child from default pro representations (also used by adults to reduce complexity) to full PF copies in the elided domain. The acquisition path for the elided structure then resembles or perhaps reproduces the acquisition path for child’s first language-specific representations leading to maximal projections of VP and NP. The acquisition process obeys an overarching notion of economy that seeks to minimize use of pragmatic inference and maximize the efficiency of inherently complex linguistic mechanisms that operate both within and across modules. The path which results shows these rough stages that we outlined: ()

a. b. c. d. e.

Pure inference: ‘oh, dear’ Fragments: ‘milk?’ + inference Minimal representations: pro ⇒ clause + inference Maximal representations: full VP/NP + inference Full interface: VP/NP + LF representations

Ellipsis constitutes one of the most dramatic examples of poverty of the stimulus in as much as it is built upon silence. While ellipsis is omnipresent, its variety is so intricate when pragmatic and interface factors are added that examples of each type remain very rare in what the child hears. Therefore, it must be built into UG interfaces. This extension of UG to include interfaces adds to the many arguments for the innateness of grammar.

A Thanks to L. Amaral, L. Green, J. Hartman, K. Johnson, J. de Villiers, A. Pérez-Leroux, A. van Hout, A. Weir, L. Renato, editors J. van Craenenbroeck and T. Temmerman, and reviewers and members of the Language Acquisition Research Center (LARC) at UMass. Thanks to Jaieun Kim for proofreading, formatting, and many other improvements. Misinterpretations and errors are mine.

  ......................................................................................................................

   ......................................................................................................................

 

. I

.................................................................................................................................. W contextual conditions are right, natural languages permit speakers to omit linguistic material that expresses information that they nonetheless intend to communicate to their hearers. This is a significant design feature of language, in that it allows speakers to minimize articulatory effort while still being expressive in communicating their message (Zipf ), at least in cases in which the hearer is able to perform the intended meaning recovery. An important class of such cases falls under the cover term of . Unlike some other forms of content enrichment (implicatures, etc.), ellipsis phenomena are among those in which the grammar itself provides a tip-off to the hearer that normally required linguistic material is in fact missing. Successful interpretation in such cases is enabled by the fact that context allows for the recovery of already activated (precomputed or otherwise predictable, and hence redundant) meanings. There are many types of ellipsis found in the world’s languages, with an impressively diverse set of constraints on their use. One instance that is prevalent in English (less so in other languages) is   (VP) , which will be the primary focus of this chapter: ()

John loves his mother, and Bill does too.

Here, the stranded auxiliary in the second clause (henceforth referred to as the  clause) marks a vestigial verb phrase, a meaning for which must be identified to recover the proposition denoted by the sentence. In many cases of VP-ellipsis, the ability to identify the meaning is enabled by the occurrence of another linguistic expression (the ), which in this case is located in the first clause. In certain cases, an antecedent is capable of making more than one possible meaning available for a VP-ellipsis interpretation. This is in fact the case for (). Assuming that the assignment of the pronoun his in the antecedent clause leads to a reading in which John loves John’s mother, two possible meanings become available for the ellipsis clause: one in which Bill loves John’s mother (the  reading), and one in which Bill loves his own mother (the  reading). As we will soon discover, the facts become more complicated when the antecedent contains multiple coreferring pronouns.

  



Despite the considerable attention paid to VP-ellipsis in the literature (not to mention a wide variety of other types of ellipsis), the conditions under which a representation of an utterance may serve as a suitable antecedent for interpreting a subsequent ellipsis remain poorly understood. Three prominent views found in the literature that we will focus on include: Syntactic: The recovery of the elided VP meaning is dependent on there being a suitable syntactic VP to serve as an antecedent in the discourse. This need may arise from a constraint that allows for deletion under identity at surface structure or syntactic logical form (e.g., Sag a), or a recovery procedure that reconstructs the missing VP at the ellipsis site at one of these levels (e.g., Williams b; Fiengo and May ). Anaphoric: VP-ellipsis is essentially a null proform, and hence interpretation is governed by the same types of processes used to resolve other types of referential expressions such as pronouns (Schachter b; Chao ; Dalrymple ; Hardt b; Kehler a; Lobeck , inter alia). Deaccenting: VP-ellipsis is essentially deaccenting taken to the point of complete lack of articulation. The constraints on ellipsis are then predicted to be based on some version of the same processes that govern the deaccentuation of overt material (Rooth a; Tancredi ; Chomsky and Lasnik ; Merchant ). Whereas analyses in this category typically posit the existence of unpronounced syntactic structure at the ellipsis site, they differ from syntactic deletion approaches in that the constraints on elision only enforce semantic identity between the elided VP and its antecedent. Note that these categories are offered only to provide prototypical characterizations of approaches, presented so as to allow comparison of their predictions without getting bogged down in the details of various particular analyses. Indeed, some analyses span multiple types; for instance, Rooth (a) offers a deaccenting analysis that is augmented with a syntactic redundancy condition. As the editors explain in the handbook’s introduction, the central questions in ellipsis research can be categorized in terms of three issues: (i) the structure of the ellipsis site (i.e., whether covert syntactic structure exists at the ellipsis site), (ii) recoverability (i.e., how the ellipsis site gets its meaning), and (iii) licensing (i.e., syntactic restrictions on ellipsis beyond those bearing on recoverability). Whereas this chapter will have little to say about the third question, the potential for an article entitled “Ellipsis and discourse” to shed light on the other two is dependent on to what extent discourse considerations play a role in enabling ellipsis. On the syntactic view, the interest level might be pretty low, as such analyses straightforwardly require structure at the ellipsis site (in answer to the first question) and that a suitable VP exist in the immediate context (in answer to the second question). On the other hand, the questions become more interesting on the anaphoric and deaccenting views. We will survey some of the evidence that informs these questions, focusing on issues bearing on the discourse context in its role of making meanings available to enable subsequent ellipsis. Section . first addresses these questions by analyzing constraints on the linguistic form of the antecedent and whether or not they entail the existence of syntactic structure at the ellipsis site. Section . then examines these questions by analyzing the role of syntax in shaping the anaphoric dependencies that give rise to strict and sloppy ambiguities. Consideration of the data will run us straight into the arms of fundamental questions about the interaction among grammar, meaning, and processing: To what extent are infelicitous





examples due to grammatical constraints, versus semantic or pragmatic factors, versus mere interpretation difficulty? Whereas we will not resolve these questions here, our examination will give rise to a set of adequacy criteria with which future work can engage.

. C   

.................................................................................................................................. Determining whether or not VP-ellipsis requires a syntactically matching antecedent should be straightforward enough. We can simply look to cases in which a salient VP meaning exists in (or is easily constructible from) the discourse context, but the antecedent is not in the required syntactic form. Syntactic approaches predict that such cases should be bad, whereas the anaphora and deaccenting approaches suggest they should work. The data, however, are not at all clear. Consider the following examples: ()

a. b. c. d.

#This problem was looked into by John, and Bob did too. [looked into the problem] #Ali blamed himselfi, and George did too. [blamed Al] #Bill defended Georgei and hei did too. [defended George] #John read everything which Bill believes the claim that he did. (Haïk )

These examples all provide seemingly prima facie evidence for the syntactic account. Example (a) involves a syntactic mismatch: the ellipsis clause is in the active voice, whereas the antecedent clause is in the passive. Therefore the VP necessary on the syntactic account (looked into the problem) is not present in the context, hence the example’s infelicity. The VPs in examples (b,c) are viable antecedents, on the other hand, but seem odd on the noted strict readings: exactly what Conditions A and C of the binding theory would respectively predict if syntactic structure exists at the ellipsis site. The example of antecedent-contained deletion (ACD) in (d) is not only unacceptable, but has the same feel of a subjacency violation as its unelided counterpart (*John read everything which Bill believes the claim that he read).1 All of this is just what we would 1

Note that acceptable examples of ACD are among a variety of cases that demonstrate that, in principle, there is no constraint against elided VPs containing traces: (i)

John read everything that Bill did .

On a standard syntactic analysis, the quantified NP everything raises, leaving behind an antecedent VP containing a trace, thereby licensing the deletion (or reconstruction) of a parallel VP in the ellipsis site. Another type of example involves VPs containing traces left behind by topicalization; (b) is from Merchant (this volume; see also Merchant b): (ii)

a. Abby likes dogs, but cats she doesn’t . b. *Abby knows five people who have dogs, but cats she doesn’t .

Like (i), example (iia) is felicitous, in this case despite the fact that the antecedent does not contain a trace. The variant in (iib) is unacceptable, on the other hand, a fact predicted by an island violation on a syntactic theory. Both facts can be explained by a deaccenting theory in which unpronounced syntactic structure resides at the ellipsis site. On the other hand, whereas anaphoric analyses can be constructed that allow for cases like (iia), it is far from clear how such an approach could simultaneously rule out (iib).

  



expect on a syntactic analysis. On the other hand, it seems hard to argue that the event representations that correspond to these VPs are not made available by the discourse for subsequent anaphoric reference. In (c), for instance, a discourse representation for defend George should be made available by the antecedent clause; the particular syntactic form used to introduce the event (which here, crucially, involves a name rather than a reflexive pronoun) should not matter. This would all be convincing if it were not for the fact that there is equally compelling evidence for the anaphoric theory. Consider: ()

a. This problem was to have been looked into, but obviously nobody did. [look into the problem] (Vincent Della Pietra, in conversation, cited in Kehler b) b. Ali defended himselfi because Bill wouldn’t. [defend Al] (adapted from Dalrymple : ex. a) c. George expected Ali to win the election even when hei didn’t. [expect Al to win the election] (adapted from Dalrymple : ex. c) d. Which problem did you think John would solve because of the fact that Susan did? [solved the problem] (Rooth )

Example (a) features the same passive–active mismatch as (a), but is acceptable. Similarly, (b,c) should result in the same respective Condition A and C violations as (b,c), but speakers judge them to be acceptable on the strict reading. Like (d), (d) is a case in which the syntactic account would predict a trace violation (consider its unelided counterpart *Which problem did you think John would solve because of the fact that Susan solved?), but again it is acceptable. The anaphoric view correctly predicts these cases to be acceptable for all of the same reasons that it gets the predictions for (a–d) wrong. How about the deaccenting theory? We can also evaluate this approach easily enough, by comparing the felicity of ellipsis examples against their variants in which the VP is overt but deaccented.2 Interestingly, not only do the judgments in some cases diverge from those for their elided counterparts, they do so in different ways depending on the type of violation. On the one hand, the deaccented versions of both mismatch cases (a) and (a) seem acceptable: ()

a. This problem was looked into by John, and Bob looked into the problem too. (cf. a) b. This problem was to have been looked into, but obviously nobody did look into the problem. (cf. a)

2 Here I set aside the well-known fact that a wider range of VPs can be deaccented than can be elided; for instance (ib), unlike (ia), cannot be used to describe a situation in which Bill insulted Mary by doing something other than calling her a Republican (here and throughout I use smaller font to indicate deaccented material):

(i) a. John called Mary a Republican, and Bill insulted her too. b. John called Mary a Republican, and Bill did too. Deaccenting in (ia) is licensed because (under suitable political assumptions) the antecedent clause renders insult Sue as Given. The VP cannot be elided, however, as the hearer would have no way of distinguishing which of many deaccentable VPs (said something to Mary, offended Mary, etc.) was the one intended by the speaker.





Whereas example (a) might be less than a perfectly constructed discourse (more on that a bit later), it seems considerably more acceptable than its unelided counterpart. On the other hand, the binding theory cases seem bad with deaccented VPs on all sides: ()

a. b. c. d.

#Ali blamed himselfi, and George blamed himselfi too. (cf. b) #Ali defended himselfi because Bill wouldn’t defend himselfi. [defend Al] (cf. b) #Bill defended Georgei and hei defended George too. (cf. c) #George expected Ali to win the election even when hei didn’t expect Al to win the election. (cf. c)

Not only does Condition A block the strict reading where we would have expected it (a), but apparently also in (b) despite the fact that the effect is ameliorated in its elided counterpart (b). Examples (c,d) show the same pattern for Condition C. What is common to all of these cases is that the normal syntactic constraints governing overt material apply to deaccented VPs, in contrast to some of the elided versions.3,4 Since the foregoing data do not settle the question, we can also look to independent criteria for additional insight. For instance, the anaphoric approach predicts that the behavior of VPellipsis should pattern with other types of anaphora like pronominal reference. This appears to be the case. For instance, as noted by Lakoff () and Jackendoff (), VP-ellipsis and pronouns may be cataphoric in similar circumstances, as shown in examples (a–d). ()

a. If McCain will ϕi, Obama [will make a statement criticizing President Putin]i. b. If hei makes a statement criticizing President Putin, Obamai will make a fool of himself. c. *Obama will ϕi, if McCain [will make a statement criticizing President Putin]i. d. *Hei will make a fool of himself, if Obamai makes a statement criticizing President Putin.

Cataphoric VP-ellipsis is allowable when it is embedded as in sentence (a), as it is for pronominal reference in sentence (b). Cataphora is not allowable when the ellipsis is not embedded as in sentence (c), on the other hand, as is the case for pronominal reference in sentence (d). This pattern receives no independent explanation on a syntactic approach. It is also mysterious on a deaccenting approach: Whereas the anaphora and deaccenting analyses share a number of properties, cataphoric deaccenting of a VP is not permitted in contexts in which its meaning has not already been given (e.g., via previous evocation) in the discourse at the time of mention:

3

One could posit an analysis that allows for referring expressions that are compatible with binding theory to occur as the objects of the elided VPs instead (e.g., blame him in (a); see for instance the V C proposal of Fiengo and May ). In this case, all of the examples would be predicted to be acceptable, again not capturing the contrast between (b,c) and (b,c). 4 It is worth noting that the deaccenting theory captures the difference between (d) and (d). Whereas the unelided version of (d) is unacceptable if a trace is left in the ellipsis site, it is fine with a full VP— Which problem did you think John would solve because of the fact that Susan solved it?—and hence can be deleted under identity. Example (d), on the other hand, is as unacceptable on the full VP variant as the trace-containing variant (*John read everything which Bill believes the claim that he read it.).

  



() [Context: What should we do this afternoon?] a. #If you’re willing to go to the mall, I’d like to go to the mall. b. If you’re willing to go the mall, I’d like to (go to the mall). c. If you’re willing to, I’d like to go to the mall. That is, assuming that going to the mall has not been previously discussed, cataphoric deaccenting is infelicitous in (a); the fact that the VP meaning is not Given means that accent is required (b). The version with VP-ellipsis (c) is nonetheless acceptable under the same conditions. So whereas we normally think of the constraints on deaccenting as being less restrictive than for VP-ellipsis (see n. ), here is a case in which the constraints on deaccenting are more restrictive. Again, this is expected if VP-ellipsis is anaphoric in the same way that pronouns are. A second characteristic of anaphora is that, while preferring local antecedents, it can access referents evoked from clauses other than the most immediate one. Such reference is not uncommon for VP-ellipsis; Hardt () reports that  percent of the examples in the Brown corpus have an antecedent that is at least two sentences back in the discourse, as in example (). ()

The thought came back, the one nagging at him these past four days. He tried to stifle it. But the words were forming. He knew he couldn’t.

Likewise, the subject pronoun he in the final sentence in passage () locates its referent from two sentences back. A third property of anaphora is that it allows for split antecedents: in (), the pronoun them successfully refers to Mary and Fred as a group, even though they are mentioned from distinct constituents. ()

Mary wants to go to Spain and Fred wants to go to Peru, but because of limited resources, only one of them can. (Webber )

Analogously, the VP-ellipsis in () refers to a plural referent composed of meanings evoked by non-conjoined VPs. Fourth, under certain conditions an anaphoric pronoun can access a referent that is not coreferent with, but is nonetheless semantically derivable from, an antecedent expression: ()

Jean is a Frenchman, although he hasn’t lived there for many years. (Ward et al. )

Here, the pronoun there successfully refers to France, which has a transparent semantic relationship to the meaning of the antecedent Frenchman. Similarly, Webber () offers an example of a related sort involving VP-ellipsis: ()

Martha and Irv wanted to dance with each other, but Martha couldnʼt because her husband was here.

Whereas the antecedent clause does not contain a VP that denotes the intended referent— i.e., the meaning of dance with Irv—this meaning can be transparently derived from the





meaning of the antecedent clause. Such behavior is thus consistent with the anaphoric account in principle, whereas it is more difficult to explain on a syntactic analysis. Finally, and more controversially, there is evidence that VP-ellipsis can refer to situationally evoked referents; examples (a,b) are from Schachter (b): ()

a. [John tries to kiss Mary. She says:] John, you mustn’t. (= Schachter b: ex. ) b. [John pours another martini for Mary. She says:] I really shouldn’t. (= Schachter b: ex. )

Obviously there is no syntactic VP available to serve as an antecedent in such examples. However, Hankamer () disputed Schachter’s data, arguing that such cases are, in his terms, either formulaic or conventionalized, occurring only as    and not generally as declarative statements or informational questions. For instance, the elliptical expressions in (a,b) are infelicitous, even though the contexts are the same as for Schachter’s examples: ()

a. [John tries to kiss Mary. She says:] #John, you’re the first man who ever has. (= Hankamer : aʹ) b. [John pours another martini for Mary. She says:] #John, are you aware that no one else has? (= Hankamer : bʹ)

More recently, these data have been argued by Miller and Pullum () to admit to a parsimonious explanation whereby VP-ellipsis with situationally evoked referents is subject to the same discourse conditions as cases with linguistic antecedents. Specifically, they argue that VP-ellipsis requires that the discourse context give rise to a highly salient set of alternatives that contains the meaning of the ellipsis clause (the nature of this set depends on whether the ellipsis is subject-focused or auxiliary-focused; see the discussion of Kertz  below, and of Question-Under-Discussion models in section .). The illusion of a constraint against VP-ellipsis with situationally evoked referents results from the fact that only in limited circumstances (e.g., in (a,b), but not (a,b)) does the discourse context raise the salience of the alternative set to the required level. All of these properties are consistent with an anaphoric theory of VP-ellipsis, rather than being indicative of a more restricted, local syntactic operation. Crucially, such behavior is not found with certain other forms of ellipsis, such as bare remnant ellipsis (e.g., gapping and stripping), which are far less sympathetic to non-local antecedents, cataphora, situationally evoked antecedents, and so forth. This is an important observation, since it might be tempting to draw conclusions about one form of ellipsis from the behavior of another. Different forms of ellipsis have different syntactic, semantic, and discoursal properties, and hence theorizing should not be guided by the assumption that a unified theory of ellipsis is possible. That is to say, there is no a priori reason why it could not be the case that VPellipsis is anaphoric and other types of ellipsis are not. As an ensemble, the aforementioned facts create a quandary. It seems clear that, as it stands, syntactic approaches undergenerate whereas anaphoric approaches overgenerate. Two types of approach to reconciling the facts therefore suggest themselves. First,

  



proponents of syntactic accounts (and to some extent, deaccenting approaches that posit unpronounced syntactic structure) might say that the fact that we see any apparent effects of syntactic structure demonstrates that syntax must be relevant to licensing and recovery, since any sensitivity to syntax would seem out of the reach of anaphoric accounts. Proponents of syntactic accounts would therefore seek to explain why examples like (a–d) are judged as grammatical even though they are not. On the other hand, proponents of the anaphoric account might argue that the paradox presented by (a–d) and (a–d) is fundamentally more difficult to resolve on a syntactic theory than on an anaphoric one, since if examples like (a–d) are ruled out by the grammar, then no semantic or pragmatic factors should be able to rescue them. Proponents of the anaphoric account would thus claim that all of the cases just reviewed are potentially acceptable, and look for independent causes for the unacceptability of (a–d) and other problematic cases. This opposition gave rise to a lineage of research that bounced back and forth between the two approaches, in search of the key factors that determine VP-ellipsis felicity. Kehler (, , ), concluding on the basis of examples of the sort in ()–() that VP-ellipsis is anaphoric, took the route of attempting to explain the contradictory data with a theory that offered an independent reason why some examples are ruled out. Specifically, he argued that whereas anaphoric interpretation requires no syntactic reconstruction of an antecedent VP, there is a second, independent interpretation process that does under certain conditions. This process—the establishment of  — requires that constraints associated with one or more   be established between the antecedent and ellipsis clauses, and part of this process is the identification of the arguments to which these constraints apply. The constraints associated with certain relations—for instance, C–E relations such as Violated Expectation (a), Explanation (b,d), and Denial of Preventer (c)—only require access to the propositions denoted by the clauses. Since anaphoric resolution of the VP-ellipsis will result in a complete propositional meaning for the ellipsis clause, no syntactic construction of the VP is predicted to be necessary. On the other hand, the identification of the arguments to R relations, including the Parallel relation operative in (a–d), is a more complex process that requires that subsentential arguments be retrieved from each clause and their correspondences identified. It is this requirement that triggers the recovery of syntactic structure, explaining the sensitivity we see to syntactic constraints in (a–d). Although Kehler’s analysis potentially explained both the mismatch data and the binding theory violation facts, it was unsatisfying in several respects. For one, it is not obvious why syntactic reconstruction would be necessary for the establishment of Resemblance relations, as it is possible to have Parallel coherence without parallel syntax in discourses with no ellipsis. Second, as noted by Kehler (), there are cases which appear to have properties of more than one relation (all examples and judgments from the paper): ()

a. ? This problem was looked into by John, even though Bob did (too). [look into the problem] b. ??This letter provoked a response from Bush because Clinton did (too). [respond] c. This problem was looked into by John, even though Bob already had (#too). [responded] d. This letter provoked a response from Bush because Clinton already had (#too). [responded]





Examples (a,b), which contain passivized and nominalized antecedents respectively, seem more marginal than their counterparts in (c,d). Kehler conjectured that despite the fact that the connective in all four cases signals a Cause–Effect relation, a Parallel relation is also triggered in (a,b) in light of existence of the corresponding agents and parallel tenses (hence the congruence with the adverbial too). On the other hand, (c,d) lack these triggers for Parallel, and hence no reconstruction occurs. While this conjecture is possible in theory, the idea that multiple coherence relations would be operative in some cases is not entirely natural, and its use here is somewhat post hoc. So whereas the data suggest that coherence relations may be a relevant factor, the muddled middle ground presented by (a–d) indicates that the coherence analysis might not make quite the right slice through the data. A series of papers by Lyn Frazier, Charles Clifton, and colleagues took the opposite approach, particularly in advocating a syntactic theory that sought to explain mismatched cases by way of other means. To start, Frazier and Clifton () set out to evaluate Kehler’s coherence theory experimentally, concluding that there was little to support its predictions.5 Their first experiment, for instance, used examples that paired cases like (a), shown in (a), with clear Parallel examples (b): ()

a. The problem was looked into by Kim because Lee did. b. The problem was looked into by Kim just like Lee did.

After hearing these sentences read aloud, participants were asked to say whether they “got” the sentence or did not “get it.” The results showed no benefit for examples in which the connective indicates the Explanation relation (a) over those in which it indicates Parallel (b); in fact, there was a small effect in the other direction. Noting that some of their causal stimuli (e.g., (a)) may have been pragmatically odd, they ran a follow-up written acceptability study for which they modified their stimuli (e.g., changing because to even though in (a)) and paired the mismatch cases with syntactically matched versions. Whereas the mismatched examples were rated significantly worse than the matched ones, the stimuli that employed Cause–Effect relations were again rated no better than the Parallel cases. Per the foregoing discussion, the results of these studies were not a surprise on Kehler’s analysis, as Kehler () had already acknowledged the unacceptability of cases like (a) (cf. (a)) and attempted to explain why they were bad. These experiments therefore confirmed previously acknowledged intuitions about such examples. Importantly, Frazier and Clifton did not conduct experiments that paired examples like (a,b) with those like (c), and hence the central contrast between examples like (a) and (a) upon which Kehler’s theory was based was left unresolved. A second paper published around the same time (Arregui et al. ) aimed to explain mismatch facts with a different tack—one that maintains the fundamental assumptions of a syntactic approach. Specifically, Arregui et al. posit the R H, which states that although VP-ellipsis imposes a requirement for a suitable syntactic antecedent, the language processor has a repair strategy for structure that allows it to “recycle” whatever linguistic materials are at hand to create one when one does not already exist. The degree of

For additional experimental evaluations of the coherence analysis, see Kim and Runner (, ) and SanPietro et al. (). 5

  



acceptability of VP-ellipsis is then predicted to be proportional to how close the existing syntactic material in the discourse context is to the VP required to license the ellipsis, and hence the complexity of operations that need to be performed to fashion this VP from those materials. This characterization, according to Arregui et al., means that we should expect declining rates of acceptability across the four types of example in (a,d), as they require increasingly greater amounts of repair. ()

a. b. c. d.

Almost nobody approached the lion, but the trainer did. (Available VP) Approaching the lion was nearly impossible, but the trainer did. (Embedded VP) The lion was nearly impossible to approach, but the trainer did. (VP with trace) The lion was nearly unapproachable, but the trainer did. (Negative adjective)

This prediction was confirmed in an acceptability study.6 A few provisos are in order, however. First, it is not completely clear that their results refute the predictions of an anaphoric theory. For one, as the authors themselves note, the difference between (a) and (b) does not concern whether a matching VP exists, but whether it appears in the expected position—an issue bearing on referent accessibility, and not syntactic mismatch.7 Second, it is not at all unnatural to think that form of reference would impact the relative level of accessibility of events for subsequent reference—for instance, whereas the adjective unapproachable in (d) might activate the concept of approaching in the hearer’s mental model of the discourse, it might do so less strongly than when evoked by a verb as in (a,b).8 On the assumption that successful anaphoric reference depends on a sufficient degree of activation of and/or attention to the referent in the hearer’s mental model of the discourse, such cases might therefore simply require additional contextual support to become accessible.9 Third, when making the case that a Although “unacceptable” in this task was defined as “violating the normal rules of everyday English”—a task description that runs the risk of participants bringing prescriptive biases to bear. 7 It should be noted that while there is a syntactic match between the VP in the antecedent clause (approaching the lion) and the form required by the ellipsis clause (approach the lion), there is not a perfect morphological match. Such mismatches are typically unproblematic for successful VP-ellipsis; indeed even examples like (a) lack a perfect match. 8 This issue is familiar from the debate on the status of so-called anaphoric islands for nominal reference. Whereas Postal () appealed to a grammatical constraint against ‘word-internal’ reference (# Fritz is a cowboy. He says that they can be difficult to look after), Ward et al. () point out that such cases are acceptable if a sufficient degree of semantic transparency exists between the antecedent expression and its word-internal counterpart (Do parental reactions affect their children?; see also example ()). That is to say, the difference results from the fact that parental activates the concept of  more strongly than cowboy does for . In the domain of event reference, Kehler and Ward () cited the same factors in their treatment of do so anaphora with nominalized antecedents. They argue that the reason that examples like Four out of five female smokers do so in order to lose weight are acceptable but Most professors will do so even when no one is listening is not is because smokers activates the concept of smoking more strongly than professors activates the concept of professing. This type of sensitivity of an anaphor to the level of activation of its referent is just what one expects on an anaphoric theory. 9 However, Arregui et al. () are quite right when they say: “A semantic account that could mirror our results would need to find the same level of relevant fine-grained distinctions in purely semantic representations. Such an account would also have to deal with some nontrivial issues. Since all conditions have event-denoting verb roots, why would these event properties be made more salient or 6





form is not anaphoric, it is important to have a control condition that uses a form that is uncontroversially anaphoric. Do it anaphora is a natural choice; consider the variants of (a,d) in (a,d): ()

a. b. c. d.

Almost nobody approached the lion, but the trainer did it. Approaching the lion was nearly impossible, but the trainer did it. The lion was nearly impossible to approach, but the trainer did it. The lion was nearly unapproachable, but the trainer did it.

The question is whether we would find a similar reduction of acceptability in (a,d) as was found for (a,d). If we do, then one cannot attribute the effect to a condition on ellipsis. Indeed, it would suggest instead that the linguistic form of the antecedent affects the accessibility of the event referent similarly in both cases, which would actually support the anaphoric view of VP-ellipsis. Resolution of the issue awaits further study. Finally, Arregui et al.’s study also did not attempt to resolve the conflicting evidence regarding mismatch: there is no difference in acceptability predicted between examples (a,d) and their counterparts in (a,d), which despite the results of Frazier and Clifton (), intuitions suggest still exists. In another paper, however, Grant et al. () posit an analysis based on what they call N-A I (NAIs). They note that in many cases of mismatch offered in the literature, the antecedent clause does not assert that the event it describes actually happened, which in turn could be taken to implicate that it did not. For instance, the antecedent clause in (a), repeated below as (), might be taken to suggest that no one did look into the problem, an implicature that the ellipsis clause reinforces. ()

This problem was to have been looked into, but obviously nobody did. [look into the problem]

Grant et al. () state their hypothesis as follows: NAI hypothesis: The presence of an NAI will set up an alternative which implicitly focuses the antecedent clause, and sets up a potential QUD [AK: Question Under Discussion], thereby motivating and guiding the processing of a later clause which comments on the QUD, especially when the antecedent of an ellipsis needs to be repaired.

Note that, as indicated by the final clause, Grant et al. build their analysis on top of the Recycling Hypothesis. As such, despite the semantico-pragmatic character of the process of providing answers to QUDs that arise from implicatures, their analysis is based on a syntactic, and not semantic, interpretation procedure. NAIs only serve the role of “motivating and guiding” the processor to repair an ungrammatical ellipsis.10 available in an active sentence than a passive, and why would they be made more available by an inflected VP than by a gerund, by a gerund than by a nominalization, etc.” (p. ). Whereas I am less pessimistic than Arregui et al. that a plausible account of this type could be constructed (see n. ), the onus is indeed on anaphoric approaches to specify the principles that link linguistic form to level of accessibility in the discourse model, a topic that has received little attention to date. What Grant et al. mean by implicatures “motivating and guiding” the language processor is admittedly not clear to me. 10

  



Grant et al. argue that their analysis is supported by the results of an acceptability study they conducted using stimuli like (a,d): ()

a. b. c. d.

This information was released but Gorbachev didn’t. (passive–active) This information should be released but Gorbachev didn’t. (NAI, passive–active) Someone released this information but Gorbachev didn’t. (active–active) Someone should release this information but Gorbachev didn’t. (NAI, active–active)

As expected, they found that mean acceptability ratings for the mismatch cases were higher when an NAI was present (b) than when it was not (a). The pattern was reversed for the match cases: Examples like (d) were judged worse than examples like (c). Results of an online study that tracked eye movements while reading also provided evidence that cases involving NAIs are easier to process. Whereas the intuitions concerning such examples seem clear, it is important to distinguish the predictions of this analysis from others that cover the same set of data. For one, there is considerable overlap between the examples that support the coherence analysis and the NAI hypothesis, so we need to examine cases in which they make the opposite predictions. Example (a), which is a Violated Expectation case, is predicted to be good on Kehler’s analysis, and it indeed seems marginal. But is this due to the mismatched ellipsis? Consider the variant in (): ()

I know that this information was released by Gorbachev, even though most people think he didn’t.

This version, which the NAI hypothesis predicts to be as bad as (a), seems much improved, suggesting that (a) is awkward for reasons independent of the ellipsis mismatch.11 Second, we can ask what happens if we modify examples like () such that there is no longer an NAI evoked from the first clause: ()

Obviously this problem wasn’t looked into, even though someone should have. [looked into the problem]

This change alters the coherence relation from one Cause–Effect relation (Violated Expectation) to another (Denial of Preventer), and hence the example is predicted to still be acceptable on Kehler’s analysis. It should be unacceptable on Grant et al.’s analysis, however, as it is crucial to their story that it is the antecedent clause (and not the ellipsis clause) that gives rise to the NAI.12 Third, we can ask what happens in passages in which the antecedent gives rise to an NAI, but which participates in a Parallel relation. In such cases Kehler would predict mismatches 11

More on this in a moment. Grant et al. (: ) briefly address a similar case, saying that “the negation in the first clause may have the effect of introducing alternatives and thus behaving much like the cases of NAI we have discussed.” No reason is given for this conjecture, however, and indeed one has to ask why a clause that asserts that an event did not occur would give rise to a QUD that it already provides an answer to, whereas a clause that asserts that an event did occur would not. 12





to be bad, but Grant et al. would presumably predict them to be good, since we should see the same effect of raising the salience of the antecedent as we did before. Consider the variant of () in (): ()

This problem should have been looked into by John, and Bob should have too.

This version seems bad by my judgment, although confirmation will have to await experimental study.13 Finally, whereas many of the examples that motivate Kehler’s analysis involve NAIs, not all of them do. Examples include (a,b): ()

a. The problem was looked into by John, even though Bob already had. (=c) b. Actually I have implemented it [= a computer system] with a manager, but it doesn’t have to be. (Kehler )

We have already discussed example (a), which seems on par with other cases of acceptable mismatch but which does not have an antecedent that gives rise to an NAI. Example (b) is likewise at least a marginally acceptable case of mismatch, featuring an active antecedent and passive ellipsis clause rather than the other way around. At this point we have examples of the two types of approach to reconciling the facts introduced earlier: Kehler’s analysis treats VP-ellipsis as anaphoric and then appeals to coherence establishment to explain the existence of syntactic constraints, whereas the approaches of Frazier, Clifton, and colleagues maintain a syntactic analysis and then appeal to external strategies to explain the acceptability of mismatch cases. Kertz (, ) offered a different analysis that took yet another slice through the data, appearing to capture many of the aforementioned facts in a more straightforward way. According to her account, the crucial factor that determines acceptability pertains neither to the type of coherence relation between the clauses nor the existence of NAIs, but is instead based on the information-structural properties of the ellipsis clause: specifically, whether the elided clause participates in a   or auxiliary focus construction. Examples like (a), repeated in (a), display subject focus: accent falls on the subject and not the auxiliary. On the other hand, examples like (), repeated as (b), feature auxiliary focus: accent falls on the auxiliary rather than the subject. ()

a. #This problem was looked into by John, and Bob did too. b. This problem was to have been looked into, but obviously nobody did.

For Kertz, the problem with (a) does not bear directly on coherence relations, but is instead due to defective information structure. Specifically, Kertz points out that utterances participating in Parallel relations typically feature contrastive topics, and a defining feature 13

To explain the unacceptability of (), one might note that in this case the ellipsis clause does not ultimately answer the QUD that Grant et al. posit to be evoked by the NAI. However, it is hard to see how the mechanics of an explanation based on this fact could work, since the role of the NAI is to encourage the processor to repair the VP as a necessary prerequisite to determining congruence of the ellipsis clause with the QUD.

  



of a contrastive topic is being the sentence topic. Unlike syntactically matched active-voice cases of VP-ellipsis, however, in (a) the speaker has chosen to express the proposition denoted by the antecedent using a passive, which has the effect of demoting the logical subject (i.e., John) to a non-topical position in favor of another event participant. As a result, interpretation is then disrupted when, in the ellipsis clause, the processor has to recognize that it is in a contrastive topic structure and reanalyze John as a topic. There is no such issue in (b), on the other hand, where the focus in the ellipsis clause falls on the auxiliary, and hence no similar information structure principle is violated. One of Kertz’s experiments was an acceptability study that employed tough alternations rather than a voice manipulation: ()

a. Venomous snakes are easy to identify, and poisonous plants are as well. [subject focus, match] b. It’s easy to identify venomous snakes, and poisonous plants are as well. [subject focus, mismatch] c. It’s easy to identify venomous snakes, and most experienced hikers can. [auxiliary focus, match] d. Venomous snakes are easy to identify, and most experienced hikers can. [auxiliary focus, mismatch]

Kertz found main effects of both mismatch and focus, as well as the predicted interaction whereby mismatched examples were judged to be significantly worse in the subject focus condition (b) than in the auxiliary focus condition (d). Further, two follow-up studies compared judgments on mismatch cases like (d) against variants in which the ellipsis clauses contained unelided VPs (Venomous snakes are easy to identify, and most experienced hikers can identify them). These studies revealed that there is a penalty for mismatch even in the unelided versions, a crucial prediction of her theory that is unshared by previous ellipsis-specific proposals. Finally, she noted that her analysis makes a novel prediction concerning processing. Specifically, if defective topic structure is part of the reason for the unacceptability of cases with mismatch, such structures should not only disrupt processing at the ellipsis site, but also on the subject noun phrase in the ellipsis clause, since this is the point at which the defective topic structure can first be detected. A self-paced reading time study that used modified versions of (a,d) as stimuli confirmed this prediction. The fact that mismatch disrupts processing even before the comprehender could know that an ellipsis will ensue is particularly striking evidence that the source of unacceptability in such examples is partially independent of ellipsis. Whereas further work is needed to confirm or refute the predictions of the information structure analysis, it does explain a number of the puzzles we have encountered thus far, without the need to invoke coherence relations, recycling processes, or NAIs. First, consider again the contrast between examples (a) and (c), repeated as (a,b): ()

a. #This problem was looked into by John, even though Bob did. b. The problem was looked into by John, even though Bob already had.

Recall that whereas Kehler had based his analysis on cases like (b), Frazier and Clifton () used examples like (a) in their experiments. The conflicting judgments are





precisely what Kertz’s analysis predicts: (a) is a case of subject focus (nuclear accent is placed on Bob), whereas (b) is a case of auxiliary focus (nuclear accent is placed on had). Her analysis therefore makes the right slice through the data without having to posit multiple coherence relations; examples like (a) simply feature a Cause–Effect coherence relation but subject focus information structure. Second, recall that Grant et al.’s study compared judgments on example pairs like (a,b): ()

a. This information was released but Gorbachev didn’t. b. This information should be released but Gorbachev didn’t.

Here we see that Kertz’s analysis predicts the difference in acceptability as well as Grant et al.’s does, since (a) is subject focus whereas (b) is auxiliary focus. Indeed, Kertz also captures the relatively acceptable status of () for the same reason.14 Finally, there is a range of other cases that are acceptable even though the antecedent clause does not give rise to an NAI. Two cases were pointed out already in (a,b); both feature auxiliary focus. The same is true of the stimuli that Kertz used in her experiments; none of the antecedent clauses in (a,d) yields an NAI, for instance, and as such the NAI hypothesis cannot explain the relative acceptability of (d) compared to (b). Kertz’s generalization therefore appears to capture the pattern found in the mismatch data much more robustly than either the coherence analysis or the NAI hypothesis. This notwithstanding, a question remains as to how to explain why mismatched ellipsis does seem to degrade acceptability to at least some degree in all context types (cf. nn.  and ), which appears to be a robust finding across the experimental work just described. Another remaining question is how binding theory violations like (b–c), in so far as they are robust (cf. Frazier and Clifton ), would be captured by these accounts. A range of related issues pertaining to binding relations will in fact play a prominent role in the set of questions addressed in section ..

. S   

.................................................................................................................................. The previous section addressed the question of under what conditions an antecedent can license the felicitous use of VP-ellipsis. As mentioned in the introduction, a second and related question concerns the space of possible readings that a particular antecedent can 14 Grant et al. briefly address Kertz’s account (as described in Kertz ) based on the results of their Experiment b, which demonstrated that cases of VP-ellipsis that involve modals that give rise to NAIs (ib) are judged to be more acceptable than ones that do not (ia):

(i)

a. The cookies may have been made, but the babysitter didn’t. b. The cookies needed to be made, but the babysitter didn’t.

Grant et al. argue that this is a problem for Kertz, since “examples with the modal may arguably involve auxiliary focus” (: ). Their judgment does not at all accord with mine, however; I find that (ia) requires subject focus to be felicitous (i.e., if someone made them it wasn’t the BABYSITTER who did). Example (ib), on the other hand, is more natural with auxiliary focus, at least in a context in which the babysitter was the person expected to make the cookies.

  



give rise to, particularly with respect to so-called strict and sloppy ambiguities. Recall example (), repeated as (): ()

John loves his mother, and Bill does too.

Assuming that the pronoun his corefers with John, this sentence admits of two interpretations: the strict reading (Bill loves John’s mother) and the sloppy reading (Bill loves his own mother). The question that immediately arises is where this sloppy reading comes from, in light of the fact that the antecedent clause says nothing about Bill. A classic analysis posits that these readings arise from an ambiguity in the antecedent clause, specifically whereby the pronoun can have either a  or  interpretation (Keenan ; Sag a; Williams b, inter alia). These interpretations lead to the VP meanings in (a,b) respectively: () a. λx.loves(x,mother(john)) b. λx.loves(x,mother(x))

[referential] [bound]

Both meanings lead to the same interpretation for the antecedent clause, so the ambiguity only reveals itself when there is a subsequent ellipsis. On a theory in which VPs are reconstructed (or deleted under identity) at a level of syntactic representation (surface syntactic or syntactic logical form), either of these meanings is available. Using (a) will derive the strict reading for the ellipsis clause, whereas (b) will derive the sloppy reading. A number of facts follow straightforwardly from the referential/bound ambiguity analysis (henceforth RBAA).15 First, it predicts that choice of linguistic form matters: a variant of example () in which John is rementioned with a name does not license the sloppy reading:16 ()

John loves John’s mother, and Bill does too.

This is predicted by the fact that there is no possibility of a bound interpretation for the second occurrence of John. Second, no sloppy interpretation is possible if a pronoun in the antecedent refers extrasententially. That is, if his in () refers to George, then the ellipsis clause can only mean that Bill loves George’s mother—the pronoun cannot rebind in the ellipsis clause. Again, this follows directly from the lack of a bound variable interpretation for the pronoun in the antecedent. Third, when the pronoun can bind to more than one possible antecedent, sloppy interpretations must exhibit a parallel binding pattern. First consider (): ()

John said that Fred loves his mother, and Bill did too.

The RBAA is related to what Dalrymple et al. () call “identity-of-relations” analyses, a term that they, following Dahl (), use more generally for any analysis in which the possible readings in the ellipsis clause result from a parallel ambiguity in the antecedent clause. Such approaches contrast with non-identity accounts, which posit no ambiguity in the antecedent but in which ambiguities result instead from the mechanics of the VP-ellipsis meaning recovery procedure itself. 16 Of course, the antecedent clause here violates Condition C of the binding theory, but that fact is orthogonal to the question at hand. 15





If his in the antecedent refers to John, the only sloppy reading available for the elided clause is Bill said that Fred loves Bill’s mother; the interpretation Bill said that Fred loves Fred’s mother is out. The reverse is true if his refers to Fred in the antecedent clause. A fourth argument is more controversial. Specifically, the RBAA predicts that a licit binding configuration must hold between the pronoun and its antecedent, which generally requires that the antecedent c-command the pronoun. This predicts that example (), which admits only of a referential relationship between the pronoun and its antecedent, gives rise only to a strict reading: ()

John’s mother loves him, and Bill’s mother does too.

Judgments on such cases vary, however. For instance, whereas Lasnik () claims that an example similar to () only has a strict reading, Tomioka () reports that his informants find the sloppy reading acceptable.17 Rooth (a) presents the similar sentence John’s coach thinks he has a chance and Bill’s coach does too as having a marginal status. Other facts seem more problematic for the RBAA. One is what Dalrymple et al. () term  . Example () was provided by Dahl (), which he adapted from an example due to Schiebe (): ()

John realizes that he is a fool, but Bill does not, even though his wife does.

An acceptable, and perhaps preferred, reading for () is: ()

John realizes that John is a fool, Bill does not realize that Bill is a fool, even though Bill’s wife realizes Bill is a fool.

Example () contains two ellipses; the reading in () results from the second clause receiving a sloppy interpretation from the first and the third clause receiving a strict interpretation from the second. The RBAA, however, specifically predicts that reading () should not exist. Since the second clause will only have the sloppy VP meaning received from the first, the strict meaning that the third clause requires from the second is not present. A second problem is that strict readings are possible even when there is no possibility for a referential interpretation in the antecedent: ()

a. Whenever someone thinks he is a failure, his colleagues probably do too. (Dahl , adapted from an example from Schiebe ) b. Every department chair thinks that she will need to buy supplies for her office, and that the dean will too.

17

However, it may be that speaker judgments covary with those for ordinary quantificational binding, whereby the sloppy reading for (), for example, is available to just those speakers for whom () admits of a bound variable interpretation: (i)

Every boy’s mother loves him.

In this case, the facts would support the RBAA even for those speakers that get a sloppy reading for ().

  



Both cases have a strict reading in addition to the (unproblematic) sloppy one; for instance, (b) can mean the dean will buy supplies for the chair’s office or her own (see Heim  for a discussion of similar cases). However, because the pronoun his can only be bound by the quantifier in the antecedent, there is no possibility for a referential interpretation. A third problem is that sloppy readings are licensed within domains that extend further than the minimal clause that contains the ellipsis. Consider (): ()

John told a man that Mary likes him, and Bill told a boy that Susan does. (Prüst )

()

First John told Mary I was bad-mouthing her, and then Sue heard I was. (Rooth a)

Whereas him in () can be bound to the c-commanding John in the antecedent, the elided VP has no mechanism for rebinding the bound variable to the subject of the ellipsis clause in light of the intervening overt material. Rooth’s example ()—which admits of the sloppy reading Sue heard I was bad-mouthing Sue—has the same property, and further illustrates the difference between constraints on deaccenting and ellipsis: Whereas the VP was badmouthing her is elidable, the remainder of VP that contains it—heard I was—is merely deaccentable, as it can be inferred from, but is not identical with, material in the antecedent clause. Such examples have inspired a variety of analyses that resolve ellipsis by way of parallelism constraints applied to expanded parallelism domains and discourse structures (Dalrymple et al. ; Prüst ; Rooth a; Hobbs and Kehler ; Asher et al. ; Hardt and Romero ; Hardt et al. , inter alia). A fourth problem is that there are clear cases in which sloppy readings are possible in the absence of a licit binding relationship between a pronoun and its antecedent: ()

a. The policeman who arrested John read him his rights, but the policeman who arrested Bill didn’t. (Wescoat ) b. If Tom was having trouble in school, I would help him. If Harry was having trouble, I wouldn’t. (Hardt ) c. I’ll help you if you want me to, but I’ll kiss you even if you don’t. (Hardt )

Example (a) exhibits a sloppy reading despite the lack of c-command between him and John.18 Example (b) is a case in which the required dependency applies across clause boundaries. Finally, example (c) demonstrates that sloppy readings—in this case, I’ll kiss you even if you don’t want me to kiss you—can result from anaphoric dependencies at the event level.19 Whereas some requirement for parallelism is clearly at play in such examples, each has a sloppy reading despite the lack of a possible binding relationship. Other problems result when the RBAA comes into contact with treatments of related phenomena. For instance, Reinhart (a) proposed what has come to be known as the Coreference Rule, which rules out referential interpretations for pronouns in clauses when a bound reading is possible and yields an equivalent interpretation. This condition 18

Tomioka () and Elbourne () give alternative analyses of examples like (a). Schwarz () provides a detailed discussion of such examples, and Charlow () a more recent treatment. 19





eliminates the possibility of ‘accidental’ coreference in cases in which Condition B should apply (i.e., to disallow him coreferring with John in John likes him by virtue of a referential interpretation). It’s clear that the RBAA and Coreference Rule cannot both hold.20 Due in part to such issues, various approaches have used methods for identifying strict and sloppy readings without positing a referential/bound ambiguity for pronouns. Dalrymple et al. (), for example, propose a semantic analysis that resolves VP-ellipsis at the clause level. Roughly speaking, interpretation involves the solving of an equation in which the meaning of the antecedent clause as a whole is equated with the meaning of the ellipsis clause VP as applied to the meanings of the elements in the antecedent clause that are parallel to overt elements in the ellipsis clause. For example (), this identity is captured by equation (), which under suitable assumptions has (a,b) as its two solutions for the meaning P of the elided VP. () P(john)=loves(john,mother(john)) Because the method operates on purely semantic interpretations (at which binding relations and distinctions among referential forms are lost), however, it does not capture various facts we have discussed. For instance, it is insensitive to whether a binding relationship exists in generating sloppy readings (which might be a good thing or not, depending on one’s judgments for examples like ()). More importantly, it will also generate sloppy readings based on coreferential expressions even when a pronoun is not used (cf. ()) (Tancredi ; Kehler a). Fox (), on the other hand, takes a different tack. Working within the assumptions of a syntactic approach, he posits a parallelism condition on the syntactic representations of elided VPs: () NP Parallelism NPs in the antecedent and elided VPs must either a. have the same referential value (Referential Parallelism), or; b. be linked by identical dependencies (Structural Parallelism). As is no doubt apparent, options (a) and (b) respectively recreate the effects of the referential and bound options in the RBAA without locating the ambiguity in the antecedent clause.21 The problem, as Fox admits, is that the condition is purely stipulative: it does not follow from other properties of the grammar. In addition, it doesn’t work for examples like (b), where there is an ambiguity even though the pronoun does not have a referential value, and hence option (a) cannot apply.

That is unless we extend the ‘equivalent interpretation’ restriction to apply to subsequent clauses (in this case, the ellipsis clause) rather than just the one containing the binding relation, a possible but unattractive option (Büring ). 21 Kehler (b) proposed a similar system, except implemented in a version of event semantics that maintains referential dependencies between pronouns and their referents. Unlike Fox’s system, however, all such dependencies are represented, and thus the system derives sloppy readings even when there is no c-command relationship between the pronoun and its antecedent. 20

  



A more radical way of handling the data was proposed by Hardt (a, b, , ). He posits an anaphoric theory, but one in which VP meanings are copied to the ellipsis clause with any pronoun representations contained in them left unbound. These pronouns are then reinterpreted in the ellipsis clause according to normal rules of pronoun interpretation. For example, in () the semantic representation for loves his mother would be copied to the ellipsis clause and then his re-resolved in its new context, in the same manner as if the elided VP appeared overtly. The fact that both John and Bill are salient in the discourse context yields the strict and sloppy ambiguity. Note that some care must be taken to restrict the possible readings on this approach—for instance, to disallow sloppy readings when a pronoun in the antecedent clause refers extrasententially, sloppy readings that do not maintain binding parallelism in examples like (), and so forth. We will revisit Hardt’s approach in section ... These are but a few of many approaches that attempt to explain strict/sloppy ambiguities without appealing to a referential/bound ambiguity in pronouns. The problem with all of these accounts is that they each go beyond what could be seen to naturally fall out from independent properties of the grammar, i.e., all have additional machinery tailored specifically to the data at hand. This is an unsatisfactory situation. Now when it comes to building a theory from first principles, it may stand to reason that syntactic analyses are on the right track, in light of the fact that matters that bear on the linguistic form of the antecedent— e.g., choice of referring expression and binding relationships—are at least in part relevant to the question of when sloppy readings are available. It therefore seems all the more surprising that we find that strict/sloppy readings are far from limited to elliptical contexts (Dalrymple et al. ; Tancredi ; Kehler a; Hobbs and Kehler , inter alia): ()

a. b. c. d.

John revised his paper before Bill did it/that/so. (Event Anaphora) John revised his paper and then Bill followed suit. (‘Same as’ Anaphora) John called his teacher an idiot and Bill insulted his teacher too. (Deaccenting) Only/Even John revised his paper. (Focus-sensitive operators)

As there is no reason to think ellipsis is involved in any of these constructions, not only does the existence of strict/sloppy readings not specifically favor syntactic approaches to VP-ellipsis, but any approach that recovers such readings as a side effect of reconstructing (or resulting from the deletion of) syntactic material misses a significant generalization. The question then is why these other phenomena would display the same sensitivity to (presumably syntactically governed) binding relationships in the antecedent. One property common to these phenomena is that they are all sensitive to information structure, i.e., they all depend on their denotations being Given in the discourse. Indeed, the fact that these ambiguities are shared not only by event-referential forms (a,b) but also overt, deaccented VPs (c) and other focus-sensitive constructions (d) points us directly toward deaccenting theories, as we expect the other forms to also be constrained by principles governing focus marking and accent placement. Rooth (a), for instance, proposes an analysis based on such principles. In his analysis, VP-ellipsis is subject to a constraint that requires that the ellipsis clause   the antecedent clause whereby, roughly speaking, there exists a replacement for the focused elements in the elided clause that gives rise to the meaning of the antecedent. Consider how this would





work for (). Let us assume that, per the Coreference Rule, c-commanded pronouns in antecedent clauses can only be bound. Example (a) represents the strict reading. ()

a. John λ1 loves his1 mother, and Bill loves John’s mother too. b. John λ1 loves his1 mother, and Bill λ1 loves his1 mother too.

Focus matching the ellipsis clause subject Bill against the antecedent subject John will give rise to the proposition that John loves John’s mother. Since this is the meaning of the antecedent clause, ellipsis is licensed. Example (b), on the other hand, represents the sloppy reading. Focus matching will give rise to the meaning by which John loves his own mother. This also corresponds to the meaning of the antecedent clause, and so again ellipsis is licensed. As pointed out by Büring (: section .), Rooth’s analysis handles at least the simple cases in an independently motivated way and without additional stipulations about binding and parallelism.22 However, the situation is about to become more complicated.

.. Missing readings examples Dahl () famously introduced what have come to be known as   puzzles in ellipsis. Consider example (): ()

John thinks he loves his wife, and Bill does too.

Assuming that both pronouns are bound in the antecedent clause, a naive theory of strict and sloppy interpretations such as the RBAA—in which each pronoun can be interpreted strictly or sloppily—predicts the four readings shown in (a,d). () a. Bill thinks John loves John’s wife. λx.thinks(x,loves(john,wife(John))) b. Bill thinks Bill loves Bill’s wife. λx.thinks(x,loves(x,wife(x)))

(all-strict: BJJ) (all-sloppy: BBB)

c. Bill thinks Bill loves John’s wife. λx.thinks(x,loves(x,wife(John)))

(mixed sloppy–strict: BBJ)

d. #Bill thinks John loves Bill’s wife. λx.thinks(x,loves(john,wife(x)))

(#mixed strict–sloppy: BJB)

22 As Rooth notes, reliance on the focus-matching condition alone is insufficient in that it is insensitive to the forms of reference used in the antecedent clause. As such, the condition predicts (), repeated below as (i), to be acceptable on a sloppy reading.

(i)

John loves John’s mother, and Bill does too.

Rooth suggests that the principle be augmented with a second that requires syntactic identity. As we saw in section ., however, such a condition would have to be reconciled with substantial data that points away from such a requirement.

  



The all-strict and all-sloppy readings in (a,b) are clearly acceptable. Dahl’s judgments are that the mixed reading in (c) “seems rather dubious” and that (d) “is clearly not possible.” Since then, the literature has generally considered (a–c) to all be acceptable, leaving the puzzle of why (d) is not.23 To explain this fact, a number of researchers have appealed to a locality condition on pronoun binding (Kehler a; Fox ; Schlenker , inter alia). Specifically, the ability to derive all four readings for () is dependent on the antecedent clause having a binding configuration whereby both pronouns are bound directly by the matrix subject John. Note, however, that his could also be bound by the embedded subject he. Let us stipulate, as Fox (: ) does in his Rule H, that his can in fact only take this more local binder in such a circumstance: ()

Rule H: A pronoun, γ, can be bound by an antecedent, α, only if there is no closer antecedent, ϕ, such that it is possible to bind γ to ϕ and get the same semantic interpretation.

Now the missing reading cannot be derived, because the binding configuration in the antecedent that would be necessary—in which the second pronoun is bound directly to John—is no longer considered licit. With his bound locally, the combination that would have otherwise lead to the missing reading—where, using Fox’s terminology, the first and second pronouns satisfy referential (a) and structural (b) parallelism respectively— will instead derive the all-strict reading (redundantly with the scenario in which both pronouns satisfy referential parallelism), since the interpretation of the second pronoun will necessarily follow the interpretation of the first.24 In cases in which local binding is not possible between two pronouns, analyses of this sort predict that all four readings should be available: ()

John thinks his wife loves him, and Bill does too.

()

John thinks his wife loves his brother, and Bill does too.

In these cases, neither pronoun c-commands the other, and hence both can only be bound by John. As predicted, examples () and () admit of all four expected readings for most informants.25

23

An exception is Tancredi (), who offers a context which he claims allows for the fourth reading of a similar example. His judgments are at odds with mine, however. 24 There are a number of variations within this class of analyses that would take us too far afield to discuss here. See, for instance, Kehler (a), Fiengo and May (), Büring (), Schlenker (), Roelofsen (), and Drummund (), inter alia. 25 Kehler and Shieber (), however, note that example (i) only has two readings instead of the expected four: (i) John loves his mother and his father, and Bill does too. Like () and (), neither possessive pronoun in the antecedent can bind the other, but unlike those cases, the readings in which the two pronouns have disjoint reference are clearly out. Interestingly, this fact follows from the deaccenting theory, since the unelided, deaccented variant of (i) is also missing the





In light of examples (a,d) discussed previously, we are led to ask whether other phenomena also exhibit the missing readings pattern, including event anaphora, deaccenting, and focus-sensitive operators. Indeed they do (Kehler and Büring ): () a. John said that he loves his wife, and Bill did it/that/so too. b. John said that he loves his wife, and then Bill followed suit.

(Event anaphora)

(‘Same as’ anaphora) c. John said that he loves his wife, and BillF said that he loves his wife too. (Deaccenting) d. Only/Even John thinks that he loves his wife. (Focus-sensitive operators) So the data again indicate that the constraint is coming from somewhere other than ellipsis per se. For instance, the fact that the deaccented version in (c) and event anaphora in (a) are missing reading (d) suggests that perhaps the constraint is informationstructural, as again all of these forms require their antecedent to be Given in the discourse.26 This yields an odd state of affairs, however, if we consider what is Given to be that information which is entailed by the context (Schwarzschild ). That is, we are then forced to the conclusion that the antecedent of (c) renders as Given the properties said that John loves John’s wife (BJJ reading), said that oneself loves one own’s wife (BBB reading), and said that oneself loves John’s wife (BBJ reading), but not said that John loves one own’s wife (BJB reading), even though obviously all four are equally entailed by the context.27 Kehler and Büring () attempt to explain these facts with a different type of information-structural analysis, particularly one that utilizes QUDs (Roberts a; Büring , inter alia). The proposal relies on the fact that parallel clauses such as those in () are plausibly analyzed as each providing partial answers to a (usually implicit) QUD. As a result, one would only expect an antecedent/ellipsis pair to be acceptable under a particular interpretation for the ellipsis if a suitable QUD can be inferred to which each clause provides a felicitous answer. It is here, according to the analysis, that things go wrong for reading (d). Specifically, the QUD that is required to license the reading—Who1 thinks that John loves his1 wife?, where crucially, the possessive is bound by the wh-expression and not John—is one that is normally understood to be asking about people other than John himself. For instance, an answer of no one to this question can be judged true even if John mixed readings—each would require accent on at least one of the two pronouns in the ellipsis clause. It is admittedly not clear to me what the source of that constraint is, although it likely bears on the fact that the two pronouns occupy parallel positions in a coordinated noun phrase. This leads us to ask how Rooth’s (a) focus-matching constraint discussed in section . fares on these more complex examples. Unfortunately, it will straightforwardly derive all four readings. 27 The missing reading becomes available in the unelided version, however, if the second pronoun is accented: . . . and Bill said that hej loves HISb wife. This version is also compatible with the all-sloppy reading, and other variants in which one or both pronouns are accented are also compatible with various readings that are acceptable with the fully deaccented version. Since the lack of the missing reading seems to result from the fact that the second pronoun must receive accent in the unelided version, coming to grips with this phenomenon may require that we come to understand why it is that bound pronouns would ever receive accent in the first place. Sauerland (, ), Jacobson (b), Mayr (), and Keshet () offer proposals and discussion. 26

  



thinks that he loves his own wife (as long as no one else does); and indeed, answering Well John does, of course! in such a circumstance seems snarky. Hence, according to this analysis, it is the meaning of the antecedent clause of (), and not the ellipsis clause, that creates the problem for reading (d), since this meaning is presupposed to not be in the alternative set denoted by the QUD required to license the reading. Whatever the relative merits of this particular analysis, we see again how models of discourse coherence may play a role in determining the felicity of VP-ellipsis. We will return to the topic of QUDs momentarily. According to the analyses just discussed, the anaphoric dependencies that eliminate missing readings result from binding relationships. We have already seen examples (i.e., a–c) that suggest that that notion must be broadened, however. A different missing readings example that has been discussed in the literature, due to Gawron and Peters (), provides another case in which the space of readings appears to be dependent on the establishment of anaphoric dependencies, but not in a way that can be readily captured by binding relations: ()

John revised his paper before the teacher did, and Bill did too.

There are two VP-ellipses at play in (). The first (the teacher did) has the first simple clause (John revised his paper) as its antecedent. The second (Bill did too) then has the entire previous sentence as its antecedent, including the first ellipsis site. We are interested in the set of possible readings that result when his corefers with John in the first conjunct. In theory, () could have as many as six readings:28 () a. John revised John’s paper before the teacher revised the teacher’s paper, and Bill revised John’s paper before the teacher revised the teacher’s paper. (JTJT) b. John revised John’s paper before the teacher revised the teacher’s paper, and Bill revised Bill’s paper before the teacher revised the teacher’s paper. (JTBT) c. John revised John’s paper before the teacher revised John’s paper, and Bill revised John’s paper before the teacher revised John’s paper. (JJJJ) d. John revised John’s paper before the teacher revised John’s paper, and Bill revised Bill’s paper before the teacher revised Bill’s paper. (JJBB) e. John revised John’s paper before the teacher revised John’s paper, and Bill revised Bill’s paper before the teacher revised John’s paper. (JJBJ) f. John revised John’s paper before the teacher revised John’s paper, and Bill revised John’s paper before the teacher revised Bill’s paper. (*JJJB) Gawron and Peters themselves claim that () has only three readings (b, c, and d), which are those generated by their situation-theoretic analysis. Dalrymple et al. (), however, argue that there are five readings, and Kehler (b) concurs—all readings except (f) are possible, although (a) and (e) are perhaps less immediately accessible. A straightforward application of Dalrymple et al.’s analysis generates all six readings, 28

Each reading contains descriptions of four papers that were revised. The notation to the right of each example indicates the respective owners of these papers.





although they appeal to a mechanism-external constraint on relation formation to rule out (f). Fox’s system appears to generate four readings: (a–c) and (e)—missing (d), a reading that appears uncontroversially to exist. Kehler’s (b) system, while being similar to Fox’s but cast within an event semantics instead of at LF, generates the five readings in (a–e). The key difference between the accounts comes when the strict reading of the first ellipsis is recovered; unlike Fox’s system, in Kehler’s system the reconstructed pronoun will be anaphorically linked to its parallel pronoun in the antecedent clause despite there being no binding relationship, which ultimately allows reading (d) to be derived when the second ellipsis is resolved.

.. Hardt’s puzzle Although approaches to resolving strict and sloppy readings vary in their details, one thing seems to be clear: there is some sort of parallelism constraint that governs the possibility of sloppy readings. We see evidence for this in the lack of sloppy readings when a pronoun in the antecedent clause refers extrasententially, the lack of readings with non-parallel binding in (), and the missing readings cases. Whereas it may not be clear whether the requisite parallelism constraints are best stated in terms of syntax, semantics, or discourse structure, it would seem that constraints on parallelism are applied at some level. Hardt (a, b, , ) offered a collection of examples that were problematic for the early treatments of strict/sloppy ambiguities of Sag (a) and Williams (b). Most of these, including his (b), can be captured by treatments involving more extended notions of parallelism (Dalrymple et al. ; Prüst ; Rooth a; Hobbs and Kehler ; Asher et al. ; Hardt and Romero ; Hardt et al. , inter alia). One of his examples, however, provides a more direct challenge for any analysis based on either syntactic or semantic parallelism: ()

Every boy in Mrs Smith’s class hoped she would pass him. In John’s case, I think she will.

The elided clause in this example clearly admits of what would appear to be a sloppy interpretation, that is, I think Mrs Smith will pass John. Yet this reading arises without the existence of the type of syntactic or semantic parallelism between the clauses that all theories we have discussed require; in particular, the anaphoric relationship that is required in the ellipsis clause does not mirror the binding configuration between the pronoun and quantified noun phrase in the antecedent clause. Despite the fact that this example provides a clear adequacy criterion for any theory, substantive discussion of it is conspicuously absent in the literature.29 29 Fiengo and May (: , n. ) are an exception, arguing that a similar example “preclude[s] a strict reading, and so appears to be subject to somewhat different constraints on anaphoric resolution,” and that therefore “such cases . . . are irrelevant to the matters of parallelism under discussion.” It is unclear to me why an infelicitous strict reading would necessarily license an otherwise unavailable sloppy reading (as opposed to simply being infelicitous on any interpretation), and no alternative account of the example is offered.

  



Hardt relied on such examples to argue that the interpretations of pronouns in VP representations that are copied to the ellipsis clause are not generated by virtue of parallelism constraints, but instead using the same discourse principles that apply to overt pronouns. In (), for instance, the semantic representation of the VP pass him would be copied to the ellipsis clause representation with the pronoun unresolved, which would then be resolved to John via ordinary pronoun interpretation mechanisms on analogy with how we would expect the unelided version in () to be interpreted. ()

Every boy in John’s class hoped she would pass him. In John’s case, I think she will pass him.

Kehler and Shieber () subsequently argued against this proposal, however, claiming that it can only be maintained if it is assumed that the unelided versions of reconstructed VPs are fully deaccentable. They considered minimal pairs such as (): ()

a.

Every boyi was hoping that Mary would ask himi out, but the waiting is over. Last night at Bob’sb party, she asked himb out. b. #Every boyi was hoping that Mary would ask himi out, but the waiting is over. Last night at Bob’sb party, she did.

The reading where Mary asked out Bob at Bob’s party, while readily available in example (a), is not available in its elided counterpart (b). As Kehler and Shieber note, however, the intended interpretation (a) appears to require accent within the VP, specifically on the pronoun. This is predicted on standard theories of focus marking and accent placement (e.g. Schwarzschild ); whereas both the verb meaning (ask X out) and the object NP meaning (Bob) represent Given information, the result of combining the two (ask Bob out) does not, and hence some part of the VP must bear accent. As such, Kehler and Shieber argued, a deaccenting analysis already gets these facts right without the need to posit a less constrained pronoun interpretation process on top of it. The real mystery is why the VP can be deaccented in the unelided version of () shown in (), but not in (a), in light of the fact that the reasoning we gave for why the VP in (a) should carry accent should apply equally to () as well. To explain this, Kehler (, ) argued for an analysis that makes crucial reference to QUDs. Briefly, the idea is that phrases like in X’s case (to which one might add as for X, regarding X, etc.) are not ordinary prepositional phrases (cf. last night at Bob’s party in ()), but instead introduce QUDs into the discourse that are created by substituting X for a parallel entity within the meaning of an anaphorically identified, contextually salient proposition. In the case of (), for instance, the phrase in John’s case is anaphorically resolved to the antecedent clause (shown again in (a)); establishing parallelism between John and the bound pronoun and taking pass to be the matrix of the QUD then yields the question in (b). The remainder of the clause then provides an answer to this QUD. ()

a. Every boy in Mrs Smith’s class hoped she would pass him. b. Will Mrs Smith pass John? c. I think she will. [pass John]





The treatment of the ellipsis in (c) is now straightforward: Despite appearances, it is not (a) that serves as the antecedent of the VP-ellipsis, but instead the meaning of in John’s case given in (b). As such, the example doesn’t involve rebinding of a pronoun at all; indeed with this antecedent the reading is an otherwise ordinary strict interpretation. The difference between this case and () is that last night at Bob’s party is not anaphoric—it is a normal adverbial phrase—and hence does not similarly introduce a QUD. The aforementioned analogous deaccenting facts follow directly. In so far as this analysis is on the right track, it provides more evidence that analyses of VP-ellipsis must account for higher-order coherence properties of the discourse.

. C

.................................................................................................................................. Despite the fact that it has remained a focus of attention in linguistics for over forty years, the analysis of VP-ellipsis is still rich and fertile ground, with many unresolved questions. We have seen examples that point specifically to syntactic, anaphoric, and deaccenting analyses, with the data supporting each often being problematic for the other two. Further work will obviously be required to sort it all out. For now, the data in this chapter is offered as a partial set of adequacy criteria for any theory, with which future work can hopefully engage. As is fitting for the topic of this chapter, one theme we have touched on repeatedly is the idea that constraints bearing on the establishment of discourse coherence may play a role beyond those imposed by the linguistic properties of the antecedent and ellipsis clauses themselves. Kehler (), which operated within a relational theory of discourse coherence, was an early example; more recently QUD analyses are emerging as a potential source of explanatory power. We have seen a varied set of works begin to converge in this regard. For instance, although Grant et al. () posit a syntactic account of interpretation, they appeal to QUDs in their explanation for why their Non-Actuality Implicatures would play a role in processing. Kertz’s () information structure theory is also closely related, in that the difference between subject focus and auxiliary focus constructions corresponds directly to the sets of focus alternatives that are commonly considered to be the meanings of wh- and polar QUDs respectively. The same is true for Miller and Pullum’s () treatment of VP-ellipsis with situationally evoked antecedents, which builds on Kertz’s insights. In related work, Miller and Hemforth () invoke QUDs in their treatment of VP-ellipsis with nominalized antecedents; according to their analysis, such cases are acceptable in just those cases in which the nominalization functions as a concealed question. In the realm of accounting for strict/sloppy readings, Kehler and Büring () appealed to QUDs to explain Dahl’s puzzle, and Kehler () used them in his account of Hardt’s puzzle. Other work along these lines has recently emerged as well: see, e.g., Keshet () and Elliott et al. (). Whereas it is the linguist’s temptation to ascribe the behavioral properties of a linguistic phenomenon to the syntax and semantics of the form itself, sometimes the picture only clarifies when one identifies the role of independent discourse interpretation processes with which discourse-sensitive phenomena like ellipsis would be expected to interact. Having said that, our understanding of QUD models of discourse coherence is still in a relative state of infancy; indeed the study of VP-ellipsis may be as useful for developing theories of QUDs as the other way around.

  



Research on VP-ellipsis also provides an exemplar of how theoretical and experimental linguistics can profitably interact. Whereas experimental research on language is burgeoning, much of it bears on processing issues rather than the sorts of fundamental questions about the structure and meaning of language that have been the primary concern of theoretical linguistics. VP-ellipsis is rich in this regard precisely because approaches differ as to whether infelicity is predicted to be caused by the grammar or by processing, and hence it is among a small set of phenomena at the center of the grammar vs performance debate. As discussed earlier, Kertz’s self-paced reading study—in showing that processing difficulty occurs at a place that her theory predicts, but before the comprehender could know that an ellipsis will ensue—illustrates how experimentation can add to the empirical base for theory building beyond what one can contribute from one’s armchair. Perhaps what is most fascinating about ellipsis is how a process that seems so conceptually simple can give rise to such complexity in the data. This is no doubt due in part to the fact that it interacts with so many other complex things—spread across the domains of syntax, semantics, anaphora, and information structure—and hence why the benefits of studying ellipsis go far beyond gaining an understanding of the phenomenon itself. Indeed, studying ellipsis is so exciting precisely because it gives us a window into big questions across such a wide range of language phenomena.

  ......................................................................................................................

    ......................................................................................................................

 

. I

.................................................................................................................................. C linguistics differs from other approaches to linguistics in that it always keeps in view a computational implementation. This can provide a useful and different perspective on ellipsis, both because the computational perspective can bring different problems to the fore, and also because the emphasis on computational techniques can result in new ways to approach these problems. For example, in the approach of Dalrymple et al. (), the higher-order unification techniques used in automated theorem proving and related fields provided direct inspiration for a new and fruitful perspective on the resolution of ellipsis. Also, the requirements of a concrete application require consideration of issues that have otherwise been ignored, such as the identification of ellipsis occurrences and the identification of the antecedent. In what follows, I begin with the structure of the ellipsis site—a topic that has received little attention in computational linguistics. I turn then to the question of recoverability: here there are two prominent accounts, that of Lappin and McCord () and Dalrymple et al. (). The topic of licensing follows. This topic has not been directly addressed in the computational literature, but I consider two related issues of direct computational interest: the identification and generation of ellipsis occurrences. Finally, I consider three additional topics: identifying the antecedent, dialogue, and text mining.

. S    

.................................................................................................................................. Many theorists (Fiengo and May ; Merchant ) have posited elaborate syntactic structure at the ellipsis site, while others have argued for a WYSIWYG approach (see Culicover and Jackendoff  and references therein), where there is no implicit

   



structure at the ellipsis site. This question has not received much direct attention in the computational linguistics literature. There is a general reluctance in computational approaches to posit implicit syntactic structure, and indeed theorists who argue against implicit structure sometimes argue that their position is more congenial to computational approaches. The account of Dalrymple et al. (), which will be treated in depth in section .., is often named as a prominent representative of the no implicit structure view.1 However, Dalrymple et al. () do not stake out a clear position here: in fact, they note: “The incorporation of an ellipsis analysis such as ours into a transformational framework is quite conceivable, merely requiring the ability to form abstractions over the syntactic objects representing semantic construals of sentences, that is, LF trees” (p. , n. ). Thus it appears that Dalrymple et al. () are agnostic on the question of implicit structure in ellipsis. One exception to this general agnosticism is Lappin and McCord (), who propose “an analysis of VP anaphora that involves applying rules of interpretation directly to S-structure (parsed surface structure)” (p. ). Thus Lappin and McCord argue that there is implicit syntactic structure, and they go on to present an algorithm for resolving VP-ellipsis that relies on the fact that the ellipsis site contains syntactic material. In particular, the VP-ellipsis resolution algorithm interacts with filters on pronoun resolution based on syntactic constraints on binding.

. R

.................................................................................................................................. There are two general computational approaches to recoverability: copying, represented primarily by Lappin and McCord (), and unification, represented primarily by Dalrymple et al. ().

.. Copying Lappin and McCord () present “an algorithm for interpreting elliptical VP’s in antecedent-contained deletion structures, subdeletion constructions, and intersentential cases” (p. ). They describe a complete implementation of the algorithm, which operates on the output of a Slot Grammar parser. The algorithm consists of three parts, A, B, and C, which we consider in turn. Part A identifies an elliptical-verb–antecedent pair . Since Lappin and McCord are describing a fully implemented system for VPE interpretation, they must address the VPE identification and antecedent location problems that are not addressed by most theoretical accounts, including computational accounts such as Dalrymple et al. (). The VPE “is identified by the presence of an auxiliary verb or infinitival ‘to’ without a

1

For example, Kennedy () gives Dalrymple et al. () as an example of an approach that claims that “elided constituents have no syntactic representation at all.”





realized verb complement” (p. ). The antecedent A is defined as a non-elliptical verb, occurring in one of three configurations relevant to the VPE: . VPE is contained in a clausal complement to a clause containing A. . VPE is contained in a relative clause to a clause containing A. . VPE is contained in the right conjunct of a sentential conjunction S, and A is contained in the left conjunct of S. The first two conditions describe cases where the VPE clause is in the same sentence as the antecedent, while in the third condition they are in adjacent, conjoined sentences. Hardt (a) points out that these configurations are incomplete, giving numerous examples from the Brown Corpus that do not satisfy any of these conditions. For example, Hardt notes that in  percent of the examples in his data, at least one sentence intervenes between the antecedent and VPE. Furthermore, when the VPE and the antecedent are in the same sentence, there are many cases that are not covered by the first two conditions, such as Hardt’s example (), repeated here as (): ()

In the first place, a good many writers who are said to use folklore, do not, unless one counts an occasional superstition or tale.

Here the VPE is the matrix VP, while it is the antecedent use folklore that is contained in a relative clause. One unusual aspect of Lappin and McCord’s approach is that they offer a characterization of the possible VPE–antecedent configurations, instead of characterizing the impossible configurations. This is the approach taken, for example, in Hardt a; in the theoretical literature a common view is that the impossible VPE–antecedent configurations are similar to the impossible pronoun–antecedent configurations. For example, Kim () points out that VPE obeys the Backwards Anaphora Constraint (Langacker )—it can precede, but not command its antecedent. Similar observations are made in Lobeck () and Hardt (). Thus, while Lappin and McCord’s account is notable in being a fully implemented system, and therefore must confront the issue of antecedent location, it is too limited in the configurations that it permits. Furthermore, it does not address the issue of ambiguity: when more than one potential antecedent occurs, some means for selection is required. We turn now to the central part of Lappin and McCord’s algorithm, steps B and C. Here the missing material at the ellipsis site is filled in: first, the missing head V of the VPE is copied from the antecedent. Then, each argument for V is also copied from the antecedent, unless an argument is already present at the VPE site. The same is done for adjuncts. Since Lappin and McCord allow the possibility that an argument is present at the VPE site, it seems incorrect to describe this as VP-ellipsis, since in such cases what is elided is not a VP, as illustrated by Lappin and McCord’s (a), shown here as (): ()

John writes more books than Bill does articles.

Lappin and McCord use the term subdeletion to describe such examples, which more typically are described as pseudogapping. There is an admirable simplicity to this approach: for each argument or adjunct of the antecedent, it is copied to the ellipsis site unless it

   



is already present. However, there are known to be differences in the distribution of pseudogapping and VPE, and these differences are not accounted for here. This is particularly relevant for the characterization of VPE–antecedent configurations, which, as I noted above, are too restrictive for VPE. These configurations are also meant to apply to pseudogapping, and here the configurations are not restrictive enough. For example, Hardt (b) points out that backwards pseudogapping does not occur—the antecedent must precede the ellipsis site. Furthermore, while the VPE can be separated from the antecedent by one or more sentences, this is not possible for pseudogapping. The VPE algorithm can be combined with two other algorithms implemented by Lappin and McCord: a syntactic filter on pronominal anaphora and an anaphor binding algorithm. The result is an “Integrated System for Anaphora Resolution” (p. ), which can be illustrated by Lappin and McCord’s examples () and (), given here as () and (): ()

John talked to him, and Bill did too.

()

The girl will write a book about herself, and Mary might too.

In (), Lappin and McCord’s system ensures that him does not corefer with John in the first conjunct, and that him does not corefer with Bill in the second. In (), the system ensures that herself corefers with the girl in the first conjunct, and that herself corefers with Mary in the second. In constructing this integrated system, Lappin and McCord seek to show that the ellipsis site contains syntactic material that is subject to ordinary syntactic constraints, such as the constraints on pronominal and anaphor binding. This became a controversial topic in subsequent literature: Kitagawa () makes a somewhat similar argument for copying syntactic material from antecedent to ellipsis site, to which binding theory constraints apply. Fiengo and May () also claim that ellipsis sites contain syntactic material to which constraints on binding can apply. However, Fiengo and May note many cases where such constraints do not apply to elided material in this way, and as a result, introduced some complex machinery to attempt to account for these observations, including the mechanism of “vehicle change.” The issue of the extent to which binding constraints apply to elided material remains controversial.

.. Unification The unification-based approach to ellipsis, originally introduced by Dalrymple et al. (), is probably the most influential computational contribution to the study of ellipsis.

... Ellipsis: An abstract characterization Dalrymple et al. () begin with what they call an “abstract statement of the ellipsis problem”, starting with (). ()

Dan likes golf, and George does too.





Referring to (), Dalrymple et al. () give the following characterization of ellipsis: The sentence is interpreted as meaning that Dan and George both like golf. The source clause, ‘Dan likes golf ’, parallels the target ‘George does too’, with the subjects ‘Dan’ and ‘George’ being parallel elements, and the VP of the target sentence being vestigially represented by the target phrase ‘does too’. (p. )

This establishes some important terminology for Dalrymple et al. (): • Source Clause: the antecedent clause for the ellipsis occurence • Target Clause: the clause containing the ellipsis • Parallel Elements: the source and target clauses contain one or more pairs of parallel elements. Parallel elements typically share the same grammatical role, but this is not necessary. Dalrymple et al. () go on to describe “the problem of ellipsis interpretation” as follows: Given this abstract view of ellipsis, the problem of ellipsis interpretation is just to recover a property of (or relation over) the parallel element (respectively, elements) in the target that the missing or vestigial material stands proxy for. Of course this property is not arbitrary. We know that the application of the property or relation to the parallel elements in the source constitutes the interpretation of the source clause. In example () above, we know that the property P being predicated of George in the second sentence is such that when it is predicated of Dan, it means that Dan likes golf.

Dalrymple et al. () state this with an equation, as follows: ()

P(dan) = like(dan, golf )

They go on to point out that a possible value for P is the property represented by the lambda term λx.like(x, golf ). The general, abstract problem of ellipsis is given as follows: ()

P(s1, s2, . . . , sn) = s

Once P is determined, the interpretation of the target clause is P(t1, t2, . . . , tn), where t1 through tn are the parallel elements in the target clause. This is the abstract characterization of the ellipsis problem offered by Dalrymple et al. (). Remarkably, it essentially constitutes their entire analysis, because, as they show, the equational statement completely determines a set of interpretations for target clauses.

... Generating interpretations with unification In general, unification provides a technique for determining the values of variables in equations. In the case of ellipsis, as in the general equation given in (), the value of the variable is a lambda term, in particular, a function from one or more individuals to a

   



sentence meaning type (for example, a truth value). Dalrymple et al. () refer to Huet (), which provides a method to enumerate solutions for such equations—that is, possible substitutions for P, which represents the ellipsis. Thus for () the solution for P is this: λx1, x2, . . . , xn.s[s1/x1, s2/x2, . . . , sn/xn]

There are two steps to the construction of such a solution: first, each parallel element must be abstracted over, giving lambda-bound variables x1, x2, . . . , xn. Second, the body of the lambda term is the same as s, except that each parallel element is replaced with the corresponding lambda-bound variable. A complication arises when a parallel element occurs more than once in s. If there are multiple occurrences, at least one occurrence must be substituted with the lambda variable. The other occurrences may or may not be substituted. This issue of multiple occurrences is crucial, because it provides the basis for strict and sloppy readings in ellipsis. This is illustrated with (): ()

Dan likes his wife, and George does too.

We assume that the interpretation of () is: () likes(dan, wife-of (dan)) As Dalrymple et al. point out (: ), there are in principle four possible substitutions for P that solve the equation in (): () . . . .

λx.likes(dan, wife-of (dan)) λx.likes(dan, wife-of(x)) λx.likes(x, wife-of(dan)) λx.likes(x, wife-of(x))

The first two solutions do not correspond to attested readings. To capture this, Dalrymple et al. () introduce the notion of primary occurrences, and introduce the primary occurrence constraint which states that primary occurrences must be abstracted over. In the first two solutions for (), the underlined occurrences of dan are primary occurrences that are not abstracted over, and thus they are ruled out. For Dalrymple et al. (), a primary occurrence is a term that is directly associated with a source parallel element. However, neither of the key terms, directly associated and parallel element, are given a formal definition, and they note that “Intuitively, the distinction between primary and secondary occurrences is clear: . . . At present, however, we have no way of making [this] precise . . . ” (p. ).2 I will return to this issue below.

2

See Gardent and Kohlhase () for a proposal using Higher-Order Coloured Unification to provide a general theory for the primary occurrence constraint.





To produce the interpretation for the target clause, we select one of the two remaining solutions ( or  in ()), and apply the property P to the parallel element, George in the target clause, as follows: ()

λx.likes(x, wife-of(dan))( george) = likes( george, wife-of(dan)) (George likes Dan’s wife) λx.likes(x, wife-of(x))( george) = likes( george, wife-of( george)) (George likes George’s wife)

Dalrymple et al. (: ) go on to present the following algorithm to produce solutions under certain restrictions (“second-order matching”):3 Given an equation of the form: P(s1, s2, . . . , sn) = s0 Construct a term s from s0 by replacing zero or more instances of the si by xi. For each such s, construct a possible binding for P given by λx1 . . . λxn.s. Clearly, there are at most c such unifiers . . .

This algorithm provides the set of possible readings for ellipsis occurrences. Next we will see how it differs from previously proposed approaches.

... Comparison with “Identity of Relations” accounts Dalrymple et al. () contrast their account with what they term “Identity of Relations” accounts; the most prominent example at the time of their writing was undoubtedly that of Sag (a). In an important sense, Sag’s account actually involves the same algorithm as Dalrymple et al. (): first a lambda abstract is produced—for Sag, this is performed by the Derived Verb Phrase rule, which always produces a lambda abstract with a single lambda-bound variable, associated with the subject. Next, there is an optional Pronoun Rule, where indexed pronouns can be replaced with a lambda-bound variable. If we look back at example (), it’s easy to see that Sag’s algorithm produces the same two solutions as the Dalrymple et al. () algorithm above. There are, however, two key differences between Dalrymple et al.’s () account and Sag’s: these have to do with generality and directionality. Generality. Sag imposes two requirements in the application of his algorithm: . There is a single parallel element: the Subject; . The variable to be solved for is the VP meaning. Dalrymple et al. () do not impose these requirements; from their perspective, these conditions are typically (but not necessarily) met in the case of VP-ellipsis. By relaxing these requirements Dalrymple et al. () can apply their algorithm to other forms of 3

An anonymous reviewer points out that Stirling () has shown that higher-order matching is decidable, so the restriction to second-order matching is no longer required.

   



ellipsis, and also capture certain cases of VP-ellipsis where other parallel elements are required, such as Dalrymple et al.’s (: ) example (b), given here as (): ()

The person who introduced Mary to John would not give her his phone number, nor would the person who introduced Sue to Bill. (Wescoat )

The reading of interest is this: ()

The person who introduced Mary to John would not give Mary John’s phone number, and the person who introduced Sue to Bill would not give Sue Bill’s phone number.

For Sag’s account, the sloppy reading is not possible—the only way to derive a sloppy reading is to replace a pronoun with a lambda-bound variable, but only when the pronoun corefers with the subject. Here there are two pronouns, her and him, which get sloppy readings. But neither of them corefer with the subject. Dalrymple et al. () show how the desired reading can be derived here, with parallel elements other than the subject. The first clause receives the following representation: refuse(pwi(m, j), give(m, phone(j)))

It is determined that there are three parallel elements: in the source clause, they are ‘the person who introduced Mary to John’, ‘Mary’, and ‘John’. The equation is therefore: P(pwi(m, j), m, j) = refuse(pwi(m, j), give(m, phone(j)))

The desired sloppy reading is given by the following solution for P: λx.λy.λz.refuse(x, give(y, phone(z)))

When applied to the parallel elements of the target clause, we get the desired reading: P(pwi(s, b), s, b) = refuse(pwi(s, b), give(s, phone(b)))

Directionality. We arrive now at what I consider the most important contribution of the Dalrymple et al. () approach: it is the first account in which the problem of ellipsis is approached in what I call a backward-looking fashion. Previous accounts, including that of Lappin and McCord discussed above, and the influential account of Sag (a), are what I call forward-looking accounts. What I mean by a forward-looking account is that the relevant interpretative decisions are made in the construction of the antecedent, which are then carried forward in providing the interpretation of the ellipsis. In particular, the Pronoun Rule in Sag’s account is applied in the construction of the antecedent, and this decides whether a given pronoun will receive a strict or sloppy interpretation under ellipsis. In the backward-looking approach of Dalrymple et al. (), the decision of strict or sloppy is not made in constructing the antecedent, but is only made as part of the interpretation of the ellipsis.





Forwards vs backwards might seem like a distinction without a difference, or perhaps merely a matter of processing. And indeed, in many cases the predictions are the same, as we saw with example (). However, Dalrymple et al. () demonstrate convincingly that the forward-looking approach is too restrictive in certain key cases—these are cases where a given clause serves as source for more than one elliptical target clause. Here there is a clear difference in the prediction of the forward-looking vs backward-looking approach. This is shown for two key examples in Dalrymple et al. (): the case of “Cascaded Ellipsis” (p. ) and their example () (p. ). In their discussion of these examples, Dalrymple et al. () show that forward-looking accounts such as Sag’s are insufficiently flexible. Consider Dalrymple et al.’s () example (), given here as (): ()

John realizes that he is a fool, but Bill does not, even though his wife does.

This example is originally due to Dahl (), who points to the following reading, which I will term the mixed reading: ()

John realizes that John is a fool but Bill does not realize that Bill is a fool, even though Bill’s wife realizes that Bill is a fool.

Two other readings are possible, which I will term across the board strict () and across the board sloppy (). ()

John realizes that John is a fool but Bill does not realize that John is a fool, even though Bill’s wife realizes that John is a fool.

()

John realizes that John is a fool but Bill does not realize that Bill is a fool, even though Bill’s wife realizes that Bill’s wife is a fool.

The forward-looking approach of Sag only permits the across the board strict and sloppy readings. On the mixed reading, the pronoun he first gives rise to a sloppy reading and then a strict reading. For Sag’s approach this would not be possible. If the Pronoun Rule applies in the derivation of the source clause, then we have the across the board sloppy reading. If the Pronoun Rule does not apply, we have the across the board strict reading. This argument applies not only to Sag’s approach, but indeed to any approach in which strict and sloppy readings are tied to a specific choice made in the derivation of the source clause. In my view this constitutes a compelling argument that readings in ellipsis are determined in a backward-looking fashion—a forward-looking account simply cannot provide the necessary flexibility for the “cascaded ellipsis” example. It should be noted that Dalrymple et al.’s () account appears to overgenerate with respect to this example, permitting other mixed readings, such as: ()

John realizes that John is a fool but Bill does not realize that Bill is a fool, even though Bill’s wife realizes that John is a fool.

   



This reading is easily generated, if the first clause functions as the source clause for both ellipsis clauses (this is pointed out by Fiengo and May ). One counterargument to this is that there is a requirement or strong preference that the second clause functions as the source for the third clause, because of a general preference for more recent antecedents. While no such restrictions are mentioned by Dalrymple et al. (), one might imagine that such preferences or constraints might arise independently from a model of discourse. I will say more below about how the Dalrymple et al. () account is best integrated with a model of discourse.

... Non-elliptical counterparts In characterizing the interpretive possibilities for ellipsis examples, Dalrymple et al. () fail to address what might seem to be an obvious question: what are the interpretive possibilities for similar examples that do not involve ellipsis? At the time when Dalrymple et al. () was published, they were not alone in this omission—this was essentially a problem with the entire literature on ellipsis up to this point. To my knowledge, it is Tancredi () that first clearly brings this issue to the fore. Tancredi points to examples like the following: ()

a. Johni said hei is a genius because Bill did. b. Johni said hei is a genius because Bill said he’s intelligent.

The italicized said he’s intelligent is deaccented, because, Tancredi argues, it is sufficiently salient in prior discourse. Tancredi points out that accounts such as that of Dalrymple et al. () correctly characterize the two readings of the ellipsis example in (), but do not apply to the apparently very similar case involving deaccenting, where one observes the exact same strict/sloppy ambiguity options for the embedded pronoun he in both examples: it can be strict or sloppy. And this would appear to hold for many or perhaps all of the ellipsis examples considered by Dalrymple et al. () and prior literature. Thus Tancredi argues that any account, like that of Dalrymple et al. (), that is specifically tailored to ellipsis occurrences, is missing a generalization. This suggests a revisioning of the Dalrymple et al. () account: rather than an account of ellipsis occurrences, it could be thought of as a way of applying higher-order unification to determine interpretive possibilities for related clauses in discourse more generally. And this is perhaps the primary contribution of Dalrymple et al. ()—the argument for a backward-looking approach to the interpretation of parallel structures in discourse, within a unification-based framework. This perspective has been a fruitful one in subsequent literature, with works such as Prüst et al. (), Hobbs and Kehler (), Asher et al. (), and Hardt et al. () all taking the general approach that the interpretive possibilities of a target clause can be characterized by applying a unification-like mechanism to determine common material. Possible readings, as well as preferences among them, can be seen as side effects of applying the unification mechanism. A discourse process of this sort has the additional benefit of providing a way to address the important issue of parallel elements. While this is left unspecified by Dalrymple et al. (), these discourse approaches all provide mechanisms for this, as a part of the general process of determining a more general semantic structure.





. L

.................................................................................................................................. What factors must be present in the surrounding environment to permit an ellipsis occurrence? While this question has not been directly addressed in the computational literature, some closely related topics have been addressed. In this section I look at identifying ellipsis occurrences and the generation of ellipsis.

.. Identifying ellipsis occurrences For any computational application, the identification of ellipsis occurrences is a prerequisite for any further processing of ellipsis. Thus the study in Hardt (a) concerning identification of ellipsis antecedents includes a simple approach to the identification of ellipsis occurrences. This work applies to a corpus of WSJ texts from the Penn Treebank, and uses Treebank parse tree patterns to identify ellipsis occurrences. This approach was only partially successful: on a small sample Hardt (a) reports recall of  percent and precision of  percent. Nielsen () improves on this, extending Hardt’s patterns with POS tags and lexical information to achieve recall of . percent and precision of . percent—a considerable improvement over Hardt’s results. Bos and Spenader () report on an annotated corpus of the complete Wall Street Journal Corpus distributed with the Penn Treebank. This is a resource that can be used to study VP-ellipsis phenomena including VP-ellipsis resolution systems. Bos and Spenader () make some interesting observations about the frequency of various constructions. They note that comparative and equative constructions are very frequent, constituting  percent of the data, although such constructions are not prominent in theoretical discussions of VP-ellipsis. They point out that “the standard example of VPE in theoretical work consists of two sentences conjoined with and, where the second sentence is marked by the presupposition trigger too” (p. ). But this structure is actually quite infrequent in the data, with only five cases. They also note that the strict–sloppy ambiguity which has received so much attention in the theoretical literature is virtually entirely absent: only nine examples include a bound pronoun in the source clause, and in all nine cases, the sloppy interpretation is strongly preferred according to Bos and Spenader (). Of course, as Bos and Spenader () point out, the fact that some of these VPE types are infrequent doesn’t necessarily mean they are not worth studying. But surely an awareness of the actual distribution of VPE types is potentially valuable, and perhaps could provide theoreticians with some new perspectives. Fernández et al. () report on a study of sluicing in the BNC, focusing on what they call “bare sluices”, i.e., sentences that consist of a single wh-word (or a phrase of the form which N). They use regular expressions to identify  bare sluices in the approximately  million word dialogue transcripts of the BNC. Fernández et al. () point out that two very common types of such bare sluices in dialogue have not received much attention in the theoretical literature, where focus has been on embedded sluices. Fernández et al. () define the direct sluice as one “used in queries to request further elucidation of quantified parameters,” and the reprise sluice is “used to request clarification of reference of a

   



constituent in a partially understood utterance.” Fernández et al. () give (a) and (b) to illustrate direct and reprise sluices: () a. Direct Sluice Cassie: I know someone who’s a good kisser. Catherine: Who? b. Reprise Sluice Sue: You were getting a real panic then. Angela: When? Fernández et al. () report that direct and reprise sluices constitute over  percent of all sluices in the BNC.4 Fernández et al. () manually annotate the sluice category for  bare sluices. Using machine learning techniques, Fernández et al. () produce a classifier with success rates of approximately  percent.

.. Generation Hardt and Rambow () explore the generation of VPE, and “present conditions under which verb phrases are elided based on a corpus of positive and negative examples” (p. ). Using data from the Penn Treebank, Hardt and Rambow () draw on positive examples from previous work (Hardt a), and for negative examples they identify cases where ellipsis had not occurred, although two VPs with the same meaning appeared within reasonably close promixity. A variety of syntactic and discourse factors were explored using machine learning techniques, and it was found that “the decision to elide VPs is statistically correlated with several factors, including distance between antecedent and candidate VPs by word or sentence, and the presence or absence of syntactic and discourse relations” (p. ). They argue, “These findings provide a strong foundation on which to build algorithms for the generation of VPE.” They go on to examine where the decision to elide should occur in the generation process, and they argue that the decision should not be made in the text planning phase, but rather in the sentence planner or realizer phase. This is because the decision appears to depend on syntactic information that would not be available to the text planner.

. O 

..................................................................................................................................

.. Antecedent location Hardt (a) reports on a study of  examples of VPE from the Penn Treebank, describing a system for determining the antecedent of ellipsis occurrences. The system applies to parsed sentences from the Penn Treebank, and the goal of the system is to determine the syntactic material that constitutes the antecedent for a given VPE example. 4

In addition to Direct and Reprise, the categories of bare sluices include Clarification, Unclear, and Wh-Anaphor.





The system determination is compared with the determination of a human coder, with different levels of strictness: . Head Overlap: either the head (i.e., the main verb) of the system choice is contained in the coder choice, or the head of the coder choice is contained in the system choice. . Head Match: the system choice and the coder choice have the same head. . Exact Match: the system choice and the coder choice are exactly the same. Hardt reports results of . percent, . percent, and . percent on these three measures, and notes that a baseline approach of always picking the most recent antecedent achieves much lower results of . percent, . percent, and . percent. The antecedent location system filters potential antecedents with a syntactic filter that eliminates potential VP antecedents that contain the VPE occurrence in a sentential complement, as in the following example: ()

She said she would not VPE.

Here, the VP headed by “said” is ruled out as a potential antecedent for the VPE occurrence. Remaining candidates are then ordered according to the following preference features: . . . .

Recency Clausal Relations Parallelism Quotation

The Clausal Relation preference is illustrated by Hardt’s example (), given here as (): ()

All felt freer to discuss things than students had VPE previously.

Here the correct antecedent, the VP headed by felt, is modified by the comparative clause containing the VPE. The VP headed by discuss is actually closer to the VPE, but the Clausal Relation preference causes the correct antecedent to be selected. Hardt suggests that there is a general preference for similar Parallel Elements, and indeed such a preference is either explicit or implicit in works such as that of Dalrymple et al. (), Asher (), Prüst et al. (), and Hobbs and Kehler (). A typical VPE configuration contains a subject and an auxiliary and sometimes other elements; Hardt suggests that there is a preference for similarity for each element’s parallel counterpart. Hardt only implements a preference for similar or identical auxiliaries; similarity of subjects is also suggested, but was not implemented. The Quotation preference is that a VPE within quotation prefers an antecedent that is also quoted. Overall, Hardt points out that the success rate of his system for VPE compares favorably with pronoun resolution systems. The preference factors have interesting relationships to discourse factors: the similarity of parallel elements is a basic feature of many leading theories of discourse relations and the existence of a syntactic clausal relation is one way to signal a discourse relation. Nielsen () builds on this work, describing an end-to-end solution that identifies VPE occurrences, and locates the antecedent.

   



.. Dialogue Dialogue systems have long been an important area for computational linguistics, and there has been an awareness that ellipsis occurs frequently in dialogues. Carberry () examines intersentential sentence fragments in the context of task-related dialogues, arguing that discourse context and conversational goals play an important role in interpreting such fragments. Frederking (: ) notes that “Natural language interfaces need to be able to interpret elliptical utterances, both because they tend to be quite frequent in dialogue and because they allow briefer and more natural interactions.” The emphasis on dialogue can provide a fresh perspective on the study of ellipsis, because it brings to light frequently occurring forms that haven’t received much attention in the theoretical literature, and these different forms can help to inform the theoretical debate in new ways. As was noted in section .., Fernández et al. () found that “bare sluices” like Clarifications are quite frequent in the BNC, and in an examination of the dialogue form of Clarification ellipsis, Ginzburg and Cooper () raise fundamental issues about the nature of utterances and semantic representations.

.. Text mining Chae et al. () describe a method for resolving ellipses in coordinated NPs, for the purpose of disambiguating Named Entity mentions in the biomedical domain, and they point out that Named Entity Recognition is important in many text-mining applications. An example of an elliptical coordinated NP is alpha- and beta-globin, where the intended interpretation is alpha-globin and beta-globin. Bergsma et al. () also investigate the problem of elliptical coordinated NPs. They report on an approach in which systems are trained on the Penn Treebank, the multilingual Europarl texts, and Google N-grams. As Bergsma et al. () point out, “ . . . if an Internet search engine is given the phrase rocket attacks as a query, it should rank documents containing rocket and mortar attacks highly, even though rocket and attacks are not contiguous in the document” (p. ). These recent works are interesting in perhaps suggesting some new general directions for ellipsis research in computational linguistics. They take as their point of departure concrete practical challenges, such as the optimization of search queries and the improvement of parsing accuracy. Furthermore, in the case of Bergsma et al. (), they address the problem using machine learning techniques applied to large collections of data.

. C

.................................................................................................................................. The perspective of computational linguistics has provided significant insights in the investigation of ellipsis over the past several decades. The higher-order unification approach has brought important empirical material to the fore, and also has given rise to a backward-looking approach to ellipsis which I believe has been a very fruitful one. Furthermore, computational linguistics has brought focus on problems that have otherwise





been given little attention, such as the identification of ellipsis occurrences and the identification of their antecedents. Most of the work described in this chapter focuses on VP-ellipsis, which is also the most well-studied form of ellipsis in the theoretical literature. As we have seen, VP-ellipsis is fairly infrequent in English, and indeed does not seems to exist in most of the world’s languages. On the other hand, sluicing is a very widespread form of ellipsis—as Fernández et al. () report,  percent of all wh-queries are direct or reprise sluices in the BNC. Thus it makes sense for computational linguistics researchers to focus more on sluicing than on VP-ellipsis; the analysis and processing of elliptical questions is a topic that has evident practical importance for NLP systems. It is not always clear what to conclude from distributional facts about ellipsis phenomena: for example, I find it surprising that the strict–sloppy ambiguity in VP-ellipsis is apparently absent from the entire million-word Penn Treebank, but of course that does not make it a less interesting object of study for theoretical linguistics. But surely such distributional analyses can be relevant for theoretical investigations. Consider the ongoing corpus development described by Anand and McCloskey (), where they are in the process of extracting and annotating  sluicing examples from the Gigaword Corpus. Anand and McCloskey () report that, while only  percent of the data has been annotated, they have “already encountered phenomena of real theoretical interest” (p. ). For example, they describe a previously unknown phenomenon of modal mismatch, where the interpretation of the elided material requires a modal that is different from the corresponding modal in the antecedent, as in Anand and McCloskey’s (), shown as (): ()

Texas A&M coach Tony Barone unabashedly predicted that . . . the Aggies could be better than a year ago. He just forgot to say when .

Here there is a shift from could to would. This kind of difference between antecedent and ellipsis has not been previously observed, and Anand and McCloskey () argue that it presents a challenge for leading theories. It seems to me that this represents a promising direction for future work on ellipsis in computational linguistics: focusing on frequent types of ellipsis with a large data collection effort. Not only does this sort of work have the potential to provide new insights of theoretical relevance, but it provides the kind of empirical foundation necessary to develop systems to process ellipsis in ways that have real practical relevance.

  ......................................................................................................................

                 ......................................................................................................................

 

. I

.................................................................................................................................. T study of ellipsis and prosody explores the interrelatedness of the presence and the absence of structure and sound in relation to meaning. The term prosody is most frequently described as “the organizational structure of speech” (Beckman : ; Speer and Blodgett ). It has been defined in terms of three independent factors of the phonological representation: “intonation, phrasal rhythmic patterning and prosodic phrasing” (Selkirk : ). Prosody, thus, also constitutes a cover term for intonation that more narrowly refers to the pitch movement (Beckman and Venditti : ). Prosody and intonation affect the interpretation of sentences (Beckman and Pierrehumbert ; Pierrehumbert and Hirschberg ; Féry ; Ladd ; Wagner and Watson ). The claim is that prosody plays an important role in the interpretation of elliptical utterances and bridges the gap between what is overtly expressed and what is understood, as is shown in () and (). The omission of linguistic material is marked by an underscore, and pitch accents by capitals. () On the principle that ONE swallow doesn’t make a summer, but TWO _ probably DO _. (Walters : ) () In der Folge deutet Stoiber erneut vor den Abgeordneten an in the following indicates Stoiber again to the members-of-parliament PRT [dass er bei der Landtagswahl  kandidieren WILL, aber nicht _ MUSS.] that he in the elections  run want, but not must ‘In what follows, Stoiber once again indicates to the members of the parliament that he wants to run in the  elections, but that he doesn’t have to.’ (News commentary,  January ) This discussion of ellipsis and prosody is couched in the phonological model of intonation and prosodic structure developed by Pierrehumbert (), Beckman and Pierrehumbert (), Pierrehumbert and Hirschberg (), Hirschberg (), and others, as one strand of autosegmental phonology (cf. Goldsmith ; Ladd ). The general phonological system proposed there has been further developed in an analysis system by Beckman and





Elam (), known as the Tones and Break Indices (ToBI) analysis system for English (cf. also Beckman et al. ; Brugos et al. ), and GToBi by Grice, Baumann, and Benzmüller () for German and by Jun () for other languages. The ToBI prosodic system implements two basic functions of prosody: (i) pitch accent assignment and (ii) prosodic phrasing. Pitch accents mark the most prominent segments in an utterance, which are often also perceived as focused. Prosodic phrases, usually delimited by prosodic boundaries (breaks or pauses), divide speech into sense units (cf. Selkirk , ). Together they make up the prosodic contour, often also referred to as intonational contour. More precisely, an intonational contour consists of only two phonemic tones, H(igh) and L(ow). The prosody of a sentence containing a gap, such as example (), is made up of several possible sequences of pitch accents (H*, L*, L+H*, L*+H, H+L*, H*+L). The target of the pitch accent, indicated by a * in ToBI, is associated with the stressed syllable of a lexical element to which it is assigned (according to certain rhythmical and focus-structural considerations) and functions as an anchoring point for the fundamental frequency (f-) contour, measured in Hertz. Another reference point for the phonological description is the phrase-final boundary tones, indicated by the percentage sign %, as in H% and L%. The phrase accent, indicated by the dash – (H-, L-), controls the pitch movement between the last pitch accent of the intermediate phrase and the beginning of the next intermediate phrase, or the end of the utterance. Thus, there are two levels of prosodic phrasing over which the intonational contour is defined. The larger intonational phrase (IPh) is marked by longer breaks and the smaller intermediate phrase (ip) by shorter breaks. The idea is that an IPh is composed of one or more phrase-like prosodic units that are smaller than the IPh and larger than the prosodic word. These intermediate phrases provide a structure for the intonational phrase, as in (): ()

[IPh (ip) . . . (ip)]

A ToBI-style pitch extraction analysis of the verb phrase ellipsis (VPE) in example () is provided in Figure .. Example (), which can be completed as . . . but two swallows probably do make a summer, contains two cases of ellipsis in the second conjunct—a noun phrase ellipsis (NPE) in the subject position and a VPE. These two types of ellipses occur in a coordinate construction where the end of the first conjunct, or antecedent clause, is prosodically marked by a pause and a low boundary tone (L L%) and the end of the second conjunct, or elliptical clause, is also marked by a low boundary tone. The overall prosodic structure thus consists of two IPhs which both contain two parallel pitch movements: a fall (H*+L) on the number head one in the antecedent clause and a fall (H*+L) on the number head two in the elliptical clause, and a fall on the most deeply embedded DP summer in the antecedent clause and a fall on the auxiliary do immediately preceding the ellipsis site in the elliptical clause. In both IPhs, the phrase accent L- controls the pitch movement of the last H* accent and the boundary tone. The accent assignment in VPE follows from information-structural considerations. In this particular instance, do contrasts with doesn’t in the first conjunct. Different distributions of pitch accents in VPE are discussed in section ... The prosody of the German example in () bears some similarity to that of the English example in (). The overall melody of the embedded dass(that)-clause (bracketed) consists

  

Antecedent Clause

H*+L accent on DP correlate



Elliptical Clause

H*+L accent on DP remnant

H* accent on most deeply embedded DP

H* accent on auxiliary

L–L% boundary marking

L–L% boundary marking

320 300

Pitch (Hz)

250 200 150 130 h*+l one 0.893

l–l%

h* swallow

doesn’t make a

summer

but

Time (s)

h*+l two

h* probably

l–l%

do 3.78

 . Pitch extraction analysis of the VPE in (): Distribution of pitch accents and prosodic boundaries

of two intonational phrases with two parallel pitch accents (H*) which are realized as falls on the modals will (want) and muss (have to). Note that, although the second conjunct looks like an instance of VPE at first glance, the prosodic deletion has targeted almost all of the structure including the subject and the main verb, and therefore constitutes a case of conjunction reduction in German (cf. Hudson a). The relevant observation here is that the elliptical clauses in () and () contain a means of picking up the missing material from the antecedent clauses, that is, the missing material is interpreted in the elliptical sites at the level of semantic interpretation. At the same time, the interpretations are supported pragmatically and by the parallel prosody of the contrastively focused remaining constituents, referred to as remnants, in each case. The central questions with respect to ellipsis and prosody are given in (i–iii). ()

i. How does prosody contribute to the licensing of ellipsis? ii. How does prosody contribute to the process of recoverability of the elided material at the syntax–prosody and syntax–LF interfaces? iii. What role does prosody play in the derivation and interpretation of ellipsis? Can the investigation of the prosodic requirements for extraction from ellipsis provide evidence for structure in the ellipsis site?





The answers to these questions will be discussed in the following sections. Section . investigates the prosodic characteristics of the different types of ellipses and shows that the two main groups, the contrastive and the givenness-marking ellipses, are subject to different prosodic licensing conditions (location and type of pitch accent and prosodic phrasing). Section . discusses the prosodic evidence for recoverability of deletion. Section . investigates evidence for structure in the ellipsis site from the perspective of prosody. Section . concludes.

. L  

.................................................................................................................................. Research on prosody and ellipsis has mainly concentrated on the effects of the deaccentuation and deletion of given and redundant material and on the distribution of pitch accents realized on the remnant(s) and correlate(s). In the classic view set out by Chomsky (see Chomsky , , b), the syntactic component of the grammar accounts for the matching of sound and meaning. More precisely, syntactic structures are interpreted at two different levels, the Phonological Form (PF) and the Logical Form (LF), which constitute interfaces with other systems, the articulatory–perceptual and the conceptual–intensional systems (Chomsky b: ). Under this conception, it is one of the most important issues to find an explanation of how it is possible for speakers to produce ellipses and for listeners to interpret them in the absence of form. One central claim in the literature is that the prosody of the elliptical utterance is relevant in answering this question (cf., e.g., Hartmann ; Merchant ; Schwabe and Winkler ; Féry and Hartmann ; Repp ; Winkler  and references quoted there). In the linguistic literature, there are two main observations with respect to the relevance of prosody in explaining the nature of ellipses: information-structurally given material is subject to prosodic reduction (givenness-marking hypothesis) and material that remains must be prosodically highlighted (contrastive remnant condition). These conditions play an essential role in the prosodic description of ellipsis and will be considered in more detail in ...

.. Types of ellipses and their prosodic characteristics The prosodic and information-structural literature on ellipsis distinguishes at least two different kinds of ellipses, contrastive ellipsis as in () and givenness-marking ellipsis as in (). The strikethrough marks the unpronounced material: () MANNY plays the PIANO and ANNA plays the FLUTE.

(Gapping)

() Manny plays the piano but Anna DOESN’T play the piano.

(VP-Ellipsis)

Example () is an instance of gapping, which typically occurs in coordinate structures with parallel information structure; here the verb plays is deleted in the second conjunct and the

  

L+ H* accent on DP subj correlate

H* accent on DP obj correlate

L+H* accent on DP subj remnant



H* accent on DP obj remnant

L–H% boundary

L–L% boundary

380 350 300 Pitch (Hz)

250 200 150 130 h*

l+h* Manny 0.157

plays

the

l–h%

piano

h*

l+h* and

Anna

Time (s)

the

l–l%

flute 2.14

 . Pitch extraction analysis of gapping in (): Parallel contrastive accents and prosodic boundaries

remnants Anna and flute bear a contrastive pitch accent. Example () is an instance of VPE, where the redundant VP play the piano is information-structurally given and therefore prosodically deleted. Since here a complete VP-constituent is elided, this type of ellipsis has also been termed constituent ellipsis. Example (), thus, is also called nonconstituent ellipsis (Ross ) since only the verbal head has been elided (cf. Chao , and for a more recent discussion Reich ). The intonational contour of () is visually illustrated with a pitch extraction contour in Figure .. The antecedent clause and the gapping clause are semantically parallel except for those points where contrastive accents signal their difference. Contrastive prosody is realized on the subject Anna and on the direct object flute in the second conjunct. They occur in a contrastive relationship to their correlates Manny and piano in the first conjunct. The parallel interpretation allows the verbal head plays to be gapped. Kuno (, ) first observed that contrastive focus and deletion of redundant strings play a crucial role in the discourse appropriateness of gapping (see also Hartmann ; Johnson , this volume). Compare the paradigm in () without gapping and that in () with gapping: ()

a. MANNY bought APPLES, and ANNA bought BANANAS. b. Manny bought APPLES, and Manny bought BANANAS. c. MANNY bought apples, and ANNA bought apples.





The unreduced (a) can occur as an answer to the multiple wh-question Who bought what? The subjects of the coordinate clauses Manny and Anna each receive contrastive rising pitch accents (L+H*) and the objects apples and bananas contrastive falling pitch accents (H*). Examples (b) and (c) show that these prosodic characteristics are not mandatory for the conjoined and nonelliptical clauses. Manny can occur as the subject of both conjoined clauses in (b) and apples as the object in both clauses in (c). There is obviously no requirement for contrast or newness on the subject and object in the second clause. This is different for the gapped clauses as shown in (b,c): ()

a. √MANNY bought APPLES, and ANNA bought BANANAS. b. * Manny bought APPLES, and Manny bought BANANAS. c. * MANNY bought apples, and ANNA bought apples.

The set in () corresponds to (), but the verb bought is gapped from the second conjunct in each case. Gapping needs to be a matching response to a multiple wh-question Who bought what? Example (a) fulfills this requirement and is a well-formed answer to this multiple focus question. In contrast, the examples in (b) and (c) are highly marked and perceived as ungrammatical. In (b), the second coordinate clause does not contain a contrastive subject (Manny is the subject of both clauses) and in (c), the second conjunct does not contain a contrastive object (apples is the object of both clauses). Question–answer congruence of this type will be used as a heuristic test for discourse-appropriate prosody throughout (cf. Klein and von Stutterheim ; Rochemont ). The ungrammaticality of (b, c) results from the violation of two information-structural requirements on gapping which are expressed prosodically: the redundant verb in the second conjunct has the information-structural status of given, as defined in (), and can therefore remain unpronounced. The remnants in gapping must be interpreted as contrastive with respect to their correlates, as defined in (). This contrastive interpretation requirement has a prosodic reflex: the contrastive remnants must be prosodically highlighted and bear a strong contrastive pitch accent. The information-structural notion of givenness that has been central to the discussion of deletability and ellipsis (see, e.g., Halliday : ; Kuno , ,  for early observations) is informally defined as in () (see Krifka ; Féry and Ishihara ; Rochemont ): ()

Givenness A referent or part of a sentence is given if it is anaphoric to linguistic material mentioned previously in the discourse, or if it is entailed by the previous discourse.

The notion of contrastive focus is defined by Krifka (: ) as in (): ()

Contrastive focus Focus indicates the presence of alternatives that are relevant for the interpretation of linguistic expressions.

These information-structural notions can be understood as instructions to the PF component. The prosodic consequences are defined in the givenness-marking hypothesis (GMH)

  



in () and the contrastive remnant condition (CRC) in () (cf. Hartmann , ; Neijt ; Pesetsky : ff.; Reich , ; Sag b; Winkler ). ()

Givenness-Marking Hypothesis (GMH) Given material is deaccented or deleted at Phonological Form.

()

Contrastive Remnant Condition (CRC) Given information licenses a contrastive focus interpretation of the remnant(s). The contrastive remnant(s) must be assigned a strong contrastive pitch accent.

The GMH in () states that syntactic material is deaccented or deleted at PF if the material is either given or redundant. The CRC in () functions as an instruction to PF and requires that contrastive remnants must be prosodically highlighted and bear a strong contrastive pitch accent. The CRC is functionally dependent on the GMH. Gapping is an instance of contrastive ellipsis and subject to the CRC. The violation of the CRC leads to ungrammaticality as in (b) and (c). The pitch extraction contour of (c) is provided in Figure .. 300

Pitch (Hz)

250 200 150 120 h*–

l+h* Manny 0

bought

l–l% apples

l–l%

l+h* and

Anna

apples

Time (s)

2.569

 . Pitch extraction analysis of (c): Violation of pairwise contrast in gapping

The pitch extraction contour of the gapping example in Figure . shows that the subjects of both conjuncts are contrasted (signaled by L+H*-pitch accents), but the objects are not. In particular, the DP apples in the second conjunct is given and therefore deaccented by virtue of being mentioned in the first conjunct (see ()). The missing contrastive pitch accent on the second remnant in the gapping construction is perceived as highly marked if not ungrammatical. In this sense, the pairwise contrastive pitch accents prosodically license gapping. A case in point is provided in (): ()

A: Who bought what? B: MANNY bought TOMATOES [təˈmɑː.təʊz] and ANNA TOMATOES [təˈmeɪ.t ̬oʊz].





The phonetic contrast, which can be found in different varieties of English, is sufficient to license gapping in (). Note, however, that the semantic requirement of the CRC that the remnants must be interpreted as contrastive foci and semantic alternatives is still violated, causing a humorous effect. Next, consider the prosodic features of givenness-marking ellipsis. The question is: how can the ungrammatical example (c), repeated in (a), be saved? Information-structurally, the verb bought and the object apples in (c) have the information-structural status of given in both conjuncts, as in the response to the question Who bought apples? There is one straightforward way. We apply the givenness-marking hypothesis in () and delete the object, as in (b). Note that (b) remains ungrammatical. If the complete VP is deleted, as in (c), the result is still highly marked. Obviously, there are further syntactic requirements on givenness-marking ellipses. As is well known from the syntactic literature, a functional head (here the inflected auxiliary did) and the affirmative particle, too, are needed in order to license VPE, as in (d) (cf. Lobeck ; López and Winkler ; Aelbrecht and Harwood, this volume). ()

a. b. c. d.

*MANNY bought apples, and ANNA apples. *MANNY bought apples, and ANNA bought apples. *MANNY bought apples, and ANNA bought apples. MANNY bought apples, and ANNA did, too.

The conclusion of this discussion is twofold: first, there are two main prosodic licensing conditions: the givenness-marking hypothesis in () and the contrastive remnant condition in (). In the class of contrastive ellipsis, the contrastively focused elements are highlighted by omitting the given material. In the class of givenness-marking ellipsis, the givenness-marked complements of a functional head are prosodically reduced, as stated in (). In addition to these prosodic licensing conditions on ellipsis, there are additional syntactic and semantic requirements that must be observed (cf. Chao ; Aelbrecht and Harwood, this volume; Johnson, this volume). The next two sections will discuss the prosodic licensing mechanisms of further instances of contrastive ellipsis (section ..) and givenness-marking ellipsis (..).

.. Prosodic licensing of contrastive ellipsis The term contrastive ellipsis applies to the set of constructions in (): ()

a. MANNY plays the PIANO and ANNA _ the FLUTE. b. MANNY plays the piano and ANNA _, TOO. c. A: What did you buy? B: A new PIANO _ .

(Gapping) (Stripping) (Fragment Answer)

The set of contrastive ellipses in () have also been referred to as clausal ellipses in the syntactic literature. Clausal ellipses have been defined by van Craenenbroeck and Merchant () as “a subspecies of ellipsis whereby an entire clause is missing, including the canonical subject position and the agreement domain, but often to the exclusion of one or more clause-internal constituents. Those constituents are usually argued to move to the

  



left periphery of the clause prior to deletion” (p. ). On the basis of this syntactic account, it has been argued that the moved constituents are semantically interpreted as contrastive foci, as defined in (). In addition, they adhere to the CRC in () in that they are prosodically highlighted and marked by a high pitch accent.1 This analysis seems straightforward for the contrastive focus requirement on the remnants in the elliptical constructions in (), since here the remnants have been moved to the left periphery of the clause. In the gapping example in (a), both remnants are assigned contrastive focus accents (cf. Johnson , this volume, for references). In (b) the complete conjoined clause is elided except for Anna and the focus particle too; this phenomenon is called stripping (cf. Hankamer and Sag ; Johnson, this volume; Wilder, this volume). Both remaining constituents are mandatorily assigned a pitch accent (cf. Konietzko and Winkler ) and must be interpreted contrastively (cf. Depiante ; López and Winkler ; Konietzko and Winkler ; Konietzko , among others). Example (c) is a fragment answer where the DP a new piano in the answer corresponds to the wh-expression in the question and where the head noun piano is prosodically highlighted and focused (cf. Merchant a and Temmerman  for a movement account and Culicover and Jackendoff , this volume, for a non-movement account). The term contrastive ellipsis has also been applied to the set of constructions in (), although they are not instances of clausal ellipsis: () a. They play the PIANO better than they DO _ the FLUTE. b. ANNA played much faster than could have MANNY _. c. Manny PLAYS _ and Anna TUNES the antique PIANO.

(Pseudogapping) (Inversion Ellipsis) (Right-Node Raising)

Example (a) is known as pseudogapping (cf. Kuno ; Levin , ). It is parallel to gapping, since the verb play is missing, but it also bears some features of VPE since the missing string occurs after an auxiliary. The complement of the missing verb, flute, is overt and contrastively focused in relation to its correlate, piano. Syntactic approaches to pseudogapping assume that the remnant has been moved out of the verb phrase to the left edge of vP (Lasnik b), where the prosodic component assigns a contrastive pitch accent (cf. Gengel ).2 Example (b) is an instance of comparative inversion ellipsis (cf. Merchant b). A strong pitch accent is realized on the subject Manny, which occurs to the right of the auxiliary complex in the comparative clause (cf. Gergel et al. ). Culicover and Winkler () propose that the contrastively focused subject undergoes a similar kind of movement to the object remnant in pseudogapping before the vP is deleted. Example (c) is an instance of right-node raising (RNR). Here the predicate plays and the predicate tunes are assigned high pitch accents since they occur in a contrastive relationship (cf. Wilder , this volume; Hartmann ; Selkirk ; Féry and Hartmann ; Ha a, b, c). 1 For the syntactic literature on A’-movement, see e.g. Chomsky (, ). For the syntaxinformation structure and prosody interface of remnant movement in contrastive ellipsis, see Winkler (, ). 2 Although pseudogapping is not an instance of clausal ellipsis, it clearly shows the same behavior at the syntax–information structure–prosody interface. The remnant is moved out of the ellipsis site, which is the vP in this instance.





There are two characteristic features that the constructions in () and (a,b) have in common. First, the syntactic analyses propose that the remnants are derived by A’-movement either to the edge of the clause () or the edge of vP (a,b) before the clause or the verb phrase is deleted. Second, the syntax–prosody interface highlights the remnants with a pitch accent that marks their special discourse status. The remnants are interpreted contrastively with respect to their parallel antecedents. This feature gave rise to the observation that contrastive ellipses occur in coordinate constructions and are subject to the Parallelism Condition (PC) in (), which has the prosodic dimension in (). ()

Parallelism Condition (PC) A constituent satisfies the parallelism condition if it is semantically identical to another constituent, modulo focus-marked constituents. (adapted from Takahashi and Fox )

()

Prosodic Requirement of the Parallelism Condition A constituent satisfies the prosodic requirement of the PC if the focus-marked constituents are prosodically highlighted by a contrastive pitch accent. The nonfocus-marked constituents are deaccented or deleted.

The condition in () provides a prosodic requirement on the discourse appropriateness of contrastive ellipsis. Prosodic and information-structural licensing conditions have explicitly been formulated with respect to gapping as in (a) by Kuno (), Sag (b), Neijt (), Pesetsky (), Hartmann (, ), and Reich (, ). The violation of () and () leads to ungrammaticality if gapping occurs in non-parallel structures, as in (a,b) (cf. Johnson ). In (a) the gap occurs in a subordinate weil (because)-clause and not in a coordinated clause; in (b) the gap occurs in a topicalized construction which changes the word order. ()

a. *MANNY hat MANDARINEN gegessen, weil Manny has mandarins eaten because ‘Manny ate mandarins because Anna carrots.’ b. *MANNY hat MANDARINEN gegessen Manny has mandarins eaten ‘Manny ate mandarins and carrots Anna.’

und and

Anna Anna

Karotten carrots

Karotten. carrots Anna. Anna

The parallelism requirement also accounts for the prosodic and information-structural markedness of the so-called dangling remnant cases in () (cf. judgements by Schwarz : ; accents added): () a. SOME talked with YOU about POLITICS and OTHERS talked with ME about MUSIC. b. *SOME talked about POLITICS and OTHERS talked with ME about MUSIC. While (a) answers the question Who talked with whom about what?, (b) doesn’t. In (b) the first conjunct answers the question Who talked about what? but the gapped

  



conjunct requires a wh-question with three operators. That is, the markedness of (b) results from the violation of the PC in () and (), which is supported by the fact that no uniform multiple wh-question can be found which is answered by both conjuncts. An attested German gapping example with mismatching parallelism is provided in (): ()

ER ist ihr TREU — SIE nicht he is her faithful — she not ‘He is faithful to her, she not always.’

IMMER. always (ZDF,  January )

This example constitutes a case of contrastive ellipsis with mismatching semantic and prosodic parallelism because the remnants do not occur in a pairwise contrast. The first conjunct has a subject–aux–dative object–predicate structure. The elliptical conjunct, however, only mentions the subject–negation–adverb. Thus, the deletion involves the predicate plus the dative object. The parallelism condition requiring a pairwise contrast between the remnants and the antecedents is therefore violated. Due to this fact, this example is ambiguous in three ways: it can mean that she wasn’t always faithful to him (sloppy reading), that she wasn’t always faithful to her (external reference), or that she wasn’t always faithful to herself (strict reading). Pragmatically, the last two readings are probably marked. Thus, licensing of gapping requires a parallel syntactic, semantic, and prosodic structure and contrastive accents on the remnants. Mismatches show that prosodic licensing is essential for the discourse appropriateness of gapping constructions. Another type of contrastive ellipsis for which prosodic licensing conditions have been proposed is RNR as in (c) (cf. Wilder, this volume). The phenomenon of RNR refers to a coordinate construction in which parts of the first conjunct are omitted, but are spelled out at the right periphery of the second conjunct. RNR is characterized by a specific prosodic pattern as in (), where the capitalized verbs receive distinct contrastive pitch accents (cf. Selkirk ). The pitch accents are followed by clearly distinguishable intonational breaks, as shown in () and illustrated by the pitch extraction contour in Figure .. 400

Pitch (Hz)

300

200

120 l+h* John 0.29

h*

l–h%

l+h*

bought and Mary

h*

broke

an

Time (s)

l–l%

h*

l–h% expensive

Chinese

vase 3.9

 . Pitch extraction contour of RNR in (): Licensing pitch accents and boundary tones





() John BOUGHT ║ an expensive Chinese vase and Mary BROKE ║ an expensive Chinese vase. (Abels : ) In response to the context question Who did what? the subjects of both conjuncts are realized with rising accents (L+H*) and the verbs bought and broke each receive a contrastive pitch accent (H*) preceding a high boundary tone, the latter of which precedes the common target (cf. Wilder, this volume). Proceeding from left to right, the first conjunct corresponds to an IPh which ends in a high boundary tone (H%) after bought followed by a pause. The second conjunct, starting with the conjunction and, is also realized with a high boundary tone after the verb broke signaling continuation. The subjects John and Mary are interpreted as contrastive topics and the high pitch accents on the verbs bought and broke are interpreted as contrastive foci. The third IPh includes the target DP an expensive Chinese vase. Since it has not been mentioned in the context question, it is realized with a fall on vase which signals new information. If the context question mentions the target as in Did you hear the story about an expensive Chinese vase? the target is realized with L* accents and a longer deaccented target string, as is expected for given information. For RNR constructions like (), there exist two different versions of prosodic licensing analyses. The first is a phonological reduction account by Hartmann () and Féry and Hartmann (). They propose on the basis of the GMH in () that the missing target in the first conjunct an expensive Chinese vase is phonologically reduced by “radical deaccentuation” (Féry and Hartmann : ). The second is an E-feature account, provided by Ha (a, b). Ha proposes that RNR is licensed by a contrastively focused element which bears an E-feature as originally proposed by Merchant () for VPE and sluicing. According to Ha, the E-feature in RNR “instructs PF to leave the RNRed element unpronounced” (Ha a: ). From the perspective of prosodic licensing, both accounts make the same predictions. The structural parallelism of the construction and the prosodic highlighting of the verbs license the phonological silence of the first of two redundant strings. Féry and Hartmann () propose the licensing conditions for RNR and gapping in (): () Condition on ellipsis in RNR and gapping . The conjuncts must exhibit a parallel syntactic and focus structure. . The focus constituents in the two conjuncts must allow for a contrastive interpretation. (Féry and Hartmann : ) Féry and Hartmann observe that the prosodic phrasing in RNR generally seems to follow the same pattern: there is a high boundary tone preceding the covert target, which is followed by a pause. They propose that this specific prosodic RNR pattern at the end of the first conjunct “signals not only the incompleteness of the utterance at this stage, as well as the concomitant intention of the speaker to continue the sentence, but also the fact that some material is missing. In other words, part of the upstep could be motivated by the incompleteness of the utterance and part of it by the ellipsis” (Féry and Hartmann : –). Although Ha (a, b) challenges Féry and Hartmann’s approach and claims that an E-feature approach fares better with respect to empirical coverage, from a prosodic perspective the proposals are difficult to distinguish since they make the same empirical predictions. The typical prosody

  



with the contrastive pitch accents on the verbs signals to the PF-component that the segmental and suprasegmental features of the target may remain unrealized. As for mismatches in RNR, there is an interesting case in which the typical RNR prosody seems to license an otherwise highly unpredicted instance of RNR, given in ().3 () Four or five girls took SWINGS at and PUSHED each other OUT into the MIDDLE of oncoming TRAFFIC. (Irish Independent,  September ) In (), the ellipsis occurs at the end of the first conjunct immediately following the pitchaccented noun swings and the preposition at realized with a rise. The second conjunct is subjectless (conjunction reduction); there is a pitch accent on the verb pushed followed by the target each other and it continues with high pitch accents on the particle out, middle, and on traffic. This example is an interesting case of RNR because the target is the reciprocal each other and not the prepositional phrase into the middle of oncoming traffic at the right edge of the clause. The interpretation is as follows Four or five girls took swings at each other and pushed each other out into the middle of oncoming traffic. This example calls the generalizations about RNR into question, particularly that the pivots must be final in all conjuncts (see Chaves ). Note that the seemingly parallel structure of the two intonational phrases disguises the fact that the example in () does not match the generalizations on RNR in the literature.

.. Prosodic licensing of givenness-marking ellipsis The theoretical precursor of the givenness-marking account of elliptical constructions (cf. () and ()) is the so-called phonological reduction hypothesis, as in (). ()

Phonological Reduction Hypothesis Elliptical sentences are formed by a rule of the PF-component that deletes the phonologically redundant information which is characterized by a “distinguished low-flat intonation”. (Chomsky and Lasnik : )

In the early PF-deletion accounts (cf. Tancredi ; Chomsky and Lasnik ; Chomsky b), phonological deletion was applied to syntactically identical or redundant material. Chomsky and Lasnik () discuss the so-predicate phrase construction in () and propose that deletion should be treated as an optional phonological rule which applies to deaccented material (i.e., the material which occurs in the square brackets and arguably lacks phonological accent).4 ()

3

John said that he was looking for a cat and so did Bill [say that he was looking for a cat].

The prosody stems from a pilot production study with three English native speakers. Note that (26) is not a particularly straightforward example to demonstrate the analogy between elliptical constructions and prosodic reduction in redundancy contexts, because the do-so construction functions as a predicative anaphor and occurs in complementary distribution with VPE (see López ). 4





The proponents of this theory assume that deaccentuation and deletion take place at PF and follow from identity conditions that are stated at PF. Alternatively, it is assumed that the relevant constraints follow from focus considerations (cf. Rooth ; Tancredi ; W. Klein ; Hartmann , ; Merchant ; Romero ; Winkler ; Repp ; Konietzko ). In particular, Tancredi () claims that the constraints on VP-deletion can be reduced to those on deaccentuation, where deaccentuation is a process that takes place at PF. Tancredi (: ) formulates this claim with respect to VPE more radically by proposing that “VP-ellipsis is no more than an extreme case of deaccenting where a VP ceases to be audible altogether.” Thus, Tancredi subsumes VPE under other cases of deaccentuation by proposing that VP deaccentuation shows restrictions similar to those active in VP deletion, as seen in the paradigm in () (cf. Tancredi : –). There are two arguments relevant for Tancredi’s claim: the first one is that the process of deaccentuation is triggered by parallel interpretation, as is deletion. This can be seen by the ambiguous paradigm in () (originally from Sag b). Italics signal deaccentuation, strikethrough signals deletion: ()

John likes flying planes but Bill doesn’t like flying planes. a. . . . but Bill doesn’t [VP like [NP flying planes]]. b. . . . but Bill doesn’t [VP like [NP flying planes]]. c. . . . but Bill doesn’t [VP like [S PRO flying planes]]. d. . . . but Bill doesn’t [VP like [S PRO flying planes]].

Following Sag’s basic insights, Tancredi observes that although the first conjunct in () is ambiguous between an NP-complement reading and a gerund reading, the resulting elliptical examples in (b,d) are also only two-ways (and not four-ways) ambiguous. However, Tancredi’s particular claim is that similar conditions hold not only for ellipsis, but also for deaccentuation, as in (a,c). Tancredi’s second argument is that the semantic interpretation of variables in the ellipsis site and in the deaccented string follow the same interpretative rules. He investigates strict and sloppy readings of pronouns in examples like (). ()

a. b. c. d.

Manny believes he is brilliant and Leo does believe he is brilliant too. Manny believes he is brilliant because Leo does believe he is brilliant. Manny believes he is brilliant and Leo believes he is brilliant. Manny believes he is brilliant and Leo believes he is a smart guy.

Tancredi observes that similar readings result in both reduction contexts. Examples (a,b) are more extreme cases of the type of deaccentuation which can be witnessed in (c). The difference between deletion and deaccentuation is that “identity is not required for deaccenting” (Tancredi : ), as seen in (d). The discussion shows that the phonological reduction hypothesis is directly linked to the information-structural notion of givenness as provided in (). Note, however, that within the e-GIVENness account (cf. Merchant ), Tancredi’s condition on deaccentuation which encodes a one-directional entailment relation has been strengthened by going both ways, from the antecedent to the elliptical clause and from the elliptical to the antecedent clause, as is formulated for VPE in ().

   () A VP α can be deleted only if α is e-GIVEN.



(Merchant : )

The condition in () states that a VP can be deleted if it qualifies as e-GIVEN as defined in (). () e-GIVENness: An expression E counts as e-GIVEN iff E has a salient antecedent A and, modulo ∃-type shifting, (i) A entails F-closure (E), and (ii) E entails F-closure (A). (Merchant : ) E-GIVENness in () states that a VP can be deleted if it qualifies as e-GIVEN, where e-GIVENness requires a mutual entailment relation to hold between the antecedent (A) and the elided VP (E). Implementing e-GIVENness in the grammar of ellipsis, Merchant (, a, b) proposes a syntactic E-feature on heads that licenses ellipsis. Egivenness applies straightforwardly to the complement of the functional heads in VPE, noun phrase ellipsis, and sluicing in (a,b,c). ()

a. Manny plays the piano but Anna DOESN’T _. b. Manny plays a solo with one hand and Anna with TWO _. c. Someone is playing the piano, but I don’t know WHO _.

(VPE) (NP-Ellipsis) (Sluicing)

The syntactic literature on VPE has paid considerable attention to the functional elements that occur at the left edge of an elided VP and function as licensors of VPE. The most influential proposal, by Lobeck (), argues that VPE is licensed by agreement between the functional head of the clause and its VP-complement (cf. Zagona b; Aelbrecht ; Aelbrecht and Harwood, this volume). From the information-structural perspective there is agreement that the deletion of the VP is licensed by e-GIVENness as spelled out by Merchant () (see also Rooth a, b, ; Schwarzschild  for earlier proposals). There is much less work on the prosody of the remnants and the question of how they contribute to the licensing of the construction. Rooth (a, b) proposes that in instances where VPE occurs in parallel constructions, the deleted VP licenses contrastive focus on the subject, as in example (a) with the corresponding LF in (b). At LF, the scope of the focus feature F (MaryF) is marked by the focus interpretation operator ~ (informally referred to as squiggle operator) which ensures that congruence is satisfied with respect to the coindexed antecedent.5 () a. First John came up with a good idea, then [MARY\]F did [VP e]. b. [S [S First, John [VP come up with a good idea]], and then [S [S MaryF did [VP come up with a good idea]] ~ ]].

5 The accent assignment in VPE follows from information-structural considerations. The main contrast can be realized on the remaining subject in the second conjunct. As pointed out by an anonymous reviewer, it would be ungrammatical to assign prominence only on did as in (i).

(i) *First John came up with a good idea, then Mary DID. The subjects, John and Mary, occur in a contrastive topic relation, typically expressed by a prosodic rise.





Two alternative proposals have been made that consider VPE in non-parallel constructions. López and Winkler () propose that these instances of VPE are licensed by focus on the negative/affirmative features (polarity) situated in a functional category called Σ, which takes a VP as a complement, as in () and the attested example in ():6 ()

a. Anna promised to do the dishes, but she DIDN’T\. b. I doubted that Ben can swim, but he CAN\.

() Harry: Hermione:

Hermione, that’s the man who betrayed my parents! You don’t expect me to just sit here . . . YES, and you MUST! (Rowling )

While in () and () the subject in the elliptical clause is pronominalized, it is deleted by the process of conjunction reduction (Hudson a) in (). ()

Rick: Ilsa:

All the same, some day you’ll lie to Laszlo. You’ll be there. No, Rick. No, you see Victor Laszlo’s my husband. And WAS, even when I knew you in Paris. (Casablanca)

Gergel () extended this proposal by suggesting that the aspectual feature [+tense] licenses VPE in modal–auxiliary combinations as in () (see also Aelbrecht ; Temmerman ; Aelbrecht and Harwood ): () a. Did he come? b. He MUST həv.

(Hofmann : )

It seems that a unifying proposal is missing. Note, however, that the theory of focus and givenness marking in question–answer contexts would predict exactly the behavior observed in these examples. The prosodic requirements in VPE are not fixed by the CRC and the PC, but they follow from the theory of information structure and discourse appropriateness (cf. Kehler, this volume), as seen in (). ()

A: Who could have been being blackmailed? B: JOHN could have been being blackmailed, but GEORGE COULDN’T həv been.

In (B), the subjects of both conjuncts are contrastive topics, the negated modal in the elliptical clause is contrastively focused (could vs couldn’t), and perfective have and progressive been are deaccented. In (B), it is shown that if the focus is realized on been, the elliptical clause is marked, since progressive been is obviously not a licensor of VPE.

See King (), who first observed that contracted auxiliaries which typically don’t bear a pitch accent cannot license VPE as seen in (i): 6

(i) *John is not happy, but I’m _.

   ()



A: Who couldn’t have been being blackmailed? B: JOHN couldn’t have and GEORGE couldn’t həv *BEEN, either.

The GMH also applies to sluicing in () and deletes the phonologically redundant material it exists following the wh-expression where. () There is a lot of talk about freedom. It’s like the Holy Grail, we grow up hearing about it, it exists, we’re sure of that, and every person has his own idea of WHERE. (Winterson, The Passion, ) According to Merchant (, b), the E-feature on C licenses the deletion of the TP in () (see also Vicente, this volume). ()

. . . it exists, we’re sure of that, and every person has his own idea of [CP WHERE [C0[wh, Q, E] [TP it exists]]].

According to the GMH, the TP it exists is deleted at PF and the remnant of the elliptical construction is prosodically highlighted. The remaining overt wh-expression where bears a downstepped H* pitch accent immediately followed by an L- phrase accent and an L% boundary tone. Note that sluicing is traditionally analyzed as a givenness-marking ellipsis (cf. Merchant ). At the same time, sluicing is also an instance of contrastive ellipsis since the wh-constituent is moved to the left periphery and the CRC in () also holds. The remnant must carry a pitch accent which is interpreted as a contrastive focus (cf. Molnár ; Repp , ; Molnár and Winkler ). This section has answered the question asked in (i), namely how does prosody contribute to the licensing of ellipsis? The discussion has shown that prosody participates in the licensing process of ellipsis together with syntactic, semantic, and informationstructural constraints. In particular, it has been shown that there are two different types of constructions, where the types of accents fulfill a specific licensing function. The contrastive ellipses are licensed in parallel contexts by a contrastive accent on the remnants with respect to their correlates, as stated in the CRC (). The givenness-marking ellipses are licensed by deaccentuation or deletion of a given complement of a functional head, as was first proposed in the phonological reduction hypothesis. This hypothesis has been incorporated into a more general information-structural requirement, the givennessmarking hypothesis in (). The prosodic regularities of both types of ellipsis in discourse follow from question–answer congruence.

. P     

.................................................................................................................................. The general assumption is that the syntax–semantics interface is involved in recovering ellipsis (e.g., Reich ; van Craenenbroeck and Merchant ). This section will investigate whether the syntax–prosody interface takes any part in the reconstruction of the





ellipsis site and how it contributes to the process of ellipsis recovery (cf. ii). In particular, prosody has been claimed to be relevant with respect to conditions of recoverability of deletion fed by the prosodic requirements of the contrastive remnant condition, the givenness-marking hypothesis, the parallelism condition, and the position of pitch accents and prosodic boundaries. The discussion concentrates on different types of mismatches that allow one to draw specific conclusions about the division of labor between the syntax– phonology and syntax–semantics interfaces. Where relevant, the discussion also refers to psycholinguistic research.7

.. Phonetic mismatches At first glance, it seems straightforward that the representation of the elliptical site cannot simply be a surface identity of phonetic sounds as represented at PF. An intriguing phonological surface structure mismatch is provided by van Craenenbroeck and Merchant (: ) in (a), where the second of the homophonous words (right and write) is deleted in VPE (see also Arregui et al. ). The second verb pronounced [raɪts] cannot be recovered. A similar point can be made with respect to the German gapping example in (b), which illustrates that the phonologically similar verbs mahlen (grind) and malen (paint) both pronounced as [maːlt] in the third person singular do not allow the recovery of the ellipsis in the second conjunct. ()

a. *Injustices, he rights, but books he doesn’t write. b. *Der Bäcker mahlt das Korn, der Künstler malt das Bild. the baker grinds the grain the artist draws the picture ‘The baker grinds the grain, the artist draws the picture.’

However, it is too early to conclude that phonetic and phonological representations are completely excluded from the recovery process of ellipsis. Although the identity relation cannot be determined by the phonetic structure at PF alone, there are examples where the syntax–phonology and syntax–semantics interface seem to interact in recovering the ellipsis site. Consider the recovery processes of the VPE in (), an attested example in English. ()

‘I was proud the other day when both Republicans and Democrats stood with me in the Rose Garden to announce their support for a clear statement of purpose: You disarm, or we will _ . —’ (George W. Bush, Manchester, NH,  October )

The unmarked recovery of the VPE in () You disarm or we will uttered out of context is the non-causative reading You disarm or we disarm. However, the speaker is George W. Bush and the statement is addressed to Saddam Hussein. In the political context of the time, the unmarked interpretation cannot be the right one. Therefore, a lexical retrieval process which involves the activation of phonetic features (cf. Frauenfelder and Floccia ) sets in. The lexical element disarm not only has a non-causative reading but 7 The discussion of prosody in relation to the different scopal interpretations in gapping is an interesting issue but goes beyond the scope of this chapter (cf. Johnson a; Winkler : ch. ; Repp ; Temmerman ).

  



also a causative one, You disarm or we will disarm you. The interpretation of this utterance as a threat is the contextually adequate one. The observation in the processing literature is that lexical retrieval processes activate the phonetic features. This can also be observed in the gapping example () in German. () Der EINE ging nach BREMEN, der ANDERE ging von dort WEG. the one went to Bremen the other went from there away ‘One player went to Bremen, the other player left Bremen.’ (U. Wickert,  October ) Example () shows that gapping in German can apply to the verbal head ging (left) alone, leaving the subject NP der andere (the other), the PP von dort (from there), and the particle of the verb weg behind. Here it seems that the phonetically identical verb ging in the second conjunct is gapped although weggehen (leave) is semantically the opposite of the verb gehen nach (go to) in the antecedent clause. The observation is that gapping can occur in the second conjunct in () despite the fact that the semantic meanings of the lexical entries (gehen vs weggehen) differ. The discussion shows that the identity relation between the deleted string and its antecedent is a matter of syntax and semantics and is not uniquely determined by phonetic and phonological representation. However, it seems that the lexical encoding of the identity relation interacts with the phonological component and the syntactico-semantic component in the antecedent retrieval. In particular, the homophonic representations seem to suggest that phonetic representations play a role in the search process for an appropriate interpretation of these examples.

.. Prosodic disambiguation The underlying argument regarding the prosodic disambiguation of syntactically ambiguous clauses is that prosody, and in particular prosodic phrasing and pitch accent assignment, can influence semantic interpretation (cf. Winkler ). Prosodic phrasing and the placement of pitch accents influence the interpretation of ellipsis.

... Prosodic phrasing and disambiguation Féry () observed that prosodic phrasing may disambiguate homophonous syntactic structures as in (): () Maria Maria

schaukelt swings

Martin nicht. Martin not

Example () has two different prosodic representations resulting from the different instantiations of the argument frame of the verb schaukeln (swing) as in (a,b): () a. [IPh Maria SCHAUkelt Martin NICHT]. ‘Maria isn’t pushing Martin on the swing.’ b. [IPh MaRIa SCHAUkelt][IPh MARtin NICHT]. ‘Maria is swinging, Martin isn’t.’





The prosody of (a) consists of one IPh, signaling a single clause in response to a question such as What’s wrong with Maria and Martin? The verb schaukeln is interpreted as a transitive verb Maria isn’t pushing Martin (on the swing). Example (b) is an answer to a question such as Who does what? The prosodic realization contains a break after the verb and thereby signals the division of the clause into two IPhs, where the second one is an instance of stripping in German. The prosodic realization follows the parallelism condition, where the subjects Maria and Martin are realized with rising pitch accents which are interpreted as contrastive topics. The verb schaukelt (rd pers. sing.) in the first conjunct is assigned a focus accent, as is the negative particle nicht (not) in the second conjunct. The parallel prosody supports the interpretation of the two IPhs as two coordinate clauses, where the intransitive verb schaukeln is reconstructed in the second conjunct: Maria is swinging, Martin isn’t swinging. The relevant observation with respect to the prosodic disambiguation of () is that the transitive single IPh-reading in (a) is the preferred interpretation. The stripping reading in (b) is brought about by prosodic phrasing in two IPhs and parallel pitch accent assignment. The recovery of the ellipsis site in (b) is dependent on the givennessmarking hypothesis and the contrastive remnant condition. In addition, it conforms to the prosodic requirement of the parallelism condition in (). The underlying argument is that this particular structural ambiguity can be prosodically disambiguated (cf. Wiedmann and Winkler ; Remmele et al. forthcoming). The parallel prosodic realization of the two IPhs in (b) supports the recovery of the ellipsis site.8

... Pitch accents and disambiguation Elliptical constructions have been observed to cause ambiguous readings since Ross () and Hankamer (a). Kuno () observed that information structure and prosody provide hints as to how to recover the ellipsis site. Hankamer proposed a general No Ambiguity Condition based on the observation that the example in (a) has the reading provided in (b) and not the one in (c). () a. Max gave Sally a nickel, and Harvey a dime. b. ___ and √[Max gave] Harvey a dime. c. ___ and Harvey *[gave Sally] a dime.

(Hankamer a: )

Note that (a) can receive the reading that Hankamer marked as impossible (c) if an appropriate discourse question that requires a particular prosodic realization precedes the utterance, as in (): ()

8

a. Who gave Sally what? b. MAX gave Sally a NICKEL, and HARVEY _ a DIME.

From a psycholinguistic perspective one could argue that there is a temporary ambiguity at the point where the verb schaukeln is parsed.

  



The multiple wh-question in (a) requires the deaccentuation of the verb gave and the indirect object Sally and the pairwise prosodic highlighting of the subjects and the direct objects in (b). Thereby, the reading that Hankamer excluded arises as the preferred one. The type of ambiguity shown in (a) is referred to as subject vs object ambiguity in the literature on ellipsis and has been widely observed with case-ambiguous DP-remnants. The reading in (b) is generally referred to as conjunction reduction, the one in (c) and () is an instance of gapping. Psycholinguistic studies have investigated the factors which influence the preferred reading in ambiguous examples, as in () (cf. Carlson ; Stolterfoht ; Hoeks et al. ; Stolterfoht et al. ; Hoeks et al. ; Frazier, this volume). ()

Lenny insulted the guests during dinner and Manny during the dance.

The psycholinguistic studies show that the question of whether Manny is interpreted as the subject or the object of the second clause in () mainly depends on prosodic and contextual factors, as seen in the small dialogues in () (gapping) and () (conjunction reduction, or non-gapping). ()

A: Who insulted the guests when? B: LENNY insulted the guests during DINNER and MANNY during the DANCE.

()

A: Who did Lenny insult when? B: Lenny insulted the GUESTS during DINNER and MANNY during the DANCE.

Subject vs object ambiguities also occur in stripping as in () (cf., e.g., Radford : ; Carlson ; Lechner ; Konietzko ). Note that the ambiguity of the comparative ellipsis in (a,b) cannot be resolved by the accentuation of the remnant alone. The pronoun you must be stressed in both readings. The prosodic pattern of the antecedent clause is relevant. In the subject reading, the contrasting subjects are accented as in (a), in the object reading, the contrasting objects as in (b): ()

a. HE loves me more than YOU love me. b. He loves ME more than he loves YOU.

The ambiguous stripping examples in English are disambiguated by nominative case assignment in (a) and accusative case marking in (b): () a. Er he

liebt loves

mich me

b. Er he

liebt loves

MICH meAcc

mehr more mehr more

als than als than

DU youNom er he

mich liebst. me love

DICH youAcc

liebt. loves

A similar type of subject vs object ambiguity has been observed with respect to sluicing in () and () by Merchant (: ) (cf. also Frazier and Clifton ; Carlson et al. ). ()

A: Who called Manny an idiot? B: ANNA called Manny an idiot, but I don’t know who ELSE called Manny an idiot.

 ()

 A: Who did Anna call an idiot? B: Anna called MANNY an idiot, but I don’t know who ELSE Anna called an idiot.

Note that the disambiguation in these cases depends on the context question and the contrastive prosodic highlighting of the correlate in the antecedent clause. A different type of ambiguity has been observed with respect to the remaining subject in VPE. The subject of the second clause can function either as a matrix subject as in (B) or an embedded subject as in (B) (cf. Frazier and Clifton ; Frazier et al. ; Traxler and Frazier : ). ()

A: Who said that Fred went to Europe? B: JOHN said that Fred went to Europe and MARY did too. (Frazier and Clifton : )

()

A: What did John say about Fred and Mary? B: John said that Fred went to EUROPE and Mary did TOO.

In the example (B) the subject reading is brought about by the context question, which requires a pitch accent on the subject John in the matrix antecedent clause and on the parallel subject Mary in the elliptical clause. In () John is the matrix subject of both conjuncts. Mary is the subject of the subordinate clause. All the NPs are given, except Europe, which is highlighted. The general observation is that the distribution of the pitch accents in the antecedent clause influences the interpretation of the elliptical clause. The relevant prosodic requirements are influenced by the context questions and formalized by the GMH, CRC, and the prosodic requirement of the PC.

.. Active–passive mismatches The literature on active–passive mismatches, such as in () and (), is extensive (cf. Dalrymple et al. ; Hardt ; Merchant a, , a; Kim et al. ). () This information could have been released by Gorbachev, but he chose not to. (Hardt ) () The janitor must remove the trash whenever it is apparent that it should be. (Merchant b) Merchant’s empirical observations of active–passive mismatches constitute strong evidence for a syntactic identity account of ellipsis. Note that there is an alternative account by Kertz () which suggests that information-structural and prosodic factors play a crucial role in the interpretation of these mismatches. Kertz (: ) investigates the puzzle given in ()–(). Example () is an instance of VPE. Example () is identical, except that it contains an active–passive mismatch and is considerably marked. The puzzle arises in (),

  



which also contains an active–passive mismatch but does not seem to be degraded. Pitch accents are added following Kertz (: ). ()

The DRIVER reported the incident, and the PEDESTRIAN did TOO (report the incident).

()

#The INCIDENT was reported by the driver, and the PEDESTRIAN did too (report the incident).

()

The INCIDENT was reported by the driver, although he didn’t really NEED to (report the incident).

The example sentences ()–() differ with respect to their information structure and intonation. This can best be seen by the question–answer test. Example () answers the question Who reported the incident? The subject remnants occur in a contrastive relationship in a parallel structure (contrastive topics, cf. Kertz). The deleted and e-GIVEN VP in the second conjunct can be readily recovered by the antecedent VP. Example () also occurs in a parallel construction but there is no single context question which it could answer. The subject of the first conjunct the incident and the subject pedestrian of the second conjunct do not contrast because they do not form alternatives of the same set (cf. Rooth a). This is a violation of the contrastive focus requirement in () and causes a severe mismatch. Example () constitutes a non-parallel construction involving a main clause which takes an embedded clause (cf. Reich and Reis ). The VPE occurs in the embedded clause. Example () answers the context question What was reported by the driver? and provides the additional information that the driver didn’t really need to report it. The subject of the subordinate clause is a given topic he that refers back to the thematic subject of the main clause the driver, and the focus accent is realized on the auxiliary need. The VPE in the embedded clause is easily recoverable because it finds its e-GIVEN antecedent in the matrix clause. The prosody supports this interpretation, as discussed in section ...

.. Pitch accents and the interpretation of pronouns under ellipsis Prosodic research has focused on the relation between pitch accents and the interpretation of pronominal reference for almost five decades (Lakoff ; Akmajian ; Hirst ; Kameyama , ; Grosz et al. ; Kehler , ; Beaver ). In an early paper, Hirschberg and Ward () investigate the relation of pitch accent assignment and semantic interpretation in VPE contexts. They propose that speakers use the stressed pronominal form to signal a marked interpretation, as originally observed by Lakoff () for the well-known example John called Bill a Republican and then he/HE insulted him/HIM. They propose two tendencies: first, the sloppy interpretation is available where no c-command relation between anaphor and antecedent obtains; second, the presence or absence of a pitch accent on the anaphor affects whether the utterance is assigned a sloppy or strict interpretation (Hirschberg and Ward : ). More precisely, they propose that “pitch accent facilitates a strict interpretation when the unmarked case (i.e. the





interpretation favored in the written condition) is sloppy, or a sloppy interpretation when the unmarked case is strict” (p. ). Their observations tie in with earlier observations by Reinhart (b). She showed that for bound pronouns, such as the reflexive in (), the referential reading is systematically absent under VPE. That is, in () only the reading in which John voted for himself and his lawyer voted for himself is available, as provided in the paraphrase. Accentuation of the reflexive does not change the reading. ()

Johni voted for HIMSELFi and his lawyer did, too. ‘[ . . . ] and his lawyer voted for himself ’

However, this prediction does not hold for all cases, as () from Hardt () shows: ()

Johni defended HIMSELFi, because his lawyer didn’t. ‘[ . . . ] because his lawyer didn’t defend him (John)’

VPE in example () occurs in a subordinate clause and allows either a strict or a sloppy reading. Here the stressed pronoun condition by Hirschberg and Ward would predict that the referential reading because his lawyer didn’t [defend him] where him refers to John is not only available, but even preferred. Note, however, that the study by Hirschberg and Ward () has shown that speaker intuitions on binding relations are not as robust as they may seem at this point. This has also been one of the major points of criticism by Hardt () and others with respect to the data discussed above. Other restrictions seem to play a role, such as accent assignment, focus, and discourse structure (cf. Hardt and Romero ; Kehler, this volume). The general answer to the second main question in (ii) is that prosody contributes to the process of ellipsis recovery by showing that the syntax–semantics and the syntax– phonology interfaces follow specific prosody-related mapping relations.

. P      

.................................................................................................................................. Research on the question of whether there is structure in ellipsis sites usually targets syntactic and semantic arguments. This section investigates the third main question raised in (iii) regarding whether there is evidence stemming from prosody suggesting that ellipsis sites contain syntactic representations that are unpronounced. Prosody provides indirect evidence for structure in the ellipsis site. By investigating the prosody and information structure of the remaining elements, the prosodic structure of the elided material can be inferred (cf. Hartmann ; Carlson ; Féry and Hartmann ; Winkler , ). One core idea is the complementary distribution of the complete phonological reduction of those parts that are contained in the ellipsis site and the highlighting of the remnants, as discussed in section . with respect to the givenness-marking hypothesis, the contrastive

  



remnant condition, and the prosodic requirement of the parallelism condition. However, complete deaccentuation of syntactic material is not an entirely convincing argument, since empty proforms are also characterized by phonological silence (cf. Lobeck ). Therefore, this section presents evidence from prosody-related research that concentrates on two main topics: first, on prosodic conditions that allow movement out of ellipsis sites, and second, on locality restrictions and the question of prosody-related repair.

.. Extraction from ellipsis site and prosody It has been widely observed in the literature that A’-extraction from an ellipsis site is possible (Sag b; Williams b; Fiengo and May ; Lasnik ; Merchant , b; Schuyler ; Takahashi and Fox ; see also Abels, this volume, for an overview). In most generative accounts, the remnants of contrastive ellipsis (i.e., gapping, stripping, pseudogapping, and sluicing) are assumed to be extracted from the ellipsis site (cf. section ..). The prosodic requirement on contrastive ellipsis is that it obeys the prosodic requirement of the PC and that the movement-derived remnant must be accented and contrastively interpreted in relation to its correlate. The underlying argument is that extraction from a VPE site is sensitive to prosodic and information-structural constraints. The logic of the argument is as follows: if the ellipsis site contains unpronounced (completely deaccented) structure, it should be possible to extract those elements that are not deaccented because they are not given. It follows that these extracted elements must be accented and contrastively focused. This is exactly what we observe in contrastive ellipsis. In extraction from VPE the wh-expression is inherently focused and requires contrastive focus on the auxiliary. The most explicit focus-related proposal for extraction from VPE is provided by Schuyler (). She proposes the Contrast-Locality Condition as in (): ()

Contrast-Locality Condition For A’-extraction out of the site of VPE to be licensed, there must be a contrastively focused expression in the c-command domain of the extracted phrase. (Schuyler : )

The contrast-locality condition in () accounts for the grammaticality of extraction out of VP in (a) and the ungrammaticality of (b): () a.

I don’t know which puppy you SHOULD adopt, but I know which one you SHOULDN’T [VP adopt ti]. b. *I think you should adopt one of these puppies, but I don’t know which one you should [VP adopt ti]. (Schuyler : )

The contrast-locality condition in () also accounts for the examples in ()–(), where the extracted wh-phrases, which are considered focus operators, take another focused element in their scope. In () and () the licensing pitch accent of the VPE is realized on the functional head.





() As Hagrid had said, what would come, would come . . . and he would have to meet it when it DID. (Rowling : ) () A: What about John? Did he hit a home run? B: John DIDN’T hit a home run, but I know a wóman who DID.

(Sag : )

The pitch extraction contour for () is provided in Figure .. It shows that the negative auxiliary in the first conjunct and the positive polarity item in the second conjunct are both heavily accented. 320 300

Pitch (Hz)

250

200

150 120 h*

John 0.22

didn’t hit a

h–h%

home

run

but I

h*

know a woman who

h*

l–l%

did

Time (s)

2.93

 . Pitch extraction contour of extraction from VPE

Alternative prosodic realizations are marked, because they violate (), as seen in (a, b): () a. *John DIDN’T hit a home run, and I know a wóman who DIDN’T. b. *John hit a home run, and I know a WOMAN who did. The contrast-locality condition for VPE extraction also applies to Johnson’s example in (B) (from Johnson b: ). (B) is a discourse-adequate answer to (A), while (B) is not. ()

A: Do you know which book was read by whom and which book wasn’t? B: I know which book MANNY réad, and which book LENNY DIDN’T. B: *I know which BOOK MANNY réad, and which PAPER he did.

The contrast-locality condition also applies to topicalization from VPE, as seen in (B) (from Merchant ). The parallelism of the construction is also expressed prosodically with rising topic accents on the topicalized elements and further contrasting accents on the

  



auxiliaries. The violation of () leads to the ungrammatical (B) where the auxiliary in the second conjunct is not focused. ()

A: Which classes should you take and which shouldn’t you? B: GREEK, you SHOULD take; DUTCH, you SHOULDN’t. B: *GREEK, you SHOULD take; DUTCH, you should.

Building on Schuyler’s insight, Merchant (a) proposes the MaxElide constraint, which states that under A’-extraction out of ellipsis the largest possible constituent must be elided (cf. also Lasnik ; Merchant ; Takahashi and Fox ). This observation seems to capture the contrast in () at first sight (cf. van Craenenbroeck and Merchant : ). ()

a. *They attended a lecture on a Balkan language, but I don’t know which they did. b. They attended a lecture on a Balkan language, but I don’t know which.

However, a more principled explanation of the facts is still missing. It is unclear whether the notion of economy implicit in the MaxElide constraint can account for the optionality requirement which holds for ellipsis in general. For example, one would expect (a) to be felicitous in the fully spelled version, as in (). ()

#They attended a lecture on a Balkan language, but I don’t know which they attended.

However, () is still marked because it does not fulfill the discourse appropriateness conditions in the undeleted version. An alternative explanation is that extraction from VPE only occurs in parallel structures and must therefore strictly follow the PC and the CRC, while sluicing, which occurs in an indirect question, doesn’t have to strictly follow these constraints (cf. Winkler ). Under this view, the ungrammaticality of (a) follows from the fact that the focus requirements on sluicing differ from those on VPE extraction: while extraction from VP is subject to the PC and CRC, the wh-remnant in sluicing must itself be a pitchaccented contrastive focus which occurs in a set relation to the focused antecedent. That is, there are diverging focus and prosody requirements on the extraction from VPE and sluicing which are difficult (but not impossible) to fulfill at the same time. The diagnostic test of question–answer congruence makes the focus requirements for (a) explicit. The relevant observation is that each of the elliptical constructions with wh-extraction must be an answer to an appropriate question, as seen in ()–(): wh-extraction from VPE answers an implicit or explicit parallel VP question like What did Ed do and what did Mary do? as in (A). () A: What did Ed do and what did Mary do? B: ED attended a lecture on cárpenting, but I don’t know what MARY díd. (van Craenenbroeck and Merchant : ; secondary accents added) Sluicing can only occur in an indirect question and must have an indefinite antecedent that is questioned.

 ()

 A: What did they want to attend? B: They wanted to attend some LECTURE, but I don’t know WHICH ONE.

Bringing together both information-structural conditions requires a complex parallel context question which induces a parallel focus structure and parallel prosody in the answer, as in (): ()

A: Which of the lectures did Ed attend and which of the lectures did Mary attend? B: I heard that ED attended a lecture on a BALKAN language, but I don’t know WHICH one MARY díd.

The prosodic structure of the wh-extraction from VPE in (B) conforms optimally to the focus requirements of the question in (A). Parallel contrastive prosody is realized on Ed and on Mary as well as on the complex noun phrase lecture on Balkan languages and which one. Thus, the perfectly grammatical (B) raises some doubt about the validity of the grammatical principle MaxElide and provides an argument for the assumption that information-structurally controlled context and prosody are the factors that determine the appropriateness of extraction from ellipsis. There are two important conclusions of this discussion: First, the extraction from ellipsis examples show that there must be structure in the ellipsis site. Second, it shows that specific information-structural and prosodic requirements must be met for extraction from VPE and from sluicing.

.. Locality restrictions and prosody The observation that A’-extraction is possible out of contrastive ellipsis and VPE under certain syntactic and prosodic conditions constitutes a strong argument for movement. However, it must also be shown that A’-extraction is subject to the same locality restrictions as the fully spelled-out versions of the ellipsis clause. Research over the last three decades has shown that A’-movement out of an island under VPE results in ungrammaticality, as in () and () (cf. Haïk : ; Johnson b: ; Merchant b: –). () a. *John met everyone that Peter wondered when he could. (Haïk : ) b. *John read every book which Bill believes the claim that he did. () a. *I know which book Mag read, and which book Mr Yunioshi asked why you hadn’t. b. *I know which book Mag read, and which book Mr Yunioshi read my report that you hadn’t. c. *I know which book Mag read, and which book Mr Yunioshi discussed after I had. (Johnson b: ) Kennedy (: ) shows that similar observations hold for topicalization. An object DP cannot be topicalized out of a VPE in a relative clause, as seen by the ungrammaticality of ().

   ()



*DOGS, I underSTAND, but CATS, I don’t know a single person who DOES.

To the best of my knowledge, there has not been an explicit discussion of the prosodic properties of these types of extractions. Also rating studies that manipulate the prosody and verify the grammaticality judgments reported here remain a desideratum. A visualization of the second conjunct of () in Figure . shows that the island violation cannot be prosodically repaired. 325 300

Pitch (Hz)

250 200 150 100 l + h* but

cats

h–

h*

I don’t know a

single

h* person

h* who

l–l% does

1.58

4.15 Time (s)

 . Pitch extraction contour of topicalization from VPE in a relative clause

Although the prosodic requirements on this extraction construction are met (parallel contrastive accents on topicalized phrases, accent on VPE licensing auxiliary does), it seems that prosody cannot save extraction out of the island configuration. Example () remains highly marked. This is different in sluicing where the island is fully contained in the ellipsis site, as in the widely discussed island repair cases (see Abels, this volume, and references cited there). An example that shows that the wh-remnant and its antecedent are obligatorily focused: () The police officer arrested a REPEAT OFFENDER who robbed a STORE, a. do you know WHO? b. do you know WHICH (one)? In (a), the remnant who refers to the object antecedent a repeat offender who robbed a store. In (b), the remnant which (one) refers to the object DP a store inside the relative clause. As relative clauses are islands to movement, the version of (b) without ellipsis, given in (), is ungrammatical. The extraction site is marked by underscore: () *The police officer is looking for a repeat offender who robbed a store, do you know [CP [which store]i [the police officer is looking for a repeat offender [CP who robbed _i ]].





As already observed by Ross (b), (b) is considerably better than (). Previous analyses of sluicing have established that antecedent selection in sluicing is determined by information structure (e.g., Romero ; Winkler ). A recent empirical investigation of sluicing in German showed that sluicing with relative clause antecedents is as acceptable as main clause antecedents if parallel focus assignment is respected and if the relative clause is extraposed which is argued to add additional prosodic weight to the focused antecedent (cf. Konietzko et al. forthcoming). The general answer to the question (iii) is that prosody provides indirect evidence for the structure in ellipsis site. There are two arguments: The first one is theoretical. The theoretical account of ellipsis as a phonological deletion process at the level of phonological form is prosody-related. The second is empirical. Close investigation of the main syntactic argument for structure in the ellipsis site, namely extraction from the ellipsis site, shows that there are prosodic requirements and discourse requirements that license extraction. Thus, there is prosodic evidence that supports the assumption that there is structure in the ellipsis site.

. C

.................................................................................................................................. This chapter has investigated ellipsis from the perspective of prosody. It has provided a summary of the prosodic system of English and reviewed the research on how prosody bears on the central issues of the theory of ellipsis. The chapter has focused on three prosody-related questions raised by the research on ellipsis: (i) How does prosody contribute to the licensing of ellipsis? (ii) How does prosody contribute to the recovery of the ellipsis site? (iii) What evidence can prosody contribute for the assumption that there is structure in the ellipsis site? With respect to the first question, it could be shown that most theories assume that the deletion of linguistic material is subject to formal syntactic, semantic, and prosodic conditions. There are in particular two prosodic constraints that interact with contextual features, the givenness-marking hypothesis and the contrastive remnant condition. In addition, there is a prosodic requirement on the parallelism condition. As for the second question, the discussion has shown that prosodic features play a considerable role in the recovery of ellipsis. In particular, it has been shown that prosody figures prominently in the disambiguation process of different readings of elliptical constructions. With respect to the third question, the prosodic evidence that there is structure in the ellipsis site is indirect. However, it could be shown that A’-extraction out of an ellipsis site is subject to prosodic constraints. Finally, the discussion has shown that the interaction of prosody and absence of sound contributes to the understanding of the theories of syntax–sound–meaning correspondence.

A This chapter has benefited from the valuable comments of the editors of this handbook, Jeroen van Craenenbroeck and Tanja Temmerman, and two anonymous reviewers. I am also grateful to Peter Culicover, Andreas Konietzko, Jason Merchant, Valéria Molnár, and Michael Rochemont for comments, fruitful discussions, and feedback. This material is based upon work supported by the German Research Foundation (Deutsche Forschungsgemeinschaft DFG) under the following grants: SFB , RTG .

  .............................................................................................................

ELLIPSIS AS A DIAGNOSTIC TOOL .............................................................................................................

  ......................................................................................................................

                 ......................................................................................................................

 

. I

.................................................................................................................................. T chapter discusses two strands of research where the interaction of ellipsis with movement has been used to construct arguments pertaining to the implementation of movement in the grammar, the architecture of the grammar, and the nature of the grammar more broadly. Such discussions stand in contrast to the simpler, more customary arguments where the possibility of extracting elements from an ellipsis site (or lack thereof) is used as a probe in the analysis of a particular elliptical construction (keeping assumptions about the analysis of movement and the architecture of the grammar essentially constant). For example, the possibility of (apparent) extraction from a particular ellipsis site might be taken as evidence for the presence of articulated syntactic structure at the ellipsis site (see Aelbrecht ; Baltin  for clear examples of this line of argumentation). Here we take the other perspective. We ask what ellipsis can teach us about the nature of movement and the grammar in general. Of course, the distinction is one of emphasis rather than one of principle, since analytical discussion informs theory and vice versa. Both lines of research discussed in this chapter take as their empirical starting point asymmetries in the behavior of moved constituents depending on whether movement originates within an ellipsis site or not. Movement from an ellipsis site is sometimes less restricted than movement from an overt phrase hypothesized to correspond to the silent structure at the ellipsis site and sometimes more restricted. The chapter is structured as follows. Section . introduces the phenomenon of movement. W-movement will be shown to be unbounded in some sense, though subject to certain constraints. We briefly review the notion of a syntactic island, that is, of a domain from which -movement is impossible. Then we introduce the idea that movement dependencies are created in a successive cyclic fashion; it will be contrasted with some important alternative views. Section . serves as background. It can be skipped by readers familiar with unbounded -movement, syntactic island phenomena, and the ways in which unbounded dependencies are modeled in different syntactic frameworks. Section . discusses the interaction of the elliptical phenomenon of sluicing with syntactic island constraints. Under certain assumptions about the analysis of ellipsis in general and sluicing in particular, sluicing ameliorates island effects (Ross b). The





section introduces some basic facts about sluicing, concentrating on case connectivity between antecedent and remnant and on Ross’s conjecture: the idea that ellipsis repairs island violations. Subsection .. explores the two canonical approaches to Ross’s conjecture, these are first the idea that ellipsis literally repairs island violations and second the idea that island violations cannot arise under ellipsis because there is no syntactic structure at the ellipsis site. Subsection .. introduces two systematic counterexamples to Ross’s conjecture—contrast sluicing and multiple sluicing—and reviews approaches to sluicing that take these counterexamples at face value. Subsection .. outlines the contour of a recent debate on the question of how damaging the prima facie counterexamples to Ross’s conjecture are. It is shown that the counterexamples to Ross’s conjecture tilt the balance in favor of approaches to ellipsis that assume the presence of unpronounced syntactic structure at the ellipsis site and against approaches that do not assume the presence of such structure. Among the approaches that do posit syntactic structure at the ellipsis site, a consensual answer to the question of whether the identity condition on ellipsis is syntactic or semantic (or both) has not been reached. Section . turns to VP ellipsis and related phenomena. VP ellipsis allows certain types of extraction from the ellipsis site but not others—at least not in an unrestricted way. The complex restrictions on movement from the ellipsis site have been used as indirect arguments for a particular implementation of movement, cyclicity, and ellipsis. Facts and arguments in this area have not enjoyed the prominence of the island amelioration phenomenon. The section discusses some observations, sketches the arguments, and points out some open questions. Any conclusions must, at this point, remain extremely tentative.

. M, , 

.................................................................................................................................. This section serves as a brief reminder of some of the basic properties of the phenomenon that the generative literature calls movement and which is, in various contexts and theories, also referred to as “long distance dependency”, “unbounded dependency”, “displacement”, or “extraction”. A slightly more in-depth introduction to the phenomenon can be found in Abels (a). Example () illustrates movement of the interrogative pronoun ‘what’ to the front of the sentence: ()

What did John {make | burn | *sleep}?

There is a dependency between the interrogative pronoun and the verb: It depends on the choice of verb whether the interrogative pronoun can appear at all—impossible with an intransitive verb like ‘sleep’—and, if it is possible, which interpretation is assigned to ‘what’—the thing created with ‘make’ and the thing destroyed with ‘burn’. Verbs that obligatorily take objects, like ‘devour’ in (), allow their object position, exceptionally, to be empty when a suitable interrogative pronoun introduces the sentence, (b), and, indeed, in such a case the object position must remain empty, (c). In a nutshell, the interrogative pronoun acts semantically and for purposes of selection as the object of the verb in these examples. Gaps can occur in various positions in the clause, with correspondingly different interpretations, and different grammatical functions.

   ()



a. John devoured *(something). b. What John devour? c. *What did John devour something?

We can describe the relation between the interrogative pronoun and the object position in terms of fillers and gaps. In () ‘what’ is the filler and the object position of the verbs ‘make’ and ‘burn’ is the gap, as shown in (). The gap is indicated by an underscore. () What did John {make | burn} ___?   We have seen so far that the filler fulfills the selectional requirement associated with the gap and receives the thematic interpretation associated with the gap. In languages that show case morphology on interrogative pronouns, the form of the interrogative pronouns varies in the same way with the verb as the form of a non-interrogative phrase at the gap site would: () shows that the German verb unterstützen ‘support’ takes an accusative object while helfen ‘help’ takes a dative object. Questions about those being supported or helped must be introduced by interrogative pronouns in the appropriate form: accusative for unterstützen ‘support’ and dative for helfen ‘help’, (). () a. Der Junge will {einen | *einem} the. boy wants a. a. ‘The boy wants to support a man.’

Mann man

unterstützen. support

b. Der Junge will {*einen | einem} the. boy wants a. a. ‘The boy wants to help a man.’

Mann man

helfen. help

() a. {Wen | *Wem} will der who. who. wants the. ‘Who does the boy want to support?’ b. {*Wen | Wem} will der who. who. wants the. ‘Who does the boy want to help?’

Junge ___ unterstützen? boy support Junge ___ helfen? boy help

The syntactic category of a clause will here be abbreviated as CP (complementizer phrase). All CPs in () are enclosed by square brackets. Questions introduced by an interrogative pronoun or an interrogative phrase are called -questions. In such questions the -phrase occupies the specifier of CP in English. ()

a. [What should I bring ___]? b. [John asked [ what he should bring ___ ]].

Furthermore, the complementizer position is occupied by an auxiliary in direct questions (a), while the complementizer position remains empty in indirect -questions, (b). The category of the remainder of the clause will be abbreviated here as TP (tense phrase).





Thus, the very approximate structure of the indirect question in (b) can be represented as in (). The gap is represented in the structure by what. Strikethrough encodes the idea that the gap site is occupied by a silent copy of the filler. The dependency itself is represented by the arrow pointing from the silent copy to the filler. It represents the movement relation holding between the filler and the gap. Theories and analyses differ in how exactly the movement relation is represented and whether silent elements (copies or traces) occupy the site of the gap or not. Some of these issues will be discussed in subsection .. and section .. The notation with a struck-out copy and with (or usually without) an additional movement arrow to indicate the long-distance dependency will be used for convenience and perspicuity when nothing hinges on the choice. ()

CP DP

C’ C

what

TP he

should

bring

what

.. Islands The examples up to this point have all contained a filler and a corresponding gap in the same clause. It is important to note, though, that filler–gap dependencies can be unbounded, i.e., there is no grammatically determined upper limit to the linear and structural distance between filler and gap, (). In (a) the linear distance between filler and gap is seven words, in (b) it is ten words, in (c) twelve, and in (d) fifteen. The structural distance between filler and gap, as measured by how deeply embedded the gap is compared to the filler is one clause in (a), two in (b), three in (c), and zero in (d). In the latter, filler and gap are both in the main clause. ()

a. b. c. d.

What did Mary claim [ John tried to devour what ]? What has Bill heard [ that Mary claimed [ John tried to devour what ]]? What do you think [ Bill heard [ that Mary claimed [ John tried to devour what ]]]? Who does the allegation [ that Mary sold state secrets to the Russians for personal gain ] annoy who?

The observation that the distance between filler and gap can be unbounded does not entail by any means that it is unconstrained. Ross () was the first to systematically catalogue constraints on movement. The discovery of the existence of systematic structural constraints on filler–gap dependencies is among the most important empirical discoveries of modern syntactic theory. Ross () called a particular structural configuration an island if the gap cannot grammatically be situated inside of the configuration while the filler is outside. (For an excellent overview and introduction to the topic, see Szabolcsi .)

  



Among the island constraints operative in English, the following four are illustrated here, since they play a particular role in the discussion later on in the chapter: the Complex NP Constraint (going back to Ross : ), which bans movement of elements contained in a clause dominated by a noun phrase (DP) out of that noun phrase; the Subject Condition (see Ross :  for clausal and Chomsky : – for DP subjects), which bans extraction from subjects; the Adjunct Condition (Huang ), which bans extraction from clausal adjuncts; and the Left-Branch Condition (Ross : ), which bans movement of the leftmost constituent of a DP containing other material. The constraints are illustrated below. In each case, the ungrammatical extraction is accompanied by relevant semantic and structural controls. The Complex NP Constraint is illustrated in (a); the remaining examples in () serve as controls. The Subject Condition is illustrated in (). In each case, the grammatical subject is enclosed in square brackets. The grammatical and ungrammatical examples differ in what constituent acts as the subject. The grammatical controls show that extraction from those constituents is possible in principle, the ungrammatical versions show that the sentence becomes degraded when the constituent in question acts as the subject. There are two examples of DP subjects, (b) and (d), and one of a clausal subject, (f ). The Adjunct Condition is illustrated in (). The grammatical and ungrammatical examples are synonymous. In the grammatical cases, the extraction site is in the main clause; in the ungrammatical cases, in the adjunct clause. Finally, () illustrates the Left-Branch Condition. In the ungrammatical example, a left branch within a DP has been extracted. The grammatical example is synonymous; the violation is avoided by moving the entire containing DP under a strategy that Ross () called ‘pied-piping’. ()

Complex NP constraint a. *Which language does Abby want to hire [ someone [ who speaks which language ]]? b. Abby wants to hire [ someone [ who speaks Greek ]]. c. Which language is Abby looking for [ a speaker of which language ]?

()

Subject Condition a. Which of the Marx brothers will [they] publish a biography of which of the Marx brothers later this year? b. *Which of the Marx brothers will [ a biography of which of the Marx brothers ] be published later this year? c. Who is [ there ] a picture of who on the wall? d. *Who does [ a picture of who ] hang on the wall? e. What is [ it ] illegal to chain what to the fence? f. *What is [ to chain what to the fence ] illegal?

()

Adjunct Condition a. Who had John talked to who [ before he went home ]? b. *Who did John go home [ after he had talked to who ]? c. What does Friederike listen to what [ while she does her homework ]?



 d. *What does Friederike do her homework [ while she listens to what ]?

()

Left-Branch Condition a. *How big did you empty [ how big a cup]? b. How big a cup did you empty how big a cup?

In the immediately preceding discussion, we saw that movement of -phrases in English is characterized simultaneously by its unbounded nature and its sensitivity to island configurations. The same simultaneous characterization also describes a substantial number of other constructions in English involving long-distance dependencies and gaps (see Chomsky b). Even more significantly, this characterization describes long-distance dependencies in many other languages as well (Ross ). While there can be no doubt about the importance of island effects to linguistic theory, the question of the ultimate cause of island effects remains disputed. There are three different styles of answer to this question. One line of thinking holds that island effects reflect constraints on syntactic derivations or representations (Chomsky , ; Huang ; Pesetsky ; Cinque ; Rizzi ; Manzini ; Starke ; Stepanov ; Müller ). Under such an approach, island violations cannot be derived and/or always lead to a deviant output. According to a second tradition, island violations are not syntactic violations proper. Instead, islands are viewed as derivable and syntactically well-formed but deviant either at the interface between syntax and a different system (interpretation for Lasnik and Saito ; Chomsky and Lasnik , information structure for Erteschik-Shir and Lappin ; Van Valin and LaPolla ; Erteschik-Shir , ; Goldberg , and phonology for Chomsky ; Uriagereka ; Hornstein, Lasnik, and Uriagereka ; Abe and Hornstein ) or as deviant representations in a non-syntactic module (semantics in the account of weak islands in Szabolcsi and Zwarts ; Abrusán ). A third strand of thinking treats some island effects as the result of processor overload: island structures are not ungrammatical in any sense but lead to a breakdown of the parser due to their complexity (see Pritchett ; Kluender ; Sag, Hofmeister, and Snider ; Hofmeister and Sag ; Hofmeister, Staum Casasanto, and Sag ; Kluender and Gieselman ; among others). We will see in section . how the interaction of island effects with ellipsis has been used in arguments concerning the nature of island effects.

.. Successive cyclicity and punctuated paths As mentioned in section .., the discovery of unbounded filler–gap dependencies and of island effects proved extremely important to the field of modern syntax. Once their existence is recognized, the question immediately arises how such unbounded dependencies are implemented in the grammar. We saw above that the filler has semantic properties usually associated with the gap site, (), and that it is morphologically realized in the case form usually associated with the site of the gap, (). Clearly, syntactic theory must explain how the grammatical properties usually associated with the gap site end up being expressed on the filler and why the filler must obey constraints associated with the gap. Two main metaphors have guided theorizing about the question of how filler–gap dependencies are mediated: The first metaphor is movement, the second metaphor is percolation.

  



The movement metaphor says that, as part of the process that derives sentences with filler–gap dependencies, the filler moves from the position of the gap to the position of the filler, occupying each at distinct points of this derivation. In elaborating this metaphor, theorists have posited that unbounded movement is implemented as an arbitrarily long sequence of short, bounded steps. The idea of an arbitrarily long sequence of individually bounded movement steps has been exploited to explain some of the island effects mentioned above (see, for example, Chomsky , b, ), to explain why a filler–gap dependency may influence the shape of elements along the structural path of movement that are locally related neither to the position of the filler nor to the position of the gap (see, for example, McCloskey ), and to explain why the filler itself may have properties associated uniquely with certain positions along the path of movement (see, for example, the discussion of intermediate reconstruction and chain binding in Barss ; Fox ; Lebeaux ). With the ancillary assumption that movement targets the edge of VP and the edge of CP (Chomsky , , ), example () would roughly have the representation in (). It is important to note how the filler leapfrogs through the structure, entering very local relations with some nodes along the path, those boxed gray in the diagram, but not with others, particularly the TP nodes in the structure in (). ()

I must know what Mary will claim John has tried to devour.

() TP NP

T’ T

VP V

CP NP N

C’ C

TP NP N

T’ T

VP NP N

VP V

CP NP

C’

N C

TP NP N

T’ T

VP NP N

VP V

TP T

VP NP N

VP V

NP N

I must know what ∅ Mary will what claim what ∅ John has what tried to what devour what





Under the percolation metaphor, the filler never occupies the site of the gap. Instead, the information that there is a gap, what its category and interpretation are, and what restrictions are placed on it, percolates through the syntactic structure locally, from node to immediately dominating node. This percolation of information is depicted by the gray shading in ().

()

TP NP

T’ T

VP V

CP NP N

C’ C

TP NP N

T’ T

VP V

CP TP

C NP N

T’ T

VP V

TP T

VP V ___

I must know what ∅ Mary will claim ∅ John has tried to devour

There are three main differences between the percolation conception and the movement conception of filler–gap dependencies. First, under the movement conception but not under the percolation conception, the filler occupies the position of the gap at some stage of the syntactic derivation. Second, under the movement conception but not under the percolation conception, the filler occupies positions between the position of the gap and that of the filler at some stage of the syntactic derivation. Third, under the percolation conception but not under the movement conception, the information that a particular constituent contains a gap, its nature, case, and interpretation are locally available at every node along the structural path between filler and gap. Abels () coined the terms punctuated versus uniform paths for the movement conception and the percolation conception of movement, respectively. Section . discusses arguments from the interaction between movement and ellipsis that purport to demonstrate the superiority of a model with punctuated paths, that is, a model based on the movement metaphor.

  



. S  

.................................................................................................................................. Ross (b) was the first generative paper to discuss sluicing. Sluicing is a type of elliptical -question, in which only the -phrase is pronounced and the rest of the question is elided. For an introduction to the phenomenon and leading analyses, see Vicente (this volume). (a–c.i) provide typical examples of sluicing. The understood question in each case is spelled out in (a–c.ii), but elided save for the -phrase in the sluice. () a. (i) John bought a car, but I don’t know which one. (ii) John bought a car, but I don’t know which one John bought. b. (i) Somebody just left. Guess who. (ii) Somebody just left. Guess who just left. c. (i) Joe just left, but I am not sure why. (ii) Joe just left, but I am not sure why Joe just left. I will use the following terminology: The entire elliptical question will be called the sluice; the pronounced -phrase will be called the remnant; the unpronounced remainder of the sluiced question will be called the ellipsis site. The clause providing the content for the sluice will be called the antecedent; finally, the phrase in the antecedent that intuitively corresponds to the remnant, if there is such a phrase, will be called the correlate. ()

[John bought [ a car ]] but I don’t know [[which one] [correlate] [ remnant ] [ antecedent ] [ sluice

] ellipsis site ]

Furthermore, the following three types of sluices will be distinguished terminologically whenever this is necessary. In what I will call merger-type sluices following Chung, Ladusaw, and McCloskey (), the remnant has an overt, indefinite correlate whose identity is in question. Examples were provided in (a.i) and (b.i). In contrast sluices, the remnant has an overt, usually definite correlate. The remnant asks for alternatives to the correlate. An example is given in (). Finally in the sprouting type of sluicing, the remnant has no overt correlate, (c.i). ()

John read War and Peace, but I don’t know what else.

With terminology settled, we turn to three important facts discovered by Ross. First, Ross (b) argued that sluices are clauses as far as their syntactic category is concerned, a conclusion that is consensual in the literature (for a review of the arguments see Levin ; Vicente this volume). Beyond this, Ross (b) contains two claims about sluicing that, taken at face value, point to mutually incompatible conclusions about the nature of the ellipsis site. The first claim is given a general statement by Merchant (: ch. ): The remnant must bear the case that its correlate bears. This observation is illustrated in () with German examples. Recall from () that the verb helfen ‘help’ takes its object in the dative case while unterstützen ‘support’ takes an object in the accusative case and that this case marking is





maintained under movement. Now observe the behavior of the remnant -phrase when the correlate is either the object of helfen ‘help’ or unterstützen ‘support’. ()

a. Hans will jemandem helfen, aber ich weiß nicht {wem | *wen}. Hans wants someone. help but I know not who. who. ‘Hans wants to help someone, but I don’t know who.’ b. Hans will jemanden unterstützen, aber ich weiß Hans wants someone. support but I know {*wem | wen}. who. who. ‘Hans wants to support someone, but I don’t know who.’

nicht not

As can be seen, the case of the remnant is the same as the case of the correlate. I will refer to this phenomenon as the case-matching effect. Ross (b) argues that this behavior can be accounted for on the assumption that the sluice has the same syntactic structure as the synonymous full question but that the structure may remain unpronounced just in case it is structurally identical to the antecedent. Under these assumptions, no new mechanisms of case assignment or interpretation are needed. The only mechanism needed to derive sluicing is a deletion/non-pronunciation operation under syntactic identity.1 We can represent this analysis of sluicing as deletion under syntactic identity schematically as follows, where grayed-out material in angle brackets indicates the analysis of the unpronounced syntactic structure present at the ellipsis site: ()

John bought a car, but I don’t know which one .

Under a theory of ellipsis under syntactic identity, examples like those below motivate the claim that merger-type sluicing does not obey islands. The crucial behavior is illustrated here only for those island constraints that were introduced above in ... Other island effects are discussed in Ross (b), Chung, Ladusaw, and McCloskey (), and Merchant (). Sauerland () concentrates on so-called weak islands. In ()–(), the example with sluicing is always given first. It is followed by the sentence which, under a syntactic identity analysis of sluicing, would be the syntactic representation of the sluice, the pre-sluice in the terminology of Dayal and Schwarzschild (). Example () offers a third example, whose function is to show that independent ellipsis of the noun phrase ‘a mug’ is unavailable. ()

Complex NP Constraint a. Abby wants to hire someone who speaks a Balkan language, but I can’t remember which (Balkan language). b. *Abby wants to hire someone who speaks a Balkan language, but I can’t remember which (Balkan language) Abby wants to hire [someone who speaks which (Balkan language)].

1 We review some of the reasons (see Merchant : ch. ; Vicente this volume, and for related arguments from other elliptical constructions Culicover and Jackendoff : ch. .) to doubt that this analysis is correct. Some such examples were already noted in Ross (b) but interpreted differently there.

   ()



Subject Condition a. A biography of one of the Marx brothers will be published later this year, guess which (of the Marx brothers). b. *A biography of one of the Marx brothers will be published later this year, guess which (of the Marx brothers) [a biography of which of the Marx brothers] will be published later this year.

()

Sentential Subject Constraint a. That certain countries would vote against the resolution has been widely reported, but I am not sure which ones. b. *That certain countries would vote against the resolution has been widely reported, but I am not sure which ones [that which ones would vote against the resolution has been widely reported].

()

Adjunct Condition a. Ben will be mad if Abby talks to one of the teachers, but I can’t remember which (of the teachers). b. *Ben will be mad if Abby talks to one of the teachers, but I can’t remember which (of the teachers) [Ben will be mad if Abby talks to which (of the teachers)].

()

Left-Branch Condition a. He emptied a big mug, but he wouldn’t tell me how big. b. *He emptied a big mug, but he wouldn’t tell me how big he emptied [ how big a mug]. c. *He emptied a big mug, but he wouldn’t tell me how big *(a mug) he emptied how big a mug.

The examples with sluicing are much more acceptable than their putative sources.2 The phenomenon that sluices are much more acceptable than the island-containing sources that an analysis of sluicing as deletion/non-pronunciation under syntactic identity would posit is known as island amelioration or island repair under sluicing. The phenomenon is surprising because, if the ellipsis site is indeed occupied by syntactic structure identical to the antecedent and if island effects are syntactic in nature, then we expect ellipsis to give rise to the same robust type of effect found in examples without ellipsis. Ross assumed that island repair characterizes all types of sluicing and, in fact, other elliptical processes as well. Consequently, his paper ends (Ross b: , ex. ) by reformulating his own earlier () syntactic theory of islands as follows: “If a node is moved out of its island [footnote omitted, K.A.], an ungrammatical sentence will result. If the island-forming node does not appear in surface structure, violations of lesser severity will (in general) ensue.” The leap from merger-type sluices, which are discussed in detail in Ross (b), to all types of sluicing and to all ellipsis phenomena is not motivated in Ross’s paper. However, the generalized form of the claim—that is, that there is a phenomenon of 2 The more modern literature, such as Levin () and Chung, Ladusaw, and McCloskey (), usually treats sluicing examples like those in ()–() as fully acceptable, while Ross (b) only claimed that the sluices were much improved compared to their putative source while still falling short of full acceptability.





island repair by ellipsis—has proved very influential in the literature. I will refer to the claim that island violations are repaired by ellipsis in general as Ross’s conjecture. I will now briefly introduce a cross-classification of theories of sluicing that structures the following discussion. Subsection .. explores some consequences of assuming that Ross’s conjecture is essentially correct. Subsection .. will question whether Ross’s conjecture is factually correct and explore consequences of the assumption that it is incorrect. Subsection .. attempts to assess how serious the challenges to Ross’s conjecture really are. Theories of sluicing can be cross-classified by the answers they give to the following three questions: Is there contextually variable syntactic structure present at the ellipsis site? Is ellipsis conditional on a syntactic identity relation between the antecedent and the sluice? Is ellipsis fed by regular -movement (regular movement of the remnant)? Ross’s analysis of sluicing assumes that the answer to all three of these question is yes. The table in () shows where various approaches situate themselves.3 Question : Is the identity condition on ellipsis syntactic?

Question : Is sluicing fed by regular movement? Theory

yes

yes

yes

yes

yes

no

yes

no

yes

yes

no

no

no no no

yes no no

yes/no yes no

() Question : Is there contextually variable syntactic structure at the ellipsis site?

Ross (b), Lakoff (), Chomsky (a), Lasnik (), Müller (), and Griffiths and Lipták () Chung, Ladusaw, and McCloskey () and Abe () Baker and Brame (), Merchant (), Fukaya (, ) van Craenenbroeck (a,b), Barros (b), Barros, Elliott, and Thoms (), and Abels (b) – N/A (incoherent) Erteschik (: –, n. )4 Levin (), Dalrymple, Shieber, and Pereira (), Ginzburg and Sag (), Culicover and Jackendoff (), and Barker ()

3 The table is essentially adapted from the two-dimensional matrix in Merchant (this volume). Question  is added as a third dimension. The option of having a partially syntactic and partially semantic identity condition is suppressed here for simplicity. 4 For reasons given in Merchant (: ch. ), this idea has never been seriously pursued for English, though an analysis in this spirit is pursued in Adams and Tomioka () for Mandarin Chinese.

  



The next subsection (subsection ..) explores the extreme positions marked out in this table: ‘No’ to all questions and ‘yes’ to all questions. Subsection .. then turns to mixed positions.

.. Ross’s conjecture and its consequences The first question in the table in () distinguishes theories that assume that the ellipsis site is occupied by syntactic structure that varies in some systematic way with the structure of the antecedent (‘yes’ for question ) from theories that assume that there is either no syntactic structure at all at the ellipsis site or that ellipsis targets some fixed, contextually invariable structure (‘no’ for question ). Proponents of the latter type of approach must assume that the identity condition on ellipsis is non-syntactic. This is why one set of positions is marked as ‘incoherent’ in the table. I will not comment further on these. Furthermore, the question whether or not there is -movement at the ellipsis site is of minor importance in accounts that don’t assume the presence of variable syntactic structure at the ellipsis site, and I will gloss over the distinction in the discussion to follow. Proponents of such theories usually cite the lack of island effects under sluicing (and Ross’s conjecture in general) as prima facie evidence for the absence of (variable) syntactic structure. The logic is simple: there is no island at the ellipsis site, hence no island effect. While lack of (contextually variable) structure provides an attractive solution to the island amelioration problem, it raises the question of how to account for the case-matching effect. Theories without (variable) structure at the ellipsis site require some form of transmission of case from the correlate to the remnant. Usually, syntactic theories allow for the full expression of morphological case only in a strictly circumscribed set of structural configurations. There must be a case licensor and case bearer and they must stand structurally in a case-licensing relation. This case-licensing relation has all the hallmarks of a well-behaved syntactic relation (Koster ; Neeleman and van de Koot ): It requires c-command, locality, and uniqueness of the licensor. However, in sluicing, the remnant does not stand in this type of relation to its putative case licensor—the remnant or the remnant’s case licensor in the antecedent. We are left with the question of whether case is a syntactic phenomenon, as all other evidence suggests, or a non-syntactic one, as case matching suggests under approaches to sluicing without syntactic structure at the ellipsis site.5 A further possibility formulable under some but not all theories of syntax would be the position that case is syntactic except

5

Pronominal forms in English fragments do not appear in the form that would be expected in the non-elliptical versions of the answers, as shown in (i). (i) a. Who wants an ice cream? b. Me. This may be an instance where the case-matching generalization breaks down. However, analyses exist that treat pronominal forms as contextual allomorphs (see Parrott , ) rather than different case forms. If such an analysis is correct, English fragments cease to be a counterexample to case matching. Evidence for this position comes from the somewhat erratic behavior of pronouns in coordinated structures (My brother and {me | I} will be in town. Joe invited Fred and {me | I}.), a behavior that is not shared by pronouns in languages with bona fide case.





in elliptical constructions. What this shows is that a definitive argument that there is no (contextually variable) syntactic structure at the ellipsis site would provide a strong argument against the assumption that case is (always) a syntactic phenomenon.6 It is difficult to overstate the importance of this issue for syntactic theory. Nevertheless, I will not pursue the issue further, since the evidence surveyed later in this chapter strongly suggests that there is contextually variable syntactic structure at the ellipsis site. Culicover and Jackendoff () stress a further consequence of their approach to ellipsis, which admits no syntactic structure at the ellipsis site at all: the syntax-semantics mapping must be flexible and powerful enough to allow the sluice to be interpreted as a -question despite its incompleteness. We thus end up with two important conclusions. If it can be shown that there is no (variable) syntactic structure at the ellipsis site, then case cannot be (strictly) syntactic. And if there is no syntactic structure at the ellipsis site at all, then the syntax–semantics mapping must be quite powerful and flexible. While we have just sketched some consequences of assuming that there is no (variable) syntactic structure at the ellipsis site, historically the first analysis of sluicing was Ross’s. Ross answered all three questions in table () positively. We will now briefly survey the consequences this move has for the nature of the grammar and the nature of movement. The most immediate consequence of his analysis was pointed out by Ross himself. Syntactic processes (like the -movement rule or the ellipsis rule involved in sluicing) were seen as operations that map one phrase marker onto another with the additional assumptions that rules could be conditioned only by properties of the input phrase marker. At the time, phrase markers did not contain any indications that movement from one position to another along a particular path had taken place. In other words, the representational devices of copies or traces (), movement arrows (), or feature percolation along the path of movement (), were unavailable. Without such representational devices, Ross’s analysis of sluicing entails that “ungrammaticality is a property not of merely deep or surface structures, or of pairs of trees which are related by rules, but rather of derivations” (Ross b: ). In the subsequent debate of the early s, the researchers that espoused the view that ellipsis identity should be syntactically defined adopted one of two solutions to the island amelioration problem. Some researchers essentially endorsed Ross’s solution, which was to expand the types of permissible syntactic rules: that is, these researchers advocated a derivational solution to the problem. Thus, Lakoff (: –) used island amelioration under sluicing as one among several arguments for the claim that constraints on syntactic rules should be allowed to access entire syntactic derivations rather than just adjacent derivational steps. Such rules were called global rules. By contrast, Chomsky (b: –) suggested enriching the representational vocabulary of the theory. In essence, he proposed that when movement crosses an island, the island is marked with the diacritic #. When a 6

The debate about the distribution of preposition stranding under sluicing and its explanation (Merchant ; Sjepanović ; Rodrigues, Nevins, and Vicente ; van Craenenbroeck a; Sato b; Nykiel ; Abels b) can also be viewed in this light. On the joint assumption that the ban on preposition stranding is an indicator of movement and that a remnant PP can only have moved if there is (contextually variable) syntactic structure at the ellipsis site, evidence that preposition stranding is disallowed under sluicing (or a well-defined subset of cases) will then count as evidence for the presence of syntactic structure at the ellipsis site.

  



phrase containing # is subsequently pronounced, the sentence is perceived as deviant. In the case of sluicing (and ellipsis more generally), the # diacritic marking an island violation remains unpronounced, therefore, the deviance disappears.7 The crucial idea here is that movement out of an island is not ungrammatical per se and neither is the # diacritic. Rather, a phrase carrying the # diacritic is perceived as deviant only if it is pronounced. The question of whether global rules were needed and desirable was a highly contentious and charged issue at the time. However, all modern theories of syntax have sufficient representational power (the copies, traces, movement arrows, feature percolation; see section ..) to make both global rules and the diacritic for ungrammaticality superfluous. Note that it does not follow that special diacritics for ungrammaticality are no longer used by syntacticians—they are (see for example Merchant b; Bošković ). However, the fact of island amelioration under sluicing can be stated without a special diacritic and without powerful new rule types because of the enhanced representations employed by current syntactic theories. While modern theories have sufficient power to state Ross’s claim, that is, that islands are repaired by ellipsis, they do not automatically explain this fact. Indeed, Culicover and Jackendoff () highlight just this point: What is it about the nature of movement and islands that makes sentences ungrammatical when the island is pronounced, but acceptable when it is not pronounced?8 The question is addressed in the programmatic and speculative Hornstein, Lasnik, and Uriagereka () and in Müller (). Both approaches rely heavily on the assumed derivational dynamics of syntactic structure building, albeit in very different ways. Hornstein, Lasnik, and Uriagereka () assume, following Chomsky (a), that as far as the relation between the lexicon and the meaning interface is concerned, syntactic structures are purely hierarchical; they are partially ordered by dominance but not by precedence. The fact that words are ordered linearly, Chomsky suggests, is an imposition of the articulatory and perceptual interface (phonology) which can only realize words that are linearly ordered. Uriagereka () then argued that what has to be ordered are the words, that is, the terminal elements of a syntactic structure. Following Kayne (), Uriagereka assumes that asymmetric c-command maps onto temporal precedence.9 Under this approach, since A in () asymmetrically c-commands all of B, C, and D, A precedes all of them. ()

A B

C

D

The # diacritic had been introduced by Chomsky (: ) as a boundary symbol together with a principle whereby # could not appear internally to a pronounced sentence; being a boundary symbol, it could only appear at the edge of a sentence. Chomsky (b) uses the diacritic to make the application of certain rules manipulating it obligatory: In the case at hand, the otherwise optional ellipsis rule becomes obligatory because otherwise # remains internally to a pronounced sentence. 8 Recall from above that Culicover and Jackendoff ’s own analysis posits no syntactic structure at the ellipsis site and therefore explains island amelioration rather than just describing it. 9 A node in a syntactic structure asymmetrically c-commands all and only the nodes that are (irreflexively) dominated by its sister. 7





A difficulty arises in a structure where both sisters are internally complex, (a), because there are no c-command relations (asymmetric or otherwise) between A, B, and C on the one hand and D, E, and F on the other.10 Assuming a bottom-up structure-building mechanism along the lines of Chomsky (a), Uriagereka () suggests that the syntactic derivational engine overcomes this problem as follows: The internal structure of one of the phrases is removed and the phrase is turned into a single (huge) terminal before it is combined with its complex sister, (b). By assumption, the internal order of elements in this new terminal is rigid and its internal parts inaccessible to (context-sensitive) syntactic operations. This move reduces (a) to (). A phrase that has thus lost its internal structure and flexibility is said to have been spelled out. Uriagereka compares the structure A-B-C in (b) to a compound, as compounds, too, are made up of more than one lexical item but behave like syntactic terminals. ()

a.

A

B

C

D

E

F

b. A-B-C D

E

F

Uriagereka suggests that the proposed analysis of complex–complex junctures under which one of the complex phrases has to be linearized internally can derive island effects that hold at such junctures. In particular, following a line of work exemplified by Cattell (), Huang (), Kayne (), and Chomsky (), Uriagereka suggests that there is a unified explanation for the subject condition, the sentential subject constraint, and the adjunct condition. According to Uriagereka, that unifying property is that in each case there is a complex–complex juncture: subjects and adjuncts from which extraction is to take place are internally complex and they are sisters to phrases that are internally complex as well. According to Uriagereka, the syntactic engine must spell out subjects and adjuncts before combining them with their sisters, thus fixing the linear order of their component parts and removing their internal structure from the purview of context-sensitive syntactic operations. This derives the island effects just mentioned. Hornstein, Lasnik, and Uriagereka () build on Urigereka’s theory just outlined. They observe that a phrase that is not pronounced doesn’t need to have its terminals linearly ordered—given that, by assumption, linear order is an imposition only needed to allow pronunciation. They suggest that material that ends up being elided need not be linearized at all. In particular, complex subjects and adjuncts within elided constituents need not be spelled out prior to becoming part of the structure. Consequently, their parts need not be removed from the purview of context-sensitive syntactic operations but can remain 10

Another type of problem arises at the bottom of each complex branch, where two terminals will unavoidably meet. The case is irrelevant for what follows and will not be discussed here.

  



accessible to such operations. Thus, for those island effects that arise because of early spellout of the island, island amelioration under ellipsis is expected. Müller () relates island amelioration to a very different aspect of the dynamics of syntactic computation. He assumes a model of movement roughly along the lines of that depicted in (). However, instead of skipping certain phrases (TPs in ()) there is a copy of the moving item at the edge of each and every phrase. Special movement-inducing features on the head through whose projection the moving element must pass are responsible for inducing these movements. These movement-inducing features are added to the relevant heads in the process of the derivation. They can be added to a head only while that head is still active, that is, while it still has other features that need to be satisfied. In addition, simplifying somewhat, the movement-inducing features must be satisfied as soon as they are added to the head. These joint demands sometimes create a catch-: movement from a given phrase can only be triggered when the head into whose projection the phrase is merged possesses the movement-inducing feature. The movement-inducing feature cannot be added to an inactive head and must be used as soon as it has been added. Therefore, a phrase that is merged and thereby satisfies the last active feature of a head can never be extracted from. Since movement-inducing features have to be used immediately, they have to be added to the relevant head after the phrase from which extraction is to proceed has been merged. But in the cases that interest us, once that phrase has been merged, the head is no longer active, hence the movement-inducing feature can no longer be added. In this way, Müller () explains the subject condition, the sentential subject constraint, and the adjunct condition, briefly reviewed above in section .. According to Müller (), ellipsis, like movement, proceeds on a phrase-by-phrase basis and is feature-driven. In order to elide a clause, it is not sufficient that the head of the clause bear the ellipsis feature. Instead, each and every phrase making up that clause must bear an ellipsis feature. Like movement-inducing features, ellipsis features are added to each head during the derivation. They make a head active. Since their effect is to elide the entire phrase, they are systematically the very last feature of a head that is satisfied. It follows from this that heads carrying the ellipsis feature never create the catch- for movement that heads without the ellipsis feature do. Therefore, the islands analyzed in terms of the catch- discussed above never arise within an ellipsis site. This, in outline, is Müller’s account of island amelioration under ellipsis. This section has discussed some prominent consequences of accepting Ross’s conjecture. On the assumption that there is no (contextually variable) syntactic structure at the ellipsis site, we are led to the conclusion that case is not a syntactic phenomenon and are likely to endorse a very powerful syntax–semantics mapping. On the completely contrary assumption that the ellipsis site is occupied by syntactic material identical to the antecedent, island amelioration becomes an important probe into the nature of island effects.

.. The island amelioration generalization The discussion in the preceding subsection has shown that Ross’s claim that islands are ameliorated by ellipsis, if true, has important theoretical consequences. It is therefore of paramount interest to ascertain whether the claim is true.





The theories discussed in the previous subsection, whether they do or do not posit variable syntactic structure at the ellipsis site, assume that Ross’s conjecture is true and derive it: Everything else being equal, all forms of ellipsis should give rise to island amelioration.11 This subsection briefly introduces a number of elliptical phenomena, including certain types of sluicing, that do not show island repair. These phenomena are prima facie counterexamples to both types of analysis discussed above. I outline briefly how some of the approaches to sluicing from the table in () not discussed so far propose to deal with the emerging phenomenon of variable island repair (Merchant b). Subsection .., finally, assesses how damaging these facts really are for the two classical approaches discussed in the previous subsection. The claim that not all types of ellipsis repair islands has a long history. Thus, Reinhart (), Depiante (), and Fukaya () contain examples like () that suggest that stripping is island-sensitive. () *The fact that some politician has resigned got much publicity but not the defense minister. (Reinhart : ) Similarly, Merchant (a), Barross, Elliott, and Thoms (), and Griffiths and Lipták () claim that contrastive fragments are island sensitive (though Culicover and Jackendoff ; Jacobson b deny this, and Weir  remains skeptical). A typical example is the following: ()

a. Does Abby speak the same Balkan language that Ben does? b. *No, Charlie. c. No, she speaks the same Balkan language that Charlie speaks.

VP ellipsis, too, shows restrictions on movement from the ellipsis site: ()

*They want to hire someone who speaks a Balkan language, but I don’tremember which (Balkan language) they do want to hire some one who speaks. (Fox and Lasnik : , ex. )

While there might be interfering factors in example () that would make it ungrammatical quite apart from the island violation (see Schuyler ; Merchant b), the relevant factors are controlled in the following example (Merchant b: , ex. ), which remains ungrammatical: ()

*Abby DOES want to hire someone who speaks {GREEK/a certain Balkan language}, but I don’t remember what kind of language she DOESN’T want to hire some one who speaks.

Thus both stripping and VP ellipsis are prima facie counterexamples to Ross’s conjecture.

11 Hornstein, Lasnik, and Uriagereka () and Müller () derive, more precisely, that, everything else being equal, all islands that are ameliorated by one form of ellipsis should be ameliorated by all forms of ellipsis.

  



Further evidence against Ross’s conjecture comes from sluicing itself. Above we based the discussion on merger-type sluices with a single -phrase as the remnant. While the island-insensitive behavior of such sluices constitutes a very stable cross-linguistic fact, sluices with contrastive remnants and sluices with multiple -phrases as remnants show locality sensitivity. We look at multiple sluicing first. The following is a typical English example of multiple sluicing: ()

I know that in each instance one of the girls got something from one of the boys. But which from which? (Bolinger )

In multiple sluicing, the two -phrases must originate in the same clause (see Bolinger ; Nishigauchi ; Richards ; Lasnik  for discussion). Certainly, the original traces of the two -phrases must not be separated by an island, (b), even though there is no (general) ban on multiple questions with -phrases from different clauses, as (a) shows. ()

{Each | Some} linguist will get offended if we award the prize to a particular philosopher, a. . . . but I can’t remember which linguist will get offended if we award the prize to which philosopher. b. * . . . but I can’t remember which linguist to which philosopher.

English is not alone in showing the restriction that the remnants in multiple sluicing must come from the same clause. Indeed, for a language to obey this restriction is the crosslinguistic norm, whether we are looking at languages that usually front one -phrase (see Lasnik  for English; Rodrigues, Nevins, and Vicente  for Brazilian Portuguese and Spanish; E. Callegari, p.c., for Italian; the same also holds for German), at languages that generally do not front their -phrases (Takahashi : –; Nishigauchi  for Japanese; Bhattacharya and Simpson  for Bangla with some qualifications; V. Dayaal, p.c., for Hindi), or at languages that generally front all of their -phrases (Marušič and Žaucer  for Slovenian; Adliene  for Lithuanian).12 Surprisingly, though the multiple remnants must originate in the same clause, that clause may be contained in an island. This is illustrated with the Slovenian example (). ()

Kupil je konja, ki je nekje nekoga brcnil, pa ne vem bought  horse which  somewhere someone kicked but not know kje koga. where who ‘He bought a horse that kicked someone somewhere, but I don’t know who where.’ (Marušič and Žaucer : , ex. a)

We return to this point after introducing the second type of island-sensitive sluicing. 12 Abe (: ch. ) discusses Japanese data and Lasnik (: ) discusses Serbo Croatian data where multiple sluicing violates the same clause condition. As far as is known, the -phrases must still originate in positions fairly local to each other in both cases.





According to Fukaya (), Merchant (b), Griffiths and Lipták (), and Winkler (), contrast sluicing is island-sensitive. A typical example is given in (). The example is ungrammatical on the island-violating construal, where the speaker cannot remember the identity of a third person on whose recommendation the speaker tried a third restaurant. On this reading, the structure would have to be the island-violating (a). For some people the sentence is grammatical under a non-island-violating construal, which can be paraphrased as in (b). The interpretations are clearly distinct, as (a) is about three or more restaurants with three or more distinct recommenders, while (b) is about a second recommender for the last restaurant mentioned. () * I have been sampling restaurants that various people recommend to me. Last week I tried three restaurants: one that Gill had recommended and one that Catherine had recommended, but I can’t remember right now who else. a. . . . I tried a restaurant that who else had recommended. b. . . . recommended it. We find a similar effect in the variant on (a) given in (). Some speakers report that this sentence is ungrammatical on any construal, others can get one of the readings paraphrased in (a). Crucially, the reading paraphrased in (b) is absent. This is the reading that we would expect if islands could freely be violated by ellipsis. The easiest way to distinguish the two readings is by paying attention to the fact that in (a) there is only one person being hired who speaks (or needs to speak) two languages, while in (b) there are two hires each of which speaks (or needs to speak) one language. The latter reading would result from island amelioration and it is absent. ()

Abby wants to hire someone who speaks Greek, but I don’t know what other language. a. . . . but I don’t know what other language he or she {should speak | speaks} b. . . . but I don’t know which other language is such that Abby wants to hire a second person who speaks that other language.

Examples of other types of islands are given in the literature cited. It seems safe to conclude that both multiple and contrast sluicing fail to conform to Ross’s conjecture. Multiple sluicing and contrast sluicing are therefore prima facie challenges to the two approaches discussed in the previous subsection. We will now see that approaches to sluicing that assume that there is contextually variable syntactic structure at the ellipsis site and combine this with a negative answer to one of the two remaining classifying questions from table () are potentially advantaged in accounting for these facts. Chung, Ladusaw, and McCloskey () give an account of sluicing that posits syntactic identity between antecedent and sluice but that does not require -movement of the remnant—at least in merger-type sluicing. The account is built essentially around the observation that merger-type sluices involve an indefinite correlate. Indefinites, unlike for example universal quantifiers, are known to be able to take very wide scope; in particular, the overt position of an indefinite and its scope position can cross an island boundary (Fodor and Sag , though see in particular Winter  for more nuanced discussion, and Szabolcsi  for a recent overview). Thus, (a) allows a reading where the indefinite

  



“a particular paper” is referentially independent and is thus seen to scope over the negative quantifier “nobody.” The indefinite is inside of a relative clause and thus inside of a complex NP but can still scope over the negative quantifier. One prominent approach to this fact is to model indefinite scope after binding relations, which are also not islandsensitive, (). Such an approach provides a possible implementation of Chung, Ladusaw, and McCloskey’s theory. ()

a. Nobody met the reviewer who rejected a particular paper. b. Nobodyk met the reviewer who rejected hisk paper.

Under such an approach wide-scope indefinites are bound by existential operators determining their scope. The indefinite correlate in merger-type sluicing is associated with just such an operator, potentially across an island. The -remnant in the sluice takes on the function of this operator in the sense that it binds the indefinite at the foot of the chain and is semantically restricted by it. Except for the operator at the root of the antecedent, the entire syntactic structure of the antecedent can be copied into the ellipsis site.13 The fact that the indefinite correlate exists and acts as a semantic restriction on the remnant is crucial. In contrast sluicing, binding of the correlate is either impossible (if it is definite) or the correlate cannot restrict the remnant semantically (if the correlate is a contrasting indefinite). Therefore instead of binding the correlate, a new branch has to be inserted in the structure. This new branch must be licensed through a chain rather than through binding. Island sensitivity arises from the assumption that the chain relation is islandsensitive while the binding relation is not.14 This accounts straightforwardly for the facts in () and (). While Chung, Ladusaw, and McCloskey’s account is thus very successful in predicting the island insensitivity of merger-type sluicing on the one hand and the island sensitivity of contrast sluicing on the other hand, it is unsuccessful at predicting the behavior of multiple sluicing. If one -phrase can bind into an island, why shouldn’t the same be true when several remnants need to enter into a binding relation, as in (b)? And if multiple binding could be shown to be island-sensitive, then why is multiple sluicing allowed into an island when all of the correlates are situated in the same clause, whether this clause is contained in an island or not?15 The final type of approach to be discussed in some detail is the island evasion approach. The island evasion approach is based on the idea that there is contextually variable structure at the ellipsis site, that the -phrase reaches its position through -movement, but that the 13 To account for case connectivity, Chung, Ladusaw, and McCloskey () assume that case is transmitted via the binding chain to the remnant, despite the fact that binding relations usually do not transmit case. 14 Chung, Ladusaw, and McCloskey () use the same type of account for the fact that sprouting is island-sensitive, but here an alternative explanation in terms of semantic/pragmatic incoherence of the antecedent–sluice pair is available. See Romero (). 15 Using Chung, Ladusaw, and McCloskey’s model, Nishigauchi () attempts to reduce the locality of Japanese multiple sluicing to the locality of so-called pair list interpretations of multiple questions. Such a reduction remains unsuccessful from a cross-linguistic perspective, because pair list readings have a wider distribution than multiple sluicing. This is shown by (a), which has a pair list interpretation, yet (b) is unacceptable.





recoverability condition on ellipsis is semantic rather than syntactic. The idea was pioneered in Baker and Brame () and developed in various ways in the subsequent literature (Merchant ; Fukaya , ; Rodrigues, Nevins, and Vicente ; van Craenenbroeck a; Barros b; Barros, Elliott, and Thoms ; Abels b). Under this approach, the ellipsis site does not have to contain a faithful syntactic copy of the antecedent but may contain a suitable paraphrase instead. Thus, example (a) is analyzed as follows, (): the ellipsis site cannot contain the structure of the overt question (a) because it violates the complex NP constraint. Instead it might contain (b) or (c); both have an interpretation suitably matching that of the antecedent.16,17 ()

Abby wants to hire someone who speaks a Balkan language, but I can’t remember which (Balkan language) a. * . . . Abby wants to hire someone who speaks. b. . . . {that | it} is. c. . . . they should speak.

Barros, Elliott, and Thoms () call the strategy employed here island evasion as opposed to island repair, because the inaudible structure at the ellipsis site evades an island violation through paraphrase instead of repairing it. The strategy works in simple cases because short paraphrases like (b) and (c) often have meanings that are sufficiently close to the meaning of the structures isomorphic to the antecedent, (a). The expectation of the theory is that the appearance of island repair should disappear in cases where there is no suitable paraphrase. Consider in this light example (a). The strategies used above in () to get around the island fail here. (a) violates the complex NP island and is therefore not a contender. (b) and (c) are grammatical, but cannot be construed as questions about the recommender for the third restaurant. () * I have been sampling restaurants that various people recommend to me. Last week I tried three restaurants: one that Gill had recommended and one that Catherine had recommended, but I can’t remember right now who else. a. * . . . I tried a restaurant that who else had recommended. b. # . . . recommended {it | one}. c. # . . . {it | that} was.

16

See Romero (), Merchant (), AnderBois (), and Barros (b) for discussion of how to define what is a suitably matching interpretation. 17 I should note some limitations of the discussion of island evasion approaches here. For space reasons I concentrate exclusively on sluicing. It is not logically necessary that all types of ellipsis allow the same freedom for paraphrases at the ellipsis site. Detailed discussion of each case would lead us too far afield. The same is true, potentially, for variation across languages: not all languages necessarily allow the same range of paraphrases at the ellipsis site. Temmerman () might be informative in this regard. It should also be noted that there are mixed approaches that combine a semantic identity condition with limited syntactic identity (Chung ). To keep the exposition manageable, I have to set such proposals aside (see Thoms  for discussion).

  



Clearly, the account is built on the assumption that the space of allowable paraphrases is constrained. If there were no constraints, the following pre-sluice would have to be considered: ()

Who else recommended one of the restaurants that I tried last week.

With this as a candidate pre-sluice, () should be acceptable. In other words, proponents of the island evasion approach will have to supply a suitably restrictive theory of which paraphrases are and which ones are not contenders as pre-sluices; there cannot be a true theory of island evasion without a solution to this problem, which we might call the toomany-paraphrases problem. The literature offers no worked-out, explicit solution to the too-many-paraphrases problem, but proponents of the island evasion approach appear to assume in practice that paraphrases are either copulative structures or simplifications of the antecedent. On these assumptions, () can be analyzed in a way quite similar to (). A pre-sluice identical to the full antecedent leads to an island violation and neither the copulative paraphrase (‘what other language it is’) nor the short paraphrase (‘what other language they (should) speak’) supports an interpretation corresponding to the intended reading. Island amelioration is thus viewed as an illusion under island evasion approaches. Consequently, proponents of such approaches are at pains to show that when no suitable paraphrase is available, the illusion of island amelioration breaks down. Recall from subsection .. that sluicing appears to repair violations of Ross’s LeftBranch Condition, () repeated as (a). Island repair approaches must posit (b) as the pre-sluice because of parallelism. Island evasion approaches cannot posit (b) as the presluice, because it violates the Left-Branch Condition. Instead, island evasion approaches assume (c) as the pre-sluice. ()

a. He emptied a big mug, but he wouldn’t tell me how big. b. * . . . how big he emptied [ how big a mug]. c. . . . how big it was.

Under evasion approaches, the sluice is thus really built on a predicative adjectival structure rather than the attributive adjectival structure in the antecedent. Prima facie evidence for this position comes from languages like German and Dutch, where there is a morphological distinction between attributive and predicative adjectives. Example (a) illustrates that German attributive adjectives are always suffixed with a schwa and a further (though sometimes null) affix indicating number, gender, and case. Example (b) shows that by contrast predicative adjectives are bare: they have neither the schwa suffix nor the agreement suffix. () a. Er hat ein-e-n {großen |*groß} he has a-ə-.. big-ə-.. big ‘He has emptied a big mug.’

Becher mug

ausgeleert. emptied

b. Der Becher ist {groß |*groß-e-Ø |*groß-e-n |*groß-e-m}. the mug is big big-ə-.. big-ə-.. big-ə-.. ‘The mug is big.’





In the German translation of example () the bare predicative form of the adjective is used, (a), (see Merchant , who also discusses parallel facts from Dutch), which strongly suggests that the pre-sluice can be the equivalent of (c), as expected under the evasion approach. The attributive form of the adjective cannot be used, (b), which strongly suggests that the pre-sluice cannot be the pre-sluice expected under island repair approaches. That is, it cannot be the equivalent of (b). We are also not dealing with N-ellipsis, since, although N-ellipsis is independently available, it requires the presence of the article and the attributive form of the adjective, (c).18 The island evasion approach offers an elegant explanation both of the acceptability of apparent left-branch extraction under sluicing and of the morphological observation. () a.

. . . aber ich weiß nicht wie but I know not how ‘ . . . but I don’t know how big.’

b. * . . . aber ich weiß but I know |groß-e-m}. big-ə-.. c.

nicht not

wie how

groß. big. {groß-e-Ø big-ə-..

|groß-e-n big-ə-..

. . . aber ich weiß nicht *(ein-e-n) wie großen (Becher). but I know not a-ə-.. how big-ə-.. mug

We have seen above that the clausemateness condition on multiple sluicing poses a challenge for syntactic identity accounts and for accounts with no structure at the ellipsis site. Though there is currently no worked-out proposal about the syntax of multiple sluicing that could claim cross-linguistic validity (though Takahashi ;19 Rodrigues, Nevins, and Vicente ; Marušič and Žaucer ; Lasnik  provide languageparticular solutions), the outline of such an account emerges quite clearly from these papers, and it involves island evasion. The idea in the papers cited is that the additional -phrase moves from the ellipsis site. The task then is to identify and characterize precisely the movement operating that lifts the additional -phrase from the ellipsis site and to attribute the clausemate condition to independently known constraints on this movement operation (see Abels and Dayal ). While this body of work makes it clear in principle what logical structure an account of the clausemateness condition of multiple sluicing will likely have, the logic of island evasion is invoked to deal with the island insensitivity of multiple sluicing. Under the island evasion approach, the pre-sluice for an example like (), repeated here as (), can be assumed to be the equivalent of ‘who it kicked where’. This structure evades the complex NP island in the antecedent rather than repairing it.

18

For Italian the conclusion that apparent violations of the left-branch condition under sluicing are due to predicative pre-sluices can be based on the relation between possible adjective position in the NP and their ability to appear in predicative structures (Montali ). For English, the conclusion that apparent left-branch extractions are based on copulative pre-sluices can be supported by considering adjectives that are both gradable and restricted to attributive position (on the relevant reading) (Thoms, Barros, and Elliott ). 19 See also Park () and Park and Kang ().

  



In the general case, violations of the clausemateness condition cannot be evaded by paraphrase. The island evasion approach is thus well positioned to develop an explanation for the otherwise paradoxical locality insensitivity of multiple sluicing combined with the clausemateness condition. () Kupil je konja, ki je nekje nekoga brcnil, pa ne vem bought  horse which  somewhere someone kicked but not know kje koga. where who ‘He bought a horse that kicked someone somewhere, but I don’t know who where.’ (Marušič and Žaucer : , ex. a) It is difficult to see how either of the theories discussed in the previous subsection, .., can be reconciled with the fact that multiple sluicing is island-insensitive while at the same time being subject to the clausemateness condition. If there is no syntactic structure at the ellipsis site, all sensible multiple -questions should in principle correspond to multiple sluices. This will explain island insensitivity but stumbles on the clausemateness condition, because there are many grammatical and sensible multiple questions where the -phrases are not clausemates. Crucially, these cannot generally be turned into multiple sluices, as () illustrated. Essentially the same problem arises for syntactic identity approaches. These approaches are built around the assumption that ellipsis repairs island violations. This assumption can be invoked to explain the island insensitivity of multiple sluicing. However, the assumption that syntactic locality constraints are repaired by ellipsis makes it difficult to see why the additional -phrase would have to originate in the same clause as the primary phrase. Whatever syntactic locality constraint might underpin this fact under a syntactic identity account, the violation should be repaired by ellipsis (see Marušič and Žaucer  on this point). While there is then substantial suggestive evidence for an island evasion approach, the case cannot be considered closed. There are two main reasons for this. First of all, there is no worked-out, explicit, and predictive solution to the too-many-paraphrases problem. Second, the island evasion approach does not by itself offer an approach to case connectivity. This is pointed out in Lasnik’s () discussion of Merchant (). Lasnik notes that case connectivity between correlate and remnant is not predicted by the evasion approach, again, because of the too-many-paraphrases issue. The fact that syntactic identity accounts do explain case connectivity gives them an explanatory advantage over island evasion approaches.20 Researchers have since tried to solve the case connectivity and related problems in a number of different ways (see van Craenenbroeck a; Chung ; Abels b), but a consensus view has not yet emerged. In this subsection we have briefly looked at elliptical phenomena that fail to conform to Ross’s conjecture. The failure of island amelioration provides prima facie evidence against

Some of Lasnik’s other arguments rely on scope and binding properties of complex -remnants. As discussed in Barros, Elliott, and Thoms (), these arguments are tenuous. 20





theories that posit no contextually variable syntactic structure at the ellipsis site and also against theories that impose strict syntactic parallelism between antecedent and ellipsis site and assume a grammatical mechanism of island repair. The facts favor theories that assume contextually variable syntactic structure at the ellipsis site but either do not impose strict syntactic parallelism between antecedent and ellipsis site (island evasion) or allow the island ameliorating examples to be derived without movement.21 While many aspects of these proposals remain to be worked out, the examples where Ross’s conjecture breaks down suggest that the ellipsis site contains (contextually variable) syntactic structure and that island violations cannot be repaired by ellipsis. It would follow from this that island amelioration under ellipsis is an illusion and should be explained neither by the absence of syntactic structure at the ellipsis site nor as a consequence of the nature of movement (as in Hornstein, Lasnik, and Uriagereka  and Müller ). On the contrary, the cases where Ross’s conjecture fails have suggested to proponents of the island evasion approach that locality constraints are robustly attested within ellipsis sites. The facts discussed in this and the previous subsection also interact with the debate on whether island constraints are syntactic or arise from parsing considerations. In conjunction with the assumption of strict syntactic parallelism between antecedent and ellipsis site, Ross’s generalization might be construed as support for the idea that island constraints are not a grammatical phenomenon at all but instead arise from constraints on online processing. One could then assume that ellipsis resolution involves copying the potentially island-containing parse tree for the antecedent into the ellipsis site. Copying would sidestep the need to build a new parse tree online, thus avoiding the processing difficulty. It is hard to see, however, how such an account could be extended to the cases where ellipsis fails to ameliorate islands, especially since the relevant structures are often semantically perfectly well-formed, as () shows.

.. Ceteris paribus In the previous section, we have taken Ross’s conjecture to say that everything else being equal, all forms of ellipsis should give rise to island amelioration. We have further seen that stripping, contrastive fragments, VP ellipsis, multiple sluicing, and contrast sluicing are prima facie counterexamples to the generalization. Theoreticians defending a syntactic parallelism account have routinely invoked the idea that these counterexamples are only superficial and that not everything is equal in these cases (Fox and Lasnik ; Merchant a; Temmerman ; Griffiths and Lipták ; Abe ). These authors suggest that the failure of Ross’s conjecture should not be explained as an island effect at the ellipsis site but in one of the three following ways instead: (i) the ungrammaticality arises (or is visible) along the movement path outside of the ellipsis site, (ii) ungrammaticality

21

In the text we discussed Chung, Ladusaw, and McCloskey () but we should also mention Abe () as a member of this family of approaches.

  



arises in the antecedent, or (iii) it is an effect of syntactic mismatch between antecedent and ellipsis site. To understand what is intended by case (i), consider the contrast in island amelioration between sluicing, (a), and VP-ellipsis, (b), again. The crossed-out material represents material at the ellipsis site under a theory demanding syntactic identity. The overt material distinguishing sluicing from VP-ellipsis is italicized in (b). Notice that the -phrase ‘what kind of language’ crosses the italicized material and the structure associated with it. In particular, ‘what kind of language’ crosses nodes  and , which form part of the -movement path. ()

a. Abby wants to hire someone who speaks a Balkan language, but I don’t know which Balkan language [0 Abby [0 T0 [wants to hire someone who speaks]]]. b. *Abby wants to hire someone who speaks a Balkan language, but I don’t know what kind of language [ she [ doesn’t [want to hire someone who speaks] ]].

If the island violation arises or is made visible on node  or node  (or both), then the fact that these nodes are outside of the ellipsis site in (b) but the corresponding nodes ’ and ’ in (a) are inside of the ellipsis site can be used to explain why (b) is ungrammatical while (a) is grammatical. Fox and Lasnik () use this idea in their explanation of the difference between VP ellipsis and sluicing. Merchant (a) and Temmerman () use the same logic, coupled with a version of Chomsky’s #-mark, to make the island violation visible outside of the ellipsis site in fragments. Recall that Chomsky’s suggestion of using a diacritic feature representing the fact that islandviolating movement has occurred as an explicit part of the syntactic representation stems from a time when syntactic representations did not carry their derivational history on their sleeves (in the forms of copies of movement, traces of movement, movement arrows, a path of feature percolation, . . . ). However, modern theories of syntax all make these representational devices available. As a consequence, it is always possible to determine by simple inspection of the output structure whether an island violation has taken place. The use of the diacritic features in modern theories therefore introduces a representational redundancy, which should be avoided if at all possible and which needs to be carefully motivated if it is to be assumed (see n. ). Given these considerations, it is unclear whether the #-diacritic can carry any truly explanatory weight. To understand case (iii) it is useful to consider the following contrast. Example (a) is well-formed while (b) is deviant. The difference clearly has to do with the different antecedents, in particular the difference between ‘a certain’ and ‘any’. Indeed, ‘a certain colleague’ in (a) takes scope above negation (‘There is a certain colleague that Bob didn’t consult’) but ‘any colleague’ in (b) takes scope below negation (‘It is not the case that there is a colleague that Bob consulted’). ()

a. Bob didn’t consult a certain colleague, but I won’t tell you who. b. *Bob didn’t consult any colleague, but I won’t tell you who.

The difference can be represented syntactically with an existential operator taking scope below negation for (b) and taking clausal scope for (a):





()

Schematic structure for (47a)

∃k

α

Bob ¬ consult

colleaguek

and for (47b)

Bob ¬

∃k consult

colleaguek

If syntactic identity is sensitive to operators in the antecedent, then the contrast above in () may find a syntactic explanation. The -operator in the sluice takes clausal scope and c-commands negation. It is outside of the ellipsis site and there is no (other) operator binding the object of ‘consult’ within the ellipsis site. Under a syntactic identity account, we can take node a in (a) as the antecedent for (a). However, there is no clausal node excluding the existential operator in (b). The antecedent provided by the structure with the low existential operator cannot be used to resolve ellipsis under a syntactic identity account, since in this structure the object of ‘consult’ in the sluice would be bound twice: once by the existential below negation and a second time, vacuously, by the -operator. It is reasonable to assume that this structure can either not be derived or not be interpreted or both. The structure for the ellipsis site that can be derived and interpreted has no existential operator binding the object of ‘consult’ below negation. It is therefore not syntactically identical to the antecedent provided in (b). Therefore, the sluice based on (b) cannot meet the syntactic identity requirement and is correctly ruled out. This illustrates case (iii) above: the failure of syntactic parallelism.22 22

The discussion here is intended to illustrate the logic of case (iii). It is not intended as an endorsement of the view that wide-scope indefinites have the syntactic representation in (a). The issue will be taken up briefly in connection with Griffiths and Lipták () just below. Given the difference in interpretation between the two structures in (), the facts in () can be approached in semantic terms under a semantic identity account.

  



Notice that the diagrams in () associated the existential operator with the indefinite noun phrase through coindexing rather than through movement. This was intended to reflect the two claims we already encountered in our discussion of Chung, Ladusaw, and McCloskey () above (see discussion around ()), namely, that indefinites can take island-insensitive scope and that binding relations are not island-sensitive. Other operators such as -phrases, universal quantifiers, and comparative operators are often taken to be scoped by island-sensitive movement operations—though these movements can be covert and not have a reflex in word order. Case (ii) above, where the ungrammaticality of an island-violating sluice is attributed to a violation in the antecedent, builds on the idea that certain types of scope are represented as movement. Specifically, consider the implication for the sluicing remnant if its correlate is an operator that is scoped by an island-sensitive movement operation. If the sluice itself obeys all island restrictions, then there is a structure for the antecedent with the corresponding syntactic scope for the correlate; both sluice and antecedent are well-formed and they match: no problem arises. However, if the remnant relates to a trace inside of an island in the ellipsis site, then the situation is quite different. By assumption, the antecedent is only grammatical if the correlate does not violate the island, but the syntactic identity constraint mandates just that. Thus, if the remnant relates to a trace inside of an island and the correlate is scoped in an island-sensitive way, then the two demands of island sensitivity and syntactic identity cannot be met simultaneously. Syntactic identity can only be achieved if the antecedent is ungrammatical. This is case (ii) above.23 Griffiths and Lipták () and Abe () account for the difference in island sensitivity between merger type and contrast sluicing exploiting this logic. Griffiths and Lipták () argue that the difference between the island-insensitive merger type of sluicing and the island-sensitive contrast sluicing is part of the following larger pattern: Elliptical constructions in which the remnant contrasts with the correlate are generally island-sensitive and constructions where the remnant specifies or queries the identity of the correlate are generally island-insensitive. They argue convincingly that treating this generalization as a violation of type (i) is neither motivated nor insightful. To account for their generalization, Griffiths and Lipták assume that the indefinite correlates in merger-type sluicing and other non-contrastive structures are syntactically scoped by an island-insensitive mechanism (an idea we have already encountered in the discussion of Chung, Ladusaw, and McCloskey  in ..), while the correlates in contrast sluicing are scoped by the island-sensitive focus movement operation. Under this view, merger-type sluices are island-insensitive because the scoping mechanism in the antecedent is islandinsensitive and because ellipsis repairs island violations. Contrast sluicing is island-sensitive because the correlate is scoped by an island-sensitive mechanism which means that either parallelism is violated (violation of type iii) or there is an unrepaired island violation in the antecedent (violation of type ii). Griffiths and Lipták’s attempt to uphold Ross’s conjecture is theoretically important. The analysis extends the scope of a syntactic identity account of sluicing to a number of cases where Ross’s conjecture fails and that we treated as prima facie counterarguments to a syntactic identity account. 23

Fox and Lasnik’s account of the locality sensitivity of VP-ellipsis builds on a similar catch-; although the details are quite different: either parallelism is violated (violation of type iii) or movement outside of the ellipsis site violates an island constraint (violation of type i).





Despite its ingenuity, Griffiths and Lipták’s account is open to a number of criticisms which, on balance, appear to disfavor the syntactic identity approach. First, the account needs to make a distinction between (island-insensitive) merger-type sluicing and (islandsensitive) contrast sluicing. This distinction is related to the nature of the correlates: indefinites in merger-type sluicing and focused DPs in contrastive sluices. Indefinites are analyzed as having syntactically represented island-insensitive scope while foci are analyzed as being scoped syntactically with an island-sensitive mechanism. This analytical distinction between indefinites and foci seems to be driven by theoretical expediency rather than by empirical necessity. Empirically, indefinites and foci behave in rather similar ways. As is well known, indefinites inside of islands can be referentially independent of c-commanding operators outside of islands (Fodor and Sag ). This behavior can— but by no means needs to (see Schwarzschild )—be modeled in terms of islandinsensitive syntactic scope of an existential operator. Equally important, though, is another aspect of the scopal behavior of indefinites: their distributive scope. The distributive scope of indefinites is island-sensitive (see Ruys ; Winter ) and tracks the locality of quantifier raising in many ways. The distributive scope of indefinites is a strong candidate for modeling in terms of a syntactically scoped operator. Foci show similarly mixed behavior. When we measure their scope in terms of the ability of a focus inside of an island to associate with a focus-sensitive operator, then foci come out as island-insensitive (Anderson ; Jackendoff ). On other measures, such as overt focus movement (Griffiths and Lipták ) and the ability of several foci from the same domain to associate with separate operators (Krifka ), foci come out as island-sensitive. Thus both foci and indefinites show a complex mix of island-sensitive and island-insensitive behavior. These empirical considerations show that Griffiths and Lipták’s choice to represent syntactically the island-insensitive aspects of the behavior of indefinites and the island-sensitive aspects of the behavior of foci is arbitrary at best. The similarities between indefinites and foci suggest a theoretical assimilation of the two phenomena rather than the dissimilation necessary for Griffiths and Lipták’s account. Indeed, Winter () and Krifka () explain the island-sensitive aspects of the behavior of foci and indefinites, respectively, by reducing these aspects to standard movement operations; and Schwarzschild () and Krifka () explain the islandinsensitive aspects of the behavior of foci and indefinites, respectively, by not representing them in the syntax at all. In light of this, Griffiths and Lipták’s analysis is not only arbitrary, but would represent a step backwards in our understanding of the similar behavior of indefinites and foci. A second issue for Griffiths and Lipták’s () analysis is raised by Abels and Thoms () and Barros, Elliott, and Thoms (), who report that contrast sluicing loses its island sensitivity in certain cases in some languages. Concretely, Abels and Thoms and Barros, Elliott, and Thoms show that contrast sluicing loses its island sensitivity in languages that allow contrastive -phrases to bind resumptive pronouns inside of islands. They claim further that the acceptability of island amelioration in contrast sluicing correlates neatly with the island repair potential of a resumptive pronoun in a given language. All of this would follow, if the status of island-insensitive contrast sluicing does not depend on the properties of focus movement in the antecedent (as in Griffiths and Lipták’s approach) but depends instead on the availability of a grammatical pre-sluice (see works cited for details). The pattern strongly suggests (a) that there is structure at the

  



ellipsis site, (b) that the status of island-insensitive contrast sluicing is not tied to properties of the antecedent, (c) that extraction at the ellipsis site is island-sensitive, and (d) that the grammaticality of island-insensitive contrast sluicing depends on language-specific properties of the available pre-sluices, as expected under the island evasion approach. This argument turns Griffiths and Lipták’s defense of Ross’s conjecture into an argument against it. It weakens the syntactic identity approach but at the same time fails, in the absence of a solution to the too-many-paraphrases problem, to lend crucial support for the island evasion approach. Subsection .. briefly introduced two kinds of sluicing that violate Ross’s conjecture: contrast sluicing and multiple sluicing. The existence of and constraints on these structures were taken as prima facie evidence against syntactic identity accounts with -movement and against accounts without (contextually variable) syntactic structure at the ellipsis site. We have taken these examples to provide prima facie evidence for island evasion approaches or for approaches with syntactic structure but without -movement. Given that to date there is no developed theory of multiple sluicing under any of the existing approaches to ellipsis, we could not pursue the issue of multiple sluicing further in this subsection (though see n. ). This current subsection has briefly looked at the possibility of neutralizing contrast sluicing as a counterexample to Ross’s conjecture by invoking the implicit ceteris paribus clause. We first identified three general ceteris paribus strategies that have been used in the literature: (i) violations arise in the structure above the ellipsis site; (ii) violations arise in the antecedent; (iii) violations arise as a result of mismatch between antecedent and structure at the ellipsis site. We then saw that the most convincing rebuttal of the argument from contrast sluicing, namely Griffiths and Lipták (), relies on a combination of (ii) and (iii): the antecedent is either ill-formed (ii), being derived via islandviolating focus movement, or it mismatches the antecedent (iii). We also saw that Griffiths and Lipták () suffers both from conceptual and empirical weaknesses. This leaves largely unchanged the strength of the claim that syntactic identity accounts cannot give a proper analysis of contrast sluicing. The argument from contrast sluicing is a prima facie argument against syntactic identity accounts. It cannot be turned into a compelling argument for island evasion approaches until the too-many-paraphrases problem has been solved. We can also evaluate to what extent the discussion from the present subsection impacts our evaluation of the prospect of a theory without (contextually variable) syntactic structure at the ellipsis site. Notice that Griffiths and Lipták’s account of the island sensitivity of contrast sluicing relies crucially on the presence of syntactic structure at the ellipsis site: Either antecedent and ellipsis site contain syntactically different material, in which case the syntactic identity condition is violated. Or antecedent and ellipsis site contain syntactically identical structure, and then the antecedent violates locality conditions. This logic cannot be reproduced in an account without (variable) syntactic structure at the ellipsis site. Nor can a directly semantic account rely on the notion that the elided material is semantically deviant as an explanation, since the intended interpretations of contrast sluicing are perfectly coherent. And nor can theories without syntactic structure at the ellipsis site avail themselves of Abels and Thoms () and Barros, Elliott, and Thoms’ () rebuttal of Griffiths and Lipták () since this rebuttal, too, relies crucially on the nature of the syntactic representation at the ellipsis site. The counterexamples to Ross’s conjecture from subsection .. remain as problematic as they were for accounts without (variable) syntactic structure at the ellipsis site.





.. Conclusion: Sluicing and islands In this section we have surveyed facts concerning and theoretical reactions to Ross’s conjecture, that is, the claim that ellipsis quite generally ameliorates island violations. In its simple, general form the generalization has elicited two types of response (subsection ..): (i) there is no (contextually variable) syntactic structure at the ellipsis site and (ii) there is a copy of the antecedent at the ellipsis site but island effects are masked or never incurred at the ellipsis site. Position (i) has important consequences for the theory of syntactic relations like case and for the syntax–semantic interface. Position (ii) has been taken to be revealing of the nature and derivational provenance of island constraints. The fact that Ross’s generalization has systematic counterexamples (subsection ..) poses problems for both of the above positions and lends prima facie support to theories that assume syntactic structure at the ellipsis site but do not assume that island violations are literally repaired by ellipsis. The latter type of theories undermine the conclusions drawn from the more traditional first two positions. Subsection .. explored the severity of the problems caused by the counterexamples to Ross’s conjecture for the traditional two positions on ellipsis. While non-syntactic accounts of ellipsis have nothing to say about these cases, syntactic identity accounts have been defended by invoking the ceteris paribus clause. We have seen that these moves run into some difficulties, both conceptual and empirical. However, the issue remains open since proponents of the island evasion approach, which has the strongest prima facie support, have not conclusively answered the two main questions arising under the approach: how to account for case matching and how to solve the too-many-paraphrases problem. As a consequence, none of the theoretical arguments from ellipsis for the nature of case, the flexibility of the syntax–semantics mapping, or pertaining to the nature of island effects holds water at the moment.

. P 

.................................................................................................................................. Section . discussed a class of cases where movement is disallowed in non-elliptical structures and allowed in intuitively corresponding elliptical ones. This section turns to cases where movement is allowed in non-elliptical structures and selectively disallowed in elliptical structures. Such effects have been discussed particularly in relation to VP ellipsis and related constructions (Aelbrecht ; Abels ; Baltin ; Bošković ). A typical example of this type of interaction between movement type and ellipsis comes from Aelbrecht’s () discussion of modal complement ellipsis in Dutch. Modal complement ellipsis is a type of ellipsis of the main verb, its dependents, and modifiers in the specific context where these appear as the complement of a modal verb. Modal complement ellipsis is illustrated in () (from Aelbrecht : –): () a. Emiel wou Sarah wel een cadeautje geven, maar hij mocht niet. Emiel wanted Sarah  a present give but he may. not ‘Emiel wanted to give Sarah a present, but he wasn’t allowed to.’

   b. Die broek MOET nog niet gewassen worden, maar those pants must still not washed become but al wel. already  ‘Those pants don’t have to be washed yet, but they can be.’

hij he

 MAG may

Example (a) features the active VP Sarah een cadeautje geven ‘give Sarah a present’ and (b) the passive VP gewassen worden ‘be washed’. Sentences in the passive voice, under many analyses, involve movement of the thematic object of the verb from object to subject position. Under those analyses, die broek ‘those pants’ in (b) originates within the elided VP and moves out of it. Similarly, -question formation is usually taken to involve movement. But such movement cannot originate within the ellipsis site of modal complement ellipsis, (a), while such movement is perfectly licit in non-elliptical contexts, (b).24 () a. ?*Ik weet niet wie Kaat WOU uitnodigen, maar ik weet wel wie I know not who Kaat wanted invite but I know  who ze MOEST. she must. Intended: ‘I don’t know who Kaat WANTED to invite, but I do know who she HAD to.’ b.

Ik weet niet wie Kaat WOU uitnodigen, maar ik weet wel wie I know not who Kaat wanted invite but I know  who ze MOEST uitnodigen. she must invite ‘I don’t know who Kaat WANTED to invite, but I do know who she HAD to invite.’

The accounts for such asymmetries in Aelbrecht (), Abels (), Baltin (), and Bošković () differ in their details but they share a number of common, crucial assumptions about the nature of syntax, ellipsis, and—importantly—movement. The first assumption common to the above accounts is that syntax is a derivational, bottom-up system, i.e., phrase markers are built along essentially minimalist lines (Chomsky a; Epstein ) by merging lexical items bottom-up. Second, ellipsis is analyzed as involving regular syntactic structure at the ellipsis site. This structure is phonologically deleted and marked as syntactically inaccessible as a consequence of ellipsis. Third, movement from the ellipsis site requires access to the syntactic representation at the ellipsis site. It follows that when movement precedes ellipsis of the domain from which an item is to be moved, movement is possible, but when movement follows ellipsis of the domain from which an item is to be moved, movement is impossible, because the domain has been elided and is therefore syntactically inaccessible by the time movement happens. In other words,

24

Note that the status of (a) is not attributable to a general incompatibility between modal complement ellipsis and -question formation, since -questions allow modal complement ellipsis just as long as the -movement does not originate inside of the elided VP (Aelbrecht : –).





whether an item can or cannot be extracted from an ellipsis site depends on whether extraction happens before or after ellipsis. This is schematized in (): ()

a. Licit extraction from ellipsis site: (i) [ X [ZP Y Z ]] (ii) [ Y [ X [ZP tY Z ]]] (iii) [ Y [ X [ZP tY Z ]]]

movement of Y ellipsis of ZP

b. Illicit extraction from ellipsis site: (i) [ X [ZP Y Z ]] (ii) [ X [ZP Y Z ]] (iii) *[ Y [X[ZP tY Z ]]]

ellipsis of ZP movement of Y blocked

The theory in Bošković () for example would deal with the facts in () and (a) as follows: movement to the subject position shown in () targets the specifier of TP while movement illustrated in (a) targets the structurally higher specifier of CP. The syntactic head triggering ellipsis is situated between those two positions (and identified with C0 in Bošković ). Because of the bottom-up nature of syntax, movement of the subject to the specifier position of TP happens before ellipsis is triggered and is therefore licit, (a). On the other hand, movement to the specifier of CP happens only after ellipsis is triggered (again because of the bottom-up nature of the syntactic computation) and is therefore illicit, (b).25 The models in Aelbrecht (), Abels (), Baltin (), and Bošković () all crucially rely on an implementation of movement in the leapfrogging way illustrated and discussed in connection with structure (). Structure () claimed that -movement stops over at the edge of VP and at the edge of CP but not at the edge of TP. This is crucial. It is only because movement to the specifier of TP escapes the ellipsis domain before ellipsis can be triggered by the complementizer and because -movement to the specifier of CP happens after ellipsis is triggered that the distinction between the two movement types can be derived. If movement stopped over at the edge of every phrase along the path (as in Müller , discussed in ..) both types of movement would target a position at the edge of TP and thus happen before ellipsis is triggered. The same reasoning applies to theories where movement is represented in terms of percolation of features, (). Under percolation approaches, there is no obvious way of deriving the fact that movement to the specifier of TP is allowed from the ellipsis site, but movement to the specifier of CP is not. Extraction asymmetries from ellipsis sites can potentially provide an interesting argument about the punctuated or continuous nature of movement paths. Seductive though this argument may be, the case is far from closed. Thus, the simplified account of the Dutch modal complement ellipsis facts presented above predicts that any movement to a position outside of VP but below TP should be compatible with modal complement ellipsis while movements to a position above TP should not. As discussed at length in Aelbrecht (), the first of these predictions is wrong: scrambling of objects and adverbials out of the VP in Dutch is not licit under modal complement ellipsis. Worse still, even the second prediction is inaccurate, since modal complement ellipsis is possible in relatives with an abstract relative 25

For details of implementation, the interested reader is referred to the primary literature mentioned in this section.

  



operator in contexts of so-called antecedent-contained deletion, (a), it is possible in (certain) free relatives (see Abels  for discussion), (b), and in comparative clauses, (c). () a. Olaf heeft elk boek gelezen Olaf has every book read ‘Olaf read every book he had to.’

dat that

hij he

b. Olaf zal uitnodigen wie hij mag. Olaf will invite who he is.allowed ‘Olaf will invite who he’s allowed to.’ c. Hij moet meer (boeken) lezen dan he must more books read than ‘He has to read more books than he can.’

hij he

moest. must. (Aelbrecht : , ex. b)

(Aelbrecht : , ex. c) kan. can (Abels : , ex. )

The difference between those movement types that do and those that do not allow extraction from the domain of modal complement ellipsis does not follow from any of the analyses mentioned above (see Aelbrecht ; Abels  for discussion). Unfortunately, the other empirical cases are equally equivocal. Thus, Baltin () claims that British English do is an ellipsis phenomenon whose behavior follows neatly from the logic  outlined in (). He claims that A-movement is and A-movement is not allowed from the ellipsis site. However, Abels (: –) argues, largely on the basis of published examples, that the generalization is not true: A-movement does and -movement does not allow  extraction from the ellipsis site in British English do. Some of the remaining A-movements pattern with A-movement, some with -movement. This leaves the overall picture muddled and the theory in need of re-evaluation. Other potential cases of ellipsis/movement-type interactions that could follow from the theory of punctuated paths involve verb phrase pronominalization in Scandinavian (Houser, Mikkelsen, and Toosarvandani 2007; Bentzen, Merchant, and Svenonius 2013), modal complement ellipsis in German, French, Italian, and Spanish (see Napoli 1985; Aelbrecht 2010; Dagnac 2010) and the high type of VP ellipsis in English discussed in Bošković (). In these cases, the database is not rich enough to reach firm conclusions about the nature of the effects. Although the theoretical question of the punctuated versus the continuous nature of movement paths is of great importance to syntactic theory, the empirical picture remains too fragmented for any firm conclusions at the moment. Moreover, the analyses presented in the literature, although exploiting the same logic, are parochial and contradict each other. It is to be hoped that ellipsis/movement-type interactions will be studied in more detail for more languages, at which point they might bear on such important questions as the derivational nature of syntax and the punctuated versus continuous nature of movement paths.

. C

.................................................................................................................................. This chapter has surveyed two types of interaction between movement and ellipsis. The first case study was devoted to cases where movement from an elliptical domain is less restricted than might be expected on the basis of comparable examples with fully





pronounced structures. Such facts go by the name of island amelioration and have variously been taken as evidence for the phonological nature of island constraints, the derivational nature of syntax, the non-syntactic nature of case, and the flexibility of the syntax/ semantics interface. It was shown that the most immediate empirical consequence of the two canonical theories of ellipsis, that is of theories that posit no (variable) syntactic structure at the ellipsis site and theories that demand syntactic identity between antecedent and ellipsis site, radically different though they are, is Ross’s conjecture. Ross’s conjecture says that, ceteris paribus, all cases of ellipsis give rise to island amelioration. On the surface, the conjecture is false. As a result, the various conclusions based on canonical theories of ellipsis are weakened. The most promising competing approach, the approach based on island evasion, cannot be considered a successful theory until a principled account of case connectivity is presented and the too-many-paraphrases problem is solved. The second case study involved structures where movement from an ellipsis site is less liberal than movement from the corresponding pronounced structure. Some researchers have taken such structures to argue for the derivational nature of syntax and the hypothesis that movement paths are punctuated. At the moment, the facts do not line up neatly with the predictions of the various theories that try to make this argument and the theories, though building on the same logic, are mutually incompatible. For the time being, conclusions about the nature of syntactic computation and the representation of movement based on these facts must therefore be considered as highly tentative and preliminary.

  ......................................................................................................................

   ......................................................................................................................

 ,  ,   . 

. T    

.................................................................................................................................. T chapter is about verb phrase ellipsis (VPE) in speakers whose language faculty is incomplete—patients with a language deficit who have lost part of their linguistic ability subsequent to (mostly focal) brain damage, and developing children who have not yet reached the steady state of adult language. We show that despite certain syntactic limitations, the ability of both patients with Broca’s aphasia and children to reconstruct an elided VP is intact. While the present data set is of a relatively limited size, this result is nonetheless remarkable, we argue, and is important for our understanding of the nature of language loss subsequent to focal brain damage, as well as language acquisition. We briefly review some properties of VPE, move on to their psycholinguistic relevance, and then describe two pilot experiments—in adult aphasic patients and in children—that tested the comprehension of VPE. We establish that VPE (a) requires a comprehender to go beyond the information given on the surface; and (b) requires complex syntactic and semantic analysis. As such, it provides an excellent testing ground for claims regarding language-incomplete individuals. With that in mind, we consider the comprehension of complex linguistic material in patients with aphasia, and in developing children. We report data that indicate that adult patients with Broca’s aphasia and young children are quite capable with VPE, despite deficiencies elsewhere in syntax. We then consider these data in a broader context, and conclude by reflecting on the potential significance of the results of these experiments.

. VPE

.................................................................................................................................. The context of the present volume permits us to be relatively brief in discussing structural issues regarding VPE, and we restrict ourselves to just those aspects of the construction that are relevant to our endeavor. Our desire for brevity and focus leads us to be somewhat



, ,  

lenient in terms of precision and generality. The ingredients of the analysis of VPE are many, as are the associated questions and debates: What kind of operations are involved in the recovery of elided material? What is the type, size, and constituency of the elided piece? And what constrains the relation between the ellipsis clause E and the overt antecedent clause A? We concentrate on the last question, through an experimental exploration with language-impaired adults and developing children. In our experimental investigation, we started off by studying relatively simple constructions, that is, coordinate VPE, in which the VP of the antecedent clause A must be interpreted in the ellipsis clause E (a) (Sag a; Williams b; Rooth b; Hardt ; Fox ; Johnson b; Merchant ; see van Craenenbroeck  for a recent survey). Once the VP constituent in A (VPA) cannot be identified as such in E, the sentence becomes unacceptable. The piece sold a car in (b) and told in (c) are not constituents, and the resulting sentence is marked: ()

a. [A The detective [VP found a child ]] and [E the cop did [VP_] too]. b. *[A The detective [VP sold a car to Jim]] and [E the cop did [_] to Mary too]. c. *[A The detective [VP told Mary to leave]] and [E the cop did [_] to stay, too].

Most (though likely not all) of the properties associated with the constructions in () carry over to VPE sentences with so and some version of auxiliary verb fronting: ()

[A The detective [VP found a child]] and [E so did the cop [VP_]].1

Ellipsis does not only have syntactic reflexes. Meaning is also at issue: the content of the elided VPE must parallel that of VPA. In (a), VPE is structurally identical to VPA, and yet it is clear that (a) cannot be interpreted as (). ()

[A The detective [VP found a child]] and [E the cop [VP saw a bird] too].

To account for these facts, assume that the form of the elided VP is governed by a constraint that requires parallelism between VPA and VPE (Fox ): ()

Parallelism VP-ellipsis is licensed only if at LF, the VP of the ellipsis clause E is identical to the VP of the antecedent clause A.

If “identical” means “containing the same syntactic constituents”, then the acceptability of (a), and the unacceptability of (b,c) follows directly—VPE must consist of the same 1 Inverted constructions like () are distributionally more restrictive than (i.a), e.g., they disallow negation:

(i)

a. The detective found a child, and the cop didn’t. b. *The detective found a child and so didn’t the cop. c. *The detective found a child and neither did the cop.

A multitude of reasons—syntactic and semantic (perhaps even phonological)—may conspire to lead to this restrictiveness. Yet these issues are obviously beyond the scope of the present study.

  



syntactic pieces as VPA. The unacceptability of () follows if Parallelism requires (roughly) semantic identity. Next, consider inflectional features (tense, person, number). These seem to be exempt from the Parallelism requirement, as the ellipsis clause needn’t parallel the antecedent clause. In (), both tense and number are different in each of the coordinated clauses: () [A The detective [VP found a child]] and [E the cops will [VP_] too]. Data like () suggest that the relevant features are left outside of the ellipsis, providing a first indication that the size of ellipsis may not amount to the complete size of VPA (see van Craenenbroeck  for a recent review). Adjusting the size of ellipsis allows Parallelism to be maintained in its strict form (). Still, a Parallelism constraint that forces content identity must allow for certain holes. That is, the VPE sometimes allows an interpretation that is different from that of the VPA. As is well known, anaphoric elements in object position of the antecedent clause bring about such situations (Williams b; Bach and Partee ; Reinhart a). Thus, the meaning of VPE in () is ambiguous between (a) and (b): ()

Jim walked his dog, and Bill did [VPE] too. a. . . . Bill walked Jim’s dog. b. . . . Bill walked Bill’s dog.

Given the strict Parallelism in (), how can the ambiguity in VPE arise in the (apparent) absence of ambiguity in VPA? Put differently, how can we reconcile between the need for a rigid Parallelism that would force the facts in ()–(), and the need for flexibility to account for ()? One key idea (Reinhart a; Grodzinsky and Reinhart ) allows us to maintain rigid Parallelism by playing with the interpretation of the antecedent pronoun in VPA. Assume that this pronoun is ambiguous: it may be assigned to Jim, but it may also be viewed as a variable, bound by a local antecedent t1. In this clause, the two meanings happen to converge. ()

a. [A Jim [VP walked his!Jim dog]] b. [A Jim [1 t1 [VP walked his1 dog]]]

Now, consider the ellipsis clause. For each meaning, Parallelism guarantees an identity relation between VPA and VPE, and the ambiguity of the former () leads to these LFs of the latter: () a. [E Bill [VP walked his!Jim dog]] b. [E Bill [1 t1 [VP walked his1 dog]]] The difference in reading comes from the fact that while the pronoun in (a) is assigned to Jim as before, in (b) it is locally bound. The ambiguity follows, as the resulting LFs of the ellipsis sentence are roughly those in (): ()

a. [A Jim [VP walked his!Jim dog]] and [E Bill [VP walked his!Jim dog]] b. [A Jim [1 t1 [VP walked his1 dog]]] and [E Bill [1 t1 [VP walked his1 dog]]]



, ,  

The meaning in (a), known as strict identity, allows only for one dog to have been walked—Jim’s; whereas the meaning of (b), known as sloppy identity, allows two. The ambiguity is thereby obtained, and Parallelism is nonetheless observed. While this analysis is not problem-free (cf. Heim , ; Fox , among many others), it illustrates how Parallelism interacts with other assumptions to account for a range of VPE phenomena.2 Lastly, there is the critical issue of recoverability of the elided part of VPE. Even on its broadest formulation, Parallelism only constrains the form and content of ellipsis; it says nothing about the elision itself. But why and how can phonologically overt material be missing? Is the ellipsis site itself an empty category of some sort (Lobeck  and especially van Craenenbroeck , which propose that the conditions that license it are akin to those that license pro)? Do the conditions that allow for elision need to be specifically formulated for VPE, or are they, rather, general to all ellipses (cf. van Craenenbroeck and Merchant )? Moreover, is VPE a result of PF deletion, or rather, of copying at an abstract syntactic or semantic level? At present, we remain agnostic on these issues, as the goals of this chapter are modest. Our experimental evidence from developing children and brain-damaged aphasic patients will mostly bear on developmental and neuropsychological issues. Still, we will argue that the aphasic deficit may shed some light on the nature of ellipsis licensing.

. T    VPE: B’ 

.................................................................................................................................. Next, we turned to test the comprehension of patients with Broca’s aphasia (Walenski et al. ). Why can VPE be informative in this context? We can see two reasons: first, results from a comprehension experiment on VPE may bear on our view of the comprehension deficit in this syndrome, and as a consequence, on our understanding of the role of Broca’s region in language processing. Second, the pattern of impairment and sparing that would emerge might point to a neurological natural class that would bear on the theory of VPE.

.. Broca’s aphasia as a clinical entity Using patients with Broca’s aphasia requires some discussion regarding the status of this clinical entity. Some have questioned the validity of Broca’s aphasia as a stable and identifiable syndrome complex, presenting a pattern that is replicable across patients. The main argument behind the challenge of this -year-old syndrome has been that

2 Among the drawbacks of this analysis is that it has the consequence that a possessive pronoun in object position is always ambiguous (between the bound variable and referential readings), or else the copying of two different LF representations into E would not be possible (cf. Fox ; Heim  for discussion).

  



the observed inter-patient variability is too large for this collection of phenomena to count as unitary (Caramazza , passim). Caramazza’s paper paved the way not only for an intense, protracted conceptual debate, but also for several quantitative analyses in which the inter-patient variability, mostly in the receptive domain, was measured. The result that seems to have settled this round of debate was based on a series of retrospective studies on a large data set (n = , Drai and Grodzinsky a, b): while there is considerable variation between patients who have been assigned this diagnosis, certain performance contrasts in the receptive domain remain very stable. The grouping of these patients, then, seems justified.

.. The comprehension deficit in Broca’s aphasia: TDH vs WM/generic complexity Broca’s aphasia was traditionally thought of as a deficit in production (cf. Goodglass and Kaplan  for the most frequently used diagnostic test). Later, this deficit was shown to extend to comprehension (Caramazza and Zurif ). Since then, its precise nature—and the subsequent role of Broca’s region in language processing—has been hotly debated. At present, there seem to be two dominant views of the comprehension deficit, and the role of Broca’s region: the movement view, embodied in the Trace Deletion Hypothesis (TDH) and its processing ramifications (e.g., Grodzinsky , , ; Shapiro et al. ; Zurif et al. ; Shapiro, Swinney, and Borsky ; Grodzinsky and Santi ) and the syntactic complexity view, that relates to Working Memory (WM, e.g., Stromswold et al. ; Caplan and Waters ; Friederici ). A brief description of the positions is followed by a discussion of the relevance of a test of VPE comprehension in aphasia. In brief, the movement account is based on the observation that the core comprehension deficit is restricted to structures that involve the displacement of a phrasal constituent. Patients are presented with binary-choice interpretive tasks (essentially ϑ-role assignment, evinced through the choice of the correct ϑ-order in a set-up with two scenarios, one of which makes the sentence true, and the other false via ϑ-reversal). Patients with Broca’s aphasia are typically at chance on object-gap relative clauses (whether these are on the subject to form a center-embedded sentence, or on the object, thereby right-branching); their performance, however, is above chance when the gap in the relative clause is in subject position (see Drai and Grodzinsky a for a review). The idea behind the TDH is to set up the representational conditions that would derive this performance pattern— chance in displacement and above-chance otherwise. As the typical task involves proper ϑ-assignment, chance performance must stem from ϑ-confusion. This is obtained by the following: ()

a. Trace-Deletion Hypothesis (TDH) Delete all traces from agrammatic representations. b. Direct ϑ-Link strategy Assign a default ϑ-role to a ϑ-less NP (subsequent to trace deletion): Given a grid , and a linearized surface order NP1>>NP2>> . . . >>NPn, if NPi is ϑ-less, link it to ϑi.



, ,  

The consequence of the TDH is a limited role to Broca’s region in syntax—it only involves displacement.3 It has further been argued that the actual role of this region is the online linking of non-adjacent positions (Shapiro et al. ), most notably relating displaced constituents to extraction sites. The evidence comes from deficiencies in real-time tasks that critically rely on the linking of traces to their antecedent. Such deficiencies are observed for patients with Broca’s aphasia through a host of reaction time studies that so indicate. As we shall see, this type of task has also been used in testing VPE in aphasia (Poirier et al. ). Regarding comprehension performance on VPE structures, it is clear that a limited deficit is predicted, observed only when comprehension depends on the analysis of trace–antecedent relation in a manner that is not properly compensated for by the Direct ϑ-Link strategy. On the other side of the debate, it has been argued (e.g., Caplan and Waters , and much subsequent literature) that Broca’s aphasia is a Working Memory (WM) failure. As WM for “language” is said to reside in Broca’s region (cf. Smith and Jonides ), its demise leads to problems with “syntactically complex” structures (the precise nature of complexity is rarely made explicit). The idea is that syntactically complex material taxes WM in an incremental manner, and hence its failure results in an impairment in speakers’ ability to process “complex sentences”. The absence of a formal complexity metric is justified empirically: based on reaction time data, certain constructions are deemed more complex than others. E.g., object-gap relative clauses take longer to process than their subject-gap analogues, and are hence viewed as more complex and more taxing on WM (and to increased activity in Broca’s region in neurologically unimpaired participants; see Makuuchi et al.  for recent discussion). A relatively widespread comprehension failure subsequent to damage to Broca’s region follows. That is, any construction that is deemed as “syntactically complex” is likely to cause comprehension problems to patients that suffer from this brain disease. In particular, that VPE is a complex construction appears uncontroversial: any reasonable construal of the notion of syntactic complexity must include VPE, as this construction, at a minimum, is biclausal, involves copying/deletion of large chunks, and potentially LF movement. WM must be taxed: for Parallelism to be implemented, a comparison of the VP in A to the one in E is mandatory. At present, little evidence is available that pertains to this debate.4

3 No empirical test has been conducted to examine whether deletion extends to other empty categories and elements that are represented on the surface (Grodzinsky ). 4 We are aware of one attempt to study VPE in aphasia: Vasić, Avrutin, and Ruigendijk (b) used what they claimed are Dutch VPE constructions (although see van Craenenbroeck : – for recent discussion of whether or not Dutch has VPE). Vasić et al. tested patients with Broca’s aphasia in a picture selection task. They were interested in the interpretation of pronouns, hence all their sentences contained anaphora, with distractors as described in (i). The relevant foil is the “related distracter” (i.c), which forces the possessive pronoun in the antecedent clause to have a local antecedent, but the possessive pronoun in the ellipsis clause to have a deictic antecedent. For this meaning to match the sentence, Parallelism must be violated.

(i)

a. Here are L, M, N. L photographed her horse and M did, too. b. Correct: LL’s horse & M L/M’s horse (strict and sloppy readings). c. Foil: LL’s horse M N’s horse (deictic reading).

  



Hence, a WM failure due to damage to Broca’s region should lead to serious comprehension problems (whose precise nature depends on the complexity metric that is assumed). We can see that the two accounts contrast in prediction regarding VPE: the predicted TDH-based deficit is structured and restricted, whereas the predicted WM-based deficit is rather widespread.

.. Testing VPE with patients suffering from Broca’s aphasia We based ourselves on methodology developed in Grodzinsky (), and used a verification task with VPE sentences with diagnosed Broca’s aphasic patients. Focusing on Parallelism, we presented VPE structures along with situations whose interpretation forces the material in A to have its parallel in E. That is, we presented pairs to patients, where the denotation of the elided part sometimes matches the situation, and sometimes does not. The patient’s task was to assign a truth-value to the sentence. We now elaborate on this task. Verification is an experimental paradigm that aims to test the patient’s ability to fill in content from A into E: all scenarios make A true, and the only potential sentence/picture mismatch pertains to E, the ellipsis part. This set-up forces the patient to consult Parallelism in order to assign an interpretation to the sentence and determine its truth-value. If sentences that accompany situations requiring improper reconstruction are deemed true, we can infer that Parallelism does not exist in the patient’s grammar (or at least cannot be put to use at the right time); otherwise, we can conclude that she possesses the relevant knowledge and can put it to use. To illustrate, consider the sentence in (), and the accompanying scenarios (a–d) in Table .. If constraints on ellipsis are not known, or unavailable for use, then any interpretation of E that is consistent with an E whose elided part contains a man as either subject or object may be legitimate. This can be put to the test, in which the evaluation of VPE would provide the critical clue to the patient’s knowledge of Parallelism. If Parallelism is not known, and is hence violable, then the range of possible interpretations of VPE is extended. This is what was put to the test here. In (), we present a sentence, and four accompanying scenarios (a–d). Each scenario consists of two events, one that always makes A true, and a second one that makes E true (a) and false (b–d) in three different ways. A patient lacking Parallelism should accept all scenarios (a–d) as true, because they all feature a man—the only constraining factor for VPE interpretation under this assumption. Conversely, a patient that masters Parallelism should only deem true (a). The difference between knowledge of Parallelism and lack thereof is testable, and presented in Table ., which marks the possible responses of a patient with and without Parallelism:

These scenarios may have led to infelicity: a silent pronoun (the copy in the elided VP) refers deictically, even though emphatic stress (cum pointing) is necessary for deixis to work in such contexts. (NB. controls’ error rates were ~ percent, a non-negligible level). This work, at any rate, did not test Parallelism.



, ,  

Table . Test conditions: VPE sentence + images Parallelism available? ()

[A The woman kicked a tiger] and [E the man did [VP__] too]

Yes

No

Match

“A woman kicking a tiger, and a man kicking a tiger”

T

T

MisMatch 

“A woman kicking a tiger, and a tiger kicking a woman”

F

T

MisMatch 

“A woman kicking a tiger, and a woman kicking a man”

F

T

MisMatch 

“A woman kicking a tiger, and a tiger kicking a man”

F

T

This part of our test consisted of Match cases (a), as well as MM (b), depicting role reversal in E, the ellipsis clause; MM (c), corresponding to an incorrect object in E, and MM (d), corresponding to an incorrect subject and object in E. Given the world created in the images, both yes and no responses are plausible. As controls, we used non-elided coordinate structures, whose primary meaning was identical to the elided ones. In the absence of ellipsis, the invocation of Parallelism is not necessary for interpretation, no error is attributable to a Parallelism failure, and therefore, all the relevant populations are expected to perform at ceiling (see Table .).

Table . Control condition: Coordination + images ()

[IP1 The tiger kicked a tiger] and [IP2 the man kicked a tiger] Expected response

Match

“A woman kicking a tiger, and a man kicking a tiger”

T

MisMatch 

“A woman kicking a tiger, and a tiger kicking a woman”

F

MisMatch 

“A woman kicking a tiger, and a woman kicking a man”

F

MisMatch 

“A woman kicking a tiger, and a tiger kicking a man”

F

.. Experiment As we discussed, our test required subjects to verify sentences against depicted visual scenarios—participants responded “yes”/“no” to pairs. In trying to adhere to principles of good design, we took extra measures to ensure that Parallelism must indeed be consulted, and that no tricks or heuristics would help the aphasic patient get around it. First, the design of scenarios was crucial: four scenarios were coupled with each experimental sentence—one scenario made the sentence true (the Match M) whereas the remaining three scenarios (MisMatches –) represented three different ways of making the sentence false. Each scenario was shown separately, and thus every sentence was presented four times—once for each scenario.

  



Second, as all our stimuli contained two conjoined clauses, they required two situations per scenario—one for each of the clauses A and E (), or IP1 and IP2 in the control condition (); the scene for A always made it true, as the mismatches were made false only by the scene that related to E. Third, we tried to ensure that there is congruence between NPs in the sentences and the characters in images—every character in the pictures is mentioned, and every NP is depicted. This way, the presence or absence of mentioned referents cannot be used as a heuristic. Fourth, to avoid a violation of the uniqueness requirement imposed by definites, the object of the antecedent clause A was headed by an indefinite. Fifth, every character appeared in pictures as either actor or recipient of action. Sixth, animacy and plausibility were perfectly balanced (rotated across positions in the sentence). Seventh, we had one M and three MM conditions, which could lead to a “yes”/“no” imbalance. We therefore maintained a : ratio between tokens that required a correct “yes” (M) vs those requiring a “no” (MM) response, thereby preventing strategic responding. We had  Match and  (*) Mismatch tokens (= items per condition), for a total of * = trials, as in Table ., and the accompanying images in Figure .. As control populations, we tested healthy and right-hemisphere damaged patients, who were at- or near-ceiling in all conditions. Next, we started testing patients with a clinical diagnosis of Broca’s aphasia. At present, we have results for three patients, as finding candidates for testing is exceedingly difficult (Walenski et al. ). These patients’ clinical profiles can be found in Table .. All three subjects exhibited the familiar TDH-based comprehension pattern: in a sentence-to-picture matching task with “reversible” sentences, they performed better on actives than on passives, and better on subject than on object relative clauses. Once clinical and TDH-based diagnoses were established, we proceeded to the VPE test. Subjects were tested in several sessions, with a short training session, and many breaks during testing. The experiment was conducted with an ethics approval and informed consent in keeping with the ethical principles of testing that apply at McGill University, as well as San Diego State University, where the subjects were tested. Healthy education- and age-matched control participants performed near or at-ceiling on all conditions. Table . details how our patients performed on the VPE test.

Table . Experimental conditions Match MisMatch  MisMatch  MisMatch  a. COORDINATION: The girl kicked a tiger and the boy kicked a tiger









b. VPETOO: The girl kicked a tiger and the boy did too









c. VPESO: The girl kicked a tiger and so did the boy











, ,  

Match:

MisMatch 1

MisMatch 2

MisMatch 3

 . Experimental stimuli

Table . Patients’ demographic and clinical profiles Gender

BDAEa severity level ( = severe,  = mild)

Years post stroke

Age at test (years)

Educational level



M







College + 



F







High School



M







PhD

Subject

Note: a = Boston diagnostic aphasia examination

Table . Number of accurate responses per condition for each patient on the VPE test Control condition (number correct)

VPE (number correct)

Subject M (/) MM (/) MM (/) MM (/) 

 

























M (/) MM (/) MM (/) MM (/) TOO









SO









TOO









SO









TOO SO

 

 

 

 

  



Overall, Patient  performed well (a total of / [ percent] correct); Patient  performed slightly worse (a total of / [ percent] correct); and Patient  performed best (a total of / [ percent] correct). Patient  leaves little room for discussion, being near- or at-ceiling on all conditions. There can be little doubt that he knows and respects Parallelism. The other patients require a more detailed examination to reveal the source of their errors. Patient  was near- or at-ceiling on M, MM, and MM across all three conditions, and erred systematically on MM across the board—he was at  correct on both ellipsis and overt coordination. That is, he accepted as true every MM trial, which is judged False by healthy speakers, because it features a recipient of action that does not correspond to the object of VPA (hence the covert object of VPE did not refer to it). What is crucial in this case is the fact that he judged these sentence–image pairs as True, in both control and two test conditions. This highly systematic error, whose source is unknown, generalizes over both VPE and coordination, hence seems unrelated to VPE. Viewed against his performance on the control, then, we can safely assert that  demonstrated knowledge of Parallelism, and an additional yet unrelated deficit. Patient  presents a more complex picture. On the Match trials, she was not as apt as the previous patients, but nevertheless performed above chance. This relatively weak performance (across both VPE and COORD conditions) provides the backdrop against which her performance on the MisMatch trials must be evaluated. She was at chance on MM and MM, and above chance on MM. These results are not easy to analyze quantitatively or interpret, and will be left in their raw form. Performance on the COORD condition (expected to serve as a baseline) is rather low; performance level on the test conditions must be evaluated accordingly. Two aspects of the result provide hints regarding the preservation of Parallelism: first, on the Match trials of both +VPE conditions, the patient scores are well above chance; second, performance levels are almost identical for both +VPE conditions. Thus, from a perspective that focuses on Parallelism, the performance of this small group of patients provides no evidence for failure in both knowledge and execution. Moreover, the virtually identical performance scores of all three patients across the two VPE types ( . . . did too, . . . so did) strongly suggests that the problem does not arise from a lexical deficit (e.g., misunderstanding of the word too). Together with the fact that the main sources of error were independent of VPE (as they occurred as frequently in the control condition), this cross-condition identity also lends credibility to our interpretation, according to which a Parallelism failure is not part of the deficit in Broca’s aphasia.

.. Patients’ relative success in analyzing VPE and the complexity/WM view We conducted this experimental study with the hope of shedding new light on the comprehension deficit in Broca’s aphasia, and of gleaning new neurological insights regarding the theory of VPE. We can now clearly assert that the patients were able to use Parallelism for the comprehension of VPE. This assertion helps to rule out the (rather vague) view that Broca’s aphasic patients fail to comprehend complex sentences—VPE can hardly be viewed as a simple construction. Regarding a WM deficit: an overt VP in clause A and its elided copy in clause E are large non-adjacent chunks; VPA cannot be reconstructed online in VPE with WM.



, ,  

Our results thus exclude an account of Broca’s aphasia as a generic deficit in analyzing syntactically complex stimuli, or as a plain WM impairment.5

.. Patients’ relative success in analyzing VPE and the movement (TDH) view What about a movement, TDH-based account? VPE contains a relation between the antecedent VP and the ellipsis site (VPA and VPE); we might therefore wonder whether this dependency relation is affected by the deficit in Broca’s aphasia. Standard approaches to VPE view the relation between overt and elided material as being akin to pronominal binding, albeit a silent one. Schwarz () and Elbourne () argue against a movement analysis on the basis of highly complex data.6 If the deficit in Broca’s aphasia is restricted to traces, this approach predicts that patients’ comprehension ability of VPE as tested here is preserved. Prima facie, then, VPE stays outside the scope of the TDH, hence its representation is expected to be intact in Broca’s aphasia, as our experiment indeed found.7

.. VPE in reaction time experiments So far, we have looked at experiments in which comprehension error rates are the dependent measure. We now turn to the time domain. Here, we find two reaction time studies on VPE with Broca’s aphasic patients (Poirier et al. ; Walenski et al. ). These present an interesting new puzzle. Consider first some results about priming-at-a-distance—a paradigm in which participants are tested in a priming paradigm, however the target is not adjacent to the prime, but rather appears further downstream. Neurologically intact populations do show remarkable priming-at-a-distance that is based on trace–antecedent relations. A displaced antecedent can prime not only adjacently, but also at the position of their trace: ()

The policeman saw [the boy] [who [the crowd at the party the crime.

[1]

] accused

[2]

] of

Nicol and Swinney (), and many other authors since, have used the Cross-Modal Lexical Priming technique to determine whether listeners fill gaps online. They presented subjects with spoken sentences such as (). At the positions marked by [] and [], visual 5 Other pieces of empirical evidence point to a similar conclusion regarding a WM deficit (cf. Grodzinsky and Santi ). 6 Grodzinsky () argues, based on preliminary data from an experiment with patients, that VPfronting at LF is mandatory in VPE, as VPE is impaired in Broca’s aphasia. He argued that if the TDH is the right deficit analysis, then mandatory VP-fronting in VPE would predict the pattern he observed. However, the data presented above, which come from a study with much improved design, seem to point to the intactness of VPE in Broca’s aphasia, and hence no conclusions regarding VP-fronting follow. More refined experiments are needed to bear on this issue. 7 Tanja Temmerman suggests that if we add a trace to a VP-ellipsis construction, the deficit would manifest. She proposes to test cases of remnant movement:

(i) I don’t remember what Ryan made for our Valentine’s Tea, but I know what Alice did. (ii) I know how fast Adam could run, but I don’t remember how fast Hilary could.

  



probes that were semantically related to the correct antecedent (boy), as well as unrelated control probes, were presented. As always with this task, subjects were required to make lexical decisions to the visual probes. It was discovered that there was significant priming for probes related to the correct antecedent at the gap ([]), but not at the pre-gap ([]) probe site. The same group later tested patients with Broca’s aphasia, who failed to prime in this way (e.g., Zurif et al. ), in keeping with the predictions of the TDH. With this picture in mind, Shapiro et al. () turned to VPE. They found that in neurologically intact populations, the copy at the ellipsis site induces priming to a noun related to the elided object. Participants are given a Cross-Modal Priming task with sentences as in (): ()

The locksmith photographed the babysitter and the friendly [1] neighbor did [2] too, according to [3] the clumsy plumber.

Healthy adults reacted faster to a related word when it is presented at point [], but not when presented at points [] or []. This pattern seems to be due to the fact that this word was primed by the copy of the elided object, as priming was observed only at the ellipsis site (point []). Against this background, Poirier et al. () tested patients with Broca’s aphasia on a similar task: Broca’s patients evinced no priming of the object from the antecedent VP at the elided position in the ellipsis clause, though in a similar study (Walenski et al. ), Broca’s patients evinced delayed priming (that is, priming at a probe position downstream from the elided position). The conflict between these results and ours is puzzling: previously, data from CrossModal Priming aligned with comprehension data. That is, patients failed to prime exactly in constructions where their comprehension was impaired. In VPE, we find evidence for the intactness of Parallelism in comprehension in the face of priming deficiencies. At present, we have no account of this performance split, but we note that the patients with Broca’s aphasia do prime for the elided material, though it is later in the time course of the sentence relative to unimpaired participants. Thus, while the results we obtained are highly structured, and help to exclude certain accounts, a full explanation of the experimental findings is not yet available. Experimentation with ellipsis that would go beyond VPE—gapping, sluicing, N’-ellipsis—would hopefully help to shed light on the nature of the deficit in Broca’s aphasia.

. T    VPE: A

.................................................................................................................................. We now move to discuss another population with incomplete linguistic abilities— developing children, whose abilities in VPE bear on yet another central issue—the “innateness debate”. If children only imitate input, as many have argued (cf. Gennari and MacDonald  for a recent statement), and if “input” only means overt phonological strings, then there is little room for ellipsis in children’s language, for it requires the reconstruction of missing material in the ellipsis clause (E) from the antecedent clause



, ,  

(A)—an operation for which no source of evidence is available on this approach. To produce and properly analyze VPE, children must go “beyond the information given,” especially given the paucity of ellipsis constructions in the input. To see how directly children’s abilities on ellipsis, and on VPE in particular, bear on this issue, consider first a recent exchange on “innateness” in the context of children’s comprehension of relative scope of logical operators, and specifically on whether they are able to represent “inverse scope” of quantifiers. At issue are scopally ambiguous sentences—ones that contain more than one Scope-Bearing Element as in (). Representations of different orderings of the bolded words not and some produce distinct meanings—(a) entails that no child was found by the detective, whereas (b) does not: ()

The detective did not find some children.

() a. It is not the case that the detective found some children. ¬[∃x[child (x) & [detective find (x)]]] b. There are some children such that the detective did not find them. ∃x[child (x) & ¬[detective find (x)]] At issue is whether children who are confronted with () are able to represent not only surface scope representations (a), but also inverse scope relations (b). Going “beyond the information given” (Bruner ) amounts to analyzing () as (b). Phonological material is identical in both cases, but (b) requires an abstract (LF) reordering operation. Details aside, Musolino and Lidz (), as well as Conroy, Lidz, and Musolino (), have claimed that children (up to : years of age) are only able to represent surface scope (a).8 Gennari and MacDonald () argue that this result provides empirical support for the claim that language development is entirely dependent on input, rather than determined by innate knowledge. Specifically, they claim that distributional patterns observed in adult speech production influence both adult and child comprehension. Given that inverse scope is rarely produced by adults, children cannot be expected to possess it early on. Children’s performance on VPE may shed new light on whether syntax and semantics acquisition is a purely input-driven process, or is, rather, privy to prior, innate, linguistic knowledge. While relatively infrequent, VPE comprehenders must invariably go beyond the information provided by surface form—they must complete material missing in E, and guarantee that it parallels A. No surface-based strategy, allegedly used by children in the “inverse scope” case discussed above, can rescue the child. She must reconstruct elided material in a highly constrained fashion. Parallelism, a crucially necessary principle, cannot be attested in the input, and the computation is not easy to perform if one’s parsing system is still budding. Moreover, if abilities depend on exposure, as Gennari and McDonald () would have it, then VPE should be very difficult to master, as it is rather infrequent in use. Gennari and MacDonald, then, would expect constraints on ellipsis to be absent from the child’s syntactic arsenal.9 A radical innatist, by contrast, would expect ellipsis to be known at an early age. At hand is, therefore, a prediction that distinguishes the two approaches. 8

Gualmini and colleagues () argue against this position, on empirical grounds. Gennari and MacDonald’s position would thus be unable to explain how Parallelism (or any other unattested constraint) is ever attained by humans. 9

  



And yet, the few studies of VPE in children (most notably, Thornton and Wexler ; cf. also Göksun et al.  for some experimental work, and Santos  for work on Portuguese) have not directly tested their ability to invoke Parallelism. These studies have by and large focused on the interpretation of pronouns in VPE (in the context of the so-called “condition B debate” of the s; Chien and Wexler ; Grodzinsky and Reinhart ). Little experimental work has been done on other aspects of VPE (with the exception of Santos a; see subsection ..). We therefore set about providing experimental evidence from VPE that would bear on the debate.

.. Spontaneous production of VPE: Parallelism is used in early childhood Our first step was to investigate young children’s abilities to check that VPA=PVPE (see Grodzinsky ). To anticipate our findings, we found that their comprehension performance—for reconstruction under Parallelism—was near-perfect at a surprisingly early age. This remarkable finding also emerged in a search of the CHILDES database, which indicated that children use VP-ellipsis in free speech very early on, even before they turn . There were not many such uses ( instances out of  utterances for the -yearolds, and  out of  for the -year-olds), but VPE was nonetheless attested in the spontaneous speech of several children (see Table . for examples). It is difficult to see how any of these utterances can be analyzed as anything other than VPE. Their meaning, moreover, cannot be reconstructed in the absence of Parallelism: Abe

Table . Examples of spontaneous production of VPE by children Name of child

CHILDES file name

Example

Abe

Kuczaj_Abe

CHI: I don’t know. FAT: you didn’t do anything? CHI: we did too. FAT: tell me. CHI: we did a lot of stuff.

Ben and Deb

Garvey_Bendeb

DEB: I will be the policemen okay? BEN: no I wanta [:want to] I wanta [:want to]. DEB: I will too okay? DEB: okay? DEB: okay.

Nina

Suppes_Nina

CHI: what kind of bus does he have? MOT: he has a Volkswagen bus. CHI: Dennis does too. MOT: that’s right Gary and Dennis have the same car. CHI: uhhuh [=yes].



, ,  

responds to a full negative question by eliding the VP and excluding its negation. To be sure, he then elaborates on the father’s request, and reiterates the VPE as a full sentence. Deb asks for consent to her being the police(wo)man, and Nina uses VPE by way of confirming that Dennis has a VW bus. And these, recall, are -year-olds, said to be unable to carry out operations that involve unattested stimuli and use inverse scope because it is too taxing on their parser. These data add to evidence by Santos (a), in which examples of child productions in Portuguese are given, that are claimed to be unambiguous instances of ellipsis.

.. Children comprehend VPE very early on: Experimental evidence Our success in finding evidence that children are capable of constructing VPE in production led us to a comprehension study, in the hope that it would bear directly on the “innateness/input dependence” debate. As -year-old children have difficulties in sitting still and concentrating, we turned to -year-olds. Focusing on Parallelism, and basing ourselves on methodology developed in Grodzinsky (), we planned to give children a verification task with VPE sentences, presented along with situations whose interpretation forces the material in A to have its parallel in E. The method was to present pairs to children, where the denotation of the elided part sometimes matches the situation, and sometimes does not. The child’s task would be to assign a truth-value to the sentence. We now elaborate on this task, which has also been used by Santos (a), who found that by and large, children interpret ellipsis in an adult-like fashion. As this experimental paradigm aims to test the child’s ability to fill in content from A into E, all scenarios make A true, and the only potential sentence/picture mismatch pertains to E, the ellipsis part. This set-up forces the child to consult Parallelism in order to assign an interpretation to the sentence and determine its truth-value. If sentences that accompany situations requiring improper reconstruction are deemed true, we can infer that Parallelism does not exist in the child’s grammar (or at least cannot be put to use at the right time); otherwise, we can conclude that she possesses the relevant knowledge and can put it to use. As an example, consider the sentence in (), and the accompanying scenarios (a–d) in Table .. If constraints on ellipsis are not known, or unavailable for use, then any VP may be

Table . Test Condition: VPE + images Parallelism available? () [A The detective found a child] and [E the policeman did [VP__] too]

Yes

NO

a.

“A detective finding a child, and a policeman finding a child”

T

T

b.

“A detective finding a child, and a policeman finding a detective”

F

T

c.

“A detective finding a child, and a detective finding a policeman”

F

T

d.

“A detective finding a child, and a child finding a policeman”

F

T

  



Table . Control condition: Coordination + images () [IP1 The detective found a child] and [IP2 the policeman found a child] a.

“A detective finding a child, and a policeman finding a child”

T

b.

“A detective finding a child, and a policeman finding a detective”

F

c.

“A detective finding a child, and a detective finding a policeman”

F

d.

“A detective finding a child, and a child finding a policeman”

F

a legitimate completion for E. If so, then she should accept all the situations (a–d) as true (this deficit would become apparent when their errors on VPE are compared to the control coordination condition); conversely, if the child knows and uses constraints on ellipsis, she should only deem (a) as true. The difference between these situations is depicted in Table ., which marks the possible responses of the child with and without Parallelism. Note that all the situations in Table . are plausible, and the child’s task is to check the sentence against them, with constraints on ellipsis being the sole factor determining truth conditions. To make sure that sentence length, or number of clauses, are not an intervening factor that increases difficulty, we used non-elided coordinate structures with the same meaning as controls, as detailed in Table .. Our test, then, required subjects to verify sentences against depicted visual scenarios— participants responded “yes”/“no” to pairs. In trying to adhere to principles of good design, we took extra measures to ensure that Parallelism must indeed be consulted, and that no tricks or heuristics would help the child get around it. We ran this experiment at the Centre de La Petite Enfance de McGill (with an ethics approval and informed parental consent in accordance with the McGill Research Ethics Board). All participants were right-handed with normal hearing. They range in age from : to : ( boys, mean=  months). Our participants were seven children. They were tested by one of us (I.D.) one child at the time, in a secluded area within their classroom. Sentences were read aloud, and the pictures shown concomitantly. The children responded to each pair verbally, and their responses were recorded immediately. The relatively large number of test items (n =), and the young age of the children, forced – testing sessions per child. Prior to each testing session, the instructions and four training sentences were presented to each child. Each session lasted no more than  minutes. Results (number of correct responses) are presented in Table .. As can be easily seen, there were hardly any errors on any of the conditions. In both control and test conditions, children correctly verified the Match pictures, and rejected the MisMatches. We can safely conclude, it seems, that in every instance where the analysis of conjoined clauses was not an obstacle, Parallelism was used for ellipsis reconstruction. In conclusion, then, children invoke Parallelism in VPE in both production and comprehension. VPE is an infrequently used sentence type, and the task forced children to use a linguistic constraint to which they have no direct evidence in the input.



, ,  

Table . Experimental results Control condition

VPE

Child M (/) MM (/) MM (/) MM (/) 























 











































M (/) MM (/) MM (/) MM (/) TOO









SO









TOO









SO









TOO









SO









TOO









SO









TOO









SO









TOO









SO









TOO









SO









Our results seem to provide clear evidence for the use of a highly abstract, unobserved, linguistic principle—Parallelism—in the processing of VPE. The significance of this result is further underscored by the relative rarity of VPE in the child’s ambient language. Early mastery of VPE speaks clearly against Genarri and MacDonald’s position, and weakens the force of the “isomorphism” claim, that can no longer cover VPE, even though VPE would be a natural candidate for such a requirement.

. W  ? W   ?

.................................................................................................................................. We have seen relatively good comprehension performance in our pilot study of Broca’s aphasia. We have also seen that very young children use and analyze VPE properly, despite the fact that such use requires highly abstract principles that are never attested in the data. Theoretically, the evidence from aphasia speaks against a generic “complexity” deficit in these patients, and suggests that VPE can be used as a research tool with patients, which would hopefully explore more refined structural issues, aimed at providing evidence that is relevant to theory construction. With children, we have provided yet another demonstration that highly complex knowledge is attained in the absence of evidence. Experimentally,

  



it is clear that our perspective must first be broadened to other types of ellipsis—gapping, sluicing, N’-ellipsis. In children, we must look at how abstract principles like Parallelism interact with children’s presumed deficiencies with inverse scope. In aphasia, we must gain deeper understanding of the discrepancy between on- and offline performances. In the meantime, the preliminary hints at hand suggest that ellipsis and its kin are likely to provide important clues on language development and breakdown.

A We gratefully thank the reviewers, as well as the editors of this volume, whose comments helped to improve this manuscript. This work was supported by ELSC, and by a grant from the Israel Science Foundation (ISF) to Y.G.

  ......................................................................................................................

                ......................................................................................................................

 

. I

.................................................................................................................................. T chapter summarizes what experimental studies on ellipsis can tell us about the mechanism of online sentence processing.1 Grammatical studies on ellipsis have revealed many properties of ellipsis constructions (see Chapter  and Chapter ). Among others, the following three properties are prominent. First, when ellipsis is licensed, the ellipsis site normally has a salient linguistic antecedent (see Hankamer and Sag  for detailed discussion). Second, related to the first property, the ellipsis site and its antecedent site must stand in a certain parallelism relation. In other words, the ellipsis site is not licensed when the ellipsis site does not have a parallel antecedent (see Merchant ; Chung  for detailed summary and discussion on the parallelism requirement on ellipsis). Third, the ellipsis site is often licensed in a specific syntactic configuration (Lobeck , among many others). Thus, in order to successfully comprehend sentences involving ellipsis, the online sentence-processing mechanism must be able to identify the ellipsis site, find its antecedent, and recover the content of the ellipsis site by referring to the antecedent. The psycholinguistic study of ellipsis aims to reveal aspects of the mechanism of online sentence processing by studying how ellipsis constructions are processed online.2 The major questions in the psycholinguistic study of ellipsis can be formulated in the following ways. What structure does the parser build in the ellipsis site? How does the parser build the structure of ellipsis? When does the parser recognize ellipsis? Where does the parser find the antecedent 1

This chapter is crucially NOT about what the psycholinguistic study of ellipsis can tell us about grammatical analyses of ellipsis. Given this purpose of this chapter, I will try to avoid going into detailed discussion on issues of the syntax and semantics of ellipsis throughout this chapter, and, as such, citations on these issues will be limited to representative works. It is important to note that, many times, the interpretation of experimental results depends on the theory of ellipsis adopted. However, given the limited space, I will not be able to discuss the details of alternative interpretations of each experimental study that I will touch on. 2 Phillips and Parker () provide an extensive survey of the psycholinguistic study of ellipsis. What this chapter covers overlaps with some of Phillips and Parker (). This chapter, however, focuses more on issues of how ellipsis constructions are processed.

 



of ellipsis? In what follows, I discuss these four questions one by one and try to provide an overview of what the study of ellipsis has been revealing about the mechanism of online sentence processing.3

. W         ,  ?

.................................................................................................................................. The content of the ellipsis site must be recovered from the antecedent. Therefore, during online processing, the parser must find the antecedent and recover the content of the ellipsis site from the antecedent. Thus, it is possible to infer the structure in the ellipsis site by looking at the structure of the antecedent. In other words, the degree to which the parser is sensitive to the structural details of the antecedent can indicate how extensive the unpronounced structure is. Making use of the Cross-Modal Priming (CMP) paradigm, Shapiro and colleagues (Shapiro and Hestvik ; Shapiro et al. ; Poirier et al. ) have investigated whether the content of the antecedent is “reactivated” at the ellipsis site. Their study is motivated by CMP studies in other domains. In the study of anaphora and empty categories, it has been argued that the content of the antecedent of a pronoun or of a gap linked to a wh-phrase is reactivated once the pronoun or the gap is processed (McElree and Bever ; Nicol et al. ; Nicol and Swinney , among others).4 The interpretation of the pronoun or the gap depends on its antecedent, and to achieve the interpretation of the sentence, the processor needs to find the antecedent and link the pronoun or the gap to the antecedent. This process of finding the antecedent and linking the dependent element to the antecedent leads to the reactivation of the antecedent. Adopting similar logic, Shapiro and colleagues asked whether the content of the antecedent for the ellipsis site is reactivated. If the reactivation of the antecedent is observed, this means that the parser accesses the information of the antecedent in memory and tries to link the ellipsis site and the antecedent online. Shapiro et al. () investigated the processing of VP-ellipsis (VPE). In their experiment, while sentences like () were audibly presented a word was visually presented. The participants performed a lexical decision task while listening to the sentence. Shapiro et al. found faster lexical decision times when the word semantically related to the object NP in the first clause (e.g., neck for a tie) was presented at the VPE site (at the point of did). This faster lexical decision time was not observed when the word was presented in the earlier position in the sentence, prior to the VPE site. Furthermore, they did not find a priming effect of the subject the mailman. ()

The mailman [VP bought a tie for Easter], and his brother, who was playing volleyball, did [VPE ø] too, according to the sales clerk.

As far as I can tell, the vast majority of psycholinguistic studies of ellipsis have focused on the first two questions, and the latter two questions have not been asked until recently. 4 Note that McKoon et al. () argue against the position that anaphors show reactivation of the antecedents (see also Nicol, Fodor, and Swinney  for related discussion). 3





Their result suggests that when the parser recognizes VPE, it searches for the appropriate antecedent, the VP in the first clause, and links the VPE site to the antecedent, resulting in the reactivation of the antecedent VP.5 Therefore, reactivation of the object was observed while the reactivation of the subject, which is outside of VP, was not observed.6 This study suggests that the parser accesses the antecedent of an ellipsis, and also that the parser is sensitive to the content of the antecedent site, recovering only a VP as the antecedent of VPE to the exclusion of its containing clause. However, studies of this type do not tell us how the structure of ellipsis is built. In a series of studies on the processing of VPE and sluicing, Frazier and Clifton argue that the ellipsis site in these constructions is associated with syntactic structure and that the parser employs a copying mechanism to build the structure of ellipsis (Frazier and Clifton , , ). Essentially, what they argue is that the structure of the antecedent clause is literally copied into the ellipsis site. As evidence for this proposition, Frazier and Clifton () show that the distance between the ellipsis site and the antecedent impacts the processing of the ellipsis site.7 In one of the experiments in Frazier and Clifton (), they tested the following paradigm of sluicing: () a. Michael slept and he studied but he didn’t tell me what. b. Michael studied and he slept but he didn’t tell me what.

(near) (far)

In a self-paced moving-window reading study, they found that when the position of the verb (studied) associated with the wh-phrase is farther from the ellipsis site, the processing speed of the elided (sluiced) clause is slower. Assuming that either just one of the conjuncts or the whole coordinated structure can serve as the antecedent of ellipsis (see Merchant : – for related discussion), Frazier and Clifton argue that when the parser cannot find the antecedent in the nearest conjunct, it considers the whole coordinated structure as the antecedent.8 As a result, in (b) the bigger size of the antecedent (i.e., the whole A recent structural priming study also supports this point (Xiang et al. ; but see Cai et al.  who report that VPE in Chinese does not show structural priming effects and thus conclude that VPE is not associated with syntactic structures). 6 Snider and Runner () report similar results using a visual world eye-tracking paradigm (for details, see Phillips and Parker  and Snider and Runner ). 7 Frazier and Clifton () suggest that the process of copying is cost-free in the sense that the size or complexity of the antecedent does not impact the processing time of the ellipsis site. Martin and McElree, on the other hand, argue that such a result points to a content-addressable memory retrieval mechanism (Martin and McElree , , ), which is not affected by the syntactic structure of the retrieved materials. However, as I will discuss, the existence of antecedent complexity effects is controversial as some studies do show such an antecedent complexity effect (e.g., Frazier and Clifton  in their study on sluicing). Note that recent cue-based memory retrieval studies have shown that the parser utilizes structural cues as well as non-structural cues (Dillon et al. ; Kush et al. ; see also the discussion in Jäger et al. ). Thus, an antecedent complexity effect can be predicted by cuebased memory retrieval theories as well as by copying theories. I will avoid going into the details of the debate, but see Frazier and Clifton (), Martin and McElree (, , ), and Phillips and Parker (). 8 Frazier and Clifton () specifically assume that wh-movement out of a conjunct is possible when a wh-phrase is moved out of the nearest conjunct, but not when it is moved out of the more distant conjunct, based on the observation in (i). 5

 



coordinated structure) is taken as the antecedent by the parser, which gives rise to the reading slowdown. In a study of gapping, Carlson, Dickey, and Kennedy () also argue that the size of the antecedent impacts the processing of the ellipsis site.9 They found that in gapping sentences like (a), the NP in the second conjunct (Sam) was preferably interpreted as the object ( . . . insulted Sam). However, in sentences like (b), where the PPs are fronted in both of the conjuncts, this strong bias for the object interpretation of the NP in the second conjunct (Sabrina) was absent. ()

a. Bob insulted the guests during dinner and Sam during the dance. b. [PP At Marshall Field’s], Melissa saw a classmate, and [PP at J. C. Penney’s], Sabrina.

They interpret this result as indicating that the structural complexity of the antecedent clause impacts the processing of the ellipsis site. They argue that in (a), the VPcoordination structure is available, and thus, it is easy to interpret the NP Sam as the object of the verb insult. However, in (b), due to the fronted PP, which is adjoined to the clausal projection, only the bigger clausal coordination structure is available. Thus, the subject interpretation of Sabrina is more accessible and the strong object interpretation bias is not seen. This result, like Clifton and Frazier’s results, indicates that the structure of the antecedent affects the processing of the ellipsis site. Even though these studies show that the size of the antecedent matters in terms of the processing of the ellipsis site, they do not show whether and to what extent the parser is sensitive to the structural details of the antecedent. If the parser employs a copying mechanism, it is expected that the detailed structure of the antecedent will be copied into the ellipsis site, and thus that the parser will be sensitive to the structural details of the antecedent site while processing the ellipsis site. Dickey and Bunger’s () study compared the processing of merger-type sluicing (henceforth just sluicing: Chung, Ladusaw, and McCloskey ), in which the antecedent clause contains an indefinite NP as the antecedent of the wh-phrase, with the processing of sprouting-type sluicing (henceforth sprouting: Chung, Ladusaw, and McCloskey ), in which an explicit antecedent of the wh-phrase is missing. This research suggests that the structural parallelism of the antecedent and the ellipsis site affects the processing of the ellipsis site. They argue that in sluicing, the antecedent clause and the ellipsis site are structurally parallel because the antecedent clause contains an object NP, but in sprouting, they are not, because the projection of the object NP is missing in the antecedent clause. They tested the following paradigm in a selfpaced moving-window experiment. ()

a. The secretary typed something/quickly, but I don’t know what exactly. b. The secretary typed something/quickly, but I don’t know what she typed.

(i) a. *Who did John see him at the party and Mary see t_who at the office? b. ?Who did John see t_who at the party and Mary see him at the office? They further assume that the nearest conjunct can be determined in terms of the surface position of the conjunct. Thus in the case of sluicing in (), the second conjunct is the nearest conjunct. 9

There is a long-standing debate on whether gapping involves ellipsis or not. I leave this issue open (see Carlson et al.  and Johnson , this volume for detailed discussion).





In this paradigm, they compared the sluiced sentences and the non-elliptical sentences. They found that when the first clause (the clause before but) involves an implicit object (typed quickly), the reading of the region including the wh-phrase (what exactly/what she typed) was slower both in sluiced and non-elliptical conditions. Based on this finding they conclude that structural parallelism impacts the processing of an ellipsis site (similar results are reported in VPE contexts by Tanenhaus and Carlson ; Arregui et al. ; see also Kim et al.  on the discussion of voice mismatches in VPE). Furthermore, they emphasize that the effect of structural parallelism is not specific to ellipsis. Given the similar result between the sluiced and non-elliptical conditions, they argue that sluicing and non-elliptical wh-interrogatives involve the same syntactic structure and operations. Importantly, for our purpose, Dickey and Bunger’s study shows that the parser is sensitive to the structural details of the antecedent clause when the ellipsis site is processed. Their result supports the claim that the parser builds the structure of the antecedent site inside the ellipsis site, as argued for by Frazier and Clifton. The slower reading time of the sprouting wh-phrase is expected if the parser copies the structure of the antecedent into the ellipsis site. In the sprouting structure, the object NP position is missing in the antecedent clause. Thus, the parser needs to reanalyze the copied structure and add an object NP node, because the wh-phrase needs a position where it can receive its thematic role and form an interpretable wh-gap chain. In sluicing, because the object NP position is available in the antecedent clause, such reanalysis need not take place. Dickey and Bunger argue that this reanalysis process leads to the slower reading time of the wh-phrase in the sprouting structure. Extending these previous findings, Yoshida and colleagues asked whether the parser is sensitive to further structural details of the antecedent site when the ellipsis site is processed (Yoshida et al. a; Yoshida et al. b). First, Yoshida, Lee, and Dickey (b) asked if the parser is sensitive to structural constraints such as islands (Ross ) in the processing of ellipsis. They contend that if the effect of an island is seen during the processing of an ellipsis site, this would strongly support the position that the parser builds a sufficiently detailed syntactic structure in the ellipsis site to support island constraints. It has long been known that while sluicing is insensitive to islands, sprouting is sensitive to them (Chung, Ladusaw, and McCloskey ). For example, in (), when an indefinite NP (something) that serves as the antecedent of the wh-phrase is inside an island (clausal adjunct island), the example is acceptable. However, when the antecedent of the wh-phrase is an implicit object (i.e., in sprouting) and this implicit object is within an island, the example is not acceptable. ()

Nick’s father was startled [because he smoked something/*ø in the garden], but it wasn’t clear what precisely.

Roughly put, Chung, Ladusaw, and McCloskey’s argument is as follows. In () the antecedent involves an island structure. Under LF-copying theories, the whole structure of the antecedent is copied into the ellipsis site at LF and the wh-phrase and the implicit object inside the copied structure form a dependency. As a result, this dependency must go across the adjunct island boundary in (), resulting in the unacceptability due to an island violation. On the other hand, sluicing has different means to form a dependency that is not affected by islands.

 



Taking advantage of this island sensitivity of sprouting, Yoshida, Lee, and Dickey investigated whether the parser will build an island-violating structure inside an ellipsis site. They tested the sentences in () in a word-by-word self-paced moving paradigm. ()

a. Nick’s father discovered [that he smoked something/secretly in the garden] . . . b. Nick’s father was startled [because he smoked something/secretly in the garden] . . . . . . but it wasn’t clear what precisely he got out of smoking in hiding.

In (a), the indefinite NP (something) and the implicit object are in a non-island complement clause. In (b) they are embedded in an adjunct island (because-clause). In all of the conditions, the wh-phrase, what, is followed by a full clause. That is, none of the conditions involves ellipsis. Importantly, however, at the point of what precisely, each sentence is ambiguous between sluicing and non-sluicing. Faced with the ambiguity of the wh-phrase, does the parser choose one structure at the point of the wh-phrase, or does it delay structure building until decisive bottom-up information becomes available to disambiguate the structure? This can be tested by looking at the processing of the wh-phrases. A wh-phrase in a sprouting structure is known to be read slower than one in a sluicing structure (Dickey and Bunger ). Thus, if the parser chose the ellipsis structure at the point of the wh-phrase in (a), the wh-phrase should be read slower when the first clause involves an implicit object than when the first clause involves the indefinite NP something: the former leads to a sprouting structure and the latter leads to a sluicing structure. Dickey and Bunger found that the wh-phrase is indeed read slower in the implicit object condition than in the indefinite NP condition. The parser, therefore, picks the ellipsis structure at the point of the wh-phrase. Importantly, at the same time they found that this contrast in slowdown of the wh-phrase disappears in (b). They argue that the lack of reading time slowdown in the sprouting version of (b) means that the parser did not attempt to construct the structure of sprouting. This follows from the island sensitivity of sprouting plus the well-known observation that the parser does not attempt to form a wh-gap dependency across an island in online sentence processing (Stowe ; Traxler and Pickering ; Philips ). The difference between (a) and (b) is whether the first clause contains an adjunct island or not. Building a wh-gap chain in the implicit argument version of (b), unlike in the overt argument version, would result in an island-violating sprouting structure. To avoid the island-violating structure, the parser did not attempt to build a sprouting structure in (b). Thus the slowdown of the wh-phrase associated with the sprouting structure was not observed. This result suggests that the processing of the ellipsis site is sensitive to the detailed syntactic structure of the antecedent. Furthermore, this result is predicted by the copying approach whereby the parser copies the structure of the antecedent into the ellipsis site and builds silent syntactic structure within the ellipsis site. The parser avoids positing the structure of sprouting-type sluicing because the antecedent clause contains an adjunct island. This indicates that the parser refers to the antecedent clause and evaluates its structure. In terms of the copying approach, this means that the parser decides whether to copy the structure of the antecedent into the ellipsis site or not. If the result of the copying would lead to an island violation, the parser does not do so.





Adopting similar logic, Yoshida, Dickey, and Sturt (a) asked whether the hierarchical structure of the antecedent clause is recovered in the ellipsis site. They examined whether the effect of the c-command relation underlying anaphora resolution (Chomsky ) is seen during the processing of an ellipsis site. Using a word-by-word self-paced moving window paradigm, they examined sentences like (). ()

a. Jane’s grandfather/grandmother told some stories at the family reunion but we couldn’t remember which story about himself from the party his brother was so impressed with. b. Jane’s grandfather/grandmother told some stories at the family reunion but we couldn’t remember with which story about himself from the party his brother was so impressed.

In () the gender of the subject NP (grandfather/grandmother) is manipulated. Furthermore, in (a) the wh-phrase is an NP, but in (b), the wh-phrase is a PP. Yoshida, Dickey, and Sturt argue that at the point of the wh-phrase (which story about himself ), (a) is locally ambiguous between sluicing and a non-elliptical wh-question. But (b) is not ambiguous, because the wh-phrase is a PP, and there is no PP in the antecedent clause that can serve as the antecedent for the wh-phrase. Thus, it is possible that the parser predicts that (a) is sluicing at the point of the wh-phrase, but in (b), the parser does not consider sluicing as a possible structure. They further argue that if the parser picks the sluicing structure and if the structure of the antecedent is copied into the ellipsis site, the parser will try to link the reflexive to the potential antecedent. It has been reported that the parser tries to link a reflexive to its antecedent as soon as the reflexive is encountered (e.g., Sturt ). Furthermore, the parser will try to link the reflexive to a grammatically sanctioned antecedent, one which c-commands it and is located in the local domain of the reflexive (Sturt ). In (a), the subject NP in the first clause is the only possible antecedent for the reflexive if the parser is building the structure of sluicing. This is because the structure of the antecedent clause is copied and reconstructed into the ellipsis site, and in the reconstructed structure, the subject NP c-commands the thematic position of the wh-phrase, so the reflexive is c-commanded by the antecedent NP. On the other hand, in (b), a sluicing structure is not built and thus the subject NP in the first clause is not a candidate for the antecedent. Yoshida, Dickey, and Sturt found that the reflexive in (a) is read significantly slower when the subject NP in the first clause and the reflexive mismatch in gender, the Gender Mismatch Effect (GMME: Sturt ; Kazanina et al. ). The GMME is not found in (b). They interpret this result as indicating that the parser does indeed copy the structure of the antecedent into the ellipsis site and tries to link the reflexive to the antecedent. The GMME reflects the failure of linking the reflexive to the antecedent due to the gender mismatch. In this section, I have reviewed several studies on the processing of ellipsis. All of these studies suggest the following: First, the parser accesses the information of the antecedent when the ellipsis site is recognized. Second, the parser copies the structure of the antecedent into the ellipsis site. In other words, the parser is using the strategy of copying to rapidly resolve the ellipsis site. These are the answers to the first two questions (what structure the parser builds and how the parser builds the structure).

 



. W     ?

.................................................................................................................................. The discussion so far has revealed that the parser accesses and copies the structure of the antecedent site into the ellipsis when the ellipsis site is recognized and processed. Now we turn to the next question, namely when does the parser recognize ellipsis? Looking at environments where ellipsis is licensed, it is clear that these environments are compatible with both elliptical and non-elliptical structures. In other words, in many cases, the potential position of ellipsis shows local ambiguity. An example is an embedded wh-question. A fragment of a sentence like () is locally ambiguous between ellipsis, i.e. sluicing, and a non-elliptical wh-question. ()

John was eating something, but I don’t know what . . .

() can be continued as a non-elliptical wh-construction like ‘I don’t know what he likes’, or it can be sluicing, in which no overt material follows. The same holds for VPE and other types of ellipsis constructions. This means that the processing of an ellipsis construction has an aspect of ambiguity resolution. Given this ambiguity, there are roughly two possible strategies that the parser can employ. One is what we can call the delay strategy. In the delay strategy, the parser does not decide immediately whether the wh-phrase is part of a sluicing structure or a non-elliptical wh-question structure. In other words, the delay strategy dictates that the parser waits on its decision until decisive bottom-up information becomes available. For example, if the wh-phrase in () is followed by a subject NP, it is clear that the wh-phrase is part of a non-elliptical wh-question structure. The other strategy can be called the incremental strategy. In the incremental strategy, the parser decides whether the wh-phrase is the part of a sluicing structure or a non-elliptical structure immediately upon encountering the wh-phrase, without waiting for the later information. The work by Yoshida and colleagues that I discussed previously directly investigates this point. Recall the sentences that they tested in () and (). These sentences do not involve ellipsis. However, at the point of the wh-phrase these sentences are ambiguous between sluicing and a non-elliptical wh-question. They found that in such environments effects of islands and effects of c-command and binding occur at the point of the wh-phrase, but not later. This suggests that the parser decides that these sentences involve clausal ellipsis upon encountering the wh-phrase in an ambiguous context without waiting for the later information. Furthermore, Yoshida, Dickey, and Sturt’s study showed that the effect of binding and c-command is not found in an unambiguous context where the wh-phrase is a PP and is not compatible with a sluicing structure. These studies suggest that the parser decides that the ambiguous wh-phrase is part of a sluicing context only when sluicing is possible. In other words, the parser recognizes the ellipsis site whenever the ellipsis is grammatically possible and does not wait for bottom-up information later in the sentence.10 10

In Yoshida and colleagues’ studies, it is not clear why the ellipsis structure is preferred over a nonellipsis structure. This point requires further investigation.





Kaan and colleagues studied gapping constructions in an EEG experiment (Kaan, Wijnen, and Swaab ; Kaan et al. ) and reached a similar conclusion. Kaan, Wijnen, and Swaab () asked whether semantic incompatibility between the gapped verb and the object NP is detected immediately when the object NP is processed. They tested the following pair of sentences: ()

Ron took/sanded the planks for the bookcase, and Bill the hammer with the big head.

The verb took is semantically compatible with the NP the hammer; however, the verb sanded is not. Thus, if the parser inserts the elided verb immediately when the NP-NP sequence (Bill the hammer) is processed, the semantic incompatibility of the gapped verb and the object NP should be detected, and the signal of such incompatibility, the N effect, should be seen. They indeed found the N effect, and thus they conclude that the gap is recognized immediately when the NP-NP sequence is processed. In line with these studies, Yoshida, Carlson, and Dickey () asked when the ellipsis site in gapping is recognized. They point out that, like examples of sluicing, the environment where gapping is licensed can be compatible with a non-elliptical structure. For instance, the example in () involves the sequence of an NP the lawyer and a PP near the telephone pole. ()

The banker stood near the newspaper stand, and the lawyer near the telephone pole . . .

This NP-PP sequence can be analyzed as gapping in which the verb stood is elided. At the same time, this NP-PP sequence can also be analyzed as an NP that is modified by an adjunct PP. Yoshida, Carlson, and Dickey asked whether the parser decides that this NP-PP sequence has the structure of gapping or the structure of an NP with a PP modifier. Using a word-by-word self-paced moving-window paradigm, they examined the following sentences: ()

a. The banker stood near the newspaper stand, and the lawyer near the telephone pole . . . b. The banker stood near the newspaper stand, while the lawyer near the telephone pole . . . . . . stood with crossed arms.

In (a) the two clauses are connected by a coordination connective and, but in (b), they are connected by a subordination connective while. It is well known that gapping is not possible in a subordinated clause (Johnson  among many others). Thus, the NP-PP sequence can potentially be analyzed as gapping in (a), but not in (b). After the NP-PP sequence, a VP follows. When the verb stood is processed, it is clear that the NP-PP sequence must be an NP with a PP modifier. Yoshida, Carlson, and Dickey argue that if the parser employs the delay strategy, it can wait until the verb stood is encountered. On the other hand, if the parser employs the incremental strategy, it decides whether the NP-PP sequence has the structure of gapping or the structure of an NP with a PP modifier at the point of the NP-PP sequence. If the parser chooses the NP structure, the presence of the verb at the later

 



point is expected as the NP in that position is analyzed as the subject. The verb stood, therefore, should not surprise the parser. On the other hand, if the parser chooses the gapping structure, then the presence of the verb stood is not expected. This is because a coordinated or subordinated clause in sentences like () cannot have two verbs. As a result, the verb stood should surprise the parser if the gapping structure is chosen. However, the subordination condition (b) does not allow a gapping structure, and so in (b) the NP structure is the only possible structure. As a consequence, in (b), the verb stood should not surprise the parser. Yoshida, Carlson, and Dickey found that the verb stood in (a) was read significantly more slowly than in (b). They thus conclude that the parser employs the incremental strategy and picks the gapping (i.e., ellipsis) structure immediately when the NP-PP sequence is encountered. In summary, these ambiguity studies all suggest that the parser prefers an ellipsis structure when the grammar allows one. In the sluicing studies, when sluicing is grammatically possible, the parser chooses the sluicing structure at the point of the wh-phrase, and in the gapping studies, when gapping is grammatically possible (in a coordination context, not in a subordination context), the parser chooses the gapping structure. Together, these results suggest that the parser employs the incremental strategy for ellipsis resolution.

. W        ?

.................................................................................................................................. For the parser to successfully process sentences involving ellipsis, it has to search for an antecedent for the ellipsis and decide which constituent is the antecedent of the ellipsis. In simpler examples of ellipsis, finding the antecedent looks like a trivial matter. For example, in the sluicing example in (), the only possible antecedent for the elided clause ([IP ø]) is the whole IP in the first clause. Thus, in such a case, there is an unambiguous antecedent, and determining which constituent is the antecedent of the ellipsis should be easy. ()

[IP John was eating something], but I don’t know what [IP ø].

A somewhat more complex example such as () reveals that the parser can need to decide which of multiple IPs is the antecedent of the ellipsis site. ()

[IP Mary said that [IP John was eating something]], but I don’t know what [IP ø]. a. . . . but I don’t know what [IP John was eating]. b. . . . but I don’t know what [Mary said that [John was eating]].

In principle, both of the IPs can be an appropriate antecedent for the elided clause, and thus () is compatible with both of the interpretations in (a) and (b). Such ambiguity in the interpretation of the ellipsis site can be found in other ellipsis constructions as well. This ambiguity arises clearly because both the long interpretation and the short interpretation are allowed for ellipsis (see Merchant ; Lasnik  for related discussion).





The problem of antecedent search exhibited by examples like () and () still does not look like a big problem, as the search space is bounded by the already-processed context. The situation becomes more complicated, however, in the context of backward sluicing (Giannakidou and Merchant ) in which the ellipsis site precedes the antecedent clause, as in (). ()

I don’t know which book [IP ø], but [IP the professor was reading a book].

In the processing of (), at the point of the connective but it becomes clear that the first clause involves ellipsis of the clause following the wh-phrase. Once the parser recognizes the ellipsis site, it has to search for the antecedent of this IP. In an example like (), it is the clause that immediately follows the connective. However, a potential antecedent for the elided IP can be embedded, and at the point of the connective but it is impossible for the parser to decide where the antecedent IP is located. For example, in () there are multiple IPs that can serve as the antecedent for the elided IP, and the parser needs to decide which IP is the antecedent. ()

I don’t know which book [IP ø], but . . . a. . . . [IP the professor said that [IP the student was reading a book]]. b. . . . [IP the professor said that [IP the TA was angry because [IP the student didn’t read a book]]].

Facing this situation, there are at least two possible antecedent search strategies. The first strategy is the delay strategy. In many sluicing cases, the wh-phrase has an antecedent, normally an indefinite NP. For example, in (), the indefinite NP a book is the antecedent for the wh-phrase, which book. If the antecedent clause of sluicing normally contains an antecedent for the wh-phrase, then an indefinite NP can be a reliable cue to tell the parser that the IP that contains the indefinite NP is most likely the antecedent clause for the ellipsis clause. Thus the parser can rely on the presence of an indefinite phrase in identifying an antecedent clause. In this case, it waits for the indefinite NP to come into the input, and starts copying the materials in the antecedent IP into the ellipsis site only after recognizing the indefinite phrase. The delay strategy guarantees safe and accurate processing because it relies on the decisive bottom-up input in deciding which IP is the antecedent IP. The delay strategy, on the other hand, sacrifices rapid processing of ellipsis sentences. The second strategy is the incremental strategy. Looking at examples like () and (), it is clear that immediately after but, an IP follows. Thus it is possible that once the parser recognizes the ellipsis site by means of the connective but, it starts copying the materials that follow but, without any delay. Using this strategy, the parser can rapidly process ellipsis sentences, but at the same time, it is at risk of making a wrong decision, i.e., it is always possible that the IP following but is not an appropriate antecedent for the ellipsis site. So far, there have been two studies on the processing of backward sluicing (Gullifer ; Yoshida et al. a). Both of these studies support the incremental strategy. First, Gullifer () asked whether backward sluicing and non-elliptical wh-questions are processed similarly or not. He compared the following type of sentences in a phrase-byphrase self-paced moving-window paradigm:

  ()



a. I can’t remember what, but the fisherman fitted something to his boat upon arriving at Old Crystal Lake. b. I can’t remember what the fisherman fitted to his boat upon arriving at Old Crystal Lake.

Gullifer reports that the reading time of the region fitted something to his boat in (a) was slower than that of the region fitted to his boat in (b).11 This result, Gullifer argues, suggests that the processing of backward sluicing is more difficult than the processing of non-elliptical wh-questions. He contends that this difference is attributed to the difference in how these two constructions are processed. In backward sluicing, unlike in non-elliptical wh-questions, the parser has to bind the indefinite NP to the wh-phrase and copy the antecedent IP into the ellipsis site. On the other hand, in the processing of non-elliptical wh-questions, these processes are not involved. Gullifer suggests that these extra steps are the source of the processing difficulty of backward sluicing sentences. Additionally, Gullifer argues that the parser copies the structure of the antecedent IP as soon as it can identify the complete IP. Thus he basically argues for the incremental strategy. Even though Gullifer’s study is compatible with the incremental strategy, his result is also compatible with the delay strategy illustrated above. It is possible that the parser starts copying the materials in the antecedent IP only after the indefinite NP is processed. This point, however, cannot be tested in Gullifer’s experiment as the indefinite NP and other material are in the same critical region of interest.12 Yoshida et al. (a) approached the problem of antecedent search from a slightly different angle. First, Yoshida et al. point out that both backward sluicing and non-elliptical wh-questions involve a long-distance dependency between the wh-phrase and the verb. For example, in (), the wh-phrase what in both of the constructions must be linked to the verb fitted because the wh-phrase receives its thematic role from the verb. Furthermore, in the sentence-processing literature, it is well known that the parser tries to actively link the whphrase to the closest verb (Pickering and Barry ; Traxler and Pickering ; Phillips , among others). Thus, if the verb closest to the wh-phrase is semantically incongruent with the wh-phrase, the verb gives rise to a reading time slowdown. Given that backward sluicing is a kind of wh-question, the processing of backward sluicing should involve a dependency formation process between the wh-phrase and the verb. Yoshida et al. argue that if the parser employs the delay strategy, this wh-verb dependency formation process should also be delayed. In the delay strategy, therefore, the semantic incongruence between the wh-phrase and the verb that is closest to the wh-phrase should not matter. On the other hand, if the parser employs the incremental strategy, the wh-verb dependency formation process should be launched immediately after the wh-phrase is encountered. Therefore, the semantic incongruence between the wh-phrase and the verb should surprise the parser, and cause a reading time slowdown at the incongruent verb. To test these two possibilities,

11

Note, Gullifer’s () experimental paradigm is a little bit more complicated, and here I am simplifying his study so that I can highlight its main finding. For the details and the full paradigm of this experiment, see Gullifer (). 12 Note that Gullifer’s main interest is whether the processing of backward sluicing is similar to that of non-elliptical wh-questions. Thus, the problem of antecedent search is not the main point for him.





Yoshida et al. examined the following type of sentences in a word-by-word self-paced reading paradigm: ()

a. I don’t know which book/author, but the editor notified the publisher about a new book/author. b. I don’t know which book/author the editor notified the publisher about.

(a) is the backward sluicing condition. (b) is the non-elliptical wh-question condition. In both conditions, the type of wh-phrase is manipulated. Which book is not compatible with the verb notify, but which author is. In (), if the parser employs the incremental strategy, it tries to link the wh-phrase to the verb. Thus, the verb should be read significantly slower when the verb and the wh-phrase are semantically incompatible than when they are semantically compatible. On the other hand, if the delay strategy is employed, such a reading time difference is not expected. Yoshida et al. found that the verb, when semantically incongruous, is read significantly slower both in the backward sluicing condition and in the wh-question condition. This result supports the incremental strategy. Furthermore, the observation that the parser is sensitive to the lexical content of the verb supports the copying strategy of ellipsis resolution. These two backward sluicing studies suggest the answer to the fourth question (Where does the parser find the antecedent of ellipsis?): the parser tries to find the antecedent IP immediately after the connective but, so the parser locates the antecedent IP in the closest grammatically possible position. Moreover, these findings indicate that the parser employs an incremental copying strategy in resolving ellipsis during online sentence processing, beginning the copying process before all the material to be copied is encountered.

. C

.................................................................................................................................. In this chapter, I have tried to survey some recent psycholinguistic studies on ellipsis. I have tried to show what the psycholinguistic study of ellipsis can tell us about the mechanism of online sentence processing. The studies I have discussed in this chapter have revealed the following properties of the mechanism of online sentence processing (which we can call parsing strategies). First, the parser builds the structure of ellipsis sites by copying the structure of the antecedent site. The copying strategy properly captures the parser’s sensitivity to the structural (and lexical) details of the antecedent during the processing of the ellipsis site. Second, the parser recognizes the ellipsis site immediately when it recognizes a structure in which ellipsis can be licensed. Thus, the parser processes the ellipsis site incrementally without delay. Finally, we have seen that the parser looks for the antecedent in the position closest to the ellipsis site. The backward sluicing studies suggest that the parser decides that the clause that is closest to the ellipsis site is the antecedent of the ellipsis site without referring to the information in the later portion of the sentence. In general, these findings support the claim that the parsing strategies that the human parser employs achieve incremental, rapid, and grammatically detailed structure building. Thus, the study of online processing of ellipsis constructions points to a conclusion that is in line

 



with those reached in the sentence-processing studies in other domains. For example, processing studies of cataphoric constructions have shown that the parser tries to link a pronoun to its antecedent, and the process of linking the pronoun to the antecedent obeys grammatical constraints such as Binding Conditions (Kazanina et al. ; Yoshida et al. ). Similarly, studies of filler–gap dependencies have revealed that the parser tries to link wh-fillers to the closest gap site as long as the gap is not embedded inside an island (Stowe ; Traxler and Pickering ; Phillips , among others). The study of the online processing of ellipsis can provide further support for the position that the parser builds grammatically detailed structure rapidly and accurately (see Phillips ; Phillips and Lewis  for a detailed discussion on this point). Finally, I would like to discuss what remains in the study of ellipsis processing. The study of ellipsis processing has been developed mostly based on research of clausal ellipses, such as sluicing (see Harris ; Harris and Carlson , which investigate a different type of clausal ellipsis construction—the let alone construction). Therefore, there are not many studies on other types of ellipsis constructions. For example, the processing of NP-ellipsis has not been investigated. One of the open questions is whether the same mechanism is responsible for the online processing of clausal ellipsis constructions and other ellipsis constructions. Some studies on sluicing (Yoshida, Dickey, and Sturt a; Yoshida, Lee, and Dickey b) have suggested that ellipsis is preferred when ellipsis and non-ellipsis are both possible. Does the same hold true in other ellipsis constructions? Why does the parser prefer to pursue ellipsis when other structures are possible? Pursuing these questions can shed new light on the study of ellipsis processing.

A I am most grateful to Michael Frazier for his valuable discussion of and comments on an earlier version of this work. This work has been supported in part by NSF grant BCS- awarded to Masaya Yoshida. All the remaining errors are, of course, my own.

  ......................................................................................................................

 ......................................................................................................................

 ́  -   . 

. I

.................................................................................................................................. M people in the world are bilingual (Grosjean ; Romaine ). Very often, when two individuals of a bilingual1 speech community speak to each other, they switch from one language to another effortlessly back and forth. This phenomenon is known as codeswitching or codemixing (Muysken ; Edwards ). Broadly speaking, codeswitching refers to the simultaneous use of two or more languages by fluent bilinguals within a discourse (Edwards ). A Taiwanese/Spanish example of codeswitching can be found in ():2 () Gua chia el postre que hizo mi I eat the dessert that made my ‘I eat the dessert my grandmother made.’

abuela. grandmother (Bartlett and González-Vilbazo )

Codeswitching is to be differentiated from other types of language contact phenomena, such as borrowing, the integration (morphological, syntactic, and phonological) of loanwords into a recipient language (see Poplack  and Romaine  for further discussion). Further, codeswitching can be categorized as inter-sentential, i.e. involving switches between clauses, or intra-sentential, i.e. switches within a single clause. Competent codeswitchers have the ability to judge the acceptability of any codeswitched utterance, in a manner similar to how monolingual speakers can discriminate between constructions that are part of their I-language and those that are not. Put in another way, codeswitching is not random, but reflects the linguistic competence of codeswitchers. This in turn has motivated much theoretical work on codeswitching, for as Woolford () noted, linguistic theory should account for codeswitching as just another expression of the faculty of language,

1

While we will use the terms bilingualism and bilinguals throughout just for ease of exposition, note that what is asserted includes multilingualism and multilinguals as well. 2 As per the conventions in the literature, we will mark codeswitches with italics.





while at the same time codeswitching can provide a rich new source of data against which to test existing theories. Indeed, several models have been proposed to explain the constraints on codeswitching and account for much data, particularly on intra-sentential codeswitching, given how linguistic theory is mostly concerned with the grammatical relations within a clause. One phenomenon that has received much attention in linguistic theory is ellipsis. To this day, however, very little work has been done on elliptical constructions under codeswitching. This is unfortunate on two counts. First, because codeswitching can and should inform our theories on ellipsis, just in the way Woolford argued. Second, because ellipsis data may contribute to a better understanding of the ways in which codeswitching is possible, even at the syntactic level prior to any phonology. In this chapter we review the work on ellipsis under codeswitching to date to show that (i) ellipsis under codeswitching exhibits several of the connectivity effects well known in the ellipsis literature since Ross (b), which we argue is further evidence not only for a deletion theory of ellipsis but for a hybrid identity requirement (see Chung ), and (ii) these data can naturally be accommodated in a constraint-free approach to codeswitching, without the need to posit codeswitchingspecific constraints. The chapter is organized as follows: we first provide a historical overview of the literature on codeswitching, focusing on some of the most important theories proposed to date. We then examine the studies that have been pursued on ellipsis under codeswitching, and we illustrate how a constraint-free approach can account for the data. Finally, we conclude with a brief discussion on possible future directions of research. Before we review the particular theories, we believe a note on the methodology of codeswitching research is warranted. Many methodological challenges are unique or exacerbated when studying codeswitching. Some of these challenges include what to count as codeswitching, how to select participants (i.e. what constitutes a bilingual codeswitcher), the types of tasks to perform, how to design the stimuli, whether participants are in what is known as the bilingual mode, the type of data to collect among others (see Grosjean ; MacSwan , , ; Poplack ; and MacSwan and McAlister  for relevant discussion on these issues, and González-Vilbazo et al.  for potential solutions to some of these challenges). We take the position that while the issues are real, they are not insuperable, and the potential rewards from studying codeswitching outweigh the difficulties.

. T  

.................................................................................................................................. The study of codeswitching has progressed from a primary focus on social and pragmatic aspects to a much more comprehensive view, where theoretical attempts to understand its underlying mechanisms inch closer to, and can borrow from, more general considerations on linguistic theory. This was highly implausible fifty years ago. In fact, early discussion about codeswitching suggested no restrictions on where a switch might occur. As Pfaff () noted, the idea that the alternation between languages was not rule-governed, suggested by researchers such as Labov () and Lance (), endured well into the s. It was nevertheless during this decade that important fieldwork on codeswitching (Timm ; Gumperz ; Wentz and McClure ; McClure , among others)



́-  

demonstrated that codeswitching was not random but rather constrained, both at the sociolinguistic level and also at the points at which a switch could occur within an utterance. An example of the extra-syntactic significance of codeswitching can be seen in Gumperz (), which examined how in some Spanish/English communities, Spanish had become the language for affective expression, while English was used to communicate in a more objective or impersonal manner. And although works like these were mostly focused on the sociolinguistic role of codeswitching, as MacSwan () observes, they nevertheless made important descriptive observations about some types of switches which bilinguals rejected. These restrictions were language-pair-specific, e.g. Timm () reports that Spanish/English switches between a personal pronoun and a finite verb are strongly restricted (). () a. *Él works too much. he ‘He works too much.’ b. Ese hombre works too much. that guy ‘That guy works too much.’3

(Timm )

Similarly, Wentz and McClure () and McClure () investigated the use of codeswitching by children in a Spanish/English bilingual community and observed that codeswitching could function as an index of identity and could allow bilinguals to use different languages for different situations or to mark different discourse styles. These authors were also the first ones, to our knowledge, to note that codeswitched elliptical constructions could also be part of the grammatical inventory of bilinguals, as in some of the answers reported in the following dialogues: () A: ¿Quién quiere grape? who wants B: I do! / Me! () A: ¿Quién tiene hambre? who has hunger ‘Who’s hungry?’ B: I am / I’m hungry / *I do / (*I have).

(adapted from Wentz and McClure )

Note that while in () consultants could respond in English in two different ways to the question in Spanish posed by speaker A, Wentz and McClure reported that in () bilingual children responded in three different ways: I am, I’m hungry and I do. But adult bilinguals consulted later only considered two of those responses acceptable, namely I am and I’m 3

For a more complete review on the question of codeswitches of pronouns, see González-Vilbazo and Koronkiewicz ().





hungry. The response I have, suggested to adults by the researchers, was not accepted by adults and not produced by any of the children. We return to these findings in more detail in the following section. What is important to note at this point, is that constraints on the possible responses for these elliptical constructions under codeswitching exist, dispelling the idea that codeswitching is random. Subsequent accounts have tried to capture the codeswitching data in a more systematic and generalizable way, suggesting rules not as mere descriptions of the data but rather as being fundamental parts of a more general theory of codeswitching (cf. MacSwan  for discussion). Some of these theories were based on word order and were very much specific to codeswitching. Poplack (), for instance, proposed that codeswitches must respect the word order of both languages, as well as a general restriction against codeswitching below the word level, i.e. no intra-word codeswitching. In this view, codeswitching seems to be understood as a linear phenomenon as in () (where > represents a sequencing operator and / a switch between languages): () L > L > L > L > L/

L > L/

L > L > L > L/

L > L (González-Vilbazo and López : )

While this theory set out a more ambitious goal, that of explaining codeswitching in more general terms rather than just describing the data, several counterexamples in the literature exist, where neither constraint is violated and the utterance is nevertheless unacceptable. Consider the Spanish/English examples in (): () a. *The students had visto la película italiana. seen the movie Italian b. *Los estudiantes habían seen the Italian movie. the students had ‘The students had seen the Italian movie.’

(Belazi, Rubin, and Toribio )

Both sentences in () only involve a switch between Aux and V, and the word order of both languages is respected, and yet they are both rejected, contra Poplack’s constraints. Further, Poplack’s constraints are specific to codeswitching and thus involve essentially a third grammar, i.e. rules that are not part of any of the grammars of the languages involved in the codeswitch. After Poplack’s seminal work, newer approaches incorporated hierarchical relations into their descriptive framework. These theories rely on concepts that at the time were considered to be part of UG, such as government or functional head selection. Di Sciullo, Muysken, and Singh () proposed an anti-government constraint for intra-sentential codeswitching, so that complement DPs must be in the same language as their governors, be they verbs or prepositions. Conversely, Belazi, Rubin, and Toribio () appeal to feature checking theory to propose that functional heads need to check, among other features, a language feature against their complements. These approaches greatly contributed to the progress of codeswitching research. Despite their theoretical appeal, however,



́-  

these theories made the wrong predictions. Two naturalistic examples from a German/ Spanish codeswitching corpus are given in () and () (González-Vilbazo ): () No estoy sicher, si nächste Woche habrá Schulfest not am sure if next week be school.party ‘I’m not sure if there’s a school party next week or not.’ () Me ha dicho que ha verkauft me has said that has sold ‘She’s told me that she sold the car.’

el the

o or

no. not

coche. car

The German/Spanish example in () would violate Di Sciullo et al.’s anti-government constraint, since the verb and its complement are in different languages, and yet the utterance is acceptable. On the other hand, a theory like Belazi et al.’s would have difficulty accounting for an example such as (). This sentence contains a switch between a functional head (Aux) and its complement VP, and would thus be predicted by Belazi et al. to be ungrammatical, contrary to fact. Another line of approach to codeswitching proposes an asymmetry between the languages involved. These theories (Joshi , Klavans , Myers-Scotton , Myers-Scotton and Jake , and subsequent work) rely on the idea that a codeswitched utterance reflects an asymmetry between a matrix language ML, from which the template for the utterance is derived, and an embedded language EL, which only inserts lexical items into the utterance in the ML structure. The model known as the Matrix Language Framework, MLF (see MyersScotton  and Jake, Myers-Scotton, and Gross ) is perhaps the best known of these approaches. The MLF model is purported to be a model of competence, and it aims to account specifically for intra-sentential codeswitching, seen as the only instance of true codeswitching. This constraint-based model aims to find the conditions limiting when codeswitching can occur. While there have been different variants of the theory (see Myers-Scotton and Jake  for one of the latest iterations), it has been a constant that these conditions can be reduced to a typology of morphemes: content morphemes and system morphemes; and within system morphemes, a subtype known as outsider late system morphemes (case-marking, verb agreement) determine the ML and the grammatical properties of the clause.4 The elements from the EL would be inserted in the structure provided by the ML. There is much discussion of the model and some empirical evidence against it in the literature (for an extended discussion on the MLF, see Jake, Myers-Scotton, and Gross , ; MacSwan a, b; and González-Vilbazo and López ). It nonetheless remains difficult to conceive of terms such as ML or EL islands—typically adjunct phrases well-formed by EL grammar—as anything but codeswitching-specific. Indeed, one key question that constraint-based proposals raise is whether these constraints can be derived from more basic grammatical principles. 4 In the MLF, system morphemes are further divided into two types: early system morphemes and late system morphemes. “[Early system morphemes] are divided from late system morphemes because, along with content morphemes, they are conceptually activated to express a part of speakers’ meanings that they wish to communicate” (Myers-Scotton and Jake ). On the other hand, “late SMs [system morphemes] build hierarchical grammatical structure rather than enlarging or restricting the core meanings of referents” (Myers-Scotton and Jake ).





We should favor simpler theories; ideally, already established theories for monolingual data would be extended to also cover the facts of codeswitching. If that is not possible, codeswitching data force us to revise current theories in order to account both for monolingual and bilingual data. Most recent generative codeswitching theories have aimed at explaining codeswitching without appealing to any third grammar factor.5 For instance, MacSwan (, , ) and González-Vilbazo and López (, ) made two such proposals with no codeswitching-specific rules. Following the minimalist framework as laid out in Chomsky (, , et seq.), these authors argue that only the interface conditions have to be met by the derivation of a codeswitched utterance in order for it to be grammatical. It is worth emphasizing, especially when studying bilingualism, that the term ‘language’ is not a primitive of linguistic theory, but a social, political, and/or cultural construct. Each speaker has individual competence. To the extent that competences overlap, this overlap is often referred to as a ‘language’. Given this, what does it mean to be bilingual, i.e., to speak two (or more) languages? What makes the difference between a monolingual and a bilingual competence? A possible answer would be to say that a bilingual has two grammars. This is still too vague. Do we mean syntax, morphology, semantics, and/or phonology? Rules, operations, lexica? Whatever is a universal property of human languages cannot make the difference between mono- and bilinguals. We have to look for the difference in those areas where we find variation across speakers. Current advances in generative linguistic theory take it that there is only one set of universal syntactic, morphological, and semantic operations, invariant across languages. Consequently, these operations/rules cannot constitute the difference, as both monolingual and bilingual speakers will necessarily have this set of operations as part of their linguistic system. There is, however, consensus among phonologists that phonological systems (e.g., constraint rankings) vary across languages. This would entail that bilinguals would have at least two phonological systems, whereas monolinguals only have one. Finally, the lexicon is of course another source of variation across languages. It is not surprising that most theories of codeswitching assume that bilinguals have two lexica. But this is not a logical necessity. It is possible to assume that bilinguals only have one lexicon (regardless of whether one assumes a DM-style lexicon or not) with lexical items (vocabulary items) from both ‘languages’. To the linguistic system it would make no difference whether the lexical items merged into a derivation are from what we call language X or Y. Summarizing, it looks like bilinguals have two phonological systems and possibly two lexica. Everything else is similar to the monolingual linguistic system.6 5 More recently, several theories of this sort have looked at different aspects of codeswitching, such as: codeswitching and language acquisition (Cantone ; Eichler ; Eichler and Müller ; Sunny Park-Johnson ), clausal structure (Mahootian ; González-Vilbazo and Struckmeier ; González-Vilbazo and López , ; den Dikken and Bando Rao ), the DP (González-Vilbazo ; Liceras et al. ; Herring et al. ), pronouns (van Gelderen and MacSwan ; GonzálezVilbazo and Koronkowiecz ), wh-dependencies (Woolford ; Toribio and González-Vilbazo ), and classifiers (Bartlett and González-Vilbazo , ). 6 Of course, this does not deny typological differences in grammars in languages (e.g., V/non-V). Rather, it states that the grammatical components are the same for monolingual and bilingual systems. In this view, the differences would be encoded in features or feature combinations in the lexicon. For further discussion, cf. González-Vilbazo and López ().



́-  

If we take these ideas on bilingualism seriously, one could say there are very few truly monolingual speakers in the world, as most speakers will know at least two linguistic varieties (either registers, dialects, or languages). If we stick to the idea that a speaker’s competence is individual, then there is no difference in principle between language, dialect, or register for the linguistic system. This is not to say that we cannot distinguish those concepts and they have certainly proven to be fruitful in sociolinguistics, but they are extralinguistic concepts. For the linguistic system there is no difference between these concepts. And everybody is competent in different linguistic varieties (language, dialects, registers). That means that almost everybody has two phonologies (to the extent that the dialects or registers they speak have different phonologies) and they certainly have two or more lexica (or an enlarged single lexicon). If this is true, these speakers are bilinguals. Thus, true monolingualism would represent the marked, not the unmarked, case. Based on these considerations and the research mentioned above, we take it that codeswitching is a natural instantiation of the faculty of language. That is, it is binary and hierarchically structured, and it undergoes the same universal linguistic operations (Merge, Agree) as monolingual derivations. In the generation of a codeswitched sentence, lexical items are drawn from two or more lexica,7 then merged into the derivation, and finally, the resulting structure is transferred to the interfaces, as represented in (). ()

Structural representation of a codeswitched utterance CP

C'

Spec

TP

C° [l1]

T'

Spec [l2]

vP

T° [l1]

v'

Spec [l2]

VP

v° [l1] Spec [do l1]

V'

XP V° [io l2] [l2] (González and López : ) 7 There is a continuous debate about the number of lexica in bilingual linguistic systems. Den Dikken () argues for one lexicon for bilinguals, while the vast majority of the literature argue for two (Woolford ; MacSwan ; Bartlett and González-Vilbazo ).





This is just the same as for a monolingual structure; the derivation will crash if the conditions of the interfaces are not met, be it a monolingual or a bilingual structure.

. W     

.................................................................................................................................. Ellipsis is a widely studied phenomenon in linguistics, as this volume can attest. Broadly, ellipsis refers to structures in which a part of a sentence appears to be missing. () a. b. c. d.

John can play something, but I don’t know what. John can play the guitar, and Mary can, too. John can play five instruments, and Mary can play six. A: Who can play the guitar? B: (Not) John.

(sluicing) (VP-ellipsis) (NP-ellipsis) (fragment answer) (Merchant : )

Ellipsis can apply to clausal structures as in sluicing (a), which is the deletion of a full interrogative clause with the exception of the wh-phrase; it can apply to verbal phrases (b) as well as inside the nominal domain (c); and it can also result in just one constituent conveying propositional information, such as in fragment answers (d). A thorough review of the literature on ellipsis is well beyond the scope of this chapter, but we refer readers to Merchant (), van Craenenbroeck and Merchant (), and Merchant (this volume) for recent overviews of some of the questions raised and tools used for the study of ellipsis. Chung, Ladusaw, and McCloskey () distinguish three different approaches for the study of ellipsis: a pragmatic approach, a copying approach, and a deletion approach. The pragmatic approach would propose that the ellipsis lacks any unpronounced structure and its interpretation depends only on pragmatic inference; the copying approach posits that the ellipsis site is empty at the time of the derivation but its content can be recovered by means of copying an already constructed structure at the level of interpretation. Finally, in the deletion approach, at some point of the derivation the ellipsis has full syntactic structure, which is then deleted or made inaccessible to PF under certain conditions. Currently, the copying and deletion approaches are favored. In a copy theory of ellipsis, or non-deletion theory, the structure of the ellipsis site amounts to a null element (or more than one) at the level of syntax, which can then either be given a fuller structure so it can be interpreted (as in Chung, Ladusaw, and McCloskey ), or resolved post-syntactically by an interpretation algorithm, similar to anaphora (Hardt ). In deletion approaches, the ellipsis contains full syntactic structure, as a corresponding non-elided sentence would, and the structure is then subject to a deletion operation, either at Spell-Out or at PF. Representative examples of deletion theories are Ross (b), Sag (a), and Merchant (), among others. Note that, both for approaches positing a deletion mechanism for ellipsis and for those adopting a copy system, one of the main questions is what type of relation of identity between antecedent and ellipsis must hold for the unpronounced material to be



́-  

recoverable. Put another way, the question is whether the relation between antecedent and ellipsis need only be a semantic one (as proposed in Merchant ), a syntactic one (as in Saab , or Thoms ) or whether a hybrid theory requiring semantic as well as other types of identity is necessary (as in Chung , ; Merchant d; González-Vilbazo and Ramos ). Codeswitching can be illuminating when answering this question. To our knowledge, there are only a few studies that look at ellipsis in codeswitching: Wentz and McClure (), Nee (), González-Vilbazo and Ramos (), and Merchant (b). We can say that all studies so far have provided evidence against both the pragmatic approach and the LF-copying approach and support for a deletion approach to ellipsis. Furthermore, within the deletion approach, some of the most recent literature provides evidence that the identity between the ellipsis and its antecedent must be morphosyntactic as well as semantic. Below, we discuss each study and how it relates to different approaches to ellipsis. Wentz and McClure () first reported elliptical constructions under codeswitching in a Spanish/English bilingual community. The data stem from interviews with bilingual children, and their responses were later presented to adult bilinguals for verification. Some of their findings are shown below: ()

A: ¿Quién tiene hambre? who has hunger ‘Who’s hungry?’ B: I am / I’m hungry / *I do / (*I have).

()

A: Who are you calling? B: Al doctor / *El doctor. the. doctor the. doctor

In () we see that the question ‘who’s hungry?’ in Spanish is formed with the equivalent of have and not with a copular verb. I am was accepted both by children and adults. It seems to be the elliptical result of the English sentence I am hungry, both of which (I am/ I’m hungry) were deemed acceptable by children and adults. I do, only accepted by children, is unlikely to be the result of an unelided English counterpart, since I do be hungry or I do hunger would be a heavily marked construction. The answer I have was not produced by any of the interviewees and was deemed unacceptable by adults. Similarly, in (), children offered two very similar Spanish answers to the English question, the only difference being the case of the DP. Adults only accepted the answer bearing overt accusative case. From these judgments, Wentz and McClure concluded that children may take longer to develop the ability to codeswitch, thus explaining the difference between children and adult answers, and thus they deemed adult judgments to be more reliable. Furthermore, the authors concluded that the fragment answer must obey the grammar of its own language, regardless of the language of the question. Unfortunately, they provided no further analysis beyond this generalization. The authors further suggested that an account such as Morgan’s () might be on the right track. Morgan () proposes that the missing part of a clause in a fragment must be present at one point in the derivation, after which it is deleted. It is worth noting, however, that the bilingual children interviewed seemed to more readily





provide a wider range of possible answers, whereas the adult bilinguals tended to narrow down the array of acceptable answers when compared to the bilingual children interviewed. For over thirty years, no further investigation of ellipsis in codeswitching was carried out. It was only in  that González-Vilbazo and Ramos reconsidered the topic following a suggestion by Jason Merchant. González-Vilbazo and Ramos () conducted an experiment to test connectivity effects on German/Spanish codeswitching, specifically case mismatch resolution. In order to test these connectivity effects they used codeswitched sluicing data. Both Spanish and German show case effects in sluicing: ()

Juan amenazó a alguien pero no sé {*quién / a quién}. Juan threatened someone. but not know. who. who. ‘Juan threatened someone but I don’t know who.’

()

Juan hat jemandem gedroht aber ich weiß nicht {*wer / *wen Juan has someone. threatened but I know not who. who. / wem}. who. ‘Juan threatened someone but I don’t know who.’

In some instances, a Spanish verb () assigns a different case to its object than the equivalent German verb does (). Take, for instance, the verb threaten in these languages. German drohen assigns dative to its complement while Spanish amenazar assigns accusative. Now, consider a codeswitched utterance with a Spanish antecedent clause and a full non-elided embedded clause, as in ()–(). In this case, a German wh-phrase will bear the case that the verb in the embedded clause assigns, dative for German () and accusative for Spanish (), regardless of the verb of the antecedent clause. ()

Juan amenazó a alguien aber ich weiß nicht {*wen / wem} er Juan threatened someone. but I know not who. who. he gedroht hat. threatened has ‘Juan threatened someone but I don’t know who he threatened.’

() Juan amenazó a alguien aber ich weiß nicht {wen / *wem} Juan Juan threatened someone. but I know not who. who. Juan amenazó. threatened ‘Juan threatened someone but I don’t know who he threatened.’ If the sentence is sluiced, however, the wh-phrase can only bear accusative, which in this instance is the case the Spanish antecedent verb assigns: ()

Juan amenazó a alguien aber ich weiß nicht {wen / *wem}. Juan threatened someone. but I know not who. who. ‘Juan threatened someone, but I don’t know who.’



́-  

This result strongly suggests that the verb elided must be Spanish amenazar. This is puzzling, as () shows that the non-sluiced embedded clause can in principle contain the German verb drohen. Why then is it not possible to just delete in ()? This would result in a sentence like () with the wh-phrase bearing dative case: ()

*Juan amenazó a alguien aber ich weiß nicht wem . Juan threatened someone. but I know not who. he threatened has ‘Juan threatened someone but I don’t know who .’

If a semantics-only theory of ellipsis (as in Merchant ) were correct, a sentence like () should be grammatical, as the antecedent and the ellipsis are semantically identical. However, () is ungrammatical. González-Vilbazo and Ramos argue that a semanticsonly theory of ellipsis is insufficient to capture the data, for () and () are semantically equivalent in terms of truth conditions. Instead, the authors claim that alongside a semantic identity condition, at least some morphosyntactic identity has to hold. More precisely, for these data, at least the case assigner of the wh-phrase has to be identical to the case assigner of its correlate in the antecedent clause. This analysis points in the same direction as other hybrid theories such as Chung’s () ban on eliding words new to the derivation, Chung’s () case and argument structure constraints, or Merchant’s (d) structure constraint. In Chung (), besides the semantic conditions, it is proposed that words which have no overt correlate in the antecedent clause cannot be left inside the ellipsis site. This proposal aims to explain, among other things, cases of sprouting: instances of sluicing where the wh-remnant has no overt correlate in the antecedent clause. In English, for instance, P-stranding, otherwise allowed (), is prohibited in these constructions (). ()

a. They’re jealous of someone, but it’s unclear of whoi . b. They’re jealous of someone, but it’s unclear whoi .

() a. They’re jealous, but it’s unclear of whoi . b. *They’re jealous, but it’s unclear who(m)i . (Chung : ) According to Chung (), the ungrammaticality of (b) is explained by the fact that the preposition of does not have an overt correlate in the antecedent, and is thus unable to stay inside the ellipsis. Chung () and Merchant (d), for their part, argue that no argument structure differences are possible between the antecedent and the ellipsis site. For Chung (), semantic constraints do not allow the ellipsis to have a different argument structure than its antecedent clause. As for case, if the remnant of the sluice is a DP, its licensing head within the ellipsis must be identical to the licensing head of its correlate in the antecedent clause. Evidence for these restrictions is provided by some intransitive verbs in Chamorro that have derived transitive counterparts. Consider (): ()

a. Ekgu’ si Joe, lao ti hu tungu’ [håyi ha ekgu’i]. .jealous  Joe but not  know who? []. jealous.of ‘Joe is jealous, but I don’t know who he is jealous of.’

 b. *Ekgu’ si Joe, lao ti hu .jealous  Joe but not  ‘Joe is jealous, but I don’t know who.’

tungu’ know



[håyi __ ]. who? (Chung )

The verb ekgu’ is an intransitive predicate which can introduce an internal argument as a PP, though this argument may be omitted. Its counterpart ekgu’i is semantically similar yet it introduces its internal argument as a DP. Now, in (a) we have an unsluiced example of an embedded question with the transitive ekgu’i preceded by an antecedent with intransitive ekgu’. (b) shows how a sluice in this situation results in ungrammaticality, even though a semantic condition should be satisfied. Chung concludes that the ungrammaticality must lie in the argument structure difference between the two clauses. In Merchant (d), this structural constraint accounts for the difference in grammaticality that voice mismatches produce in VP-ellipsis and sluicing. In VP-ellipsis (a), unlike in sluicing (b), voice mismatches are allowed. () a. The janitor must remove the trash whenever it is apparent that it should be. b. *Joe was murdered, but we don’t know who. (Merchant d) The argument here is that in VP-ellipsis the ellipsis site is smaller than in sluicing. In VPellipsis (a), a VP, the complement of the head determining voice, is elided. In sluicing, a full clause, including the voice head, is elided. Because in sluicing the voice head is elided, no mismatch with respect to voice between antecedent and ellipsis is possible. In a similar vein, Nee () conducted research on Spanish/Zapotec codeswitching in Teotitlán del Valle, México. This language pair is interesting for the study of sluicing because some Zapotec transitive verbs () have a Spanish equivalent that selects a PP as its complement (). ()

Juan gunien María. Juan spoke María ‘Juan spoke to María.’

()

Juan habló con María. Juan spoke with María ‘Juan spoke with María.’

In () the Zapotec verb gunien selects a DP for its complement whereas the corresponding Spanish verb habló selects a PP as its complement (). Nee takes advantage of this difference between Zapotec and Spanish in constructing codeswitched sluicing utterances. () Juan habló con alguien, per kednanadia Juan spoke with someone but not.know.s ‘Juan spoke with someone, but I don’t know who.’

tu who

cuni with

. spoke



́-  

() *Juan habló con alguien, per kednanadia Juan spoke with someone but not.know.s ‘Juan spoke with someone, but I don’t know who.’ ()

tu who

. spoke with

Juany gunien, pero no sé quién . Juan spoke but not know.s who spoke ‘Juan spoke, but I don’t know who to.’

Nee analyzes example () by adopting Chung’s () lexical requirement. According to her, the embedded clause contains the Spanish verb habló and its PP complement tu cun. Notice that the Spanish verb hablar, which is in the antecedent, selects for a PP complement, which in this case is satisfied by the Zapotec PP cun tu.8 The wh-phrase must then move up to Spec CP along with its preposition. The TP is then elided. In this case, Merchant’s semantic condition, Chung’s lexical requirement, and González-Vilbazo and Ramos’s morphosyntactic condition are all observed. On the other hand, Nee takes () to be ungrammatical because neither Zapotec nor Spanish allows for P-stranding. In (), the antecedent clause contains the Zapotec transitive verb, and Nee argues, assuming Chung’s lexical requirement, the wh-phrase must be bare. This is so because the verb inside the ellipsis must be the Zapotec gunien, and thus selects a DP. This derivation would also satisfy all semantic, lexical, and morphosyntactic conditions mentioned above. As we can see from these examples, Nee also advocates for a deletion approach to sluicing. Moreover, she embraces the view that identity between antecedent and ellipsis goes beyond semantics, and must also include some sort of lexical identity. Merchant’s recent paper on codeswitching and ellipsis (b) looks into verb phrase ellipsis in Greek/English codeswitching. This paper is consistent with the studies described above and finds further evidence not just for structural identity, but also evidence that in some instances the elided structure could not be pronounced. The data in this study are Greek/English codeswitching dialogues uttered by balanced bilingual children and contain a Greek antecedent and a verb phrase ellipsis response in English. An example is shown below: () A: Píres tin tsánda mazí su? took. the bag with you ‘Did you take the bag with you?’ B: Yes, I did. In (), the elided VP in the response appears to be in English, for two reasons. First, the corresponding unelided sentence would be fully grammatical (a). Second, a Greek VP at this point produces an ungrammatical result (b,c).

8

Nee reports that in Valley Zapotec the prepositions are optionally inverted in interrogative contexts; it is only in sluicing contexts that P-inversion is mandatory.

 () a.



Yes, I did take the bag with me.

b. *Yes, I did píra take.... c. *Yes, I did pern take[.]

tin tsánda the bag

tin tsánda the bag

mazí mu. with me

mazí mu. with me

In (b), Merchant argues, the Greek verb píra cannot agree with the subject, since the auxiliary did effectively blocks agreement. The Greek pern (c), for its part, is a bare stem form that cannot remain unbound in Greek. However, if (a) were the underlying representation for B’s reply in (), this result would contradict González-Vilbazo and Ramos’s condition as well as Chung’s () lexical requirement and Chung’s () argument structure and case condition. Chung () would expect the ellipsis to be in Greek just as the antecedent, whereas Chung () and González-Vilbazo and Ramos () would expect at least part of the morphosyntax of the ellipsis to match the Greek antecedent. Other Greek/English data from Merchant (b) suggest that (a) cannot be the elided structure in the ellipsis in (). () a. Mother: Pinás? hunger.. ‘Are you hungry?’ b. Daughter: Yes, I do. The example in () is similar to one of the answers given in the English/Spanish example in () reported by Wentz and McClure, although one adult speaker later rejected it. The unmarked way of expressing hunger in Greek is via an intransitive verb (a), unlike in English where an adjective is used (I am hungry). In this case, if the underlying representation of the response in (b) were either *I do hunger or *I do (be) hungry, neither of these options would correspond to the common way of responding to this question in English: I am hungry. In fact, it seems to correspond to a verb phrase ellipsis response (I do) which suggests a non-copula verb is inside the ellipsis. Similar to the previous example, the codeswitched non-elided counterpart to (b) would also be unacceptable. ()

*Yes, I do pináo. hunger..

Merchant offers a detailed analysis within the Distributed Morphology framework in which the verbal complex cannot enter into a local relation with T, and as a consequence the verb is not inflected. This makes the root unpronounceable, as Greek lacks infinitives. Again, these data suggest that a semantic condition alone is not sufficient and a structural constraint is necessary, echoing González-Vilbazo and Ramos’s proposal. In another example, the impossibility of an English VP inside the ellipsis with a Greek antecedent becomes clearer.

 ()

́-   [Context: a son attempts to turn on the air-conditioning one morning] a. Mother: To proí ðe xriázete the morning  need.... klimatizmó. air-conditioning. ‘In the morning there’s no need for air-conditioning.’ b. Son:

Yes, it does!

c. Mother:

Éxi have.... ‘It’s a little cool.’

d. Son:

No, it doesn’t.

ðrosúla. coolness.

Let us focus on the instance of ellipsis in (d). As Merchant notes, in Greek the existential predicate is expressed via an impersonal éxo ‘have’, which takes an accusative object. This can be seen in the remark about the temperature in (c). Just like in the previous examples, the elliptical response in (d) is acceptable but it is not initially clear what the ellipsis must contain. Crucially, no unelided counterpart to this response, codeswitched or not, is grammatical.9 ()

a. *No, it doesn’t be a little cool. b. *No, it doesn’t have a little coolness. c. * No, there doesn’t be a little coolness. d. *No, there isn’t a little coolness. e. *No, it doesn’t éxi ðrosúla. have.... coolness. (Merchant b)

As seen earlier in () and (), the ellipsis cannot be in English if it has a Greek antecedent. This explains (a–d). However, it seems it cannot be in Greek either as (e) shows. Similar to (), Merchant proposes the following structure for the attested ellipsis in (d): ()

TP it

T′ doesn’t

VoiceP Voice

v

VP √ex

DP √ drosja (Merchant 2015b)

9

The editors point us to the well-known observation that ellipsis has been argued to repair otherwise ungrammatical sentences (cf. Richards ; Lasnik , ; van Craenenbroeck and den Dikken ; Merchant a). Indeed, Merchant also considers this, though operating “at the morphological level” (Merchant b).





This ellipsis contains the roots √EX and √ÐROSJA that correspond to the Greek antecedent. That is the only way to satisfy the structural condition on ellipsis, i.e., to have at least the same relevant morphosyntactic features in the antecedent and the ellipsis. However, these roots are unpronounceable, as mentioned earlier. Merchant shows that the Greek/English codeswitching data offer more evidence for an irreducible syntactic condition on ellipsis. He further shows that this condition is so strong that it must hold even if the structure within the ellipsis would be otherwise unrealizable. This would thus be a new instance of elliptical repair,10 i.e., a situation in which the English answer to the Greek question as given in (d) is only possible with an elliptical construction. In this section, we have seen that codeswitching can unveil some of the inner mechanisms of ellipsis. It has been known since at least Chung () that a semantics-only condition for ellipsis still faces challenges (specifically for sluicing). However, as shown by Nee (), González-Vilbazo and Ramos (), and Merchant (b), codeswitching offers some of the strongest evidence yet that a theory of ellipsis requires some morphosyntactic condition as well, beyond whatever semantic conditions must hold. Further, codeswitching (Merchant b) also offers evidence of elliptical repair in previously unknown contexts.

. W     

.................................................................................................................................. One goal of this chapter is to show how ellipsis informs our understanding of codeswitching. Indeed, the research done so far already suggests surprising insights. First, it is possible to identify the language of the material within the ellipsis. This is shown by the case of the German wh-remnant in the data from González-Vilbazo and Ramos (), as well as the fact that ellipsis is what seems to cause the ungrammaticality of (), repeated as (b). ()

Juan amenazó a alguien aber ich weiß nicht {*wen / wem} er Juan threatened someone. but I know not who. who. he gedroht hat. threatened has ‘Juan threatened someone but I don’t know who he threatened.’

() Juan amenazó a alguien aber ich weiß nicht {wen / *wem} Juan Juan threatened someone. but I know not who. who. Juan amenazó. threatened ‘Juan threatened someone but I don’t know who he threatened.’ 10 An editor wonders if any insight could be gained from studying elliptical island repair on codeswitching. This is very possible, although it would be necessary first to look into island effects in codeswitching without ellipsis. To our knowledge, no such study exists. Looking into codeswitching between two languages, which differ with respect to their island effects, we would expect to gain insights into the elements that play a role in licensing those effects.



́-  

() a. Juan Juan

amenazó a threatened

alguien someone.

aber but

ich I

weiß know

nicht not

wen. who.

nicht not

wem. who.

‘John threatened someone, but I don’t know who.’ b. *Juan amenazó a alguien aber ich weiß Juan threatened someone. but I know ‘Juan threatened someone, but I don’t know who.’

Given the remnant’s case, we can infer that the verb in the ellipsis has to be Spanish; as we have seen, the Spanish verb amenazar ‘threaten’ assigns accusative, thus wen, whereas German drohen assigns dative (wem). Hence the ellipsis is in Spanish. The generalization that the ellipsis site has to be in the same language as in the antecedent is corroborated by the data from Merchant (), as seen in the previously discussed examples () and (). Further, what these results also indicate is that codeswitching can occur inside the ellipsis site. In all the data from González-Vilbazo and Ramos () and some of the data in Nee (), some element, the remnant, has been moved out of the ellipsis site prior to deletion. That in turn means that within the elided structure in (), we find a Spanish verb amenazar and the lower copy of the German wh-phrase. This is a new observation. Finally, ellipsis provides further evidence that surface realization of structures plays a role in codeswitching. Take the structure in (). This structure contains a switch between English it doesn’t and Greek exi drosula. Although the switch is perfectly grammatical, it is only so if the Greek verb phrase is not realized overtly. It appears that the codeswitched utterance is structurally fine but the surface realization of the VP is unacceptable for PF reasons. This problem is solved by eliding the Greek verb phrase and thus not realizing the codeswitch overtly. Before the recent studies on ellipsis in codeswitching, it could have been speculated that codeswitching would prove to be fundamentally different from monolingual cases in ellipsis. For instance, in a theory of codeswitching as a surface-oriented phenomenon, the lack of overt surface realization in ellipsis could have entailed the impossibility of ellipsis in a codeswitching context. The results reviewed in the preceding section show that this is not the case, and ellipsis can occur in the context of codeswitching and codeswitching can occur within ellipsis. Ellipsis is another phenomenon that shows that bilingual linguistic structures are in principle no different from monolingual structures. Eventually it comes down to features and operations on those features. Whether those features (or feature bundles) are extralinguistically assumed to belong to one language or another is irrelevant to the linguistic system in general and to the derivation in particular.

. O

.................................................................................................................................. As we have seen in this chapter, codeswitching has opened new avenues of research on ellipsis. At the same time, the extent to which ellipsis may be constrained in codeswitching remains unknown. More research is obviously necessary. Possible topics for further research are, among others, fragment answers and ellipsis in the nominal domain. Currently, an ongoing project (Ramos ) suggests that fragment answers in English/ Spanish codeswitching may be subject to the same constraints found in González-Vilbazo





and Ramos () for German/Spanish sluicing. Specifically, the morphosyntactic condition has to be observed with respect to the antecedent clause. In this ongoing study, constructions which would require a PP in one language but not in the other are investigated. () A: ¿Who did John marry? B: (a) mi hermana /  my sister

*con with

() A: ¿Con quién se casó with whom  married B: With my sister / my sister.

mi my

Pepe Pepe

hermana. sister el the

domingo? Sunday

The answer in () can be easily explained following the literature reviewed above. In this instance, the verb in the question, English marry, does not select for a PP, and the subsequent Spanish answer cannot be headed by a preposition. This is in line with previous findings, where the structure of the antecedent must match that of the ellipsis: the underlying structure left unpronounced in the answer must contain the English verb marry and not the Spanish verb casarse, which does select for a PP complement. In (), conversely, a new puzzle arises. The question contains a Spanish verb that selects for a PP. The possible codeswitched answers given, however, are an English bare DP or an English PP. The fact that a bare DP can be an acceptable response to the question in (A) could suggest that the Spanish verb casarse inside the ellipsis can select for an English bare DP. This, however, contradicts what we know of the Spanish verb casarse, i.e., that it necessarily selects for a PP complement. The puzzling alternative would be that the preposition has been stranded inside the ellipsis. This in turn also runs counter to what we know about Spanish prepositions, namely, that they cannot be stranded.11 Other future research involves the resolution of gender in NP-ellipsis in codeswitching. In German/Spanish codeswitching, the following examples are attested: () Der Tisch aus Bolivien y el __ de the. table of Bolivia and the. of ‘The table from Bolivia and the one from Germany.’ ()

Das Mädchen aus Madrid y la __ de the. girl. of Madrid and the. of ‘The girl from Madrid and the one from Barcelona.’

Alemania. Germany

Barcelona. Barcelona

In (), the elided noun appears to be masculine, which can be inferred from the remaining Spanish determiner. The German antecedent is also masculine, unlike the Spanish equivalent noun mesa, which is feminine. Consequently, the elided noun would have to be

An anonymous reviewer suggests the answer ‘my sister’ in both () and () could be the result of an underlying cleft or cleft-like structure, similar to cases investigated by Rodrigues et al. () and van Craenenbroeck (a). This is certainly a possibility, and future research could control for this possibility by using multiple remnants or non-exhaustive modification. 11



́-  

German Tisch, not Spanish mesa. On the other hand, the German neuter Mädchen ‘girl’ does not trigger the same effect (): in the ellipsis, the remaining Spanish determiner bears feminine gender, suggesting that the elided noun is Spanish niña, and not German Mädchen. The differences between these two cases might reside in the fact that Tisch/mesa is inanimate and therefore only has grammatical gender, whereas Mädchen is animate, and in German its grammatical gender (neuter) differs from its biological gender (feminine). These preliminary results are examples of the many questions that remain to be studied.

A We want to thank Jason Merchant, Luis López and the UIC Bilingualism Research Lab for valuable feedback on this chapter. Our special gratitude to the editors and two anonymous reviewers for their comments and suggestions.

  .............................................................................................................

ELLIPTICAL CONSTRUCTIONS .............................................................................................................

  ......................................................................................................................

              ......................................................................................................................

  ({)

. I

.................................................................................................................................. T approach to sluicing that I discuss in this chapter is the one originally advanced by Ross (b) and later on revived and explored extensively by Merchant in a series of publications (see especially Merchant , , , , a, b, , d, and Giannakidou and Merchant , plus Merchant a, this volume, and Chapters  and  of this volume for overviews of data and arguments).1 Abstracting away from frameworkrelated details, the core of Ross’s and Merchant’s proposals is the same: sluicing is a PF-side operation, namely, deletion (non-pronunciation) of a syntactically complete wh-interrogative to the exclusion of the wh-expression, under identity with some suitable antecedent.2,3 1

Throughout this chapter, I use the following terminology: antecedent

sluice

Someone just left — guess whoi [ti just left] ! correlate 2

remnant sluicing site

Throughout this chapter I assume a general P&P/Minimalism framework (as do a majority of the works I cite), but readers should keep in mind that this is not a necessity. The essentials of a deletion analysis can be ported to other frameworks, so long as those frameworks allow us to talk about phonetically null constituents in a meaningful way (cf. Jacobson b: § for a nuanced discussion of this issue). 3 There are a number of approaches to sluicing that propose a much less articulate, or even nonexistent, syntax for the sluicing site. For example, Chung et al. (, ) propose that sluicing sites consist only of an atomic proform that inherits the LF of the antecedent (this approach effectively amounts to treating sluices as deep anaphors, contrary to Hankamer and Sag ); more radically, Ginzburg and Sag (), Culicover and Jackendoff (), Sag and Nykiel (), Barker (), and Jacobson (b) have argued that sluicing sites, qua independent syntactic constituents, don’t exist; this line of attack requires accepting the existence of rules that can map a bare wh-phrase to the semantics of a full wh-question. This whole chapter can be viewed as an extended argument against such non-deletion approaches: if the syntactic properties of any given sluice follow from the kind of





Somewhat unimaginatively, I will refer to this line of analysis as the Ross–Merchant approach. By way of illustration, suppose that sluicing applies to (a) and produces the surface string (b), where [__] marks the position of the sluicing site. What Ross and Merchant are telling us is that the best analysis for (b) is going to be one along the lines of (), where the light grey font represents lexical material that fails to be phonetically realized.4 ()

a. Someone left—guess who just left! b. Someone left—guess who [__]!

VP () Someone just left—guess

CP who

C’ C° [+wh]

IP t just left

More precisely, both Ross and Merchant define deletion as constituent deletion (i.e., a specific node in the tree, and every node dominated by it, fails to receive a phonetic realization), which requires the remnant to evacuate the sluicing site prior to deletion, lest it also gets deleted. This requirement is easy to satisfy (in the general case) in languages with overt wh-fronting, given that the regular mechanism of wh-question formation independently takes the remnant to a position external to the sluicing site. More problematic, however, are sluices in wh-in situ languages, illustrated in () with Japanese (but see Kirchner , Gribanova c, Gribanova and Manetta , and many of the chapters in Merchant and Simpson  for comparable examples in other languages). How come sluicing is possible in the absence of overt wh-movement? () Hanako-ga nanika-o katta rasii ga . . . Hanako- something- bought likely but ‘It is likely that Hanako bought something, but . . . ’

syntactic structure that underlies the sluicing site, then analyses that deny the existence of such structure are inadequate, at least in the general case. See also Merchant (, a, a), Agüero-Bautista (), Abels (), Barros et al. (b), and references therein for more detailed arguments, and Vicente (b) for a summary of these arguments. See also the discussion of French sluicing in Chapter  of this volume for some problematic patterns. 4

This particular tree comes from Merchant (), but the exact category that the remnant moves to depends largely on the granularity of the functional structure one wishes to assume (as does the category that undergoes deletion). For example, Hartman (), who assumes that Rizzi’s () decomposition of the Italian CP layer carries over to English, proposes that the remnant moves to the specifier of ForceP after an intermediate stop in the specifier of FocusP, with deletion targeting a lower TopicP. See also Radford and Iwakasi () for a further exploration of this line of analysis in English, as well as van Craenenbroeck (, b) for Dutch.

    a. boku-wa [IP kanojo-ga nani-o I- she- what- ‘I don’t know what she bought.’ b. boku-wa nani-o I- what- ‘I don’t know what.’

katta] bought

ka Q



wakaranai. not.know

[__] ka wakaranai. Q not.know

A different version of the same problem arises in wh-fronting languages when we combine sluicing and islands. Ross (b: ff.) already noted that (b) is a grammatical sluice, even though the unsluiced question (a) incurs a strong island violation. How come the sluice is grammatical if the putative movement step required to derive it is illicit? ()

a. * She kissed a man that bit one of my friends, but Tom doesn’t realize which one of my friends she kissed a man who bit. b. She kissed a man that bit one of my friends, but Tom doesn’t realize which one of my friends [__].

There are, in principle, three different ways in which (b), (b), and comparable examples can be brought into the deletion fold. First, one could simply deny that constituent deletion is a necessity; instead, one could define a process of non-constituent deletion that elides everything inside IP except for the in situ remnant (Kimura , Abe , Ott and Struckmeier , and in a different context, An ), as illustrated in ()–(). On the assumption that sluicing bleeds A-bar movement in wh-fronting languages (cf. Kimura ; Abe ), the lack of island effects follows directly from the fact that no movement is taking place. I will not discuss this solution any more in this chapter, given the accumulated evidence that regular A-bar movement is an integral part of sluicing in all wh-fronting languages and at least some wh-in situ languages (cf. Agüero-Bautista , Abels , Barros et al. , plus Chapters  and  of this volume). ()

. . . boku-wa [IP kanojo-ga nani-o katta ] ka wakaranai. I- she- what- bought Q not.know

()

. . . but Tom doesn’t realize [IP she kissed a man who bit which one of my friends].

Alternatively, one could preserve the constituent deletion requirement and assume instead that sluicing invariably licenses movement of the remnant to an IP-external position (as illustrated in ()–()), even if such movement is otherwise illicit in unsluiced questions—see Takahashi (, ), and Takahashi and Lin () for an application of this idea to wh-in situ languages, plus Merchant (, a) and Almeida and Yoshida (), among others, for an application to sluicing out of islands in wh-fronting languages. This approach faces the problem of not being fully generalizable: both across and within languages, the observed distribution of grammatical and ungrammatical sluices is inconsistent with an analysis where sluicing licenses exceptional, island-insensitive movements on a regular basis (Rodrigues et al. ; Abels ; Gribanova c; Barros et al. ; and references). However, we will see in §§. and . some environments where an appeal to a weak version of this approach





(i.e., sluicing exceptionally licenses only a small subset of otherwise illicit movements) might be necessary. ()

. . . boku-wa I-

()

. . . but Tom doesn’t realize [which one of my friends]i [IP she kissed a man who bit ti].

nani-oi what-

[IP kanojo-ga ti she-

katta ] ka wakaranai. bought Q not.know

Note that the two approaches above presuppose a high degree of syntactic isomorphism between the sluice and the antecedent, i.e., the sluice is invariably a question built on the same predicate type as the antecedent. Our third approach, which constitutes the core of this chapter, rejects this assumption: so long as the standard semantic identity conditions are satisfied, sluices may stem from a different predicate (more precisely, various types of clefts and copular clauses),5 and any required movements need not be wh-movement in the strict sense (e.g., sluicing can be derived by scrambling or focus movement; see van Craenenbroeck and Liptak , , Toosarvandani , Ince , , and Chapter  of this volume). For example, Kizu (a) and Merchant (), among others, propose that (b) stems from a copular clause exhibiting simultaneous subject drop and copula drop (both independently possible in Japanese) as illustrated in (); similarly, Merchant () and Barros et al. (), among others, propose that (b) stems from deletion of a cleft, as in (). Figure . partially illustrates the range of underlying structures that sluices have been argued to stem from across languages (although, due to space restrictions, I will not be able to go through this taxonomy in detail). ()

. . . boku-wa [IP pro nani-o I- it what- ‘I don’t know what it is.’

da-aru ] ka wakaranai. be- Q not.know

() . . . but Tom didn’t realize [which one of my friends]i [IP it was ti]. This line of attack is interesting in that, unlike the other two approaches, it commits us to a very clear and specific prediction—i.e., that across languages the taxonomy of sluices should parallel the taxonomy of unsluiced questions. It follows from this that any given sluice should have the same properties as the corresponding unsluiced question. This prediction has been explored extensively, with notable results, in a great variety of languages and sluicing types over the last fifteen years, starting with the seminal work of Shimoyama (), Nishiyama et al. (), Kizu (a), and Merchant () on Japanese, 5 There is a tendency in the literature to refer to sluices derived from clefts and copular clauses as pseudosluices, which is a perversion of the original meaning of the term, as popularized by Merchant (). Merchant used pseudosluicing to refer to a class of Japanese sluices that do not involve deletion of IP or a comparable clausal constituent, but rather simultaneous subject drop and copula drop, which under certain circumstances can produce the same surface result as genuine sluicing (hence the pseudoqualifier). Here I am going to use pseudosluicing in the sense originally intended by Merchant and use the more accurate labels cleft sluicing and copular sluicing to cover the subtypes of sluicing other people use pseudosluicing for.

   



wh-movement isomorphic focus movement predicative copular clause

single

specificational equative

deep non-isomorphic

cleft shallow

Sluicing pseudocleft

multiple wh-/focus movement genuine multiple clefting multiple single wh-/focus movement + HNPS fake null coordination of single sluices

 . A partial taxonomy of sluicing, based on the underlying syntax of the sluice

and following with Merchant (, a), Hiraiwa and Ishihara (, ), van Craenenbroeck (, a), van Craenenbroeck and Lipták (, ), Potsdam (), Toosarvandani (), Vicente (), Ince (, ), Rodrigues et al. (), Barros (a, b), Hoyt and Teodorescu (), Paul and Potsdam (), Gribanova (c), Barros et al. (), Barros and Vicente (), and Gribanova and Manetta (), among many others—see also the bibliography in Merchant and Simpson () for an extensive list of related publications. As Merchant (a: ) aptly points out: These parallels in distribution are immediately and straightforwardly accounted for by the theory of sluicing discussed above, since the grammatical constraints [ . . . ] will be operative uniformly in both elliptical and non-elliptical structures.

The goal of this chapter is to present a small subset of the data and arguments discussed in this body of literature (§§. and .), as well as some patterns that seem to put limits on the generalizability of a pure deletion analysis as described in Merchant’s quote above, i.e., grammatical sluices for which no grammatical unsluiced question seems to exist (§.). Due to space restrictions, I am going to deliberately (and unfortunately) ignore a number of important issues, such as the different discourse functions of sluices (Ginzburg and Sag ; Schlangen ; Fernández ), the status of sluices without a linguistic





antecedent (Ginzburg ; Ginzburg and Sag ; Stainton b; Merchant ; Weir ), correlate–remnant congruence restrictions (Chung et al. , ; Romero ; Dayal and Schwarzschild ; Barros b; Winkler ; Messick et al. ; Jacobson b), voice and argument structure mismatches (Chung et al. , ; Chung ; Barros b; Barros and Vicente ; Merchant this volume), the conditions on, and causes of, deletion (cf. Rooth a; Tancredi ; Johnson a; Merchant ; Hartman ; AnderBois , ; Chung ; Barros b), certain minor variants of sluicing like sprouting (Albert ; Chung et al. , ; Romero ; Chung ; Barros and Vicente ), swiping (Merchant ; Culicover and Jackendoff ; Beecher ; Hartman and Ai ; van Craenenbroeck b; Larson ; Radford and Iwakasi ), and sluices derived from relative clauses (Szczegielniak ; van Craenenbroeck and Lipták , ), modal existential constructions (Šimík : §..), or amalgams (Kluck ). Readers interested in these issues are referred instead to the works just cited and the references therein.

. M     

..................................................................................................................................

.. Non-isomorphic sluicing and island repair effects The idea that sluices can also stem from deletion of non-isomorphic predicates (specifically, clefts and copular clauses) is hardly a new one: Erteschik-Shir (: ), Pollmann (: ), and Rosen () proposed this much in order to account for Ross’s observation that sluicing is able to circumvent island effects. Their claim, elaborated in more detail by Merchant (: §..), Rodrigues et al. (), and especially Abels () and Barros et al. (), is that repair effects are only a superficial illusion: whenever sluicing appears to void an island violation, it is because the sluice in question stems from deletion of a nonisomorphic clause that doesn’t incur the pertinent violation (an evasive source, in the terminology of Barros et al. ). This line of attack makes a very clear prediction: because repair effects are necessarily linked to the availability of an evasive source, we should expect island effects to reappear whenever the evasive source can be independently blocked. Given that locality effects are discussed in more detail in Chapter  of this volume, I offer only a brief illustration of this pattern here: in many languages, extraction of an attributive adjective out of its containing DP is ungrammatical (a violation of Ross’s  LeftBranch Condition), but the corresponding sluice is grammatical. Call these left-branch extraction (LBE) sluices, illustrated here with English. ()

They hired a diligent worker . . . a. * . . . but I don’t know [how diligent]i they hired [DP a ti worker]. b. . . . but I don’t know how diligent [__].

Merchant (: §...) proposes a literal repair analysis of these sluices, but Barros et al. () observe that they exhibit various properties that support an evasive analysis instead. First, the repair effect is suspended for adjectives with non-intersective readings (a), and

   



adjectives ambiguous between an intersective and a non-intersective reading are invariably disambiguated in favor of the intersective reading (b). ()

a. * They hired a hard worker, but I don’t know how hard [__]. b. Jack is visiting an old friend, but I don’t know how old [__]. [= I don’t know the age of Jack’s friend] [6¼ I don’t know how long this friendship has been going on]

Notably, non-intersective readings of adjectives are also disallowed in the predicate position of copular clauses. ()

a. The worker is diligent. b. * The worker is hard. c. Jack’s friend is old. [= Jack’s friend is of an advanced age] [6¼ this friendship has been going on for a long time]

This parallelism suggests that (b) doesn’t stem from deletion of (a) but rather from deletion of a predicative copular clause, which doesn’t incur an LBC violation in the first place. ()

. . . but I don’t know how diligent [IP that worker is t].

Barros et al. reinforce this analysis by pointing out that, in languages where adjectives inflect differently depending on whether they are used predicatively or attributively, LBE sluices require the predicative inflection. I illustrate this below with German, where attributive adjectives bear the appropriate case morphology and both predicative adjectives and remnants of LBE sluices are bare.6 This pattern follows directly from an evasive analysis, but a literal repair analysis would predict sluicing remnants to bear the attributive inflection. ()

a. Elke hat ein-en groß*(-en) Elke has a- big-

Mann man

gesehen. seen

b. Der Mann ist groß(*-en). the man is big- c. Elke hat ein-en groß-en Mann gesehen, aber ich weiß nicht wie Elke has a- big- man seen but I know not how groß(*-en) [__]. big-

Merchant (: ) contests this claim. He provides (i), with the indicated judgment, and the comment that “sluicing is fairly degraded, with or without inflection.” I have been unable to replicate this judgment in my own research: all the speakers I have consulted agree with the judgments that Barros et al. indicate. 6

(i)

?? Sie haben ein-en groß-en Center eingestellt, aber ich weiß nicht wie groß(-en). they have a- big- center hired but I know not how big-





To complete the argument, note that languages that allow extraction of attributive adjectives in unsluiced questions (and therefore do not require an evasive source) exhibit the opposite pattern, i.e., they allow LBE sluices with intersective adjectives7 and attributive inflection. I illustrate this pattern here with Serbo-Croatian (data from Boban Arsenijević, p.c.). This asymmetry between language types is predicted to arise only in an analysis where the repair effect is an illusion.8 () Jovan Jovan

je zaposlio tvrdog radnika, ali ne znam koliko tvrdog [__].  hired hard. worker. but not know. how hard.

I refer the reader to Lasnik (), Merchant (: §.), Fox and Lasnik (), Lasnik and Park (), van Craenenbroeck and den Dikken (), Agüero-Bautsta (), Nakao and Yoshida (), Merchant (b), Nakao (), Rodrigues et al. (), Abels (), Barros (a), Fukaya (), Nakamura (), Yoshida et al. (b), Cantor (), Marušič and Žaucer (), Rottman and Yoshida (), Barros (a), Barros et al. (), Griffiths and Lipták (), and especially Chapter  of this volume for additional discussion of the evasive approach to island repair, as well as of its limitations. See also §§... and ... for a brief illustration of two environments where literal island repair might be necessary.

.. Non-isomorphic sluicing and wh- in situ languages Beyond providing a means to evade island violations, cleft- and copula-based sluices have also been successfully applied to the analysis of sluicing in wh-in situ languages. Consider Japanese, which I already mentioned in §..

7 This generalization is somewhat of an idealization, in the sense that some adjectives in these languages lose their non-intersective reading upon extraction, regardless of whether sluicing happens. What is important for this particular argument is whether the availability of the non-intersective reading of a certain adjective under LBE sluicing correlates with the availability of this reading under unsluiced left-branch extraction. As far as I have been able to determine, this is indeed the case. 8 As an additional argument in favor of this view of LBC violation repair, note that many Slavic languages disallow multiple LBE. Example (i) illustrates this for Russian. This restriction persists under sluicing (ii), which would be surprising if sluicing could actually repair LBE violations (all data from Grebenyova b).

(i) * [Naskol’ko bogatyj]i [naskol’ko doroguju]k [ti aktër] kupil how.much rich how.much expensive actor bought ‘How rich an actor bought how expensive a car?’

[tk mašinu]? car

(ii) * Včera odin aktër kupil mašinu, no ja ne pomnju [naskol’ko botyj] yesterday one actor bought car but I not remember how.much rich [naskol’ko doroguju] [__]. how.much expensive ‘Yesterday, an actor bought a car, but I can’t remember how rich how expensive.’

    ()



Hanako-ga nanika-o katta rasii ga, boku-wa nani-o ka wakaranai. Hanako- something- bought likely but I- what- Q not.know ‘It is likely that Hanako bought something, but I don’t know what.’

Shimoyama (), Nishiyama et al. (), Kizu (a), Merchant (), Fukaya (, ), Hiraiwa and Ishihara (), and Nakamura (), among others, have developed a number of arguments in favor of a non-isomorphic cleft source (but see Iseda  and Hasegawa , as well as the discussion in §.. here). For example, Shimoyama notes that these sluices support a copular verb, which would be surprising if they didn’t stem from an underlying cleft. ()

Hanako-ga dareka-o kubinisita rasii kedo, boku-wa Hanako- someone- fired seem but I- da ka siranai.  Q not.know ‘It seems that Hanako fired someone, but I don’t know who.’

dare-o who-

Additionally, Kizu (a) and Merchant () note, following earlier literature, that cleft pivots tend to resist being overtly case-marked, each case morpheme triggering a specific degree of deviance (and some degree of cross-speaker variation). Compare, for example, the different acceptability of nominative -ga and accusative -o: () a. Bungo-ni Aya-o syookaisita no-wa Bungo- Aya- introduced C- ‘It is Kota that introduced Aya to Bungo.’ b. Kota-ga Bungo-ni syookaisita no-wa Kota- Bungo- introduced C- ‘It is Aya that Kota introduced to Bungo.’

Kota(*-ga) Kota-

da. 

Aya(??-o) Aya-

da. 

As expected under a cleft-based analysis, sluices exhibit the same case-marking pattern: () a. Dareka-ga sono hon-o yon-da ga, watashi-wa dare(*-ga) ka wakaranai. someone- that book- read- but I- who- Q not.know ‘Someone read that book, but I don’t know who.’ b. Taroo-ga dareka-o nagutta ga, watashi-wa dare(??-o) ka wakaranai. Taroo- someone- hit but I- who- Q not.know ‘Taroo hit someone, but I don’t know who.’ This line of analysis extends easily to other wh-in situ languages. I refer the reader to Kirchner () and Tomioka () for Mandarin Chinese, and Gribanova (c) for Uzbek.





.. Non-isomorphic sluicing as a default strategy As is the case with other Austronesian languages, wh-extraction of non-subjects in Malagasy requires a pseudocleft (). Potsdam () and Paul and Potsdam () argue that this restriction remains active under ellipsis, so sluices like () must stem from a comparable non-isomorphic, pseudocleft source. ()

nanontany aho hoe iza no ask. .  who  ‘I asked who it is that is buying the goat.’

mividy buy.

ny the

osy goat

() a. nandoko zavatra i Bao fa hadinoko hoe inona [__] paint. thing Bao but forget..  what ‘Bao painted something, but I forget what it is that was painted by Bao.’ b. nangalarin’ ny olona steal. the person hoe viza [__]  who

ny the

fiarako car.

fa but

tsy 

fantatry know

ny the

polisy police

‘My car was stolen by someone but the police don’t know who it is that stole my car.’ Potsdam () and Paul and Potsdam () support this analysis by pointing out that sluices exhibit certain restrictions that are otherwise specific to pseudoclefts; as should be obvious, such parallelisms follow without stipulation if sluices are derived from a pseudocleft base. First, there are certain constituents (e.g., accusative-marked arguments) that can neither be questioned through the pseudocleft strategy (a) nor sluiced (b). () a. * an’iza no nanasa who.  invite ‘Who did Rabe invite?’

Rabe? Rabe

b. * nanasa olona Rabe ka nanontany invite someone Rabe and ask ‘Rabe invited someone and I asked who.’

aho I

hoe 

an’iza who.

Second, pseudocleft pivots (qua fronted predicates) can be directly followed by a variety of elements, such as the modifiers daholo ‘all’ and foana ‘always’ (). Potsdam’s and Paul and Potsdam’s analysis of this pattern relies on the well-supported assumption that daholo, faona, and similar modifiers are contained in the constituent that undergoes fronting in pseudoclefts (PredP in their terminology), which allows them to escape deletion of IP1. As expected, the same range of elements can also follow sluicing remnants (). () a. iza daholo no namaky who all  read ‘Who all read the book?’

ny the

boky? book

   



b. iza foana no any an-tsena? who always   -market ‘Who is always at the market?’ () a. nahandro zavatra maro Rasoa fa tsy fantatro hoe inona daholo [__] cook thing many Rasoa but  know.  what all ‘Rasoa cooked many things, but I don’t know what all the things are.’ b. any an-tsena matetika ny mpivarotra sasany fa tsy fantatro hoe there -market often the merchant some but  know.  iza foana [__] who always ‘Some merchants are often at the market but I don’t know who is always there.’

.. The balance between isomorphic and non-isomorphic sluicing Given the pervasivity of cleft and copular sluicing, there might be a certain temptation at this juncture to jump to the conclusion that all cases of sluicing can be reduced to cleft or copular sluicing (i.e., there would be no proper isomorphic sluicing). This is a conjecture that can be safely rejected. Merchant (: –) already provides ten independent arguments against a reduction along these lines. In the interest of brevity, I reproduce only two of them here: first, he notes that adverbial wh-phrases (e.g., how, why, when . . . ) are allowed in wh-questions (a), but not in clefts (b). The fact that sluices with adverbial wh-remnants are grammatical (c) suggests that they do not stem from an underlying cleft. ()

He fixed the car, . . . a. . . . but I don’t know { how / why / when / where } he fixed the car. b. * . . . but I don’t know { how / why / when / where } it was. c. . . . but I don’t know { how / why / when / where } [__].

Second, regular wh-questions allow else-modification (a), but clefts do not (b). As above, sluicing remnants pattern with regular wh-questions in allowing elsemodification (c). ()

Harry was there, . . . a. . . . but I don’t know who else was there. b. * . . . but I don’t know who else it was. c. . . . but I don’t know who else [__].

Importantly, Merchant notes that these arguments only contraindicate a general reduction of sluicing to cleft sluicing, but leave open the possibility that some sluices stem, in fact, from a cleft (or a copular clause); van Craenenbroeck (a) and Barros (b) make





the same point.9 This much suggests, as already mentioned above, that sluicing is a syntactically heterogeneous construction both cross- and intra-linguistically. Merchant’s arguments can be replicated in other languages—see, for example, Rodrigues et al. () and Saab () on Spanish, or Lipták (a) on Hungarian. Especially interesting in this respect is the status of Farsi (Toosarvandani  and Chapter  of this volume) and Turkish (Ince , ). Both are wh-in situ languages, and in both of them the battery of tests discussed above contraindicate deriving sluicing from a nonisomorphic source, at least in the general case. Toosarvandani and Ince conclude from this combination of factors that Farsi and Turkish sluicing is of the isomorphic kind. The difference with English-style isomorphic sluicing is that the movement of the remnant out of the sluicing site is not a case of wh-movement proper; rather, it is a case of focus movement, which is independently attested and productive in both languages. Takita () develops a comparable analysis for the specific case of Japanese sluices embedded under control predicates (). () Taroo-wa [dono zyaanaru-ni zibun-no ronbun-o das-oo ka] kimeta Taroo- which journal- self- paper- submit- Q decided ga, Hanako-wa [dono zyaanaru-ni [__] ka] kimekaneteiru. but Hanako- which journal- Q cannot.decide ‘Taroo decided which journal to submit his paper to, but Hanako can’t decide which (to submit hers to).’ Takita notes that these predicates cannot take cleft or copular complements (), which contraindicates deriving sluicing from a non-isomorphic source; this conclusion is reinforced by the fact that the relevant sluices disallow the optional presence of a copular verb (), unlike the cases discussed in §... From this, Takita concludes that these sluices stem from an isomorphic source, with extraction of the remnant out of the sluicing site being a case of scrambling.10 () a. * Taroo-wa [(sore-ga) doko-e Taroo- it- where-to ‘Taroo can’t decide where it is.’

da 

b. * Taroo-wa [iku no-ga doko-e da Taroo- go C- where-to  ‘Taroo can’t decide where it is to go.’

ka] Q ka] Q

kimekaneteiru. cannot.decide kimekaneteiru. cannot.decide

9 Although van Craenenbroeck and Barros make slightly different proposals about the availability of cleft sources: van Craenenbroeck defines an analysis where cleft sources are available only if using the corresponding non-cleft source would have produced an ungrammatical result; on the other hand, Barros proposes that both cleft and non-cleft sources are equally available in principle. 10 He treats the relevant movement as scrambling, and assumes that scrambling is itself a subtype of wh-movement (pace Takahashi , ). It is unclear to what extent this assimilation is justifiable.

   



() Taroo-wa [dono zyaanaru-ni zibun-no ronbun-o das-oo ka] kimeta ga, Taroo- which journal- self- paper- submit- Q decided but Hanako-wa [dono zyaanaru-ni [__] (*da) ka] kimekaneteiru. Hanako- which journal-  Q cannot.decide ‘Taroo decided which journal to submit his paper to, but Hanako can’t decide which (to submit hers to).’ These results suggest that non-isomorphic sources are not the only means to circumvent the limitations of wh-movement—one can also resort to an isomorphic source coupled with a different type of movement. This conclusion, however, is hardly new: Merchant (: ) has briefly hinted at a similar analysis of Turkish and Hindi, writing that “to the extent that sluicing structures pattern with those found in overt wh-movement languages like English, [they] are employing a scrambling-type movement to create the input structures for deletion, and not using ‘true’ wh-movement to specCP.”

. M 

..................................................................................................................................

.. Genuine multiple sluicing ... Multiple focus/wh–fronting Given the discussion so far, it is unsurprising that multiple wh-fronting languages allow sluices with multiple remnants. I illustrate this possibility in () with Serbo-Croatian, but comparable paradigms can be constructed in other Balkan/Slavic languages (cf. Richards  and Grebenyova b, among others).11 () Serbo-Croatian (Merchant ; Stjepanović ) Neko je vidio nekog, ... someone. is seen someone. a. . . . ali but

ne not

znam know.

koi who.

kogak who.

[TP ti je vidio tk]. is seen

b. . . . ali but

ne not

znam know.

ko who.

koga who.

[__].

11

Here it is appropriate to note the work of Ortega-Santos et al. () and Yoshida et al. (b) on English ellipses with two remnants, out of which only one is a wh-item. (i)

A: John ate natto. B: Why natto [__]?

(ii)

A: Lou will ask Doris about syntax. B: And who about phonology [__]?

Ortega-Santos et al.’s and Yoshida et al.’s proposal is that such examples involve a combination of sluicing and stripping, the latter involving focus movement rather than wh-movement. In this sense, they are not cases of multiple sluicing; however, given the close parallelisms between sluicing and stripping (Merchant a), this might be the closest that English gets to genuine multiple sluicing.





Given that there are different subtypes of multiple wh-fronting languages, each one with a different cluster of properties (Rudin  et seq.), the Ross–Merchant approach to sluicing predicts that multiple sluices in each of these languages will inherit the properties of the corresponding overt multiple questions. In support of this prediction, Merchant (: ) notes that Bulgarian multiple sluices (a) respect Superiority, just as their unsluiced counterparts do (b). ()

Bulgarian (Merchant ) a. * Njakoj vidja njakogo, no ne znaw kogo koj [__]. someone. saw someone. but not know. who. who. b. * Kogo who.

koj who.

e 

vidjal? seen

The reverse effect also obtains, i.e., Superiority violations under multiple sluicing are possible only in languages that also allow them in unsluiced multiple questions. I illustrate this pattern here with Russian (data from Grebenyova ).12 ()

Russian (Grebenyova ) Každyj kto-to priglasil na tanec, no ja ne pomnju kogo everyone. someone. invited to dance but I not remember who. kto [__]. who.

Grebenyova (a: ch. ) further points out that Russian unsluiced multiple wh-questions allow pair-list, but not single-pair readings. The examples below demonstrate that, as expected, this restriction carries over to multiple sluices. Note that the manipulation of the subject in the antecedent clause (každyj ‘everyone’ vs ktoto ‘someone’) is done deliberately to induce the pair-list and single-pair reading, respectively. () Russian (Grebenyova a) a. Každyj priglasil kogoto na tanec, no ja ne pomnju kto everyone. invited someone. to dance but I not remember who. kogo [__]. who. [každyj induces pair-list reading]

12

Note that this is only a one-way generalization. For example, Grebenyova points out that Russian Superiority-violating multiple sluices require the order of the remnants to parallel the order of the correlates. She attributes this restriction to a combination of Fox’s () scope parallelism constraint and the fact that scope in Russian is defined on surface structures (Ionin ). Similarly, Stjepanović () observes that Serbo-Croatian, which allows Superiority-violating unsluiced questions, bans Superiority-violating multiple sluices, given that scrambling of an object across the subject in the antecedent blocks sluicing altogether.

   



b. ?? Ktoto priglasil kogoto na tanec, no ja ne pomnju kto someone. invited someone. to dance but I not remember who. kogo [__]. who. [ktoto induces single-pair reading] In contrast to Russian, Serbo-Croatian allows single-pair readings in multiple wh-questions (Bošković ; Grebenyova a). As expected, this enables multiple sluices with singlepair readings.13 () Serbo-Croatian (Boban Arsenijević, p.c.) a. Svako je nekog pozvao na pies, ali ne secam se everyone.  someone. invited on dance but not remember.  ko koga [__]. who. who. [svako induces pair-list reading] b. Neko je nekog pozvao na pies, ali ne secam se someone.  someone. invited on dance but not remember.  ko koga [__]. who. who. [neko induces single-pair reading] For a more extensive and detailed discussion of the correlations between multiple whfronting and genuine multiple sluicing, I refer the interested reader to Grebenyova (a, ), van Craenenbroeck and Lipták (), and references therein.

... Multiple clefting Some wh- in situ languages also exhibit sluices with multiple remnants. Here, I will concentrate on Japanese, for which Takahashi () and Kuwabara (), Hiraiwa and Ishihara () and Takahashi and Lin () have noted examples like (). () Tetsuo-ga [dareka-ga nanika-o katta to] itta. Kaneda-wa Tetsuo- someone- something- bought that said Kaneda- [dare-ga nani-o [__] ka] siri-tagat-e iru. who- what- Q know-want is ‘Tetsuo said that someone bought something. Kaneda wants to know who what.’

13 It is not clear to me to what extent this correlation can be preserved in analyses where sluices do not have a regular syntax. Bošković (, ), Grebenyova (a), and others have argued at length that the availability of a single-pair reading is contingent on wh-fronting being triggered by a [] feature, rather than a [-] feature. If this conjecture is cross-linguistically consistent (see Bošković ,  for discussion), then the contrast between Russian-type languages and Serbo-Croatian-type languages can be construed as an additional argument in favor of sluices having a regular syntax.





The standard analysis of such examples relies on the fact that multiple clefting is independently possible in Japanese ()—see Kuwabara (), Hiraiwa and Ishihara (), and references therein.14 () [Taro-ga ageta-no]-wa Hanako-ni Taro- gave-- Hanako- ‘It is three apples to Hanako that Taro gave.’

ringo-o apple-

san-tu three-

da. 

Hiraiwa and Ishihara () discuss a number of parallelisms between multiple clefts and multiple sluices that suggest that the latter ought to be derived from the former. For example, Japanese DPs can appear without case markers under certain circumstances, but multiple clefting requires case markers to be retained (a). In the same way, remnants of multiple sluicing necessarily each appear with their corresponding case markers (b). () a. [Taro-ga ageta no]-wa Hanako*(-ni) Taro- gave - Hanako- ‘It is three apples to Hanako that Taro gave.’

ringo*(-o) apple-

san-tu three-

da. 

b. Taro-ga dareka-ni nanika-o ageta rasii ga, boku-wa [[__] dare*(-ni) Taro- someone- something- gave seem but I- who- nani*(-o) (da) ka] wakaranai. who-  Q know.not ‘Taro gave something to someone, but I don’t know what to who.’ Similarly, pivots of multiple clefts generally need to be clausemates except if they are wh-expressions—see especially Ishihara () and references for an analysis of this pattern. As expected, remnants of multiple sluicing pattern with wh-pivots of multiple clefts in not being subject to the clausemate restriction.15 () a. * [Mari-ga [Naoya-ga nonda to ] iituketa no ] wa sensei-ni Mari- Naoya- drank  told   teacher- wine-o da. wine-  ‘It is to the teacher, wine, that Mari told that Naoya drank.’ 14 Specifically, Hiraiwa and Ishihara make a distinction between clefts and pseudoclefts, which exhibit a number of asymmetries (e.g., possibility of multiple pivots, possibility of case dropping, possibility of nominative–genitive conversion, etc.). Given that sluices in general, and multiple sluices in particular, exhibit the same range of properties as clefts, Hiraiwa and Ishihara propose that (multiple) sluices stem exclusively from clefts, rather than pseudoclefts. 15 Although Takahashi (: §.) makes the opposite claim, citing (i) as evidence. At present, I do not know why judgments differ in this way.

(i)

* Dareka-ga [John-ga nanika-o katta to] itteita ga, Mary-wa [dare-ga someone- John- something- bought that said but Mary- who- nani-o [__] ka] oboeteinai. what- Q not-remembers ‘Someone said that John bought something, but Mary doesn’t remember who what.’

   



b. [Mari-ga [Naoya-ga nonda to ] iituketa no ] wa dare-ni nani-o Mari- Naoya- drank  drank  top who- what- na no?   ‘To who, what is it that Mari told that Naoya drank?’ c. Mari-ga dareka-ni [Naoya-ga nanika-o nonda to ] iituketa Mari- someone- Naoya- something- drank  told rasii ga boku-wa [[__] dare-ni nani-o (da) ka] wakaranai. seem but I- who- what-  Q know.not ‘It seems that Mari told someone that Naoya drank something, but I don’t know to who what.’ As above, I refer the reader to Kuwabara () and Hiraiwa and Ishihara () for a more detailed discussion of these examples. What is relevant for the purposes of this chapter is the fact that languages with non-isomorphic sluicing allow multiple sluicing remnants, so long as the corresponding unsluiced structure (in the case of Japanese, clefts with multiple pivots) is independently available.

.. Fake multiple sluicing ... Single sluicing plus Heavy NP Shift Contrary to what one might expect from a Ross–Merchant analysis, multiple-sluicing-like configurations are also allowed in languages that do not otherwise allow multiple fronting (whether qua wh- or focus-driven movement) or multiple clefting. The like qualifier is important here: it turns out, upon closer inspection, that what these languages exhibit is a case of single sluicing supplemented with some additional process that produces the illusion of genuine multiple sluicing, as in the English examples in ().16 ()

a. ? I know that, in each instance, one of the girls got something from one of the boys. But which [__] from which? b. ? One of the students spoke to one of the professors, but I don’t know which student [__] to which professor.

Note that the placement of the [__] diacritic in between the two remnants is not accidental. Both Merchant () and Richards () originally proposed that sluicing licenses exceptional multiple overt wh-fronting in English (and, by extension in other non-multiple wh-fronting languages). Against this background, Lasnik () proposes that only the first Takahashi (: ) marks (i) as ungrammatical, which leads him to argue that English lacks multiple sluicing altogether (Takahashi and Lin  makes the same claim). While (i) is indeed ungrammatical, it only shows that the distribution of English multiple sluices is not totally free; for one, Nishigauchi () and Merchant () point out that licit multiple sluices need to have a pair-list reading, which is not available in (i). 16

(i) * John said someone bought something. Mary wonders who [__] what.





remnant undergoes regular focus/wh-fronting; the second one escapes the sluicing site through rightward Heavy NP Shift (). ()

. . . but I don’t know [CP [which students]i [ti talked ti] [to which professors]k]

Among Lasnik’s arguments in favor of () is the fact that the second remnant doesn’t allow P-stranding (). This is a surprising restriction, given that P-stranding under sluicing is otherwise possible in English (cf. Merchant ); however, it follows directly from (), given that P-stranding under HNPS is likewise impossible (b)/(c). ()

a.

Some of the students talked to some of the professors, but I don’t know [CP which students [__] *(to) which professors]. b. Some of the students talked ti yesterday [PP to some of the professors]i. c. * Some of the students talked [PP to ti] yesterday [DP some of the professors]i.

Additionally, Lasnik also notes that multiple sluicing becomes impossible if the two remnants are not clausemates (a); other languages I have checked (i.e., Farsi, Maziar Toosarvandani, p.c., and Turkish, as described in Ince ) behave in the same way.17 This restriction follows directly from Lasnik’s proposal, as the second remnant would have to undergo HNPS across a finite clause boundary in violation of Ross’s () Right Roof Constraint (illustrated in (b) for English). () a. * Some of the students say that Harvey talked to some of the professors, but I don’t know [CP which students [__] to which professors]. b. * Some of the students said that Mary will speak ti yesterday [to some of the professors]i. Note that, under Lasnik’s analysis, this clausemate restriction is predicted not to hold in languages where multiple sluicing can be derived through regular multiple wh-/focus fronting (see §...), at least to the extent that fronting of non-clausemate multiple wh-phrases is possible in non-elliptical contexts. The paradigm in () illustrates this effect for Serbo-Croatian. Lasnik points out that not all speakers he consulted accept (a). Importantly, though, those same speakers didn’t accept (b) either, which provides additional support for the hypothesis that there is a direct connection between the acceptability of the sluiced and unsluiced forms of these questions.

17 Norwegian is a special case, as some of the speakers I polled do not report a clausemate restriction (i). Other than this, all the speakers I surveyed reported Lasnik-like judgments in accepting multiple sluicing with clausemate remnants and rejecting P-stranding on the second remnant. I have currently no explanation for this pattern.

(i)

% Noen some hvilke which

studenter sier at Per har snakket med noen professorer, men jeg vet ikke students said that Per has spoken with some professors but I know not studenter [__] med hvilke professorer. students with which professor

    ()



Serbo-Croatian (Lasnik ) a. % Neko misli da je Ivan nesto Pojeo. Pitam se [CP ko sta [__]]. someone thinks that is Ivan something ate ask self who what ‘Someone thinks Ivan ate something. I wonder who what.’ b. % Ko sta misli da je who what thinks that is ‘Who thinks Ivan ate what?’

Ivan Ivan

pojeo? eaten

... Null coordination of simple sluices Merchant (: ) argues that the Turkish multiple sluice in (a) is likely to be two separate simple focus sluices joined by a null coordinator. In support of this idea, he points out that the two remnants need to be separated by either a strong pause or an actual overt coordinator (b). () a. Biri birşey gördü ama, [kim ne [__]] bil-mi-yor-um. someone something saw but who. who. know--- ‘Someone saw something, but I don’t know who what.’ b. Biri birşey gördü ama, [kim [__] { ve / veya } ne [__]] someone something saw but who. and or who. bil-mi-yor-um. know--- ‘Someone saw something, but I don’t know who and/or what.’ Gribanova (c) proposes the same analysis for the Uzbek example in (), which appears to instantiate multiple pseudosluicing. The evidence she offers is the same Merchant does for (a), i.e., that “when larger, more weighty conjuncts are coordinated [ . . . ] a large pause (a) or an overt coordinator (b) becomes necessary.” () Kecha kim-dir kim-ga-dir pul ber-di, lekin [kim kim-ga [__] yesterday some-one some--one money give-. but who who- lig-i-ni ] bil-ma-y-man. -.- know--- ‘Yesterday, someone gave money to someone, but I don’t know who to whom.’ () a. Kecha bir bola bir qiz-ga pul ber-di, lekin [qaysi bola [__]] *(#) yesterday one boy one girl- money give-. but which boy [qaysi qiz-ga [__] lig-i-ni ] bil-ma-y-man. which girl- -.- know--- ‘Yesterday, some boy gave money to some girl, but I don’t know which boy (it was and) which girl (it was).’ b. Kecha bir bola bir qiz-ga pul ber-di, lekin [qaysi bola [__]] *(va) yesterday one boy one girl- money give-. but which boy and [qaysi qiz-ga [__] lig-i-ni ] bil-ma-y-man. which girl- -.- know--- ‘Yesterday, some boy gave money to some girl, but I don’t know which boy (it was and) which girl (it was).’





Obviously, if Merchant’s and Gribanova’s analyses are correct, then (a) and () instantiate multiple sluicing only superficially, not in any analytically meaningful sense of the term.

. L     

.................................................................................................................................. As should be obvious at this stage, the unambiguous prediction of the Ross–Merchant analysis is that sluices should invariably exhibit the same syntactic properties as the corresponding unsluiced question. This is admittedly a very strong prediction, so it is unsurprising that it runs into difficulties every now and then. The goal of this last section is to present a few of these difficult cases. The question of whether there is a way to subsume them under a pure deletion analysis (or whether they require extending the deletion analysis in specific ways) is one that runs beyond the limits of this chapter.

.. Repair effects without an evasive source ... Parasitic gaps in the remnant Yoshida et al. () argue at length that the remnant-internal gap in (), notated [PG2__] is a parasitic gap licensed by a real gap [RG2__] contained inside the sluicing site (and thus not directly visible), paralleling the licensing of the parasitic gap in the first conjunct [PG1__] by its corresponding real gap [RG1__]. Note that standard licensing conditions on parasitic gaps prevent [PG1__] from being licensed by [RG1__]. ()

The editor told me which book I must review [RG1__] soon after receiving [PG1__], but I don’t remember exactly how soon after receiving [PG2__][__].

Note, however, that the unsluiced counterpart of () is ungrammatical, a fact that Yoshida et al. are well aware of (). Moreover, the source of the ungrammaticality of () is easy to pinpoint—namely, the illicit movement of the wh-phrase containing the parasitic gap across the wh-island boundary created by which book. ()

* The editor told me which book I must review [RG1__] soon after receiving [PG1__], but I don’t remember exactly how soon after receiving [PG2__] the editor told me which book I must review [RG2__].

The acceptability of () and similar examples suggests that the theory of sluicing must be supplemented with some island repair capabilities (as proposed by Ross b and Merchant , but contrary to the evasion approach discussed in §..). This conclusion, if valid, raises the difficult question of why some island effects require an evasive source to be “repaired,” whereas others allow literal repair.

   



... P-stranding effects in non-P-stranding languages Merchant () argues that P-stranding effects under sluicing are observed only in those languages that allow P-stranding in unsluiced questions. A P-stranding language like English allows prepositions heading the remnant to be freely dropped (), whereas a non-P-stranding language like German does not (). Previous literature has referred to this as the P-stranding generalization (PSG), but here I will use Merchant’s generalization instead.18 ()

a. Whoi did Jack talk [PP to ti]? b. Jack has talked to someone, but I don’t know (to) who [__].

()

a. * Wemi who b.

hat has

Peter Peter

[PP mit ti] with

gesprochen? talked

Peter hat mit jemandem gesprochen, aber ich weiß nicht *(mit) wem [__]. Peter has with someone talked but I know not with who

Some counterexamples to Merchant’s generalization have been reported, but the relevant question here is whether they are genuine counterexamples or simply instances of evasive sluices (in the sense discussed in §..), where a cleft/copular source creates a superficial P-stranding illusion. Rodrigues et al. () argue at length that the second option is the correct one for Romance languages, focusing on Spanish and Brazilian Portuguese. Their argument is based on the fact that P-stranding sluices in these languages exhibit a number of properties typically associated with clefts. To give a single example (from Spanish), they observe that wh-phrases in non-cleft questions can be modified by más ‘else’, but those in cleft questions cannot. () a. Juan ha hablado con María, pero no sé con qué chica (más) ha hablado. Juan has talked with María but not know with what girl else has talked b. Juan ha hablado con María, pero no sé qué chica (*más) es. Juan has talked with María but not know what girl else is Under sluicing, the presence vs absence of a preposition correlates with the possibility vs impossibility (respectively) of más modification. This parallelism, among others, leads Rodrigues et al. to conclude that P-stranding effects under sluicing stem from an evasive source, rather than being direct counterexamples to Merchant’s generalization. ()

a. Juan ha hablado con María, pero no sé con qué chica (más) [__]. Juan has talked with María but not know with what girl else b. Juan ha hablado con María, pero no sé con qué chica (*más) [__]. Juan has talked with María but not know with what girl else

18 Merchant () himself uses the more opaque term Form-Identity Generalization II, where FormIdentity Generalization I is the observation that remnants of sluicing bear the same case morphology as their correlates (see §..).





Rodrigues et al. conclude their study by noting that “the strongest implication is that all languages that appear to violate this generalization should be reducible to a pseudosluicing [i.e., evasive, LV] analysis.” A number of later studies have challenged this conjecture. For example, Sag and Nykiel () and Nykiel () argue that, in Polish P-stranding sluices, the remnant must have the case assigned by the missing preposition, rather than the characteristic instrumental of cleft pivots.19 () Adam regularnie dostaje prezenty Adam regularly gets presents { ✓ kogo / * kim }[__] . who. who.

od from

kogós, someone.

ale but

nie not

wiem know

Similarly, Sato (b) argues that P-stranding effects in Indonesian are not reducible to a cleft source, on the grounds that they also obtain in other types of ellipsis (e.g., pseudogapping), where a cleft source is clearly not at play. ()

a. Saya ingat Ali berdansa dengan seseorang, tapi saya tidak tahu (dengan) I remember Ali dance with someone but I  know with siapa [__]. who b. Esti harus berdansa dengan Fernando tapi Fatimah bias [__] (dengan) Ali. Esti must dance with Fernando but Fatimah can with Ali

See also Stjepanović (, ) and Leung () for a comparable line of argumentation in Serbo-Croatian and Emirati Arabic, respectively. Sato takes these data as an indication that the nature of the ban on P-stranding in unsluiced questions varies across languages. He hypothesizes in Polish, Indonesian, and other languages, this ban can be characterized as a surface restriction that ellipsis can void, whereas in the languages that Rodrigues et al. examine, it is a “deeper” constraint that also holds under ellipsis. The nature of the ban, then, dictates whether P-stranding effects under sluicing require an evasive source or not. This approach, if correct, implies that literal repair effects under sluicing are possible (see also the conclusion of §...), contrary to the conjecture outlined in §...

... German Philipp () observes that the fronted wh-phrase in the unsluiced question (a) can contain an anaphor (sich) bound by the subject; note that this anaphor cannot be replaced with a coindexed pronoun (ihm), as this would induce a Condition B violation. The prediction of a Ross–Merchant analysis is that this anaphor/pronoun asymmetry should 19 Nykiel’s article reverses Szczegielniak’s () previous claims that Polish patterns together with Spanish and Brazilian Portuguese. Note, however, that Szczegielniak introduces an unnecessary confound by focusing exclusively on the preposition z ‘with’, which assigns the same instrumental case that cleft pivots receive. Independently of this issue, I have attempted to replicate Nykiel’s reported judgments in an informal setting, but the two Polish speakers I polled rejected P-stranding sluices except in cases where both the remnant and the correlate carried instrumental case morphology.

   



be preserved under sluicing, but as (b) shows, it is not—in fact, Philipp reports that some of the speakers she tested had a slight preference for the pronoun over the anaphor (see also Ginzburg and Sag :  for a similar pattern in English fragments). () Peteri hat einige Bilder von sichi an der Wand aufgehängt . . . Peter has many pictures of self on the wall hanged a. . . . aber ich weiß nicht, welche Bilder von { ✓ sichi / * ihmi } eri but I know not which pictures of self him he aufgehängt hanged

hat. has

b. . . . aber ich weiß nicht welche Bilder von { ✓ sichi / ✓ ihmi } [__]. But I know not which pictures of self him It is tempting to analyze the pronoun version of (b) as stemming from deletion of a cleft, given that a pronoun is grammatical in this environment. ()

. . . aber but

ich I

weiß know

nicht, not

welche which

Bilder pictures

von of

ihmi him

es it

sind. are

However, this is not a general solution, given that the grammaticality of the pronoun persists even in sluices where a cleft source is not possible—e.g., sluices with a prepositional remnant (a) or with a morphologically non-nominative remnant (b). These two types of remnants block a cleft source because both are illicit cleft pivots. () a. Peteri Peter über about

hat über einige Bilder von sichi gesprochen, aber ich weiß nicht has about many pictures of self talked but I know not welche Bilder von { ✓ sichi / ✓ ihmi } [__]. which pictures of self him

b. Peteri Peter weiß know

hat ein-en Schnappschuss von sichi an der Wand aufgehängt, aber ich has a- snapshot of self on the wall hanged but I nicht, welche-n Schnappschuss von { ✓ sichi / ✓ ihmi } [__]. not which- snapshot of self him

The conclusion we can extract from this paradigm, then, is that sluicing interacts with Binding Theory in ways that we do not yet understand.

.. Morphological case parallelism effects Van Craenenbroeck (, b) observes that the availability of cleft-based sluices in a variety of languages is subject to a morphological case identity condition. Consider the following examples: the ungrammaticality of a dative remnant wem in (a) is unsurprising, given that it requires an illicit stranding of the dative-assigning preposition mit ‘with’ inside the sluicing site. More surprising is the fact that a nominative remnant wer is also





ungrammatical, which suggests that, unlike in Spanish (see §...), the P-stranding violation cannot be circumvented through an underlying cleft. Note that the corresponding unsluiced cleft is grammatical (b), so the ungrammaticality of a nominative remnant in (a) is purely an ellipsis effect. () Peter hat mit jemand-em Peter has with someone-

gesprochen . . . talked

a. . . . aber but

ich I

weiß know

nicht not

{ * we-m who-

b. . . . aber but

ich I

weiß know

nicht, not

we-r who-

/ * we-r } [__]. who- es it

ist. is

One cannot attribute the ungrammaticality of a cleft source in (a) to a language-wide ban on non-isomorphic sluices. For one, we already saw in §.. that German allows copulabased sluices as a means to circumvent LBC violations. Moreover, van Craenenbroeck also points out that P-stranding effects (and, by extension, the possibility of a cleft source) become licit under nominative case syncretism—consider the grammaticality of (), where both etwas ‘something’ and was ‘what’ are nominative-accusative syncretic. Van Craenenbroeck further shows that this pattern also holds in a number of languages beyond German (e.g., Greek, Czech, Hungarian . . . ). () Peter hat über etwas gesprochen, aber ich weiß nicht was [__]. Peter has about something. talked but I know not what./ On the basis of these data, van Craenenbroeck proposes that sluicing is subject to a morphological case parallelism requirement between the remnant and its correlate. Note that this requirement is a very superficial one, as it has to allow for abstract Case mismatches if case syncretism obtains, as in (); Sag and Nykiel () and Barros (b) reach the same conclusion. However, a requirement along these lines is too strong, as there exist multiple examples of licit sluices with non-identical case morphology (see Vicente a for a non-exhaustive list). We already saw in §.. that German LBE sluices are grammatical despite the fact that the correlate is case-marked and the remnant is not. One might want to dismiss these sluices on the grounds that the lack of case morphology on the remnant satisfies the case identity requirement trivially, but this line of reasoning would not apply to languages that display licit sluices where the remnant and its correlate carry distinct case morphemes. For example, Sakamoto () points out that, in Mongolian, remnants of embedded sluicing carry accusative case morphology, rather than the case morphology of their correlates.20

20

Given that Mongolian embedded subjects are typically accusative, the pattern in () suggests that Mongolian sluices invariably stem from a cleft or copular clause (i.e., I don’t know who it is). Notably, embedded subjects can be nominative, under certain circumstances (Klein et al.  and references), so one would expect () also to be grammatical with a nominative remnant. As of this writing, I have not been able to check if this is true.

   



() Bat hennegen-d ene nom-ig ug-sun, gevch bi { ✓ hen-ig / * hen-d} Bat. someone- this book- give- but I who- who- ’n med-eh-gui.  know-- ‘Bat gave someone this book, but I don’t know who.’ Accounting for the observed distribution of case (non)-identity effects probably requires integrating a more sophisticated analysis of morphological case assignment and case syncretism (e.g., Caha , among others) into our current analyses of sluicing. As far as I know, this integration has not been attempted yet.

. C  

.................................................................................................................................. We can summarize this chapter as follows: sluicing consists of a number of subtypes, each one exhibiting a number of properties and restrictions. Given that the properties and restrictions of any given subtype tend to be the properties and restrictions of some independently available strategy of wh-question formation, the most parsimonious analysis is the one advanced by Ross and Merchant: sluicing is a purely PF-side operation, i.e., nonpronunciation of a regular wh-question, without affecting its syntactic or semantic properties.21 At the same time, we need to contend with the fact that it seems to be only a tendency (albeit a remarkably strong one), rather than an exceptionless generalization: we have seen in §. that there exist cases of sluicing that appear to resist assimilation to a pure deletion approach. Determining whether they are genuine exceptions (and if so, which factors exactly allow them to be exceptional) is arguably one of the main issues that an eventual comprehensive, insightful theory of the taxonomy of sluicing must resolve.

A I want to thank the editors of this volume, Matt Barros, Sandra Chung, Masaya Yoshida, and especially Jason Merchant for very insightful comments on a previous draft. All errors and shortcomings remain my responsibility.

21

In and of itself, this is a surprising result. Sluicing in the languages that are discussed here is a case of surface anaphora, in the sense of Hankamer and Sag (). But are there languages with deep anaphoric sluicing—that is, languages where the sluicing site is an unstructured proform? These would be languages that would behave exactly as predicted by Chung et al. (), i.e., they would not exhibit any locality or connectivity effects contingent on there being an internal syntax to the sluicing site, they would regularly allow sluicing with non-linguistic antecedents, etc. At present, I do not know of any languages where sluicing behaves deep-anaphorically in the general case.

  ......................................................................................................................

                ......................................................................................................................

    

. W   ?

.................................................................................................................................. A the name leads one to suspect, predicate ellipsis is a type of ellipsis that leaves the main predicate of the clause unpronounced, most often together with one or more of its internal arguments or (low) adjuncts. Unlike clausal ellipsis, predicate ellipsis typically does not affect the canonical subject position and the finite verbal element. A prototypical example of predicate ellipsis is VP-ellipsis as it occurs in English: ()

Mr Honeypie has never eaten pumpkins, but his wife (surely) has.

What is missing in the second clause in () is the verbal predicate and its direct object, i.e. [eaten pumpkins]. The external argument which appears as the subject his wife, the (high) adverb surely, and the finite auxiliary has are unaffected by ellipsis. This phenomenon, especially in English, is probably the most widely discussed elliptical construction of all, and has a great amount of literature devoted to it. Some important publications on this topic include Hankamer and Sag (), Sag (a), Williams (c), Zagona (), Hardt (), Fiengo and May (), Lobeck (), Fox (), Johnson (b), Goldberg (), and van Craenenbroeck (), but this list is by no means exhaustive. The second section of this chapter discusses the distinctive properties and various analyses of VP-ellipsis (henceforth VPE) in more detail. VPE is not the only kind of ellipsis that targets the predicate and its internal arguments. A second type of predicate ellipsis is pseudogapping (first identified by Stump ; Levin , ). This phenomenon elides the main verbal predicate, parallel to VP-ellipsis, but leaves one argument or (low) adjunct unaffected. As we saw with VPE, the finite auxiliary and the surface subject are not elided either. An example is given in (), in which the main verbal predicate buy is elided, but the direct object lilies remains unaffected (just like the subject others and the finite auxiliary did). ()

Some brought roses, and others did lilies.

 



This type of ellipsis is called pseudogapping because it is reminiscent of another elliptical phenomenon, gapping, exemplified in (). ()

Some brought roses, and others lilies.

The difference between gapping and pseudogapping is that pseudogapping does not include the inflectional domain—in this case the finite auxiliary—in its ellipsis site, whereas gapping does (for more on gapping, see Chapter  of this handbook). Despite the name, most accounts of pseudogapping consider this phenomenon to be a subtype of VP-ellipsis. In the mainstream generative analysis, a VP-internal remnant with contrastive focus is assumed to have moved out of the ellipsis site prior to the ellipsis. Section . zooms in on pseudogapping and discusses various analyses that have been proposed, which differ mostly in terms of the type of movement that extracts the remnant out of the ellipsis site (see Jayaseelan ; Johnson ; Lasnik a, b, ; Kennedy and Merchant a; Takahashi , ; Gengel ; Merchant a; and Aelbrecht ), although Hardt () and Lobeck (), for instance, advocate a different approach. Since these two elliptical constructions occur in Standard English, they have received a high degree of attention in the ellipsis literature. However, recently two other kinds of predicate ellipsis have been brought to the fore. The first type is called British English do (henceforth BE do). The example of BE do in () shows that this type of predicate ellipsis looks identical to VPE, except for the addition of a non-finite form of the verb do (see Chalcraft , Haddican , Aelbrecht , Thoms b, and Baltin ). ()

Miss Coconut will wear her pink hat and Miss Crazyhorse should do, too.

The second elliptical phenomenon that was recently added to the spectrum of predicate ellipsis is called Modal Complement Ellipsis (or MCE), first discussed in Busquets and Denis (), Depiante (), Cyrino and Matos (), and Dagnac () for French, Italian, and Spanish, and Aelbrecht () for Dutch. As the examples in () show, MCE is reminiscent of VPE as it occurs in English, but is attested in languages which have been claimed not to display VPE, as illustrated in () (see Lobeck ). () a. Ik I

wil want

helpen, help

maar but

ik I

kan can

niet. not

b. Je I

veux want

aider, help

mais but

je I

ne 

peux can

ich I

kann can

nicht. not

c. Ich möchte helfen, aber I want help but ‘I want to help, but I can't.’ () a. *Ik heb I have b. *Je I

hem him

geholpen, helped

l’ai aidé, et him-have helped and

c. *Ich habe ihm geholfen I have him helped ‘I have helped him, and she has too.’

en and

zij she

heeft has

elle she

a has

aussi. too

und and

sie she

hat has

[Dutch] pas. not

[French] [German]

ook. too

[Dutch] [French]

auch. too

[German]



  

As will be discussed in section ., MCE also elides the main predicate of the clause, namely the infinitival complement of the modal, but differs from VPE in that it only occurs with root modals, not with aspectual auxiliaries. First, however, we focus on VPE as it occurs in English. Section . presents the main properties and analyses of VPE. Section . then turns to pseudogapping, and section . extends the VPE data range beyond English. The properties and analyses for both MCE and BE do are presented in section ., and section . concludes.

. E VP-

.................................................................................................................................. The ellipsis literature has been dominated by research on VP-ellipsis in English for many years, and although this chapter does not limit itself to dealing exclusively with the English variant, most of what is known about predicate ellipsis starts with English VPE (see Zagona , a, b; Lobeck , ; Hardt , ; Potsdam b; Johnson b, among others). First, in section .., we look at the licensing of VPE in English. Section .. discusses recoverability in VPE, and then the main approaches to VPE are discussed in section .., revolving around the question of whether or not there is syntactic structure in the ellipsis site. Finally, section .. investigates the size of the ellipsis site in VPE.

.. Licensing of English VPE VPE elides the main verbal predicate and its internal arguments, but leaves the finite verbal element and the subject intact. This implies that modals and finite auxiliaries survive the ellipsis, as illustrated in ().1 The examples also show that both direct and indirect objects are missing in the ellipsis clause, as well as low adjuncts. ()

a. Mr Honeypie was happily eating pumpkins, but his wife certainly wasn’t. b. Mr Honeypie will eat the carrots quickly, even though he shouldn’t. c. Has Mr Honeypie given some to his wife? Actually no, he hasn’t.

In cases where the ellipsis clause does not contain an auxiliary, dummy do is introduced. The lexical verb itself cannot survive the ellipsis: ()

a. *Mr Honeypie never liked pumpkins, but his wife surely. b. *Mr Honeypie never liked pumpkins, but his wife surely liked. c. Mr Honeypie never liked pumpkins, but his wife surely did.

Hence, these examples illustrate the generalization that VPE is found after tensed auxiliaries and not after lexical verbs. 1

Section .. deals with non-finite auxiliaries.

 



No other verbal elements or aspectual verbs selecting non-finite clauses allow for VPE, as the following sentences from Johnson (b: ) illustrate: ()

a. *Sally Tomato started running down the street, but only after José started. b. *Sally Tomato made Mag laugh, and then José made.

One way of capturing this empirical generalization is to say that it is the finite auxiliary in T that licenses VPE. This ties in with one of the conditions on ellipsis, namely syntactic licensing. As was first noted by Bresnan (: ), Sag (a), Williams (c), and Zagona (, a, b) and further explored by Martin (, ), Lobeck (, ), and Johnson (b), VPE is only allowed in clauses with an overt T head; or as Johnson (b: ) puts it: “the ellipsis site must be in construction with, or perhaps governed by, a member of ‘Aux.’ ” The instantiations of these “Aux” elements are argued to be modals, the auxiliaries have, be, and do, and the infinitival marker to. However, the issue of what licenses VPE in English is not as straightforward as this. A first point of concern has to do with VPE in non-finite clauses, with infinitival to. Not all instances of to allow for VPE (see Lobeck , , , ; Zagona a, b; Johnson b): when the infinitival clause is a complement, as in (), VPE is licensed, but not when it is an adjunct, as in () (examples from Lobeck ). ()

a. Bettina wants to hear Susanna’s story, and I also want to. b. Even though he doesn’t like to, Ron jogs every day.

()

a. *Meg Wildwood came to read Fred’s story and I also came to. (Johnson b) b. *Even though he could jog to, Ron doesn’t do anything to stay in shape.

Furthermore, when an infinitival clause occurs in subject position, VPE is disallowed, as in (), as observed by Zwicky (), and also in certain wh-clauses VPE is restricted; see ().2 We refer the interested reader to Zagona (, a, b), Zwicky (), Lobeck (), and more recently, Thoms (b) for further discussion. ()

a. *You shouldn’t play with rifles because to _ is dangerous. b. You shouldn’t play with rifles because it is dangerous to _.

()

a. *John wants to go on vacation, but he doesn’t know when to _. (Zagona a: ) b. *Mary was told to bring something to the party, so she asked Sue what to _. (Lobeck : ) c. ??Ron wanted to wear a tuxedo to the party, but Caspar couldn’t decide whether to _. (Lobeck : ; judgement from Johnson b: )

Moreover, whilst control sentences involving a to-infinitive permit VPE, this is not the case for subject-to-object raising constructions that favour stative VP complements (see Martin ): 2

We indicate the ellipsis site with an underscore here, for clarity’s sake.

 ()

   a. Bob is not certain he can fix the car, but he will definitely try to _. b. *I consider Ron to enjoy banoffee pie, and I believe Bob to _ as well.

Note, however, that VPE following infinitival to in other types of subject-to-object raising constructions is permissible (Miller p.c.): ()

a. I think this is a major fight in terms of the direction of the Democratic Party. And the media are responding as one would expect them to. (COCA) b. You have been corresponding with me, Robert. I could prove it, but I do not think you would wish me to. (COCA)

Another problem for the claim that T licenses VPE is the fact that negation also sometimes independently licenses VPE (example from Potsdam b, see also Johnson b): ()

Ted hoped to vacation in Liberia but his agent recommended that he not.

The exact contexts in which negation can license VPE are still unclear. A last remark concerning VPE licensing is that Gergel () has claimed that VPE is degraded when licensed by epistemic must. This is illustrated in (a), whereas (b) shows that the ungrammaticality disappears when the modal is followed by a non-finite auxiliary. ()

a. Do you think Ted will be sick today after last night’s pub crawl? – *He must.3 b. Do you think Ted will be sick today after last night’s pub crawl? – He must be.

However, this restriction does not appear to be absolute, as Miller (p.c.) notes the following counterexamples: ()

a. A sick feeling coiled in my stomach as I concluded that he already knew about me—he must. (COCA) b. Does he know the truth about their whereabouts? He must. (COCA)

Other epistemic modals, e.g. should or might, allow for VPE with the relevant epistemic reading: ()

a. Should he be home by now? – Yes, he should (be). b. Will he be arrested? – He might (be).

Recapitulating, English VPE requires the presence of a finite auxiliary, a modal, or the infinitival marker to (the latter only in certain syntactic environments). Sometimes, however, negation can license VPE, and VPE licensed by epistemic must is sometimes degraded. It is clear, therefore, that not all problems have been solved when it comes to

3

The asterisk here indicates that an epistemic reading is unacceptable. A deontic interpretation is fine.

 



VPE-licensing, nor do we have a thorough formal understanding of what ellipsis licensing actually is. In the next section we discuss recoverability in VPE constructions.

.. Recoverability Whether or not VPE is syntactically licensed is not the only factor determining whether or not it is licit in a given linguistic context. Recoverability also plays an important role. That is, an elided constituent must have a salient antecedent elsewhere in the linguistic discourse so that its meaning is recoverable by the hearer. In the following example, for instance, the elided constituent in the second clause can only be interpreted as punch Rocky since this phrase appears as a salient antecedent in the preceding clause. ()

Mr T punched Rocky, and Apollo did _, too.

Much debate has revolved around what can serve as an antecedent for ellipsis. Importantly, Hankamer and Sag () claim that VPE cannot be used exophorically. That is, it must always have a linguistic antecedent. Its meaning cannot be recovered via inference from the non-linguistic context. Consider, for instance, the following minimal pair: ()

[Context: Sag raises a cleaver and prepares to hack off his left hand.] a. Hankamer: *Don’t worry, he never actually does _. b. Hankamer: Don’t worry, he never actually does it.

[VPE] [do it]

The important point here is that (a) contains VPE whereas (b) does not. The fact that (a) is infelicitous, while (b) is not, illustrates that VPE is not recoverable from the nonlinguistic context (though see Schachter b, Hankamer , and in particular Miller and Pullum  for more detailed discussion and counterexamples). A further issue is to what extent the elided constituent must be identical to its antecedent in order for it to be recoverable. Mismatches between the two can be informative as to the nature of the recoverability requirement: if formal mismatches that don’t affect the (truthconditional) semantics are allowed, then this suggests that the recoverability requirement is semantic in nature. On the other hand, if even the slightest formal change leads to ellipsis no longer being an option, it seems justified to conclude that an ellipsis site has to be structurally identical to its antecedent in order for the application of ellipsis to be recoverable. The literature on VPE abounds in cases involving ellipsis–antecedent mismatches, but the theoretical conclusions that can be drawn from them invariably turn out to be more subtle and complex than they appear to be at first glance. Due to space limitations we only discuss one such case here, but see van Craenenbroeck () for an overview of other mismatches. The case we focus on is that of VPE with a nominal antecedent, as illustrated in () (from Hardt ). ()

David Begelman is a great [NP laugher], and when he does _ , his eyes crinkle at you the way Lady Brett’s did in The Sun Also Rises. (from You’ll Never Eat Lunch in This Town Again)



  

In this example, VPE (indicated by the underscore) does not take another VP as its antecedent, but an NP, in particular the noun laugher in the first clause. At first sight, then, this seems like an open-and-shut case in favour of a semantic recoverability requirement over a syntactic one: while the VP laugh and the NP laugher can arguably mean (roughly) the same thing, they are clearly not structurally identical. Johnson (b), however, uses these data to make the exact opposite claim. The key to understanding this example, he argues, is the fact that laugher is a deverbal noun. If, as proposed by Fu et al. (), such nouns feature in their underlying representation an actual VP, then there is in fact a level of representation at which the ellipsis site and its antecedent are structurally identical, and examples like () cannot be used as arguments against an approach in terms of structural identity. More recently, however, Miller and Hemforth (a) have taken issue with Johnson’s analysis. They point out that it wrongly predicts (a) that all deverbal nouns should be able to serve as VPE-antecedents, and (b) that underived (and hence nondeverbal) nouns should never be able to do so. Miller and Hemforth propose instead that VPE can only take a nominal antecedent when that NP expresses an implicit polar question. Since it is hard to see how this could be implemented in terms of a structural identity requirement, they contend that examples like () can indeed be used as an argument against such an approach. The next section addresses the issue of whether the elided constituent in VPE contains syntactic structure or not.

.. Deleted syntactic structure? As discussed elsewhere in this handbook, there are several ways of analysing VPE. Roughly, there is a distinction between structural and non-structural accounts, which provide opposite answers to the question of whether there is syntax in the ellipsis site. Nonstructural accounts assume no covert syntax in the VP-ellipsis site: under these approaches, the syntax of an elliptical sentence lines up with its phonology, and the computational burden is placed on the syntax–semantics interface in order to derive the correct interpretation (see Culicover and Jackendoff , but also accounts in the Head-Driven Phrase Structure Grammar and Construction Grammar frameworks: Pollard and Sag ; Ginzburg and Sag ; Sag et al. ; Schlangen ; Stainton a; Jacobson ; Kim and Sells ). The other end of the spectrum has the syntax lined up with the semantics of the elliptical clause, while ellipsis affects the phonology. This is known as the structural approach.4 According to this approach, which was first proposed by Chomsky (, ) and Ross (b), the syntax of an elliptical sentence does not really differ from the syntax of its non-elliptical counterpart. Ellipsis is an operation that marks part of the syntactic structure for non-pronunciation at PF, the phonological interface. It is entirely plausible that a structural approach suits certain types of ellipsis, while a non-structural approach suits other types, but we show here that VPE favours a structural analysis. A first argument has to do with agreement facts. Consider the following data:

4

Here we only present the two opposite ends of the spectrum. Some structural accounts exist in which there is only a limited amount of hidden syntactic structure involved (see e.g. Hardt ).

  ()



a. He wouldn’t believe there was a crocodile in the garden, but there was/*were [a crocodile in the garden]. b. He wouldn’t believe there were three pandas in the garden, but there *was/were [three pandas in the garden].

In these examples the finite auxiliary agrees in number with the associate that is left unpronounced. This indicates that the associate should be present in the underlying syntax, otherwise these facts are less straightforwardly accounted for. Another argument in favour of the structural approach involves extraction out of the ellipsis site. Although there are certain restrictions to be considered (see Schuyler  for extensive discussion), VPE allows for elements to be extracted from the ellipsis site. This is only possible if the ellipsis site contains enough structure to host the movement trace or copy. Some examples are given in (), showing subject extraction out of passive and unaccusative verb phrases, subject raising out of VPE, and also wh-object extraction (see also Fox  for covert movement out of the ellipsis site).5 ()

a. Laura was hit by the elephant trunk and Lola was [hit tLola by the elephant trunk], too. b. The train arrived on time, even though the bus didn’t [arrive tthe bus on time]. c. Arya seems to be tough, even though her sister doesn’t [seem to be ther sister tough]. d. I know which song I want to sing, but I can’t guess which one you do [want to sing twhich one].

In section . we will see that head movement out of the ellipsis site is possible too: in languages with V-stranding VPE the lexical verb moves to T from its base position. This section has provided a quick overview of the debate as to whether VPE involves hidden syntactic structure or not. Next, we have a closer look at the size of the ellipsis site, before turning to other types of predicate ellipsis.

.. The size of the ellipsis site In the introduction, predicate ellipsis was presented as a kind of ellipsis that targets the predicate of the clause, and the name VP-ellipsis suggests that it is the VP(-predicate) that is targeted. However, VPE does not necessarily involve a verbal predicate. Prepositional, nominal, or adjectival predicate phrases can be elided too, as () shows:

5 Jacobson (), Hardt (), and Lobeck () argue that these cases do not involve extraction at all, but are in fact pseudogaps with a fronted remnant. Under their account, VPE does not allow for extraction and therefore does not necessarily contain syntactic structure. However, as Johnson (b) points out, pseudogapping is more restricted than the extraction cases with VPE: not all phrases that can move out of a VP-ellipsis site can act as pseudogapping remnants. See Miller (), however, for critical discussion of Johnson (b).



  

()

a. Sam was already in the kitchen, even though he really shouldn’t be. b. Ted will be a famous writer some day, and Ed will be, too. c. Clara has always been pretty, and no doubt her daughter will be, too.

This suggests that VPE uniformly targets all types of predicates in finite clauses. But what exactly is to be considered part of the predicate is less straightforward. So far, we have seen that VPE elides more than just the (verbal, nominal, adjectival, or prepositional) predicate; it can also elide verbal complements and VP-adjuncts. What is not elided are the subject and the finite auxiliary in T . However, more elements can be present in the clause between T and VP. Within the Minimalist framework the VP-layer is standardly assumed to be dominated by a functional projection vP, which hosts the base position of the predicate’s external argument. Most analyses take VPE to include at least this vP-layer in the ellipsis site as well (see Johnson b, ; Merchant , a, d). One piece of evidence is found in there-existential sentences: the associate of there in unergative or transitive existential surfaces in spec,vP, but under VPE this associate is obligatorily elided, as shown in ().6 () a. I didn’t know there was someone talking to Rebecca, but there was [someone talking to Rebecca]. b. *I didn’t know there was someone talking to Rebecca, but there was someone [talking to Rebecca]. (Aelbrecht : ) Non-finite auxiliaries also help us identify the size of VPE. Consider the sentence in (a), with the largest possible sequence of auxiliaries in English: ()

a. Betsy must have been being hassled. b. finite modal > perfect HAVE > progressive BE > passive BE > lexical verb

Akmajian and Wasow () and Sag (a), and more recently Thoms (), Bošković (), Sailor (), and Aelbrecht and Harwood (), observed that when VPE is applied to such a sequence, the following pattern emerges (from Sag a: ): ()

Betsy must have been being hassled by the police, and . . . a. *Peter must have been being hassled by the police, too. b. Peter must have been being hassled by the police, too. c. Peter must have been being hassled by the police, too. d. *Peter must have been being hassled by the police, too.

This illustrates that perfect have cannot be elided (see (a,b)), whilst been is optionally elided (see (b,c)), and being is obligatorily included in the ellipsis site (see (c,d)). We

6 Johnson (b) provides another argument in favour of this claim involving different interpretations of the adverb again. Merchant (a, d) adds a third, showing that argument structure alternations are not allowed between the VPE site and its antecedent.

 



briefly review some accounts of this data pattern (see also Lasnik b, Potsdam b, and Nunes and Zocca  for discussion of the behaviour of auxiliary verbs under VPE).7 The non-ellipsis of have is rather straightforward under most approaches: given that have always precedes forms of the auxiliary be, the former is taken to be higher in the structure than the latter, specifically in a position that is external to the ellipsis site. Consequently, have is never elided. Most accounts also assume, one way or another, that the position of being is always included in the ellipsis site.8 Bošković () and Sailor (), following Akmajian and Wasow (), Iwakura (), Akmajian, Steele, and Wasow (), Lobeck (), Bošković (a), and Thoms (b), claim that being does not raise for inflectional purposes and instead has its inflection lowered onto it in its base position of Voice , where it is subsequently elided (cf. (a)). Aelbrecht and Harwood (), on the other hand, argue that being raises into a progressive aspectual projection in order to receive its non-finite -ing inflection, but that the ellipsis site includes this landing site: under their account VPE targets the progressive aspectual layer above VoiceP (cf. (b)): ()

a. Peter must have been [ProgP t-ing [VoiceP being hassled by the police]. b. Peter must have been [ProgP being [VoiceP tbeing hassled by the police].

The optional ellipsis of been has also received two possible accounts. Thoms (), Sailor (), and Aelbrecht and Harwood () argue that been can optionally move out of the ellipsis site, thus optionally surviving VPE: ()

a. The police must have been [ProgP tbeen hassling Peter]. b. The police must have [ProgP been hassling Peter].

On the other hand, instead of optional raising and a fixed ellipsis site, Bošković (), as well as Akmajian, Steele, and Wasow (), argues that the size of the ellipsis site can fluctuate: it does not contain been, but can be optionally extended to include the perfect aspectual projection, which Bošković () assumes been has risen into for inflectional purposes:9

7

The claim that perfect have is never elided under VPE is arguably too strong. Although the early references do not acknowledge it, subsequent literature on VPE has recognized that perfect have can be elided under certain circumstances and in some dialects (see discussion in Aelbrecht and Harwood ). In particular, Sailor (: –) shows that in certain ellipsis–antecedent configurations, ellipsis of perfect have becomes possible, and is even preferred by speakers of particular varieties of English: (i) If you’d been there, he’d have got you one too, I know he would.
(BNC) The opposite can be said about being: some authors have claimed that it can be stranded in certain dialects (see also Chapter  in this handbook). (ii) %Remember, always be respectful and courteous, even if the officer isn’t being. 8 See Baker, Johnson, and Roberts (), however, for an alternative account in which the ellipsis of being is ascribed to a ban on ellipsis following V+ing forms in general. See Aelbrecht and Harwood () for critical discussion of this approach. 9 Earlier work by Akmajian and Wasow () represents a third option: they proposed optional Auxdeletion operations, which could occur separately from VPE. See Sag (a), however, for arguments against this two-operation approach to VPE size variability.

 ()

   a. The police must have [PerfP been [ProgP tbeen hassling Peter]. b. The police must have [PerfP been [ProgP tbeen hassling Peter].

Whatever analysis one adopts, bringing non-finite auxiliaries into the picture leads to the conclusion that the VP ellipsis site must be larger than simply VP or vP, extending as far as VoiceP, ProgP, or even PerfP. In addition, Sailor () claims that the size of VPE in English differs depending on whether the ellipsis clause is coordinated with or subordinated to its antecedent. It should be clear by now that the last has not been said on this topic. This section has discussed the licensing and recoverability requirements of VPE in English, the question of whether the elided constituent involves syntactic structure or not, and the size of the ellipsis site in English VPE. The next section focuses on a different kind of predicate ellipsis that is standardly considered a subtype of VPE, but displays some notable differences: pseudogapping.

. P

.................................................................................................................................. Pseudogapping is a kind of ellipsis that was first named by Stump (), who argued that it was similar to gapping (see also Levin , ). Both kinds of ellipsis seem to target nonconstituents (or discontinuous strings) of a clause. ()

a. Some are bringing roses to the party and others are [bringing] lilies [to the party]. (Pseudogapping) b. Some are bringing roses to the party, and others [are bringing] lilies [to the party]. (Gapping)

Because of this discontinuity, pseudogapping has been viewed as a subtype of VPE involving movement: in pseudogapping a contrastive phrase has moved out of the VP prior to the occurrence of VPE, hence surviving the ellipsis (see Jayaseelan ).10 This is illustrated in (). ()

a. Some brought roses, and others did lilies. b. Some brought roses, and others did lilies [VP bring tlilies].

Many have adopted this view and the main differences between the proposals are the motivation for the movement out of the ellipsis site and the direction of this movement (see Johnson ; Lasnik a, b, ; Kennedy and Merchant a; Takahashi , ; Merchant a; Aelbrecht ; Gengel ).11 If this is on the right track, pseudogapping strengthens the argument in favour of syntactic structure inside the 10

Toosarvandani () and Potter () have recently attempted to also analyse gapping as VPE. See Chapter  of this handbook for more discussion about gapping. 11 See Miller (, ), Hardt (), Lobeck (), and Thoms (b), however, for an alternative view.

 



ellipsis site in VPE, as it instantiates another case of extraction out of that ellipsis site (see section ..). First, we discuss some basic properties of pseudogapping, especially those in which this phenomenon differs from VPE. Then we come back to the question of the movement operation responsible for extracting the remnant phrase out of the verb phrase.

.. Properties of pseudogapping Pseudogapping seems to be licensed by the same mechanisms that license VPE, namely either the modal verb, dummy do, or finite auxiliary verb; cf. (). As () illustrates, finite main verbs or other aspectual verbs do not license it, parallel to VPE. ()

a. Does that make you mad? It would me! (Gengel : ) b. Some brought roses and others did lilies. c. I’m not citing their analysis so much as I am their data. (Levin : , ex. ())

()

a. *It started bothering me more than it started her. b. *He seems to care about his job more than he seemed about his family. c. *You read the book yesterday and I read today.

However, the distribution of pseudogapping is more restricted than that of VPE. First of all, pseudogapping gives rise to variable judgements depending on the sentence and the speaker. While all speakers accept it in comparatives and other subordinate structures, they are less widely accepted (Levin : ; Lasnik a) and far less common in coordinations: in Miller’s () corpus study of pseudogapping, coordinated pseudogaps constitute only . percent of all pseudogapping examples. Second, the infinitival marker to does not license pseudogapping, unlike VPE (cf. ()). ()

a. *Bettina tried to write down a nice quote every day, just like she tried to a song title. b. *Bettina tried to read the paper before she tried to the book.

Thirdly, unlike VPE, pseudogapping generally does not allow for backwards ellipsis (examples from Levin : ):12 ()

a. *Although I don’t know if Tom does books, I know he writes magazines. b. Although I don’t know if Tom does, I know Harry writes magazines.

A fourth asymmetry traditionally believed to hold between VPE and pseudogapping is that the latter does not allow for voice mismatches, while the former does (cf. Merchant a, d; 12

Miller (), however, notes the following counterexample:

(i) Behind them, disguising her desire, one catches a poignant glimpse of the youthful, shaved-headed Cather. As it did me, work rescued Willa Cather.



  

but also Sag a; Dalrymple et al. ; Hardt ; Fiengo and May ; Kehler , ; Johnson b; Arregui et al. ; and Frazier and Clifton  for more examples and discussion): () a. This problem was to have been looked into, but obviously nobody did. (VPE; Kehler : ) b. The janitor must remove the trash whenever it is apparent that it should be. (VPE; Merchant a: ) c. * More people were invited to Beth’s reception by her mother than Beth herself did to her wedding! (Pseudogapping; Merchant a: ) d. *Beth’s mother invited more people to her wedding than were by Beth herself! (Pseudogapping; Merchant a: ) These data are brought forward by Merchant (a, d) as supporting evidence for the claim that pseudogapping elides a larger part of the structure than VPE. He assumes ellipsis to be subject to a strict syntactic identity condition, which states that the morphosyntactic specification of the ellipsis site matches that of the antecedent. He explains the contrast between VPE and pseudogapping by arguing that VPE elides vP, thus not including the Voice head in the ellipsis site, while pseudogapping targets VoiceP. Because in the latter case, the Voice head is part of the ellipsis site, the voice specification in the ellipsis site has to match that of the antecedent.13 However, Miller () has recently shown Merchant’s generalization to be incorrect, with attested examples from corpora, illustrating that voice mismatches are just as acceptable in pseudogapping as they are in VPE:14 ()

a. A whole poached wild striped bass should be taken to the table as you would a Thanksgiving turkey or a crown roast of pork. b. I mean for her to be dressed—and addressed—as we would Becky Sharp.

As a result, the voice mismatch data no longer seem to bear on the debate surrounding the size of the ellipsis site involved in VPE and pseudogapping, and if anything, only highlight the syntactic similarity between the two phenomena.

13

Recall, however, the discussion in the previous section: since being is always elided, several authors have claimed that VPE targets at least as much as VoiceP. These approaches are incompatible with Merchant’s explanation for the voice mismatch contrast. 14 An additional complication is that voice mismatches in VPE are themselves not always acceptable (Kehler ; Kertz ; examples from Sailor : ): (i)

a. *The janitor removed the trash, but the recycling wasn’t. b. *This guy’s tape should be scrutinized by John, and Bob also should.

On the basis of these data Sailor () claims that there are two types of VPE. When the ellipsis clause is coordinated with its antecedent, VPE includes VoiceP, and voice mismatches are disallowed. However, when the ellipsis clause is subordinate to its antecedent clause, VPE targets a constituent smaller than VoiceP, and voice mismatches are permitted.

 



As was pointed out above, most accounts argue that pseudogapping is a subtype of VPE. In the mainstream generative analysis, a VP-internal remnant with contrastive focus is assumed to have moved out of the ellipsis site prior to the ellipsis. What the analyses disagree on mostly is what kind of movement is responsible for the evacuation of the remnant. The different approaches are presented in the next section.

.. The nature of the movement operation Pseudogapping is traditionally analysed as movement of an element out of the verb phrase plus VP-ellipsis (Jayaseelan ; Lasnik a, b, ; Merchant a, b; Gengel ). Crucially, this movement does not seem to take place in non-ellipsis, or is at least disguised by further movement of the verb. In this section, we give an overview of the main proposals. For extensive discussion of the arguments for and against A- and A’-movement in pseudogapping we refer the reader to Johnson (b) and Gengel (). As a first proposal, Jayaseelan () claims that the remnant is moved out of the ellipsis site by Heavy NP Shift (HNPS): it adjoins to the right of the clause, as in (). ()

Some brought roses and others did [brought tlilies] lilies.

The examples below, however, show that the remnant is not necessarily a heavy NP (or PP): it can also be a pronoun, which cannot undergo HNPS (cf. (a,b)). ()

a. Did it frighten you to hear that they found a bomb as much as it did me? b. *Did it frighten you to hear that they found a bomb as much as it frightened to hear that they found a bomb me?

Furthermore, Takahashi (, ) argues against an approach with only HNPS on the basis of pseudogapping with multiple remnants. The sentence in (a) (= Jayaseelan :  ()) cannot be derived by rightward movement of both remnants because multiple applications of HNPS are disallowed in non-elliptical contexts (cf. (b) = Jayaseelan :  (); see also Gengel , and see Shiobara  and Park and Kim  for alternative views on multiple HNPS). () a. I didn’t give a dime to Mary, but I did a nickel to Jane. b. *It proved to the jury his guilt that Michel was seen with the murder weapon. (vs It proved his guilt to the jury that Michel was seen with the murder weapon.) Moreover, Lasnik (a, b) observes that unlike pseudogapping, HNPS is impossible with the first object of a double object construction: ()

a. ?Steven will give Bill a car and Susan will give Anne a car. b. *Steven [gave ti a lot of money] [the fund for wildlife and nature preservation]i.

Thus, HNPS cannot fully account for moving the pseudogapping remnant out of the ellipsis site. Another proposal, advocated by Lasnik (a, b, and subsequent work),



  

argues that the remnant undergoes Object Shift (OS) to the left of the verb phrase (Vanden Wyngaerd ; Chomsky ; Koizumi ), parallel to Scandinavian OS. This movement is triggered by an optional feature that attracts the pseudogapping remnant when present. Lasnik additionally takes the main verb in English to move to a position preceding the object in non-ellipsis. Under pseudogapping such verb movement is bled and the verb is deleted with the rest of the VP. However, the movement in pseudogapping behaves differently from Scandinavian OS. Firstly, the latter does not target prepositional objects (see ()), unlike pseudogapping in both English and Scandinavian (Gengel ). () Jón talaði John spoke to Mary ‘John didn’t speak to Mary.’ ()

ekki not

. Mary (Icelandic)

John spoke to Mary, and Bill did to Lisa.

Secondly, Scandinavian OS is blocked when an auxiliary is present in the sentence, i.e., it only occurs when the main verb moves out of the verb phrase (Holmberg’s Generalization, cf. Holmberg ). Given that in pseudogapping an auxiliary is crucially present, the two phenomena do not pattern together (see also Gengel ). In short, object shift does not seem to be responsible for the movement of a pseudogapping remnant either. Takahashi () therefore proposes a hybrid account and argues that pseudogapping is derived by either HNPS or OS, depending on the context. Another alternative is put forward by Jayaseelan (, ), who observes that pseudogapping requires the moved element to bear contrastive focus. Gengel () builds directly on this idea and argues that Focus movement is at stake in pseudogapping, not OS or HNPS. Gengel assumes that there is a clause-internal focus projection above VP that attracts the pseudogapping remnant to its specifier (following Jayaseelan  for Malayalam and cleft constructions in English). For arguments in favour of such a focus position, see Belletti and Shlonsky () for Italian and Hebrew, and É. Kiss () for Hungarian. Summing up, there is much debate as to which operation is responsible for moving the pseudogapping remnant. Jayaseelan () argues that Heavy NP Shift (HNPS) is involved, while Lasnik (a) claims the remnant to have undergone Object Shift (OS). Takahashi () takes pseudogapping to be derived by either of these two operations, and finally, Gengel () takes focus movement to be at stake (see also Jayaseelan ).15 The next section turns back to VPE, but extends the data set to languages other than English.

15 See also Johnson (b) for the hypothesis that pseudogapping involves a movement operation that resembles Dutch scrambling, and Miller () for general critical discussion of the movement analysis of pseudogapping.

 



. VPE -

.................................................................................................................................. English has dominated the literature on ellipsis, and particularly VPE. Compared to other well-documented Western European languages, such as German, French, Italian, Spanish, and Dutch, English is the odd one out, as none of these other languages display VPE. However, that is not to say that VPE is restricted to English. It has also been argued to exist in Danish (Houser et al. ), Persian (Toosarvandani ), Finnish (Holmberg ), Galicean (Rouveret ), Hebrew (Sherman (Ussishkin) ; Doron ), Hungarian (Bartos b), Irish Gaelic (McCloskey a, ), Japanese (Otani and Whitman ), Korean (Otani and Whitman ; Kim , though see Park  for an opposing view), Libyan Arabic (Algryani ), Ndenduele (Goldberg ), Polish (Sczcegielniak ), Brazilian and European Portuguese (Raposo ; Martins ; Cyrino and Matos , ; Santos b), Russian (Gribanova b), Samoan (Sailor ), Scottish Gaelic (Sailor ), Serbo-Croatian (Stjepanović ), Swahili (Ngonyani ), Taiwanese (Sailor and Kuo ), and Welsh (Rouveret ). We focus in this section on what is known as V-stranding VPE languages such as (European and Brazilian) Portuguese, Hebrew, and Irish.16 An interesting property of these languages is that VPE only appears to target the material following the lexical verb, namely its internal arguments, and does not include the lexical verb itself: () A: Šalaxt etmol et ha-yeladim le-beit-ha-sefer? send. yesterday  the-children to-house-the-book

[Hebrew]

B: Šalaxti. send. ‘A: Did you send the children to school yesterday? B: I did.’ () Dúirt mé go gceannóinn said I that buy ‘I said I would buy it and I did.’

é it

agus and

() O João leu esse livro e a the João read that book and the ‘João read that book and Ana did too.’

cheannaigh. bought

Ana Ana

também too

[Irish]

leu. read

[Portuguese]

A complication is that some of the relevant languages also exhibit object drop. Portuguese, for instance, allows for a null object in (a) (Raposo ). One could therefore argue that Portuguese does not in fact have VPE and that what looks like VPE in () is a null object construction. However, closer scrutiny reveals that in Portuguese, specific instances of ellipsis must be analysed as VPE. For instance, the typical null object cases only involve

16

Though there are differences between European and Brazilian Portuguese when it comes to their V-stranding VPE properties; see e.g. Cyrino and Matos () for discussion.



  

direct objects, while VPE elides all verbal complements; see (b) (Raposo ; examples from Cyrino and Matos ). () a. Ela tirou o anel do dedo e guardou [-] no cofre. she took.off the ring from finger and put in safe ‘She took off the ring from her finger and put it in the safe.’ b. A Ana não leva o computador para as aulas, porque os the Ana not brings the computer to the classes because the amigos também não levam [tlevam o computador para as aulas]. friends too not bring the computer to the classes ‘Ana doesn’t bring her computer to class because her friends don’t either.’ Data such as these show that proper identification is a crucial step in VPE research: one needs to distinguish cases of object and topic drop from actual VPE. In this vein, Goldberg () has argued that Japanese and Korean do not exhibit genuine V-stranding VPE, and that such cases can be ascribed to object and topic drop. Similar claims have been made for Russian (see Bailyn ; Erteshik-Shir, Ibn-Bari, and Taube , ; and Gribanova a, b for discussion), though the matter is far from closed. Nevertheless, if the Hebrew, Irish, and Portuguese examples in (), (), (), and (b) are genuine cases of VPE, how does the lexical verb survive the ellipsis in such cases? Unlike English, these languages display main verb movement: when the main verb is finite, it raises to the T -head to receive its inflection. Earlier we have seen that both A- and A’-movement are allowed out of the VP ellipsis site. The data in (), (), (), and (b) illustrate that the same holds for head movement: when the verb undergoes V-to-T movement, it raises out of the ellipsis site and so survives predicate ellipsis. Therefore these languages are called V-stranding VPE languages (see also Goldberg ).17 It is worth noting, however, that the analysis presented here is a purely transformational approach. Many other frameworks would not necessarily claim that languages such as Portuguese, Irish, and Hebrew actually exhibit VPE, though it remains to be seen exactly how those frameworks would tackle such data. The fact that the main verb can survive the ellipsis in these languages brings along a new identity restriction. Goldberg (: ) calls this the Verbal Identity Requirement: () Verbal Identity Requirement (VIR) The antecedent- and target-clause main Vs of verb-stranding VP ellipsis must be identical, minimally, in their root and derivational morphology. This is illustrated in the Irish example in (), from Goldberg (: ). The main verb thuig has moved out of the ellipsis site, and is therefore recoverable, but still VPE is disallowed, because the verb needs to be identical to the verb in the antecedent.18 17

Other languages that are only Aux-stranding, like English, are Taiwanese and Mainland Scandinavian. See Sailor (, b) for general discussion. 18 The principle in (50) requires the two verbs to be identical not just in their root, but also in their derivational morphology. See Goldberg (: –) for further illustrations of this.

  ()

*Léigh mé an dán ach níor read. I the poem but not. ‘I read the poem but I didn’t understand it.’



thuig. understand.

No such requirement applies to phrasal movement out of a verbal ellipsis site, as the following examples with subject and object extraction illustrate: the subject in (a), Summer, differs from the subject in the antecedent, Clara; and the same holds for the topicalized objects in (b). ()

a. Clara was fired, and Summer was [fired tSummer], too. b. The film I haven’t seen yet, but the TV series I have [seen tthe TV series].

This contrast in identity requirements between phrasal movement and head movement out of the ellipsis site poses an interesting puzzle that is the subject of ongoing debate in ellipsis research. It should be noted, however, that counterexamples to the VIR have been noted in the literature. Gribanova (b: ), for instance, notes the following data from Russian in which the stranded lexical verb in the ellipsis clause does not match the lexical verb in the antecedent: ()

Kto-to ètu vazu uronil, i tot fakt, čto nikto someone this vase dropped, and the fact that no-one [eë] ne podnjal, menja ogorčaet. it not picked-up me upsets ‘Someone dropped this vase, and the fact that no one picked (it) up upsets me.’

It seems therefore that the VIR is not an entirely robust generalization for all V-stranding VPE languages.19 A closely related phenomenon to V-stranding VPE is that of little v-stranding VPE, as discussed in Toosarvandani () for languages such as Persian (see also Chapter  of this handbook). This language is argued to exhibit complex predicates consisting of a light verb in v , and a nominal, adjectival, or prepositional element in its complement: () Rāmin farsh-o jāru Ramin carpet- broom ‘Ramin swept the carpet.’

kard. 

(Toosarvandani : )

Toosarvandani () argues that VPE exists in Persian, except it does not affect any verbal elements within the clause, only the nominal, adjectival, or prepositional element of the complex predicate, and its internal arguments.20 The light verb is stranded by such ellipsis (example from Toosarvandani : ): 19

See Lipták (), Rouveret (), Schoorlemmer and Temmerman (), and Gribanova (b) for more discussion. 20 Toosarvandani () argues for the existence of VPE in Persian on the basis of identity requirements, extraction, and the lack of pragmatic control.

 ()

   Sohrāb piranā-ro otu na-zad vali rostam [piranā-ro otu] zad. Sohrab shirt- iron - but Rostam shirt- iron  ‘Sohrab didn’t iron the shirt but Rostam did.’

Unlike the main verb in V-stranding VPE, however, v is not stranded by raising out of the ellipsis site. Using the distribution of the adverb dobāre (again) as a diagnostic for the position of the light verb, Toosarvandani () argues that it does not undergo v-to-T movement, and instead remains in its base position. For the light verb to survive VPE this would imply that in Persian, only the complement of v is elided. vP itself is not the target of VPE. This shows that VPE in Persian targets a much smaller constituent than in English, and, more generally, that the size of VPE can differ cross-linguistically. This section expanded the VPE data set to languages other than English. In section ., we deal with instances of predicate ellipsis that have only recently been getting more attention, namely Modal Complement Ellipsis and British English do.

. O  : M C E  B E 

..................................................................................................................................

.. Modal Complement Ellipsis The previous section has shown that VPE is not restricted to English; it also occurs in Hebrew, Irish, Swahili, Portuguese, and several other languages. It is remarkable, then, that VPE is unattested in many languages related to these, such as German and Dutch for the Germanic languages and French, Italian, and Spanish in the Romance family. The question of why VPE has such a limited distribution is one of the most prominent and long-standing issues surrounding this construction. However, these languages display a type of predicate ellipsis following modal verbs. The phenomenon was first discussed in Busquets and Denis (), Depiante (), Cyrino and Matos (), and Dagnac () for French, Italian, and Spanish, and Aelbrecht () for Dutch, under the names of Modal Ellipsis (or ME) and Modal Complement Ellipsis (or MCE). The properties of M(C)E differ slightly between these languages, and this section focuses on the Dutch variant. In Dutch MCE, the infinitival complement of root modals can go unpronounced, as in (): () Ik wil wel helpen, maar I want  help but ‘I want to help, but I can’t.’

ik I

kan niet can not

[helpen]. help

(Dutch)

Unlike VPE, MCE only occurs with modals, not with aspectual auxiliaries (see (a)), and even epistemic modals do not allow for the ellipsis (cf. (b)).

 



() a. *Ik had gezegd dat ik zou helpen, maar ik heb niet [geholpen]. I had said that I would help but I have not helped ‘I had said that I would help, but I haven’t.’ b. *Hij zegt dat hij al klaar is met zijn huiswerk, maar hij kan he says that he already ready is with his homework but he can toch niet [al klaar zijn met zijn huiswerk]; hij is net thuis.  not already ready be with his homework he is just home ‘He says he’s already finished his homework, but he can’t have; he just got home.’ When it comes to the ellipsis site, the examples in () show that MCE targets not only the main infinitive and its complement, but also the aspectual (cf. (a)) and voice (cf. (b)) auxiliaries: () a. A: Ik denk dat Charlotte haar kamer nog altijd niet opgeruimd heeft. I think that Charlotte her room still always not cleaned has ‘I think Charlotte still hasn’t cleaned her room.’ B: Goh, tegen vanavond  ze wel [ haar kamer opgeruimd hebben]. well by tonight must she  her room cleaned have ‘Well, by tonight she’ll have to have cleaned it.’ B’: *Goh, tegen vanavond  ze wel [ haar kamer opgeruimd] hebben. well by tonight must she  her room cleaned have ‘Well, by tonight she must have.’ b. Die broek  nog niet gewassen worden, maar hij  wel those pants must still not washed become but he is.allowed  al [ gewassen worden]. already washed become ‘Those pants don’t have to be washed yet, but they can be.’ b’. *Die broek  nog niet gewassen worden, maar hij  wel those pants must still not washed become but he is.allowed  al [ gewassen] worden. already washed become Curiously, MCE seems to allow for voice mismatches, even though it elides the Voice head (as well as the aspectuals dominating VoiceP; see Aelbrecht ): () a. A: Het cadeautje kan afgegeven worden door Yves. the present can off.given become by Yves B: Ralf mocht vandaag toch [ het cadeautje afgeven]? Ralf may today  the present off.give ‘The present can be handed over by Yves.’ – ‘I thought Ralf was allowed to do that?’ b. Dit programma kan gebruikt worden door iedereen die wil [ dit this programme can used become by everyone who wants this programma gebruiken]. programme use ‘This programme can be used by everyone who wants.’



  

This presents a further problem for Merchant’s (a, d) analysis of voice mismatches in VPE and pseudogapping (see section ..), since it cannot be claimed that voice mismatches are permitted only if the voice head is situated external to the ellipsis site. What is remarkable about MCE is that extraction out of the ellipsis site is allowed, but restricted in a way that extraction out of VPE is not (see Aelbrecht ; Dagnac ). It turns out that only subjects—including derived subjects originating in object position—are able to move out of the to-be-elided constituent, not objects. This is illustrated in () for object extraction, with (b) indicating that the ungrammaticality is due to ellipsis: the non-elliptical counterpart of this example is fine.21 () a. ?*Ik weet niet wie Kaat  uitnodigen, maar I know not who Kaat wanted invite but weet wel wie ze  [twie uitnodigen]. know  who she must. invite b.

ik I

Ik weet niet wie Kaat  uitnodigen, maar ik I know not who Kaat wanted invite but I weet wel wie ze  uitnodigen. know  who she must. invite ‘I don’t know who Kaat  to invite, but I do know who she  to.’

The example in () displays subject extraction, here with the derived subject of an unaccusative verb (but see also (b) for a derived subject of a passive).22 () Erik is al langsgekomen, maar Joris moet nog [tJoris langskomen]. Erik is already by.passed but Joris must still by.pass ‘Erik has already passed by, but Joris still has to.’ These data are relevant in light of the debate around the structure of the ellipsis site. The ban on object extraction at first hints at a proform analysis, without a full syntactic structure, but that would leave the subject extraction facts unexplained. Aelbrecht () develops a theory of ellipsis licensing which captures this contrast by proposing that the timing of the ellipsis is crucial: only elements that have moved out of the ellipsis site before the licensing head is merged can escape ellipsis (see Aelbrecht  for the details). As it turns out, MCE is not the only type of predicate ellipsis to display this property of limited extraction. British English do does as well, as the next section shows.

21 Note that Modal Ellipsis in Romance languages differs in several respects from Dutch (Germanic) MCE. According to Busquets and Denis (), Dagnac (), and Authier (), extraction is much less limited in Romance ME than in Germanic MCE (more than just subjects can be extracted), and voice mismatches are illicit in French (and not fully investigated in other Romance languages). This potentially has consequences for the size of the elided constituent in M(C)E and the licensing of the ellipsis operation. Whether MCE and ME constitute the same type of ellipsis remains an open question. 22 See Aelbrecht () for detailed discussion as well as evidence that deontic modals behave like raising verbs in Dutch and, consequently, that the subject in these cases is extracted from the ellipsis site.

 



.. British English do British English do (henceforth BE do) is a phenomenon reminiscent of English VPE, and it is discussed by Chalcraft (), Haddican (), Aelbrecht (), Thoms (b), and Baltin (). In the examples in (), the verb phrase run the race is elided and there is a finite auxiliary present. Hence, in the VPE sentence in (b) there is no need to insert a dummy do. In British English on the other hand, an additional non-finite do can appear, as in (a). ()

a. Luis will run the race and Nana will do [run the race], too. b. Luis will run the race and Nana will [run the race], too.

This do can only occur when the verb phrase is elided: ()

*Luis will run the race and Nana will do run the race, too.

Parallel to what we have seen for MCE, BE do yields mixed results when the extraction test for internal structure of the ellipsis site is applied to it (see Aelbrecht , Baltin ): object extraction is banned (unlike in the VPE counterparts, as in ()). Moreover, pseudogapping is disallowed with BE do (see (a)). Subjects, on the other hand, can be extracted from the ellipsis site, as in (). ()

a. *Although I don’t know who Thomas will visit, I do know who Aga will do [visit twho]. b. Although I don’t know who Thomas will visit, I do know who Aga will [visit twho].

()

a. *Although she won’t eat pizza, she will do pasta [eat tpasta]. b. Although she won’t eat pizza, she will pasta [eat tpasta].

()

a. ?Kay might seem to enjoy that, and James might do [seem to tJames enjoy that], too. b. ?The river will freeze solid, and the lake will do [freeze tthe lake solid], too.

Aelbrecht () applies the analysis proposed for MCE to BE do to capture this contrast. Hence, these recently emerged types of predicate ellipsis shed some new light on issues that have been at the centre of the VPE literature for years, such as the question of internal structure and the value of the extraction test.

. C 

.................................................................................................................................. Predicate ellipsis is a cover term for different types of ellipsis targeting the predicate part of a clause. The types presented here are VP-ellipsis (the most widely discussed elliptical phenomenon), pseudogapping, verb-stranding VP-ellipsis, modal complement ellipsis, and British English do. We have seen that these types display differences in behaviour and distribution, and it has become apparent that the question of what exactly is elided is not always easy to answer.

  ......................................................................................................................

  ......................................................................................................................

´  

. I

.................................................................................................................................. N  is the somewhat vague label applied to different types of anaphoric phenomena involving a gap within the internal structure of the nominal phrase. It is used in the literature for a set of arguably disparate phenomena within and across languages. Just to give an idea of the many constructions to which the term is (correctly or incorrectly) applied, let us consider Japanese. In this language, the term nominal ellipsis (and relatives) is used to refer to: (i) nominal gaps with at least one genitive remnant (see ()), (ii) nominal gaps filled with the pronominal -no ‘one’ (see ()), (iii) nominal gaps of place and time nouns modified by ku-inflected adjectives (see ()), and (iv) radical gaps in argument position (see ()):1,2 N’-ellipsis () a. [Taroo no taido]-wa yoi ga, [Hanako Taroo no attitude- good though Hanako ‘Though Taroo’s attitude is good, Hanako’s isn’t.’

no taido]-wa no attitude-

yokunai good.not

b. [Rooma no hakai]-wa [Kyooto no hakai]-yorimo hisan data Rome no destruction- Kyoto no destruction-than miserable was ‘Rome’s destruction was more miserable than Kyoto’s.’ (Saito et al. : )

1 I use the names that each of these constructions receives in the literature, without any theoretical commitment. 2 Through this chapter, I follow the Leipzig Glossing Rules (https://www.eva.mpg.de/lingua/ resources/glossing-rules.php). For the sake of simplicity, in most cases I use Rule  for expressing one-to-many correspondences and optional Rule C, which avoids unnecessary segmentation in the original examples. I also use Rule  for non-overt elements, especially in the Spanish examples. In the general case, I avoid modification of glosses from other sources. Abbreviations:  = accusative;  = adjective;  = attributive marker;  = classifier;  = clitic;  = complementizer;  = dative;  = determiner;  = diminutive;  = feminine;  = future;  = genitive;  = imperative;  = infinitive;  = locative;  = masculine;  = neuter,  = nominative;  = passive;  = particle;  = plural;  = present;  = past;  = subjunctive;  = singular;  = topic.

  Anaphoric-no () Akai no–o mittu kudasai red one- three give.me ‘Please give me three red ones.’ Ku-ellipsis () a. Taroo-ga too-i basyo-e Taroo- far- place-to ‘Taroo went to a distant place.’



(Takita : )

itta. went

b. Taroo-ga too-ku-e itta. Taroo- far--to went ‘Taroo went to a distant place.’ c. Kono densetu-ga huru-i zidai-kara this legend- old- time-from ‘This legend is from old times.’ d. Kono densetu-ga huru-ku-kara this legend- old--from ‘This legend is from old times.’

aru. be

aru. be (Larson and Yamakido )

Radical ellipsis () Taroo-ga Hanako-ni [e e kekkonsuru Taroo- Hanako- he her marry ‘Taroo promised Hanako that he would marry her.’

to ] that

yakusokusita. promised (e = empty) (Takahashi a: )

As the English glosses show, some of these constructions can be translated by a nominal gap or a similar nominal anaphor as in () and (), respectively, but in other cases, the gap must be overtly expressed with the corresponding noun () or pronoun (). In other languages, like Spanish and Romance in general, both () and (), for instance, have a silent counterpart: () a. La destrucción de Roma fue más :[] destruction of Rome be.[] more la de Kyoto. :[] of Kyoto ‘Rome’s destruction was more miserable than Kyoto’s.’ b. Por favor, deme please give.:.. ‘Please, give me three red ones.’

tres three

miserable miserable

que than

rojos. red::

Like English, Spanish lacks temporal/spatial elliptical nouns, but like many languages (including English and Japanese) it makes productive use of human ones (the so-called





people-deletion or human null construction: Pullum , Kester a, b, Giannakidou and Stavrou , Panagiotidis , Kornfeld and Saab , among many others). () a. Que la crisis la paguen los ricos! that :[] crisis ..[] pay:[.]: :: rich:: ‘May the rich pay for the crisis!’ b. Los/Las de al lado llamaron :: /:: of to:[.] side call:: ‘The guys/girls living next door called three times.’

tres three

veces. time:

Finally, Spanish licenses radical gaps in a productive way for subject arguments, but regarding objects, the phenomenon is only attested with indefinite objects (with the exception at least of Andean dialects; see Suñer and Yépez ): () a. e Trabajan. e work:[]: ‘They work.’ b. Juan compró manzanas y Pedro también Juan buy:[.] apple: and Pedro also ‘Juan bought apples and Pedro also bought apples.’

compró buy:[.]

e. e

Thus, in language after language we observe the ubiquity of nominal ellipsis phenomena. In this chapter, I will adopt the hypothesis—very widespread in the generative tradition—that the term ellipsis refers to a syntactic mechanism that generates some gaps on the basis of salient linguistic information. Put differently, the term ellipsis only applies to what Hankamer and Sag () call surface anaphora. In contradistinction, Hankamer and Sag argued that other anaphora are simply (silent) base-generated proforms, namely, deep anaphora. The general conditions that account for the distribution of these two types of anaphor are discussed at length in Hankamer and Sag’s original paper and much subsequent work since then. This chapter is framed in this tradition. Thus, the set of facts to be discussed in what follows raises at least two basic questions: (Q) To what extent does the term ellipsis adequately describe the entire set of phenomena illustrated in ()–()? (Q) Under what general conditions are the nominal gaps in ()–() allowed in natural languages? Given the deep/surface assumption, and the analytical tools it provides, only a subset of the examples discussed so far will be derived as instances of ellipsis. Other cases illustrated in ()–() will be conceived of as empty nouns or deep anaphora; i.e., base-generated nominals (or different projections of a nominal), whose meaning and syntactic distribution respond to different conditions from those attested for nominal ellipsis. In section ., I will provide several tests to distinguish both types of anaphora with the hope of making a descriptive contribution in an area that is still poorly understood. This will answer (Q) at least partially. In section ., I will (also partially) answer (Q) by addressing the problem

 



of the identity condition in nominal ellipses. The so-called licensing problem, also connected to (Q), will be explored in section ., where I will argue that there is no morphological licensing of any sort; the morphological reflexes attested in nominal ellipsis environments are epiphenomena arising from the interaction of various morphological, syntactic, and semantic factors. Before entering into the varieties of nominal ellipsis, let me clarify some aspects of the DP structure I will assume. Following the tradition initiated by Abney (), I adopt the minimal structure of DPs illustrated in (), in which D-features are encoded in an independent projection dominating the nominal root. Features pertaining to Number are universally encoded in an independent functional head Num above the nP (Ritter  and much subsequent work). As for the nP domain, I assume that it minimally consists of a lexical Root, √, and a category-defining head, n, and that both heads are combined via head movement in the syntax (see Embick and Marantz ).3 () [DP D [NumP [AP] Num [nP [AP][nP √ + n[gender] [√P t√ [AP/PP ]]]]]] I take adjectival modifiers to be phrasal adjuncts (or specifiers) that attach to the nP or above and AP/PP complements of the noun (e.g., la destrucción italiana vs la destrucción de Italia, ‘the Italian destruction’ vs ‘Italy’s destruction’) to be selected by the Root (= √). Gender features, when present in a language, are encoded inside the nP (Saab , , a).

. E    :  

.................................................................................................................................. Hankamer and Sag () distinguish two basic types of anaphora, namely, deep and surface anaphora. Briefly summarized, with the term deep anaphora Hankamer and Sag refer to a base-generated (c)overt proform, whose basic recovery conditions boil down to those of a (free) pronoun. Surface anaphora instead are elliptical structures derived by transformation (PF deletion in their terms) and, consequently, their recoverability conditions reduce to the theory of identity in ellipsis, whatever the right theory of identity is (syntactic or semantic, in broad terms). Since then, surface anaphora are conceived of as invisible/inaudible full-fledged structures, the result of the operation that we call ellipsis. In turn, deep anaphora are conceived of as lexical proforms made available by the Universal Feature Inventory of Universal Grammar and the combinatory system that produces syntactic objects; i.e., they are not derived by any particular transformation of the computational system. The particular abstract form of these two types of entities is sometimes obscured by their surface form. In the nominal domain, this is especially clear in languages like Spanish which make productive use of both empty noun constructions (ENs) and NP-ellipsis (NPE).

3

When irrelevant, I will just refer to the complex formed by n and RootP as NP.





() los de al lado / los tontos / los que :: of to:[.] side :: fool:: :: that cantan sing[]: ‘the ones living next door / the foolish / the ones who sing’ The three expressions in () might be ambiguous in the right contexts. Consider, for instance, the following sentences: () a. Los perros inteligentes y los tontos :: dog:: smart: and :: fool:: son indistinguibles. be[]: indistinguishable: NPE reading: ‘Smart dogs and foolish dogs are indistinguishable.’ EN reading: ‘Smart dogs and foolish people are indistinguishable.’ b. Los perros de enfrente y los de al :: dog:: of in.front and :: of to.[.] lado son ruidosos. side be[]: noisy:: NPE reading: ‘The dogs living in front and the dogs living next door are noisy.’ EN reading: ‘The dogs living in front and the people living next door are noisy.’ The sentence in (a) can mean either that smart dogs and silly dogs cannot be distinguished or that smart dogs and silly people cannot. Likewise, the sentence in (b) can be true either in a scenario in which the dogs that live in front and the dogs living next door are noisy, or in a situation in which the dogs that live in front and the people living next door are. This ambiguity is straightforwardly derived under the hypothesis that we are dealing with different types of nominal gaps in each of the DPs in the second conjunct of the sentences in (). Thus, the human reading arises because of the presence of an empty noun in the second conjunct of both sentences in () (see Panagiotidis , a,b, and Schütze b and Corver and van Koppen  for related analyses). I will assume that an empty noun is, strictly speaking, a functional nominal category which might encode some syntactic–semantic features such as [+/ human], [+/ female], [+/ count], and so on. It is the same category that provides nominal status to a given bare Root (see Embick and Marantz  and ()). In turn, the readings under which we are always talking about dogs are derived by NPE. As already mentioned, NPE entails deletion/non-pronunciation of a full-fledged nP including the n itself and minimally the √P. The two configurations are illustrated in (), where I cross out the elliptical constituents following standard conventions:

()

Empty Noun a. [DP D [NumP Num [nP n[human]]]]

NP-ellipsis b. [DP D [NumP Num [nP n [√P √] ]]]

Applied to the ambiguity of the sentences in (), we get the following underlying structures (omitting some important details):

  ()



a. Los perros inteligentes y [los :: dog:: intelligent: and :: tontos] son indistinguibles. fool:: be[]: indistinguishable:

perros / n[human] dog: : / n[human]

b. Los perros de enfrente y [los :: dog:: of in.front and :: de al lado] son ruidosos. of to:[.] side be[]: noisy::

perros / n[human] dog:: / n[human]

The two (rough) representations in () adequately describe the ambiguity in (). The next question is what other types of predictions arise by virtue of the two configurations just provided. An immediate set of predictions can be grouped under what I will call the uniformity assumption, which can be traced back to Ross’s (b) seminal work on sluicing (see Saab , b, and references therein): Uniformity Assumption: () Ceteris paribus, the syntactic dependencies in contexts of ellipsis between the remnant constituent and the ellipsis site (case, agreement, dislocation, thematic assignment, etc.) are obtained in the same way as in non-elliptical sentences. This assumption allows us to distinguish between ENs and NPE in at least three related domains, namely, (i) thematic assignment (...), (ii) extraction (...), and (iii) case and other matching effects (...). Another set of diagnostics emerges precisely from the particular conditions contemplated in the ceteris paribus proviso. Some of these conditions are quite general, like those that regulate the distribution of antecedents and remnants, while others are language-particular and have to be determined in a case-by-case fashion. In section .., some of these diagnostics are addressed with special reference to the human EN construction and NPE in Spanish.

.. Uniformity tests ... Thematic assignment A crucial difference between the two configurations in () is that ENs are not theta-role assigners. Elided NPs, in turn, conserve the thematic properties of their non-elliptical counterparts. This is a general property that distinguishes pronouns or deep anaphora from ellipsis phenomena. In the case at hand, it is predicted that the internal theta-roles of the Root (or lower N, depending on different assumptions on the DP geometry) cannot be assigned in EN configurations. As shown by Panagiotidis (, a,b), if well-known instances of one-anaphora in English were analyzed as the surface realization of an EN configuration, then the old observation by Lakoff () on the asymmetry between nominal adjuncts and complements in the licensing of one would be immediately derived, given that ENs lack theta-grids. In effect, whereas nominal adjuncts can modify one, internal complements cannot:

 ()

 a. I bought the car from England and Sam bought the one from Spain. b. *I met the king of England and Sam met the one of Spain. (Lakoff : , ex. )

In ()–(), there are some additional examples taken from Corver and van Koppen (): ()

a. *Jack met the king of England, and I met the one of France. b. Jack met the king from England, and I met the one from France.

() a. *the treatment of Bill and the one of Sue b. the treatment by the psychologist and the one by the psychiatrist ()

a. *the rumor that Bill would be fired and the one that John would keep his job b. the rumor that John heard yesterday and the one that Mary had heard the day before (Corver and van Koppen : , n. )

By virtue of these facts, Panagiotidis (, a,b) proposes the following EN analysis, according to which one realizes an empty noun incapable of assigning theta-roles:4 ()

DP D

NumP

the AP

Num'

new Num [pl]

NP N ones

4

Additional evidence for the nominal status of one is that: (i) it can be the complement of a determiner, (ii) it can receive adjectival modification, and (iii) it can occur in the plural form. (i) (ii) (iii)

a. b. a. b. a. b.

[This one] is from New Jersey. [The one I saw] is from New Jersey. [A new one] is sometimes a challenge. I find it annoying she lost [the new one]. You should carefully file [the new ones]! [New ones] are usually laser-printed.

(Panagiotidis b: , examples ()–())

Yet Kayne () has criticized the nominal status of one on empirical grounds. According to Kayne, in its anaphoric use, one is not a noun, but a complex determiner. The link with its antecedent is, nevertheless, mediated by a silent noun, so anaphoric one is still a deep anaphor on Kayne’s account.

 



It is worth mentioning that neither this analysis nor the evidence that seems to support it is uncontroversial. Empirically, examples in which one replaces deverbal nouns (a), picturelike nouns (b), and human relational nouns like queen, supporter, and student (c–e) are attested in corpora (see Payne et al.  for the source of each example): () a. This interpretation is contrary to an accepted [one of wrestling] as a sport. b. How the printers had got hold of her photograph she did not know, but they had, and now it was being sold all over along with [ones of Lillie Langtry and other noted belles]. c. Dudley himself was no more eager for the match. Yes, he wanted to marry with a queen, but not [the one of Scotland]. d. Despite the rivalry between the two sides, supporters, specially [the ones of Real Madrid] are known to show respect to the individual talents in the opposition team. e. In the case of medicine, I think there’s no other alternative than the Universidad de la República. I would think their classes are equally crowded, but haven’t ever heard any of the medicine students complain as much as the [ones of computer science]. The fact that some replacements are more frequent than others is attributed by Payne et al. to the existence of more successful competitors (NumP-ellipsis, for instance, like Rome’s destruction and Carthage’s; see section ...). The conclusion for the distinction made in this chapter would be that theta-role assignment would not be a reliable test, against what is generally assumed. In effect, if there is no argument structure projected in nominals and complements are indeed modifiers (see also Llombart-Huesca ), then there would be no basis for the distinction between empty nouns and ellipsis. I think, however, that such a conclusion would be misleading both empirically and conceptually. One way of handling the facts under the EN analysis is to reinterpret Payne et al.’s idea that frequency is accounted for in terms of blocking: more than to a frequency effect we can attribute the difference between NumP-ellipsis with genitive remnants and one-replacement precisely to the distinction between ellipsis and empty nouns. Thus, in my opinion, the sharp contrast between *the destruction of Rome and the one of Carthage and Rome’s destruction and Carthage’s is not because of some blocking effect, but because the second case is a true instance of ellipsis. However, I agree with Payne et al. that the difference between complements and adjuncts might dissolve in favor of the second when it comes to human relational nouns like student or supporters. On the empirical side, more research is needed to know whether other noun complements are really grammatical or not under onereplacement (e.g., ()). Alternatively, the paradigm in () could also be taken as evidence in favor of a surface anaphor analysis for one (see, for instance, Hankamer and Sag ). This idea is extensively discussed in Llombart-Huesca (), where one is taken as the surface reflex of a support strategy for the Number head in cases in which the NP is indeed deleted by ellipsis. According to Llombart-Huesca, who does not consider data like (), the basic pattern in ()–() shows that NP complements are part of the ellipsis site and, as a consequence, not





visible on the surface.5 On this account, the asymmetry between complements and adjuncts would follow from the size of the ellipsis site. At any rate, both the NPE and the EN analyses should account for variation in judgments with respect to the availability of one to take thematic complements (i.e., the paradigm in ()). The diagnostic we are discussing here raises no controversy when it comes to some nominal gaps in Spanish. As shown in Kornfeld and Saab () and Saab (), this diagnostic gives positive results when applied to the following Spanish examples: () a. los estudiantes de química y los :: student: of Chemistry and :: ‘the students of Chemistry and the students of Physics . . . ’

de of

física . . . Physics

b. la matanza de los leones y la de los :[] killing of :: lion: and :[] of :: tigres . . . tiger: ‘the killing of the lions and the killing of the tigers . . . ’ c. la destrucción del puente y la de la :[] destruction of:[.] bridge and :[] of :[] ciudad . . . city ‘the destruction of the bridge and the destruction of the city’ d. el rumor de que llegaste temprano y el [.] rumor of that arrive:. early and [.] de que nunca llegaste . . . of that never arrive:. ‘the rumor that you arrived early and the rumor that you never arrived . . . ’ The facts in () clearly favor an analysis in terms of NPE. If the nominal gaps in () were analyzed as containing an EN, we would expect these examples to be ungrammatical given that ENs are not theta-role assigners. In contradistinction, a nominal phrase that is elided retains its thematic dependencies after ellipsis. Thus, the (again rough) representation in (a), but not the one in (b), accounts for the occurrence of internal arguments of the Root: ()

a. [la destrucción de la ciudad] b. *[la eN de la ciudad]

Note, however, that the analysis in (a) is inconsistent with the assumptions made regarding the size of the ellipsis site and the geometry of DPs (cf. ()). Given that this type of NPE in Spanish is ellipsis of nP (see b), we predict that thematic PPs should be part of the ellipsis site, as argued by Llombart-Huesca () for English one (see above). 5

An alternative analysis for one-replacement is proposed in Harley (), where one is seen as the phonological realization of the n head specified as [+identity] and [+count]. Despite important differences with Llombart-Huesca’s analysis, one also represents a surface anaphor underlying structure under Harley’s account.

 



A possible way to overcome this problem is by assuming that the internal complement of destruction in the second conjunct has to vacate the nP domain to some higher position (see Ticio , Saab , and also Eguren  for another approach): () [DP

la :[]

[XP

[PP

de of

la ciudad] :[] city

[nP

destrucción t]]] destruction

On this approach, this kind of Spanish NPE parallels the behavior of pseudogapping in English (i.e., VP-ellipsis + movement of an internal complement; see Chapter ). The same analysis has also been defended for English NumP-ellipsis by Yoshida et al. (b) (e.g., John’s books of poems and Peter’s of theater). In both languages, the evidence goes against any assimilation of these constructions to gapping.6 Space limitations prevent us from entering into the nature of such evidence, although I will present an argument in favor of the analysis in () for some putative instances of NPE in English in section ....

... Extraction tests A well-known test for surface anaphora is extraction from elliptical sites (see Depiante  and this volume, for extensive discussion and references). In short, if a constituent can establish a chain dependency with a position within a putative nominal gap, then we can safely conclude that such a gap has internal structure; i.e., the gap is indeed derived by ellipsis. In the nominal domain, however, there is no easy way to construct this type of example because extraction out of DPs requires manipulating too many variables (e.g., the relative position of the DP, its definiteness/specificity, among other poorly understood variables). However, Lipták and Saab () provide the following example from Hungarian, a language that makes productive use of NPE: () Hallottam riportot több miniszterrel. Nem emlékszem, heard interview: many minister: not remember: melyik miniszterreli hallottam [hosszút [NP riportot ti]] which minister: heard long: interview: ‘I heard an interview with many ministers. I don’t remember which minister I heard a long one with.’ Merchant (a) also gives the following example from Greek, another NPE language: () Tis istoriasi dha ton palio [proedhro__], kai . . . the history: .saw : old: chair: and ‘I saw the former chair person(masc) of the history department, and . . . ’ . . . tis glossologias tha dho  linguistics: .:see : ‘of linguistics, I’ll see the new(masc) (one).’

6

tonkenurio. new: (Merchant a: , ex. )

The reader is referred to the aforementioned works for extensive discussion.





The same result is obtained in Spanish NPE: () a. Yo sé de quién compraste dos fotos, I know[.] of who buy:[.] two picture: pero no sé de quién compraste tres fotos. but not know[.] of who buy:[.] three picture: b. Yo sé de quién no compraste ninguna I know[.] of who not buy:[.] any foto, pero sé de quién compraste algunas fotos. picture but know[.] of who buy:[.] some picture: c. Yo sé de quién no compraste ninguna foto, I know[.] of who not buy:[.] any picture pero sé de quién compraste varias fotos. but know[.] of who buy:[.] many picture: Yet, English presents a more interesting case related to subextraction facts. In principle, extraction is ungrammatical with numerals or quantifiers as remnants (sometimes taken as remnants of true NPEs). Consider the English counterparts to () (all from Gary Thoms, p.c.): () a. *I know who you bought two pictures of, but I don’t know who you bought three. b. I know who you bought two pictures of, but I don’t know who you bought three pictures of. c. *I know who you didn’t buy any pictures of, but I know who you did buy some. d. I know who you didn’t buy any pictures of, but I know who you did buy some pictures of. e. *I know who you bought no pictures of, but I don’t know who you bought many. f. I know who you bought no pictures of, but I don’t know who you bought many pictures of. On the basis of this paradigm, one would be tempted to conclude that English lacks NPE in these environments. However, as observed by Jason Merchant (p.c.), the examples become grammatical whenever the preposition is left stranded in the DP domain. As an illustration, consider the following counterpart to (a): () I know who you bought two pictures of, but I don’t know who you bought three pictures of. On an NPE account, these facts are accounted for along the lines proposed for Spanish in the previous section (see ()); i.e., the PP complement vacates the ellipsis site and then the wh-constituent is subextracted from the PP: () I know who you bought two pictures of, but I don’t know whoi you bought [DP three [PP of ti]j [NP pictures [PP t]j]].

 



Thus, the contrast observed in (a) and () would constitute evidence in favor of a pseudogapping approach, at least in this case. However, the interaction between NPE and extraction is an area almost unexplored and whatever conclusion one draws from such facts should be taken as provisory at this point.7

... Matching effects As illustrated in (), human EN constructions in languages like Spanish can contain, among other modifiers, de-PPs or adjunct CPs as modifiers. NPE gaps, however, subcategorize for the same categories as their non-elliptical counterparts. Take for instance psych nouns like amor ‘love’ which subcategorize for PPs headed by the preposition por ‘by’, or deverbal nouns like insistencia ‘insistence’ which inherits PP complements headed by en ‘in’ from the Root: () El amor de Juan por los autos es [.] love of Juan for :: car: be[.] grande que el mío por los libros. big than [.] mine for :: books ‘Juan’s love for cars is bigger than mine for books.’

más more

() la insistencia de Juan en sus problemas y :[] insistence: of Juan in his problems and la de Pedro en los suyos . . . :[] of Pedro in :: his ‘Juan’s insistence on his problems and Peter’s insistence on his . . . ’ Case markers are also inherited in nominal gaps. Thus, the deverbal noun entrega ‘delivery’ takes a dative complement headed by the dative marker a ‘to’. ()

La entrega de los libros de Cortázar a María :[] delivery of :: books of Cortázar to María y la de los libros de Borges a Juan tienen and :[] of :: books of Borges to Juan have que hacerse el mismo día. that do:: [.] same day ‘The delivery of Cortázar’s books to María and the delivery of Borges’ books to Juan have to take place the same day.’

As with the other diagnostics connected to the Uniformity Assumption, the test has to be checked language by language, taking into account the particular lexical and non-lexical properties of the languages under consideration. In principle, and as expected, the test gives positive results in clear cases of NumP-ellipsis in English. 7 NumP-ellipsis in English genitive constructions seems to be less uncontroversial, since in these cases extraction of internal arguments gives grammatical results (e.g., Rome’s destruction and [DP Carthage’s [NumP Num [NP destruction t ]] ]).

 ()

 a. John’s dependency on drugs and Peter’s dependency on his mom . . . b. John’s dependency on drugs and mine dependency on my mom . . .

.. More diagnostics ... Antecedents Recall that a sentence like (a), repeated in (), is ambiguous between a human EN construction and an NPE one. ()

Los perros inteligentes y los tontos :: dog:: smart: and :: fool:: son indistinguibles. be:[]: indistinguishable: NPE reading: ‘Smart dogs and foolish dogs are indistinguishable.’ EN reading: ‘Smart dogs and foolish people are indistinguishable.’

Interestingly, reversing the order of the conjuncts eliminates the ambiguity: () Los tontos y los perros inteligentes :: fool:: and :: dog:: smart: son indistinguibles. be:[]: indistinguishable: ‘Foolish people and smart dogs are indistinguishable.’ This new sentence can only mean that foolish people and smart dogs cannot be distinguished. This follows from the basic division made between ENs and NPE. As observed by Hankamer and Sag (), one of the basic properties distinguishing deep and surface anaphora is the mandatory requirement of a linguistic antecedent for surface anaphora. Given that backward anaphora is not allowed in coordinated structures (at least in NPE), the contrast between () and () is derived as a matter of linguistic antecedence. Put differently, the NPE reading in () is blocked because there is no linguistic antecedent in the required configuration.

... Identity effects Given that NPE requires a linguistic antecedent, ellipsis sites are formally linked to them. Such a formal dependency has particular manifestations across languages. In the case of NPE in Spanish and other languages, the gender specification must be identical between the antecedent and the nominal gap (for Spanish see Leonetti , Depiante and Masullo , Ticio , Kornfeld and Saab , Saab , , a, and Eguren ; for Brazilian Portuguese see Zocca , Nunes and Zocca , and Bobaljik and Zocca ; for Greek see Giannakidou and Stavrou  and Merchant a). Although nouns differ as to how natural gender is morphologically represented, the ban on nominal ellipsis under gender mismatches remains constant with some subtle differences among speakers:

  ()

a. *el tío de María [.] uncle of María b. el tío de María [.] uncle of María c. *la tía de María :[] aunt of María d. la tía de María ::[] aunt of María ‘Mary’s uncle/aunt and Pedro’s.’

y and y and y and y and

la :[] el [.] el [.] la :[]

de of de of de of de of



Pedro Pedro Pedro Pedro Pedro Pedro Pedro Pedro

Number does allow for mismatches under ellipsis: () a. el perro de [.] dog:[] of ‘Juan’s dog and Pedro’s dogs’ b. los perros de Juan :: dog:: of Juan ‘Juan’s dogs and Pedro’s dog’

Juan Juan

y and

los ::

de of

Pedro Pedro

y el de Pedro and [.] of Pedro

Gender and number asymmetries have been accounted for in at least two ways in the literature. On one approach, the difference boils down to the lexical vs syntactic nature of gender and number, respectively. The general idea is that gender is a lexical property of nouns and, as such, it has to respect lexical identity under ellipsis. This suggestion has been made by Giannakidou and Stavrou (), Depiante and Masullo (), and Kornfeld and Saab (). This type of analysis has been criticized for a number of reasons by Saab (, , a), who proposes that NPE in Spanish and other languages is ellipsis of the nP, to the exclusion of other extended projections of the nominal domain such as NumP and DP (see also Ticio  and, more recently, Merchant a). Given that gender is specified on the nP level and number on the head of NumP (see ()), the asymmetry follows from the licensing condition on ellipsis. See the following representation, where the [E] feature on Num is just a convention to indicate the head licensing the elliptical phrase (see Merchant  and section ..): ()

DP D

NumP Num # [E]

→ NPE domain nP

n √P [gender]

As mentioned, this analysis captures the gender–number asymmetry on the basis of the ellipsis site size. Yet, it still remains to be seen whether identity in nominal ellipsis must be captured in semantic or lexico-syntactic terms, but this will be the topic of section ... For our purposes here, what matters is that identity effects of this type are a property of NPE, but not of ENs, which, as already noted, are subject to the general recoverability conditions on pronouns. When used as a free expression, the natural gender specification





on ENs would be consistent only if the female or male presupposition encoded in the empty noun meets its conditions. Therefore, whenever you say las tontas de al lado ‘the fools (female people) living next door’ in an out-of-the-blue context, the expression would be felicitous only under the condition that the people living next door are indeed female. Note that this difference between NPE and human EN predicts that the ambiguity observed in () would vanish whenever gender features between antecedent and elliptical gap differ. This is borne out: () las perras inteligentes y los the:: dog:: smart: and the:: ‘the smart female dogs and foolish people . . . ’

tontos . . . fool::

Here, the second conjunct can only have a human EN reading; the identity constraint on gender discussed above blocks an NPE reading.

... Lexical restrictions and productivity As noticed by Giannakidou and Stavrou () and others, ENs, but not NPE, are lexically restricted.8 On the one hand, they express some meanings but not others. As we have already seen, ENs in Spanish and other languages are commonly used to express a human entity. The types of meanings that ENs might express are, however, a lexical matter. Thus, there is no other obvious reason, beyond lexical specification, why Japanese has empty nouns of time and space (see ()), but Spanish does not. The meanings that empty nouns might encode correspond to general concepts such as HUMAN, SPACE, TIME, FACT, MANNER, etc. On the other hand, ENs are especially productive with certain types of modifiers but not others. Thus, as discussed in detail by Panagiotidis (), poor and rich are especially good EN modifiers in out-of-the-blue cases like the poor and the rich, under generic readings. Such effects could lead us to conclude that these are not manifestations of a particular empty noun syntax, but just nominal uses of adjectives. Given that there are indeed certain lexical and superficial similarities between these two phenomena, it is important to know whether there are arguments for keeping them apart. As shown by Panagiotidis (: ), adjectives like poor maintain their adjectival properties in EN environments; thus, they license degree modification but not plural morphology (e.g., the very poor vs *the poors). This follows if in the relevant cases poor is just an adjective modifying an empty noun and not a nominalized adjective: () [DP the [NumP [AdjP poor] [NP Ø] ] ]

8

Giannakidou and Stavrou do not propose an empty noun analysis. Instead, they refer to a substantivization process to account for some of the constructions that I analyze as involving an underlying empty noun. Other constructions explored by Giannakidou and Stavrou also do not seem to instantiate EN constructions. Concretely, what they call the abstract construction of Greek, which is translated by the so-called neuter article lo ‘it’ in Spanish (e.g., lo desconocido, ‘the: unknown’), does not have the properties of EN constructions (see Kornfeld and Saab ). Space limitations prevent us from providing an overview of the properties of this construction.

 



There are other morphological particularities that are relevant when it comes to making this distinction between ENs and nominalizations clearer. The diagnostics, once again, depend on language-internal properties. The use of diminutive forms in Dutch, for instance, is a good diagnostic to distinguish pure EN constructions from nominalizations, because adjectives in this language do not tolerate diminutive suffixes. () a. een blinde a blind ‘a blind person’ b. *een a

blindetje blind:

(Kester a: , ex. ())

Moreover, Dutch human ENs present an irregular form of the plural; instead of using the [‑s] ending that characterizes most nouns with final schwa, they show the plural form [-en]. ()

a. de blinden  blind: ‘the blind’ b. *de blindes  blind: ‘the blind’

(Kester a: , ex. ())

Compare with true nominalizations which can bear diminutive morphemes and show the regular plural ending: () a. het centraletje  central: ‘the small power station’ b. de centrales  central: ‘the power stations’

(Kester a: , ex. ())

However, it is not always easy to keep empty noun constructions and nominalizations apart. In Spanish, for instance, there are good reasons to think that adjectives like rich, poor, and others can participate in both types of underlying configurations. Thus, a nominal phrase such as los ricos (lit. ‘the rich’) can be ambiguous between an empty noun structure, where the adjective modifies the EN (a), or a nominalization structure, where ricos occupies the head noun position (b): () a. [DP los [NumP [AdjP ricos] [NP Ø] ] ] b. [DP los [NumP [NP ricos] ] ]

Empty noun Nominalization

As shown in (), the second conjunct is ambiguous between a human EN construction according to which we are talking about people in general and an ellipsis reading according





to which we are talking about rich people. This second reading follows if ricos is analyzed as a noun like in (b) and, as such, can be a legitimate antecedent for another occurrence of the same noun in the elliptical gap. Of course, an EN reading is also available in this particular context: () los ricos que ahorran toda la vida y :: rich:: that save:[] all ::[] life and los [NP ricos] / [n[human]] que no pagan impuestos . . . :. rich: that not pay:[]: taxes ‘the rich that save money their entire life and the rich/ones who do not pay taxes . . . ’ We can also disambiguate in the opposite direction. As the following examples from Kester (a) show, degree adjectival modification disambiguates in favor of an EN analysis: () [Los extremadamente ricos [NP Ø] ] no :: extremely rich:: not viven en este barrio. live:[:] in this neighborhood ‘The extremely rich do not live in this neighborhood.’ (adapted from Kester a: , ex. ()) As expected, NPE cannot apply in this case, because rich, when acting as an adjective, is not a legitimate nominal antecedent. Consider (): () Los muy ricos de este país y los del :: very rich:: of this country and :: of:[.] país vecino . . . country neighbor ‘The very rich of this country and the people from the neighboring country . . . ’ Here, the reading according to which people from the neighboring country are very rich is just not available. Dropping the intensifier muy ‘very’ makes the elliptical reading fully available again, showing that Spanish distinguishes adjectives that modify ENs (e.g., los muy ricos, lit. ‘the very rich’) from deadjectival nouns. Only the latter can be the target of NPE. In summary, besides superficial similarities, modified empty nouns and nominalizations are different phenomena. Once these possible confusion factors are set aside, we can safely conclude that ENs, as a phenomenon distinct from nominalization, are subject to two basic lexical restrictions: (i) they encode general concepts (TIME, MANNER, HUMAN, etc.), and (ii) they can be subject to particular morphological processes (irregular plurals, for instance). None of these properties applies to NPE, a systematic and productive process of deletion/non-pronunciation quite unrestricted semantically.

 



.. Summary I have provided several tests to distinguish NPE from ENs. As shown in section .., some diagnostics follow from uniformity considerations (theta-role assignment, extraction, and matching effects) and, as shown in section .., others follow from the general conditions that license either NPE or ENs (the need for an antecedent, identity effects, productivity, and so on). Unfortunately, there is some degree of sloppiness in the literature when it comes to making this basic division. Even though I am aware of the controversial status of some of the diagnostics discussed in this section, I still think that the distinction between NPE and ENs is robust and should be taken seriously when discussing these phenomena within and across languages. In the next section, I discuss the nature of the recoverability conditions that regulate NPE and ENs.

. R    

.................................................................................................................................. In this section, I briefly discuss the problem of recoverability for empty anaphora in general. Building on the basic distinction made in the previous section, I will focus on the identity condition for NPE (..) first, and then on some particular cases of pragmatic ENs in Spanish (..).

.. Identity in NPE As is well known, the proper nature of the identity condition on surface anaphora is a matter of controversy. Broadly speaking, the debate centers on whether identity should be formulated in purely semantic terms (Merchant ), in purely lexico-syntactic terms (Chomsky  and much subsequent work in transformational grammar), or in mixed ones (Chung ,  for a recent approach). Curiously, NPE has not been the focus of such a debate, even though it constitutes an ideal scenario to evaluate competing theories on identity in ellipsis. An important exception is Giannakidou and Stavrou () who propose the following condition: Recoverability of the Descriptive Content in Nominal Subdeletion: () An elided nominal subconstituent α must recover its descriptive content by an antecedent γ previously asserted in the discourse. (Giannakidou and Stavrou : ) Importantly, for Giannakidou and Stavrou recoverability is semantic and not lexicosyntactic. This is so, they argue, because case and number mismatches, as in Spanish, are attested in Greek NPE:9 9

Yet this argument does not hold if number (and also case) mismatches are the result of the fact that ellipsis only deletes the NP layer, to the exclusion of NumP and K(ase)P, as discussed in section ....





() Htes irthe enas filos mu na me dhi ki yesterday came: a friend: mine  me see: and ego meta episkefitka alus dio [filus]. I then visited: other two: friends: ‘Yesterday, a friend came to see me and I afterwards visited two more [friends].’ (Giannakidou and Stavrou :  ex. ()) According to (), the descriptive content must be given and asserted in the previous discourse and cannot be entailed; otherwise, the following example should be grammatical with the intended meaning that Andreas bought three books, given that buying three dictionaries entails buying three books: () *O Andreas agorase dio :. Andreas:. bought: two:. lexika ke i Maria agorase dictionaries:. and :. Maria:. bought: tria [vivlia]. three:. books:. ‘*Andreas bought two dictionaries and Maria bought three.’ (intended meaning: ‘three books’) (Giannakidou and Stavrou :  ex. ()) One may wonder in what sense we can say that an antecedent A and an ellipsis site E are semantically identical when it comes to evaluating NPs for ellipsis. Assuming that in the general case NPs denote properties, we can understand the condition in () as making reference to equivalence of properties. In this respect, like other semantic approaches to ellipsis (e.g., Merchant ), the theory predicts that lexical mismatches should be legitimate in NPE whenever the properties denoted by the antecedent and the elided phrase are identical. Empirical evidence demonstrates that this formulation of the identity condition is too weak and that some sort of lexical identity is needed. The main argument comes from synonymy relations among pairs of words. Consider, for instance, the synonymous nouns doctor ‘doctor’ and médico ‘medic’ in Spanish as they occur in the following DPs: () El médico/doctor de Juan se [.] medic/doctor of Juan  el médico/doctor de Pedro. [.] medic/doctor of Pedro ‘John’s doctor/medic met Pedro’s doctor/medic.’

reunió meet:.

con with

As it stands, () predicts that médico would be a legitimate antecedent for elision of doctor and vice versa, given that, for instance, John met Peter’s medic entails that John met Peter’s doctor and vice versa. The prediction cannot be (dis)confirmed in this particular case because both nouns are masculine. Yet, it is easy to find pairs of synonymous nouns that differ in gender specification. Take, for instance, the masculine noun casamiento ‘marriage’ and the feminine one boda ‘wedding’, which are synonymous in

 



Spanish under the ceremony and party meanings.10 Ellipsis under strict lexical identity is of course allowed: ()

a. el casamiento de [.] marriage:[] of de Pedro . . . of Pedro . . . ‘Juan’s marriage and Pedro’s . . . ’

Juan Juan

y and

el [.]

casamiento marriage:[]

b. la boda de Juan y la boda de Pedro . . . :[] wedding:[] of Juan and :[] wedding:[] of Pedro Intended: ‘Juan’s wedding and Pedro’s . . . ’ Lexical mismatches are however strongly ungrammatical in both directions: () a. *la boda de Juan y el :[] wedding:[] of Juan and [.] de Pedro . . . of Pedro Intended: ‘Juan’s wedding and Pedro’s marriage . . . ’

casamiento marriage:[]

b. *el casamiento de Juan y la boda de Pedro . . . [.] marriage:[] of Juan and :[] wedding:[] of Pedro Intended: ‘Juan’s marriage and Pedro’s wedding . . . ’ It is worth noticing that the ungrammaticality of () cannot be triggered only by gender. Unlike the mismatches discussed in (), gender specification in () is arbitrary (i.e., not related to natural gender or sex). Therefore, no semantic explanation for this particular gender mismatch seems to be available here. English NumP-ellipsis behaves similarly in this respect. As shown by Merchant (this volume), pairs of synonyms like nuptials/wedding differ arbitrarily in number specification: i.e., nuptials is a plurale tantum noun that triggers plural agreement with the verb in spite of being semantically singular. ()

Beth’s wedding was in Bond Chapel, and Rachel’s wedding was in Rockefeller Chapel.

() Beth’s nuptials were in Bond Chapel, and Rachel’s nuptials were in Rockefeller Chapel. Verbal agreement mismatches, however, are not allowed under NumP-ellipsis: ()

a. Beth’s wedding was in Bond Chapel, and Rachel’s was in Rockefeller Chapel. b. Beth’s nuptials were in Bond Chapel, and Rachel’s were in Rockefeller Chapel.

() a. *Beth’s wedding was in Bond Chapel, and Rachel’s were in Rockefeller Chapel. b. *Beth’s nuptials were in Bond Chapel, and Rachel’s was in Rockefeller Chapel. 10

Note that wedding and marriage are not synonymous in English.





The fact that an arbitrary property of nouns, such as being a plurale tantum, must be preserved under ellipsis gives support to the hypothesis that semantic identity alone is not enough to identify the ellipsis site. In summary, we are led to conclude that NPE requires some type of lexical/formal identity condition, more along the lines of the lexical–syntactic approach proposed by Merchant (d) (see also Saab , , and Chung ) than in the semantic theory defended by Merchant () (and much subsequent work) or in even more radical semantic/pragmatic views (Culicover and Jackendoff  and this volume, or Dalrymple et al. ). Notably, this type of evidence also goes against laxer syntactic approaches, like the recent proposal by Thoms (), according to which syntactic identity does not require making reference to the content encoded in the syntactic nodes but only to the syntactic distribution of variables.

.. Recoverability and ENs I have claimed that ENs are particular instances of deep anaphora. According to Hankamer and Sag (), recoverability conditions for deep anaphora are similar to those applying to pronouns and indexicals in general. In the nominal domain, standard cases of deep anaphora are indeed pronouns of different sorts like she, we, I, this, that, and so on. In fact, the semantic and syntactic literature on pronouns agrees in the postulation of an empty noun in the underlying structure of pronouns. Whether all pronouns contain an empty noun is something that we cannot resolve here (see Elbourne  and references therein). The intricacies of the semantics and syntax of pronouns in general are beyond the limits of the present chapter. Instead, I would like to make a brief remark on some putative instances of pragmatically controlled NPEs in Romance and beyond. Consider the following examples: () a. [pointing to a red skirt; pollera is feminine in Spanish] Dame la roja. give..:. :[] red:[] ‘Give me the red one.’ b. [pointing to a toy; juguete is masculine in Spanish] Yo quiero ese con rueditas.  want:[.] that:[.] with wheel:: ‘I want that one with small wheels.’ c. [pointing to a bike; bicicleta is feminine in Spanish] Cuando era chico, tenía una como esa. when was[.] boy have:[.] one:[] like that:[] ‘When I was a child, I had one like that.’ Some clarifications are in order. First, I have avoided using cases of human EN constructions, given that we have already provided several tests in favor of their non-elliptical nature in section .. Second, note that the remnants of each example are modifiers and not internal arguments of some putative elliptical noun. Third, this entire set of examples can be translated into English by one or by just a demonstrative.

 



The examples in () seem to pattern like putative cases of English VP-ellipsis controlled pragmatically (e.g., Shall we? as an invitation to dance; from Merchant a) and other surface anaphora (e.g., sluicing). In view of this, some researchers have proposed abandoning the requirement of a linguistic antecedent (see section ...) as a reliable test for distinguishing surface and deep anaphora (see Chao  and Lobeck , among many others) and claim that surface anaphora can also be controlled pragmatically. However, Merchant (a) has provided evidence for analyzing putative examples of English VPellipsis controlled pragmatically as instances of silent deep anaphora. I think that a similar approach is generalizable to (): they are empty nouns pragmatically controlled. I propose then that the sentences in () contain ENs minimally specified for some features like gender. Thus, the DP in (a) can be analyzed along the following lines: ()

DP D

NumP

la Num [sg]

nP nP

n[+gender]

AP roja

It seems that one crucial property that these empty nouns encode is that the entity pointed to must be a discrete entity, something that can be indeed signaled and individuated depending on properties of the context and an act of demonstration. In other words, the difference between cases like () and, say, human ENs is that the former are pure demonstratives (in Kaplan’s  sense) that require an associated demonstration to be complete. As is well known, the notion of associated demonstration is vague as it does not necessarily imply a stereotypical act of pointing; some salient properties of the context seem to be enough in order to license a correct use of a demonstrative. Again, more research is needed in this almost unexplored area with the detail it deserves. In any case, we can reasonably conclude that pragmatically controlled ENs should not be confused with NPE.

. L  

..................................................................................................................................

.. Some background A crucial question for any theory of ellipsis is to what extent languages can vary in the elliptical constructions they allow. Although not always explicit, the idea is that natural languages do not parameterize the identity condition. We assume that anaphoric recoverability in general is not subject to different language-particular conditions. In other words, recoverability conditions are universal (see Chomsky ). The obvious next question is





then why some languages have productive NPE or VPE while others do not. The answer involves what we might call the licensing component of the theory of ellipsis. Two prominent formal theories in the literature on NPE are Saito and Murasugi’s (a) Specifier-Head agreement approach (see also Lobeck ) and Lobeck’s () proper government approach. According to the first view, a given head licenses the ellipsis of its complement whenever such a head has an agreeing specifier. The following pattern is taken as evidence in favor of this line of analysis: () a. b. c.

I have read Bill’s book, but I haven’t read [DP John’s [NP book]] *I have edited a book, but I haven’t written [DP a [NP book]] *I have seen the book, but I haven’t had a chance to read [DP the [NP book]] (Saito et al. : )

Note that only in (a) is the specifier of the DP filled and, as predicted, it allows for NPE. The appeal of this proposal is that it seems to generalize to other types of ellipsis— concretely, to VP-ellipsis and sluicing. For instance, it is well-known that only some C heads license sluicing, in particular, those that are in a Specifier-Head configuration: () a. John bought something, but I don’t know [CP what [TP he bought t]] b. * John insisted that he turned in his homework, but I wasn’t sure [CP whether [TP he turned in his homework]] c. *John insisted that he turned in his homework, and Bill reported to Mary [CP that [TP he turned in his homework]] (Saito et al. : )

Similar patterns are attested in VP-ellipsis contexts. Schematically, the theory can be illustrated for each type of ellipsis as follows: ()

NP-Ellipsis DP XP

VP-Ellipsis TP

D’ D

XP NP

Sluicing CP

T’ T

XP VP

C’ C

TP

Beyond its initial appeal, Saito and Murasugi’s idea that Specifier-Head agreement is what accounts for some types of elliptical phenomena across languages has some shortcomings. Its main weakness is that its empirical coverage is too restricted: languages show many other forms of ellipsis than those that would depend on such an abstract configuration. Moreover, there are Specifier-Head agreement configurations that do not license ellipsis (e.g., sluicing is not generally allowed in relative clauses). These reasons might be behind Lobeck’s () reformulation of the theory of formal licensing. First, for Lobeck, ellipses (i.e., NPE, VPE, and sluicing) are not derived by deletion. On her account, ellipsis sites are instances of the so-called little pro that must be licensed and identified through government (see Rizzi ). Here is Lobeck’s particular implementation:

 



Licensing and Identification of pro: () An empty, non-arbitrary pronominal must be properly head-governed, and governed by an X0 specified for strong agreement. (Lobeck : ) The crucial notion is strong agreement. Lobeck proposes that a given head X0 is specified for strong agreement if and only if the head or the phrases and heads with which X0 agrees morphologically realize agreement in a productive number of cases (Lobeck : ). Note that this approach explains why the configurations in () license ellipsis, but also extends to cases where the X0 itself is inherently specified for certain formal features. In English, for instance, [+plural] is taken by Lobeck as a strong feature, which would derive why certain demonstratives and quantifiers license NPE without the need for Specifier-Head agreement: () a. Mary likes those books, but I like [DP these [e]]. b. The books were new, and [DP all [e]] were on syntax.

(Lobeck : )

This contrasts with the or a we saw in (b,c) where the plural feature is not realized on the relevant D head. Other strong features in English are [+possessive] and [+ partitive]. I refer the reader to Lobeck for detailed analyses of particular paradigms. At any rate, the problem with this government-based approach is not only that the conceptual apparatus is not consistent with current approaches to formal syntax, but also that it makes the empirically suspicious claim that the particular morphological requirements some NPE gaps show are an indisputable indication of the role that morphology plays in licensing ellipsis. So the fact that Germanic languages like Dutch or German license NPE ellipsis under the condition of having inflectional adjectives (see also Kester a,b) or that Romance languages use designated inflected determiners and not uninflected ones in NPE environments is taken as a demonstration that inflectional morphology is needed in order to license ellipsis. As we will see in section .., these facts seem to be independent of the licensing problem per se. Other current approaches to the licensing problem in ellipsis are also controversial. Theories based on some putative contrast condition on remnants (Giannakidou and Stavrou  or Eguren ) are falsified by the mere existence of non-contrastive remnants (see Saab  and Saab and Lipták ). Theories based on the quantificational nature of adjectival remnants (Sleeman , ) fail because of the existence of nonquantificational remnants (such as simple thematic PP remnants). More morphologically oriented approaches, which attribute a crucial role to word marker projections (Bernstein ) or gender/classifier ones (Alexiadou and Gengel ) are not only too weak but also empirically incorrect at least for Romance. Attributing licensing to a gender or classifier phrase in Spanish, as Alexiadou and Gengel do, overgenerates gender mismatches under nominal ellipsis (see ... and Saab a for more discussion). Finally, the proposal to assimilate licensing to D-linked functional heads has a flavor of circularity in part because of the vagueness of the notion of D-linked functional heads (López ). Of course, I do not wish to deny the important contribution of the aforementioned works. The fact that not every complement/modifier can be a legitimate remnant under NPE is an important discovery of the past decades. There are two aspects to be distinguished: (i) on the one hand, some remnants must show some sort of inflectional





morphology, (ii) on the other hand, remnants must bear some particular semantic import. Government approaches (e.g., Lobeck , Kester a,b) focus on the first aspect of the problem. Semantic accounts (e.g., Giannakidou and Stavrou  or Eguren ) center on the second one. I claim here that neither of these aspects forms part of the theory of ellipsis licensing. As we will see in the subsection that follows, morphological effects in NPE are epiphenomena arising from the way in which morphology resolves some stranded affix filter configurations. As for the second aspect, the issue is still poorly understood, but the restrictive character of remnants seems to be a necessary condition for ellipsis to apply, as already observed in Hernanz and Brucart (). For Sleeman (), this would follow from the need to look for an antecedent. So, in el auto rojo y el verde ‘the red car and the green one’, the restrictive nature of the color adjective in the second conjunct makes the linguistic antecedent salient. This does not happen with non-restrictive modifiers which cannot license ellipsis. Thus, prenominal adjectives like pobre ‘poor’ in el pobre hombre ‘lit. the poor man’ cannot be legitimate remnants for a gap like el pobre ‘the poor’ which can only be restrictive when understood as an instance of NPE (i.e., Había varios hombres en la fiesta: uno pobre, uno rico . . . ‘There were several men in the party: a poor one, a rich one . . . ’); otherwise, it is interpreted as a human empty noun. This kind of semantic-discursive effect on remnants led some grammarians to exaggerate the role of contrast and focus in NPE (Giannakidou and Stavrou  and Eguren , among others). This position is criticized in Saab () and Alexiadou and Gengel (), among others, on empirical grounds. At any rate, whatever the ultimate explanation for the distribution of remnants, it is important to have the morphological and the semantic factors separated. In the rest of this section, I address some morphological effects in nominal ellipses and try to show that they are epiphenomena derived from the way in which morphology and syntax interact in NPE contexts.

.. Licensing as selection Licensing can be implemented in a purely mechanistic way. Call this (maybe skeptical) way to think the E-feature approach, which is pursued in Merchant () and subsequent work. It simply states that some phrases are eligible to be elliptical in a given language for the syntactic occurrence of a specific ellipsis feature, [E]. Thus, English, but not Spanish, has VP-ellipsis because the functional node T can be optionally specified for taking elliptical complements. This mechanistic way of thinking has a positive consequence, namely, that ellipsis is a matter of phrasal selection. In the nominal domain, it predicts different sorts of nominal ellipses depending on the loci of the E selection feature—plus, of course, other conditions on possible remnants, on the one hand, and legitimate morphological outputs, on the other: ()

DP → NumP ellipsis D[e]

NumP nP

Num[e] n[e]

→ nP ellipsis → RootP ellipsis √P

 



The explicative force of such a view depends on the empirical justification of this formal feature and also on the grammatical correlations that it implies within and across languages.

.. Government effects in nominal ellipses as epiphenomena We must wonder then how the selection theory of ellipsis licensing may account for the morphological effects attested in different nominal ellipses across languages. Space limitations prevent us from doing justice to the empirical richness of this aspect of nominal ellipsis, but we will advance some lines of research.

... Selection by Num: nP ellipses Let’s start with nP ellipses (see also ()): ()

DP D

NumP Num[e]

nP n

√P

Depending on some properties of the morphological make-up of DPs, we predict different government effects. There is, indeed, a clear systematic pattern according to whether (i) the language at hand has nP ellipsis or not, and (ii) the language is agglutinative or inflectional. For agglutinative languages with productive nP ellipsis, it is common to observe that number and case morphemes get stranded whenever nP ellipsis applies. Consider the following examples from Hungarian taken from Saab and Lipták (): () Mari a régi kis ház-ak-at látta. Én az új nagy-[__]*(-ok-at) Mari the old all house-- saw I the new big -- ‘Mari saw the small old houses. I saw the big new ones.’ () Mari a régi kis ház-ak-at látta. Én az Mari the old small house-- saw I the ‘Mari saw the small old houses. I saw the (small) new ones.’ () Mari egy ház-Ø-at látott. Én Mari one house- saw I ‘Mari saw one house and I saw four.’

négy-[__]*(-Ø-et). four --

új-[__]*(-ak-at). new --





() Én a János mellett-i szék-Ø-en ültem. Ők a I the Janos ne- chair-- sat they a Péter mellet-i-[__]*(-ek-en). Peter next- -- ‘I sat on the chair next to János. They on the ones next to Péter.’ What is interesting about this type of example is that Hungarian adjectives do not inflect in number or case in non-elliptical contexts; such an inflection is mandatory when ellipsis applies, as in ()–(). As shown in detail in Saab and Lipták (), strictly speaking, these are not agreement markers (pace Kester a) but the stranded case and number affixes that are attached to the adjacent adjectival remnant under specific conditions. This is not a particularity of Hungarian grammar but rather is attested in other agglutinative languages like Turkish (Saab ), Quechua (Weber ), and Persian (Ghaniabadi ). We conclude then that the special adjectival inflection that NPE remnants show is the result of nP ellipsis, which leaves the number and case morphemes stranded. Inflectional languages with productive nP ellipsis, like most Romance languages, resolve this stranded affix scenario by deleting the stranded number affix via number identity with an agreeing morpheme. This would explain the observation that determiners accompanying nominal gaps must minimally inflect in number, as illustrated in the following Spanish example from Kornfeld and Saab (): () ¿{Qué/cuáles} libros de Borges y {*qué/cuáles} libros de Bioy . . . ? what/which: book: of Borges and *what/which: book: of Bioy ‘Which books by Borges and which ones by Bioy . . . ?’ Number agreement on cuáles ‘which.pl’, then, licenses the deletion of the conflictive plural feature specified on the Num head: ()

DP D

NumP

cuáles[+pl] Num[e]

nP √P

n

[+pl]

If this additional deletion operation did not apply, then a stranded affix filter violation would arise at PF. This is exactly what happens in the case in which uninflected qué fills the D head position: ()

*DP D qué

NumP nP

Num[e] [+pl]

n

√P

 



Crucially, the ungrammaticality qué triggers seems to be enough to reject a NumP-ellipsis analysis for these ellipses, at least under our assumptions (see Eguren  for another account). If the Num head were part of the ellipsis site, then it would be deleted under identity with the antecedent NumP.11 Under the NPE analysis, then, government effects are illusory, the surface reflex of a stranded affix filter configuration.

... Selection by D: NumP ellipses Suppose now that the licensor is D and not Num. The predictions about the syntactic and morphological correlations we should expect are quite different from those for nP ellipsis, because now there is no need for “rescuing” stranded number affixes, so no government effects would show up. The syntactic/LF correlate of this is that, as a counterpart, number mismatches should not be allowed.

()

DP NumP

D[e] Num

nP n

√P

The best-known example of NumP-ellipsis is English (for a first approach in terms of ellipsis, see Jackendoff , , and Lobeck  for NumP-ellipsis analysis in terms of the theory of pro): ()

Rome’s destruction and [DP Carthage’s [NumP [nP destruction t]]]

As already mentioned, the morphological correlate of selecting NumP as elliptical is absence of government effects for the category of number. Put differently, this type of nominal gap does not require number-inflected adjectives or determiners, as number is part of what is being deleted. In turn, the syntactic/LF correlate of NumP-ellipsis is that number mismatches should not be allowed. In simple cases, like John’s book/s and Peter’s, the elliptical gap is interpreted as singular or plural depending on the number information encoded in the antecedent. Interestingly, adding grammatical information— through, for instance, verbal agreement—makes number mismatches acceptable for the

11 Of course, another crucial argument against the NumP-ellipsis analysis is the asymmetry between gender and number with respect to identity effects already discussed in section ... (see the contrast between () and ()).





speakers I consulted, with a preference for the identical cases for some of them ((a) and (c)):12,13 ()

a. John’s book is on the table but Peter’s is on the desk. b. % John’s book is on the table but Peter’s are on the desk. c. John’s books are on the table but Peter’s are on the desk. d. % John’s books are on the table but Peter’s is on the desk.

The % symbol ranges from speakers that found the sentences perfect to those that found a subtle difference between the identical sentences and the non-identical ones. The fact that we find subtle variation across speakers contrasts the behavior of gender with that of number features in nominal ellipses. In effect, recall that gender concord does not improve the basic cases we saw in section ... (e.g., *el tío de Juan y la de María ‘the. uncle of Juan and the. of María’). The question is why number specification in verbal agreement improves the number mismatches but gender concord does not. I will not try to formulate an answer here as the matter was not discussed in the previous literature. Instead, I will provide a robust piece of evidence in favor of distinguishing NumP and nP ellipses. Assuming, as is standard, that numerals are specifiers (or modifiers) of the Num head, a clear prediction arises. Concretely, numerals should be part of the ellipsis site whenever NumP-ellipsis applies. According to the informants I have consulted, this is correct: the example in () is strongly deviant followed by the assertion that Mary has ten books of poems.14 Since the NumP antecedent contains the numeral three, the ellipsis site must be also modeled as containing an identical numeral. Evidently, this is inconsistent with the follow-up assertion of Mary having ten books.15

12

(i) (ii) (iii) (iv)

Arnold Zwicky contrasts the following examples in his blog: I accept the first argument, but reject the other two ___. [understood arguments] I accept the first two arguments, but reject the third ___. [understood argument] That was your dream. Kim’s ___ were all nightmares. [understood dreams] Those were your dreams. Kim’s ___ was a nightmare. [understood dream] (http://arnoldzwicky.wordpress.com////nominal-ellipsis)

According to Zwicky, (i) and (ii) are fully grammatical, but (iii) and (iv) require extra processing work, even when verbal agreement provides the relevant information for the elided number feature inside the DP. It is not clear to me whether the particular grammatical status that Zwicky assigns to (iii) and (iv) is in accord with the judgments that my informants provided for (). At any rate, it is interesting that he finds (i) and (ii) fully grammatical. If confirmed, this contrast would follow from the size of each gap, NP in (i)–(ii) and NumP in (iii)–(iv). 13 Thanks to Dave Embick, James Griffiths, Jason Merchant, and Gary Thoms for judgments and comments. 14 I follow Yoshida et al.’s (b) hypothesis that the remnant moves to the right of the ellipsis site in English, but the point is orthogonal to this particular argument. For Spanish NPE, I will continue assuming that remnants move to the left, as in Ticio () and Saab () (see the discussion surrounding () in section ...). 15 Thanks to Patrick David Elliott for this particular example. Other informants share exactly the same intuition.

  ()



#John had time to read [DP David’s [NumP three [NP books of music]]], but not [DP Mary’s [NumP three book t] of poems], since Mary has TEN books of poems.

Interestingly, similar examples are perfectly coherent in Spanish NPE: () Juan compró los tres libros de Borges y yo compré Juan buy:[.] the:: three books of Borges and I buy:[.] [DP los [XP [de Cortázar] [NP libros t ] ] ], que eran más de diez the:. of Cortázar books that were more of ten ‘Juan bought the three books by Borges and I bought Cortázar’s, which were more than ten.’ There is then evidence for distinguishing these two ellipses, beyond the nature of number and gender mismatches, which, needless to say, deserves careful examination.

... Selection by n: RootP ellipses As shown by Saito and Murasugi (a), some types of Japanese NP-ellipses seem to parallel what we call in the previous section NumP-ellipsis (see the Japanese examples in (), repeated in (), and Saito et al.  for a recent approach), where ellipsis is licensed by the sole presence of genitive remnants. As we have seen in the introduction, the form of the genitive is homophonic with the no anaphor (see () repeated in ()): N’-ellipsis () a. [Taroo no taido]-wa yoi ga, [Hanako no taido]-wa yokunai Taroo no attitude- good though Hanako no attitude- good-not ‘Though Taroo’s attitude is good, Hanako’s isn’t.’ b. [Rooma no hakai]-wa [Kyooto no hakai]-yorimo hisan data Rome no destruction- Kyoto no destruction-than miserable was ‘Rome’s destruction was more miserable than Kyoto’s.’ (Saito et al. : ) Anaphoric -no () Akai no–o mittu kudasai red one- three give.me ‘Please give me three red ones.’

(Takita : )

An important difference between both types of phenomena is that anaphoric no cannot be used to replace eventive nouns: () *Taroo-no koogeki-wa [DP totemo Taroo- attack- very ‘Taroo’s attack was a very strong one.’

tuyoi strong

[NP no]] data. one was (Takita and Goto : )

I will assume here that the anaphoric no construction is amenable to an EN analysis. In turn, () has all the properties of a surface anaphora (NumP-ellipsis, strictly speaking). Interestingly, in dialects that morphologically distinguish the genitive and the pronominal





form, such as the Nagasaki dialect of Japanese, co-occurrence of a genitive remnant (‑n) plus the putative proform to is attested. Consider (). In (a), the particular -n form of the genitive marker is shown, whereas (b) illustrates the particular form of the anaphoric empty noun to.

()

Nagasaki Japanese a. Takuya-n {keitai /hahaoya /taido /aizyoo} Takuya- cell.phone /mother /attitude /love ‘Takuya’s {cell phone/mother/attitude/love}’ b. Mariko-wa aoka to-ba Mariko- blue one- ‘Mariko bought a blue one.’

katta. bought (Maeda and Takahashi )

In (), the genitive marker and to co-occur: () Haruna-n taido-wa Mariko-n to yorimo rippayatta. Haruna- attitude- Mariko- one than good. ‘lit. Haruna’s attitude was better than Mariko’s one.’ (Maeda and Takahashi ) At first sight, one could be tempted to adopt an EN analysis for (). However, Maeda and Takahashi () present evidence in favor of an analysis in terms of ellipsis. One such piece of evidence comes from extraction of internal arguments in multiple genitive constructions: () a. Haruna-n piano-n toriatukai-wa teineiya Haruna- piano- handling- careful ‘Though Haruna’s handling of the piano is careful,’ b. Mariko-n to-wa sozatuya Mariko- one- rough ‘lit. Mariko’s one is rough.’

kedo, though

ne. 

c. Mariko-n furuuto-n to-wa sozatuya Mariko- flute- one- rough ‘lit. Mariko’s one of the flute is rough.’

ne.  (Maeda and Takahashi )

As mentioned in section ..., extraction from an ellipsis site can be considered as robust evidence in favor of ellipsis. As proposed by Maeda and Takahashi, the derivation for the elliptical gap in (c) should minimally contain the trace of the internal argument: ()

[DP Mariko- [DP flute- [D’ [ . . . [XP tflute handling]to] D]]]

As the reader can check, this analysis resembles the pseudogapping analysis already discussed for Spanish NP-ellipsis and English NumP-ellipsis (see section ...). As in those ellipses, here the internal argument of the noun is also extracted to some peripheral position in the DP structure, which is an indication of the surface anaphor behavior of this particular construction.

 



As also noticed by Maeda and Takahashi, the form to cannot co-occur with overt nouns: () a. *Haruna-n taido-wa Mariko-n taido to yorimo rippayatta. Haruna- attitude- Mariko- attitude one than good. ‘lit. Haruna’s attitude was better than Mariko’s one attitude.’ b. *Haruna-n taido-wa Mariko-n to taido yorimo rippayatta. Haruna- attitude- Mariko- one attitude than good. ‘lit. Haruna’s attitude was better than Mariko’s one attitude.’ Moreover, to occurs below classifiers like satu, which by assumption occupy a designated functional head, a Class(ifier)P projection (see Tang  and much subsequent work): () a. Haruna-wa [Murakami-n hon san-satu]-ba Haruna- Murakami- book three-- ‘lit. Haruna bought Murakami’s three books.’ b. Mariko-wa [Isaka-n to ni-satu]-ba Mariko- Isaka- one two-- ‘lit. Mariko bought Isaka’s two ones.’

katta. bought

katta. bought

The solution proposed by Maeda and Takahashi is that to is the surface form of the n head, which contains the [E] feature licensing ellipsis. The elliptical DP in (b) is represented as follows: ()

DP D’

Isaka-n

NumP ClassP nP

Class

NP

n

hon

to

D

Num 2

satu

Thus, to is seen as the nominal counterpart of English do in VP-ellipsis. Just like English do, Nagasaki to does not have any semantic import; it just supports a stranded head in elliptical contexts. This analysis accounts for all the properties seen above; i.e., the extraction facts, the impossibility of co-occurrence with overt nouns, and its relative position with respect to classifiers. A similar analysis could be extended to some dialects of Dutch. Consider the case of Afrikaans and Frisian, both languages that have been argued to use a pronominalization strategy. According to this analysis, Afrikaans, among other strategies, makes use of the proform een ‘one’ to fill some nominal gaps:





() Jan het [’n wit konyn] gekoop en Pieter het Jan has a white rabbit bought and Pieter has [’n swart een ] gekoop. a black one bought ‘Jan bought a white rabbit and Pieter bought a black rabbit/a black one.’ (adapted from Corver and van Koppen : , ex. (b)) However, as noticed by Corver and van Koppen (), Afrikaans een can combine with internal arguments: () a. % Ek vind [die koningin van Nederland ] meer aangenaam I find that queen of the.Netherlands more kind as [die een van Engeland ]. than the one of England b. Ek vind [ die koningin uit Nederland ] I find that queen from the.Netherlands as [ die een uit Engeland ]. than the one from England

meer more

aangenaam kind

() [Die gerug dat Bill ontslaan sou word] en [ die een dat Marie the rumor that Bill fired would be and the one that Marie haar pos sou behou], veroorsaak baie oproer. her job would keep caused much commotion (Corver and van Koppen : , n. , examples (i)–(ii)) Frisian behaves similarly. Among other ways to produce nominal gaps in the language, Corver and van Koppen observe that the form -en attached to adjectival remnants can be followed by a nominal gap optionally realized by a zero morpheme or the form ien ‘one’. () Jan hie in witte auto en Geart in swarten (ien). Jan has a white:e car and Geart a black:en (one) ‘Jan has a white car and Geart a black one.’ (adapted from Corver and van Koppen : , ex. (b)) As shown in () and () both strategies can take internal arguments of the missing noun as remnants: () Jitse wiisde him op [in posityf besprek Jitse pointed_out him to a positive review Jitske op in [negativen (ien) fan syn Jitske to a negative:en (one) of his ‘Jitse pointed out to him a positive review of his novel negative review of his collected poems.’

fan syn roman] en of his novel and samle fersen]. collected poems and Jitske pointed out a

 



() Jan krige [in mûnlinge meidieling dat syn omke siik wie] Jan got an oral announcement that his uncle ill was en [in skriftliken (ien) dat syn heit stoarn wie]. and a written:en (one) that his father died had ‘Jan got an oral announcement that his uncle was ill and a written announcement that his father had died.’ (Corver and van Koppen : , n. , examples (i) and (iii)) According to Corver and van Koppen, the Afrikaans and Frisian facts are an indication that, as opposed to English one (although see the discussion in section ...), some empty nouns like een and ien can inherit the argument structure properties of the antecedent noun (Corver and van Koppen : , n. ). This stipulation is needed in a theory like Corver and van Koppen’s and related ones, according to which nominal ellipsis sites are empty proforms.16 An alternative way to account for this particular paradigm could assume that these variants of Dutch make use of RootP-ellipsis, which would bring the analysis of these facts in line with the other data discussed in this subsection. So the underlying representation for a case like () and () in Frisian, for instance, would have a full-fledged RootP in the nominal gap position with all its theta assigners abstractly represented in the syntax. Again, as in other cases explored in the chapter, the internal arguments of the elided RootPs would vacate this projection in the relevant cases (i.e., when they are visible as in () and ()). Evidently, this view would have to assume that the occurrence of a putative proform like een or ien is the morphological reflex of the surviving n head that selects the elided RootP, as in Nagasaki Japanese.17 Yet such an account would still have to explain the doubling effect we observe in Frisian, among other potential problems. Hence, the final form of the analysis would depend on the results of other NPE diagnostics in these Dutch dialects.

.. Summary The E-selection approach to the licensing problem predicts different sorts of nominal ellipses depending on the size of the elliptical site, which is selected by the [E] feature. It seems that there is good evidence to postulate the following elliptical sizes: nP ellipses, NumP ellipses, and RootP ellipses, among other options (e.g., Class(ifier)P-ellipsis). In turn, this approach predicts different morphological effects depending on the size of the elliptical site and internal properties of each language (inflectional or agglutinative, for instance). Thus, we have seen that there are at least three morphological strategies that give us what in other accounts are considered as government effects, namely: (i) morpheme dislocation (Hungarian), (ii) morpheme deletion (Spanish), and (iii) support strategies (Nagasaki Japanese). If correct, this approach allows us to dispense with 16

Indeed, this problem applies to Lobeck’s () general theory of ellipsis, according to which all types of ellipses (i.e., NPE, VPE, and sluicing) have underlying pros. 17 As discussed in section ..., Llombart-Huesca () also defends an analysis along these lines for English one.





the morphological licensing component of the theory of nominal ellipses proposed in previous analyses.

. C

.................................................................................................................................. In this chapter, I have presented several diagnostics to distinguish empty noun constructions from different sorts of nominal ellipses. Ultimately, nominal gaps within and across languages are derived under one or the other analysis and their subtypes. Thus, first of all, we have seen that empty nouns are lexically restricted to express some general concepts and are sensitive to particular morphological processes. Moreover, their structure is that of a pronoun (i.e., a set of nominal functional rootless projections) as is also their semantic distribution. Second, nominal ellipses are derived from full-fledged nominals via an operation that gives us their silent final forms (i.e., deletion or non-pronunciation). Such an ellipsis operation is minimally subject to a lexical identity condition, although other pragmatic and semantic accommodations may be at work. Third, phrasal selection by an [E] feature gives us different forms of nominal ellipses, depending on the locus of such a feature: nP ellipsis, NumP ellipsis, RootP ellipsis and so on. This account predicts different illicit morphological outputs which are resolved by different strategies across languages, like morpheme dislocation, morpheme deletion, and morpheme support. Thus, the so-called morphological licensing in NPE (Lobeck ; Kester a,b) is illusory. The survey of nominal ellipsis phenomena we have explored here is, of course, incomplete.18 I hope, however, that the diagnostics presented and discussed throughout this

18 For instance, I have remained silent about the nature of null arguments in languages like Spanish or Japanese. As we have seen in the introductory section (cf. examples in () and ()) both languages make use of null arguments in subject and object position. However, it is well known that the distribution of such arguments differs between these two languages. An important observation is that while null arguments in Japanese allow for sloppy readings, Spanish null subjects do not. This observation is due originally to Oku (), who provided the following minimal pair.

Japanese: strict reading OK, sloppy reading OK (i) a. Mary-wa [zibun-no teian-ga saiyo-sare-ru-to] omotteiru. Mary- [self- proposal- accept---] think ‘Mary1 thinks that her1 proposal will be accepted.’ b. John-mo [e saiyo-sare-ru-to] John-also [e accept---] Lit. ‘John also thinks e will be accepted.’

omotteiru. think

Spanish: strict reading OK, sloppy reading * (ii) a. María cree que su propuesta será aceptada Maria believes that her proposal be:. accepted ‘Maria believes that her proposal will be accepted.’ b. Juan también cree que e será aceptada. Juan also believes that it be:. accepted ‘Juan also believes that it will be accepted.’

(Oku : )

 



chapter make some contribution to a better understanding of nominal ellipsis and its subtypes in future research.

A I would like to thank Tanja Temmerman and Jeroen van Craenenbroeck for inviting me to write this chapter and for their endless patience and support at every stage of this long process. I also thank Marjo van Koppen and an anonymous reviewer for comments, criticisms, and corrections that considerably improved the final result. My gratitude extends to Dave Embick, Patrick David Elliott, James Griffiths, Jason Merchant, and Gary Thoms, who helped me a lot with English judgments and references. Finally, many thanks to Verónica Ferry for proofreading the chapter. The usual disclaimers apply.

Thus, while the null subject in the Japanese example in (ib) is ambiguous between a strict and a sloppy reading, according to which either John thinks that Mary’s proposal will be accepted or his (=John’s) own proposal will, the null subject in (iib) only admits the strict reading, according to which the empty subject can only refer to María’s proposal and not to Juan’s. A prominent line of analysis, mainly represented by Oku (), Saito (), and Takahashi (), among others, claims that the difference between Spanish and Japanese must be done on the basis of the surface/deep anaphora distinction. Thus, while Japanese null arguments would derive from a radical DP-ellipsis analysis, Spanish null subjects would be base-generated proforms (i.e., they are deep anaphora). Given that deep anaphora do not allow for sloppy readings in the general case, the contrast in (i) and (ii) would be accounted for (although see Tomioka  for another analysis of sloppy readings in Japanese). If this is on the right track, then the distinction between nominal ellipses and empty nouns would generalize to cover also the entire DP structure.

  ......................................................................................................................

   ......................................................................................................................

 

. I

.................................................................................................................................. T chapter looks at two constructions which are sometimes characterized as involving ellipsis. One is known as gapping, a name Ross () suggested. And the other is variously known as stripping or Bare Argument Ellipsis.1 An example of gapping is (), and stripping is illustrated by (). ()

Jones likes seafood and Smith, bread.

()

Jones likes seafood a lot, and bread too.

The second conjunct of () is understood to have the same verb that the first conjunct has: likes. Similarly, in the second conjunct of (), there is understood to be both a verb and a subject, each identical to those found in the first conjunct: Jones likes. These constructions are sometimes alleged to have the same source. They each involve material in a conjunct that is allowed to go unexpressed when there is identical material in the preceding conjunct. Gapping and stripping are typically described as being found only in conjunctions or disjunctions. Cases like () and () contrast with examples such as () and (). ()

a. b. c. d.

* Jones likes seafood because Smith, bread. * Jones likes seafood whenever Smith, bread. * Jones likes seafood although Smith, bread. * If Jones eats seafood, then Smith, bread.

()

a. b. c. d.

* Jones likes seafood because bread too. * Jones likes seafood whenever bread too. * Jones likes seafood although bread too. * If Jones eats seafood often, then bread too.

1 The term “stripping” might have been first coined by Hankamer (), though he uses it only to refer to the cases that would today be called “fragment answers” (see section .).

  



To the extent that this is correct, gapping and stripping fit the description of a process that has been hypothesized to give conjunctions and disjunctions their meaning. This process is sometimes called “Conjunction Reduction,” which allows the mappings in (). ()

Smith left and Jones left ! Smith and Jones left. Smith ate seafood and ate bread ! Smith ate seafood and bread.

Hankamer () is the first attempt to reduce gapping and stripping to Conjunction Reduction. One fact such an account might be able to capture is the putative generalization in Hernández () that for a language to have gapping it must use the same morpheme for coordinating DPs and clauses. English, for instance, uses and for both clausal coordination and DP coordination, as do all the other Germanic languages, the Romance languages, Hindi, Persian, and Turkish. Yoruba, Wolof, Chinese, Korean, and Japanese, however, have one morpheme that coordinates DPs and another that coordinates clauses. The first group of languages all have gapping and the second does not.2 If Conjunction Reduction cannot change the morphological form of the coordination morpheme, then this typology would be explained. An important component of Hernández’s survey is her assumption that gaps only arise in the second conjunct; she assumes that what Ross () calls “Backwards Gapping” doesn’t exist. Ross argued that examples like the Japanese () arise from gapping the verb from the first conjunct. () Watakusi-wa sakana-o soshite I-top fish-acc and ‘I ate fish and Bill ate rice.’

Biru-wa Bill-top

gohan-o rice-acc

tabeta. ate.

Ross suggested that whether a language has “Forward Gapping,” where the gap shows up in the second conjunct, or backwards gapping, correlates with whether the language has Verb–Object word order or Object–Verb word order.3 Hankamer (, ) and Maling () argue, however, that examples like () are not gapping, but have a different source: Right-Node Raising. Right-Node Raising is a process that causes material at the right edge of an initial conjunct to be expressed in a conjunct that follows.4 Because languages whose VPs end with the verb will allow Right-Node Raising to apply to their verbs, Right-Node Raising and backwards gapping will produce the same surface strings. Ross’s observation that only verb-final languages have backwards gapping can be explained if all such cases are in fact Right-Node Raising. Hernández adopts this now standard view and for this reason doesn’t classify Japanese as a language that has gapping.5

2

See Paul () for discussion of a limited form of gapping in Chinese. Koutsoudas () supports this correlation and seeks to build it into gapping. 4 See Hartmann (), Wilder (, ), Phillips (), Bachrach and Katzir (), Sabbagh (), and Ha (b) for discussions of Right-Node Raising. 5 But see Abe and Hoshi () for a recent reinvocation of Ross’s typology. 3





Reducing gapping to Conjunct Reduction predicts that it will be found only in coordinations and disjunctions, but there are examples which resemble gapping and stripping in other environments. For instance, comparatives include cases like (). ()

a. Jones eats seafood more often than Smith bread. b. Jones eats seafood more often than bread.

And there are dialogues whose independent sentences, like those in (), appear to arise from gapping and stripping. () a. Elizabeth: Who ate what? Claire: Jones seafood. b. Elizabeth: Jones ate seafood. Claire: Bread too. For the most comprehensive investigation of gapping in the context of comparatives, see Lechner (, ). For a recent sketch of how gapping might arise in question/answer pairs, see Reich (), Boone (), and Weir (). Those analyses of gapping and stripping that tie them directly to coordinations would require these cases to have a different source. Gapping and stripping also have superficial resemblances to certain other ellipsis processes. Gapping, for instance, looks similar to “pseudogapping,” which () illustrates. ()

Jones likes seafood and Smith does bread.

Pseudogapping, given that name by Stump (), and investigated more extensively with corpora in Levin (), in turn resembles “VP-ellipsis,” exemplified by (). ()

Jones likes seafood and Smith does too.

The salient difference between pseudogapping and gapping is the presence of the word that stands in the highest position of what I’ll call the “verbal series.” In (), that word is does; in non-tensed sentences, that word is a modal or the infinitival marker to. ()

a. Jones should like seafood and Smith should bread. b. ? For Jones to like seafood and for Smith to bread is improbable.

Gapping examples, by contrast, do not have this word. One strategy we’ll see is to make gapping the name we give to examples of pseudogapping in which the highest word in the verbal series is also elided. Stripping also looks very much like various forms of clausal ellipses. The short answers to questions like that in (), for instance, can have the superficial appearance of stripping. ()

Elizabeth: Who left? Claire: Jones!

  



See Merchant (a) for an account of () that involves ellipsis. There are also discourses like those in () which have a superficial similarity to (). ()

a. Elizabeth: Jones left. Claire: Smith too! b. Elizabeth: Jones left. Claire: No, Smith!

Questions themselves can, in the right contexts, involve an ellipsis that appears like stripping. Ross (b) called these cases “sluicing”; examples are in ().6 ()

a. Jones ate something but I won’t tell you what. b. Jones: I’ve eaten something! Smith: What?

How to situate gapping and stripping among the panoply of ellipsis types remains open, as does the question of whether they belong together as two cases of the same process. In sections . and ., we examine some of the problems that remain in understanding how to classify gapping and stripping.

. S

.................................................................................................................................. One early analysis of stripping, in Ross (), is that it involves a rightward movement rule. This putative rule would move the string made up of and and the second conjunct to the right in the way indicated in (). See Hudson (a, b) for an extended defense of this approach. () a. Jones ate seafood [and Smith] too.

b. Jones ate seafood today [and bread] too. Reinhart’s () proposal is similar; it suggests that Quantifier Raising (QR) can create an unspoken, but semantically interpreted, representation by moving a nominal to the left of and, as illustrated in (). ()

a. ate seafood Jones and Smith too. b. Jones ate today seafood and bread too.

Under both views, the idea is to generate a representation that would make these sentences semantically equivalent to (). 6

See Hoji and Fukaya () for the relationship between sluicing and stripping in Japanese.





()

a. Jones and Smith ate seafood. b. Jones ate seafood and bread.

But the meaning of stripping sentences and sentences with conjoined nominals is not exactly the same, suggesting that they shouldn’t be equated. For instance, because the subject argument of (a) and the object argument of (b) are plurals, they can support semantic material that depends on a plurality. (a,b) are therefore grammatical. ()

a. Jones and Smith ate seafood together. b. Jones ate seafood and bread together.

This isn’t true of the stripping versions, however. ()

a. * Jones ate seafood together and Smith too. b. * Jones ate seafood together and bread too.

Stripping has two “singular” conjoined clauses rather than one “plural” clause, like these accounts claim. There are other more syntactic problems with an approach that gives strips an underlying form like ()—Depiante () has a good list—but this is reason enough to abandon it.7 The other approach to stripping gives the second conjunct of a stripping construction its meaning through anaphora. This view is perhaps first defended in Ross (b), and given more extensive support in Hankamer and Sag (), Fiengo and May (), Heim and Kratzer (), Depiante (), Kolokonte (), and Algryani (). It assumes that there is a silent anaphor in the second conjunct whose meaning is recovered from the material in the first conjunct. If we represent that anaphor with ‘Δ’, the hypothesis is that stripping involves representations like those in (). () a. Smith ate seafood, and Jones Δ too. Δ = ate seafood b. Smith ate seafood yesterday and Δ bread too. Δ = Smith ate yesterday This correctly captures the fact that stripping can involve the conjunction of singular sentences. If () is correct, the anaphor involved is one that Hankamer and Sag () dubbed a “surface anaphor.” Surface anaphors get a denotation from their antecedent instead of referring independently to their antecedent’s referent. Thus, for instance, in (), ‘Δ’ is understood to have a denotation that, like its antecedent, includes an indefinite. ()

7

Smith ate some seafood, and Jones Δ too. Δ = ate some seafood See Merchant (a) for a similar semantically based argument against accounts like these.

  



As a consequence, the event of seafood eating that Smith and Jones participate in doesn’t have to be the same one in (), and the seafood they eat can, thankfully, be different. One way of deriving the fact that the anaphor in stripping is a surface anaphor is to let it be an ellipsis. Ellipses are uniformly surface anaphors, and this is typically derived by taking ellipses to be unspoken syntactic phrases. The antecedence conditions on ellipsis require that the elided phrase have a denotation that “follows” from the denotation of its antecedent. In the limiting case, the denotations of antecedent and ellipsis are the same. It is this view of stripping that dominates, and is the one we shall explore more fully here. If stripping removes a portion of the second conjunct whose meaning follows from that of an antecedent, then the words it elides should combine to have a meaning. This poses a problem for those examples in which the words that have elided don’t seem to form a constituent, as in (b). If the string ‘Smith ate yesterday’ which has elided in (b) is to form a meaningful constituent, we should expect the spoken form of (b) to be (). ()

Smith ate yesterday bread.

But the word order in () is generally judged ungrammatical, though the meaning it has is precisely the one needed in (b). There are cases of stripping and, as we shall see, gapping too, in which the meaningful constituent that is alleged to have elided is difficult to find among the corresponding ellipsis-less sentences of English. This is a cloud that hangs over the view that strips are kinds of surface anaphors. If they are anaphors, then the material that has elided must form a semantically contentful phrase, and in many cases the phrase cannot be found in unelided sentences. This occasionally encourages the idea that what has elided is merely a collection of words, and that how those words compose semantically is irrelevant. On this view, what would have elided in (b) are the words Smith, ate and bread independently. Williams () proposes a system like this and Ackema and Szendrői () employ it to analyze cases of determiner gapping that we will see in (). There are two problems that lie in taking up an approach along these lines, however. First, it needs to be prevented from running amok. In the majority of cases of stripping, the set of words that elide is very nearly just those that can compose into meaningful phrases. From (), for instance, only (a) is a possible strip. ()

Smith ate the stale bread before noon and ate the stale roll before noon too. a. Smith ate the stale bread before noon and the stale roll too. b. * Smith ate the stale bread before noon and the stale too. c. * Smith ate the stale bread before noon and the too.

This follows if only strings that form meaningful phrases can strip. A second problem is that there are cases which suggest that it’s not just the words that make up a strip and its antecedent that matter, but also how they combine semantically. For instance, () is four ways ambiguous. ()

Smith read Jones the riot act and Smith read Wier the riot act.





The words ‘read the riot act’ have an idiomatic meaning, roughly equivalent to ‘chastise’, as well as their compositional meaning: to read that famous  act of the British Parliament to a riotous assembly in order to dispel it. Both meanings are available to each of the conjuncts in (), and they can be independently chosen. But this is not the case for the stripped version of ().8 ()

Smith read Jones the riot act and Wier too.

Whatever meaning is given to the words ‘read the riot act’ in the first conjunct must be applied to those elided words in the second conjunct. Strips seem to depend on how the elided words are put together. We should restrict the groups of words that can strip to just those that make phrases. A virtue of taking strips to be ellipsis is that it provides a ready-made analysis of the cases where the strings that have elided do not seem to make a meaningful phrase. Because elided phrases are just those phrases unspoken, it is possible for material within them to move out. This provides a way of characterizing the strings that have elided in stripping constructions that is consistent with normal views of constituency. The sentence in (a), for instance, could arise from moving the object out of the elided constituent, as indicated in (). ()

Smith ate the stale bread before noon and ate before noon the stale roll.

This correctly predicts that the elided material in a strip should not contain an island from which the remnant moves. The ungrammaticality of the stripping examples in () matches the ungrammaticality of the island violations in (). (The sentence-final adverbs in the first conjuncts of () must be understood as belonging to the root clause, otherwise we cannot be sure that the elided phrase will contain the island.) ()

a. *Smith asked who ate seafood once, and asked who ate once bread too. b. *Smith danced after he ate seafood wildly, and danced after he ate wildly bread too. c. *Smith complained about the person who ate seafood unfairly and complained about the person who ate unfairly bread too d. *Smith noticed that eating seafood is dangerous suddenly and noticed that eating bread is dangerous suddenly too.

()

a. b. c. d.

* What did he ask who ate? * What did she dance after she ate? * What did she meet the person who ate? * What did he notice eating is dangerous?

8 The word too is present in (), but not (). Without too, () is either ungrammatical or requires an intonation (stress on and). If either this intonation, or the word too is added to (), then () also loses the meanings in which ‘read the riot act’ has a different denotation in each conjunct. I interpret this to mean that stripping requires either intonation or something equivalent to too to signal that the meanings of the deaccented portions of the two conjuncts are the same. That amounts to the same problem for the semantically independent theory of ellipsis that I am describing in the text.

  



Island effects are diagnostic of movement, and their presence in stripping implicates movement in forming the strings that elide. There are different kinds of movement operations, however, and they are sensitive to slightly different locality conditions. If the movement of the remnant out of the phrase which elides in stripping is like the movement of the interrogative pronouns which form the questions in (), then we should expect the phrases that can elide under stripping to be just those that this kind of movement can escape. We should expect, for example, the strings traversed by movement in () to be strippable. () a. What did he tell Jones that Smith ate in a letter?

b. What did he tell Jones to eat in a letter? But that doesn’t seem to be the case; (a,b) are degraded. (In these examples, in a letter must be understood to be part of the root clause, otherwise the stripped material would include only the embedded clause.) () a. * He told Jones that Smith ate seafood in a letter, and told Jones that Smith ate bread too. b. ?* He told Jones to eat seafood in a letter, and told Jones to eat bread too. If movement of the remnant is what is responsible for producing the constituents that stripping can elide, then the type of movement operation involved must be different than that used to form constituent questions. To identify what movement operations are implicated, one might try to match the constraints on these strings with the constraints that identify different kinds of movement operations. Indeed, there is a rough match between the strings that can strip and those that can elide in pseudogapping (= ()). () a. b. c. d. e. f. g.

Smith will eat seafood, but he won’t Δ bread. * Smith will tell Jones that Smith ate seafood, but he won’t Δ bread. * Smith told Jones to eat seafood, but he won’t Δ bread. * Smith asked who ate seafood, but he didn’t Δ bread. * Smith danced after he ate seafood, but he didn’t Δ bread. * Smith complained about the person who ate seafood, but he didn’t Δ bread. * Smith noticed that eating seafood is dangerous, but he didn’t Δ bread.

A popular account of pseudogapping (see Jayaseelan , Lasnik a, , and Aelbrecht ) is that it is simply VP ellipsis, from which the phrase that doesn’t elide has moved. Takahashi () argues that the operations responsible for moving the remnant out of a pseudogap are “Heavy NP Shift” and “Object Shift.” These two operations obey Ross’s islands and also cannot move things out of (most) clauses, so they have the same properties we’ve seen are necessary for stripping. This also correctly captures the fact that pseudogapping cannot strand a preposition or move a predicate, as () shows.

 ()

 a. * Smith stood near Jones, but he didn’t Δ Brown. b. * Smith made Jones a judge, but he didn’t Δ an executioner.

Neither Heavy NP Shift nor Object Shift can strand prepositions or move predicates. Interestingly, the grammaticality judgments for stripping a preposition without its complement, as Kuno (: , n. ) first noted, are not clear-cut. There are speakers who find () grammatical. ()

a. John talked about Mary but not Susan. b. This article appeared in the NY Times but not the Daily Telegraph.

Depiante () argues that the cross-linguistic grammaticality of examples in which the preposition is included in the strip, leaving its complement behind, correlate with the ability of the language to strand a preposition with movement. She finds that English examples like () contrast with Spanish examples like (). () a. * Pedro sabe sobre geografia más que nadie y/e Peter knows about geography more than anyone and también. too ‘Peter knows more than anyone about geography and history too.’

historia history

b. * Juan escribe para Clarín y La Nación también. John writes for Clarín and La Nación too ‘John writes for Clarin and La Nacion too.’ (Depiante : , (), (a,b)) Unlike English, Spanish doesn’t allow preposition stranding under any kind of movement. ()

a. Who did you talk to? b. * Quién hablaste con? who you talked to ‘Who did you talk to?’

(Depiante : , (a,c))

Similarly, I believe stripping has the ability to strand predicates too. I and my consultants find examples like () grammatical. ()

Smith made Jones a judge yesterday, and an executioner too.

If stripping has the ability to strand prepositions and predicates, and pseudogapping doesn’t, then a movement account of the strings that can elide must involve more movement kinds than just Heavy NP Shift and Object Shift: the movement operations that allegedly feed pseudogapping. One popular view is that stripping employs a movement operation that is tied to the information structure of the construction. Depiante (),

  



Kolokonte (), and Algryani (), among others, argue that the movement involved is to a position reserved for focus-marked phrases. This responds to the observation that the material left over from the strip is focus-marked, and the fact that in many languages, focused phrases are moved leftwards to either the edge of the local clause or the edge of the local VP. (See, e.g., Rizzi ,  and Belletti .) While this movement doesn’t occur overtly in English, we might speculate that it is nonetheless available. Adopting the view that the focus position is at the left edge of VP, this would give to examples like (), representations like (). ()

Smith spoke to Jones today and Brown too.

()

IP DP Smith

IP I

FocP

FocP

FocP VP

Foc V spoke

FocP

and DP1

PP P

DP

to

Jones

FocP

Brown Foc

VP V spoke

PP P

t1

to Brown has moved to the Specifier of a Focus Phrase in (), leaving a variable ‘(=t)’ in the position it moves from. Stripping would presumably elide the boxed VP in the second conjunct. Recall that ellipsis is allowed just when the denotation of the elided constituent matches, or follows from, the denotation of an antecedent constituent. This requirement is not trivially satisfied in (38): the constituent that elides has a trace of the moved phrase Brown in the same place that the antecedent has the DP Jones. One way of addressing the issue is to recognize that Jones in () is focused too, and allow it to covertly move, forming the representation in ().9

9

The way this hypothesis is usually expressed is by positing a focus position at the left edge of the clause, rather than the left edge of a VP. (See Depiante , for instance.) But there’s no reason for this, so far as I can see, and assuming a lower position enables a “low conjunct” analysis of stripping—see section ..





()

IP DP Smith

IP I

FocP

FocP DP2

FocP and

FocP

Jones Foc

FocP DP1

VP V spoke

Brown Foc

PP P to

FocP

t2

VP V spoke

PP P

t1

to It is known from other kinds of ellipses that phrases with variables in them can be in the right matching relationship for one to serve as the antecedent to the other even if those variables are bound by different things. When an elided phrase, and its antecedent, have different bound variables in them, we have a situation that Ross () called “sloppy identity.” Another solution would be to use an antecedence condition on ellipsis that allows focused phrases to behave like bound variables and partake in sloppy identity without moving. That would allow the focus-marked Jones in () to “match” the trace of Brown. Merchant () offers an account of this type. If movement to a focus phrase can make strippable strings, we should understand why this normally non-overt movement becomes possible in ellipsis contexts.10 And if it is to give us a handle on the locality conditions—like the inability of strips to include part of an embedded finite clause—then we should be able to discover that this locality condition holds of covert focus movement too. This is presently the most popular view of stripping. It is a form of ellipsis that results when a movement operation removes a focused item from the ellipsis. Many of the considerations that lead to this conclusion show up again in gapping. Indeed, certain of the properties that seem diagnostic of stripping—that it is confined to certain kinds of coordinations, for instance—are found in gapping too, fueling the hypothesis that they are the same phenomenon.

10

See Weir () for a solution to the similar problem in fragment answers that could be applied here.

  



. G

.................................................................................................................................. Like stripping, gapping is a surface anaphor, as can be seen from the fact that two books can be involved in (). ()

Some read a book to Melissa and others, to Paul.

This example also demonstrates that the verb gapping elides can be part of a longer string. Just as with stripping, it is necessary to find a way of characterizing the strings that can gap, and there is a rough match with the strings that can strip. Neijt () showed, for instance, that island effects are obeyed. ()

a. * Some asked who ate seafood, and others asked who ate bread. b. * Some danced after they ate seafood, and others danced after they ate bread. c. * Some complained about the person who ate seafood, and others complained about the person who ate bread. d. * Some noticed that eating seafood is dangerous and others noticed that eating bread is dangerous.

As with stripping and pseudogapping, then, we can hypothesize that the second phrase left behind by gapping has moved from the constituent that elides. Neijt also discovered that, as in stripping, the constraints on gapping are stricter than the islands which govern wh-movement would predict. For instance, the remnant of a gap cannot have moved out of an embedded finite clause, though, as we’ve seen, that is possible for wh-phrases.11 () * Some decided that she liked beans and others decided that she liked bread. Most of the literature takes gapping to be like pseudogapping in not permitting preposition stranding. Lasnik and Saito (), Jayaseelan (), and Abe and Hoshi () characterize examples like () as ungrammatical. ()

a. * Some talked about Jones, others Δ Brown. b. * Some built it with a hammer, and others Δ a crowbar. c. * Some put it on Jones, and others Δ Brown.

Steedman (), by contrast, provides examples of preposition stranding in English that seem perfectly grammatical, such as (). ()

11

Harry went to London and Barry, Detroit

She notes, however, that there are exceptions that arise when the embedded clause has a subject that corefers with the subject of the higher clause. See Moltmann () for discussion.





Vanden Wyngaerd () argues that the availability of preposition stranding in gapping tracks a language’s ability to strand prepositions under movement. I will assume that gapping in English does permit preposition stranding, but it is worth noting that this requires more investigation. Predicates are also capable of being stranded by gapping. ()

Some made Jones a judge, and others Δ an executioner.

If the remnants of gapping can include predicates and separate prepositions from their complements, then the class of remnants for gapping matches that for stripping. Like the remnants in stripping, the remnants of gaps are also contrastively focus-marked. Kuno () argued that the remnants of a gap must present discourse-new information, and Hartmann () and Winkler (, , ) show that the remnants of a gap must be in a contrastive focus relationship to matching material in the antecedent. If the explanation for this fact in stripping is that the remnant focus moves to a position outside the ellipsis, then we should be tempted to give the same explanation for gapping. An important question to be settled is what are the focus structures of gapping and stripping and to what extent they can be wedded to a theory of what remnants are possible. This is especially well discussed in Susanne Winkler’s work. There is an interesting argument in Yoshida () for the more general conclusion that the remnants of a gap have moved from the elided phrase. His argument is based on the generalization that the contents of phrases that have moved cannot themselves be moved. For instance, if (b) is derived from (a) by rightward movement of the PP, then the contrast in () indicates that the PP becomes impenetrable for movement in its derived position. ()

a. John talked about something yesterday. b. John talked yesterday [about something].

()

a. I wonder which topic John talked about yesterday. b. * I wonder which topic John talked yesterday about.

Using this as a diagnostic for movement, Yoshida argues that the remnant of a gap has moved. His centerpiece example is (). () a. I wonder which topic John talked about and Mary talked about too. b. * I wonder which topic John talked about and Mary about too. (Yoshida : ()) In (a), which topic has moved in across-the-board fashion out of a PP that resides in its unmoved position. (“Across-the-board movement” describes a scenario in which one term moves from two parallel positions in conjoined or disjoined phrases. See Ross .) The contrast with (b) suggests that when the verb gaps, the PP that remains has become an island. If this PP moved, then this will follow from whatever is responsible for the contrast in ().

  



As in the case of stripping, then, there is evidence that the strings which can gap correspond to those that are phrases from which something has moved. There are, however, a class of cases discovered by McCawley () which seem to be clear counterexamples. These are cases where a determiner seems to gap leaving the rest of its nominal behind.12 ()

a. b. c. d. e.

Too many Irish setters are named Kelly and German shepherds Fritz. The duck is dry and mussels tough. Your daughter is  and son ½. No representative voted for the proposition or senator against it. Few dogs eat Whiskas or cats Alpo.

If the gapped predicate is restored in these examples, they either become ungrammatical, or acquire a different meaning. ()

a. Too many Irish setters are named Kelly and German shepherds are named Fritz. b. The duck is dry and mussels are tough. c. * Your daughter is  and son is ½. d. * No representative voted for the proposition or senator voted against it. e. Few dogs eat Whiskas or cats eat Alpo.

The grammatical (a,b) and (e) differ semantically from the corresponding examples in () in having bare plurals as the subject of the second conjuncts. In the matching () examples, the subject of the second conjunct is understood to have a determiner: too many in (a), for instance. The absence of prenominal material in (c) and (d) is what makes these cases ungrammatical. McCawley’s conclusion is that gapping has removed the determiner from the subject of the second conjunct in (). While that might be a way of characterizing the examples in (a–c), which are synonymous with the examples in (), it doesn’t extend to (d–e) which aren’t synonymous with (). ()

a. Too many Irish setters are named Kelly and too many German shepherds are named Fritz. b. The duck is dry and the mussels are tough. c. Your daughter is  and your son is ½.

()

a. No representative voted for the proposition or no senator voted against it. b. Few dogs eat Whiskas or few cats eat Alpo.

Instead (d) and (e) have the meanings indicated in (). ()

12

a. ¬ [any representative voted for the proposition or any senator voted against it] b. ¬ [many dogs eat Whiskas or many cats eat Alpo]

(a–c) are from McCawley (). Determiner gapping has not been discovered in stripping, and this has no explanation if gapping and stripping have the same source.





The negation part of the meaning of no and few falls outside the scope of the disjunction. McCawley noted a similar fact about his (). () Not enough linguists study Russian, literary scholars French or engineers Japanese. (McCawley : , (a)) () is not a disjunction of denials (‘Not enough linguists study Russian or not enough literary scholars study French or . . . ’) but a denial of disjunctions (‘It’s not the case that enough linguists study Russian or enough literary scholars study French . . . ’). An account of these cases should derive the fact that the negation part of these determiners must be outside the coordination. Johnson (a,b) suggests that what is happening in (d) and (e) involves a more complicated syntax for the determiners no and few. If we adopt the view in Bech () that the determiner no is the exponent of two morphemes, one of which has the meaning of not and the other the meaning of any, then we can see (d) as involving a surprising way of spelling out that exponent. Suppose, for instance, that we assume no to arise when the not and any parts of the meaning are adjacent at some point of the derivation. ()

Let: a. [[ϕ]] = [[not]], and b. [[ψ]] = [[any]] pronounce ϕ as not and a ψ in the scope of ϕ as any, unless ϕ and ψ are adjacent, in which case pronounce ϕ as no and leave ψ silent.

If (d) is parsed as in (), then () causes the ψ in the first subject to be pronounced as no, but leaves the ψ in the second subject silent. () ϕ [ [DP ψ representatives] voted for the proposition or [DP ψ senator] voted against it.] A parallel syntax–semantics can be put together for (e); few could be seen as the exponent of a morpheme that means not and another that means many. Because this account lives on the complex two-part syntax of the determiners, it predicts that the only prenominal material that should be able to “gap” in this way are those terms which have this two-part syntax. McCawley discovered that neither prenominal adjectives nor the determiner a is able to gap. () a. b. c. d.

* A soup too salty and pie too sweet, but otherwise the food was outstanding. * An Irish setter should be called Kelly and German shepherd Fritz. * Italian red wines are outstanding and white wines excellent. * Red wines from Italy are outstanding and wines from France excellent. (McCawley : , () and , ())

Many other determiners, however, can.

   ()

a. b. c. d. e.



Every child carried a toy and father a box. Several women saw the accident and men the aftermath. Each marble rolled down the chute and ping-pong ball around the loop. Most students read books and professors articles. Which boy left and girl stayed?

The list of terms that can gap roughly match the class of items that trigger Quantifier Raising (QR)—the rule that fixes the scope of quantificational determiners. In some analyses of Quantifier Raising, the determiner involved does indeed have two syntactic parts: one that makes the DP involved a variable and the other that creates a binder for that variable. One possibility, then, is that the complex syntax of determiners quite generally is responsible for their ability to appear to gap. The implementation of this idea won’t be straightforward. It’s not clear, for instance, how to get it to remove the possessive pronoun in (c), or how to reconcile the ability of indefinite DPs headed by a to scope in the way that QR predicts but not be able to gap. A fact that may follow from an account along these lines is that a determiner may gap only if it is in a position consistent with it being at the left edge of the coordination. McCawley points out, for instance, that a determiner associated with an object cannot be gapped. ()

a. * The cat chased the mouse and Fido chased the cat. b. * Tabby chased the mouse and Fido chased the cat.

But if that object is at the left edge of the coordination, then its determiner can gap. ()

a. She told [the girl to stay] and [boy to leave]. b. He’ll give [no book to Smith] or [pamphlet to Jones].

If the relationship between the two parts of these complex determiners is part of the syntax of QR, then one way of viewing this fact is that it follows from conditions on across-theboard movement. Williams (a) shows that across-the-board movement is subject to a condition that requires the moved item to come from parallel positions in the coordinations. For instance, an item cannot move across-the-board from a subject position in one conjunct and an object position in the other conjunct; examples like () are ungrammatical. ()

* Who did she say [t left] and [Mary visited t]? compare: Who did she say [t left] and [t visited Mary?] Who did she say [he left t] and [you visited t]?

If the relationship between the part of the complex determiner that is outside the coordination (the ϕ in ()) and the parts of the determiner within the coordination (the ψs in ()) is subject to the same condition, then the parts of the determiner inside the coordination will have to be in parallel positions. Because the two parts of the determiner must be adjacent to get mapped onto their exponent, the DP in the left conjunct will have to be at





the left edge of that conjunct. This will force the DP in the right conjunct to also be at the left edge of its conjunct. On this view, the determiners have not gapped in (). Instead, the coordinations allow for a syntax that reveals the otherwise hidden complex relationship between the morphology of a determiner and its syntactic/semantic components. If that is the case, then we must understand why the coordinations in () without gapping do not allow this revelation: what is it about gapping that allows a determiner to be unexpressed? Levine and Kubota () argue that gapping is not required, but instead that a kind of parallelism which gapping is a special instance of is the relevant requirement. They offer the examples of successful determiner gapping without any obvious gap in (). ()

a. Some dog barked and donkey brayed last night. b. No dog barked or donkey brayed last night.

Similarly, Centeno () finds that wh-determiners in Spanish can also gap without an accompanying predicate gapping. () Cuántos estudiantes acuden a la clase de sintaxis y profesores how many students go to the class of syntax and professors faltan a la reunión de cada mes? miss to the meeting of each month ‘How many students go to the syntax class and how many professors miss the monthly meeting?’ (Centeno : , ()) Note that in both of these examples, the tenses of the conjoined clauses are the same, and at least for the English cases, this looks like a requirement. So whereas () and () are grammatical, cases like () where the tenses are different are worse. ()

Some dogs bark and donkeys bray at night.

() a. * Some dogs bark and donkeys brayed at night. b. * Some dogs barked and donkeys bray at night. A feature of gapping that Hankamer () discovered is that the tenses must be the same. In (), he noted, the conjuncts are either both present or both past.13 () The Tunisians cut off the tails of their cattle and the Muranians the testicles of their prisoners. (Hankamer : ) If, as is standardly assumed, the highest functional head in a clause is the one that carries the tense morpheme, then we might see this sameness in tense as a sign that gapping has applied 13

See also Chao ().

  



to just this tense morpheme, leaving the rest of the clause behind. Perhaps, then, the gapped determiners in () and () are accompanied by a gapped portion of the verb: its tense.14 If this method of analyzing gapped determiners is successful, then our working hypothesis can be that a movement operation is responsible for forming the strings that gap. Those same movement operations also form the strings that strip. Gapping and stripping are then cases of remnant ellipsis, just like pseudogapping, but with a wider inventory of movement rules. If the conclusion that gapping cannot elide material in the first coordinate is correct, then this also matches the inability of a pseudogap to be in this position: () * Some have beans and others have eaten rice. These similarities encourage the view that gapping, and stripping too, are kinds of remnant ellipsis constructions. A salient difference between gapping and pseudogapping is that the highest term in the verbal sequence elides in gapping but not pseudogapping. This is related to another property that distinguishes gapping and stripping from pseudogapping. If gapping or stripping occurs in the second of two conjuncts, then it must elide the highest term of the verbal sequence in that conjunct. Examples like () are ungrammatical. ()

a. * Smith left and everyone thought that Jones too. b. * Smith ate sushi and everyone thought that Jones rice.

Their pseudogapping and VP ellipsis correlates, by contrast, are fine: ()

a. Smith left and everyone thought that Jones had too. b. Smith ate sushi and everyone thought that Jones had rice.

Hankamer (: ), who discovered the constraint, calls it “downward bounding.” I’ll formulate it with (). ()

Downward Bounding Let α be some member of the verbal sequence of the right conjunct, and β be the set of elements in the sequence that c-command α. If gapping or stripping includes α then it must include β. (A “verbal sequence” of X is the set of functional heads and auxiliary verbs that is included in X and the main verb of X. This might be the same thing that Grimshaw () calls an “extended projection.”)

Downward Bounding is violated in () because thought is not included in the ellipsis. The Downward Bounding Constraint has inspired one class of analyses of gapping that sometimes goes by the name of the “small conjuncts” account.

For accounts of determiner gapping, see also Lin (, ), Ackema and Szendrői (), Citko (), Arregi and Centeno (), and Centeno (). 14





. A

..................................................................................................................................

.. Small conjuncts account The small conjuncts account gets started with Siegel’s () observation that some cases of gapping can be captured with no ellipsis at all. In these cases, just the highest item in the verbal sequence is missing from the second conjunct.15 ()

Smith can’t make natto or Jones eat it.

Siegel argues that () should be given a representation that, in today’s syntax, would look like ().16 ()

TP TP

DP Smith

T can’t

vP vP

vP vP

make natto

or

vP DP

vP

Jones

eat it

This syntax assumes that the Specifier of TP requires a subject argument to move into it. For the account illustrated in () to work, there must be a constraint on this movement that would prevent the subject of the second conjunct from moving into the Specifier of TP, for otherwise we would expect the ungrammatical ‘Jones can’t Smith make natto or eat it’ to arise as well. The movement of the subject in () appears to violate John Ross’s Coordinate Structure Constraint, which prohibits movement out of a single conjunct. But Lin () argues that () satisfies the Coordinate Structure Constraint, once it is formulated properly. She

15

(i)

A reviewer points out that similar examples can be found in Dutch: Ik heb Stein het boek getoond, maar de foto niet gegeven. I have Stein the book shown but the photo not given ‘I have shown Stein the book but not given the photo.’

See Heycock and Kroch (), Büring and Hartmann (), and Johnson (). 16 Siegel did not have the “derived subjects hypothesis” at her disposal, but conjectured that the right conjunct in () is what is called a small clause in contemporary syntax.

  



follows Ruys (), who argues that the Coordinate Structure Constraint is a condition that holds of binders rather than movement. We can formulate it with (). ()

Coordinate Structure Constraint Let C1 and C2 be two coordinates of the coordination C. If α is outside C and binds a variable in C1, then it must also bind a variable in C2.

If movement of Smith in () leaves a variable bound by Smith, this condition is violated. Lin argues that in contexts like (), the movement of the subject is semantically vacuous, leaving no bound variable in the first conjunct. This can be seen by considering examples like (). () Bob or Mary can’t eat rice and Jeremy eat potatoes. (Lin : , (a)) If the first conjunct of () occurs on its own, it is ambiguous. ()

Bob or Mary can’t eat rice.

The two readings of () depend on the relative scope of the subject and can’t. When Bob or Mary scopes outside of can’t, the meaning can be paraphrased by (a). When it scopes within can’t, the resulting meaning can be paraphrased by (b). ()

a. Bob can’t eat rice or Mary can’t eat rice. b. Neither can Bob eat rice nor can Mary eat rice.

Lin’s consultants report that () only permits a reading in which the first conjunct has the meaning paraphrased by (b). That would follow if the movement of the subject in () is forced to be semantically vacuous because of the Coordinate Structure Constraint in (). The syntax in () seems like a viable analysis of (), then. Siegel () argues on its behalf from the observation that hasn’t and can’t in () scopes over the conjunction. () a. Oh, no! The villain hasn’t ended up with the whole ranch and the hero ended up with only one horse! b. Ward can’t eat caviar and his guest eat dried beans. (Siegel : , ()–()) This is just what the constituency in () predicts. There are, however, apparent counterexamples to this prediction studied by Hulsey (), that involve disjunctions. One of these is (). For the Red Sox to make the playoffs . . . ()

The Sox must beat the Yankees or the Angels lose to the Mariners.

This sentence is ambiguous. It has a reading in which must scopes over the disjunctions in line with the meanings found in (). On this interpretation, () describes a requirement that can be





met in one of two ways: what must happen for the Red Sox to make the playoffs is either that they beat the Yankees or that the Angels lose to the Mariners. This interpretation arises if must scopes over the disjunction. But () can also get an interpretation in which the disjunction scopes over must. It can be synonymous with The Sox must beat the Yankees or the Angels must lose to the Mariners. On this interpretation, () is expressing ignorance about which of two requirements hold; this meaning allows the continuation: but I don’t know which. Hulsey explains this apparent counterexample by making appeal to an analysis of disjunctions that uses an “alternative semantics.” On this analysis of disjunctions (see Alonso-Ovalle  and references therein), the semantic scope of disjunctions can be wider than their syntax alone would allow. Crediting the disjunct-wide reading of () to this analysis of disjunctions explains why we don’t find a similar wide-scope reading for conjunctions, whose scope is constrained more rigidly by the syntax.17 If Hulsey is right, the generalization that emerges is that the modal and negation which appears to be in the first coordinate is in fact outside the coordination. This supports Siegel’s proposal that these sentences have the structure in (). Putting () together with the version of the Coordinate Structure Constraint in () predicts that when the second conjunct contains something that the moved subject of the first conjunct can bind, it will be able to do so. That is correct, as () shows.18 ()

No father should make okonomiyaki or his daughter eat it.

In this context, the Coordinate Structure Constraint is satisfied when () has the representation in () (no father binds a variable in both conjuncts: its trace in the first, and the pronoun his in the second). ()

TP TP

DP1 no father

T

vP vP

should t1

vP vP

or

make okonomiyaki

vP DP

vP

his1 daughter eat it

17

(i)

It also explains (i), where the disjunction cannot include the negated modal. John hasn’t seen Harry or Bill Sue.

(Oirsouw : , (b))

Negation blocks the scope-widening effect of an alternative-based semantics for disjunction. 18 Although, as Jeroen van Craenenbroeck notes, this is inconsistent with the account I offered for the ungrammaticality of ().

  



And, indeed, () does have an interpretation in which his is a variable bound by no father. (That this is possible in gapping contexts is discovered by Oehrle  and McCawley  and part of Lin’s  discussion.) This class of cases, then, could be handled without ellipsis of any kind. The material that is shared by both conjuncts is simply outside the coordination; the appearance of ellipsis is achieved by letting the material that has conjoined be sufficiently small. If this is the only way gaps of this kind can be formed, then we also have an understanding of why they only arise in coordinations and why they obey Downward Bounding. An example like (), which illustrates a violation of Downward Bounding, cannot be formed from a low coordination. () * Smith can’t make natto or everyone knows that Jones eat it. If a way of spreading this treatment to all other gaps can be found, then we could derive these attributes for all gaps. Consider first how the low conjuncts account might be applied to a case, like (), in which the highest term in the verbal sequence is the main verb. ()

Some ate rice and others natto.

One approach to this case involves the hypothesis that main verbs move a short distance in English. Imagine, for instance, that the verbal root combines with the external θ-role assigner, ‘v’, and they jointly move to a higher head position, X0. This would allow a small conjuncts analysis of () like that in (), where the complex made of the verbal root and v have, in each conjunct, moved in across-the-board fashion into X. ()

TP DP1 Some

TP T

XP

past

X

v V eat

vP

X

v

vP t1

vP and

vP

vP DP

VP DP rice

others

vP VP DP natto

With the addition of verb movement and the presence of XP, the cases of gapping that Siegel described would have the slightly different analysis shown in ().





()

TP TP

DP Smith

T

XP

can’t

X

v

make

v

or

vP

X

V

XP

XP

XP DP

vP VP

XP

Jones X DP natto

X

v V

vP

v

eat

vP VP DP it

The coordinates are large enough in () that movement of the main verb does not remove them from the coordination. As before, the subject of the first conjunct moves into the Specifier of TP. But we must now also conjecture that the subject of the right conjunct also moves, for otherwise it would remain lower than the main verb of that conjunct. There is no satisfactory account for this. (But see the discussion of () in section ...) This technique for producing a gap out of the main verb, in addition to an object which has shifted, preserves the account of Downward Bounding, as well as ensuring that gaps only occur in coordinations. We are left with the task of extending this account to more complex gaps, where more than just the highest term in the verbal sequence, plus possibly a shifted object, gaps. There have been two approaches: ellipsis and rampant across-the-board movement.

... Ellipsis Adding ellipsis to a small conjuncts analysis generates many of the cases we’ve seen. In Coppock (), this ellipsis is just garden variety pseudogapping; in Lin (), it is a purpose-built ellipsis process. In either case, the remnant moves out of the phrase to be elided in the way we’ve reviewed in the previous sections. To see how this works, let’s walk through (), which cannot be produced with just low coordination and movement of the main verb and/or object. ()

Some will put books on a table and others magazines.

We’ll start the derivation by Object Shifting the objects in each of the coordinated XPs. (I’ve called the phrase into whose Specifier objects shift: µP.)

   ()



TP DP some

TP XP

T XP

will

XP μP

X

and

XP

μP

DP2 books

μ

XP

DP vP

others

X

μP

vP v

DP1 VP

magazines μ

VP V

μP vP

PP

vP

on a table

t2

v

VP

VP

put V

PP t1

on a table

put

The vPs in this representation are close enough a match for ellipsis to apply, forming (). ()

TP DP some

TP T

XP

will

XP

XP μP

X DP2 books

XP

and μP

μ

DP

XP

others

vP

magazines

VP

μ

PP

VP V

μP

DP1

vP v

μP

X

t2

on a table

put

The main verb will do its movement in the left conjunct, forming ().

vP





()

TP TP

DP some

T will

XP XP μP

X μ

V

X DP2 μ

v v

XP

books

and

XP

μP

DP

vP

others

vP

VP

μP

X DP1

VP

put

XP

magazines μ

μP vP

PP t2 on a table

This is the correct outcome. Because the highest term in the verbal sequence still gaps by way of a low coordination on this view, it preserves the effects we’ve seen credited to low coordinations: its restriction to coordinations and its compliance with Downward Bounding. It’s not that Downward Bounding is a constraint on gapping on this view, but rather that Downward Bounding names the environment where pseudogapping and small conjuncts come together to form the outputs we call gapping. When Downward Bounding isn’t honored, examples such as () arise, where the combination of small conjuncts and pseudogapping is obvious. ()

? Some must [vP eat seafood] and [vP others, think that Mary should Δ bread].

This is the proposal in Coppock ().19 Note that if movement of the verb in the second conjunct could happen, we would expect to get ().20

David Pesetsky was the first I know of to make this proposal, but his suggestion never made it to print: it was part of a seminar he gave at MIT in . 20 To the extent that () is grammatical, it has an interpretation paraphrased by ‘Some put books on a table and others put magazines on it’, not the interpretation that would be derived by gapping: ‘Some put books on a table and others put magazines on a table’. 19

   ()



* Some put books on a table and others put magazines.

In many of the world’s languages, it is possible to elide a VP from which the verb has moved (see Goldberg ) and () is just where we’d expect it to occur in English. This is a problem that comes with the small conjuncts analysis. I don’t know that this problem has been observed or addressed, so it is an open problem. There are other problems as well. The first is, simply, that it predicts that gaps which involve more than one term will not arise in contexts where pseudogapping isn’t licensed. This doesn’t appear to be correct. For instance, neither VP-ellipsis nor pseudogapping can occur after the auxiliary verb have in Icelandic, and yet gapping can occur in a sentence in which have is the highest term in the verbal sequence.21 ()

a. * Sumir hafa borðað natto en aðrir some have eaten natto and others ‘Some have eaten natto and others haven’t.’

hafa have

ekki. not

b. * Sumir hafa borðað natto en aðrir some have eaten natto and others ‘Some have eaten natto and others have rice.’

hafa have

hrísgrón. rice

c. Sumir hafa borðað natto en aðrir hrísgrón. some have eaten natto and others rice ‘Some have eaten natto and others rice.’ Similarly, whereas VP-ellipsis and pseudogapping can occur in English to VPs and APs when they are embedded under modals, finite auxiliaries, and other similar terms (see ()), they cannot when VPs or APs are complements to lexical verbs (compare ()). ()

a. b. c. d.

Sally must eat tomatoes whenever I see that Holly does eat tomatoes. Sally must eat tomatoes whenever I see that Holly does eat apples. Sally is fond of the tomatoes because she knows that Holly is fond of the tomatoes. ? Sally is fond of the tomatoes even though she knows that Holly is fond of the apples.

()

a. b. c. d.

* Sally must eat tomatoes whenever I see Holly eat tomatoes. * Sally must eat tomatoes whenever I see Holly eat apples. * Sally is fond of the tomatoes because she considers Holly fond of the tomatoes. * Sally is fond of the tomatoes even though she considers Holly fond of the apples.

Gapping, on the other hand, can apply to verbs and adjectives that are embedded as complements to lexical verbs, as () illustrates.

21

My thanks to Johannes Gisli Johansson for the examples.





()

a. b. c. d.

I saw some eat tomatoes on Tuesday and others eat apples. I saw some eat them on Tuesday and others eat them on Wednesday. ? I consider some fond of the tomatoes and others fond of the apples. ? I consider some especially fond of the tomatoes and others especially fond of the apples

If pseudogapping is the source of a gap that includes more than just the highest term in the verbal sequence, this isn’t what we should find. Another problem for the pseudogapping-based account is one that Siegel () notes. We have seen that when the highest term in the verbal sequence is all that gaps, the resulting semantic interpretation puts that term outside the scope of the coordination. The example in () has just the interpretation in which can’t scopes outside the disjunction. () Ward can’t eat caviar and his guest eat dried beans.  It’s not allowed for Ward to eat caviar while his guest eats dried beans. (Siegel : , ()) But when the highest term in the verbal sequence gaps along with the main verb, as in (), the sentence can have a reading in which that term doesn’t scope out of the disjunction. () Ward can’t eat caviar and his guest dried beans. (Siegel : , ()) Like (), () has a reading in which can’t scopes over the conjunction; it can be synonymous with (). But it also has an interpretation in which can’t is in each conjunct; a reading synonymous with Ward can’t eat caviar and his guest can’t eat dried beans. Neijt concludes from this contrast that gapping is something different than a mere low coordination. If she is right, () is low coordination and () isn’t—it is an independent process instead: gapping. A third problem is that the antecedence conditions on pseudogapping and gapping don’t seem to be the same. The antecedence conditions on gapping are, as Hernández () points out, stricter than those for pseudogapping. Like VP-ellipsis, pseudogapping is able to have a deverbal noun as its antecedent, as in (). () Sal may be a talented forger of passports, but surely he can’t Δ paintings. (Hernández : , ()) See Fu et al. () for an analysis. Gapping, by contrast, cannot; as () demonstrates.22 () * Sal saw a forger of passports and Holly forge paintings. (based on Hernández : , ()) 22 Hernandez’s example illustrating this point is *Sal is a forger of passports and Holly paintings. This example, however, also violates Upward Bounding, which we will encounter next. Indeed, my own example might too, depending on the exact parse of deverbal nominals and what derives Upward Bounding.

  



And a final problem is that there is a condition like Downward Bounding that holds of the antecedent for the gap in the left conjunct. That condition requires that the antecedent for the gap not be embedded. It is responsible for the ungrammaticality of (). ()

* Either [Smith will require that I bring natto], or [Jones rice].

() is only grammatical if the disjunction is interpreted as the complement to require; it cannot have an interpretation in which the disjunction is the one indicated by the brackets. Pseudogapping is possible in this context, as we can see from (). ()

Either [Smith will require that I bring natto], or [Jones will rice].

If gapping is just a low coordination in tandem with pseudogapping, then () should be able to get the representation in (). (I’ve suppressed here the irrelevant XP, µP, and the movements they incur.) ()

TP TP

DP Smith

T will

vP vP

vP or

vP v

vP DP

VP V

CP

require C that

Jones TP

DP I

vP

v

vP

bring DP1

vP v

VP V

TP T

VP V bring

DP2

vP

natto t1

rice t2





An ellipsis-based account of gapping doesn’t explain what Hankamer (), who first discovered this constraint, called the Upward Bounding Constraint: ()

Upward Bounding Constraint The antecedent for a gap must include the highest term in the verbal sequence of the first conjunct.

... Across-the-board movement In an attempt to solve these problems with the ellipsis-based account, Johnson () proposes that gapping only involves across-the-board movement. Following proposals in Kayne (), Johnson suggests that the syntax of English can involve short movement of a verbal projection. A sentence like (), for instance, can have a representation like (). ()

She must read the book to Smith quickly.

()

TP DP

TP

she T

XP

must

XP

vP v

PP

VP V read

X

VP

μP μ

vP

DP to Smith the book

vP quickly

We can imagine that this is the phrasal version of movement of the single verb into X. Here, I’ll assume that English allows either head movement of the V to X, or phrasal movement of vP to XP, to bring the verb into its surface position.23

I’ve remained cryptic about the identity of X, but one could adopt the view in Zwart’s (, ) work on Dutch, or the similar ideas in Bowers (), that it is ‘Pred’. 23

  



If this movement is available, then it can happen in across-the-board fashion in contexts where there are low coordinations, and this will produce gaps of verbal projections, not just verbs. For instance, the gap in () can be derived by across-the-board moving give a book as shown in (). ()

Some gave a book to Jones and others to Smith.

()

Some [ gave a book]1 [XP [t1 to Jones] and [others [t1 to Smith]]]

The complete set of relevant cases aren’t given in Johnson , however, and it is not transparent how the right word orders are always achieved—a problem that has been pointedly raised by Vicente (a) and Boone (). For instance, to form the gap in () requires gapping give and to me. ()

Some gave books to me and others magazines.

That can be done in one of two ways on Johnson’s account. Either the verb and the PP independently move across-the-board, as in (), or they move across-the-board as a phrase, out of which the contrasting objects have moved, as in the derivation in (). ()

Some gave1 [to me]2 [XP [[t1 books] t2 ] and [others [[t1 magazines] t2 ]]]

()

a. Some [ xp [[[gave t ] to me ] books] and [others [[gave t ] to me ] magazines]]. b. Some [vp gave t to me]1 [xp [t1 books] and [others [t1] magazines]].

In neither case is the word order in () achieved, and it is not transparent how to derive that word order. This is the same problem we saw in () for stripping, where the movement operations necessary to form the string that elides in the second conjunct cannot be mirrored by overt movement operations in the first conjunct’s antecedent to the ellipsis. On an ellipsis-based account, this can be solved by letting the movement operations in the first conjunct happen covertly. But that’s not a possible solution in the across-the-board account, since it is the overt form of the elided phrase that is spoken in the left conjunct on that account. The across-the-board account does, however, provide a way to capture the fact that a modal can be interpreted inside the coordination when gapping involves both the modal and the main verb. In that scenario, the phrase that moves across-the-board could be made to include the modal. If the standard view that modals surface in T is correct, then the easiest way of getting this to work would be to assume that modals have a lower underlying position from which they move. If that lower position is inside the phrase that moves across-the-board, then representations like () could produce the desired interpretation.





()

TP TP

DP some T

XP XP

MP vP

M can’t v

XP VP

VP V read

X PP

XP

μP μ

and

books

can’t

DP

MP

DP to Smith M

XP

others X

vP v

XP

VP

DP

μP μP

read booksF magazines μ MP to Smith M vP can’t v

VP

read magazinesF to Smith The surface form could be produced from () by moving can’t from the phrase it heads (i.e., MP) into T. If movement of MP and movement of can’t into T are semantically vacuous, this syntax will achieve the desired reading, one in which can’t is interpreted in both conjuncts.24 The across-the-board account also derives Upward Bounding. Because the movement of vP to XP is by definition clause-bound, it will not be able to produce violations of Upward Bounding. It also doesn’t make the erroneous prediction that gapping of this type will only be possible in languages, and syntactic contexts, that have pseudogapping. Instead, it makes the more difficult to confirm prediction that it only happens in languages that have predicate movement of the sort that () illustrates. Both executions of the small conjuncts analysis harbor problems that have not yet been overcome. The across-the-board based analysis has its most serious difficulties in explaining the word orders that arise, and the pseudogapping-based analysis has its greatest difficulties in explaining why pseudogapping qua gapping becomes available in 24 A problem with letting a modal move from an underlying position lower than T is that it allows for derivations in which it moves across-the-board out of conjoined MPs. This would be another way of achieving a gap of the modal alone, and this could wrongly allow it to be interpreted in both conjuncts of a gapping construction. Letting modals move threatens to unravel Siegel’s explanation for the fact that a modal has just wide scope in these cases. In addition, as Jeroen van Craenenbroeck notes, these derivations violate the Freezing Principle.

  



environments where it is not otherwise available. In addition to these problems there are various more particular problems of execution which we will review in section ... But let’s highlight first the reasons for maintaining a small conjuncts analysis. It explains why gapping is found in coordinations, and not other sentential connectives. It also explains Downward Bounding, as we’ve seen, and if the across-the-board account is adopted, Upward Bounding as well. And, finally, it explains why the movement operations that seem to be responsible for forming the strings that can gap are just Heavy NP Shift, Object Shift, and the putative short focus movement, normally covert in English. If gapping can be fed by just these movement operations, and not by movement operations that can span longer distances, then we will explain why gapping and stripping cannot remove part of an embedded finite clause. How does the small conjuncts analysis restrict the class of movement operations that can form gaps and strips? The small conjuncts analysis traps these operations into structures where T remains outside coordinated phrases that contain a vP, as in (). ()

TP T

XP XP

X

XP vP

and

XP X

vP

If the constituent that gaps or strips is vP, then the only material that can remain behind will have to reside somewhere in XP but outside vP. The only established movement operations in English that target this region are Heavy NP Shift, Object Shift, and presumably Focus Movement. The small conjuncts analysis therefore restricts the class of movement rules available to form gapped phrases by forcing those movement rules to be ones whose landing site is within the small conjunct. This predicts that the kinds of gaps a language has should vary with the inventory of short movement rules it possesses. Languages with short scrambling, for instance, should be able to use this rule to form a gapped constituent. We should also see differences crosslinguistically in what a gap can include that will track what material can be put between vP and XP. This might be the reason that German, as Repp (a, ) shows, can have gaps in which the sentential negator is interpreted in just the first conjunct. ()

Carl hat meine Katze nicht genommen, aber Harry meinen Hamster. Carl has my cat not taken, but Harry my hamster. ‘Carl didn’t take my cat but Harry took my hamster.’ (Repp : , (.))

This could reflect the fact that sentential negators in German are low adverbs and can therefore be positioned within the first conjunct while still allowing the vP of that conjunct





to serve as the antecedent for the gap. See Repp () for a different approach, and for a discussion of the factors that influence the availability of this reading.25

.. Large conjuncts Let’s now consider an account that allows gapping and stripping to occur when two full clauses are coordinated: a “large conjuncts” account. An across-the-board account is incompatible with this possibility, so the gap or strip will have to be created by ellipsis. Because Heavy NP Shift and Object Shift do not move their phrases far enough out of the VP to escape an elided clause, it relies on Focus Movement that targets the left edge of the clause. A stripping example like (), repeated here, will have the structure in (). ()

Smith spoke to Jones today and Brown too.

()

TP TP DP Smith

TP TP

T

and

FocP DP1

VP V

PP

spoke

to Jones

FocP

Brown Foc

TP DP

TP

Smith T

VP V

PP

spoke

P

DP

to

t1

Ellipsis silences the boxed TP. A gapping example like () will get a representation like ().

25

(i)

Repp suggests that a similar interpretation is available in the English (i). Pete wasn’t called by Vanessa but John by Jessie.

(Repp : , (.))

I and most of my consultants find this sentence ungrammatical. It can be marginally improved by adding rather after but. Consultants who find this sentence grammatical do report that it has the meaning ascribed to it by Repp. I haven’t found an example with and or or that has a similar effect. See Repp (b) for some discussion.

   ()



Some ate beans and others rice.

()

TP TP TP

DP some

TP

T

FocP

and VP

DP1

V

DP

ate

beans

FocP

others DP2 rice

FocP Foc

TP t1

TP T

VP V

t2

ate Note that both remnants of the gap are in a contrastive focus relationship to the parallel arguments in the first conjunct, and this is captured by moving both of them to the Specifier of Focus Phrase. Because there can be an indefinite number of remnants to a gap, we will have to allow there to be an indefinite number of focus positions to be moved into.26 These are the analyses advocated for by Depiante () and Boone (). Something must ensure that the remnants of the gap are ordered in a particular way. It isn’t possible to move the remnants out of the gapped TP in () so that the object precedes the subject: Some ate beans and rice others is ungrammatical. A parallel requirement is needed for the small conjuncts account as well, since the remnants of the gapped vP on that account must be similarly ordered (witness the contrast in ()). ()

a. Some brought pickles for Sam and others olives for Mary. b. * Some brought pickles for Sam and others for Mary olives.

The generalization seems to be that the linear order of the gap’s remnants must be the same as the linear order of the phrases they are contrasted with in the antecedent. That is what paradigms like () imply.27

26

Jackendoff () suggests that gapping can have no more than two remnants, but I believe the facts suggest a steady decline in grammaticality as the number of remnants increase, with no sharp cut-off point. 27 Hudson (b) discusses gapping in inversion structures like (a).





()

a. Into the study walked Mary and into the closet, Sam. b. * Into the study walked Mary and Sam, into the closet. c. Mary walked into the study and Sam, into the closet. d. * Mary walked into the study and into the closet, Sam.

As far as I know, this effect is unexplained. Allowing for a large conjuncts analysis gives an immediate remedy for several of the afflictions of the small conjuncts account. It permits an explanation, for instance, of Siegel’s paradigm concerning the scope of modals. Recall that when only the modal gaps, as in (b), it necessarily scopes outside the conjunction. But when both modal and verb gap, as in (), it can scope either outside or inside the conjunction. (b) ()

Ward can’t eat caviar and his guest eat dried beans. Ward can’t eat caviar and his guest dried beans.

If both a large conjuncts and small conjuncts syntax for gapping are available, then the ambiguity of () emerges. The wide-scope reading for can’t arises when it is outside the conjunction and the small conjuncts syntax is used. The narrow-scope reading for can’t arises when his guest and dried beans have focus-moved out of a TP and the large conjuncts syntax is used. The unambiguity of (b) can be ensured if the large conjuncts syntax can be blocked in this configuration. That might be achieved by disallowing focus movement of VPs, which is what would be required if (b) were to get a large conjuncts analysis. Allowing gapping and stripping to come by way of large conjuncts also removes the problem that gapping is possible in places where pseudogapping isn’t. Recall that there are languages which have gapping and stripping, but not pseudogapping (e.g., Icelandic) and there are contexts in English where gapping is possible but pseudogapping isn’t (e.g., complements to causative or perception verbs). If large conjuncts, and an accompanying TP ellipsis, can produce gaps, then this could be the source of the gaps in these situations. Of course, this requires that there be independent justification for the TP ellipsis in these languages and environments that the large conjuncts account requires. A candidate for this independent evidence is fragment answers, which we will look at in section .. This is the ellipsis needed to produce fragment answers, of the kind that () illustrates. ()

A: Jones ate seafood. B: Bread too.

Merchant (a) makes a convincing case that the question–answer versions of these conversations derive from an ellipsis which removes everything but Bread from Bread, Jones ate. (See Weir ,  for an extension of Merchant’s analysis to cases like these.) It’s conceivable, then, that the gaps and strips could be achieved with this ellipsis process applying to clause-sized coordinates. Boone () is an extended argument for this approach. If languages like Icelandic only have gapping in clause-sized conjuncts, we expect that gaps of the form in () will unambiguously have a narrow-scope reading for can’t in

  



languages like Icelandic.28 As noted above, Repp (a) reports that German, which also fails to have pseudogapping, has facts inconsistent with the small conjuncts analysis when negation and gapping are combined. Finally, allowing a large conjuncts analysis alongside a small conjuncts analysis gives an account for the fact that the subject in a clause with gapping can be either accusative or nominative in English. ()

a. She likes rice, and he beans. b. She likes rice, and him beans.

On a large conjuncts account, the subject of the clause with gapping will have moved into the nominative case-marked position and, from there, into its focus position. On a small conjuncts account, the conjunct with the gap will be too small to contain Specifier of TP, where nominatives appear in English, and we can expect the default accusative case form instead. When this is put together with the account for Siegel’s contrast between (b) and () sketched above, we make the prediction that (b) should only allow the accusative form, as this example only has a small conjuncts syntax. I don’t believe that prediction is confirmed, as there is only the slightest difference in grammaticality between the nominative and accusative examples of (). ()

a. She can’t eat caviar and him eat dried beans. b. ? She can’t eat caviar and he eat dried beans.

It also predicts that using a nominative or accusative subject should disambiguate (). A nominative subject in the second conjunct should require a large conjuncts analysis, forcing can’t to be within each conjunct, and an accusative subject in the second conjunct should require a small conjuncts analysis, forcing can’t to scope outside the conjunction. I haven’t examined this prediction with enough thoroughness, but it is in only partial accord with my judgments about the cases in (). ()

a. She can’t eat caviar and he dried beans.  She can’t eat caviar and he can’t eat dried beans., or  It’s not allowed for her to eat caviar and him to eat dried beans. b. She can’t eat caviar and him dried beans.  It’s not allowed for her to eat caviar and him to eat dried beans.

As expected, the accusative subject in (b), which is only compatible with a small conjuncts syntax, forces can’t to scope outside the conjunction. But the nominative subject in (a) allows for both readings, suggesting that it is compatible with either a small conjuncts or large conjuncts syntax. The grammaticality of (b) and the wide-scope reading for can’t in (a) would both follow if English subjects can be nominative when they remain in their underlying position. It isn’t trivial to discover whether this is the case 28

Similarly, it predicts that examples like No father should make okonomiyaki or his daughter eat it won’t allow the bound variable reading for his in Icelandic and similar languages.





since the scenarios where subjects in their low position can be pronouns—the only terms that show Case morphology—are rare. These are some of the advantages to letting gapping and stripping occur with clausesized conjuncts. There are also certain kinds of examples that are admitted by widening the set of coordinations that host gapping to clause-sized ones. We should, for instance, expect gapping to be possible in coordinated CPs. This does not generally seem to be the case, however. Gapping is blocked in coordinated complement CPs (= (a) and (b)), adjunct CPs (= (c) and (d)), and relative clause CPs (= (e)).29 ()

a. * Smith said that some like beans and that others rice. compare: Smith said that some like beans and others rice. b. * Smith wanted for some to bring beans and for others rice. compare: Smith wanted for some to bring beans and others rice. c. * If some bring beans and if others rice, we’ll have everything we need. compare: If some bring beans and others rice, we’ll have everything we need. d. * Smith stayed because some brought beans and because others rice. compare: Smith stayed because some brought beans and others rice. e. * This is the woman who some gave books and who others magazines. compare: This is the woman who some gave books and others magazines.

But gapping does seem possible in coordinated CPs if they are interrogatives:30 ()

a. Smith asked which guest had brought rice and which beans. b. Which guest has brought rice and which beans?

Sluicing is a clausal ellipsis that is found only in questions, and so it might be the source of examples like (). If sluicing is responsible for these constructions, and not gapping, then there are no examples of coordinated CPs that support gapping. That is an unsolved problem for the view that gapping can occur in clause-sized conjuncts. Allowing gapping and stripping to occur in clause-sized conjuncts brings real benefits. But we lose some of the features that drive a small conjuncts account. Because a small conjuncts account claims that the highest head in the verbal sequence isn’t elided, but is instead simply outside the coordination, it derives the fact that gapping and stripping are found only in coordinations and it derives Downward Bounding. The across-the-board variant of the small conjuncts account also derives Upward Bounding. If gapping and stripping aren’t restricted to just small conjuncts, then there is no explanation for these three properties.

29

The fact that a complementizer in coordinated CPs blocks gapping is reported in Hartmann (). In all the examples of gapping we’ve seen so far, the remnants are in a contrastive focus relation to parallel phrases in the antecedent. If that holds for (), then the interrogative determiners, which, in the second conjunct should be contrastively related to something in the first conjunct. See Romero () for an account. For a discussion of gapping in question contexts, and an exploration of some of their constraints, see Pesetsky (). 30

  



. F 

.................................................................................................................................. While the small conjuncts analysis predicts that gapping and stripping show up in coordinations but not adjuncts, it doesn’t explain why fragment answers and similar discourses, as well as comparatives, should have superficially similar-looking ellipses. If these cases are to be brought into the analysis, then a force that uniquely singles out these environments should be found. The only attempt I have found in the literature to do this that doesn’t use the small conjuncts analysis points to the antecedent conditions on ellipsis instead. Perhaps the most salient of these accounts is in Andrew Kehler’s work, which I will briefly sketch here. Kehler argues that the antecedence conditions on ellipsis track relations that hold discourses together, and that they are responsible for segregating one kind of ellipsis from another. In Kehler () (and see also Kehler ), he adopts a typology of discourse relations from Hobbs () and applies them to the antecedence conditions on ellipsis. He singles out two relations in particular: “cause–effect” and “parallel.” The “cause–effect” relation is invoked to connect the two sentences in (a), and the parallel relation is invoked to connect the sentences in (b). ()

a. Bill was about to be impeached. He called his lawyer. b. Bill likes to play golf. Al enjoys surfing the net. (Kehler : (a), (a))

Kehler () suggests that the antecedence conditions on VP-ellipsis are tied to either one of these relations. I’ll frame that hypothesis informally with (). () Condition on VPE For VPa to be the antecedent of elided VPe, VPe must be embedded in a clause that is either parallel to, or the result of, a clause VPa is embedded in. Cases which satisfy () are in (). ()

a. Bill drank milk because Sam did Δ. b. Bill drank milk and Sam did Δ too.

Gapping, by contrast, is subject to a narrower condition, one that forces the gapped material and its antecedent to be in a parallel relation. I’ll frame this condition with (). ()

Condition on Gapping For XPa to be the antecedent of a gapped XPe, XPe must be embedded in a clause that is parallel to a clause that XPa is embedded in.

This captures the contrast in (). ()

a. * Bill drank milk because Sam gin. b. Bill drank milk and Sam gin.





If the notion “parallel” can be formalized correctly, it might manage to restrict the range of connectives that gapping can exist in to just those involving and and or. This approach to the question of what derives the environments where gapping can occur is wholly semantic, and Kehler argues that is supported by the fact that gapping has an effect on the meanings of the coordinations that host it. To see this, consider the range of meanings that () can have. ()

a. Sue became upset and Nan became downright angry. b. Al cleaned up the bathroom and Joe cleaned up the mess. c. One of the students was accepted at Bryn Mawr and the high school was praised on TV. (Kehler : , ())

In each of these cases, the conjunction can have a strictly Boolean interpretation: the sentences report merely that both propositions are true. But it is also possible to understand these sentences as communicating more: that there is a temporal or causal connection between the propositions each conjunct conveys. In (c), for instance, the second conjunct could be construed as the result of the first. Kehler reports that Levin and Prince () discovered that this second reading disappears when gapping applies. The sentences in () have only the strictly Boolean interpretation. ()

a. Sue became upset and Nan downright angry. b. Al cleaned up the bathroom and Joe the mess. c. One of the students was accepted at Bryn Mawr and the high school praised on TV. (Kehler : , ())

This follows from (), if the parallel relation can be defined so that it precludes all but the Boolean construal. For Kehler’s system to explain why gapping is restricted to coordinations an adequate definition of the parallel relation is required, but an explanation for why gapping is only capable of using this relation is needed too. Kehler suggests that the cause–result relation—and all the other similar discourse relations—cannot be used when syntactic reconstruction of the anaphor is at play and that this is why only the parallel relation can be employed in gapping. He assumes, along the lines described above, that gapping necessarily involves ellipsis of a constituent from which the remnants have moved and that the only method of achieving this is for the gap to have syntactic structure that is reconstructed from the antecedent. VP-ellipsis, by contrast, he argues can arise without syntactic reconstruction, and is therefore able to make use of more discourse relations in finding its antecedent. Note, however, that if pseudogapping is similarly an instance of ellipsis of a constituent from which the remnant has moved, this would wrongly predict that pseudogapping is restricted to coordinations in the same way that gapping is. ()

Smith ate natto because Bill had rice.

  



An explanation for the putative connection between the restriction to parallel discourse relations and gapping remains wanting. A definition of “parallel” that might do the job needed comes from Craige Roberts’ notion of “Question Under Discussion.” Roberts () sketches a model of discourse cohesion that relates each sentence of a discourse to an implicit goal that the contributors to that discourse share. Those goals, she suggests, can be modeled as the answer to an implicit question: the “Question Under Discussion.” Reich () suggests that the antecedents to gapping constructions come from these implicit Questions Under Discussion. Perhaps we can take Kehler’s “parallel” relation to be the sentences that are related to the Question Under Discussion in the way that Reich suggests. See Weir () for an elaboration of Reich’s idea that solves some of its problems, like those in AnderBois (). Consider how this model will apply to the environments we have seen gapping licensed in. When the Question Under Discussion is overt, we have the special case of gapping being licensed in answer contexts: ()

A: Who ate what? B: Jones, seafood.

A declarative sentence can sometimes also be used to invoke a Question Under Discussion, thereby also providing the antecedent to a gap. Thus, for instance, (a) could invoke the Question Under Discussion in (b), licensing the gap in (c). ()

a. A: Someone ate something. b. QUD: Who ate what? c. B: Yes, Jones, seafood.

And (a) could invoke the Question Under Discussion in (b), thereby licensing the strip in (c). ()

a. A: Jones ate seafood. b. QUD: What did Jones eat? c. B: Bread, too.

Coordinations could be understood as having a first conjunct which invokes a Question Under Discussion that, in turn, licenses gapping in the second conjunct. To formalize Kehler’s parallel relation with the Question Under Discussion model, and to use it to explain gapping’s restriction to coordinations, requires that only root assertions be able to contribute to the Question Under Discussion.31 Imagine that coherent discourses consist of assertions offered by participants which constitute partial answers to the Question Under Discussion. Under normal circumstances, these assertions are the root, independent, sentences uttered by the participants. A rough description of the condition on gapping could then be ().

31

Jeroen van Craenenbroeck points out that this does not lead to the expectation that gapping can be inside embedded contexts, as in ‘John regrets that no one asked whether Sally likes beans or Bill rutabagas’.





()

QUD condition on gapping and stripping XP can gap or strip if it is proffered as a partial answer to the Question Under Discussion, Q, and XP’s meaning can be recovered from Q.

Because coordinated clauses in a strictly Boolean coordination act like root assertions, they could each be understood as independently providing a partial answer to the Question Under Discussion. The first conjunct, then, would offer information about what the Question Under Discussion is by providing a partial answer to it, and the second conjunct provides another partial answer to that Question Under Discussion and is thereby able to house a gap. This is roughly the direction that Reich () pursues. () also comes close to deriving Upward and Downward Bounding. It requires that the clause with the gap and the clause containing the antecedent are parallel and this, as Toosarvandani () has argued, could capture many of the cases Downward and Upward Bounding are designed for. If the typical means of making a contribution to a discourse is with the meaning of a root clause, then only root clauses will be able to be gapped. But, as noted, conjuncts of a coordination are straightforward exceptions to that, as their semantics allows them to be understood as independent contributions to the discourse when they get a strictly Boolean interpretation.32 This will effectively block violations of Downward Bounding like those in (). ()

* Smith ate sushi and everyone thought that Jones rice.

The first conjunct can license the Question Under Discussion who ate what?, but the second conjunct can only obliquely be construed as a partial answer to that question. In a parallel way, () can explain why violations of Upward Bounding like () are blocked. ()

* Either [someone thought that Smith ate sushi] or [Jones rice].

A Question Under Discussion that the first disjunct licenses is who thought that who ate what?, to which the second disjunct cannot be construed as a partial answer. There are no other Questions Under Discussion that the first coordinates of () or () can license that the second coordinates can be partial answers to, and () will therefore correctly block them. An advantage of deriving Downward Bounding in this way is that it offers a method for relaxing it, and there are reports of cases where it does seem to be relaxed. Weir (), for instance, reports that the clausal complements to certain verbs allow relaxation of Downward Bounding when the complementizer is unexpressed. ()

a. ? John ate oysters and I think Mary swordfish. b. ? John ate oysters and I believe Mary swordfish. c. John ate oysters and I suspect Mary swordfish. d. John ate oysters and I imagine Mary swordfish. (Weir : , ())

32 Disjunctions work similarly, though in a weaker way. Each disjunct can be linked to a Question Under Disjunction in the way necessary to license gapping, but the disjuncts are not offered as independently partial answers of the QUD, but instead potential partial answers.

  



And Wurmbrand () reports a similar effect for stripping: ()

Jane loves to study rocks, and John says geography too. (Wurmbrand : (d))

Weir and Wurmbrand propose that the obligatory absence of the complementizer is connected to the purely syntactic conditions that determine which constituents can be elided. As we’ve seen, the presence of a complementizer blocks gapping in cases where embedded CPs are conjoined, and this could be connected to these cases. () and () show that when this syntactic constraint is satisfied, Downward Bounding is sometimes relaxed. Similarly, Farudi () shows that certain cases of gapping in Farsi also allow violations of Downward Bounding. () Maamaa chaai xord va fekr mi=kon-am mother tea ate.sing and think IMP-do-sing ‘Mother drank tea and I think Father drank coffee.’

baabaa father

qahve. coffee

(Farudi : , (a))

The relaxation of Downward Bounding here could be credited to the ability of the embedded clauses in these examples to constitute a partial answer to the Question Under Discussion. Perhaps with these verbs, the propositions they embed can be understood as proffered by the speaker, thereby satisfying (). When the embedding verb presupposes the truth of the proposition they embed, however, it would not allow those propositions to be part of what the speaker is asserting. This predicts correctly that Downward Bounding holds in the factives of (). ()

a. * John ate oysters and I remember Mary swordfish. b. * John ate oysters and I deny Mary swordfish. c. * John ate oysters and I am surprised Mary swordfish. (Weir : , (b, c))

See Weir  for a worked-out account along these lines. Kush () argues that deriving Downward and Upward Bounding from something other than the small conjuncts analysis is warranted from a cross-linguistic perspective as well. He shows that Hindi-Urdu obeys Downward and Upward Bounding, and yet doesn’t have many of the other properties predicted by the small conjuncts analysis. For instance, Hindi-Urdu doesn’t permit material that appears to reside in the first conjunct to scope over the second conjunct, as we’ve seen is possible in English gapping constructions. The subject of the first conjunct in (), for instance, cannot bind the pronoun in the second conjunct. ()

* Koi bhii larka1 doctor ke-paas nahiin gayaa, na uskii1 maa some EMPH boy doctor near neg go-PFV.M nor his mom dentist ke-paas. dentist near ‘No boy went to the doctor or his mom to the dentist.’ (Kush : ())





He suggests that languages vary with respect to whether gapping elides a full clause or a smaller VP-like constituent, but that this variation doesn’t track whatever it is that derives Downward and Upward Bounding. An approach that credits the antecedence conditions on gapping to something like (), and derives Upward and Downward Bounding from it, would allow for such a typology.33 For () to successfully constrain gapping and stripping would require a formalization of its key parts, including how it is that a gap or strip recovers the elided material from the Question Under Discussion. As we’ve just seen, it offers a method for constraining gapping and stripping to just coordinations and certain cross-discourse constructions, and it also offers a way of deriving Upward and Downward Bounding that relaxes its effects in ways that might be required. Interestingly, though, it doesn’t block simultaneous violations of Upward and Downward Bounding, like those in (). ()

* Smith decided that Sally likes natto and Jones decided John likes okonomiyaki.

A Question Under Discussion that the first conjunct of () can invoke is (). ()

Who decided who likes what?

The second conjunct of () is a partial answer to this question and should, therefore, satisfy (). Moreover, though an account based on () would do a better job of allowing gapping and stripping in constructions other than coordinations—like question–answer exchanges—it wouldn’t seem to permit gapping and stripping in comparatives, which () seems to indicate is necessary. Nonetheless, a more careful investigation of the discourse and information structure of stripping and gapping seems likely to reduce the number of open mysteries these interesting constructions harbor. I recommend this direction to those embarking on a gapping and stripping career.

33

Erschler () also reports a number of languages in which violations of Downward Bounding are found. However, in the ellipses he is studying many of the other attributes of gapping are not found— they are found in adjunct clauses, Upward Bounding is violated, and backwards gapping also seems possible. I tentatively conclude that his cases have a different source than gapping does.

  ......................................................................................................................

 ......................................................................................................................

 

. I

.................................................................................................................................. F are utterances that appear to be smaller than a sentence, including short answers to questions, such as (B), and various kinds of utterances occurring without an overt antecedent, such as the apparently isolated DPs and PPs in ()–(): () A: What did you buy?

B: A new coat.

()

A coffee, please.

()

[Uttered by a waiter displaying a bottle of wine to customers] From Italy.

()

[Text on sign] No parking.

Such fragments are frequently used to express a fully propositional meaning: as has been argued by Stainton (e.g. b), this is strongly suggested by the fact that utterances of fragments can be used to make assertions, are evaluable as true or false, and do not merely communicate implicatures. The debate about fragments used in this way is whether what you see is basically all the linguistic structure you get—that is, whether they are basegenerated words or subsentential clauses—or whether they have the syntax of full sentences, part of which is subject to ellipsis.1 Sections . and . present the non-sententialist and sententialist approaches to fragments. Sections . and . present the main evidence used in the literature to support each approach, and discuss recent attempts to meet the challenges posed by these arguments. Arguments for the latter approach, which assumes silent sentential syntax, are based largely on a set of connectivity effects, particularly case marking; arguments for the former include ‘anti-connectivity’ effects, and differences between fragments and their overtly sentential counterparts. Section . concludes. 1 I set aside, then, those uses of fragments that are generally agreed not to express propositions, including labels or titles (Jane Eyre), and items on lists (e.g. a shopping list).





. N- 

.................................................................................................................................. Stainton (b) concentrates on utterances of phrases such as NPs, DPs, APs, and PPs occurring without an overt antecedent, and argues that appearances reflect reality and they are syntactically non-sentential: for instance, example (), ‘From Italy’, as uttered while holding up a bottle of wine, has the structure of just the overt prepositional phrase. This is then integrated in the central conceptual system with information from other perception modules, inference, and memory. This integration, proposes Stainton (b), occurs through function–argument application: the speaker utters either (i) a word or phrase whose content is a propositional function, and context provides the argument to this function, or (ii) a word or phrase whose content is an argument to some propositional function, and context provides the function. Applying the propositional function to the argument results in the proposition expressed. The first of these possibilities is illustrated by (), ‘From Italy’. The central system will receive from the visual module a concept α of the bottle of wine being pointed at (i.e. the argument); the input from the language faculty is a property concept (the propositional function), and the two inputs are concatenated in language of thought/mentalese to give the proposition α    (Stainton b: –).2 The second possibility is illustrated by Stainton’s example given in (): ()

[Uttered by someone trying to pick a robber out of a police line-up] The third man from the right.

Given the context, the propositional function in (a) would be salient, and the determiner phrase uttered in () would provide the argument to this function, giving the proposition (b) (where β is a concept of the speaker of ()): () a.   β b.        β Stainton remains neutral on the question of whether fragment answers contain silent syntax, though one of his arguments—that antecedentless fragments are non-sentential appeals to the leeway for different interpretations that results from their lack of overt antecedent— suggests that he might accept a sententialist account of those varieties of ellipsis where a determinate content is recoverable from an overt antecedent. Progovac et al. (c) apply a non-sententialist analysis much more broadly than the fragments Stainton focuses on, to cover not just fragment answers, but also ‘small-clause’ structures, such as ():3 ()

2

Him worry?!

Small caps indicate mental representations (as opposed to natural-language representations). Structures containing both an overt subject and overt predicate, such as (7), will not be discussed here, as the main issue about fragments is how a fully propositional meaning is recovered from structures that look like they only contain DP/NP/PP, etc. 3





The core proposal, presented in Progovac () and building on earlier work by Barton (e.g. ), is that non-sententials are generated without merger of (at least) TP, and then, as on Stainton’s account of antecedentless fragments, are combined with salient language of thought material. Among the predictions of this analysis is the appearance of non-sententials with tenseless/non-finite verb forms, as in (), and, in the kinds of fragments that are the focus of this chapter, default case on subject pronouns when these occur as apparently unembedded fragments, as in (). This latter prediction relies on Longobardi’s () argument that a DP layer on top of NP is required in an argument position only when needed to check structural nominative case in a TP (Progovac ). The idea, also argued for by Cardinaletti () and Progovac (), is that pronouns may be generated in the N position, and moved to D in order to check Case features. Where TP is absent, structural nominative case is not required, so subject pronouns would not move to D to check structural case, and will surface in default case (for example, accusative in English and French; nominative in German and Spanish). This feature systematically appears in fragments: () A: Who wants candy? B: Me! Case plays an important role in arguments about whether sententialism or nonsententialism is correct, and I return to it in section .. In the next section, I present the influential sententialist analysis of Merchant (a).

. T  

.................................................................................................................................. On Merchant’s (a) account, fragments are generated by movement of the fragment into a left-peripheral position Functional Projection, FP, followed by ellipsis of the clause that the fragment has moved out of. B’s answer in () would have the structure in (): () A: Who did you see at the party? B: Mary. ()

[FP [DP Maryi] [FP E [TP I saw ti at the party] ] ]

What licenses the ellipsis is a feature that Merchant calls the [E] feature, which is located in the projection to which the fragment has moved, and tells the phonological component not to parse its complement. What licenses the presence of [E], in turn, is the presence of the appropriate kind of similar content in the prior discourse: this is, in fragment answers, the antecedent question. This analysis can immediately be seen to improve on previous accounts of fragment answers, such as Morgan (), which involved only deletion but not movement, and therefore required that various constituents and non-constituents surrounding the





fragment be deleted. The problem with such a proposal is in giving a principled account of what kinds of constituent and non-constituent strings can be deleted that does not overgenerate by allowing for impossible deletions.4 Merchant’s move-then-delete analysis constrains the material that can be deleted and that which can appear as the fragment: the deleted material must be a TP, and the fragment must be a constituent that can undergo this kind of movement. As will be seen in section ., this analysis also elegantly accounts for a range of connectivity effects, capturing several parallels between fragment answers and fronted constituents. When it comes to fragments without overt antecedents, it appears that there is no linguistic content in the prior discourse to license [E], but Stanley () argues that they should be analysed similarly to fragment answers, VP-ellipsis, and so on: as encoding the syntax of a full sentence, part of which has undergone ellipsis. These cases do not have an overt antecedent, but Stanley downplays this difference, claiming that the felicitous use of these utterances depends on something in the non-linguistic context making salient a linguistic expression to serve as the antecedent for ellipsis. For example, they would often be infelicitous without a preceding ostensive stimulus (such as a demonstration) to draw attention to some object or person. They therefore are not discourse-initial in any relevant sense, claims Stanley, because the prior context necessary for their acceptability makes linguistic antecedents salient to serve as licensers for ellipsis. Consider an example where two people are talking, when across the room, a woman comes in. One of the interlocutors glances at the woman, then back to his partner with a quizzical look. The latter utters ‘John’s mom.’ The idea is that the implicit question ‘Who is that woman?’ is made salient by the questioning look, so can serve as the antecedent. Implicit question and overt answer, then, work similarly to an overt question–answer pair: The elliptical clause would be ‘That is John’s mom’ or ‘That woman is John’s mom’; ‘John’s mom’ would be fronted, followed by deletion of the rest of the clause. There are various kinds of data that sententialists appeal to that make a compelling case for silent syntax in many fragments—both of the answer and the antecedentless variety. Among this data is the case connectivity displayed by fragment answers, as illustrated in the German example (). Here, the determiner in the fragment answer must have the same case as it would in the full-sentence answer: ()

A: Wen hast du who.ACC have you ‘Who did you see?’ B: Den/*Der the.ACC/The.NOM ‘Hans.’ B’. Ich habe I have

4

gesehen? seen

Hans Hans

den/*der the.ACC/*the/NOM

Hans Hans

gesehen. see.PRF

Though see Abe and Tancredi () for a recent attempt to develop an approach that allows nonconstituent deletion.





As () and () show, antecedentless fragments in German are also required to have the same case marking that they would have if embedded in a full sentence. ()

Einen/*Ein a.ACC/a.NOM ‘A coffee, please.’

()

Ich hätte gerne I have.COND like ‘I would like a coffee.’

Kaffee, coffee

bitte. please

einen/*ein a.ACC/*a.NOM

Kaffee. coffee

In section ., I discuss in more detail how case and other connectivity effects are used as evidence for sententialism, and also assess non-sententialist attempts to account for them. In the next section, I review arguments that have been given in favour of non-sententialism, and how these could be dealt with on a sententialist account.

. A  -,   

.................................................................................................................................. In this section, I present arguments that non-sententialists have given against sententialism, or varieties of sententialism, and discuss sententialist responses. These arguments concern whether silent syntax is psychologically plausible, particularly in antecedentless fragments (section ..); ‘anti-connectivity’ effects (..), and the fact that fragment answers and their overtly sentential counterparts behave differently with regard to whether they inherit presuppositions conveyed by the question (..).

.. The psychological plausibility of silent syntax Non-sententialists have tended to cite concerns about the psychological plausibility of silent syntax as an argument for their approach. Stainton (b) raises the question of whether it is plausible that elided linguistic material is what combines with the fragment to produce a propositional interpretation in the overtly antecedentless cases such as ‘John’s mom’, discussed in section ., and argues instead for a non-sententialist account on which the fragment combines with a salient mentalese property or argument (see also Bezuidenhout  for similar considerations). Stainton’s argument is that, since the mentalese representation is prior to (and makes available) the natural-language representation ‘That woman is’, the more parsimonious explanation is that the process of recovering the proposition expressed is not mediated by a fully sentential linguistic representation. Merchant (a: –) and van Craenenbroeck (b) take up the question of the nature of the linguistic material that can be made salient in this way: while it might not be plausible that natural-language representations as specific as those posited by Stanley () are recovered, there are certain phrases, they suggest, that are so general that they





are available in almost any context. These phrases are ‘do it’ and ‘that/this is’. The words in these phrases are inherently underdetermining of the concepts they express: pronouns and demonstratives require reference assignment, and a more specific sense of the verb ‘do’ than the very general encoded meaning would often need to be inferred. This captures the fact that there is some leeway for hearers to recover different propositions expressed within a range that is compatible with the speaker’s communicative intention. A question about this so-called ‘limited ellipsis’ strategy, however, is whether it is too limited to capture all the data. The kinds of elements involved simply do not appear to account for many cases—consider two of Stainton’s examples in ()–(): ()

[Passenger to taxi driver] To Segovia.

()

[Parent to child holding a mug of hot chocolate precariously] Both hands!

It does not seem that the propositions expressed using () and () would be anything that could count as the content of ‘To Segovia it is/do it’ or ‘Both hands do it!/It is both hands!’. Instead, it looks like the elided phrases must be, for (), ‘Take me’, and for (), ‘Use’. Merchant () does mention a different possible analysis: that certain fragments are abbreviations of ‘scripts’: we memorize dialogues for conventional situations such as ordering things (e.g. ‘A coffee, please’ in ()) or getting into a taxi. It is the script, whose linguistic form is memorized, that supplies the antecedent for the elided clause. However, () is not such a conventionalized phrase, so it is far less plausible that there is a script for it. Stainton (b: –) gives more such examples, which seem to require greatly increasing the inventory of phrases available in ‘limited ellipsis’, thus entailing that any discourse-initial fragment is massively linguistically ambiguous. Alternatively, it might be that the sententialist would treat as non-elliptical any isolated fragment that cannot be accounted for by limited ellipsis in the original version Merchant (a) suggests. Merchant (b: –) lists a range of fragment types that he does not consider elliptical, extending beyond those (labels, etc.) on which the consensus is that they do not express propositions; I return to these briefly in section ... It is unclear, then, how far the limited ellipsis analysis can go in accounting for antecedentless fragments; however, there is still a problem for the non-sententialist argument that limited ellipsis was intended to address: considerations about whether the recovery of silent syntax is less parsimonious and psychologically plausible than Stainton’s mentalese/language of thought only become relevant if non-sententialism can successfully address the problem of evidence for silent syntactic structure in fragments, such as the case connectivity briefly presented at the end of section ., and discussed further in section .. Another objection based on psychological plausibility is raised by Casielles (), who points out that children produce fragment answers before they acquire the movement that is essential to Merchant’s account: for example, answers such as ‘Me’ in () are produced well before children are able to reliably produce wh-questions, at around  years  months, and subject cleft structures, at around  years old (see Guasti ). This is a concern for sententialist accounts of fragments if movement turns out to be inevitable, though would not affect non-movement accounts such as that currently being developed by Abe () and Abe and Tancredi ().





.. Anti-connectivity effects Connectivity effects, introduced briefly at the end of section ., provide evidence for sententialism; on the other hand, there are also certain anti-connectivity effects that do not, at least initially, appear to be explained on Merchant’s move-then-delete account. Example (), repeated here, demonstrates one such effect: the pronoun in (B) appears in accusative case, whereas nominative would be expected if the source were the full sentence, as in (): ()

A: Who wants candy? B: Me!

()

I/*Me want candy.

()

Me, I want candy.

Merchant claims (a: ) that the ellipsis account is supported by the fact that the fragment answer must bear the same case as it does when in the hanging topic leftdislocation structure in (), suggesting the latter is the sentential source for (B).5 This proposal faces the question of where this source comes from; as Progovac (: ) notes, this no longer appears to be a movement-then-deletion analysis. A possible answer is that the source is the biclausal structure (): ()

It is me, I want candy.

This analysis is based on Ott’s () argument that, in contrastive left-dislocation, the leftdislocated XP is a remnant of clausal ellipsis.6 However, a problem that remains to be solved for this account is that it appears to wrongly predict the possibility of the fragment answer in German, (B), derived from (): ()

A: Wen hast who.ACC have ‘Who did you see?’ B: *Der the.NOM ‘Hans.’

du you

gesehen? seen

Hans. Hans

() Der Hans, the.NOM Hans ‘Hans, I saw him.’

den him.ACC

habe have

ich I

gesehen. seen

For Merchant, the island sensitivity of fragments, to be discussed in section ., also supports this analysis. 6 Ott (: , n. ) suggests that HTLD could be analysed along the lines he suggests for CLD. He notes the problem that in examples such as (), the fronted element does not show connectivity into the host clause (, n. ) but does not pursue an account of HTLD in that paper. 5





To account for examples such as (B), ‘Me’, Progovac uses the idea of default case, although if fragments with non-default case are also to be treated as non-sentential, an explanation is needed for why accusative case is obligatory in the answer to (A); I return to this question in section ... It should also be noted that the idea of default case is controversial in the Minimalist Program, with Merchant (b), for one, rejecting it, though others (e.g. Schütze a) arguing for it, so the success of Progovac’s non-sententialist analysis here will hinge on the outcome of that debate. Casielles () also highlights anti-connectivity in bare infinitive answers, such as (), and negative answers, such as (): ()

A: What did John do? B: Play baseball.

() A: Who did you talk to? B: Not to John. Casielles claims that the non-elliptical source of B’s answer in () looks like it would have to be ‘John play baseball’; that for (B) would be ‘I talked not to John’, both of which are impossible. There are at least the following responses open to the sententialist, though. First, it could be that the sentential source for () is ‘Play baseball, John did’, as such movement of tenseless VPs is allowed in English. An alternative suggestion (den Dikken et al. ) is that the underlying full forms of these fragments could be pseudocleft or it-cleft constructions, as in ‘What John did is play baseball’ and ‘It is not to John that I talked’. Van Craenenbroeck (a) suggests ways of accounting for when the pre-deletion movement required by Merchant’s account results in simply fronting the pronounced fragment, as in (), and when it results in it-clefting or pseudoclefting.

.. The problem of presupposition inheritance A recent argument against silent syntax is based on the differences between fragment answers on the one hand and full-sentence and VP-ellipsis answers on the other. Jacobson (b) points out that when the question contains a presupposition, fragment answers and their corresponding full-sentence and VP-ellipsis answers have different properties: () A: B: B': B'':

Which maths professor left the party at midnight? Jill. Jill left the party at midnight. Jill did.

The fragment answer, ‘Jill’, commits the speaker to the view that Jill is a maths professor, but the full answer does not; it seems to communicate that the speaker thinks Jill is not a maths professor. Importantly, the VP-ellipsis answer also behaves differently from the fragment answer. This difference becomes more obvious when considering the following continuations:





() a. Jill left the party at midnight, but she’s not a maths professor. b. ? Jill, but she’s not a maths professor. c. Jill did, but she’s not a maths professor. Ellipsis accounts such as Merchant’s do not predict this behaviour, claims Jacobson, who takes it as evidence that ellipsis is not involved in deriving fragment answers. A problem for this conclusion, though, is that sluicing behaves the same way as fragment answers (Chung et al. ): ()

Some maths professor left the party at midnight, but I don’t know who.

In (), ‘who’ means ‘which maths professor’. Because sluicing is generally analysed as movement plus ellipsis, the parallels between fragment answers and sluicing seem to provide evidence for sententialism (though Ginzburg and Sag  propose a nonsententialist account of sluicing). Jacobson’s alternative proposal is that fragment answers compose directly with the meaning of the antecedent question. This is based on the idea that question–answer pairs are a basic syntactic category of our grammar, Qu-Ans. The question is taken to be a function (or partial function) from objects/entities to propositions, and only the fragment answer—not the full answer—can combine to give the Qu-Ans proposition. For example, the question in (), ‘Which maths professor left the party at midnight?’, denotes a partial function defined for maths professors: ()

λxx ∈ ⟦maths professor⟧ [x left the party at midnight]

This can then combine with ⟦Jill⟧ to give a Qu-Ans proposition only if Jill is a maths professor. This accounts for why the fragment answer commits the speaker to the presupposition while the full answer does not.7 A question that Jacobson herself raises for the Qu-Ans account, though, is whether it extends to fragment answers to implicit questions, such as those created by indefinites and by focus, illustrated in () and (): () A: Someone left the party early. B: Yes, Jill. () A: JILL left early. B: No, BILL. These are exactly the same kind of fragments that appear as answers to explicit questions: for example, in a language with morphological case, they would appear obligatorily with the same case marking as in the full-sentence answer. Antecedentless fragments will pose the

7

For some suggestions about how this account deals with the connectivity effects that are a problem for any non-sententialist account, see Jacobson (a,b, and this volume).





same problem for Jacobson’s account—after all, they are often answering an implicit question that is salient in the context, and they display case connectivity. To enable the sententialist analysis to account for the different kinds of fragments— fragment answers to explicit and implicit questions, and antecedentless fragments—while avoiding the problem of presupposition inheritance illustrated by Jacobson’s examples, Weir () adds to the account a semantic licensing condition on fragments based on Roberts’ () Question Under Discussion (QUD). Drawing on Merchant’s (a) e-GIVENness condition and an earlier QUD account by Reich (), Weir proposes a condition that he dubs QUD-GIVENness, where ‘a clause can be elided as long as the union of its focus value mutually entails the union of some unresolved question’ (: ). Returning to example (), the QUD is ⟦Which maths professor left the party at midnight?⟧ = {Jill left the party at midnight, Bill left the party at midnight, . . . }. Mutual entailment between the union of the QUD and the union of the focus value requires that Jill is a maths professor. Antecedentless fragments where there appears to be no existing QUD in the discourse, and fragment answers to implicit questions, can be dealt with straightforwardly on this account by allowing the QUD to be implicit, and accommodated by the hearer. Consider (), from Schlangen (). B’s answer seems to express (): () A: Why is seaweed good for you? B: Lots of vitamins. ()

Seaweed is good for you because it contains lots of vitamins.

()

Seaweed is good for you because . . . .

Merchant (b: –) claims that this example does not involve ellipsis. The question appears to provide the antecedent in (), which cannot combine with (B) to produce (). The QUD account, however, allows accommodation in the context of an alternative QUD, which could be ‘What does seaweed contain that makes it so good for you?’ The QUD is a semantic object, and has to be inferred in context, but what is inferred, on the QUD accounts mentioned here, is not just the semantics of the QUD but also the corresponding sentential syntax. Evidence for this silent syntax is presented in the next section.

. C     ,  - 

.................................................................................................................................. Evidence for silent syntax in ellipsis comes from a range of connectivity effects, which are discussed in detail in Merchant (a, this volume). One that applies to both fragment answers and many antecedentless fragments is case marking, discussed in section ...





In ..–.. I introduce some of the other main connectivity effects demonstrated by fragment answers: island effects, obligatoriness of complementizers, binding, and requirements for prepositions. I also discuss how non-sententialists have attempted to respond to these arguments from connectivity effects.

.. Case connectivity The strongest evidence for sentential structure in the class of fragments as a whole comes from the fact that, in languages with morphological case, the case that appears on a fragment is the same case that the fragment would have if embedded in a full sentence. For example, to order a coffee in German, you can utter the apparent subsentence (): () Einen/*Ein Kaffee. a.ACC/*a.NOM coffee ‘A coffee.’ Accusative case on the determiner is obligatory, as it would be if the fragment were embedded in a full overt sentence, as in (): ()

Ich hätte gerne einen/*ein I would like a.ACC/*a.NOM ‘I would like a coffee.’

Kaffee. coffee

This fact could be easily explained on the assumption that there is elided syntax in the fragment in (): it seems, then, that the fragment actually has the syntactic structure of the full sentence in (). The presence of the phonologically unrealized expression gerne haben (‘would like’), whose direct object is required to have accusative case, would explain the otherwise puzzling requirement for accusative case on the fragment. Merchant () presents evidence of such effects in a wide variety of languages, which points to the conclusion that all apparently isolated determiner/noun phrases might turn out to display some such effect, even if they are largely invisible in English. Some examples involving fragment answers are, from German () and Greek (): () A: Was hast du gekauft? what have you bought ‘What did you buy?’ B: Einen/*Ein a.ACC/a.NOM ‘A coat.’

Mantel. coat

B’: Ich habe einen/*ein I have a.ACC/*a.NOM ‘I bought a coat.’

Mantel coat

gekauft. bought

 ()

 A: Pjos idhe who-NOM saw ‘Who saw Maria?’

tin the

Maria? Maria

B: O the

Giannis Giannis-NOM

/*ton / the

Gianni Giannis-ACC

B’: O the

Giannis Giannis-NOM

/*ton / the

Gianni Giannis-ACC

idhe saw

tin the

Maria Maria (Merchant b: )

Unless case marking can be shown to be explained by interpretive effects, its presence in fragments strongly suggests that the fragment encodes the syntax of a full sentence. Stainton (b: ) considers the issue of obligatory case marking in antecedentless fragments. He discusses the German coffee-ordering example, where accusative is obligatory, and suggests that case marking plays a semantic role in fragments. Here, accusative case indicates that the denotation of the determiner phrase is the patient. Merchant () rejects this explanation on the grounds that structural cases including nominative and accusative cannot be assigned a consistent semantics and their function is purely syntactic. The idea of case having a semantic role is more plausible for inherent cases, such as instrumental case, which several languages have (Slavic languages, Basque, Finnish, Turkish, and others). Instrumental would be obligatory on, for instance, the Russian version of ‘Both hands!’ (Merchant : section ..). As Progovac (: –) notes, an object that has the theta role of goal/recipient is typically marked with dative case in Serbian and other languages, meaning that dative in such languages could be interpretable, as could certain genitives in some languages (although Merchant : section  rules this out for others). Progovac suggests that if a language has some inherent cases, then even a structural case such as accusative becomes informative, even without a consistent semantics, therefore its presence in fragments could be explained on these grounds. In Serbian, for example, which has dative case linked to objects, and instrumental case, accusative case on an argument would reduce ambiguity and help narrow down the possibilities for interpretation, by indicating that the argument ‘is not an agent, not a goal/recipient, not an instrument’ (: ). While this account remains to be developed in detail, one obstacle that it faces is that, if, for example, a language where dative indicates goal/recipient can employ another way of expressing that something is the goal/recipient, then the lack of dative case marking does not rule out this interpretation. An alternative suggestion is that structural cases encode not a semantics in the sense at issue above, but constraints on interpretation—that is, on the conceptual structure to be built. According to Kempson et al. (this volume), accusative case marking on an expression, such as the German coffee example, introduces the constraint that the immediate combination of that expression with another must result in a predicate. It is semantic expressions, rather than syntactic ones, that are combining, so this is a non-sententialist account. The case marker indicates how the fragment should fit into the proposition expressed, so provides a clue to how to enrich the fragmentary logical form that is encoded. The idea seems to be that using the nominative (which acts as the default case in German) could be misleading, depending on what constraint is proposed for nominative case, or would at





least be less informative in examples where the accusative could be used, which would predict the infelicity of nominative in an example such as this.8 However, this account, like Progovac’s, has yet to be fleshed out and does not currently appear to have an answer to why, for example, the nominative should not be able to trigger the construction of a different proposition, such as It is a coffee that I want. So far, then, case connectivity provides evidence for the silent syntactic structure assumed by sententialists. Other connectivity effects have been offered in support of the sententialist account of fragment answers, and those which I introduce briefly in the following subsections have been used specifically to support Merchant’s idea that the fragment has been fronted before elision of the TP.

.. Islandhood At least some fragment answers seem to be subject to island effects, as shown in () and (): () A: Did each candidate agree on who will ask him about taxes at tonight’s debate? B: *No, about foreign policy. (Merchant, this volume) () A: Does Abby speak the same Balkan language that Ben speaks? B: *No, Charlie. The sentential sources of B’s answers, if there is movement before deletion, would have to be () and (), the ungrammaticality of which would explain the ungrammaticality of the fragments derived from them: () *No, about foreign policy, each candidate agreed on who will ask him t at tonight’s debate. ()

*No, Charlie she speaks the same Balkan language that t speaks.

One issue that this analysis raises is that island effects might not be expected to appear in fragments at all, since sluicing does not show the same effects, as can be seen from a sluicing example similar to (): ()

Each candidate agreed on who will ask him about something at tonight’s debate, but I don’t know what.

Because Merchant analyses fragments and sluicing as TP ellipsis, this difference in island sensitivity is unexpected at first sight, but this is where the Functional Projection mentioned in section . comes in. In sluicing, the remnant moves to CP, then TP is elided; in 8

This suggestion, that it is infelicity rather than ungrammaticality strictly speaking, is also made by Merchant (b: ).





fragments, there is an additional step of movement to the FP, which leaves a non-elided trace in the CP which accounts for the island sensitivity. Griffiths and Lipták () criticize the ad hoc nature of this extra step of movement: it is posited purely to create a trace that will explain the island sensitivity of fragments. While it is ad hoc for English, however, Temmerman () presents evidence that motivates the extra movement in Dutch, as Griffiths and Lipták appear to acknowledge, despite noting some variability in judgments about the Dutch data (see Griffiths and Lipták : –, n.  for discussion). Griffiths and Lipták (: –) also show that in several kinds of example, including the following, island effects appear to be voided: ()

A: I imagine John wants a detailed list. B: I’m afraid he does. Very detailed.

() A: I hear that Abby is likely to get mad if Ben speaks to one of the guys from your syntax class? B: Yeah, John. Their conclusion is that only contrastive fragments are island-sensitive; non-contrastive ones (where the antecedent contains an indefinite correlate of the fragment answer) are island-insensitive. This mirrors the pattern identified in sluicing by Merchant (b): contrastive sluicing, illustrated in (), from Barros et al. (), is island-sensitive; non-contrastive sluicing, as in (), is not: () * The radio was playing a song that Ringo wrote, but I don’t know who elsei the radio was playing a song that ti wrote ()

The radio was playing a song that Ringo wrote, but I don’t know what song the radio was playing that Ringo wrote

As with the parallels between sluicing and fragment answers mentioned in section .., on presupposition inheritance, this looks to provide more evidence for treating fragment answers as requiring, like sluicing, a movement-then-deletion analysis. To capture the difference between the contrastive and non-contrastive fragment answers, Barros et al. () propose that what is happening in these non-contrastive cases is not that ellipsis voids island violations, but that what is elided is not isomorphic to the antecedent; instead, it is a cleft, for example: () A: I hear that Abby is likely to get mad if Ben speaks to one of the guys from your syntax class? B: Yeah, John it is. The island effects discussed here do not provide unambiguous evidence for sentential structure, as there is a lack of consensus about the empirical facts. Casielles (), Culicover and Jackendoff (), and Stainton (b) all note variation in judgments of acceptability depending on the salience of the intended contrast in examples such as ()–(). For data





that tends to elicit more uniform judgments, though, we can look to similar locality effects. Merchant (a: –) discusses predicate answers such as ()–(): () A: What did he do to the car? B: Totaled *(it). () A: What did he do for his sister? B: Funded *(her). As the object is easily recoverable from the context, it is not clear how the non-sententialist would explain why the object cannot be omitted. This follows from the movement-thendeletion analysis: in English, only the whole VP, e.g. ‘totaled it’, not the verb alone, can move to the left periphery.

.. Complementizer distribution If the move-then-delete account of fragment answers is correct, then complementizers should be required in English fragment answers, as in (). This is because the full-form equivalents of these answers only allow fronting with an overt complementizer, illustrated in (): () A: What did Katy deny? B: *(That) she had taken the keys. ()

*(That) she had taken the keys, Katy denied.

Answers of the kind in (B) are significantly more acceptable with ‘that’ than without, as confirmed in a judgment task run by Merchant et al. (). Their conclusion is that, since ‘complementizers do not obviously add any lexical semantic meaning to the answers, it is difficult to see how a syntactic theory without inaudible syntax for fragment answers could explain these results’ (: ). In response, Jacobson (b) and Progovac () point out that the correlation between the degradedness of the fragment answer and that of the full sentence in () and () does not mean that one caused the other; a third factor could be the cause of both namely, they suggest, the fact that null complementizers are licensed only under adjacency with the verb, as shown by ()–() along with data such as the following: ()

Katy denied (that) she had taken the keys.

()

He believes, not surprisingly, *(that) Mary did it.

The adjacency explanation would work for Jacobson b, which has the fragment answer composing directly with the question; verb and complementizer are not adjacent in the construction that results, so the complementizer must be overt. On the kind of analysis





developed in Progovac (), however, where there is no verb ‘deny’ in (B), and no natural-language complementizer—overt or null—it cannot be adjacency or non-adjacency of them to each other that determines whether an overt complementizer is required; another factor must be responsible. Culicover (a) gives an account of where English zero-relatives such as () are possible that hints at how a non-sententialist, non-Qu-Ans account of complementizer distribution in fragment answers might go: ()

This is a book Ø you should read.

Building on work by Jaeger and Tily () which finds that the more predictable ‘that’ is in a given context, the more likely it is to be omitted, Culicover argues that, where zero relatives are unacceptable, this reflects complexity of processing—how easy it is to identify the relative clause as a relative clause without the relative ‘that’—and such examples are syntactically well-formed. Applying this idea to complementizers in fragment answers, a Progovac-style non-sententialist analysis could try to argue that ‘that’, while not adding lexical semantic meaning, does provide an interpretive clue which could be expected to ease processing: it indicates that what follows should be interpreted as the direct answer, as opposed to the various other possibilities for interpretation available when the complementizer is absent.

.. Binding in fragment answers Another connectivity effect cited by Merchant is binding connectivity. Two of his examples are: ()

A: Who did John1 try to shave? B: *Him1. B’: *John1 tried to shave him1.

() A: Where is he2 staying? B: *In John2’s apartment. B’: *He2 is staying in John2’s apartment. The fragments in (B) and (B) are not possible as answers to the antecedent questions. On Merchant’s account, the sources for these fragments would be (B’) and (B’), which violate principles B and C of binding theory respectively, and this is seen as another strong argument that there is sentential syntax in the fragments.9

9

A recent attempt, yet to be assessed, to account for the binding facts without appeal to silent syntax was made by Jacobson (b: –), who applies to fragment answers the direct compositionality account of binding detailed in Jacobson (, a).





.. Prepositions in fragment answers Languages that do not allow preposition stranding (P-stranding) in full sentences require a preposition to appear in corresponding fragment answers. In English questions, P-stranding is possible, as in (A), and the fragment answer (B) does not require the preposition, whereas in German, which forbids P-stranding, the preposition moves together with the object in question formation, and in the corresponding fragment answers, the preposition is obligatory, as illustrated by (): ()

A: Who was Anna talking with? B: Hans.

() A: Mit wem hat die with whom.DAT did the ‘Who did Anna talk with?’ B: *(Mit) *(With) ‘Hans.’

dem the.DAT

Anna Anna

gesprochen? talk

Hans. Hans (Merchant b: )

This is easily explained on Merchant’s account of fragment answers, where the derivation of the fragment has involved movement then deletion of the phrase that has been moved out of it. In English, the object can be fronted without its preposition, as in (), but in German, only the entire PP can be fronted, as illustrated by (): ()

Hans, [Anna spoke with t]

()

Mit dem Hans [hat die Anna t gesprochen]

What remains after deletion of the TP, then, must be the fronted PP in German and other non-P-stranding languages. This requirement for prepositions in fragment answers is a major challenge for nonsententialist accounts, at least in languages such as German () and Spanish () where the preposition does not appear to contribute anything to the interpretation: () A: Con quien hablaba? with whom speak.PST.S ‘Who did he speak with?’ B: Con el Juan. with the Juan ‘With Juan.’ In Serbian, in contrast, as discussed by Progovac, there is interaction between prepositions and determiners that affects interpretation. Here is one of her examples (: ):





() A: Na on B: Na on

čega what

je did

Stefan Stefan

stolic-u / chair.ACC /

seo? sit

*Stolic-u chair.ACC

When not governed by a preposition, accusative in Serbian is associated with the theme/ patient role and whole-object affectedness; the preposition is needed to override this interpretation. Progovac does not suggest an account for those non-P-stranding languages where the preposition in fragment answers does not have an obvious effect on interpretation, such as German and Spanish, but Jacobson (a) addresses the P-stranding facts, raising several questions that remain to be answered before it can be concluded that they are evidence for silent syntax. These include, first, apparent counterexamples to the P-stranding generalization, where languages that forbid P-stranding seem to allow preposition omission in sluicing. In the Spanish answer in (), con (‘with’) is required, but it can be omitted in sluicing, as illustrated in () (Casielles : ); Brazilian Portuguese and Polish also allow this: () Hablaba con alguien, pero no sé (con) speak.PST.S with someone but not know.S (with) ‘He talked with someone, but I don’t know (with) who.’

quién. who

Furthermore, it is not known what determines whether or not a language allows P-stranding, and, as Progovac et al. (a: ) also point out, only if the explanation must invoke the full-sentence projection and not just the PP would prepositions in fragment answers constitute evidence against the non-sententialist approach. While the cross-linguistic picture for P-stranding is complex, though, there are analyses of examples such as () that maintain the P-stranding generalization by proposing that they involve an underlying cleft (Rodrigues et al. ; van Craenenbroeck a; Vicente ). On this analysis, then, the structure of () without the preposition would be (): ()

Hablaba con alguien, pero no sé quién speak.PST.S with someone but not know.S who ‘He talked with someone, but I don’t know who it is t.’

es t is

While Jacobson (a: –) raises some more questions for this account, the P-stranding facts pose a major challenge that non-sententialists have so far made little progress with, and are most naturally understood as strong support for the movement-and-deletion account of fragments.

. C

.................................................................................................................................. Sententialist accounts of fragments (Merchant a, b, ; Abe ; Barros et al. ; van Craenenbroeck a; Ott ; Temmerman ; Weir ) and the non-sententialism





of Progovac et al. (c; Progovac ), Barton (, ), Casielles (), Stainton (b) each successfully account for certain features of fragments, but face difficulties with other features. Merchant’s movement-then-deletion analysis, followed by many others, deals elegantly with many examples of case connectivity in both fragment answers and antecedentless fragments, and with binding connectivity, island violations, and complementizer and preposition requirements in fragment answers, though there are arguments which may instead motivate a non-deletion analysis (Casielles ; Abe and Tancredi ). Additional connectivity effects that have recently been cited as support for sententialism, though not yet widely discussed, are gender marking and attributive versus predicative marking on fragments (Martí ). While those who are sceptical about silent syntax have addressed the connectivity data (Progovac ; Jacobson a,b) and begun to suggest some routes for nonsententialists to explore, the facts so far strongly support the idea that fragments require a sententialist analysis.

A Thanks to the reviewers and the editors for their comments on earlier versions, and to Patrick Elliott, Matthew Reeve, and Ye Tian for discussion.

  ......................................................................................................................

                  ......................................................................................................................

 

. I

.................................................................................................................................. T surface appearance of the comparatives in () is shaped by six deletion operations— gapping, VP-ellipsis, pseudogapping, stripping, across-the-board (ATB) movement and Right-Node Raising (RNR)—which are also attested in other syntactic environments, typically coordinate structures (unpronounced parts marked by angled brackets): () a. b. c. d. e. f.

Ann liked Berlin more than Cleo Dubai. (Gapping) Ann liked Berlin more than Dubai. (ATB movement and Gapping) Ann liked Berlin more than Cleo . (Stripping/Gapping) Ann liked Berlin more than Cleo did . (VP-Ellipsis) Ann liked Berlin more than Cleo did Dubai. (Pseudogapping) More people liked than disliked the place. (RNR)

All of the deletion processes in () are, modulo the presence of do, optional. The examples in () document another type of ellipsis found in comparatives, traditionally referred to as C D (CD; Bresnan , ), which renders unpronounced the gradable property (a) or the common noun minimally including the gradable property (b,c) inside the clause following than. Unlike the ellipsis processes in (), CD is construction-specific and obligatory:1 ()

a. The ship is longer than the train is (*long). b. Korea built better ships than Greece owns (*good ships/*ships). c. Korea built more ships than Greece owns (*ships).

1 Two notable exceptions are    (see (i)), which are non-elliptical on all extant accounts, and   ((ii); Corver ; Lechner and Corver ); subdeletion will be ignored throughout.

(i) The ship is longer than m. (ii) The ship is longer than the shipyard is wide.

 



The present chapter surveys the main characteristics of the deletion mechanisms illustrated in () and (), and explores analytical options that have been pursued in attempts to account for the principles underlying these phenomena. Central to the discussions will be three questions which over the last forty years have time and again served as a practical vantage point for studying ellipsis phenomena (see recent surveys in Merchant, this volume, and Aelbrecht ): ()

a. Does the unpronounced part of the construction contain syntactic structure? b. What are the identity conditions on the ellipsis and its antecedent? c. What are the licensing requirements for ellipsis?

As already indicated by the grouping in () and (), in comparatives, these questions materialize in two partially overlapping guises. On the one hand, the triad () helps to clarify whether degree constructions afford new insights into the principles governing ellipsis. For instance, if (a)–(c) are actually the result of forward verb deletion and movement, two processes that are known to be restricted to coordinate structures, how come that comparatives, which are usually held to establish a subordination relation, emulate the behavior of coordinations? On the other hand, it is also possible to use () in an attempt to gain a better understanding of the comparative construction itself. Here, in particular, two topics have attracted a significant amount of attention in the literature, both of which will be addressed below: the nature of Comparative Deletion (see ()), and the mechanisms responsible for the formation of   (PCs), illustrated by (b) and (c). Superficially, PCs are simply comparatives in which the particle than combines with a single, typically nominal constituent. Since the early s, it has been recognized, though, that there are two competing analytical strategies to capture this observation. PCs can be parsed either as reduced clausal comparatives embedding hidden syntactic structure, or base-generated PPs that are introduced by than (Hankamer b). At the time, cross-linguistic variation had been thought to be related to typological universals. Interestingly, recent studies have revealed that PCs do not form a homogeneous class in that PCs react to routine tests for the presence of silent syntactic structure in some languages, including English, but not in others, among them Japanese and Hindi (Bhatt and Takahashi ; Bochnak ). The debate leading up to this conclusion, which also has important consequences for the semantic treatment of degree constructions more generally, will be traced in some detail in the sections to follow. In this context, a concomitant question to consider will be whether those instances of PCs for which the diagnostics indicate the presence of hidden structure are best uniformly treated as the outcome of standardly sanctioned ellipsis operations, or whether there are designated kinds of deletion processes only operative in comparatives. The chapter is organized as follows. Subsequent to some preliminaries on the syntax and semantics of comparatives in section ., section . turns to CD, comparing three competing analyses in terms of ellipsis and movement which mainly diverge in the amount of hidden structure they allocate inside the CD-site. Section . expands on the different kinds of ellipsis in comparatives. PCs will be treated in section .. As both the phenomenology of and the literature about comparatives is extraordinarily rich, the presentation will by necessity be selective and limited in depth. Complementary surveys of comparatives which also treat ellipsis phenomena are Klein (), Corver (), Pancheva (), Morzycki (), Lechner (forthcoming), and Lechner and Corver ().





. T     

.................................................................................................................................. D , which include comparatives, equatives, superlatives, and enough/too-constructions, form a class of structures which express an ordering between two (sets of ) degrees. Comparatives induce an asymmetric ordering between the degree introduced by the main clause and the degree specified by the   (henceforth than-XP). For instance, (a), repeated below as (a), expresses the proposition that the degree of the ship’s length exceeds the degree of the train’s length, while the   (b) and    (c) assert that the quality and quantity of ships built by Korea exceeds the quality and quantity, respectively, of ships owned by Greece: ()

a. The ship is longer [than-XP than the train is]. b. Korea builds better ships [than-XP than Greece owns]. c. Korea built more ships [than-XP than Greece owns].

Following a widely adhered-to terminological convention, the second degree of the comparative relation is also referred to as   , whose left edge is demarcated by the   (than). While the standard in () is , it can also be left , as in John is tall, but Sam is taller. The degrees themselves are supplied by a  , typically represented by a gradable adjective denotation, as in (a) (long) and (b) (good), or by the   much/many, as in (c). Degree adjectives have been analyzed in two ways, as measure functions from individuals to degrees (logical type ; Bartsch and Vennemann ; Kennedy ), or as relations between degrees and individuals (type ; Cresswell ; von Stechow , among others). Here, I will adopt the latter strategy, without any discernable consequences for the argumentation. On this conception, long denotes the relation which maps each degree to the individuals which are long to that degree. As mentioned above, the gradable property is removed from the comparative complement by CD; for details see section .. Degree clauses share various properties with relative clauses. Just as relative clauses denote derived predicates of individuals, clausal than-XPs can be analyzed as derived predicates of degrees, or, equivalently, as sets of degrees. In both instances, set formation is the semantic reflex of empty operator movement to a clause-initial position (von Stechow ; Heim and Kratzer ). In the case of comparatives, movement is usually interpreted as λ-abstraction over the degree variable inside the gradable property which has been silenced by CD. The LF of the degree complement for our sample comparative (a) accordingly looks as in (a), and denotes, as seen in (b), the set of degrees to which the train is long (von Stechow ; Rullmann ; Heim , inter alia):2 2

Than will be assumed to be vacuous throughout; see Alrenga et al. () for an analysis in which than has a semantic contribution. The empty operator will be ignored for the moment.

  () a. [than-XP than [CP OP [λ1 [TP the train [VP is ]]]]] b. ⟦(a)⟧ = {d|the train is d-long}



(LF for than-XP of (a))

(equivalently: λd.the train is d-long)

Syntactic evidence for OP-movement comes from the observation that comparatives respond to barriers imposed by islands, among them the Complex Noun Phrase Constraint in (a) and the Coordinate Structure Constraint (CSC) in (b) (Ross ; Chomsky b): () a. *Korea built more ships [λ1 than Greece had discussed [DP a plan to buy ]]. b. *Korea built more ships [λ1 than Greece bought [tanks and ]]. The comparative relation is expressed by an abstract degree head  which projects a   (DegP; Abney ; Corver ) and selects for a degree complement and the gradable property. A more fine-grained representation of (a) is given in (). ()

The ship is [DegP long  [than-XP than the train is [DegP ]]].

On a popular analysis, which models degree heads in analogy to quantificational determiners in the individual domain (type ),  denotes a two-place second-order relation between degrees (type ). One of various possible implementations is given in () (Gawron ; Heim , ; Bhatt and Pancheva ): () ⟦⟧ = λD.λD’ .D ⊂ D’

(Bhatt and Pancheva )

The DegP of () can then be expanded into the tree in (), with AP occupying specDegP (Izvorski ; Lechner , ) and the degree complement in complement position: ()

DegP AP long

Deg’ (= DegGQ) more than-XP

In (), the unit consisting of  and the than-XP (henceforth also DegGQ) denotes a generalized degree quantifier, that is a second-order property of degrees. Similar to generalized quantifiers in the complement position of individual predicates, this DegGQ is not type-compatible with its sister node and therefore needs to undergo Quantifier Raising (QR), as spelled out in (a) (Heim ).3 The moved DegGQ in (a) binds a degree variable which serves as the inner argument of the matrix AP. Together with the

3

Concerns that QR affects an intermediate projection (Deg’) can be defused by adopting Bhatt and Pancheva’s () Late Merge hypothesis or Alrenga et al. (); see discussion of (). Just like covert movement of nominal quantifiers, QR in (a) leaves a movement copy (not represented). Again, see the discussion of Late Merge surrounding ().





lexical semantics of  and the degree complement, the semantic computation yields (b), which demands that the set of degrees to which the train is long forms a proper subset of the set of degrees to which the ship is long (or, equivalently, that the maximal degree of the ship’s length exceeds the maximal degree of the train’s length; for maximality analyses see von Stechow ; Rullmann ). () a. [DegGQ more [than-XP than λ1 the train is ]] [λ1 the ship is d1-long] b. ⟦()a⟧ = ⟦⟧(⟦[than λ1 the train is ]⟧) (⟦[λ1 the ship is d1-long]⟧) = = λd.the train is d-long ⊂ λd.the ship is d-long A noteworthy property of this account is that comparatives parallel propositions with universally quantified objects both in their logical syntax and meaning. Discussion of empirical predictions generated by a corollary of the assumption that DegGQs undergo QR will be taken up presently. The factorization in () is one of the three main approaches towards the constituency of gradable predicates that have been pursued in the literature, all of which are associated with a distinct set of consequences. On the ‘classical’ view, sketched in () (Chomsky ; Selkirk ; Bresnan ; Heim ), which is directly compatible with the semantics above, the DegGQ is not a daughter of DegP, but is situated in the specifier of AP. If the DegP is organized as in (), QR targets the specifier of AP, instead of the sister of AP. The differences between () and () are minor, but will be seen to render the phrase structure () inconsistent with a particular perspective on CD (to anticipate, the raising analysis of CD; see section .). ()

AP (DegGQ =)

Aº

DegP

more

than-XP

Finally, the DegP has also been assigned the template in (), in which  and the AP form a constituent to the exclusion of the degree complement. The than-XP joins the derivation as a DegP adjunct, resulting in (), or is merged as the external argument of the degree head (von Stechow ; Abney ; Larson b; Corver , ; Rullmann ; Kennedy ): ()

DegP DegP more

than-XP

AP

Due to this structural modification, either the lexical entry of  has to be adjusted, or the order in which the components are assembled needs to be changed. The former option can be operationalized by adopting the meaning rule (a), which while retaining the subset

 



condition of the generalized quantifier analysis does not require scoping. Alternatively, it is also possible to switch to a semantics that treats the than-XP as a nominalized degree term, as in (b) (von Stechow  and Rullmann ; for a third option in terms of measure functions see Kennedy ). The typing in () reflects (a): ()

a. ⟦’⟧ b. ⟦’’⟧

= =

λAP.λD.λxe.D ⊂ λd.AP(d)(x) λAP.λdd.λxe.max{d’|AP(d’)(x)} >d

(von Stechow )

But the phrase structure in () can also be made consistent with the generalized quantifier analysis in () by abandoning the standard phrase structure axiom that all syntactic trees grow strictly monotonically. Specifically, a variant of () becomes interpretable if Bhatt and Pancheva’s () hypothesis is adopted that only the degree head  moves at LF and that the degree complement is counter-cyclically inserted in the scope position of  by Late Merge (Lebeaux ). On this conception, the derivation of () starts with the base-generated structure in (a). Next,  raises covertly, leaving a degree trace that serves as the inner argument of the AP-denotation (b). Finally, the than-XP is Late Merged as a sister node of  (c). ()

a. The ship is [DegP  long]. b.  [λ1 the ship is [DegP d1-long]] c. [ [than-XP than λ1 the train is ]] [λ1 the ship is d1-long]

The resulting LF-representation (c) for all intents and purposes parallels (a), except that the derivation does not create a movement copy in the base position of the than-XP. This has, as Bhatt and Pancheva point out, various desirable consequences. Among others, the Late Merge hypothesis improves on the standard QR-account in that it offers an explanation for systematic correlations between the scope of  and the size of ellipsis that the than-XP is able to embed, captured by the E‑S G () (adapted from Bhatt and Pancheva : (); Williams : ch. ; Gawron ; Fox ; Lechner : ): () Ellipsis‑Scope Generalization The LF-position of the degree complement matches the scope position of . Empirically, () manifests itself, among others, in the observation that in elliptical comparatives, possible coreference patterns reflect the height of attachment of the than-XP, which in turn regulates possible choices for the ellipsis antecedent. To illustrate, () admits coreference between Mary and the pronoun her in case the missing VP is construed with broad ellipsis, paraphrased in (b), whereas the narrow ellipsis reading (a) incurs a Principle C violation:4

The full paradigm also involves a licit reading on which  and the degree clause take matrix scope but the ellipsis is identified narrowly, by the embedded VP works hard ((c) in Bhatt and Pancheva ). This documents the well-known, yet orthogonal, phenomenon that VP-ellipsis can reach into embedded clauses in search for their antecedents (Fiengo and May ). 4





() Her father tells her to work harder than Mary’s boss does. (Bhatt and Pancheva : , ()) a. *Her father tells her3 to work harder than Mary3’s boss works. b. Her father tells her3 to work harder than Mary3’s boss tells her to work. c. *Her father tells her3 [[  than [ λ1 Mary3’s boss ]] [ λ2 PRO to work d2-hard]] (tell  ) d. [[  than [ λ1 Mary3’s boss ]] [ λ2 her father tells her3 PRO to work d2-hard]] (  tell) The paradigm is accounted for as follows. The narrow ellipsis reading in (a) requires the degree complement to be merged in the embedded clause (c), triggering a disjoint reference effect between the name and the c-commanding coindexed pronoun (underlined). By contrast, construing  with wide scope, as in (b/d), makes it possible to Late Merge the than-XP within the matrix clause, above her, which removes the name from the c-command domain of the pronoun. Essentially, what () demonstrates is that matrix scope for  does not entail that the degree complement is syntactically represented in a position next to the degree adjective, as would be expected in the montonic derivation (b): ()

a. Late Merge: [ [than-XP . . . name3 . . . ]1 [ . . . pronoun3 . . . t1 . . . ]] b. Cyclic Merge: *[ [than-XP . . . name3 . . . ]1 [ . . . pronoun3 . . . [than-XP . . . name3 . . . ]1]]

The behavior of degree complements mimics in this respect that of extraposed relative clauses (b), which are equally able to escape the verdict of Principle C by Late Merge (Fox and Nissenbaum ; Fox , , inter alia): ()

a. I gave him3 a book yesterday [that John3 liked]. b. ??I gave him3 a book [that John3 liked] yesterday.

(Fox : ())

Two remarks regarding the Ellipsis‑Scope Generalization () are in order. First, on the assumptions above, the effects of () cannot be derived from the denotation of  in (a) or (b). This supplies an argument for the generalized quantifier analysis of the degree head (). Second, not all researchers agree that the best account of the Ellipsis‑Scope Generalization () resides with the Late Merge theory. Alrenga et al. (), for example, capture () by designing a semantics that admits multiple possible attachment sites for the comparative complement. To recapitulate, although syntactically comparatives fit at least three distinct syntactic templates, the generalized degree quantifier analysis limits the analytical options: while the structures () and () can be embedded both within a conservative QR analysis and the Late Merge account, () is compatible with the latter only. Further consequences of this finding will be discussed in section .. Moreover, Late Merge and, concomitantly, the second-order degree predicate analysis of the comparative relation were seen to receive support from their ability to handle systematic correlations among scope, Principle C, and

 



ellipsis (but see Alrenga et al. ). Similar interactions between ellipsis and comparatives will be taken up again in section ., following a survey of empirical and theoretical challenges posed by Comparative Deletion (CD) in the next section.

. C D

.................................................................................................................................. It is possible to distinguish among at least three families of approaches towards CD, which diverge mainly in whether they endow the silent gradable property with syntactic structure or not, and the mechanism which identifies the elliptical gradable property. More specifically, CD has been analyzed as (i) the result of syntactic ellipsis in combination with movement of a degree operator (which, on some accounts, itself consists of a deleted constituent); (ii) a designated type of movement operation (‘head raising’), and (iii) an instance of semantic ellipsis, with the CD-site being syntactically inert. The present section provides a synopsis and outlines how the syntactic and semantic assumptions collected in section . align with these competing perspectives on CD.

.. The movement and deletion (matching) analysis of CD Proponents of the movement and deletion analysis (Lees ; Hankamer b; Chomsky b; Kennedy ) agree that CD combines fronting of a constituent inside the comparative complement with the instruction to forgo pronunciation of the degree predicate, possibly together with other nodes. What exactly moves or is deleted depends on the particulars of the theory, though. To exemplify, the classical empty operator analysis (Chomsky b; von Stechow ; Heim ; Izvorski ; Rullmann , among others) postulates movement of a null operator to specCP. From there, the operator binds a degree variable within the gradable predicate, which is deleted by a constructionspecific ellipsis operation. In a recent incarnation, developed in Kennedy (), displacement targets the whole DegP-complex instead of just the null operator, followed by ellipsis of the higher occurrence of DegP. Kennedy’s analysis also diverges from the classical account in that the variable is not located within the degree predicate, but is syntactically represented by the lower copy of the DegP. On this view, the effects of which are exposed in (), CD is a manifestation of matching the unit [DegP Deg° long] in specCP with the external head of the degree clause [ long]:5 ()

The ship is [[DegP  long] [than-XP than [CP the train is ]]]. (‘The ship is longer than the train is.’)

5 On why the two occurrences of DegP in the matrix clause and fronted position of the than-XP do not need to be strictly identical ( long vs Deg° long), see Kennedy : . For an analysis that observes strict identity, see Alrenga et al. ().





Modeling CD in terms of DegP-movement has two immediate consequences. First, it is no longer necessary to stipulate a designated ellipsis operation affecting the lowest DegPoccurrence, because deletion follows from the general principles regulating the distribution of audible copies. Second, the semantic part of the analysis, to be expanded on momentarily, includes a device which operationalizes the matching relation between the higher DegP and the external head. Adjusted to current assumptions according to which gradable adjectives denote relations between degrees and individuals, the abstract degree head Deg° of Kennedy () can be defined as in (a).6 (b) repeats from above the non-scoping version of ’ and () supplies relevant parts of the semantic calculation for (). The LF to be compositionally interpreted is (a): () a. ⟦D°⟧ b. ⟦’⟧ ()

= λAP.λR.R(λxλd.AP(d)(x)) = λAP.λD.λxe.D ⊂ λd.AP(d)(x)

(= (a))

a. LF: (than) [DegP Deg° long] λ1 the train (is) [DegP Deg° long]1 b. ⟦λ1 the train [DegP Deg° long]1⟧ = λT.T(the train) c. ⟦Deg° long⟧ = λAP.λR.R(λx.λd.AP(d)(x))(λd.λx.long(d)(x)) = λR.R(λx.λd.long(d)(x)) d. ⟦ [[Deg° long] λ1 the train T1]⟧ = = λR.R(λx.λd.long(d)(x))(λT.T(the train)) = = λT.T(the train)(λx.λd.long(d)(x)) = = λx.λd.long(d)(x)(the train) = = λd.long(d)(the train)

As shown by (b), the lower DegP translates as a variable of type which is bound by the λ-operator created by DegP-movement to specCP. Semantically, abstraction over T generates a predicate of ‘passivized’ (i.e. instead of ) gradable adjective meanings. Once this derived predicate is combined with the DegP-denotation, given in (c), the derivation produces, as desired, the predicate of degrees to which the train is long (d). The output of (d) can finally be picked up by the comparative head ’ (b). A noteworthy feature of this set-up is that the identity condition between the internal head in specCP and the external head is defined syntactically, yet the content of CD is restored in semantics, by β-converting the degree predicate inside the fronted DegP into the CD-site. Moreover, as DegP movement to specCP arguably leaves a copy, the gradable property inside the than-XP contains silent structure in specCP, as well as in its base position. As observed by Kennedy (: n. ), the derivation in () is reminiscent of the matching analysis of relative clauses (Hulsey and Sauerland ; Bhatt ) in that a category which is internal to the relative/degree clause is raised and elided under identity with an external head. Syntactically, the proposal is based on the constituency for the DegP in (), which, as exposed by (a), guarantees that the internal DegP ([Deg° AP], 6

For the original analysis, on which adjectives denote measure functions, see Kennedy (: –).

 



underlined) actually finds a constituent which supplies a matching external head ([ AP], double underlining): () a. [DP [DP  AP] [than-XP . . . . . . . ]] (à la ()) b. [DP AP [D'  [than-XP . . . . . . . ]]] (à la ()) c. [[DP  [than-XP . . . . . . . ] A(P)]] (à la ()) None of the alternative syntactic templates for DegP satisfy this requirement. Both in (b) (which models the DegP after ()) and (c) (which follows ()), the internal DegP lacks a corresponding external head that excludes the than-XP, suggesting that the matching analysis is compatible only with the phrase structure in (). While the observation above does not have any detrimental effects in itself, it relates to another, more problematic trait of the analysis: the matching account is, at least in its present form, inconsistent with the Late Merge hypothesis. This follows from the fact that the two theories impose two conflicting sets of requirements on the constituency of  and the AP. On the one hand, the Late Merge account mandates that the than-XP be inserted as the sister of , instead of as a sister to the unit [ AP]. That is,  has to form a unit with the than-XP to the exclusion of the AP. By contrast, the matching analysis makes just the opposite kind of demand since the local context must supply an external head—for matching with the internal DegP—which needs to contain both  and the AP, but not the than-XP. Put differently, counter-cyclical Merge inevitably locates the than-XP in an environment that fails to provide an external head.7 Note, moreover, that the problem generalizes into two directions. First, it is not restricted to the particular template in () but extends to the alternative two phrase structures. Neither () nor () is able to reconcile the assumption that the than-XP is Late Merged as a sister of  with the need of the internal DegP for an external matching [ AP] unit. Interestingly, while this finding poses a substantial challenge for the matching analysis, a similar (yet not identical) shortcoming will also be seen to impinge upon its main competitor to be addressed in the next subsection. Second, it is hard to see how the matching analysis derives scoping of the comparative relation  in the first place, given that (i)  and the than-XP do not form a constituent, which prohibits them from QRing as a unit, and (ii) movement of  in isolation bleeds Late Merge. Naturally, it would of course be possible to scope only the than‑XP, but there is no discernable strategy for transporting  together with the than-XP into their scope positions. This second generalization of the problem makes the matching analysis incompatible with the standard degree quantifier approach. An implementation that avoids this complication can be found in Alrenga et al. (), though.

7

The problem does not in arise in head-external relative clauses, because the node that is moved prior to Late Merge is also the head of the relative.





.. The raising analysis of CD Similarities between relative clause and comparative are also exploited by the second strand of analyses, which construe comparatives in analogy to the head-raising analysis of relative clauses (Vergnaud ; Kayne ; Bhatt ; on head raising in comparatives see Donati ). In Lechner (), it is proposed that the gradable property is attracted by a feature on the matrix degree head, landing in its specifier position, where the feature is morphologically expressed by comparative morphology under spec–head agreement: ()

The ship is [DegP [AP long-er] [Deg' Deg°[+comp] [than-XP than λ1 the train is [DegP [Deg' Deg°[E] d1]]]]]

Just like in the head-raising account of relative clauses, only the higher copy is spelled out. This can be made to follow from the assumption that the lexical specification of the lower, semantically inert degree head contains an ellipsis feature [E] (Merchant ), which instructs the grammar to forgo pronunciation of specDegP.8 Moreover, semantic considerations demand that both the higher and the lower copy of the gradable property are interpreted, resulting in an instance of movement without chain formation.9 Since the relation between the comparative AP and the CD-site is defined in terms of movement, the identity condition on CD is the same as the identity requirement operative in other instances of dislocation. Thus there is no need to grant special dispensation to the relation between the external head [ AP] and the internal head [Deg° AP] from the laws defining syntactic or semantic identity, as was necessary in the matching analysis (see also n. ). In general, the raising analysis does not make it necessary to stipulate a constructionspecific, obligatory deletion process for comparatives; rather, the core properties of CD fall out from the universal principles regulating movement. It is obvious that for head raising to succeed, the internal structure of gradable APs must be like (), because only () ensures that the pronounced, morphologically marked AP c-commands the than-XP and at the same time resides in the specifier of the degree head. Once head raising has applied, the quantificational degree determiner  in () together with the than-XP undergoes QR, providing the requisite logical syntax for interpretation. The LF-representation and corresponding interpretation were already introduced by ().

8 It has been suggested that there is an additional functional layer above DegP, which hosts measure phrases and more in synthetic comparatives (cm more distant). On this view, the [E]-feature is hosted by the head of this projection, and ellipsis affects the complement, instead of the specifier of DegP. Crucially, adopting the more fine-grained structure ensures uniformity with Merchant () and subsequent work, according to which [E]-features always affect the complement domain. 9 Multiple LF-copies are also attested in other domains. On the raising analysis of relative clauses, the higher copy can but, crucially, does not have to be ignored by the computation. The same rationale is also underlying the explanation of disjoint reference effects with A’-movement and QR. For instance, (i.a) is assigned the LF in (ii), in which both copies are interpreted (see Fox ; Sauerland  for details):

(i)

a. *Which picture of John1 did he1 like best? b. *He1 liked every picture of John1.

(ii)

[[which picture of John1] λ2 [did he1 like best]

 



Independent support for the phrase structure in () comes from an observation due to Bresnan (), who notes that in attributive phrasal comparatives, the size of CD systematically correlates with the position of the degree complement. Only (a) admits a sensible reading, while (b) entails that Sally is a man, resulting in a category mistake. This indicates that the CD-site comprises of the AP in (a), but includes the NP in (b): ()

a. Mary met a man [DegP [AP older] [than-XP than Sally ]]. b. #Mary met an [DegP [older man] [than-XP than Sally ]].

A successful analysis of the contrast above hinges on finding a way to secure that the size of CD is functionally determined by syntactic properties such that the CD-site corresponds to the sister node of the than-XP as expressed by Gawron’s (: ) ‘Hypothesis A’: ()

Hypothesis A The scope of comparison is exactly the semantics of the sister of the than‑phrase.

On the raising analysis, Hypothesis A follows directly from the c-command condition on movement. By contrast, the matching account, which posits the structure in (a), repeated below, either has to stipulate that identity in comparatives ignores the difference between ‘ AP’ and ‘Deg° AP’, or needs to assume that the matching operation reaches into the external head, targeting the complement of Deg°: ()

a. [DP [DP  AP] [than-XP . . . . . . .]]

If the latter option is chosen, matching is no longer defined in terms of c-command. But the raising analysis comes at a cost, too. Similar to what was observed with the matching account, an independent yet inherent feature of the head-raising mechanism renders the analysis incompatible with the Late Merge hypothesis. The raising account of CD presupposes that the than-XP is merged early, as a sister of , so as to allow movement out of the than-XP into specDegP. But this is clearly inconsistent with the assumption that the than-XP joins the derivation only at a later stage, once  has reached its scope position. Thus, unlike in the matching analysis, the complication is created by conflicting criteria on the sequencing of derivational steps, and not by a conflict of constituency.10 One possible response to this challenge consists in assuming that than-XPs are merged cyclically after all, but fail to create movement copies due to independent factors which are operative in comparatives, but not in relative clauses. These factors might in turn be reduced to logical properties that distinguish comparatives from other clause types. More precisely, Bhatt and Pancheva (: –) point out that LF-configurations in which the than-XP is merged cyclically lead to logical contradictions. Together with a principle that prohibits certain, formally definable, representations with trivial truth conditions (L-triviality; Gajewski ), this would be sufficient to guarantee that QR of the unit 10 Note incidentally that the same complication shows up in relative clauses, yet with diametrically opposite effects. Extraposed, i.e. Late Merged, relative clauses actually resist a raising construal, as witnessed by the absence of reconstruction effects, among others (see Bhatt  for an overview).





[ than-XP] leaves a degree variable, instead of a copy. Moreover, as contradictions are limited to comparatives, Late Merge of relative clauses would, as desired, still be possible. While this looks like a promising move for rendering the raising analysis of CD consistent with the Ellipsis Scope Generalization, a more careful investigation of the problem has to await another occasion.

.. Semantic theories of CD and diagnostics for structure On semantic theories of CD, the unpronounced gradable property is either λ-converted into the degree clause (Kennedy , ; see ()) or treated as an instance of discourse anaphora (Lerner and Pinkal ; see also Klein ; Larson b; Heim ). By nature, these accounts require a syntax in which the external head of the comparative is already present at the point where CD is interpreted. In principle, this configuration can be provided both by a raising syntax or a matching derivation. While the specific mechanism implicated in restoring the CD-site is not of direct relevance for our present concerns, there is an important property discriminating between the semantic analysis and its other competitors. Both the deletion and the raising analysis have in common that the CD-site contains hidden syntactic structure that is predicted to react to standard diagnostics for the presence of silent nodes. By contrast, semantic analyses (Lerner and Pinkal ; Kennedy ) generate the expectation that the CD-site is syntactically inert. Various tests document that CD does indeed embed unpronounced structure, in compliance with the syntactic approach (Lechner ). To begin with, names inside the CD-site induce Principle C violations (a), indicating that the degree complement contains a silent copy of the gradable AP and its complement. CD behaves in this respect just like VP-ellipsis (b). Moreover, CD also parallels VP-ellipsis in that clausal embedding of the name ameliorates the disjoint reference effect, as shown by (c) and (d) (vehicle change; see Fiengo and May ; Safir , among others): ()

a. b. c. d.

*Mary is prouder of John1 than he1 is . *Mary is proud of John1 and he1 is , too. Mary is prouder of John1 than he1 believes that I am . Mary is proud of Johni and he1 believes that I am , too.

Second, the CD-site can host reflexives and reciprocals, which are commonly assigned sloppy readings.11 For theories which express the conditions on anaphor licensing structurally, this entails that the abstract representation of the degree complement has to include unpronounced occurrences of the bound variables: ()

a. Mary1 is prouder of herself1 than Sally2 is . b. The girls1 are prouder of each other1 than the boys2 are .

Next, extraction from the CD-site is subject to the CSC, a condition which is usually held to be verified at the syntactic level of LF (Fox ). The contrast in () receives a natural 11

For exceptional strict readings see Hestvik (), Kennedy and Lidz (), and McKillen ().

 



explanation under the syntactic account, but is difficult to reconcile with the tenets of semantic theories, which lack the means to express conditions on LF-representations. ()

a. a person who1 Mary is [more proud of t1] than Peter is b. *a person who1 Mary is [more proud of t1] than Peter is (of) John2 c. *a person who1 Mary is [more proud of John] than Peter is

Finally, the fact that the silenced AP in (a) can contain a trace whose binder is located outside the elliptical node provides further, independent confirmation that the CD-site contains hidden structure. Thus, it seems very likely that CD is an operation that either consists in movement and deletion or head raising. To recapitulate, while the ellipsis process that renders unpronounced the degree predicate inside the comparative displays all the signatures of a syntactic operation, the exact nature of CD is still somewhat elusive. On the one hand, it was seen that some of the core properties of CD are successfully captured by the matching analysis. On the other hand, the raising account avoids complications regarding the identity condition on the CD-site and its antecedent, and offers at least the prospect of an explanation for Late Merge effects. After this brief, necessarily selective survey of theories of the obligatory process of CD, the discussion to follow elaborates on optional ellipsis operations in comparatives. For more information on CD, see Pancheva (), Morzycki (), and Lechner and Corver ().

. E  — C E

.................................................................................................................................. It has been observed at least since Hankamer (b) that various kinds of reduction processes that are attested in coordinate structures can also be found in comparatives. () provides a representative sample from coordinate contexts, and (), repeated from above, examples of arguably the same phenomena in comparatives (nothing bears on the particular parses): () a. b. c. d. e. f.

Ann liked Berlin and Cleo Dubai. (Gapping) Ann1 [t1 liked Berlin and Dubai]. (ATB movement and Gapping) Ann liked Berlin and Cleo , too. (Stripping/Gapping) Ann liked Berlin and Cleo did , too. (VP-Ellipsis) Ann liked Berlin and Cleo did Dubai. (Pseudogapping) Ann liked and Cleo disliked the place. (RNR)

()

Ann liked Berlin more than Cleo Dubai. (Gapping) Ann1 [t1 liked Berlin more than Dubai]. (ATB movement and Gapping) Ann liked Berlin more than Cleo . (Stripping/Gapping) Ann liked Berlin more than Cleo did . (VP-Ellipsis) Ann liked Berlin more than Cleo did Dubai. (Pseudogapping) More people liked than disliked the place. (RNR)

a. b. c. d. e. f.





Historically, three questions have been central to the study of ellipsis in comparatives. First, can all manifestations of ellipsis in comparatives be subsumed under independently attested ellipsis operations, or are there processes that are only operative in degree constructions (Comparative Ellipsis; Bresnan ; Pinkham )? Second, are all seemingly elliptical comparatives actually elliptical? More specifically, do all phrasal comparatives (PC), exemplified by (b) and (c), embed unpronounced structure, or are there also base-generated PCs? Third, it has been asked how phenomena that are generally confined to coordinate structures are licensed in the subordinate environments provided by comparatives. The present section reports findings relevant to the first and third domain, while the dispute about PCs will be taken up in section .. Before proceeding, a brief methodological note is in order. It is important to keep in mind that degree complements are (unless the remnant is a measure phrase) elliptical as a result of CD. When lining up comparatives and coordinations, the CD-site should therefore, all things being equal, be phonologically filled in the corresponding coordinate structures. Thus, the analogous coordinate structure of (a) is (b), and not (c). () a. [Ann gave him more expensive presents] than [he gave her ]. b. [Ann gave him expensive presents] and [he gave her expensive presents]. c. *[Ann gave him expensive presents] and [he gave her]. Throughout, I will assume that the right-hand-side bracketed unit in (a) corresponds to the second conjunct in a coordination, with than usurping the role of a syntactic coordinator.

.. Ellipsis in comparative and coordinate structures While it is uncontroversial that the reduction processes in () have the appearance of standard ellipses, demonstrating that the comparatives in () and their coordinate analogues () actually share a common derivational history turns out to be a less trivial task. In what follows, I will present reasons to believe that this strong hypothesis is correct at least for gapping, ATB movement, and RNR (Pinkham ; Hendriks ; Moltmann ; Lechner ).

... Gapping Gapping is a deletion operation exclusively found in coordinate structures ((a) vs (b)) which targets a proper subset of verbs in non-initial conjuncts ((c); Johnson, this volume, for an overview). ()

a. Lisa visited Millhouse and Otto Bart. b. *Lisa visited Millhouse although/after/because Otto Bart. c. *Lisa Millhouse and Otto visited Bart.

Six signature characteristics which are known to define the behavior of coordinate gapping are also attested in comparatives (Lechner ):

 



(i) Gapping operates progressively only, in coordinations as well as in comparatives: ()

a. Lisa visited Millhouse more often than Otto Bart. b. *Lisa Millhouse more often than Otto visited Bart.

()

a. More people visited Millhouse than Bart. b. *More people Millhouse than visited Bart.

(ii) Gapping cannot affect infinitives to the exclusion of finite verbs. This is shown for coordinate structures in (a) and adverbial, subject, and object comparatives in (b–d). (For expository convenience, the discussion will focus on a single exponent of the paradigm from now on.) ()

a. b. c. d.

*Lisa tried to visit Millhouse and Otto promised Bart. *Lisa tried to visit Millhouse more often than Otto promised Bart. *More people tried to visit Millhouse than promised Bart. *Lisa tried to visit more people than Otto promised .

(iii) Just like gapping is prohibited from targeting finite, embedded clauses in coordinate structures (a), it cannot do so in comparatives (b), either: ()

a. *Lisa said that some visited Millhouse and Otto claimed that others Bart. b. *Lisa said that some visited Millhouse more often than Otto claimed that others Bart.

(iv) In both constructions, the results are strongly degraded if the gap includes finite sentence boundaries: () a. ??Lisa said that some visited Millhouse and Otto Bart. b. ??Lisa said that some visited Millhouse more often than Otto Bart. (v) In V-languages such as German or Dutch, gapping must not operate across overt complementizers (Hendriks ): () a. Ich glaube daß mehr Leute das Buch lesen als (*daß) den ArtikelACC I believe that more people the book read than that the article . read ‘I believe that more people are reading the book than the article.’ b. Ich glaube daß Hans das Buch liest und (*daß) Maria den ArtikelACC I believe that Hans the book reads and that Mary the article . read. ‘I believe that John is reading the book and Mary the article.’





(vi) In V-languages, it is only marginally possible to gap clause-final, finite auxiliaries to the exclusion of the main predicate (). The results improve drastically if the finite verb undergoes V (): () a. *?weil viele Leute Sue besucht haben und einige Otto eingeladen since many people Sue visited have and some Otto invited

have ‘since many people visited Sue and some invited Otto’ b. *?weil mehr Leute Sue besucht haben als Otto eingeladen since more people Sue visited have than Otto invited have ‘since more people visited Sue than invited Otto’ () a. Gestern haben viele Leute Sue besucht und einige Otto yesterday have many people Sue visited and some have Otto eingeladen. invited ‘Yesterday, many people visited Sue and some invited Otto.’ b. Gestern haben mehr Leute Sue besucht als Otto eingeladen . yesterday have more people Sue visited than Otto invited have ‘Yesterday, more people visited Sue than invited Otto.’

... Right-Node Raising Unlike gapping, RNR removes material from non-final constituents and targets strings that contiguously extend from the right edge of the first conjunct (), largely ignoring locality conditions such as the Right Roof Constraint ((b); Hartmann ; Sabbagh ). RNR treats comparatives and coordinations alike (see Lechner  for further details): ()

a. b. c. d.

Lisa tried to visit and Otto promised to invite Bart. Lisa tried and Otto promised to visit Bart. More people tried to visit than promised to invite Bart. More people tried to than promised to visit Bart.

()

a. Some people told us that Clinton would win and others tried to convince us that she would lose the election. b. More people told us that Clinton would win than had tried to convince us that she would lose the election.

In OV-languages, RNR overrides the prohibition on auxiliary ellipsis seen in (). Thus, finite auxiliaries can be RNRed (), but not gapped (). () a. weil since

viele many

Leute people

eine a

Zeitung newspaper

gekauft bought

have

und and

 



einige ein Buch gelesen haben some a book read have ‘since many people bought a newspaper and some read a book’ b. weil mehr Leute eine Zeitung gekauft since more people a newspaper bought have ein Buch gelesen haben a book read have ‘since more people bought a newspaper than read a book’

als than

() a. weil viele Leute eine Zeitung und since many people a newspaper bought have and einige ein Buch gekauft haben some a book bought haben ‘since many people bought a newspaper and some bought a book’ b. weil mehr Leute eine Zeitung als ein Buch since more people a newspaper bought have than a book gekauft haben bought have ‘since more people bought a newspaper than bought a book’ Finally, comparatives, just like coordinations, tolerate RNR in certain non-coordinate contexts. The status of RNR in marginally acceptable subject relative clauses ((a)/(a); Hudson ) parallels that of subject comparatives ((b)/(b); Lechner ). In (b), RNR generates a PC. () a. ?weil viele Leute die ein Buch gekauft eine Zeitung gelesen since many people who a book bought have a newspaper read haben have ‘since many people who bought a book read a newspaper’ b. ?weil mehr Leute als ein Buch gekauft eine Zeitung gelesen since more people than a book bought have a newspaper read haben have ‘since more people read a newspaper than bought a book’ () a. ?weil viele Leute die ein Buch auch eine Zeitung since many people who a book bought have also a newspaper gekauft haben bought have ‘since many people who bought a book also bought a newspaper’





b. ?weil mehr Leute als ein Buch eine Zeitung since more people than a book bought have a newspaper gekauft haben bought have ‘since more people bought a newspaper than a book’ These findings corroborate the hypothesis that comparatives can be modulated by RNR.

... Across-the-board movement Direct evidence for ATB movement in comparatives comes from examples like (a), in which relativization reaches both into the matrix clause and the than-XP, suggesting an analysis that assigns to (a) a representation similar to that of the coordinate structure in (b):12 ()

a. the book OP1 [John read t1 more often] than [Bill read t1] b. the book OP1 [John read t1] and [Bill read t1]

() adds a semantic argument for the existence of ATB movement in comparatives. The missing indefinites in the coordination (a) and the comparative (b), respectively, are interpreted as variables bound by the existential subject. Given that there is no ‘small conjunct’ analysis for comparatives, which would equally be able to derive the intended reading (see Hirsch  for recent implementation), it is hard to escape the inference that the subject has been removed by ATB movement: () a. Someone1 b. Someone1

[t1 bought books for the library] and [ magazines for the school]. [t1 bought more books for the library] than [ for the school].

ATB is also implicated in the curious exception to the ban on auxiliary gapping in (). The puzzle disappears once it is recognized that the verb (haben) is not gapped but has reached its surface position by ATB verb-second movement (ATB-V), as in () (Lechner : .): () Gestern haben1 [mehr Leute Sue besucht t1] als [Otto eingeladen t1]. yesterday have more people Sue visited than Otto invited ‘Yesterday, more people visited Sue than invited Otto.’ That the deletion process at work in comparatives targets a genuine coordinate structure becomes apparent by inspecting the effects of the Coordinate Structure Constraint (CSC) in 12 A reviewer points out that the well-formedness of (i) suggests that (a) does not implicate ATB-movement, but is a parasitic gap (Hendriks ). While this is a viable analytical option, (i) is also compatible with the view that comparative coordination is optional and that the degree clause in (i) is subordinated. See discussion of German V in () and () for evidence in support of this view.

(i)

the book OP1 [John read t1 more often] than [Bill read it1]

 



paradigm (). (a) illustrates a legitimate combination of two ATB movements, viz. V and subject raising. No conflict arises, because both operations presuppose a coordinate parse. But whenever one of the two movements skips a conjunct, the results drastically degrade. In (b/c), ATB subject raising to specTP establishes a coordinate structure, such that asymmetric V is bound to incur a violation of the CSC (traces that do not abide by the CSC are marked by ☛). Conversely, (d/e) are blocked by the CSC due to illegitimate asymmetric subject movement out of a coordination established by ATB-V: () a. Zweifellos hat1 Hans2 [t2 mehr Brote bestellt t1] als [t2 gegessen t1]. doubtless has Hans more sandwiches ordered than eaten b. *Zweifellos bestellte1 Hans2 [t2 mehr Brote ☛ t1] als [t2 gegessen hat]. doubtless ordered Hans more sandwiches than eaten has c. *Zweifellos hat1 Hans2 [t2 mehr Brote bestellte] als [t2 gegessen ☛ t1]. doubtless has Hans more sandwiches ordered than eaten d. *Zweifellos hat1 Hans2 [☛ t2 mehr Brote bestellt t1] als [seine doubtless has Hans more sandwiches ordered than his Mutter gegessen t1]. mother eaten e. *Zweifellos hat1 Hans2 [seine Mutter mehr Brote gegessen t1] doubtless has Hans his mother more sandwiches eaten als [☛ t2 bestellt t1]. than ordered intended: ‘Hans undoubtedly ordered more sandwiches than he/his mother ate.’ The effects visible in () are replicated for ATB-V and object extraction in (): () a. Diesen Film2 haben1 wohl [mehr Leute t2 bestellt t1] als [t2 auch gesehen t1]. this movie have particle more people ordered than also seen ‘More people ordered the movie than saw the movie.’ b. *Diesen this [t2 auch also

Film2 movie gesehen seen

haben1 wohl [mehr Leute t2 bestellt ☛ t1] als have particle more people ordered than haben]. have

c. *Diesen Film2 haben1 wohl [mehr Leute t2 bestellten]als [t2 auchgesehen ☛ t1]. this movie have particle morepeople ordered than also seen d. *Dieses this Film movie

Buch2 haben1 wohl [mehr Leute ☛ t2 bestellt t1] als [den book have particle more people ordered than the gesehen t1]. seen

e. *Diesen Film2 haben1 wohl [mehr Leute das Buch bestellt t1] als [☛ t2 this movie have particle more people the book ordered than gesehen t1]. seen intended: ‘More people ordered the book than saw the movie.’





Finally, what () demonstrates is that the coordinate parse for comparatives is, unless forced by independent factors such as ATB movement, optional. In the absence of other ATB dependencies, the than-XP can be parsed into a position subordinate to the matrix clause, which enables syntactic operations to target the two clauses individually, accounting for the availability of asymmetric V-movement of the finite auxiliary hat (see Lechner , ,  for details): ()

Zweifellos [hat1 Hans2 t2 mehr Brote gegessen t1] als [seine doubtless has Hans more sandwiches eaten than his Mutter bestellt hat]. mother ordered has ‘John undoubtedly ate more sandwiches than his mother ordered.’

In sum, the standard diagnostic for coordination (CSC) yields solid evidence that comparatives which include ATB movement establish a coordinate structure. This is expected if the symmetric deletion processes observed in (b), (), (a), and (a) actually consist in ATB movement.

... Other ellipsis operations in comparatives Additional ellipsis operations that have been observed to target comparatives include    ((b); Aelbrecht ), and N C A (NCA; ()). Just like NCA in coordinate structures ((); Jacobson c), ellipsis is possible with control, but not with raising predicates: ()

a. She wanted to read the book but she couldn’t. b. She wanted to read more books than she could.

()

a. More people offered to support Clinton than refused/tried/remembered. b. *More people offered to support Clinton than seemed/happened/turned out/tended.

()

a. While some offered to support Clinton, others refused/tried/remembered. b. *While some offered to support Clinton, others seemed/happened/turned out/ tended.

In German, NCA co-occurs with the expletive es, a requirement that carries over to comparatives: ()

a. Maria Mary

versuchte tried

uns us

anzurufen, call

b. Maria versuchte uns öfter Mary tried us more.often

und and

Peter Peter

versuchte tried

*(es) it

auch. too

anzurufen, als Peter *(es) versuchte. call than Peter it tried

Interestingly, comparatives permit    also with verbs like think and seem, which usually do not license NCA (Kennedy and Merchant b; Pancheva , attributing () to Irene Heim):

 



()

a. Mary grew taller than I thought . b. *Mary grew tall even though I didn’t think .

()

a. We are more vulnerable than we seem . b. *We seem (to be) divided but I don’t think that our enemies seem .

The effect, yet to be accounted for, appears to be restricted to stative predicates: ()

a. *We are running faster than we seem . b. *Mary won more often than she seemed .

An additional puzzle is raised by German sentential argument ellipsis in (). The contrast in () demonstrates that the deletion may consist of a discontinuous string made up of a finite auxiliary (wurde) and the complement clause, provided that the stranded participle is embedded in a passive clause. (The active version (b) is well-formed with a finite auxiliary): () a. Dies wird sich schwieriger umsetzen lassen als this will self more.difficult to.realize let than . was that this self realized let will ‘This will be more difficult to realize than thought.’ b. *Dies wird sich schwieriger umsetzen lassen als this will self more.difficult to.realize let than . had that this self realized let will

gedacht thought

ich I

gedacht thought

To summarize, the observations collected in the last four subsections provide strong support for the claim that comparatives can be targeted by the same kinds of deletion operations—specifically gapping, RNR, and ATB movement—that are operative in coordinate structures. Sentential argument ellipsis, which is attested in comparatives only, is a to date poorly understood exception to this generalization. Note, incidentally, that in all the examples examined up to now, ellipsis and ATB movement do not discriminate between PCs and elliptical comparatives with more than one remnant. This parallelism will become relevant again in the discussion of different approaches towards PCs in section ..

.. Coordination vs subordination Even though the question why comparatives are able to emulate the behavior of coordinations is largely unexplored, there are a few speculations in the literature. In principle, three factors have been identified to influence the availability of a coordinate parse for comparatives: linearization, non-containment, and morphosyntactic properties of the standard marker (than in English). The contrast between (a/b) and (c/d) illustrates that gapping and ATB movement in comparatives is contingent upon extraposition of the than-XP, suggesting that the degree





complement must, just like regular non-initial conjuncts, be properly linearized to the right of the matrix clause. Notably, (c) can only be blocked by linearization conditions on the whole conjunct since (c) satisfies the requirement that the antecedent precede the gap: () a. [Someone visited more people at Christmas] than [Otto on his birthday]. b. Someone1 [t1 visited more people at Christmas] than [t1 on his birthday]. c. *[Someone visited more people than [Otto on his birthday] at Christmas]. d. *Someone1 [t1 visited more people than [t1 on his birthday] at Christmas]. Proper linearization can be effected in various ways: by extraposition of the than-XP (Pinkham ; Lechner ); by Late Merge of the than-XP subsequent to covert rightward movement of  (which is another strategy to model extraposition); by overt QR of the DegGQ; or by using enriched phrase structures, such as multidimensional trees (Moltmann ) or multidominance (Sabbagh ; Gračanin‑Yüksek ). Second, elliptical comparatives in which the than-XP contains hidden structure are constellations of antecedent contained deletion (ACD; Wold ). The than-XP accordingly has to attach to a node c-commanding the ellipsis site at the point in the derivation where ellipsis is resolved. Importantly, while containment is related to linearization, the two criteria do not define the same class of expressions. In the structure (a), the degree clause is properly linearized, but still contained inside the (VP)-node which embeds the antecedent. Hence, ACD-resolution requires re-bracketing, for instance as in (b): ()

a. John met [more people than [Sally ]] b. [[John met more people] than [Sally ]]

As with linearization, the proper factorization can be achieved by movement of the thanXP or the degree quantifier. Which of these options is best suited to account for the coordinate behavior of comparatives is not known at the moment. A third factor apart from linearization and containment implicated in coordination formation is the lexical inventory of standard markers provided by a language. Greek, Polish, and Serbo-Croatian, for instance, distinguish between two versions of than (see ()), only one of which licenses ellipsis. Conversely, Hankamer (b) suggested that English than is ambiguous between a coordinating complementizer and a preposition. A modern implementation of this line of thought is Alrenga et al. (). In sum, the fact that comparatives replicate the restrictions on deletion operations typically attested in coordinate structures makes it seem likely that the processes at work are the same. Naturally, it was not possible in the confines of the presentation to demonstrate that the analysis is complete. Further evidence in support of the claim that all deletion in comparatives is reducible to the union of the operations exclusively targeting coordinations (gapping, ATB movement, RNR) and more liberal deletion processes also attested in subordinate contexts like VP-ellipsis or pseudogapping (Kennedy and Merchant a) is collected in Hendriks () and Lechner ().13

13

On ellipsis in comparatives see also Napoli (a); Moltmann (); Hendriks (); Pinkham (); Lin (), among many others. Complications for Lechner () are noted in Osborne ().

 



. P 

.................................................................................................................................. From the early s on, there has been an intense debate about the proper treatment of PCs, illustrated in (). ()

a. Ann is taller than Bill. b. Ann met a taller man than Bill.

Scholars have noticed that PCs meet certain, but not all criteria typically associated with elliptical constructions (Hankamer b), which led to the formation of two groups of approaches: the D A (DA), which analyzes PCs as base-generated PPs headed by a prepositional version of than, and the R A (RA; terminology by Bhatt and Takahashi ), adherents of which hold that PCs embed hidden syntactic structure.14 In general, the standard heuristics for detecting the effects of RA has been that PC formation by ellipsis affords phenomenologically richer structures than by DA. Over the last fifteen years, a consensus has emerged that the choice between DA and RA is not universal, but subject to typological variation, such that some languages employ DA, while others opt for RA. More recent results indicate that both strategies might be employed even within a single language. This final section traces synoptically the most important steps of this still ongoing controversy, which not only affects the analysis of comparatives and ellipsis, but also has broader repercussions for the theory of the lexicon, language typology, and learnability, among others. The remainder of this section falls into three parts. Reporting findings from classical studies of the s and s, as well as two more recent studies, section .. lists in a synoptic form arguments in support for DA, accompanied by a brief critical assessment. Next, section .. presents the compositional semantics for PCs, turning from there to evidence in support of RA in ... Before proceeding, a final note on terminology: The DP following than will be referred to as the , while the category which takes up the same grammatical function as the remnant in the main clause is the . In (a), for example, Bill serves as the remnant and Ann is the correlate.

.. Arguments for DA Arguments for DA, and thereby for the claim that at least some PCs are base-generated and lack hidden syntactic structure, include observations about the morphology of the standard marker or the remnant, the syntax of PCs, and interpretive properties exclusively found with PCs. In the remainder of this section, I will list arguments in support of DA culled from the literature.

On the debate surrounding DA vs RA see also Hankamer (, b), Bresnan (), Brame (), Pinkham (), Napoli (a), Heim (), McCawley (), Kennedy (, ), Xiang (), Lechner (, ), Pancheva (, ), Beck et al. (), Lin (), Merchant (), and Sudo (), among others. 14





(i) English PCs with subject accusative remnants (a) lack a clausal source (b), and have therefore been argued to be base-generated (Hankamer b; Napoli a). But accusative subjects are also found in elliptical conjunctions (c) and fragment answers (d); Merchant a), suggesting that the contrast (a) vs (b) does not reflect the clausal vs phrasal distinction, but is an artifact of whatever mechanism is responsible for the distribution of default case (Lechner ; Merchant b). ()

a. b. c. d.

Ann is taller than me/*I. Ann is taller than *me am/I am. Ann is eager to see the movie, and me (*is) too. Who’s watered the plants? Me/*I.

(ii) Extraction and preposition stranding is possible from PCs only ((a); Hankamer b). At first sight, the best analysis of (a) seems to reside with the assumption that the remnant is a base-generated prepositional complement which has been separated from its head under preposition stranding: ()

a. Who are you taller than t? b. *Who are you taller than is t?

There is another conceivable explanation of the phenomenon, though, which treats (a) as an instance of island repair under ellipsis (Merchant a, b). One possible implementation of this idea involves the two assumptions that degree complements are inherently islands for extraction, and that the ellipsis in (a), but not in (b), is large enough to include all offending intermediate traces. As a result, the representation of (a), relevant parts of which are given in (a), does not contain any illegitimate syntactic objects, while (b), where ellipsis affects a node lower than TP (b), embeds at least one non-locally bound trace: ()

a. Who1 are you taller than b. *Who1 are you taller than [CP *t1 [TP t1 is ]]

On this conception, (a) could be elliptical, after all. Naturally, various challenges remain, most prominently the observation that ellipsis does not amend island violations in comparatives (a) in the same way it does under sluicing (b): () a. *More people live in the country that Putin governs than Obama1 . b. Ben wants to live in a state that is governed by a Democrat, but I can’t remember which1 . However, given that as to date, there is no satisfactory account for the contrast in (), the island repair analysis of () cannot be excluded as a plausible analytical alternative to DA. (iii) Reflexives can be externally licensed in PCs, but not in clausal degree complements, suggesting that PCs lack a clause boundary, as predicted by DA (Hankamer b; Brame ):

  ()



a. Nobody is taller than himself. b. *Nobody is taller than himself is.

Again, an alternative interpretation of the data exists, though, which is equally compatible with RA. It is also possible to treat (a) as a small clause isomorphic to (a). The smallclause analysis has the added benefit that it provides a reason for why the subject remnant in (a) surfaces with accusative case: it is a small-clause subject, just like the embedded subject of (b): () a. Nobody considers [small clause himself tall]. b. Ann considers [small clause me tall].

(Lechner : ; Pancheva )

That clausal degree complements do not always have to include functional heads but also come as small clauses was already seen in (), repeated below as (). In the intended readings, the remnant is followed by a single adjectival (a) or nominal (b) projection made up by the CD-site: ()

a. Mary met a man older than Sally . b. Mary met an older man than Bill .

(iv) On the DA, the PC remnant is a prepositional complement. Since prepositions generally select a single, nominal complement, one is led to expect that genuinely phrasal comparatives admit only a single remnant (single remnant restriction) and resist prepositional remnants. Although these predictions are not confirmed for English, which generally treats PCs as reduced clauses (see section .. for justification), they accurately characterize languages such as Greek, Polish, Russian, and Serbo-Croatian, where base-generated PCs and clausal comparatives are morphologically distinguished by the shape of the standard marker (Pancheva ; Merchant ). To exemplify, Greek PCs formed with the standard marker apo display all properties typically associated with the DA, while the particle ap’oti introduces elliptical, clausal comparatives. This division of labor is reflected by the fact that only apo-PCs abide by the single remnant restriction () and the prohibition on prepositional remnants ((); Merchant : , ()–()): () a. Perisoteri more [PP me with

anthropi milisan me ton Gianni tin Kyriaki ap’oti people spoke with the Giannis the Sunday than.clausal ton Anesti] [NP to Savato]. the Anesti the Saturday

b. *Perisoteri anthropi milisan me ton Gianni tin Kyriaki apo more people spoke with the Giannis the Sunday than.phrasal [PP me ton Anesti] [NP to Savato]. with the Anestis the Saturday ‘More people spoke with Giannis on Sunday than with Anestis on Saturday.’ ()

a. Perisoteri anthropi menun stis IPA ap’oti [PP sti Rosia]. more people live in.the USA than.clausal in.the Russia



 b. *Perisoteri anthropi menun stis IPA apo [PP sti Rosia]. more people live in.the USA than.phrasal in.the Russia ‘More people live in the US than in Russia.’

Furthermore, as will be seen in some detail in .., whenever a language makes available a clausal and a prepositional strategy to form PCs, RA and DA properties cluster together. (v) In some languages, among them Greek and Polish, the formation of PCs is limited to certain grammatical functions. As observed by Pancheva (), in those Slavic languages that possess a designated standard marker for base-generated PCs, the comparative DP must not surface as a subject. The same restriction is operative in attributive comparatives in German (Lechner , ):15 () a. Der DirigentNOM unterstützte bessere Komponisten als der RegisseurNOM the conductor supported better composers than the director ‘The conductor supported better composers than the director.’ b. *Bessere Dirgenten unterstützten den KomponistenACC als den RegisseurACC. better conductors supported the composer than the director ‘Better conductors supported the composer than the director.’ In Slavic (as well as in German), the judgments are gradient and subject to speaker variation, but clearly distinguish between PCs that fall under the DA and those which are derived by RA. To recapitulate, even though morphosyntactic and syntactic criteria (i)–(iii) provide some suggestive initial evidence for the existence of base-generated PCs, none of the classical arguments for DA survives exposure to closer scrutiny; in all three cases, alternative ellipsis analyses for the phenomena were seen to exist, partially with broad empirical coverage. By contrast, criteria (iv) and (v), which entered the debate more recently, provide solid evidence for the hypothesis that at least in some languages, some PCs are basegenerated, as posited by DA.

.. Interpreting PCs PCs can be interpreted compositionally by defining a homomorphic lexical entry for the comparative degree head that applies to the remnant meaning first, and then combines with the gradable property and the subject denotation, in any order (Heim ). A version of this three-place version 3, adopted from Bhatt and Takahashi (), is given in (). On this view, first explicitly argued for in Hankamer (b) and recently defended in Bhatt and Takahashi (, henceforth BT), more is lexically ambiguous between ()

15

More generally, the restriction demands that in attributive degree comparatives, the correlate c‑commands the comparative DP. For unknown reasons, the effect disappears with numerical/amount DPs, the German equivalent of More conductors supported the composers than the directors is impeccable. See Lechner () for discussion.

 



and the two-place interpretation 2 ((), repeated below), which generates clausal comparatives:16 ()

⟦⟧

=

λxe.λA.λye.λd.A(d)(y) ⊂ λd.A(d)(x)

()

⟦⟧

=

λD.λD’.D ⊂ D’

(à la BT)

(Bhatt and Pancheva )

BT defend the conjecture that the choice between  and  is subject to crosslinguistic variation, in that some languages, among them English and German, treat PCs as reduced clauses formed with the help of , while others parse them as base-generated PPs introduced by  (Hindi-Urdu). Still others, such as Greek (Merchant ) and Polish (Pancheva ) employ both options. () provides a sample derivation for predicative PCs based on the individual comparison degree head , which proceeds without changes in the surface constituency: () ⟦[TP The ship is [DegP longer  [than-XP than the train]]⟧ = = ⟦⟧ (⟦the train⟧) (⟦long⟧) (⟦the ship⟧) = = λd.the train is d-long ⊂ λd.the ship is d-long By contrast, the overt representations of nominal attributive PCs such as (a) are not compositionally interpretable, given the meaning rule (), and need to be manipulated by two covert movement operations (Bhatt and Takahashi ; Kennedy ). First, the correlate Sue moves (① in (b) and the binder index is attached to the sister node of the moved category (Heim and Kratzer ). Then, the degree quantifier [DQP  than Ann] raises (②) and lands in-between Sue and the binder index of Sue, resulting in a configuration of P S (Barker ; Sternefeld ; Nissenbaum ; Beck and Sauerland ). As detailed by (c), Parasitic Scope supplies DegQP with a two-place relation between degrees and individuals.

a. Suecorrelate read a better poem than Annremnant.

()

b. LF:

Suecorrelate DegQP more3

than Annremnant

TP

λ2 λ1

vPt

①

t1 read a d2-good poem ② c. ⟦more3⟧ (⟦than Ann⟧) (⟦λ2 λ1 t1 read a d2-good poem⟧) (⟦Sue⟧) = = λd.Ann read a d-good poem ⊂ λd.Sue read a d-good poem

Kennedy (: ()) observes that  can be defined in terms , but not vice versa, creating the expectation that no language uses  only; this seems, as Kennedy notes, to be correct. 16





Note that, somewhat surprisingly, PCs in which the remnant consists of a measure phrase instead of an individual term are not handled by  but by the clausal version , at least if measure phrases are assumed to denote sets of degrees (Schwarzschild ). ()

John is taller than ft.

From this it follows that languages which only have access to  should not sanction measure phrase remnants, a prediction which still deserves further study. Turning to a first set of empirically falsifiable consequences of the DA, the specific implementation above entails the five claims in (): ()

a. b. c. d. e.

In PCs, the CD-site is resolved in the semantic component. PCs do not embed categories that react to diagnostics for syntactic structure. The correlate and the remnant of attributive PCs undergo covert movement. Quantificational remnants are assigned wide scope. Nominal attributive PCs are derived by Parasitic Scope.

Corollary (a) predicts that base-generated PCs should not display Principle C effects of the sort seen in (a). One is accordingly led to expect that the Greek PC (a) should contrast with its ill-formed clausal variant (b). This prognosis is not confirmed, though: () a. *I Maria ine pio perifani gia ton Gianni1 apo auton1, ACC. the Mary is more proud of the Gianni than.phrasal he b. *I Maria ine pio perifani gia ton Gianni1 ap’oti autos1, NOM. the Mary is more proud of the Gianni than.clausal he ‘Mary is more proud of Gianni than he.’ However, as pointed out by a reviewer, the coreference pattern (a) is still amenable to DA, given that at LF, the pronominal remnant (auton/‘heACC’) c-commands the predicate containing the name (Gianni), inducing a Principle C violation. Second, BT demonstrate that the second consequence of the DA, according to which the than-XP does not embed unpronounced structure, is empirically supported by the absence of disjoint reference effects in PCs in Hindi and Japanese, among others. Discussion of details will be postponed to section ... Third, as Heim () pointed out, covert movement of the remnant and the correlate should be subject to syntactic island constraints. That this is correct can be inferred from the observation that the degree head  in (a) is unable to scope out of a relative clause. Movement of  together with the remnant, which is required to generate the interpretable LF-representation (b), incurs a strong violation of the Complex NP Constraint:17 () a. *[DP Someone [CP who could answer fewer questions]] made a good impression on Bill than on Fred. 17

[Heim : , ()]

() showed that remnant movement in PCs differs from remnant movement under sluicing in that—for yet undisclosed reasons—only the latter grants amnesties to island violations.

 



b. *[IP on Bill [[ than on Fred] [λ2 λ1 [[DP someone [CP who could answer ☛ d2-many questions]] made a good impression on t1]]]] (a), also discussed by Heim, demonstrates that correlate raising is equally bounded by locality. As revealed by the pertaining LF (b), the correlate (the clarinet) must not bind a trace across a complex NP barrier: () a. ?*I spent more time with [DP a woman [CP that played the clarinet]] than the lute. b. [the clarinet [[ than the lute] (Heim : , ()) [λ2 λ1 I spent d2-much time with [DP a woman [CP that played ☛ t1]]]]] While supplying a prima facie argument in support of DA, it should not go unnoticed that the paradigms in () and () also find a natural explanation under the ellipsis analysis (Lechner ). This is so because gapping in () and () removes a finite CP, in violation of constraint (iv) of section ... (see ()). The pertinent underlying representations are given in (). () a. *Someone1 [CP who could answer fewer questions] made a good impression on Bill than on Fred. b. *I spent more time with a woman that played the clarinet than the lute. Fourth, BT observe that the DA requires quantificational remnants to leave their surface position, because  selects for an individual term as its first argument. More precisely, the remnant needs to QR and is therefore obligatorily assigned wide scope with respect to the correlate. () depicts the relevant relations schematically (for examples see BT, pp. –): ()

[QP3 [correlate [ [ than t3] [ [λ2 λ1 . . . d2-many. . . . ]]]]]

BT show that effects of this requirement are visible in languages that unambiguously treat PCs as base-generated, among them Japanese and Hindi-Urdu. A further prediction, which has not been attended to in the literature, is that in these languages, NPs that resist wide scope such as bare indefinites, should not be able to function as remnants. Thus, the structure corresponding to () should be ill-formed in Japanese and Hindi-Urdu.18 ()

Girls are more avid readers than boys.

Finally, the derivation of attributive PCs should reflect properties generally characteristic of Parasitic Scope constellations. Positive evidence to this effect is discussed in Lechner . The information accumulated so far lends plausibility to the idea that PCs also have a base-generated parse. Further substantiation for this claim will be presented in the 18

A reviewer notes that the example is acceptable in Hindi, but that bare nouns in that language behave like kind names. A more complete investigation of the issue has to await another occasion.





following section, which summarizes the results of RA, and traces current and recent developments which lead to a partial synthesis incorporating aspects of both approaches.

.. Arguments for RA Proponents of RA have recruited the following arguments in support of the claim that the degree complement of PCs is endowed with abstract syntactic representation. (i) The ellipsis analysis offers a natural explanation for the fact that in languages such as German, the case of the remnant matches that of the correlate (Heim ): () a. SieNOM zeigte ihrDAT mehr Bilder she showed her more pictures ‘She showed her more pictures than he did.’

als than

b. PeterNOM zeigte ihrDAT mehr Bilder als Peter showed her more pictures than ‘Peter showed more pictures to her than to him.’

erNOM. he ihmDAT. him

This is unexpected on DA in its present form, because the representations of DA do not supply the remnant with a case assigner. While it is possible to amend this shortcoming—for example, by relegating case assignment to LF after movement of the two DPs in configurations akin to ()—RA clearly offers a more concise explanation for case matching. ()

[correlateCASE-α [ than remnantCASE-α . . . [ . . . . ]]]

(ii) English PCs obligatorily extrapose (Pinkham : ; but see also Bhatt and Takahashi ). The DA lacks the means to express this linearization and/or containment condition, which excludes (b): () a. b. c. d.

More people than bought books bought newspapers. *More people than books bought newspapers. More people bought newspapers than bought books. More people bought newspapers than books.

(Pinkham )

(iii) If the degree clause contains hidden structure, one is led to expect that names embedded in the remnant should not be able to corefer with pronouns that c-command the correlate, as expressed by generalization () (Lechner ; improved examples and () from BT): ()

Every node that c‑commands the correlate also c‑commands the remnant.

() documents that this prediction is confirmed for English PCs. As the LFs in () reveal, the pronoun c-commands the name inside the degree complement in (a), but not in (b): ()

a. *More people introduced him3 to Sally than to Peter3’s sister. b. More people introduced Peter3 to Sally than to his3 sister.

  ()



a. *More people introduced him3 to Sally than to Peter3’s sister. b. More people introduced Peter3 to Sally than to his3 sister.

By contrast, the DA assigns to both sentences representations in which the name is free. Hence, the disjoint reference effect in (a) remains unaccounted for: ()

a. Sally1 [ than Peter3’s sister]2 [λ2 λ1 d2-many people introduced him3 to t1] b. Sally1 [ than to his3’s sister]2 [λ2 λ1 d2-many people introduced. Peter3 to t1]

BT observe that not all languages react to Principle C alike. In Hindi-Urdu, for one, structures isomorphic to (a) admit coreference (Bhatt and Takahashi : ()): () Atif-ne [Ravi-kii3 behen-kii foto]-se us-ko3 [Hindi-Urdu] Atif-E Ravi-G sister-G picture-than he-D Mohan-kii behen-kii foto zyaadaa baar dikhaa-ii. Mohan-G sister-G picture more times show-P ‘Atif showed Mohan’s sister’s picture to him3 more times than Ravi3’s sister’s picture.’ They take this to signal that English assigns PCs a clausal analysis and employs , while Hindi-Urdu parametrically has access to  only. This conjecture is, as BT show, supported by the systematic clustering of properties such as scope (see ()) and the single remnant condition, illustrated for Hindi-Urdu in (): () *Tina‑ne aaj [Pim kal-se] zyaadaa TinaERG today Pim yesterday‑than more ‘Tina read more books today than Pim yesterday.’

kitaabe books

parh-ĩ. readPfv.FPI.

(iv) The ellipsis analysis leads one to expect that PCs display the same characteristics as partially reduced comparatives, i.e., constructions in which the than-XP either contains more than one remnant or a remnant that is not nominal. As shown by Lechner (; see also section ..) this prediction is borne out for reduction languages such as German and English. (v) A well-defined subclass of PCs admit tense mismatches between the matrix clause and the reconstructed degree complement (McCawley ; Pinkham ; Lechner , ): ()

John will visit more friends than SueNOM. a. . . . than Sue will visit d-many friends b. . . . than Sue (has) visited d-many friends

(atemporal reading)

The existence of atemporal readings is not expected on DA, because  λ-converts an exact copy of the matrix predicate, including the temporal specification, into the than-XP. Specifically, DA generates the LF for () in (a), which is mapped into the meaning (b): () a. LF: John [[ than Sue] [λ2 λ1 [TP t1 will visit d2-many friends]]] b. ⟦⟧ (⟦Sue⟧) (⟦ [λ2 λ1 [TP t1 will visit d2-friends]⟧) (⟦John⟧) = = λx.λA.λy.λd.A(d)(y) ⊂ λd.A(d)(x)(Sue)(λ2 λ1 t1 will visit d2-friends)(John) = λd.Sue will visit d-many friends ⊂ λd.John will visit d-many friends





Evidently, the mechanics of DA do not provide the means to derive mismatches in temporal specification, indicating that PCs that admit atemporal readings are derived by ellipsis.19 Taken together with the findings from .., where it was concluded that arguments for base generation on closer scrutiny turn out to be equally compatible with RA, the five generalizations listed in the present section provide additional support for the existence of hidden structure in the representation of PCs. Moreover, there is also clear evidence for the competing base-generation approach. In light of these two conflicting sets of facts, a consensus has emerged that the model currently best suited to handle the full range of phenomena is synthetic, incorporating both RA and DA, subject to typological variation and possibly other parameters (Kennedy ).

. C

.................................................................................................................................. Unpronounced constituents in comparatives fall into three discrete classes, none of which, it was argued, requires designated syntactic or semantic mechanisms: (i) the silent degree variable created by empty operator movement; (ii) material affected by Comparative Deletion, which targets degree predicates (including the common noun in attributive comparatives) and consists, I have suggested, in a version of the raising operative known from NP-movement; and (iii) non-construction-specific ellipsis operations. The present chapter has set out to demonstrate that attempts at bringing this idealization closer to reality have been successful in many respects. Naturally, a number of questions regarding ellipsis in comparatives remain under debate at the point of writing, among them (i) how to rein in the zoo of possible meanings for the degree head  (with important consequences for delimiting the class of PCs that are not elliptical); (ii) the inner architecture of degree predicates (affecting the analysis of how much ellipsis is tolerated in amount and degree comparatives, respectively; Lechner ); (iii) and numerous issues related to the scope of the degree relation (Beck ).20

19 Interestingly, the availability of atemporal readings is syntactically restricted in similar ways as attributive nominal comparatives in German or base‑generated PCs in Slavic: in atemporal PCs, the correlate has to c‑command the comparative DP. See Lechner  for discussion. 20 To close with a puzzle observed by Larson (b), universals in subject position can scope over matrix clause negation in PCs but not in clausal comparatives:

(i) Joe didn’t score more than everyone. a. Joe scored more than not everyone. b. Joe didn’t score more than anyone. (ii) Joe didn’t score more than everyone did. a. Joe scored more than not everyone. b. *Joe didn’t score more than anyone.

  ......................................................................................................................

   ......................................................................................................................

 

. C  N C A

..................................................................................................................................

.. The basic data The phenomenon of Null Complement Anaphora (henceforth NCA) refers to the null complements of certain verbs as in sentences () and (). () a. When mother told him to clean up his room, Tommy refused Ø (from Shopen : ) b. When Mary said she was going to change careers, Anna agreed Ø c. Mary works late nights but her sister doesn’t approve Ø d. They couldn’t persuade anyone to collect the tickets, so Susan volunteered Ø e. I may not ever have read Finnegan’s Wake, but I’ve tried Ø (examples (d) and (e) from Pullum : ) () a. Statement: John is telling lies again b. Response: I agree Ø

(Grimshaw : )

The possibility of having null complements is only available to a certain group of verbs as we see from the ungrammaticality in ():1

1

These null complements differ from the intransitive use of verbs such as eat; see (i).

(i) John was hungry, so he ate __ The null complement of eat in (i) is not interpreted anaphorically as the null complements in () and () are. In addition, the null complement of eat in (i) has a very specific interpretation (i.e., John ate something edible, not anything at all). We will not be discussing these types of null complements here.





() a. *They couldn’t persuade anyone to collect the tickets, so Susan arranged Ø b. *I may not ever have read Finnegan’s Wake, but I’ve attempted Ø (from Pullum : )

.. The questions In this chapter we will look at the behavior of these null elements. What are their properties? Which predicates allow them and which ones do not and why? What syntactic category/ categories do they belong to, and what type/s of semantic interpretation can they take? In addition, we will discuss the theoretical questions that they raise. The main question that arises with the null elements to be analyzed in this chapter is how the grammar can account for them given that they are not pronounced but nonetheless interpreted. What is their structure, how are they represented, how are they interpreted? Are they the same type of null element as other elliptical constructions, like VP-ellipsis? When asking how the syntax accounts for the non-pronounced strings that are the focus of this chapter, there are a number of logical possibilities available as answers. I list these logical possibilities below: a. Nothing in the syntax is present in the position of the null complement. In other words, what you see is what you get.2 b. The null element is the result of a deletion operation (such as PF-deletion) and the full syntactic structure is available. c. The null element is not the result of a deletion operation. Instead, a null pro-form with no internal syntactic structure is present in the syntax. d. The null element is not the result of a deletion operation. Instead, a null pro-form with some/full internal syntactic structure is present in the syntax. e. Some other combination of the options mentioned above. Some version of the possibilities given above has been put forth in different proposals by various authors in the literature since the s. We will present and discuss these proposals in the next section of this chapter.

. P 

.................................................................................................................................. The null complements illustrated in () and () have been referred to in the literature by various names. Shopen (, ) calls this feature definite constituent ellipsis, Hankamer and Sag (), Grimshaw (), and Napoli (b) call it Null Complement Anaphora. Fillmore () calls it Definite Null Complement and Saeboe (, ) uses the term definite ellipsis. Huddleston and Pullum () use the label lexical verb complement ellipsis. In this section, we present each one of the different proposals. 2

This is the approach taken in Culicover and Jackendoff () for ellipsis.

  



.. Shopen (, ) Shopen (, ) called examples like the ones in () and () definite constituent ellipsis. He notes that some verbs allow definite constituent ellipsis and others do not. Whether a verb allows it is a lexical issue, i.e., that information must be encoded in the lexical entry of the verb. So, according to Shopen, the fact that we can say Henry explained but not Henry expected has to do with the lexical properties of the verbs explain and expect. Shopen notes that definite constituent ellipsis can be used either anaphorically or deictically. If used anaphorically, the null argument has an antecedent within the linguistic context as in (). ()

When mother told him to clean up his room, Tommy refused Ø

When used deictically—as in ()—there is no linguistic antecedent present in the discourse. The speaker “relies on knowledge shared with the hearer or the obviousness of the identity of the referent within the perceivable physical context for the utterance” (Shopen : ). () Context/Situation: Knowing that Mother had again told Tommy to clean up his room, two close acquaintances have the following dialogue: A: What happened this time? B: Tommy refused Ø Shopen (: ) In addition, Shopen (, ) claims that the grammar of sentences such as () is just what you see. In other words, he does not propose any syntactic structure in the position of the null complement. The null complement is not a consequence of a deletion transformation.

.. Hankamer and Sag () ... Deep vs surface anaphora Hankamer and Sag () put forward a mixed theory of anaphora. They propose that there are two types of anaphora, independently of whether they are null or not. They call them deep and surface anaphora. In their system, deep anaphors are not derived via an operation (or transformationally, in their terms) but are present in underlying representations, while surface anaphors are derived via an operation (or transformationally). Null Complement Anaphora is—in their system—a type of deep anaphor. According to Hankamer and Sag (: –), deep and surface anaphors differ in the following ways: ()

a. Deep anaphors allow a pragmatic antecedent while surface anaphors do not. b. Deep anaphors cannot contain an antecedent for a pronoun (Missing Antecedents Test), while surface anaphors can. c. Deep anaphors do not require strict syntactic parallelism with their antecedent, while surface anaphors do.





Given their typology, Hankamer and Sag () classify different anaphoric constructions as either deep anaphora or surface anaphora, as summarized below: ()

a. Deep anaphora: Do It Anaphora, Sentential It Anaphora, Null Complement Anaphora, One Pronominalization, and ordinary personal pronouns. b. Surface anaphora: VP Deletion, Sluicing, Stripping, Gapping, Conjunction Reduction and Do So Anaphora.

What Hankamer and Sag () call deep anaphors are anaphors that can get their interpretation either from a linguistic or a non-linguistic antecedent. Independently of one’s theoretical position and perspective on the details of Hankamer and Sag’s proposal, the gist behind Hankamer and Sag’s typology is still empirically accurate in the sense that all languages have anaphors that need linguistic antecedents and anaphors that can get their interpretation from the non-linguistic context. In that sense, any analysis of NCA has to take this fact into account, because NCA is an anaphor that can get its interpretation both from the discourse and from the non-linguistic context.3 In the following section, we will concentrate on how NCA behaves with respect to Hankamer and Sag’s () three main tests for distinguishing between deep and surface anaphors: whether an anaphor can have a pragmatic antecedent or not, whether it can host a missing antecedent, and whether it needs to have an antecedent that is syntactically parallel or not.

... Hankamer and Sag’s (1976) tests (i) Pragmatic antecedents Hankamer and Sag propose that anaphors split into two groups depending on whether they can take a non-linguistic antecedent or not. NCA and the Do It pro-form can take a pragmatic antecedent while VP-ellipsis cannot,4 as illustrated in (): ()

[Hankamer attempts to stuff a -inch ball through a -inch hoop] a. Sag: It’s not clear you’ll succeed Ø (Null Complement Anaphora) b. Sag: It’s not clear that you’ll be able to do it. (Do It pro-form) c. Sag: #It’s not clear that you’ll be able to __ (VP-ellipsis) (modified from Hankamer and Sag )

(ii) Missing antecedents Hankamer and Sag () argue that only surface anaphors exhibit the missing antecedent phenomenon previously discussed in Grinder and Postal () and Bresnan (). Deep anaphors cannot contain an antecedent for a pronoun that follows it, while surface 3

See also Sag and Hankamer () on further differences in interpretation between surface and deep anaphors. 4 There has been a lot of discussion in the literature as to whether it is really true that surface anaphors can never take pragmatic antecedents. Schachter (b) challenges Hankamer and Sag () and argues that VP-ellipsis in English can take pragmatic antecedents. However, Hankamer () challenges these counterarguments in a very convincing way. See also Pullum () and Miller and Pullum () who revisit this issue, and also Thompson () for discussion.

  



anaphors can. The data in () show that NCA and the Do It pro-form pattern together with respect to this test. On the other hand, VP-ellipsis behaves differently, in that it can contain the antecedent for a following pronoun and therefore patterns as a surface anaphor according to the typology proposed by Hankamer and Sag ().5 () a. *John didn’t want to give up his seat, so Peter volunteered Ø because it was too narrow for him anyway. (Null Complement Anaphora) b. *John didn’t want to give up his seat, so Peter did it because it was too narrow for him anyway. (Do It pro-form) c. John didn’t want to give up his seat so Peter did [VP give up his seat] because it was too narrow for him anyway. (VP-ellipsis) The idea here is that a VP-ellipsis site contains enough syntactic structure to contain an antecedent for the following pronoun while the pro-form Do It or an NCA-site do not.

(iii) (Non)-strict syntactic parallelism with the antecedents Hankamer and Sag’s () third test for distinguishing between deep and surface anaphora is to see whether the anaphor in question demands strict syntactic parallelism with its antecedent. Deep anaphors do not require such strict syntactic parallelism with their antecedent, as shown with the pro-form Do It and NCA in examples (b) and (c). In contrast, VP-ellipsis must have strict syntactic parallelism with its antecedent, as we see in (a). ()

The oats had to be taken to the bin a. *so Bill did (VP-ellipsis) b. so Bill did it (Do It pro-form) c. so Bill volunteered Ø (NCA)

(from Hankamer and Sag )

.. Grimshaw () Within the context of her discussion of categorical and semantic selection, Grimshaw (: –) discusses the phenomenon of NCA. She proposes that NCA has no syntactic representation at all. In terms of how to get the interpretation for NCA, Grimshaw proposes the Null Complement Rule, which is a discourse rule that applies after sentence grammar (along the lines of the VP Deletion Rule in Williams b): () Null Complement Rule: Copy well-formed sentential formulae of logical form6 into the complement position of the verb. 5

There is discussion in the literature regarding whether the Missing Antecedent Test is reliable. Hankamer and Sag (: ) recognize that the judgments are “admittedly delicate.” See Thompson’s (: ) discussion of some of the problems that this test has. Bresnan (: ) and Johnson (b:  n. ), among others, discuss further limitations of this test. See also Merchant’s (a: ) arguments against using the missing antecedent test as a diagnostic for ellipsis. 6 In other words, sentential expressions composed of a predicate and its arguments, with or without any associated sentential operators.





Grimshaw limits the rule to copying “sentential expressions” only, which means that a predicate alone or a term alone cannot be copied. This is shown in the unacceptable examples in (b) and (b): ()

a. Question: Did John leave? b. Response: *No, but Bill __ c. Response: No, but Bill left

()

a. Question: Did John buy a book? b. Response: *No, John sold __ c. Response: No, John sold a book

(Grimshaw : )

Equipped with the Null Complement Rule, Grimshaw can account for the fact that an NCA can have a concealed question as an antecedent. In () Grimshaw proposes that the concealed question ‘what time it was’ is copied onto the NCA, not the NP ‘the time’. ()

Mary asked me the time, so I inquired Ø (cp. *Mary asked me the time, so I inquired the time)

In addition, she observes that whatever restrictions are imposed on the semantic type of the complement of these verbs will be imposed on the null complements of these verbs. So, for example, the verb ‘know’ can take a complement that is semantically a question as in ‘I know who left’ or a proposition as in ‘I know that John is telling lies again’. When this verb selects an NCA, this NCA will be able to be interpreted either as a question or as a proposition, just as if the complement had been overtly expressed, as shown in the contrasts in () and (): ()

a. Question: Did John leave? /Who left? b. Answer: I don’t know Ø

() a. Statement: Guess what, John is telling lies again. b. Response: Oh, I didn’t know Ø With respect to which verbs can take NCA, Grimshaw (: ) observes that “it is very unlikely that the difference between the predicates allowing null complements and predicates which require filled complements is predictable.” So, in her system, the subcategorization frame of the verb will need to include that information.

.. Napoli (b) Napoli proposes that examples such as () and () should be analyzed as intransitive uses of the verb. ()

a. Are you coming? b. I refuse.

(Napoli b: )

   ()

a. Is he coming? b. I suppose.



(Napoli b: )

She argues that the representations in () with deletion or with a base-generated null proform are not the right analyses for NCA. ()

a. I refuse to come/ I suppose he’s coming b. I refuse [e]VP / I suppose [e]S

(Napoli b: )

Napoli argues that the representations in () are not possible representations for NCA. What you see in () and () is what you get. According to Napoli, there is nothing present in the syntax which will later be deleted, nor is there an empty element present at the base. One of the arguments that she gives against a deletion account and a basegenerated account is that these elements do not seem to obey the Backwards Anaphora Constraint (BAC), if they were either the product of ellipsis (of the VP-ellipsis type) or base-generated elements (as is the case with pronouns). Napoli (b) presents examples such as the one in (a) in favor of her position. However, Hankamer and Sag (: ) had already shown that NCA does in fact obey the Backward Anaphora Constraint as illustrated in their examples given in (b). () a. John succeeded __, even though he didn’t try to impress. (cf. Even though John didn’t try to impress, he succeeded __) (Napoli b: ) b. *I don’t approve __, though she’s old enough to drive a truck (cf. Though she’s old enough to drive a truck, I don’t approve __) (Hankamer and Sag : ) There seems to be disagreement with respect to the possibility of BAC with NCA and that would be something interesting to explore in future research. To summarize, Napoli takes verbs that select NCA to be the manifestation of intransitive uses of those verbs and proposes that the interpretation is done in the pragmatics. To analyze NCA as an intransitive use of a predicate is incompatible with the fact that NCA is not interpreted in a generic way as are null complements of verbs like eat when used in an intransitive way. Such an analysis cannot account for the anaphoric nature of NCA. This had already been pointed out by Hankamer and Sag (: , n. ) who show that an example such as the following: I play cards and shoot dice, and my wife doesn’t approve, can only mean that the speaker’s wife disapproves of the speaker playing cards and shooting dice, not of anything in general, which a view such as Napoli’s would predict to be possible.

.. Fillmore () Fillmore () distinguishes between what he calls Indefinite Null Complement, as in () and Definite Null Complement, as in (). According to Fillmore, Definite Null Complements must be “retrieved from something given in the context” (p. ) while in Indefinite





Null Complements “the referent’s identity is unknown or a matter of indifference” and does not require a context as we can see from the contrast between () and (). ()

A: He was eating __ B: I wonder what he was eating

() A: They found out Ø B: #I wonder what they found out

(Indefinite Null Complement)

(Definite Null Complement or NCA)

Fillmore also claims that the predicates that can take NCA cannot be semantically derived since synonymous predicates do not all take NCA. However, he notes that there appear to be “some commonalities across word meanings within particular semantic domains in the semantic roles of omissible arguments” (p. ). These will be discussed in more detail in later sections.

.. Depiante (, ) Depiante (, ) revisits the issue of NCA in English and looks at NCA in Spanish and Italian as well. She concludes that in all three languages there is evidence that NCA is a deep anaphor and that the distinction between deep and surface anaphora is empirically accurate and should be maintained. In addition to the tests put forth by Hankamer and Sag (), she takes the following two facts as evidence for NCA having no internal syntactic structure: . NCA does not allow overt extraction (overt wh-movement and topicalization). . NCA does not allow inverse scope readings. These two properties will be discussed in more detail in the following two subsections.

... NCA does not allow extraction If in fact NCA—like other deep anaphors—has no internal structure at all, as proposed by Hankamer and Sag (), then we would expect there to be no extraction possible out of it. Extraction out of ellipsis sites has been shown in the literature to be a diagnostic for internal structure of that element (see Haïk , Fiengo and May , Johnson b, Houser , and others). One of the ways to show that VP-ellipsis, for example, has internal syntactic structure is by showing that extraction out of it is allowed, as shown by the following examples: () a. I know which book Max read and which book Oscar didn’t _ (VP-ellipsis) (from Fiengo and May : ) b. I think the blue papers Pete should sign and I think the green ones, Jan should __ (VP-ellipsis) (Houser : )

  



The reasoning goes like this: if a null string disallows extraction, then that is evidence that there is no internal structure in that element. However, other authors such as Aelbrecht () claim that this implication is not valid. We will discuss all this in detail in section ... Consider the following examples of lack of extraction in Do it anaphora from Johnson (b) and taken as evidence for lack of internal structure in the anaphor Do it. () a. *I know which book José didn’t read for class, and which book Lulumae did it for him. (cp. I know that José didn’t read this book for class, but that Lulumae did it for him.) b. *This is the book which O. J. Berman reviewed, and this is the one which Fred won’t do it. (cp. O. J. Berman reviewed this book but Fred won’t do it.) (from Johnson b: ) Depiante () shows that NCA does not allow overt wh-movement nor topicalization out of it, as shown in (a) and (b) respectively. This is taken as evidence for the lack of internal structure in NCA. () a. *I know which book Mary volunteered to read and Peter knows which article Sally volunteered Ø (Ø= to read) (from Depiante : ) b. *I think the blue papers Pete volunteered to sign and I think the green ones Jan refused Ø (Ø= to sign) Similar arguments have been made by others in the literature. For example, Merchant (a: ) gives the following minimal pair: a sentence with VP-ellipsis where extraction is possible and a sentence with NCA where extraction is not possible, in order to show that extraction can be used as a diagnostic for the existence of internal structure inside a null element: () a. Which films did he refuse to see and which films did he agree to? b. *Which films did he refuse to see and which films did he agree?

(VP-ellipsis) (NCA)

Haynie () also uses the lack of extraction to argue for the lack of internal structure in NCA—this will be discussed in section ...

... NCA does not allow inverse scope readings Inverse scope readings are not allowed with NCA, as opposed to their availability with VPellipsis constructions (see Hirschbühler ; Fiengo and May ; Fox ; among others). Sentence () illustrates a standard example of a sentence containing quantifiers and no ellipsis that allows two readings, as a consequence of the scope of each quantifier phrase: one reading in which ‘some doctor’ has scope over ‘every patient’ and another reading in which ‘every patient’ has scope over ‘some doctor’. ()

Some doctor volunteered to visit every patient (some doctor>every patient, every patient>some doctor)





In the examples in (), we can see the contrast between VP-ellipsis and NCA. Sentence (a) is the example with VP-ellipsis, and in this case two readings are possible: one in which ‘some nurse’ has scope over ‘every patient’ and another in which ‘every patient’ has scope over ‘some nurse’. In contrast, sentence (b) contains NCA, and only one reading is possible: the one in which ‘some nurse’ has scope over ‘every patient’. () a. Some doctor volunteered to visit every patient and some nurse did __ too (VP-ellipsis) (some nurse>every patient, every patient>some nurse) b. Some doctor volunteered to visit every patient and some nurse also volunteered Ø (NCA) (some nurse>every patient, *every patient>some nurse) (Depiante : ) In this respect, NCA behaves like other deep anaphors such as Do it in not allowing inverse scope readings, as shown in example (), where an inverse scope reading is not possible. () Many men read five books, and many women did it as well. (five>many,*many>five) (Compare: Many men read five books (five>many, many>five)) (Example from Baltin : )

.. Huddleston and Pullum () Huddleston and Pullum (: ) refer to NCA as lexical verb complement ellipsis and compare it to VP-ellipsis, which they call auxiliary-stranding ellipsis. They observe three differences between lexical verb complement ellipsis and auxiliary-stranding ellipsis, as well as a list of verbs that can license lexical verb complement ellipsis. The first difference that they observe is that lexical verb complement ellipsis is not compatible with relatives, while auxiliary-stranding ellipsis is, as we see in the contrast in (). () *They’ve asked him to mend the fuse, which he won’t even try Ø. Lexical verb complement ellipsis (cf. They’ve asked me to mend the fuse, which I can’t __) Auxiliary-stranding ellipsis (from Huddleston and Pullum : ) The second difference that they observe is that in lexical verb complement ellipsis, the full complement must be null, while in auxiliary-stranding ellipsis it doesn’t have to be, as we can see from the contrast in (). ()

*I won’t even try to attend the first session, but I will try Ø the others. Lexical verb complement ellipsis (cf. I can’t attend the first session, but I can __ the others) Auxiliary-stranding ellipsis (from Huddleston and Pullum : )

  



Finally, the third difference that they observe is that lexical complement ellipsis allows changes of voice while, according to them, auxiliary-stranding ellipsis does not—as illustrated in (): () The program needs to be corrected: why won’t Jill try Ø? Lexical verb complement ellipsis (cf. *The program needs to be corrected: why won’t Jill___?) Auxiliary-stranding ellipsis (from Huddleston and Pullum : ) Huddleston and Pullum () is descriptive in nature and no theoretical analysis is proposed.

.. Cinque () Cinque (: ) discusses NCA when it occurs with Italian restructuring verbs and proposes that in the syntax of Italian, an NCA with a restructuring verb is really the deletion of the Italian overt pro-form far lo (‘do it’). A sentence such as (a) in Italian would be represented as (b), according to Cinque (). ()

a. . . . ma io no posso Ø . . . but I not can b. ma [iok non posso . . . [VP tk [[V far] [DP lo]]]]

Under this view, the null element has some internal structure. Cinque’s analysis can account for the lack of object extraction out of the null site, which is also not possible in Italian. One could conceive of a proposal that extended Cinque’s () analysis of NCA with restructuring verbs in Italian to English7 where NCA could perhaps be conceived as the product of deletion of the pro-form Do it, which is suggested by Cinque (: ). 7 Interestingly, we independently know that Do it cannot have a stative predicate as its antecedent. The same holds for Italian ‘far lo’ (thanks to Guglielmo Cinque for the Italian judgments) and Spanish ‘hacerlo’, as shown in (i). Given this, we would be able to make an interesting prediction, which is that stative predicates could not be antecedents to NCA.

(i)

a. *I feel calm and Mary did it too. (Compare: I went for a walk and Mary did it too) b. *Me sentí tranquila y María también lo hizo. to-me felt calm and María also it did ‘I felt calm and María also did it.’ c. *Gianni è tranquillo e anche Maria lo fa. Gianni is calm and also Maria it does ‘Gianni is calm and also Mary does it.’

It appears that the prediction is borne out in Italian as we see from the examples in (ii): (ii) a. ??Io riesco a stare calmo (nonostante tutto) ma Maria non riesce Ø. I can to stay calm (in.spite.of everthing) but Maria not can ‘I am able to be calm (in spite of everything) but Maria is not able.’

Spanish

Italian





.. Haynie () Haynie () also argues that NCA is a null element without any internal structure. She does a thorough study of NCA in English where she explores the properties of a large sample of predicates that license NCA. She uses Hankamer and Sag’s () tests and demonstrates that NCA behaves like a deep anaphor. She also assumes that the lack of extraction out of an NCA-site indicates that NCA lacks internal structure. Below is a minimal pair where she shows the difference between the possibility of extraction from a VP-ellipsis site as opposed to an NCA-site: () a. I don’t know which puppy you should adopt, but I know which one you shouldn’t [VP adopt] (VP-ellipsis) b. *I don’t know which puppy John managed to adopt, but I know which one he tried Ø (NCA) Haynie () provides a thorough critique of Depiante’s semantic account of NCA which attempts to answer the question of the interpretation of these null complements. We will review her critique and her exact proposal in section .. where we will address the question of what syntactic category NCA belongs to and what its exact semantics should look like.

.. Summary and discussion of the different analyses of NCA in the literature Sections ..–.. have presented the different proposals regarding NCA that have been put forth in the literature. As we have seen, some authors propose that there is nothing in the syntax that is present in the position where the NCA appears (Shopen , ; Grimshaw ; Napoli b). Grimshaw proposes an interpretative rule that can account b. ?Io posso a stare calmo (nonostante tutto) ma Maria non può Ø. I can to be calm (in.spite.of everything) but Maria not can ‘I can stay calm (in spite of everything) but Maria can’t.’ The data for English in (iii) and Spanish in (iv) seems a bit mixed. More data would need to be explored in order to test this interesting prediction accurately. (iii) a. When mother asked him to be calm, Tommy refused Ø. b. ?Our yoga teacher asked us to be calm, so Susan volunteered Ø. (iv) a. Yo puedo estar tranquila (a pesar de todo) pero María no puede Ø. I can be-INF calm (in spite of everything) but María not can ‘I can be calm (in spite of everything) but María can’t.’ b. ?Nuestra maestra de yoga nos pidió que estuviéramos tranquilos, pero our teacher of yoga to-us asked that remain-SUBJ calm, but María se rehusó Ø. María refused. ‘Our yoga teacher asked us to remain calm but María refused.’

  



for its interpretation. Shopen does so as well. Napoli, on the other hand, takes NCA verbs to be intransitive uses of those verbs and proposes that the interpretation is done in the pragmatics. We have seen that another set of authors (Hankamer and Sag , Depiante , , and Haynie ) propose that NCA is an element with no internal syntactic structure that behaves like other deep anaphors with respect to the tests proposed in Hankamer and Sag (). Importantly, NCA behaves like other deep anaphors in that it does not allow extraction out of it, as opposed to what happens with surface anaphors. Under these analyses, NCA is not the product of a deletion operation, regardless of how we want that operation to be understood/formulated. Finally, there is another possible analysis which takes a mixed/hybrid view of NCA. In this view, NCA has some internal syntactic structure and does involve some deletion. That is the position taken by Cinque () for Italian and one that could perhaps be extended to English. Under this view, the complement of the verb would be generated as Do it, and later be deleted. This analysis is appealing in the sense that Do it is also a deep anaphor with respect to Hankamer and Sag’s () tests, and doesn’t tolerate extraction. Furthermore, it would predict that all predicates that take NCA also take Do it. However, that prediction is not borne out, because NCA doesn’t just have to be interpreted as a property but also can be interpreted as a proposition. So, such a proposal would not be able to account for sentences where the NCA is interpreted as a proposition, such as in () and to be discussed in section ...

. W   NCA  E?

.................................................................................................................................. Most authors agree that whether a predicate can take NCA or not is a lexical issue and cannot be derived from semantic facts. Fillmore () presents pairs of near-synonymous verbs that don’t both license NCA as evidence against a semantic criterion delineating the class of predicates that take NCA; see (): ()

a. She promised b. I tried c. They accepted/approved They concurred/agreed d. She found out e. I protest/object

a’. *She pledged/vowed/guaranteed b’. *I attempted c’. *They endorsed/authorized/acknowledged d’. *She discovered e’. *I oppose

(Fillmore : )

In (), I have listed the English predicates that have been thought to license NCA in the literature.

() Shopen 

Hankamer Grimshaw  and Sag 

Napoli Fillmore Saeboe b  

accept (dis)agree

Saeboe Depiante Huddleston Haynie   and Pullum   

x x

x

x

x

x

x x

x

x

x

allow answer

x x

x

apply

x

approach

x

(dis)approve

x

ask

x

x

x x

x

x x

x

x x

x

assume

x

be able

x

be certain

x

be convinced

x

be delighted

x

be offended

x

be sure

x

be surprised

x

beg begin bother

x x

x

x

x x

care

x

cease

x

check

x

complain comply

x

concur

x

x

x

x

x

confess

x

consent

x

continue

x

x

convince dare

x

x

x

x

x

x

x

debate desert

x x

discover dissent

x x

encourage

x

enter

x

escape

x

excuse

x

expel

x

explain

x

x (continued )

fail

x

find out finish

x x

x x

x x

x

x

x

force

x

forget

x

guess

x

x

x

x

hear

x

x

help

x x

inquire

x

insist

x

intrude

x

know

x

x

x

x

x

x

x

let

x

x

x x

x

x

x

x

x

x

listen listen in

x

x x

look

x

x

make (sb do sth)

x

x

manage

x

mind

x

notice object

x

x

x

x

x

x

x

oblige

x

offer

x

order

x

order (sb to do sth)

x

overhear

x

peek

x

x

x

permit

x

persuade promise

x x

x

x

protest quit

x

x

x

x

x x

x

recall

x

recover

x

refuse

x

rely

x

remember

x

x x

x x

reply rescue resume return

x

x x

x

x

x

x

x

x x

x x x

x (continued )

see

x

x

x

show start

x x

x

stay

x

x

x

stop

x

succeed

x

suppose suspect

x

x

x

x

x

x x

x

x

x

x

x

x

x

teach

x

tell try

x

understand

x

volunteer

x

x

x

x

x

x

x

x x

x

x

wait

x

x

x

x

warn

x

wonder

x

x

worry

x

x

  



As we see in the table in (), there is attested variation among speakers. This type of intra-linguistic variation is expected, since it seems that it is a lexical property of particular predicates that determines whether a predicate can take NCA or not. We also expect that there would be cross-linguistic variation in the predicates that take NCA. That seems to be the case if we look at work on NCA in Spanish and Italian for example (see Depiante , ). Authors agree that it seems difficult to find a natural class of verbs that take NCA. Fillmore () argues that the class of verbs that take NCA (in his terms Definite Null Complement) cannot be derived from a common semantics. Nevertheless, he says that verbs that take NCA seem to be “in a great many cases, verbs having to do with causing, inducing or allowing someone to perform an action” (Fillmore : ). See the examples in (). () a. b. c. d. e.

He dared me Ø. They made me Ø. They didn’t let me Ø. I asked him Ø. I ordered them Ø.

(Fillmore : )

Fillmore () proposes that the right generalization must involve something more than just causation or enablement since we don’t get NCA with verbs like cause, get, or have (which have more general meanings than the verbs that take NCA) but he says that a “social act of some sort” (Fillmore : ) is involved. He also says that most instances of aspectual complementation also allow NCA, which is true for start, stop, continue, finish, resume, stay, and begin. Somewhat related to Fillmore’s observations is Saeboe’s (, ) observation that “quite many of the verbs where zero argument anaphora is possible describe reactions and generate presuppositions as to the stimulus” (Saeboe : ) such as in the verbs agree or refuse or in predicates such as to be delighted. Saeboe () tries to distinguish the predicates that take NCA (which he calls definite ellipsis) from the predicates that select implicit complements (which he calls indefinite ellipsis) such as eat by virtue of the availability of a presupposition. He proposes that the predicates that take NCA have presuppositions involving their complements while the predicates that have implicit arguments do not. So, for verbs like agree or refuse, Saeboe says that when we say Sue agrees, that is represented as Sue agrees that p and that will be interpreted by way of the presupposition that someone else thinks that p. For Saeboe, an NCA (or a zero anaphor, as he calls it) is not really an anaphor but is instead an epiphenomenon of presuppositions. However, not all verbs that have presuppositions involving their complements will select NCA. In fact, Saeboe ends up admitting that the availability of NCA for a particular predicate cannot be predicted. Another attempt at trying to characterize the predicates that take NCA is the one in Depiante (, ), where it is proposed that the predicates that select NCA are the same predicates that do not take the overt predicate/propositional pro-forms it/so. She thus claims that NCA is in complementary distribution with it/so. In order to account for that, Depiante proposes an extension of Chomsky’s () Avoid Pronoun Principle. In later work, Haynie () shows that this characterization is not empirically accurate since there are numerous predicates which allow both NCA and it.





In agreement with other authors, Haynie () also admits that the class of predicates that select NCA cannot be defined either semantically or syntactically. Semantically, it cannot be defined because, following Fillmore, predicates with similar meanings do not all take NCA, as shown earlier. Syntactically, they cannot be defined either, because verbs with identical syntactic complementation behave in different ways with respect to whether they license NCA or not. Haynie illustrates this with the contrast in (). () a. Janet was able to ride her bike up Mount Diablo but she didn’t bother Ø b. *Janet was able to ride her bike up Mount Diablo but she didn’t endeavor Ø (Haynie : ) In example (), Haynie shows that bother selects an NCA but endeavor does not, even though they both select the same type of syntactic complement.

. H  NCA    ? H  NCA ?

.................................................................................................................................. As we have seen, there are different proposals in the literature as to what the syntax of NCA looks like. Some authors believe that nothing is present in the syntax and that an interpretative rule in the semantics or in the pragmatics provides the right interpretation. Other authors believe that NCA is represented in the syntax by a phonetically empty pro-form with no internal syntactic structure; in other words, that NCA is the null counterpart of it/so as in (). ()

John believes the war in the Middle East will end soon, but I don’t believe it/so.

As we have seen, the basis for arguing for the lack of internal structure comes from the unavailability of extraction out of an NCA-site. In the following section, we discuss the validity of this diagnostic.

.. Is lack of extraction a good diagnostic for lack of internal structure? As mentioned in previous sections, the literature on ellipsis has used the possibility of extraction out of null elements as a diagnostic for internal structure inside that element. In other words, it has assumed the following implication: ()

Availability of extraction from a phonetically null element implies internal structure inside that element.

  



The reverse implication, given in (), has been assumed by some authors but challenged by others. ()

Non-availability of extraction from a phonetically null element implies lack of internal structure inside that element.

One such challenge comes from Aelbrecht () in her work on Dutch Modal Complement Ellipsis (MCE) and British English do. She shows that these types of ellipsis allow some extractions out of them but not all. They allow A-movement (passive and subject raising) but do not allow A-bar-extraction. She claims that “the lack of extraction does not automatically mean that the ellipsis site is a null pro-form without internal structure, although it can be” (Aelbrecht : –). The gist of her account is that Dutch MCE and British English do exhibit internal syntactic structure and that the reason why A-movement is allowed and A-bar-movement is not has to do with the timing of the ellipsis operation. So, these types of anaphors in Dutch and British English are in fact, according to her, the product of a deletion operation. According to this analysis, the actual deletion operation will happen after A-movement but before A-bar-movement. The fact that Dutch MCE and British English do behave the way they do appears to call into question the implication in (). However, it doesn’t follow from this that () does not work for NCA in English. As we have seen, overt extraction out of English NCA is not possible and neither is covert extraction (recall the inverse scope facts in section ...). So, given these data, NCA in English cannot be equated to a surface anaphor, just because Dutch MCE or British English do behave differently. The point is that they behave differently and they are a different type of anaphor. The locus of contention should not be () as a diagnostic but instead, the discussion should be focused on the empirical adequacy of the typology of anaphors proposed or whether a typology is necessary at all. Along these lines, some authors such as van Craenenbroeck and Merchant (: ) have suggested that “it could well be that even null complement anaphora involves abstract syntax, but that the timing of the ellipsis process in this specific construction precludes any movement operation from targeting material inside the ellipsis site.” However, the fact that some null anaphoric elements in some languages allow some extractions but not others seems to indicate that there still is a difference between different types of anaphora. We either need to expand the typology of anaphors (see Thompson ) or propose that all null anaphors are the product of deletion. Then, the difference between deep and surface anaphors will be lost as a theoretical primitive and would be reduced to the timing of the deletion operation. Under this view, the grammar must stipulate the exact timing of each deletion. If it turns out that this is how the grammar works, then we would lose the empirically based distinction inherent in the typology of deep and surface anaphora. On the other hand, it might be theoretically appealing to say that all null anaphors are generated by just one deletion operation and that the fact that the deletion operation applies at different points in the derivation is what is perceived as the distinction between deep and surface anaphora. Future research in this area will be able to tell us more about these alternative positions.





.. On the semantics of NCA In terms of the semantics, one clear proposal is put forth by Depiante (, ), who claims an NCA-site contains a free variable that is specified for three semantic types (property, proposition, or question), as summarized in () and illustrated in ()–() respectively. () a. Type (when interpreted as a property) b. Type (when interpreted as a proposition) c. Type (when interpreted as a question) () Mother asked John to clean up his room but he refused Ø Ø = to clean up his room. Semantic type: Property () Anne said that they should leave and Bill agreed Ø Ø = that they should leave Semantic type: Proposition () a. Anne asked what time it was and Mary inquired Ø Ø = what time it was Semantic type: Question b. Anne asked the time and Mary inquired Ø Ø = the time Semantic type: Question However, Depiante (, ) claims that NCA is never specified for the semantic type e, an individual, as we can see from the ungrammaticality of (). () *John plays golf and Peter tried Ø Ø = golf Upon examination of a larger set of NCA-selecting predicates, Haynie () discovers that NCA can in fact be interpreted as an individual (type e) in certain circumstances. In this way, Haynie () challenges the proposal put forth in Depiante (, ). The interesting discovery by Haynie () is that PPs which contain a semantically vacuous preposition and an individual-denoting DP can be replaced by NCA as in example (). So, in this case, NCA is interpreted as an individual. () The board considered the new proposal but half of the members objected Ø (Ø = to the proposal; Ø 6¼ to considering the proposal) (Haynie : ) Based upon examples such as (), Haynie argues that Depiante’s “semantic” account of the nature of NCA is not viable, since it doesn’t account for the fact that NCA can have an individual interpretation in certain cases when it replaces a PP with a semantically vacuous preposition. However, she recognizes that NCA can never have an individual interpretation given by a DP. So, she proposes what she calls a “syntactic” account of NCA to account for the new

  



set of data. In her account, the grammar does not need to specify the semantic types that NCA will be interpreted as; it only needs to specify the types of syntactic elements that it can be. The semantic types follow from the interpretation of the syntactic types. According to Haynie (), NCA can be of the syntactic category CP, VP, or PP, but crucially not DP or AP. She proposes that the semantic interpretation of NCA will be determined by its antecedent and the predicate that selects it. Here is her exact proposal, which we believe to be empirically accurate: ()

NCA consists of a null, anaphoric pro-form (i.e. free variable) that can be merged in place of a PP, CP, or VP. No NCA pro-form exists with the category features D or A. There are no semantic restrictions on NCA. NCA will be interpreted as any semantic type that is compatible with its syntactic category, its selecting predicate, and its antecedent (Haynie : ).

. S     NCA

.................................................................................................................................. To summarize, here is a list of the properties that characterize NCA in English: . It is licensed by a particular group of predicates, determined lexically—no natural class has been found yet. . It can be interpreted anaphorically or exophorically: that is, it can have a nonlinguistic antecedent. . It behaves like a deep anaphor according to Hankamer and Sag’s () tests: it allows pragmatic antecedents, disallows missing antecedents and doesn’t need strict syntactic parallelism with its antecedent—as exemplified by changes in voice. . It does not allow extraction: no overt wh-extraction, no topicalization. . It does not allow inverse scope readings. . It must be completely null; no subpart of the NCA can be overt. . Syntactically, it can replace a CP, VP, or PP but not a DP or an AP. . Semantically, it can be interpreted as a property, a proposition, or a question as well as an individual—when it replaces a PP with a vacuous preposition.

. NCA -

.................................................................................................................................. NCA is common cross-linguistically. The question arises as to whether it receives the same analysis in other languages as its English counterpart or not. Because of space limitations, I will not discuss here all of the cross-linguistic work available in the literature, but will refer the reader to some of the work that is available. Research on NCA in Spanish (Zubizarreta ; Bosque ; Brucart , ; Brucart and MacDonald ; Bošković ; Depiante , ) and Italian (Cinque , ; Cecchetto and Percus ) seems to point in the direction that NCA in these languages is also a deep anaphor (according to





the results of the tests presented in the literature on English) or some version of it (see Cinque , ). However, other authors like Dagnac () seem to suggest that at least the null complements of modals in Spanish and Italian (as well as French) seem to exhibit some of the characteristics of surface anaphors. Work on Portuguese (Cyrino and Matos ) seems to point to the idea that in Portuguese, the equivalents of Spanish NCA do not behave like deep anaphors. Research on Romance thus seems to suggest that there might be a split between the way the null complements of modals behave versus the null complements of other types of predicates that seem to select NCA. The former appear to exhibit some of the characteristics of surface anaphors while the latter do not. Further research with larger sets of predicates and more data is needed to elucidate what the exact nature of these complements is and in what ways the type of predicate determines the type of null complement it takes. Cross-linguistic research is fundamental to further our understanding of the nature of NCA and its interactions with other elements of the grammar, as well as to see whether a typology of anaphors is necessary and whether there is a universal typology or not.

. C     

.................................................................................................................................. In this chapter, we looked at NCA in detail and discussed its nature and behavior. The empirical evidence reviewed points in the direction of NCA in English being a deep anaphor in Hankamer and Sag’s () sense, and thus seems to further support their typology of anaphora. The syntactic and semantic properties of NCA were presented. Syntactically, NCA appears to be a syntactic unit without any internal structure. The diagnostic of lack of extraction out of null elements as evidence for lack of internal structure was used for English NCA. Challenges to the use of this diagnostic were discussed. An alternative view that has been hinted at—but not been completely explored—in the literature by van Craenenbroeck and Merchant () is that NCA might perhaps have abstract syntax and that the lack of extraction is a consequence of the timing of the deletion operation à la Aelbrecht () for Dutch MCE and British English do. Advantages and disadvantages of such a view were discussed. Further research along these lines should be interesting and elucidating. One of the questions that still remains to be answered is how to define the class of predicates that take NCA. It seems to be lexically determined and it exhibits intra-linguistic and cross-linguistic variation. A corpus study combined with work on the lexical semantics of the predicates that license NCA is a direction in which future research might go. In this chapter, we have been able to see that NCA is clearly a phenomenon of the interface. The lexicon, syntax, semantics, and pragmatics all conspire to license it, restrict it, and provide an interpretation for it. We think future research on NCA will benefit from more research focused on the interfaces of the different components of the grammar.

  ......................................................................................................................

   -  ......................................................................................................................

 

T chapter surveys ellipsis or sharing phenomena peculiar to coordination structures, known as Conjunction Reduction and Right-Node Raising, and reviews some leading issues in their theoretical treatment, including their relation to non-coordinate ellipsis phenomena such as verb phrase ellipsis (VPE). While focusing on analyses from mainstream generative grammar, the chapter seeks to highlight empirical patterns and generalizations. No attempt is made at wide cross-linguistic coverage; the data are mostly English, though some German examples are discussed as well.

. I

.................................................................................................................................. The term Conjunction Reduction (CR), which goes back to generative grammar of the s, is nowadays associated with the idea that conjuncts in a coordinate structure can be reduced by ellipsis, in the sense of omission of repeated material. Examples like (), where material in one conjunct is repeated in another, have paraphrases in which the relevant words (underlined) are missing from one conjunct, as shown in (). ()

a. They offer money to the rich but they offer nothing to the poor. b. We are not responsible for any losses incurred and we will not reimburse you for any losses incurred.

()

a. They offer money to the rich but nothing to the poor. b. We are not responsible for and we will not reimburse you for any losses incurred.

CR phenomena fall into two classes according to directionality. Forward CR can be modelled as ellipsis in the non-initial conjuncts of a coordination licensed by pronounced material inside the initial conjunct. In an analysis of (a) that takes the conjuncts to be





sentential, the subject and verb in the second conjunct are elided, under (some notion of ) identity with corresponding pronounced material in the first conjunct (underlined).1 ()

< [ They offer money to the rich] but [ they offer nothing to the poor] >.

In backward CR, commonly known as Right-Node Raising (RNR), material is missing in non-final conjuncts, so the pattern can be modelled as ellipsis under identity with pronounced material in the final conjunct.2 ()

< [We are not responsible for any losses incurred] and [we will not reimburse you for any losses incurred] >.

An ellipsis approach cashes out the omission of words from a larger expression in terms of the non-pronunciation, or deletion, of items that are present inside a conjunct at an underlying syntactic level (indicated by strikeouts in ()–()). However, there is no consensus on whether any, and if so which, of the data in question involve an ellipsis mechanism in this sense. I use sharing (in coordination) to describe the phenomenon without prejudging a particular approach. In (), the underlined elements are shared, while the remainder (excluding the coordinators) is non-shared material, belonging solely to one conjunct. What is uncontroversial is that the underlined material in () is semantically shared by the conjuncts. In (a), the non-shared NP-PP sequences are semantically arguments of (tokens of ) the same shared predicate offer, along with the shared subject they. In (b), any losses incurred likewise figures in the meaning of both conjuncts, supplying an argument to the non-shared predicates. Contemporary approaches diverge widely on whether and in which cases shared expressions are structurally (syntactically) shared, in the sense of being structurally contained in both conjuncts, as in CR analyses like ()–(). This variety reflects different assumptions concerning constituent structure, the size of conjuncts and the treatment of movement phenomena (see .). In () and many other cases associated with CR, the shared material (underlined) is peripheral to the coordination, being located at the left (a) or right (b) periphery of the non-shared material of the conjuncts. This partly explains why these phenomena are susceptible to an alternative claim concerning the structural location of the shared items—namely, that they are outside the coordination (). ()

a. They offer < [money to the rich] but [ nothing to the poor] >. b. < [We are not responsible for] and [we will not reimburse you for] > any losses incurred.

1 Throughout the chapter, angled brackets < . . . > mark the coordination, i.e., the constituent comprising the conjuncts plus the coordinator. 2 The pattern in () was attributed by Ross () to a rightward across-the-board extraction rule which he called Conjunction Reduction (see ..); it has been known as Right-Node Raising (RNR) at least since Postal (). The term RNR is commonly applied to the phenomenon itself without implying a particular analysis of it.

   - 



Such approaches—including across-the-board movement (ATB) analyses (see ..)—I call direct analyses, in contrast with CR analyses which locate the shared material inside the coordination in surface structure. Direct approaches are challenged by cases where shared items are not peripheral but sandwiched among non-shared material in one of the conjuncts, as in gapping (). Viewed as forward CR, gapping involves medial ellipsis of the verb, auxiliaries, and possibly other elements.3 ()

a. The girls are sleeping here, and Peter, there. b. < [the girls are sleeping here] and [ Peter is sleeping there] >

CR patterns are mostly found in structures with and/or/but, hence the term coordination ellipsis. They normally do not occur in clauses linked by subordinators such as because, while, or unless (). This sets them apart from types which are not coordination ellipses— sluicing (a), verb phrase ellipsis (VPE) (b), and noun phrase ellipsis (NPE) (c). ()

a. *They offer money to the rich, while nothing to the poor. b. *The girls are sleeping here, because Peter, there. c. *They will not reimburse you for, unless they are responsible for, any losses incurred.

() a. I’m sure he met someone, but/while [I don’t know who he met]. b. She will sleep there and/because [he will sleep there, too]. c. He pays ten dollars or/unless [she pays twenty dollars]. However, “coordination ellipsis” may be a misnomer—both forward CR and RNR patterns also turn up in configurations not normally viewed as coordinations. Most notably, both figure in comparative constructions (more-than, as-as, etc.); cf. Lechner (). In addition, examples similar to (c) with the RNR pattern may, unlike (a,b), be judged acceptable to various degrees. Non-coordinate RNR is taken up in ... The divide between forward and backward CR is deep (Neijt : ch. ); the directionality difference correlates with distinct clusters of properties, reviewed in . and .. While backward CR (RNR) is typically considered a unitary phenomenon, individual forward patterns are often treated under different labels, including gapping, stripping (see ..), and what I term Phrase Cluster Coordination (PCC), illustrated by (a) (see ..). Gapping and stripping are treated in depth in Chapter . The present chapter largely ignores medial gapping, but briefly discusses stripping and PCC—firstly, to illustrate the rationale behind CR approaches to coordinations that resist direct analysis; and secondly, to demonstrate the affinity of forward CR to non-coordinate ellipses such as VPE. The chapter is structured as follows. Section . introduces assumptions and data that figure in the debate about direct vs CR approaches to sharing. Sections . and . look at forward and backward sharing in more depth, highlighting their different empirical profiles, and what appear to be promising theoretical accounts. . focuses on a CR 3

Another case of medial sharing is discussed in .. (Right-Node Wrap).





approach to forward patterns that combines conjunct-internal movement and ellipsis. . reviews arguments for and against direct (ATB movement) and ellipsis analyses of RNR, and outlines the case for an alternative approach using Multiple Dominance. Section . concludes.

. C   CR

.................................................................................................................................. This section begins with the early conception of CR and why it was abandoned, and then sketches the subsequent standard view on coordination syntax, which allows direct analysis of phrasal coordinations. Specific classes of data that resist direct analysis within that frame have motivated more recent CR proposals (..–..), which compete with alternative direct analyses utilizing ATB movement, or flexible syntactic constituency (..–..). Other kinds of motivation for CR are pointed out in ...

.. Conjunction Reduction vs phrasal coordination Many an ordinary phrasal coordination is at first blush semantically equivalent to a sentential coordination, as indicated in () by the parentheses. Matching the paraphrase relation is a syntactic equivalence which can be stated as (), adapted from Chomsky (: ). ()

()

a. John (will be there) or Mary will be there. b. They have a huge mortgage but (they have) a tiny house. c. He dries his laundry (in the attic) and (he) stores his books in the attic. If [S1 α X β ] is a sentence and [S2 α Y β ] is a sentence, where S and S are identical except for X and Y, and X and Y are constituents of the same type, then [S3 α X Conj Y β ] is a sentence.

This characterization prefigures both the early idea that subsentential coordinations are generally derived from sentential coordinations, and two issues (discussed in .. and ..) behind later CR proposals, namely whether the non-shared strings (X and Y) constitute single constituents of the same type. Early generative work pursued the idea that phrasal coordinations result from a syntactic rule applying to underlying coordinated sentences. () can be reformulated as a “Conjunction Reduction” transformation (), which takes a coordination of sentences as input, and delivers a reduced structure, a single sentence containing coordinated substrings (cf. Gleitman , Ross , Dougherty /, Koutsoudas , Hankamer , Hudson b, and others). ()

Input: Output:

< [S α X β ] Conj [S α Y β ] > [S α < X Conj Y > β ]

   - 



The approach was consistent with the framework of Chomsky (): the deep structure of a sentence determines its meaning, transformations derive the surface structure. It captures the paraphrase relation between phrasal and corresponding sentential coordinations, and the intuition that the meaning of and and or is constant across phrasal and sentential coordinations. It further suggests that their meanings can be equated with logical conjunction and disjunction of classical propositional logic. However, () posits massive restructuring of the input. Two sentences are reduced to one, and X and Y, separate in the input, are assumed to form a constituent [X Conj Y] in the output. Such derivations are incompatible with subsequent developments (e.g. the idea that syntactic operations are structure-preserving) and have long been rejected in mainstream generative grammar. The approach was also abandoned on other grounds. Underlying the rule in () is the idea that the paraphrase relation between phrasal and sentential coordinations is due to their having the same structure (at the relevant level). Yet by no means do all phrasal coordinations have sentence coordination paraphrases. One class concerns and-coordinations of singular nominals that function as the argument of a collective predicate such as alike in () (Lakoff and Peters ). Since the predicates in question select for a semantically plural argument, the corresponding sentential coordination is simply ill-formed. ()

a. John and Mary are alike. b. *John is alike and Mary is alike.

The paraphrase also fails where the phrasal coordination is interpreted in the scope of a higher negation, attitude predicate, or quantificational expression. The examples in () express something different than their full sentential equivalents in (); cf. also (). ()

a. Sam denied being involved in the crime or knowing the culprit. b. Everyone voted for Bush or for Clinton.

()

a. Sam denied being involved in the crime or Sam denied knowing the culprit. b. Everyone voted for Bush or everyone voted for Clinton.

With the breakdown in the paraphrase relation, there is no motivation from interpretation for assuming sentential conjuncts. These cases lead naturally to the assumption that phrases of all types (NP, AP, VP, PP, etc.) can form coordinated constituents embedded directly within the larger sentence, where relative semantic scope is reflected in structure. ()

a. b. c. d.

Nothing is < [AP round ] and [AP square ] >. ¬∃x ((x) ∧ (x)) < [S Nothing is round ] and [S nothing is square ] >. ¬∃x ((x)) ∧ ¬∃y ((y))

To the extent that phrasal conjuncts do not denote propositions, the semantics of and and or cannot be limited to the logical conjunction/disjunction of propositional logic. A compositional approach requires a different semantics for coordinators that link phrasal





conjuncts, such as the “generalised conjunction” of Partee and Rooth (). The denotations of (a,b) can then be captured straightforwardly on the basis of a phrasal coordination in the scope of a single shared operator. At the same time, the paraphrase relation in cases like () falls out as a result of direct semantic interpretation of the phrasal coordination. In addition, a different semantics is needed for the and of () (“group-forming and”). See Zamparelli () for an overview. The upshot is that direct phrasal coordination, in which the conjuncts contain exclusively non-shared material, can capture not only ()–(), but also (), with no need of CR to capture semantic sharing effects reflected in sentential paraphrases. Such an approach, now more or less universally adopted, underpins a common view on the core of coordination syntax. Nevertheless, a version of the old CR idea is retained by ellipsis analyses of sharing in coordination for cases in which the core approach breaks down. The structural reduction of () is replaced by a conception in which the output retains the structure of the (not necessarily sentential) input, and the reduction itself affects only the pronunciation of words contained in the input. The output in () symbolizes the combination of forward and backward CR sketched in ()–(). () Input: Output:



.. Coordination syntax: Constituency, substitutability, and coordination of likes This section reviews core assumptions concerning coordination syntax that are widely held.4 Firstly, a coordination comprises a sequence of two or more adjacent expressions (conjuncts), which are interrupted only by the coordinator which links them, and which are independent of one another in the sense that no conjunct is part of another in the coordination. Secondly, the coordination itself and its individual conjuncts are constituents. ()

a. A coordination constitutes a single syntactic constituent. b. Each of the conjuncts Xi constitutes a single syntactic constituent.

Descriptively, possibilities seem to be further constrained by a condition like (): ()

Substitutability of the conjunct for the coordination: Each conjunct Xi may occur independently (i.e. replace the coordination) in the larger expression containing the coordination.

4 To facilitate the presentation, the remainder of the chapter assumes (except where otherwise stated) a standard X'-theoretic analysis of non-conjoined expressions, in terms of a CP-TP(-vP)-VP clause structure, and a DP-NP structure for nominals.

   - 



For example, () rules out an analysis like (a) on the basis of the ill-formedness of *he a tiny house, while permitting the correct analysis (b).5 ()

a. *He < [ has a huge mortgage ] but [ a tiny house ] >. b. He has < [a huge mortgage ] but [ a tiny house ] >.

Examples like (), where each of the italicized expressions satisfies (b) and (), are nevertheless ill-formed. Such cases motivate an additional condition that the conjuncts of a coordination be syntactically and/or semantically similar (cf. ()), which has come to be known as the Law of Coordination of Likes (LCL) following Williams (). ()

a. *The scene of the movie and that I wrote was in Chicago. (Chomsky : ) b. *He drank whisky and on Sunday.

()

LCL: Coordinated constituents are of the same type or category.

What exactly “same type” means, whether for example, traditional syntactic categories, or semantic types, is subject to debate (and of course depends on what types are defined by one’s theory). There is a widespread practice in the syntax literature of assuming that “same type” means “same syntactic category.”6 The classic phrase structure for coordination is a special configuration type (), according to which for X = any category, a coordination is a constituent of type X that immediately dominates two or more (signalled by the Kleene star) constituents of the same type X (the conjuncts), plus a coordinator. ()

Classic coordination structure: [X (X*) X & X ]

() encodes both the LCL and, by identifying the type of the conjuncts with the type of coordination, substitutability. It also straightforwardly accounts for the possibility for coordinations to be nested. In frameworks that assume phrase structure uniformly conforms to X’-theory, the multiple headed () is anomalous, and endocentric alternatives have been proposed that take the coordinator to head the coordination. In one version (), & forms a phrase with the second conjunct that is right-adjoined to the first conjunct (Munn ). In another (), the conjuncts are specifier and complement to & (Larson : ; Kayne : ch. ; Johannesen ).7 5 Counterexamples to substitutability arise where an external item depends on the presence of the coordination—either, respectively, and plural agreement in (i)–(iii):

(i) He either met John *(or Mary). (ii) John *(and Mary) are happy. (iii) They drank wine *(and beer) respectively. 6

For early discussion, see Schachter (a). Whitman () has insightful discussion of semantic and pragmatic factors influencing judgements on coordinations of different types of phrases. 7 The LCL is not an inbuilt feature of () or (); the type equality of XP and YP needs to be ensured independently. See Progovac () and Borsley () for critical discussion of endocentric approaches.





()

Adjoined &P structure: [XP XP [&P & YP ]]

()

Specifier-complement &P structure: [&P XP [&' & YP ]]

It is uncontroversial that phrases of any category can be coordinated, from root clauses down to the smallest individual phrases, whether these be functional categories (CP, TP, DP) or lexical. The broader consensus breaks over whether (in X’-theoretic terms) nonmaximal constituents can form conjuncts. While the coordination of heads, for example, is often assumed under (), the endocentric approaches do not permit the conjuncts to be subphrasal.8 Cases of apparent coordination of heads, e.g. as in (a), have been argued to involve phrasal coordination with structural sharing of the direct object (backward ellipsis in Kayne : ch. ).9 Apparent coordination of sub-word units as in (a) is problematic for either approach, and has been argued to involve backward ellipsis (Booij ; Wilder ; Chaves ). ()

a. They will him. b. They will .

()

a. in the -war periods. b. in the periods.

In sum, the consensus picture has it that coordinations whose non-shared strings form adjacent constituents of the same type (perhaps further restricted to phrasal constituents) do not involve structural sharing. That picture is disturbed by three types of cases (in addition to ()), each of which has been argued to motivate a CR-type analysis. One involves conjuncts that appear not to correspond to a well-formed single constituent, thus threatening (b) (see ..). Another concerns apparently non-adjacent conjuncts (see ..). A third involves Unlike Category Coordinations (UCCs), i.e. what appear to be phrasal conjuncts of different categories in violation of the LCL.

.. Unlike Category Coordination Examples like (a) (Sag et al. ), where the non-shared elements are predicates of different categories, illustrate the UCC phenomenon. ()

a. John is a Republican and proud of it. b. John is < [DP a Republican] and [AP proud of it]>.

8 Examples like (i) suggest coordination of an intermediate projection (C’) (likewise for conjoined finite verb-initial constituents with shared initial specifier phrase in Germanic V languages); but under alternative assumptions (e.g., multiple heads in the C-domain, following Rizzi ), they may be analysed as phrasal coordination.

(i)

Why < [has the teacher left] and [are the pupils still here]>? 9

For a defence of head coordination, see Abeillé (a).

   - 



The direct analysis (b) requires a modification of LCL or of assumptions concerning categories (the latter pursued by Sag et al. ). () illustrates another less well-known UCC type, involving focus or focus particles (Grosu , ). Related cases involve whquestions (What and when does he drink?). ()

a. He drinks only  and only on S. (compare (b)) b. He drinks < [DP only whisky ] and [PP only on Sundays ] >.

Most UCCs conform to substitutability.10 Exploiting this property, a CR approach can offer an account that satisfies LCL while allowing leeway on the category of non-shared constituents within the conjuncts. If α can combine with constituents of different categories X and Y such that both [Z α X ] and [Z α Y ] are independently well-formed, and if α can be structurally shared, then Z-coordination with sharing of α can yield the effect of a UCC. This is outlined in (), where α = the verb (plus other material). ()

a. John < [VP is a Republican] and [VP is proud of it] >. (forward CR) b. He < [VP drinks only  and [VP drinks only on S ] >. (forward CR) c. < [CP What does he drink ] and [CP when does he drink ] >? (backward CR)

On (a), see Beavers and Sag (), Chaves () and for critical discussion, Levine ().11 On (c), see ...

.. Discontinuous coordination and stripping The phenomenon illustrated by () poses a related challenge to the core assumptions. ()

He left his phone at home, and his keys.

Taking the first conjunct to be a phrase left-adjacent to and (PP, VP, or TP in ()) conflicts not only with the LCL but also with the intuition that the first conjunct is the DP his phone. ()

[TP He [VP left his phone [PP at home ]]], and [DP his keys ].

On (i) (from Sag et al. ), which contradicts substitutability as well as the LCL, see ... (i) You can depend on [DP Bill ] and [CP that he will be on time]. 11 Constructions where a predicative UCC is isolated from the verb, such as fronting (i), raise questions for (a). Beavers and Sag () consider the analysis (ii) involving larger conjuncts with backward CR, but Levine () points out problems. For instance, (iii) indicates a smaller coordination, since both cannot modify conjoined root clauses. 10

(i) Tired and in a foul mood, he certainly was. (ii) (iii) Both tired and in a foul mood, he certainly was. Chaves () discusses related adjunct UCCs (Both tired and in a foul mood, Bob packed his gear).





Some have taken this to indicate that the conjuncts of a coordination need not always be adjacent. Munn (: ) suggests an approach in terms of rightward movement of the second conjunct of a DP-coordination (and his keys forms a phrasal unit &P according to ()), paralleling extraposition of a postnominal modifier (see also Johannesen : ff.). ()

He left < his phone tj > at home, [ and his keys ]j .

However, an indication that () does not involve DP-coordination is that the pattern fails with collective predicates like combine (cf. Merchant a).12 ()

a. He combined zinc and hydrochloric acid yesterday. b. *He combined zinc yesterday, and hydrochloric acid.

Forward CR analyses attribute discontinuous coordinations to the data set derived by stripping, “a rule that deletes everything in a clause under identity with corresponding parts of a preceding clause, except for one constituent (and sometimes a clause-initial adverb or negative)” (Hankamer and Sag : ).13 Under this approach, the non-shared DPs of () are contained in larger, adjacent conjuncts conforming with the LCL, and (b) is straightforwardly accounted for.14 ()

< [TP He left his phone at home ], and [TP he left [ his keys] at home ] >.

Considerations about relative scope of the coordinator and shared negation or quantifiers (cf. ..) also apply in determining the height of the coordination in stripping. The negation naturally scopes over the coordination in (a), indicating that the coordination is at VP (or vP) level. ()

a. He didn’t leave his phone at home, or his keys. b. He didn’t < [ leave his phone at home ], or [ leave [ his keys ] at home ] >.

Interpreting the negation inside both conjuncts, expected if the coordination is at clause level, is virtually impossible in (). The possibility for larger conjuncts is evidenced by (), where the negation is interpreted inside both conjuncts, in the scope of or.15 ()

a. Either J didn’t switch the light off, or M. b. < [ Either J didn’t switch the light off ], or [ M didn’t switch the light off ] >.

Stripping is discussed further in .; see also Chapter . Neijt (: –) gives further arguments against ‘conjunct extraposition’. The construction is sometimes termed Bare Argument Ellipsis, yet it is also found with predicates (He seemed angry yesterday, and rather drunk). A different approach to stripping, according to which the correlate in the first conjunct raises covertly to attach to the second conjunct, is argued for by Reinhart (); see also Fiengo and May (: –). 14 Prinzhorn and Schmitt () discuss German data problematic for this account. 15 On the interaction of either and ellipsis, see Den Dikken () and Hofmeister (). 12 13

   - 



.. Non-constituent conjunct strings The string of the first conjunct in (a) does not correspond to a constituent (Chomsky : ); the same applies to (b) and (b). ()

a. John enjoyed and my friend liked the play. b. A very tall and a very short child were playing together.

The examples in () instantiate what has come to be known as Right-Node Raising. Standard assumptions about constituent structure strongly favour a structural sharing approach, such as backward ellipsis (). While (a) is a sentential coordination, the plural agreement and collective predicate in (b) suggest a DP coordination with structural sharing of the N child only. () a. < [TP John [ enjoyed the play ]] and [TP my friend [ liked the play ]] >. b. < [DP A [very tall child ]] and [DP a [very short child ]] > were . . . Two forward patterns are found where the non-shared material in the second conjunct does not form a single constituent (in a traditional structure). One is gapping (see ()). In the other, (), Phrase Cluster Coordination (PCC) (cf. also (a)), the non-shared material is a sequence of postverbal phrases, often analysed (following Sag ) as VP-coordination with forward ellipsis of the shared verb. ()

a. She’ll fix her car today and her bike tomorrow. b. She’ll < [VP fix her car today ] and [VP fix her bike tomorrow ] >.

The non-constituent status of the non-shared material evident in RNR and PCC (and medial gapping) has provided perhaps the strongest impetus behind recent explorations of CR analyses in the form of ellipsis (see . and ..) and/or Multiple Dominance (see ..).

.. Conjunction Reduction and movement Movement delivers an alternative way of reconciling apparent non-constituent conjuncts in the RNR, gapping, and PCC patterns with the core assumptions of ... This perspective was opened up by Ross’s () discovery of the Coordinate Structure Constraint (CSC) and the across-the-board (ATB) movement phenomenon (see de Vries  for recent discussion). ()

Coordinate Structure Constraint “In a coordinate structure, no conjunct may be moved, nor may any element contained in a conjunct be moved out of that conjunct” (Ross : )—unless the same element is moved ATB (out of all conjuncts in the coordination).16

16 The ATB exception to the CSC applies only to movement of an element contained within a conjunct. Conjuncts themselves appear to be immobile (Grosu ); granted, that is, that there is no conjunct extraposition.





Examples () and () illustrate the ban on leftward movement of a conjunct and of a constituent out of a conjunct (Ross : ). () exemplifies the possibility for leftward ATB extraction, showing that the constituent must be extracted from all conjuncts.17 ()

a. *What sofa will he put the chair between < [ some table ] and [ _ ] >? b. *What sofa will he put the chair between < [ _ ] and [ some table ] >?

()

a. *The lute which Henry < [ plays _ ] and [ sings madrigals ] > is warped. b. *The madrigals which Henry < [ plays the lute ] and [ sings _ ] > sound lousy.

()

These are the grapes which a. < [Tom picked _ ], [I washed _ ], and [Suzie will prepare _ ] >. b. *< [Tom picked _ ], [I washed some turnips], and [Suzie will prepare _ ] >.

Ross proposed to treat RNR as ATB movement.18 The missing object in non-final conjuncts, giving the impression of a non-constituent string, is (in more recent terms) analysed as a movement trace; and a corresponding trace is posited for the final conjunct. The shared object is located (after movement) outside of the coordination (). Like leftward ATB movement, RNR applies in ATB fashion in a coordination of more than two conjuncts () (Ross : –). ()

[ < [ Tom picked tDP ], [ I washed tDP ], and [ Suzie will prepare tDP ] > [DP these grapes ] ].

()

Tom picked, I washed (*some turnips), and Suzie will prepare, these grapes.

The relative merits of movement and backwards CR approaches to the RNR phenomenon are discussed in .. The interdependence of assumptions on movement and coordination is illustrated by an early problem with the direct analysis of VP-coordination and its interaction with passivization (i.e. the movement assumed to derive the subject of a passive clause) (Dougherty , ). Under the direct analysis (a), passive movement violates the CSC. A CR-style analysis avoids the CSC problem (b), but runs into the interpretation problem sketched in ... The adoption of the VP-internal subject hypothesis allows a reinterpretation in terms of ATB movement out of small conjuncts (c) (Burton and Grimshaw ; McNally ). ()

a. Nobodyj will < [VP laugh ] and [VP be arrested tj ] > b. < [ Nobody will laugh ] and [ nobodyj will be arrested tj ] > c. Nobodyj will < [VP tj laugh ] and [VP be arrested tj ] >

ATB movement out of a low coordination has also been proposed to account for PCCs like () (Larson : ); and also for medial gapping (Johnson b, ), as sketched

17 Cases of (apparent) coordination that do not obey CSC, such as What did he turn round and say to you?, already noted by Ross () and later discussed by Goldsmith (), Lakoff (), Postal (), and others, are not considered here. 18 Ross’s (: ) rule (“Conjunction Reduction”) was responsible for both RNR and leftward ATB movement.

   - 



in (). Johnson’s analysis assumes that the CSC does not govern movement of the subject. See Chapter . ()

The girlsj are sleepingk < [VP tj tk here ] and [VP Peter tk there ] >.

.. Alternative: Flexible constituency The case for structural sharing approaches to non-constituent conjuncts seen in RNR, PCC, etc. is founded on the assumption that the pronounced non-shared material in the relevant conjunct does not form a constituent on its own; that there is, for instance, no wellformed syntactic constituent containing Mary praised in () that does not also contain a direct object as sister to the verb, unpronounced in that position, whether as a movement trace or target of ellipsis. ()

[ Mary [ praised John ] ] and Cathy criticized John.

An alternative hypothesis is that such conjoined strings do in fact correspond to single constituents, albeit under a non-standard view of what counts as a constituent. Such an approach is pursued in categorial frameworks, for instance Combinatory Categorial Grammar (CCG) (see Steedman , ; Steedman and Baldridge ).19 CCG syntax is based on categories defined as functions or arguments, with the possibility for the latter to be shifted to higher functional types. Complex units are formed by combinatory operations such as function application or function composition, directionally specified; cf. the sample derivations in ()–() (derived categories in bold).20 While basic categories combine in ways consistent with standard constituency as in (), typeshifting makes alternative derivations available that generate non-standard constituents as in () (based on Steedman : ff.). () Mary NP

() [S/NP

[S\NP

praised (S\NP)/NP S\NP S

Mary NP S/(S\NP)

praised

John ] NP S\NP from forward function application S from backward function application ]John

(S\NP)/NP S/NP NP S

S/(S\NP) from typeshifting the subject NP S/NP from forward function composition S from forward function application

Proposals for direct coordination accounts of PCC in terms of flexible constituency also include Pesetsky’s () “layers and cascades” and Phillips’ (, ) constituency-altering derivational merge approach, which also extends to RNR. 20 X/Y signals a function that combines with Y to its right to give X; while X\Y is one that combines with Y to its left to give X, where X and Y can both be function types. (S\NP)/NP is the category for a finite verb in an SVO language. 19





Non-standard constituents such as [Mary praised] in () are then available to be coordinated and subsequently combined with an object NP (), which forms the basis of semantic composition, to give a direct account of the rightward ATB movement effect. ()

[ [S/NP Mary praised ] and [S/NP Cathy criticized ] ] John S/NP NP S

S/NP from conjunction S from forward function application

A similar approach is taken to PCC. A double object VP can be derived as [[ V NP ] NP ] by forward function application, or the two NPs can be typeshifted so as to be able to combine by function composition to give a constituent that combines with the ditransitive verb as [ V [ NP NP ]]. The latter permits a direct analysis of a PCC (Steedman : ). For an account of medial gapping, see Steedman : ff. In short, the syntax of all coordination is given a direct analysis, involving neither movement traces nor ellipsis. A basic feature of CCG and related approaches is that “everything that can coordinate . . . is a constituent under the generalized definition of that notion that is afforded by categorial grammars” (Steedman : ). The same flexible notion of constituency is claimed to capture prosodic (intonation) units (Steedman : ch. ). On the other hand, the CCG notion of constituency is not constrained by traditional notions of dominance and c-command which underlie mainstream approaches. Flexible constituency approaches are not considered further here.21

.. Other kinds of motivation for CR Motivations for CR are not limited to considerations pertaining to adjacency and constituenthood of conjuncts and the LCL, but extend for instance to assumptions about how meaning relates to structure, and about how agreement works. Two cases which appear at first blush consistent with a direct analysis illustrate the point.22 McCawley’s () example () appears, like (), to fit a pattern [DP < X Conj Y > α ] in which X and Y corresponds to a coordination of APs modifying a single head noun. Yet in 21

The CCG approach outlined by Steedman suggests a basic right–left symmetry for ATB constructions. RNR is claimed to be island-constrained like leftward movement (Steedman : ), contra Wexler and Culicover () and many since (see ..). 22 A case which does not motivate a CR approach is the phenomenon of respectively readings in coordination, whereby (i) is paraphrased by (ii). (i) Alex and Bill met Yvonne and Zoe (respectively). (ii) Alex met Yvonne and Bill met Zoe. These readings have been described as fundamentally problematic for the syntax of coordination (Postal : ). In the logic of early Transformational Grammar, the paraphrases suggest a clausal coordination source, and syntactic solutions that assign sentences like (i) (in the relevant reading) a biclausal syntax have been suggested, e.g. by Goodall (). However, the fact that such readings also arise with non-conjoined plurals, as in The first two boys met the last two girls (respectively), indicates that their source is not to be found in coordination syntax (via some kind of CR), but in semantics (e.g., a covert distributive operator). See Gawron and Kehler () and Chaves ().

   - 



the case of (), interpretation (two kinds of knowledge) and number agreement (a plural verb agrees with an and-coordination of non-count singular DPs but not with a single such DP) indicate larger conjuncts, two DPs, each with its own head noun, reduced by backward CR. ()

a. Accurate and helpful information is hard to come by. b. [DP < AP & AP > N. ] is . . .

() a. Historical and scientific knowledge are completely different from each other. b. < [DP historical knowledge. ] and [DP scientific knowledge. ] > are . . . Sentences with corrective but provide another example. In () with and, negation is naturally interpreted as taking scope over the coordination, as reflected in the structure under the direct analysis (a). With but, however, the negation’s scope in () is restricted to the first conjunct, suggesting that this case must involve larger conjuncts—which in turn suggests forward CR (b) (Vicente b; Toosarvandani a). ()

He hasn’t drunk beer {and/but} champagne.

()

a. He hasn’t drunk < [DP beer] and [DP champagne] >. b. < [TP He hasn’t drunk beer ] but [TP he has drunk champagne ] ] >.

. F CR

.................................................................................................................................. If CR is a species of ellipsis, one expects it to share basic properties with canonical (noncoordinate) ellipsis types (verb phrase ellipsis, sluicing, noun phrase ellipsis). This section begins by considering the relation between forward CR (stripping, PCC) and canonical ellipsis, and then discusses the ‘move-and-elide’ approach.

.. Forward CR and non-coordinate ellipsis The canonical non-coordinate ellipsis types involve omission of a single constituent. In the context of a local head of a specific type, shown as X in the configuration (), YP can be elided (Lobeck ; Merchant , a). ()

. . . [X [YP . . . ] ] . . . where YP = NP in DP (NPE), VP (VPE), or TP (Sluicing)

Approximately, X stands for an interrogative wh-complementizer (sluicing), certain auxiliaries (VPE), or certain determiners, including numerals and possessors, but not the or every (NPE). The ellipsis site is a “surface anaphor” (Hankamer and Sag ), one formed during and interacting with the syntactic derivation (see van Craenenbroeck and Merchant  for an





overview). According to the phonological deletion approach championed by Merchant and others, the ellipsis site contains a fully fledged syntactic structure, whose subparts show syntactic interactions with elements external to the ellipsis, including movement out of the ellipsis site (). The ellipsis itself arises through suppression of the pronunciation of the constituent in question.23 ()

a. John was arrested because Mj was [ arrested tj ]. b. This book, Mary’s read, but [that one]j, she ’ [ read tj ].

The forward CR approach to stripping and PCC sketched in . invokes conjuncts of variable size containing a focused phrase (the pronounced remnant), with the remainder, i.e. shared material, elided by a specific type of ellipsis rule, applying in ATB fashion in non-initial conjuncts (Forward Deletion / FWD in Wilder ). FWD involves ellipsis of phrases or individual words in the second conjunct, under identity with parallel material in the initial conjunct, leaving solely the focused element(s) as remnants (cf. also Hartmann ). The idea that forward CR is of a kind with ellipsis types such as VPE24 finds support in patterns of non-identity of forms known as “vehicle change” effects (Fiengo and May ). As documented in the literature on VPE and sluicing, an ellipsis target may contain forms that differ from corresponding forms in the antecedent, as long as the semantic identity condition (Merchant’s  e-ness) is satisfied. The VP elided in (a) (italicized) contains a pronoun (as required by Binding Theory) where the antecedent VP contains a name. The antecedent and elided VPs of (b) contain bound (sloppy) pronouns which take different forms, as required by conditions on pronominal agreement. ()

a. They arrested Alexj, though hej thought they ’ [arrest himj/*Alexj ] b. Mary lost her way, and Jj did [lose { hisj/*herj } way ], .

The same effects can be observed with stripping and PCC ()—cf. inform himj about {hisj/*Johnj’s} results, and put the dog out of {its/*their} misery. ()

a. We should inform M about Johnj’s results and j, too. b. They’ll put the  out of their misery , and the  .

Verbal agreement may also diverge in stripping (a), an effect replicated in VPE (b) (Fiengo and May : ).

23 In Merchant’s () proposal, phonological deletion is triggered by an “E-feature” introduced by X in (), which simultaneously imposes the requirement for the elided YP to stand in a suitable semantic relation (“e-”) with an antecedent. 24 A central (and hitherto unanswered) question facing attempts to assimilate forward CR to regular ellipsis is why with forward CR, unlike the canonical ellipsis types, the antecedent–ellipsis relation is coordination-bound. See Johnson () for relevant discussion with regard to gapping.

   -  ()



a. The  are happy, and J [is/*are happy], too. b. The girlsj said theyj were happy, but Johnk didn’t [ say hek was/*were happy ].

Another vehicle change effect concerns indefinites (any-some, no-a, etc.) in polarityswitching VPE () (Johnson : –; Merchant b). This effect is not detectable in English stripping (perhaps because the polarity switch that would trigger it depends on the presence of an emphatic finite auxiliary in that language), but it can be observed in German. In (), the polarity switch is marked by the contrast between negation (in niemanden ‘nobody’) and the particle schon (which signals the emphatic affirmative). ()

a. John didn’t see anyone, but Mary did. (= see someone) b. I could find no solution, but Holly might. (= find a solution)

() a. Wir haben niemanden gesehen, we have nobody seen ‘We didn’t see anybody, but he did.’

aber but

b. . . . aber  hat  jemanden but he has  someone. ‘ . . . but he did see someone.’

 he

. 

gesehen seen

Furthermore, like canonical ellipsis, stripping and PCC show interaction with movement; both ellipsis sites () and remnants () may contain the trace of an item extracted (ATB) out of the coordination. () a. the children whoj he either < [sold  to tj ] or [sold  to tj ] > b. the place wherej Bill < [left his  tj on M ] and [left his  tj on T] > () a. We must work to become the place where companies are not moving Headquarters out of—but into. (Web example) b. the room whichj they < [moved J [ of tj ] ] and [moved M [ tj ] ] > In sum, forward CR shares much of the empirical profile of canonical ellipses.

.. Move-and-elide In the FWD conception, elliptical conjuncts have the same syntax as corresponding nonelliptical expressions, involving no special operations except for the elision of given material. Unlike canonical ellipsis types, what is targeted by FWD evidently need not be a single constituent (e.g. (a) involves non-adjacent gaps). However, not just any given material may be elided (cf. *He bought the book and  the book); conversely, not just any part of the second conjunct can surface as a remnant in an elliptical non-initial





conjunct, and difficulties arise in accounting for the attested patterns under the FWD approach.25 An alternative move-and-elide approach, sketched in (), is proposed for stripping in Merchant (a), building on Sag (), Depiante (), and others. ()

. . . &[XP ZPj [YP . . . tj . . . ] ] . . .

The key idea is that ZP (the focused remnant) is moved to the periphery of the conjunct (XP). What is elided is a single constituent (YP in ()), containing the trace of ZP.26 ()

< [ I bought   yesterday], and [a j [ I bought tj yesterday ] ] >

Stripping is thereby aligned with a number of other constructions analysed in terms of move-and-elide, including sluicing (Merchant ), pseudogapping (Jayaseelan ; Lasnik a), and fragment answers (Merchant a). The movement of ZP correlates with A'-movement to the clausal periphery. As discussed in .., the conjuncts in which stripping applies are in many cases smaller than a full clause (a point emphasized by Toosarvandani a). The account thus needs to be formulated in terms of A'-movement of the focused remnant to the conjunct periphery, whatever its category (CP, TP, vP, etc.), even if that movement pattern does not surface in non-coordinate environments (b). () a. Nobody < [ bought a book yesterday ], or [ [a ]j [ bought tj yesterday ] >. b. *Nobody a j bought tj yesterday. The move-and-elide approach has the twin advantages of predicting what can be a remnant (namely, any constituent within the conjunct which can undergo A'-movement to its periphery), and what can be elided (namely, the constituent YP). A further conceptual attraction is that it brings ellipsis in stripping into line with non-coordinate ellipses; deletion targets a single constituent in each case. The prediction that the class of stripping remnants should correspond to all and only (focusable) units that are licit targets for A'-movement is borne out widely. For example, a remnant may not correspond to a phrase contained in an island (b); but the island itself

25 Attempts at characterizing the possibilities are Hankamer’s () Major Constituent Condition on remnants, combined with a Head Condition prohibiting deletion of material c-commanded by an overt head (Wilder ). 26 The idea can be implemented in the Minimalist framework by assuming an abstract head X which attracts ZP, and carries the E-feature which simultaneously imposes the e- requirement on YP and triggers its phonological deletion. If the LCL is interpreted in terms of syntactic categories, this kind of solution would require the initial conjunct also to be an XP, even though the correlate to the focused remnant typically remains in situ.

(i) . . . < [XP . . . ] & [XP ZP X [YP . . . tZP . . . ] ] > . . .

   - 



may front, giving (c).27 Nor may the remnant be a unit that resists A'-movement, such as one headed by a finite auxiliary (b), (b), as opposed to a full CP (compare (a) and (a), which allow a direct, non-elliptical coordination parse). Nor may it be the object of a preposition in a language, such as German (b), that does not have preposition stranding. ()

a. He claimed that there will be a  yesterday, and , too. b. *He complained because there will be a  yesterday, and , too. c. He complained because there will be a  yesterday, and because there will be , too.

()

a. They recommended that Ed be fired and Kim be promoted in their report. b. They recommended that Ed be fired in their report, and *(that) Kim be promoted.

()

a. I didn’t say that he should see a doctor or must visit the dentist. b. I didn’t say that he should see a doctor in my letter or *(that he) must visit the dentist.

()

a. I spoke with the mayor yesterday and not (with) the chief of police. b. Ich habe mit dem Bürgermeister gesprochen I have with the mayor spoken und nicht *(mit) dem Polizeipräsidenten. and not with the police-chief

Cases like *He has bought the book and  the book also fall into place. The verb cannot raise alone (remnants arise by phrasal A'-movement), and the VP minus its object cannot be an A'-moved remnant either, in English at least (cf. *Read he has the book). The hypothesis that stripping remnants have moved is strongly supported by such patterns.28 Move-and-elide has further been argued to apply in PCCs (Sailor and Thoms ). Similar evidence motivates movement of remnants in medial gapping (Neijt ; see also Chapter ). It seems worth exploring whether the UCC types () and () (see ..) can be captured as a kind of stripping under move-and-elide. That approach may also shed new light on UCCs like () (see n. ). ()

You can depend on my assistant and that he will be on time. (Sag et al. )

For whatever reason, English does not tolerate a that-clause in the complement of a preposition (*You can depend on that my assistant will be on time)—but the clause may surface if it Sluicing can in certain cases remedy island violations (cf. Merchant ), and the question arises of why the effect is found with some kinds of ellipsis but not others. See Griffiths and Liptak () for discussion. 28 A case that deserves closer scrutiny is Reinhart’s () examples involving a remnant extracted from a subject, as in (i): 27

(i) Jokes about F were told, and F, . (ii) *Felixj, jokes about tj were told.





has moved (That my assistant will be on time, you can depend on). A move-and-elide derivation would generate () with the that-clause as a stripping remnant (): ()

. . . and [ [that he will be on time]j (you can) depend on tj ]

The movement analysis of stripping and PCC remnants raises issues requiring further study (see also n. ). For example, ATB extraction is possible from the ellipsis site (cf. ()); () shows the move-and-elide structure. Focus movement of the remnant to the conjunct periphery might be expected to create an intervention configuration, i.e. turn the conjunct into an island for extraction out of the coordination. ()

the children whoj he either < [sold  to tj ] or [ []k [ sold tk to tj ] ] >

Also, remnant movement creates a classic “freezing” configuration (see Corver ) in which the remnant itself becomes an island. Yet extraction (ATB) from a remnant is apparently possible; cf. (). Whether the move-and-elide approach can extend to other forward CR cases is an open question. Stripping is not confined to clausal or verbal coordinations; () illustrates its occurrence in coordinations of nominals, a phenomenon on which the literature is largely silent (but see Neijt : ). ()

a. The sale of alcohol on these premises, or of any tobacco product, is forbidden. b. Her reliance on Bill and that he would arrive on time proved to be her downfall.

Another case involves the sharing of the initial part of a compound word in the pattern () (Booij ; Chaves ). If coordination below the word level is rejected, a direct analysis is not possible; yet at the same time, given that parts of a word are not moveable elements, movement and ellipsis within larger conjuncts is also ruled out.29 ()

a. < [word α . . . ] & [word α . . . ] > b. Half-brothers and -sisters are considered the same as full brothers and sisters.

. RNR

.................................................................................................................................. According to Ross’s () proposal, RNR is similar to leftward ATB movement. Although the shared element α is introduced inside the conjuncts, it surfaces outside the coordination. A CR approach (a) thus competes with an ATB movement analysis (b), according to which α has undergone extraction from each of the conjuncts. 29 This pattern appears to be restricted to coordinations, unlike its mirror image RNR counterpart (cf. Chaves : ); compare ?distinguishing neurolinguistics from psycholinguistics and *distinguishing half-brothers from half-sisters. A similar pattern in Japanese is discussed under the rubric of Left-Node Raising by Yatabe ().

   -  ()



a. . . . < [ . . . α ] & [ . . . α ] > . . . b. . . . < [ . . . tα ] & [ . . . tα ] > α . . .

RNR displays properties that differ sharply from those of forward CR cases analysed in terms of ellipsis, and also from leftward movement, including leftward ATB movement. These special RNR properties pose puzzles for both ATB movement and ellipsis approaches (see .. and ..). .. discusses a third approach, according to which shared material in the RNR pattern arises through Multiple Dominance (MD). Two related and understudied phenomena, non-coordinate RNR and the mysterious Right-Node Wrap configuration, are addressed in .. and ...

.. RNR as rightward ATB movement The rightward movement approach is what Abels () terms an ex situ analysis; it makes the assumption in (). ()

The RN (i.e. the shared element α in RNR) is outside the coordination (since it has moved out of the coordination).

The movement analysis is natural in a model which assumes rightward and leftward movement operations, the CSC, and ATB movement. It directly captures the ATB nature of RNR (cf. ()–()); if α moves out of the coordination, it is subject to the CSC. Another potential advantage is the prediction that RNRed quantifier expressions can scope over the coordination. Sabbagh () cites contrasts such as that in (), where (a), unlike (b), allows the distributive (‘different ways’) reading (a) expected if every patient takes scope over or. ()

a. She will either speak to or email every patient. b. She will either speak to every patient or email every patient.

() a. ‘Patients may be contacted in different ways (or all in the same way).’ ( > ) b. ‘All patients will be contacted in the same way (not in different ways).’ ( > ) Under the movement approach, the wide scope of the RN follows as a direct consequence of its derived position, from which it takes scope over the coordination. A narrow-scope interpretation of an RN may be explained via the possibility for reconstruction, similar to what is found with leftward ATB movement (Höhle ). See also ....

.. Special RNR properties Various traits of RNR are unexpected from the viewpoint of the ATB movement approach. These relate to (i) general linear properties of the RNR configuration, (ii) possible targets of RNR, and (iii) RNR–ellipsis interaction.





... Linear properties: String-vacuous nature, Right-Edge Restriction A hallmark of movement, rightward and leftward, is that it typically results in a change in the order of constituents. For leftward ATB movement, that includes reordering the shared item α with respect to material X outside of the coordination (). RNR never reorders α with respect to elements external to the coordination; cf. (). If RNR is rightward ATB movement, it is string-vacuous—unlike its leftward counterpart, and unlike non-ATB rightward displacements (Extraposition, Heavy NP Shift). ()

a. α X < [ . . . tα . . . ] & [ . . . tα . . . ] > b. Batmanj he said that < Sue likes tj and Mary hates tj >.

()

a. *< [ . . . tα . . . ] [ . . . tα . . . ] > X α b. *He said that < Sue likes tj and Mary hates tj > yesterday Batman and Robinj.

()

String-vacuous nature of RNR: RNR of α never results in α being placed to the right of a conjunct-external element.30

The ex situ assumption () may be compatible with (), but the absence of cases like (b) means that the order facts are also compatible with the in situ assumption that the RN is inside (has not moved out of) the final conjunct (Abels ). Unlike leftward ATB movement (and forward CR), RNR is subject to an edge restriction on the placement of RN gaps (Wilder ).31 ()

Right-Edge Restriction (RER) on RNR gaps Gaps corresponding to an RN must be at the right edge of their non-final conjuncts.

The RER is illustrated by the contrast in () (cf. Oehrle ). The object of invite in (a) can appear at the right edge of its conjunct following the PP, i.e. it can undergo Heavy NP Shift (cf. I invited into my house all the winners). The first object in a double object construction resists Heavy NP Shift (*I gave a present all the winners); hence the gap in the first conjunct of (b) cannot satisfy (). ()

a. I [ invited into my house _ ] and congratulated all the winners. b. *I [ gave _ a present ] and congratulated all the winners.

The RN itself also shows a right-edge effect. Under the movement analysis, the contrast in () suggests that the rightmost position in the final conjunct must be a licit position for α if it is to undergo RNR.32 30 Kluck and de Vries () argue that RNR can feed non-string-vacuous rightward processes such as extraposition of nominal modifiers. 31 Postal (:  and , n. ) presents examples as grammatical that do not conform to RER. His judgements are contested by Levine (: ) and others. 32 Cases where the shared element surfaces in a non-final position in the final conjunct are discussed in ...

   -  ()



a. I congratulated tj and [ invited tj into my house tj ] all the winnersj. b. *I congratulated tj and [ gave tj a present (*tj) ] all the winnersj.

Properties () and () govern the RNR pattern across languages. Comparison of VO and OV languages is instructive; () and () interact with the word order of the language to produce different choices of RN, as illustrated for German in (). Verb-final contexts permit RNR of V or O+V, but RNR of the object alone is possible only if it is in final position.33 () a. Ich werde Wein bestellen und Susi wird I will wine order and Susi will ‘I will order wine, and Susi will order mint tea.’

Pfefferminztee mint.tea

bestellen. order

b. Ich sollte Pfefferminztee trinken und Susi wird Pfefferminztee trinken. I should mint.tea drink and Susi will mint.tea drink ‘I ought to, and Susi will drink mint tea.’ c. Ich hasse Pfefferminztee und I hate mint.tea and ‘I hate and Susi loves mint tea.’

Susi Susi

liebt loves

Pfefferminztee. mint.tea

An account within the ATB movement approach is developed by Sabbagh (). His theory aims to derive both the RER and string vacuity of RNR from the way movement interacts with linearization (see his work for details).

... RNR targets Inspection of possible RNs reveals a range of choices that are unexpected if RNR is movement. Units which can be RNs but undergo neither leftward movement (ATB or otherwise) nor simple rightward shift include: TP, stranding complementizers (a) (Bresnan ); T’/VP (b,c), with do-support not triggered, unlike in leftward movement; nominals, stranding predeterminers like every (d) or adjectives (cf. (b) and ()); and parts of words (e) (cf. also ()), (Booij ; Wilder ; Chaves ). ()

a. b. c. d. e.

They were discussing whether _, but wouldn’t state that, your theory is correct. John said that Mary _ and Bill said that Sue must see a doctor. John said that Mary _ and Bill said that Sue occupies this office. I don’t wash my hair every _ but every second day. (Web example) His theory over-_ and my theory under-generates. (Wilder )

RNR can also target the complement of P, stranding P—a pattern unexpected in English, since rightward DP-shift cannot strand P otherwise; and more so in languages like Irish (McCloskey ) or German (b) which otherwise generally disallow P-stranding. () a. John spoke to _ but Mary ignored the new boss. 33

This suggests that backwards V-gapping, famously restricted to SOV languages (SO&SOV vs *SO&SVO; Ross ), is RNR.



 b. Das the

Buch book

liegt lies

auf _ on

und and

die the

Zeitung newspaper

liegt lies

unter under

dem the

Sofa. sofa

The fact that RNR can target an element contained in an island inside one or both conjuncts (Wexler and Culicover ) also calls into question the idea that RNR involves movement out of the coordination, even if that element is mobile otherwise. ()

. . . < [ . . . [ISLAND . . . tα ] ] & [ . . . [ISLAND . . . tα ] ] > α

() a. I know [the man who buys _ ] and you know [the man who sells _ ] pictures of Elvis Presley. b. Politicians win [when they defend _], and lose [when they attack _] the right of a woman to an abortion. (Sabbagh ) In ()–(), the RNRed string does not correspond to a single constituent. Such cases have been taken to indicate that RNR can target a sequence of constituents at the right edge of the conjuncts. While (a) involves multiple moveable elements, (b) and () certainly do not. ()

a. Mary baked _, and George frosted,  cakes in less than an hour. (Abbott ) b. a positively _ and a negatively charged electrode (Wilder )

() Er hat einen Mann, der drei Hunde, und sie hat he has a man who three dogs and she has eine Frau, die drei Katzen besitzt, gekannt. a woman who three cats owns known ‘He knew a man who owns three dogs and she knew a woman who owns three cats.’ (German, Wesche ) In (), RNR targets the verbs of the main clause and a relative clause, which form a clausefinal sequence due to the OV nature of the language (Ha :  and Kubota :  give similar examples from Korean and Japanese respectively). Such data contribute to the case for considering an in situ approach.34

... Interaction of RNR with leftward movement In () (following standard analyses of German V), the shared verbs are followed by the trace of a non-shared auxiliary hat ‘has’ inside each conjunct, cf. (). ()

[ Er hatj einen Mann, der drei Hunde besitzt, gekannt tj ] und [ sie hatk eine Frau, die drei Katzen besitzt, gekannt tk ]

34 Sabbagh () offers an account for why () should be possible, i.e., why RNR does not respect islands. His approach does not, however, afford insight into other surprising RNR targets, including sublexical units and multiple constituent sequences.

   - 



This illustrates the fact that only overt material counts for the RER. Distinct (non-shared) null elements such as the traces in () or () do not cause RER violations. ()

< [ The drugsj John gave those poor kids tj ] and [ the picturesk Mary showed those poor kids tk ] > were illegal.

To the extent that silent elements can be said to figure in a linear string, a better formulation of () would be that ‘no RNR gap may be followed by a pronounced non-shared element in its conjunct’. The fact that only overt elements count is a further indicator that linearization principles are what underlie the restriction. Non-shared material can be extracted from a position inside the shared string, such as the two wh-phrases in (), or the auxiliary belonging to the first conjunct (). ()

I wonder < whoj the  _ and whok the  _> will accuse { tj/tk } of the crime.

() < Cj _, and will she  _ > John ( tj ) attend the meeting? (Wilder ) Such patterns are tricky to accommodate if RNR is treated as ATB movement of a single constituent (Grosz ).35 Under an ellipsis analysis, ()–() are relevant for the question of what counts as ‘identical with the antecedent’, since there is a mismatch between the RN and the gap with respect to null elements contained within them. () a. < [ whoj the  will accuse tj of the crime ] and [ whok the  will accuse tk of the crime ] > b. < [ Canj John tj attend the meeting ] and [ will she make John attend the meeting ] > ?

... Interaction of RNR with ellipsis Another asymmetry between leftward ATB movement and RNR is pointed out by Abels (). In contrast with leftward movement, an RNRed element cannot survive independent ellipsis of the VP containing its original position. In the configuration (a), the VP of the second conjunct can be elided (b), while the example corresponding to (b) is impossible (constructing test examples with P-stranding eliminates a potential pseudogapping source). ()

a. α . . . < [ . . . [VP . . . tα . . . ]] & [ . . . [VP(elided) . . . tα . . . ]] > (leftward ATB movement) b. * . . . < [ . . . [VP . . . tα . . . ]] & [ . . . [VP(elided) . . . tα . . . ]] > α (RNR)

35

A similar problem arises in leftward ATB cases like (i), with a single fronted constituent apparently containing the traces of the non-shared subjects John and Mary. See .... (i) How likely {tj/tk} to abscond was Johnj _ and is Maryk _?





()

a. Their achievements, Jane talked about t and Frank didn’t talk about t. b. Their achievements, Jane talked about t and Frank didn’t [ talk about t ].

()

a. < Jane talked about t and Frank didn’t talk about t > their achievements. b. *< Jane talked about t and Frank didn’t [ talk about t ] > their achievements.

If the RN in (a) is outside the coordination, ellipsis ought to be possible just as it is in (). That this is not so suggests that in fact the RN remains inside its conjunct.36

.. RNR as ellipsis The facts reviewed above are taken by many to indicate that () is mistaken, and that an in situ analysis of RNR is more likely to be correct. One possible in situ approach is an ellipsis analysis (a), whereby each conjunct contains a (syntactically independent) token of α, with all but the final token undergoing phonological deletion, under identity with the pronounced token. As such, RNR shows a directionality opposite to that of both forward CR and of canonical ellipsis. There are other significant differences between RNR and canonical ellipsis phenomena (NPE, VPE, sluicing). One concerns the grammatical type of the gap, which in each of NPE, VPE, and sluicing is a specific category (NP, VP, TP). RNR is promiscuous—there is quite possibly no category type that cannot undergo RNR, once independent factors such as prosody are controlled for.37 Another concerns identity. Forward CR shows form identity mismatches largely in line with those found in canonical ellipsis; RNR does not (see ...). In addition, certain types of RN are at odds with the assumption of two independent syntactic tokens (see ... and ...). Focus is seen by some to play a crucial role in RNR. Certainly, RNR involves semantic contrast between the non-shared parts of the conjuncts, which typically leads to material preceding the gap and the RN being focused. Hartmann () proposes that focus conditions the phonological deletion of the RNRed material in non-final conjuncts. Ha () suggests that the RNR is syntactically conditioned ellipsis, and that the context for RNR is identified by a syntactically encoded contrastive focus which immediately precedes the gap in each non-final conjunct. Chaves () presents arguments against taking (contrastive) focus to play a defining grammatical role in RNR. () illustrates that RNR gaps need not be directly preceded by contrastively focused elements.

36

This argument is not addressed in Sabbagh (). The failure of RNR to interact with sluicing points in the same direction: (i) I know why we talked about _ but I don’t remember when we talked about their achievements. (ii) *I know why we talked about _ but I don’t remember when _ their achievements. 37 Postal (: –) and others have proposed that certain types of expression cannot undergo RNR, whether for syntactic or prosodic reasons. Chaves (: sections .–.) gives some counterarguments.

   -  ()



My mother blushed at this small lie because  knew and  knew that the roosters had already been paid for. (Chaves : )

... Form identity RNR generally does not tolerate form mismatches (vehicle change) of the type observed in forward CR and VPE etc., discussed in ... It appears that the form of the RN must satisfy form requirements imposed by both conjuncts (Wilder ) (exceptions are discussed below). Firstly, an R-expression contained in the RN cannot corefer with a pronoun c-commanding the RNR gap in the first conjunct (Levine ): ()

*Shej said _ and I happen to agree that Maryj needs a new car.

Secondly, while an RN can contain bound (sloppy) pronouns (Höhle ; Jacobson ), mismatch of forms leads to ill-formedness (Chaves ), unlike in VPE and forward CR. ()

a. Chrisj likes hisj bike and Billk loves hisk bike. b. *Chrisj likes hisj bike and Suek loves herk bike.

Thirdly, mismatching verb agreement forms cause ill-formedness, though morphosyntactic feature mismatches are tolerated in the case of syncretism, i.e. if the form is compatible with requirements in both conjuncts (Eisenberg ; Wilder ). ()

a. *John said that I, and Mary said that she, is the best swimmer. b. John said that I, and Mary said that she, was the best swimmer.

Similar examples involving other verb forms in English, and case forms in German and other languages are discussed in Pullum and Zwicky () and Chaves (). These facts indicate that, if RNR is a kind of ellipsis, then it is subject to a different kind of identity requirement than forward ellipsis. Wilder () suggests that RNR involves deletion at a level (in terms of the Minimalist model, following Spell-Out and operations of inflectional morphology) at which grammatical categories are no longer relevant. Chaves proposes an operation of ‘backward peripheral deletion’ which ‘imposes morph form identity conditions’ (Chaves : ). A puzzling counterexample to the identity generalization is seen in () (Kayne : ). ()

a. Mary bought but John didn’t buy any books about linguistics. b. *Mary didn’t buy but John did buy any books about linguistics.

The pattern resembles the some-any vehicle change found in VPE, yet the shared constituent is a DP, not a regular ellipsis target in English. Chaves (: –) suggests that the RN in (a) is not the whole DP but the bare plural books about linguistics following any. However, natural-sounding instances of the pattern can be found which lack a suitable RN following any (examples from the Web).





()

a. [He] went into the library where he shot at but didn’t hit anybody. b. I asked for, but didn’t receive any refund [ . . . ] c. ‘What? There is no Thomas Haden Church page?’ That’s what I thought when I searched for and didn’t find any.

... A total of, same, different Ellipsis accounts break down in the face of examples like (a), involving relational adjectives like same, different (Jackendoff ), and what Chaves () calls the additive reading of (b). ()

a. John hummed, and Mary whistled, different tunes. b. I borrowed, and my sisters stole, a total of $. (Abbott )

()

a. John hummed different tunes, and Mary whistled different tunes. b. I borrowed a total of $, and my sisters stole a total of $.

In the reading in which John’s tune was different from Mary’s tune (Carlson’s  internal reading of different), (a) is not equivalent to its undeleted counterpart (a). Likewise, (b) lacks the prominent reading of (b) according to which the sum of the money borrowed and stolen was $. An analogous difference was described in .. with respect to quantifier scope. The source of the problem these examples pose is that the RN takes scope over the remainder of the coordination. This in and of itself does not force the conclusion that the RN is syntactically outside the coordination (Abels ). But ellipsis approaches suppose that each conjunct contains a syntactically independent token of the RNRed phrase; an RNR example is syntactically the same as its counterpart without deletion. It is unclear what syntactic or semantic mechanism could be responsible for the wide scope of the RN, and why such a mechanism should not be available in the absence of the deletion. These cases, in short, are highly problematic for in situ deletion theories such as Wilder () or Hartmann (b).

... Summative agreement The phenomenon of summative agreement (Yatabe ), illustrated in (), was observed in English by Postal (: ) and in German by Schwabe and Heusinger (). The RNRed string includes a plural verb which agrees with non-shared singular subjects that are not themselves coordinated but each in a separate larger conjunct. ()

The pilot claimed that the first nurse, and the sailor proved that the second nurse, were spies.

The pattern appears to exist in some but not all languages/varieties and for some but not all speakers (Grosz ). It clearly presents a problem for a deletion account, in which each subject would be in a syntactic agreement relation with a separate token of the plural verb. Postal (), Yatabe (), and Chaves () conclude that the pattern is to be handled

   - 



via an ex situ analysis of the RNRed string in which some special mechanism provides a plurality for the verb agreement. Other phenomena involving a plural form in the RN related to singular non-shared elements, illustrated by (), cause similar problems for in situ deletion accounts: ()

The Czechj and the Slovakk republicsj+k are very similar.

Relative clauses with split antecedents (Perlmutter and Ross ), insofar as it is correct to suppose that they instantiate RNR, pose a similar problem. ()

Mary met a manj and John met a womank whoj+k know each other well.

... Backwards ellipsis vs RNR Canonical ellipses (sluicing, VPE, NPE) usually follow their antecedents, but backward ellipsis is also possible. That raises the question of whether some apparent RNR gaps (specifically, RNR of TP, VP, or NP) might have an ellipsis source in addition to or instead of an RNR source. The configuration of interest is (), where YP = TP, VP, or NP, the gap is final in its conjunct, and the content of the gap is identical with overt final YP in the second conjunct. ()

< [ . . . [YP gap ]] & [ . . . YP ] >

Barros and Vicente () propose that some cases of () are derived by backward ellipsis, discussing YP=VP in English. Chaves () makes a similar claim, discussing examples where YP= NP and YP=TP as well. Sluicing, NPE, and VPE are all licensed by specific choices of X in the left-hand context (). ()

. . . [ X [YP elided constituent ] ] . . .

Certain choices of X only occur where YP undergoes ellipsis, i.e. do not permit a nonelliptical YP. These can be deployed to probe for genuine backward ellipsis, and to distinguish ellipsis from RNR gaps. Given the form identity condition on RNR (see ...), an RNR gap should behave like overt YP in this respect. In English, the inversion of a preposition and wh-word (swiping) is an instance of X that only occurs in sluicing; swiping does not occur when the TP is not elided (Merchant ).38

Giannakidou and Merchant () analysed examples like (i) as “reverse sluicing” (i.e., backward ellipsis). However, if does not license regular (forward) sluicing (ii), and (i) is better analysed as RNR. 38

(i) It’s not clear if __ and when the police arrested the demonstrators. (ii) *She said they arrested the demonstrators but it’s not clear if __. The following exemplifies backward sluicing, evidenced by wiping (Merchant : ): (iii) Although we don’t yet know who from, we know she received a package last Monday.





()

a. He opened it somehow, but I don’t know what with (*he opened it). b. What with (*did he open it)?

Gračanin-Yüksek () analyses wh&wh constructions (What and when did he eat?, cf. (c)) as a form of RNR. Chaves () suggests similar examples are derived by ellipsis (backwards sluicing) rather than RNR. However, as Gračanin-Yüksek notes (pp. –), swiping is degraded in this construction, implying that it does not have a sluicing source. ()

{ How and when / ??What with and when } did he open it?

The same contrast is seen in cases involving larger conjuncts; the deviance of (b) suggests that (a) does not have a sluicing source: ()

a. Sue knows how _ and Bill knows when he opened it. b. ??Sue knows what with _ and Bill knows when he opened it.

English NPE is licensed in the complement of prenominal possessors (), but in the case of possessor pronouns, a special ellipsis form (mine, yours, theirs) replaces the regular my, your, their. ()

a. Fred got Judy’s signature, but not { John’s/mine/yours/*my/*your } _. b. *Fred got {mine/yours} signature.

This alternation directly distinguishes RNR gaps from NPE. (a) (from the Web) illustrates a possessed NP in the RNR configuration where the gap follows the regular pronoun. The ellipsis form is degraded (b). ()

a. The most frequent mistake in creating a website is creating it from your _ rather than from your customer’s perspective. b. ?? . . . from yours _ rather than from your customer’s perspective.

German allows NPE in the complement of quantifiers like ein ‘one/a’, kein ‘no’, and pronominal possessors (mein ‘my’, sein ‘his’, etc.). These also have a special ellipsis form: the regular form ein (or mein, kein, etc.), as in ein Wagen ‘one car’ (..) or ein Buch ‘one book’ (../), is replaced by einer (..) or eins (.. /) if the NP is elided () (Lobeck : ch. ). A contrast similar to () shows up in German RNR (); again, the ellipsis form is degraded. ()

Sie she

()

a. Du hast sein __, und er hat mein you have his and he has my b. ??Du hast seins__, und er hat mein Buch.

hat has

mein my

Buch (*meins Buch) book

und and

du you

hast have

Buch. book

seins _ (*sein _). his

   - 



The licensing contexts for English VPE and RNR of VP differ in four cases. Firstly, VPE cannot target the complement of progressive being (passive or copular); cf. () (Akmajian and Wasow ; Sag ; Harwood ). RNR of a passive VP or other predicate, stranding being, is perfectly possible (): ()

John was being spied on/too careful, and Bill was/might be (*being), too.

()

a. Judy  being, and Jack  to be, as quiet as a mouse. b. Selected sites are being, and will continue to be, monitored . . . (Web example)

Secondly, RNR can take out non-finite perfect have. VPE normally cannot (Sag ; Harwood ). ()

a. They want to know who must and who may have witnessed it. b. They want to know who must have witnessed it and who may *(have).

Thirdly, RNR (unlike VPE) can take out a finite lexical verb without triggering do support: ()

a. We asked whether Judy owns a car and not whether John owns a car. b. We asked whether John owns a car and not whether Judy *(does).

Fourthly, RNR (unlike VPE) cannot trigger do-support except in emphatic do contexts. Do-support in affirmative declaratives is restricted to cases of emphatic do, with do focused (John  leave, but *John did  / *J did leave)—except in VPE contexts, where dosupport is mandatory even if do is not focused (Who left? – J did). Where VP is RNRed, do-support only occurs when focused (emphatic do, (c)).39 ()

a. S said that J, and B said that M left first. (RNR only) b. ??S said that J did and B said that M left first. (VPE only) c. Sue said that J  and that M ’ leave.

The first three contrasts indicate that RNR can elide English VPs which VPE cannot, whereas the contrast in () suggests that VPE is unable to create an RNR-like configuration where the VP gap in the first conjunct depends on the VP in the second conjunct. This matches what () and () suggest for sluicing, and () and () for NPE.40 39 In a context like (i), (b) improves considerably, presumably because the VP gap is parsed as an ellipsis whose antecedent is provided by the preceding question, rather than the following conjunct.

(i) A: Who left first? B: S said that J did and B said that M left first. A contextual antecedent can also render (b), (b), and (b) acceptable. 40 However, natural-sounding examples with but can be found that appear to involve backward ellipsis: (i)

We don’t yet know who from, but we do know she received a package last Monday. (cf. n. , ex. (iii)). (ii) I don’t know if J did, but M passed the exam. (iii) I don’t know about yours, but my wetsuit fits perfectly.





On the other hand, Bošković (b), Barros and Vicente (), and others have documented that apparent RNR of English VPs allows identity mismatches that are unexpected from the viewpoint of the form identity condition on RNR discussed in ..., and that mirror those permitted in VPE. This is illustrated by the elision of the bare form write licensed by the participle written in both examples in (). ()

a. John will _, and Mary already has written to Bill. b. Mary already has written to Bill, and John will _ tomorrow.

Barros and Vicente argue that this pattern and the way it interacts with phenomena discussed in ... and ... show that RNR is not a unitary phenomenon (they propose a hybrid ellipsis/MD approach).41 In sum, although there is evidence to suggest that the configuration () generally only arises via RNR (and not regular ellipsis), there are still unresolved issues in this domain, notably relating to facts like (a).

.. RNR as Multiple Dominance ... Sharing as MD Multiple Dominance has been explored as an alternative in situ approach to RNR (McCawley ; Wilder ; Gračanin-Yüksek ; Kluck and de Vries , and others). It differs from ellipsis/deletion accounts in that there is assumed to be only one token of the shared constituent in the structure. Thus instead of two copies of the shared element α with one deleted, there is a single element α, dominated by two different mother nodes, one in each conjunct. ()


T'

DPj Bill-sg

DPk John-sg

Tj+k have-pl

VP ...

The subjects in () do not form a single unit (a DP coordination). It is thus correctly predicted that group readings of the DPs are impossible () (Grosz reports that such examples are uniformly rejected by speakers who accept summative agreement in the pattern ()). () *Sue’s proud that Bill, and Mary’s glad that John, have finally met.

... Non-bulk sharing The MD approach permits sharing of a sequence of more than one constituent. The facts noted in ... and ... concerning multiple constituent sequences and the interaction of RNR with leftward movement are both captured in terms of multiple constituent sharing. In (), what is shared is a single complex unit, viz. [DP the book]. There is one node in each conjunct which is a mother of the shared unit. In () (cf. (b)), two units are shared, A and N, each of which has a mother in each conjunct—AP1 and AP2, NP1 and NP2 (Wilder ). ()


NP2

D a AP2

N electrode

Adv A negatively charged

pattern. The possibility for a wide scope of an RNRed QNP pointed out by Sabbagh () is consistent with that possibility, as is the possibility for a multiple wh-question such as (ii). (i) *Who said [that John bought what] and [that Peter sold what]. (Bošković and Franks : ) (ii) Who said that John hates and Peter likes what? ‘For which person x and which thing y did x say that John hates y and Peter likes y?’

   - 



Gračanin-Yüksek () distinguishes bulk sharing from non-bulk sharing. In (), a single shared constituent, formed from smaller shared units, is shared as a ‘bulk’ unit. In (), two smaller units (A and N) are shared separately (‘non-bulk’). RNR cases like () have non-shared constituents (which book and which film) extracted from a position within the shared string, posing tricky questions for an ATB movement account (cf. ...). ()

Bill asked which book _ and Celia asked which film they chose.

In an MD approach, RNRed strings containing distinct null elements involve non-bulk sharing—one trace belongs to one conjunct and the other trace belongs to the other. The shared items of the RNRed string are shared individually, a sequence of MDed constituents. Any constituent XP that dominates a non-shared trace within a conjunct is itself nonshared (i.e. distinct from its counterpart XP in the other conjunct by virtue of dominating an element not dominated by XP), even where the constituent in question (e.g. the TPs and VPs in ()) dominates no pronounced non-shared elements. ()

< [ Bill asked

CP1

] and [ Celia asked

which bookj TP1

CP2

which filmk VP1

]> TP2 VP2

they tj

chose

tk

The same account is applied to wh&wh questions like () by Gračanin-Yüksek () (see also Citko and Gračanin-Yüksek ). In (b), the positions of the individually shared elements (did, he, eat) in the first conjunct are indicated with underscores. ()

a. What and when did he eat? b. < [CP1 whatj _ [TP1 _ [VP1 _ tj ]]] and [CP2 whenk did [TP2 he [VP2 eat tk ]]] >

The non-bulk sharing approach resolves the issue of RNRed strings containing distinct (nonshared) null elements by denying that the RNRed string corresponds to a single constituent.43 Any such RNRed string will be analysed as non-bulk sharing, including examples involving A-movement (a). For the same reason, though, this approach is unable to derive (b) as ATB VP-fronting; the structure will not contain a single shared VP to input ATB fronting. ()

43

a. Johnj was and Maryk wasn’t caught { tj/tk } stealing. b. Caught { tj/tk } stealing, Johnj was _ and Maryk wasn’t _.

An ATB movement approach to RNR could in principle avoid the non-shared trace problem by denying that the shared string forms a single constituent, and assuming that the examples involve movement of multiple constituents. This is suggested by Sabbagh (: –) for cases like (). The same would apply in the leftward movement case (b).





.. Non-coordinate RNR The RNR configuration (), with a gap at the right edge of a constituent A corresponding to a string within its sister B, is also found where A and B do not form a classic coordination. One case involves comparatives (a), where B is the than-clause. Various authors argue that comparatives can be structured as coordinate constructions (e.g. Lechner , ; Phillips ; Osborne ). Comparatives also license forward CR; cf. (b). ()

. . . [A abc xyz ] [B . . . def xyz . . . ] . . .

()

a. More people [A talked to _ ] than ignored the new colleague. b. Peter has sent  to S more often than [ _ a  to M ].

RNR also turns up in environments which are non-coordinate in the sense that they bar forward CR. Some involve arguments of predicates of comparison (a,b) or motion (c), or subordinators like whereas or without (d,e), while in other cases, material is shared by a subject and its predicate (f). Many such examples have a marginal flavour. ()

a. It’s interesting to compare [A the people who like _] with the people who dislike the power of the unions. (Hudson b: ) b. How to distinguish [A neuro-__ ] from psycholinguistic claims. (Chaves : ) c. a journey [A from the southern _ ] to the northern tip of the island d. [A John throws out _ ], whereas Mary eats anything that happens to be in the refrigerator. (Goodall : ) e. [A I talked to _ ] without actually meeting all the members who voted against Hinkly. (Williams : ) f. [A Anyone who talks to _ ] really gets to like our new colleague from Sunderland.

That such examples belong to RNR is supported by several characteristics. Subword deletion as in (b) is only otherwise found in coordinate RNR. The possibility for Pstranding precludes an analysis of (e) as parasitic gap construction licensed by rightward shift of all the members across the without phrase (Williams ; Postal ). Like coordinate RNR, these dependencies are not constrained by islands; cf. (a) and many examples in Postal (). In German, non-coordinate RNR in subordinate clauses can yield the sharing of verbs in the predicate and a relative clause in the subject (a), a possibility bled in main clauses by V (Wilder ; Lechner : ). ()

a. ?weil [viele Leute, die ein Buch gekauft haben] since many people who a book bought have auch eine Zeitung gekauft haben also a newspaper bought have ‘since many people who bought a book also bought a newspaper’ b. *Gestern haben [viele Leute, die ein Buch gekauft haben] yesterday have many people who a book bought have auch eine Zeitung gekauft. also a newspaper bought

   - 



A general characterization of the environments permitting non-coordinate RNR is lacking, as is a systematic study of other factors impacting on its occurrence.

.. Right-Node Wrapping The VP coordinations in () involve a shared object that surfaces in a non-final position in the final conjunct (Wilder , ; Whitman ), a pattern which Whitman calls Right-Node Wrap. RNW shows a characteristic intonation pattern: the shared element is deaccented (cf. the pronoun in (c)) and the non-shared material that follows it is accented. ()

a. Mary congratulated _ and gave the boy the prize. b. The blast upended_ and nearly sliced an armored Chevrolet Suburban in half. (Whitman ) c. After using dishes, please wash, dry, and put them away in the proper place. (Whitman ).

Object-sharing in the RNW pattern is attested for German by Hartmann and Schmitt () and for French by Mouret and Abeillé (); it is also found in Norwegian (). ()

Torsdag kunngjør hovedstyret i Norges Bank Thursday announces executive.board. in Norge Bank om det vil kutte _ eller holde styringsrenten uendret på , prosent. whether it will cut or hold control.rate. unchanged at . percent ‘On Thursday, the Bank of Norway’s executive board announces whether it will cut or keep the key policy rate at .%.’ (Web example)

The phenomenon itself and its implications for theories of RNR have only recently begun to be explored. One reason for treating RNW as part of RNR is that non-initial conjuncts show the RER (). ()

*John gave _ a prize and recommended Mary for promotion.

The non-final placement of α in the final conjunct makes it difficult to pursue an ATB movement analysis—if α has raised out of the coordination, then the non-shared phrase following it presumably has, too, in violation of the CSC. Another RNW pattern is found in German nominals, illustrated by (b) (from the Web). Coordinated possession nominals with shared possession can take the form (a), whereby the possessor pronoun (mein) has the bare form used in RNR but not NPE (cf. ...). ()

a. < [ Possessor pronoun ] & [ Det Noun NP ] > b. Dieser Tag hat < mein _ und auch das Leben vieler this day has my... and also the... life many. anderen Menschen > verändert. other people changed ‘This day changed my life and also the life of many others.’





Sabbagh () notices that a quantifier phrase in the RNW configuration does not scope over the coordination, unlike a regular RNRed QP. Unlike (a), (b) cannot describe a prediction that some will be fired and others promoted. ()

a. b. c. d.

They will either promote or fire . ( > ) They will either promote or give everyone the . ( >  / * > ) They will either  or  everyone. ( >  / *  > ) ??They will either promote or give  the sack.

Taking wide scope to be a reflex of overt movement of the RN out of the coordination (Sabbagh ), the narrow scope of the RN in (b) indicates an in situ analysis for that case (as the word order independently suggests). However, the wide-scope reading of everyone also correlates with accenting; unstressed everyone, whether non-final or final (c), does not have the distributive (wide-scope) reading.

. C

.................................................................................................................................. Viewed as a whole, CR is characterized by striking right–left asymmetries, suggesting that different mechanisms lie behind forward and backward gaps in coordination ellipsis. Forward CR shows affinities with canonical ellipsis with respect to vehicle change etc. This, together with indications that non-elided remnants in conjuncts affected by forward CR have undergone conjunct-internal movement, renders the move-and-elide approach particularly promising, though important questions remain. If move-and-elide is correct, then forward sharing arises from an ellipsis mechanism like that involved in VPE, NPE, and sluicing. RNR is different. An ellipsis approach falters in the face of identity patterns (absence of vehicle change) and wide-scope effects found with quantifiers, relational adjectives, etc. The major alternative has been Ross’s original rightward ATB movement proposal; however, doubt is cast on that approach by a range of considerations, including the RNR ability of sharing otherwise unmoveable elements. The third possibility, an in situ MD approach, appears to offer ways out of the dilemma, though much remains to be explored in this domain. If the MD hypothesis is correct, then RNR gaps are due to a basically different mechanism than forward CR and regular ellipsis.

  .............................................................................................................

CASE STUDIES .............................................................................................................

  ......................................................................................................................

 ......................................................................................................................

     

. I

.................................................................................................................................. T chapter examines ellipsis in Dutch. We will discuss the major types of ellipsis as they have been presented in Part III of this handbook: gapping and stripping (section .), predicate ellipsis (VP-ellipsis and pseudogapping) (section .), Conjunction Reduction and Right-Node Raising (section .), sluicing (section .), fragments (section .), nominal ellipsis (section .), Comparative Deletion (section .), and Null Complement Anaphora (section .). In addition, section . will discuss some peculiarities concerning ellipsis in Dutch dialects. Since we will keep the discussion of the various types of ellipsis as theory-neutral as possible, we will provide suggestions for further reading at the end of each of these sections.

. G  

.................................................................................................................................. The phenomenon of gapping is a part of Dutch grammar. Gapping is an ellipsis operation applied to coordinated clauses, which involves the deletion of a verb in the second conjunct under identity with a verb in the first conjunct. For example, in (a), the finite verb at ‘ate’ is deleted, here represented by means of subscription, in the second conjunct under identity with the finite verb at in the first conjunct. As shown in (b), in addition to the verb, other material (in casu the direct object vis) may be silent as well. () a. J at  en P Jan ate fish and Piet ‘Jan ate fish and Piet steak.’

at ate

. steak

b.  at op  vis en  at op  Jan ate on Monday fish and Piet ate on Wednesday ‘Jan ate fish on Monday and Piet on Wednesday.’

vis. fish



   

As indicated by means of small capitals in (), the remnant constituents in the clause to which gapping has applied and its correlates in the antecedent clause are contrasted and bear stress. In the rest of section ., this contrastive relationship will not be represented orthographically. Notice also that gapping does not require strict formal identity between the antecedent verb and the deleted verb. Specifically, if the two verbs differ from each other in number (a) or tense (b), gapping can still take place. () a. Ik at vis en jullie I ate1.sg fish and you ‘I ate fish and you steak.’

aten ate.2.pl

biefstuk. steak

b. Gisteren at ik vis maar vandaag yesterday ate I fish but today ‘Yesterday I ate fish, but today I eat steak.’

eet

ik

eat

I

biefstuk. steak

In () and (), there are two remnants in the second conjunct. As shown in (), it is also possible to have only one remnant (Piet) in the second conjunct. In that case, the single remnant is often followed by a (focus) particle like ook (‘also/too’) or a polarity item like niet. This phenomenon, in which the verb has been deleted and only one remnant is left in the second conjunct, is called stripping.1 () Jan at vis maar Piet Jan ate fish but Piet ‘Jan ate fish but Piet not.’

at ate

niet not

vis. fish

The main focus of section . will be on gapping. Section .. addresses the question of which verbal material can be deleted by gapping. Section .. discusses some properties of the remnants in the clause to which gapping has applied. Section .. examines the distribution of gapping. That is, in what kinds of clauses is this ellipsis phenomenon attested? In section .., finally, the phenomenon of stripping is discussed.

.. Gapping: Deletion of verbal material As was shown in ()–(), gapping deletes the finite verb in the second conjunct. If the finite verb is an auxiliary, then gapping applies to the auxiliary, as in (a). The main verb has to be included in the gapping operation, if it is identical to the main verb in the first conjunct, as in (b). As shown by (c), even more complex verbal patterns can be deleted in the second conjunct, if this verbal complex is identical to the one in the first conjunct.

1

We simply use the term stripping as a descriptive device and will not address the question as to whether the (analysis of the) phenomenon of stripping can be reduced to that of gapping.





() a. Jan heeft de vis gebakken en Piet heeft de Jan has the fish cooked and Piet has the aardappels geschild. potatoes peeled ‘Jan has cooked the fish and Piet has peeled the potatoes.’ b. Jan heeft de vis gebakken en Piet Jan has the fish cooked and Piet (*gebakken)/gebakken. cooked/cooked ‘Jan has cooked the fish and Piet the steak.’ c. Jan zal de vis moeten bakken en Piet Jan will the fish must cook and Piet . bakken

heeft has

zal will

de the

biefstuk steak

de biefstuk the steak

moeten must

cook

‘Jan will have to cook the fish and Piet the steak.’ Observe that, as shown by examples (b,c), the elements that are deleted by means of gapping need not be contiguous. For example, the deleted verbal elements heeft and gebakken in (b) are separated from each other by the direct object de biefstuk. Gapping is also found in embedded clauses. In (), gapping has applied to the second conjunct of a direct object clause. As indicated, gapping cannot apply to the second conjunct if an overt complementizer (dat) is present. () Ik dacht dat Jan vis at en (*dat) Piet biefstuk I thought that Jan fish ate and ( that) Piet steak ‘I thought that Jan ate fish and Piet steak.’

at. ate

Importantly, gapping also applies to infinitival verbs in non-finite clauses. Note that here too, the second (infinitival) clause cannot be introduced by an overt complementizer (om). () Jan Jan

stelde voor [ om PRO op maandag vis te eten en proposed PRT for PRO on Monday fish to eat and

(*om) PRO op woensdag vlees te eten]. ( for) PRO on Wednesday meat to eat ‘Jan proposed to eat fish on Monday and meat on Wednesday.’ So far, we have seen that gapping can delete a finite verb of a main or an embedded clause, or an infinitival verb of a non-finite clause. The fact that gapping cannot apply in the second conjunct of an embedded clause if that conjunct starts with an overt complementizer seems to suggest that gapping is only operative in coordinate structures with TP-conjuncts (that is, [CP dat/om [TP & TP]]). However, the wh-interrogative main clause in () suggests that gapping is also operative within a CP-conjunct. In this example, a wh-word has been moved to specCP in both conjuncts, along with movement of the finite verb to C. In the right conjunct, the wh-word wat occupying specCP functions as one of the remnants.



   

() [[CP Wat heeft Jan op maandag gegeten] en what has Jan on Monday eaten and [CP wat op woensdag ]]? heeft Jan gegeten what has on Wednesday eaten Jan ‘What did Jan eat on Monday and what on Wednesday?’ So far we have seen that gapping can apply in finite and infinitival clauses. Interestingly, gapping is also found in coordinate structures involving (what looks like) conjoined VPs. Consider the examples in (): () A: Wat zal Jan nooit doen? what will Jan never do ‘What will Jan never do?’ B: [[VP Bloemen voor zijn vrouw kopen] en [VP chocola voor flowers for his wife buy and chocolate for zijn kinderen kopen]] his children buy ‘Buy flowers for his wife and chocolate for his children.’ In (), person B uses a coordinated VP-structure as an answer to person A’s question. Gapping has applied in the second conjunct of B’s answer.

.. Gapping and its remnants As was noted in (b), (b), and (), in addition to the elided verb(s), other material can be silent as well as a result of gapping. As shown in (), Dutch permits gapping constructions in which more than two remnants are contained within the second conjunct (see also Aelbrecht a,b). In this respect, Dutch differs from English, which only permits two remnants in the second conjunct (Jackendoff ). () Jan gaat dinsdag bij z’n broer dineren en [Piet] gaat Jan goes Tuesday with his brother dine and Piet goes [woensdag] [ bij z’n zus] . dineren Wednesday with his sister dine ‘Jan will dine with his brother on Tuesday and Piet will dine with his sister on Wednesday.’ As noted by Hankamer () for English, the remnants left behind after gapping must be major constituents of the clause (the so-called Major Constituent Condition); see Neijt () for Dutch. According to this condition, only phrases that are immediately dominated by a verbal projection on the extended verbal projection line (Grimshaw ) can function as remnants. Thus, arguments of the verb and modifiers contained within the extended verbal projection can be remnants, but subparts of these arguments and modifiers cannot. This condition on the remnants of gapping is exemplified in () for PP-complements (the same can be shown for CP-complements; see below).

 ()



Jan snakt naar een glas water . . . Jan longs for a glass water . . . ‘Jan longs for a glass of water . . . ’ a. . . . en Piet snakt [PP naar een glas . . . and Piet longs for a glass ‘ . . . and Piet longs for a glass of beer.’ b. * . . . en . . . and

Piet Piet

snakt

naar

longs

for

een a

glas glass

bier]. beer

bier. beer

There seem to be counterexamples to the Major Constituent Condition. There are gapping constructions in which one of the remnants is a subpart of a major constituent (see Zwarts  for some discussion). ()

Jan stond [  meter achter mij] en Marie Jan stood  meter behind me and Marie meter achter mij]. meter behind me ‘Jan stood  meters behind me and Marie  meters.’

stond [PP stood

 

In () a left branch constituent which acts as a modifier within a larger phrase (a major constituent: PP) has been left behind as a remnant after gapping. One could explore an approach in which these remnants are placed (e.g., via some displacement operation) in a clause-internal position outside of the major constituent (PP) before gapping applies. Independent evidence is then needed for the existence of such a constituent rearrangement operation. That the left branch modifier in () is accessible to displacement is exemplified in (); see Corver  for discussion:2 ()

Hoeveel meteri stond Marie [ti achter how many meter stood Marie behind ‘How many meters did Marie stand behind me?’

mij]? me

A final property of remnants that we would like to point out here is that they can sometimes contain a gap (more specifically, a trace) that results from subextraction. Consider the following example, in which the wh-phrase has been extracted in an acrossthe-board fashion (Ross ): ()

Waari is Jan [AP geschikt [PP ti voor]] en what is Jan suitable for and voor]]? for ‘For what is Jan suitable and Marie unsuitable?’

2

Marie [AP Marie

ongeschikt [PP ti unsuitable

Pied-piping of achter mij also yields a well-formed sentence.



   

.. Gapping: Its distribution Gapping is a forward ellipsis operation; that is, the gap must be contained in the second conjunct. Gapping cannot apply backwards: ()

*Jan Jan

at ate

vis fish

en and

Piet Piet

at ate

biefstuk. steak

Gapping is permitted in coordinate structures in which the coordinator is en ‘and’, maar ‘but’, of ‘or’, or ‘zero’ (i.e., asyndetic coordination; represented here as ConjØ). For certain speakers, gapping is blocked when the second conjunct is introduced by the coordinator want ‘because’. For other speakers, (b) is fine or just slightly marked (see also Aelbrecht a,b). ()

a. Mijn moeder bakt op zondag een taart en/maar/of my mother bakes on Sunday a pie and/but/or mijn vader bakt een cake. op zondag my father bakes on Sunday a cake ‘My mother bakes a pie on Sunday and/but/or my father bakes a cake on Sunday.’ b. */?Jan Jan

kent knows

mij me

want since

ik I

ken know

hem. him

Gapping is restricted to coordinate structures (Neijt ). In other words, it cannot apply to a verb in an embedded clause that is identical to a verb in the main (matrix) clause. This is exemplified in () where gapping applies within a direct object clause. ()

*Jan Jan

had had

mij me

verteld [CP told

dat that

Piet Piet

een a

leuk nice

verhaal story

had verteld]. had told

When gapping applies, it is subject to what Boone () calls the Equal Conjunct Requirement: the ellipsis site may not be embedded relative to its antecedent, nor may the antecedent be embedded relative to the ellipsis clause (see also Aelbrecht a,b). This requirement is exemplified in () where the gapping clause is embedded under the matrix clause headed by geloof. ()

*[CP

J Jan

at ate

] fish

en [CP and

ik I

geloof [CP believe

dat that

P Piet

 steak

at]] ate

We have already seen that gapping applies to main clauses and embedded clauses. The remnants must be clause mates. Thus, for many but certainly not all speakers of Dutch, gapping is illformed if one remnant is in the matrix clause while the other is in the embedded clause.3

3

For some speakers, the sentence improves if the embedded subject is coindexed with the matrix subject, as in (i).

 ()

*/?Ik dacht dat Jan vis at en [ jij I thought that Jan fish ate and you biefstuk at]]. steak ate

dacht

[CP

thought

dat

Jan

that

Jan



As has been noted by Hankamer () for English, and Neijt () for Dutch, the ellipsis site that results from gapping cannot contain an island (Ross ). ()

*Jan bestelde [ vis en Jan ordered fish and [ vis en aardappelen]. fish and potatoes

rijst] rice

en and

Marie Marie

bestelde ordered

In (), the ellipsis site contains a coordinate structure and the result is ill-formed, which suggests that the Coordinate Structure Constraint is active. As shown in (), gapping can apply across a sentence boundary (see van Zonneveld ). It should be noted, however, that not all speakers of Dutch find this pattern fully acceptable. As exemplified by (), the ellipsis site that results from gapping cannot be licensed by a non-linguistic antecedent; see Hankamer and Sag () for English. () A: Ik geef Jan een muts. I give Jan a cap ‘I will give Jan a cap.’ B: Dat komt goed uit. Ik geef that suits well PRT. I give ‘That is good. I will give Jan a scarf.’ ()

Jan Jan

namelijk namely

een a

sjaal. scarf

*Ik stofzuig mijn kamer. I my room vacuum

.. Stripping Stripping (also known as ‘bare argument ellipsis’), like gapping, involves the elision of verbal material in the second conjunct under identity with corresponding material in the antecedent clause (the first conjunct). All other (non-verbal) material is deleted as well, except for one phrase. In other words, there is only a single remnant. This remnant is often (i)

% Ik dacht I thought

dat ik that I

vis ging fish went

eten en eat and

jij you

dacht

dat

je

thought

that

you

biefstuk ging eten. steak went eat ‘I thought that I was going to eat fish and you thought that you were going to eat steak.’



   

followed immediately by (focus) particles like wel/ook (‘also’, ‘too’) or polarity items like niet (‘not’), as in (a). As shown by (b), the remnant can also be bare. This requires a special intonation: strong emphasis on hij and zij, and a brief pause after vis. The pattern is quite marked compared to (a). () a. Hij eet vis maar zij niet/ook. he eats fish but she not/too ‘He eats fish, but she does not/too.’ b. Hij eet vis en zij. he eats fish and she ‘He eats fish and she does too.’ Modulo the number of remnants, stripping shares many properties with gapping. For that reason, stripping is often regarded as a particular manifestation of gapping. In ()–(), some properties are exemplified which were also found above in gapping constructions. ()

Wat bestelt Jan erg vaak en wat niet? what orders Jan very often and what not ‘What does Jan order very often and what doesn’t he order very often?’

() *Piet Piet ()

ook en Jan eet op maandag vis. too and Jan eats on Monday fish

[De mededeling dat Jan vis bestelde en Piet ook] wekte verbazing. the announcement that Jan fish ordered and Piet too raised surprise ‘The announcement that John ordered fish and Piet too, surprised many people.’

() A: Jan ontmoette gisteren Jan met yesterday ‘Jan met Els yesterday.’

Els. Els

B: Dat is toevallig! Ik ook. that is coincidental. I too ‘That is a coincidence! Me too.’ In (), a direct object wh-phrase precedes the negation, which suggests that stripping can apply to entire CPs (this under the assumption that wh-phrases occupy specCP). The illformed example () shows that stripping, just like gapping, is a forward ellipsis operation. Thus, it cannot apply backwards. () shows that, like gapping, stripping is also found in embedded clauses. Example () shows that stripping can apply across sentence boundaries in a discourse (compare with the gapping example in ()). In short, many of the properties attested in gapping constructions are also found in stripping constructions. For more details and in-depth analyses of gapping and stripping in Dutch we refer the reader to Bakker (), Neijt (, a, b), Kerstens (, ), de Vries (), Haeseryn et al. (), van der Heijden and Klein (), van der Heijden (), Aelbrecht (a, b), Pieters (), Boone (), Cremers (), and Zwart ().





. P  ( VP-  )

.................................................................................................................................. In English VP-ellipsis the predicate containing the lexical verb, the arguments, and low adverbs are elided, i.e., the VP. The remnant of ellipsis consists of the subject and a finite auxiliary; see (a). If there is no finite auxiliary, the dummy verb do appears; see (b): ()

a. John has sold his house very quickly, but Pete hasn’t sold his house very quickly. b. John sold his house very quickly, but Pete didn’t sell his house very quickly.

VP-ellipsis is not possible in Dutch, as the examples in () illustrate: () a. *Jan heeft zijn huis snel verkocht, maar Piet Jan has his house quickly sold, but Piet Intended: ‘Jan has sold his house quickly, but Piet hasn’t.’ b. *Jan verkocht zijn huis snel en Piet deed Jan sold his house quickly, and Piet did Intended: ‘Jan sold his house very quickly, but Piet didn’t.’

heeft has

niet. not.

niet. not.

A potential reason for the ungrammaticality of VPE in Dutch might be that the lexical verb moves out of the VP domain into a head-final TP projection. If this analysis is correct, then this might account for the ungrammaticality of these examples: the lexical verb is not in the VP-domain and hence should not be part of the ellipsis site. This would predict that ellipsis of just the object and low adverbs should be possible (provided they stay in the VP-domain). This also does not result in a grammatical sentence as illustrated in (). () *Jan Jan

heeft has

zijn his

huis house

heel very

snel verkocht, en Piet quickly sold, and Piet verkocht. zijn huis heel snel sold. his house very quickly Intended: ‘Jan has sold his house very quickly, and Piet has too.’

heeft has

ook also

VPE with infinitival verbs is also impossible in Dutch (with all types of infinitives, but illustrated here with control infinitives):4 () *Ik probeerde te dansen en Jan probeerde ook te I tried to dance and Jan tried also to Intended: ‘I tried to dance and Jan also tried to.’

4

(i)

dansen dance

Note that these examples are also ungrammatical if te ‘to’ is part of the ellipsis: *Ik probeerde te dansen en Jan probeerde ook te dansen. I tried to dance and Jan tried also to dance



   

As discussed extensively by Aelbrecht (), VPE with the infinitival complement of modals is possible in Dutch. She calls this Modal Complement Ellipsis. An example is given in (). ()

Ik wil wel helpen, maar I want PRT help, but ‘I want to help, but I cannot.’

ik I

kan can

niet not

helpen. help

As shown by Hoeksema () and Cremers (), while VPE might not be possible in general in Dutch, it is (marginally) possible for some speakers in comparative constructions: ()

%Jan verkocht zijn huis veel sneller dan Jan sold his house much quicker than ‘Jan sold his house much quicker than Piet did.’

Piet Piet

deed. did.

The second type of predicate ellipsis is pseudogapping, i.e. VPE where one constituent has survived ellipsis by moving into a focus position. Just like VPE, pseudogapping is generally not possible in Dutch (not even in modal complement ellipsis contexts). ()

a. *Ik verkocht mijn fiets en Piet deed I sold my bike and Piet did Intended: ‘I sold my bike and Piet did his car.’

zijn his

auto. car

b. *Ik wil mijn fiets verkopen, en Piet kan zijn I want my bike sell and Piet can his Intended: ‘I want to sell my bike and Piet can sell his car.’

auto. car

Just like VPE, pseudogapping is (marginally) possible for some speakers in comparative contexts, however, as was observed by Hoeksema (): () %Jan verkocht eerder zijn huis dan Piet Jan sold sooner his house than Piet ‘Jan sold his house sooner than Piet did his car.’

zijn his

auto car

deed. did

For a thorough discussion of the relevance of comparative contexts for VP-ellipsis and pseudogapping we refer the reader to Hoeksema () and Cremers ().

. C R

.................................................................................................................................. Examples (a,b) illustrate the phenomenon of Conjunction Reduction in Dutch. ()

a. [ Jan kocht een fiets] en [ Jan bought a bike and ‘Jan bought a bike and rented a car.’

Jan Jan

huurde rented

een a

auto]. car

 b. [ Jan kocht en [ Marie een auto] Jan bought a car and Marie ‘Jan bought and Mary rented a car.’

huurde rented

een a



auto]. car

In the coordinate structure in (a), the subject noun phrase Jan in the second conjunct is deleted under identity with the subject noun phrase Jan in the first conjunct. Since deletion under identity applies in a forward direction, this ellipsis phenomenon can be characterized as Forward Conjunction Reduction (FCR). In (b) we have the reverse situation. The deleted phrase—the direct object een auto ‘a car’—belongs to the first conjunct and the “antecedent” is part of the second conjunct. Thus, (b) exemplifies the phenomenon of Backward Conjunction Reduction (BCR). We will discuss FCR in section .. and BCR in section ...

.. Forward Conjunction Reduction In (a), a subject noun phrase is absent in the second conjunct. As exemplified in (), the deleted element can also be another type of clausal constituent. () a. [ Aan mij verkocht Jan zijn fiets] en [ to me sold Jan his bike and haar auto]. her car ‘Jan sold me his bike and Marie lent me her car.’ b. [ Gisteren kocht Jan een auto] en [ yesterday bought Jan a car and hij een fiets]. he a bike ‘Yesterday, Jan bought a car and rented a bike.’

aan

mij

to

me

gisteren yesterday

leende lent

Marie Marie

huurde rented

In (a), the shared subject noun phrase carries the thematic role of agent. It is possible for the shared subject noun phrase to fulfill a different semantic role in each of the two conjuncts. In (), for example, the shared subject Jan is interpreted as the theme of the passive verb geslagen in the first conjunct and as the agent of the active verb geslagen in the second conjunct. ()

Jan werd geslagen maar Jan heeft zelf ook Jan was beaten but has self also Jan ‘Jan was beaten up but has also beaten up himself.’

geslagen. beaten.

An important constraint on FCR is that the shared material must occupy a left peripheral position in the two coordinated clauses. The right conjunct cannot be reduced if the shared phrase occupies a clause-internal (a) or clause-final (b) position: ()

a. *[[ Jan heeft een auto gehuurd] en [ Marie heeft Jan has a car rented and Marie has ‘Jan rented a car and Marie bought one.’

een

auto

a

car

gekocht]]. bought



    b. *[[ Jan Jan

huurde rented

een a

auto] car

en [ and

Marie Marie

kocht bought

een

auto]].

a

car

It should be noted that this parallelism requirement (i.e., the requirement that the shared material occupy a left peripheral position) is not always fulfilled. In so-called Tante Betje (aunt Betty) clauses of the type in (), the subject noun phrase of the first conjunct and the subject noun phrase of the second conjunct do not occupy the same (i.e., left peripheral) position (example taken from https://onzetaal.nl/taaladvies/tante-betje). Examples like these are normally considered a stylistic error, but they do occur frequently, especially in writing. () ?[ Door because.of Leiden en Leiden and

een defecte bovenleiding rijden de treinen tussen a broken overhead-line drive the trains between Amsterdam via een andere route] en [ de treinen Amsterdam via a different route and the trains stoppen niet in Schiphol]. tussen Leiden en Amsterdam stop not at Schiphol between Leiden and Amsterdam ‘Because of a broken overhead-line the trains from Leiden to Amsterdam take a different route and they won’t stop at Schiphol airport.’

In FCR, the peripheral element need not be a constituent (see also Zwart ). The string Jan heeft in (), for example, does not pass any constituency tests; e.g., Jan heeft cannot be the input to any displacement operation. () [ Jan Jan

heeft has

een a

fiets bike

gehuurd] rented

en [ and

Jan

heeft

Jan

has

een a

auto car

gekocht]. bought

The non-reduced part (i.e., the remnant) of the second conjunct must be a constituent in FCR (see also Zwart ). This constituency requirement on the remnant is exemplified in (): ()

*[ De oude man huurde een fiets] en [ de oude vrouw the old man rented a bike and the old woman een auto]. a car ‘The old man rented a bike and the old woman bought a car.’

kocht bought

This example is ill-formed because of a violation of the constituency requirement on the remnant of FCR. In (), vrouw kocht een auto does not form a constituent without the string de oude. Notice that a sentence like () satisfies the constituency requirement on the remnant: een auto gekocht forms a constituent (VP). This is shown, for example, by the possibility of fronting and pronominalizing the string een auto gekocht, as in [Een auto





gekocht] heeft Jan (a car bought has Jan, ‘Jan bought a car’) and Dat heeft Jan ook (that has Jan too, ‘Jan did so too’) respectively. As shown in (), FCR is also attested in embedded clauses: herinner me [[CP dat Jan een fiets kocht] remember  that Jan a bike bought een auto huurde]]. Jan a car rented Jan ‘I remember that Jan bought a bike and rented a car.’

() Ik I

en [CP and

(*dat) that

In this example, the second conjunct of the embedded clause has been reduced: the subject noun phrase of this conjunct has been “deleted” under identity with the antecedent Jan of the first conjunct. Notice that reduction is blocked if the complementizer dat ‘that’ is present in the second conjunct. In other words, if the subject noun phrase of the second clausal conjunct is absent, the complementizer must be absent as well. This again shows that the deleted peripheral material need not form a constituent: the string dat Jan (i.e., complementizer + subject) does not pass any constituency tests. Notice that the remnant, viz., the string een auto huurde, does form a constituent, viz., VP. The observation that the remnant must be a constituent in FCR leads to the conclusion that coordination in this type of construction is operative at the remnant constituent level. In other words, a sentence like Jan heeft een fiets gehuurd en een auto gekocht does not have the structure in () but rather the structure in (), where, in this specific case, the two coordinated phrases are VPs. () Jan heeft [[ een fiets gehuurd] en [ Jan has a bike rented and ‘Jan has rented a bike and bought a car.’

een a

auto car

gekocht]]. bought

Thus, all the (well-formed) FCR examples discussed so far do not involve a coordination of two clauses, of which the second clause has been reduced (i.e., contains an ellipsis site). Rather, these examples all involve coordination at the remnant constituent level.

.. Backward Conjunction Reduction This section discusses the phenomenon of Backward Conjunction Reduction (henceforth BCR), exemplified in (b) and repeated here as (): () [Jan kocht en [Marie een auto ] Jan bought a and Marie car ‘Jan bought and Marie rented a car.’

huurde rented

een a

auto]. car

This phenomenon is also known as Right-Node Raising. In early transformational grammar an example like () was derived by across-the-board rightward movement of the shared



   

element (een auto ‘a car’) from the left and right conjunct, as in (). According to this movement analysis, the gap in the left conjunct is not the result of ellipsis. In the rest of section .., we will use the more neutral notion of Backward Conjunction Reduction. ()

John bought ti and Mary rented ti [a car]i.

It is important to note at the very start that BCR is conjunct-final and not sentence-final. That is, the shared element is not at the end of the entire sentence but rather at the end of the right conjunct. This is exemplified in (), where the verb huurt is the shared element. () *[[ Dat Jan een auto that Jan a car

huurt rents

en Marie een fiets and Marie a bike

huurt] rents

is bekend huurt] is well-known rents

BCR has the intonation pattern in (): there is emphasis (represented here with small capitals) on the contrasting words within the conjuncts. () Jan  en Marie  Jan bought and Marie rented ‘Jan bought and Marie rented a car.’

een a

auto. car

In (), the shared material (een auto ‘a car’) qualifies as a constituent, in casu a noun phrase, but e.g. APs, PPs, CPs, and TPs can also be the remnant in BCR. For reasons of space, we only provide an example of the last of these. () Jan eet een banaan nadat hij gesport heeft en (TP) Jan eats a banana after he sported has and Kees eet een appel voordat hij gesport heeft. Kees eats an apple before he sported has ‘Jan eats a banana after working out, and Kees eats an apple before working out.’ It should be noted that the shared material does not need to qualify as a constituent. Some illustrations of this are given in (), where the subscripted (i.e., shared) material “crosses” a constituent boundary. () a. Jan heeft [NP een dik en boek ] gelezen Jan has a thick book and read dun boek gelezen. thin book read ‘Jan read a thick book and Piet a thin book.’ b. Jan heeft [NP twee jaar] in Parijs Jan has two years in Paris Piet heeft drie jaar in Parijs Piet has three years in Paris

Piet Piet

en and gewoond. lived

gewoond lived

heeft has

een a





As shown by (), the BCR gap can be contained in a (deeply) embedded clause. () a. Jan Jan

zei [CP dat said that

hij he

een a

boek book

had

gelezen]

had

read

en and

[CP

dat Piet de krant had gelezen]. that Piet the newspaper had read ‘Jan said that he had read a book and that Piet had read the newspaper.’ b. Jan Jan

zei said

[CP

dat hij dacht [CP dat Nederland de wedstrijd that he thought that the.Netherlands the game en Piet zei [CP dat hij dacht [CP dat zou winnen]] and Piet said that he thought that would win Duitsland de wedstrijd zou winnen]]. Germany the game would win ‘Jan said that he thought that the Netherlands would win the game and Piet said that he thought that Germany would win the game.’

It should further be noted that the BCR gap can be embedded in an island configuration. This is exemplified in () with a complex noun phrase. ()

Jan kende [NP een man [CP die vaak in Parijs was geweest]] Jan knew a man who often in Paris was been en Piet kende een man die nooit in Parijs was geweest. and Piet knew a man who never in Paris was been ‘Jan knew a man that had often been in Paris and Piet knew a man that had never been in Paris.’

For certain speakers, it is possible to move material out of the shared element. Other speakers judge these examples as marked. In (), for example, the pronoun waar has been moved out of the PP complement of the shared AP. ()

? Waari was Jan al jaren [ verslaafd [ti aan]] en raakte where was Jan already years and got addicted to Marie gisteren pas [ verslaafd [ti aan]]? Marie yesterday only addicted to ‘What had Jan already been addicted to for years and did Marie get addicted to only since yesterday?’

All examples of BCR given so far involve structures in which the shared material is in the right periphery of a coordinate structure whose conjuncts are clauses. It should be noted, though, that the BCR phenomenon is also attested in coordinate structures involving phrasal conjuncts. In (), for example, BCR applies within a coordinate structure with two nominal conjuncts.

 ()

    [[NP

Jongens die twee broertjes en [NP meisjes hebben] boys who two little.brothers have and girls drie broertjes hebben]] krijgen gratis toegang. three little.brothers have get free access ‘Boys with two and girls with three little brothers get free access.’

die who

Although the phenomenon of BCR is typically found in coordinate structures, it should be noted that it is also sometimes found in structural configurations that are not coordinations. This is illustrated in (), where BCR applies to an argument of the verb, and the shared material is contained within another argument that follows. The question mark indicates that not all speakers of Dutch find these sentences fully acceptable. () ? Jan stelde [ echtparen die twee kinderen hadden] [ aan echtparen Jan proposed couples who two children had to couples die drie kinderen hadden] voor. who three children had  ‘Jan introduced couples with two children to couples with three children.’ Let us finally address the question of whether the elided material in the left conjunct and the shared material in the right conjunct must be identical in form. Examples in which there is a mismatch are given in (). It should be noted that certain speakers find these sentences fully grammatical, while others consider them to be completely impossible.5 This inter-speaker variation is represented by means of % (see Kluck  for discussion of speaker variation concerning matching problems in Dutch BCR). ()

a. %Ik vind het raar dat jij wel maar bent uitgenodigd I find it strange that you PRT are invited but niet zijn uitgenodigd. not are invited ‘I find it strange that you áre invited but that we are not invited.’

wij we

b. %Jan mocht zich maar ik moest me scheren. scheren Jan could himself shave but I had.to myself shave ‘Jan was allowed to shave himself, but I had to shave myself.’

Quite surprisingly, (55a,b) are better than the corresponding examples in which the finite verb/ reflexive is overtly realized: 5

(i)

a. * Ik I zijn are b. ?* Jan Jan

vind het raar dat find it strange that uitgenodigd. invited  zich scheren could himself shave

jij wel you PRT

maar but

bent are

ik  I had.to

uitgenodigd invited

maar wij but we

niet not

me scheren. myself shave

As pointed out by a reviewer, this could be due to phonological reasons: the contrasted element in the first clausal conjunct must be the rightmost overtly realized element in that conjunct.





In (a), the perfective auxiliaries display a mismatch in person and number: bent (second person singular) versus zijn (first person plural). In (b), there is a mismatch in the form of the reflexive pronoun: zich (third person) versus me (first person singular). Even though we can conclude from () that mismatches are possible in Dutch BCR patterns (at least for certain speakers), we should add that examples can be found in which matching is required: () *Jan is gisteren en Piet zal morgen komen. gekomen Jan is yesterday come and Piet will tomorrow come Intended: ‘Jan came yesterday and Piet will come tomorrow.’ In these examples, there is a mismatch in the form of the main verb: the participial form gekomen versus the infinitival form komen. Even if the participial form and the infinitival form are phonologically identical, it is impossible for many people to have them as shared material in BCR patterns: ()

*Jan is vandaag zijn bril en Marie zal morgen vergeten Jan is today his glasses forgotten and Marie will tomorrow haar tas vergeten. her bag forget Intended: ‘Jan forgot his glasses today and Marie will forget her bag tomorrow.’

As is clear from the above discussion, BCR and FCR display different grammatical behavior in many respects. Even though they both fall under the descriptive label of conjunction reduction, it seems plausible to treat them as very different phenomena. For more information and in-depth analyses of Conjunction Reduction in Dutch we refer the reader to Bakker (), Blom (), Neijt (), de Vries (), Haeseryn et al. (), de Vries (), Pieters (), Cremers (), and Zwart ().

. S   

.................................................................................................................................. Dutch allows sluicing, i.e., ellipis in wh-clauses where the only remnant is the wh-element. Consider the example in (). ()

Ik heb iemand gezien, maar ik weet niet wie I have someone seen but I know not who ‘I saw someone, but I don’t know who.’

ik

gezien

heb.

I

seen

have

The example in () is a standard example of sluicing. The IP-part of the embedded question ik gezien heb ‘I have seen’ has been elided. The remnant consists of a wh-word, in this case wie ‘who’. Sluicing is possible with subjects, as shown in (), but also with other arguments, like indirect objects (a), and with adverbs (b).



   

() a. Ik gaf iemand een boek, maar ik weet I gave someone a book but I know ‘I gave a book to someone, but I don’t know who to.’

niet not

wie. who

b. Jan heeft mij ooit gekust, maar ik weet niet meer wanneer. Jan has me once kissed but I know not anymore when ‘Jan has kissed me once, but I don’t remember when anymore.’ Sluicing is also possible with all types of wh-items (e.g., waarom ‘why’, wanneer ‘when’, hoeveel ‘how many’), but the examples with the wh-element wat voor ‘what for’, i.e. ‘what kind of ’, are most interesting since this element can be split. Consider the sentences in (), which illustrate the wat voor pattern without ellipsis. () a. Ik weet niet wat voor boek(en) Jan I know not what for book(s) Jan ‘I don’t know what kind of book(s) Jan bought.’

kocht. bought

b. Ik weet niet wat Jan voor boek(en) I know not what Jan for book(s) ‘I don’t know what kind of book(s) Jan bought.’

kocht. bought

Sluicing is only possible with the non-split wat voor ‘what for’ variant, (a), not with the split one, (b): ()

Jan heeft bepaalde boeken gekocht, Jan has certain books bought ‘Jan has bought certain books, but . . . ’

maar . . . but

a. ik weet niet precies wat voor boek(en) Jan heeft gekocht. I know not exactly what for book(s) Jan has bought ‘I don’t know exactly what kind of book(s).’ b. *ik weet niet precies wat Jan voor boek(en) heeft gekocht. I know not exactly what Jan for book(s) has bought ‘I don’t know exactly what kind of book(s) Jan bought.’ It is not entirely clear what causes the ungrammaticality of (b). It might simply be that the parallelism between the antecedent clause and the sluice prohibits sluicing in split wat voor ‘what for’ constructions. The antecedent clause does not contain an element that is similar to the wat voor ‘what for’ wh-element, since there is no indefinite counterpart of it. Something similar is found with wh-pronouns that undergo preposition stranding. The R-pronoun waar ‘where’ and the preposition op ‘on’ can be split in the non-sluiced version of this sentence, but not in the sluiced version. Consider the contrast in () (see also Merchant , Chung , and also Kluck  who provides an analysis for this observation). () Jan geloofde ergens in, . . . Jan believed somewhere in ‘Jan believed in something, . . . ’

 a. maar ik weet niet waar Jan in but I know not where Jan in ‘but I don’t know what Jan believed in.’ b. *maar ik weet niet waar. but I know not where



geloofde. believed

The by now standard analysis of sluicing involves deletion of IP (see Merchant ). The idea is that the remnant of sluicing, i.e. the wh-phrase, has moved out of the IP before ellipsis has taken place. As such, this wh-element shows the characteristics of movement (like reconstruction) as well as the morphological properties related to the position it has moved from (i.e., the morphological case belonging to the base position of the wh-phrase). Unfortunately, Dutch wh-items do not show case distinctions and hence they do not provide any information about the base position of the wh-phrase in sluicing. However, Dutch sluicing shows the same reconstruction effects for binding as English sluicing. Consider the data in (). () [Marie en Piet]i lachen om sommige grapjes van elkaari, Marie and Piet laugh about some jokes of each.other maar ik weet niet . . . but I know not ‘Marie and Piet laugh about some jokes of each other’s, but I don’t know . . . ’ a. om welke grapjes van elkaari [ Marie en Piet]i lachen. about which jokes of each.other Marie and Piet laugh ‘about which jokes of each other’s Marie and Piet laugh.’ b. om welke grapjes van elkaari. about which jokes of each.other ‘about which jokes of each other’s.’ Since the reciprocal elkaar ‘each other’ is bound by the subject Marie en Piet in the b-example, the sluiced part of the sentence should contain this antecedent, indicating that the IP has been deleted. Another frequently used diagnostic for movement, i.e. island sensitivity, works slightly differently in sluicing. Dutch sluices, like English ones, are sensitive to weak islands (see also Sauerland ) (see () with a negative island) but not to strong islands (see (), with a complex NP-island).6 () Marie heeft niet gezegd dat Jan de auto op een Marie has not said that Jan the car in a manier repareerde . . . way repaired ‘Marie didn’t say that Jan repaired the car in a certain way . . . ’

6

bepaalde certain

Note that island sensitivity does hold for so-called contrast sluices or else sluices (see Merchant b). This is true for Dutch as well.



    a. *en ik weet and I know repareerde. repaired b. *en ik weet and I know

hoe how

Marie Marie

niet not

heeft has

gezegd said

dat that

Jan Jan

de the

auto car

hoe. how

() Ze hebben iemand aangenomen die een Slavische taal spreekt, . . . they have someone hired that a Slavic language speaks ‘They hired someone that speaks a Slavic language, . . . ’ a. *maar ik weet niet [CP [ welke Slavische taal]i but I know not which Slavic language [C'[TP ze [T' [iemand [ die spreekt ti]] hebben aangenomen]]]]. they someone who speaks have hired b. maar ik weet niet welke Slavische taal. but I know not which Slavic language ‘but I don’t know which Slavic language.’ A potential problem for the IP-deletion analysis is multiple wh-sluicing, which is grammatical in Dutch. Consider (a) (see also Merchant : ). If sluicing is the result of wh-movement followed by ellipsis, we can conclude that multiple wh-sluicing in (a) is the result of movement of the wh-items wie ‘who’ and wat ‘what’ into the CP domain where they survive the ellipsis of IP. However, multiple wh-fronting is ungrammatical in nonellipsis contexts in Dutch; see (b). () a. Jan gaf iemand iets, maar ik weet niet wie wat. Jan gave someone something, but I know not who what ‘Jan gave something to someone, but I don’t know who Jan gave what.’ b. *Jan gaf iemand iets, maar ik weet niet wie wat Jan gaf. Jan gave someone something but I know not who what Jan gave To account for this problem, Lasnik () argues that multiple wh-sluicing in English is the result of moving one of the wh-items into the CP domain via regular wh-movement, whereas the other one undergoes heavy NP-shift.7 This explanation does not hold for Dutch, however, since it allows multiple wh-sluicing of elements that do not allow heavy NP-shift (see also Boone ): () *Iemand heeft someone has gegeten wat. eaten what

7

iets something

gegeten, eaten

maar but

ik I

weet know

niet not

But see Richards (, ) for an alternative account of these data.

wie who

heeft has





Now let us consider how Dutch fares with subtypes of sluicing: sprouting and swiping.8 Sprouting, i.e., sluicing based on an implicit argument or modifier of the antecedent predicate, is possible in Dutch: () Jan wil een band vervangen, maar hij weet John wants a tire replace but he knows ‘John wants to replace a tire, but he does not know how.’

niet not

hoe. how

In this example hoe ‘how’ has as its antecedent the implicit modifier of the predicate wil een band vervangen ‘wants to replace a tire’. One subtype of sluicing that is not found in Dutch is so-called swiping (see ()). Swiping, which is an acronym for Sluiced Wh-word Inversion with Prepositions in Northern Germanic, reverses the order of wh-element and preposition in sluicing contexts. () Ik heb op iemand gewacht, maar ik weet I have on someone waited but I know ‘I waited for someone, but I don’t know who for.’

niet not

op wie/*wie op. on who/who on

Swiping is typically found in languages that allow preposition stranding. Dutch only allows preposition stranding with R-pronouns, not with regular pronouns, as is shown in the contrast between (a–b): () a. Ik heb ergens op gewacht, maar ik weet niet waar I have something on waited, but I know not where ik op heb gewacht. I on have waited ‘I have waited for something, but I don’t know what I waited for.’ b. *Ik heb op iemand gewacht, maar ik weet niet wie ik op I have on someone waited, but I know not who I on heb gewacht. have waited Intended: ‘I have waited for someone, but I don’t know who I waited for.’ c. Ik heb ergens op gewacht, maar ik weet niet waarop. I have something on waited, but I know not where.on ‘I have waited for something, but I don’t know what for.’ The example in (c) shows that the order wh-pronoun–preposition is possible with R-pronouns. However, this order is not limited to ellipsis contexts, but it is the normal order of preposition and R-pronoun, so it is a bit unclear whether we can speak of swiping in these cases.9 Another subtype of sluicing is spading (see van Craenenbroeck b). As this is only possible in dialectal Dutch we discuss it in section .. 9 Note that van Craenenbroeck (, b) provides an argument in favor of the idea that there is swiping in similar examples in Frisian. We will not go into this analysis here, but refer the reader to the original work. 8



   

For more details and in-depth analyses about sluicing, swiping, and sprouting in Dutch we refer the reader to van Craenenbroeck (, b), Boone (), and Broekhuis and Corver ().

. F

.................................................................................................................................. Fragments, i.e. non-clausal answers to questions, are also found in Dutch. Consider an example in (). ()

A: Wie heeft hij gekust? who has he kissed ‘Who did he kiss?’

B: Piet/Hem. Piet/him ‘Piet/Him.’

The fragment answer of B contains just the proper name Piet or the pronoun hem, but the answer conveys that it was Piet/hem that was kissed. Fragments are possible in Dutch with NPs (see ()), but also with APs (see (a)), PPs (see (b)), finite and infinitival CPs, VPs, etc.: () a. A: Hoe vindt Piet het eten? how finds Piet the food ‘How does Piet like the food?’ B: Lekker. good ‘It’s very good.’ b. A: Waar heeft Piet where has Piet B: In de garage. in the garage ‘In the garage.’

zijn his

auto car

(AP)

geparkeerd? parked

(PP)

Fragments, just like sluicing, are argued to be the result of movement of an element into the CP-domain and concomitant ellipsis of IP (see Merchant a). This means that the sentence in () has the underlying structure in (). ()

A: Wie heeft hij gekust? who has he kissed ‘Who did he kiss?’

B: Piet/Hem Piet/him ‘Piet/Him.’

heeft

hij

Piet/hem

gekust.

has

he

Piet/him

kissed

We refer the reader to Chapter  in this volume for a more extensive discussion of fragments. Just as with sluicing, we discuss several connectivity properties of fragments in Dutch. Fragments, just like sluicing, show some (but not all) connectivity effects. Let us first





consider binding. The example in () shows that principles A, B, and C of the binding theory are at work in fragments: () A: Wie heeft hiji geprobeerd who has he tried ‘Who has he tried to shave?’

te scheren? to shave

B: *Jani/*Hemi/Zichzelfi. Jan/him/himself ‘Himself.’

Let us now consider two other connectivity effects: sensitivity to islands and preposition stranding. Let us start with preposition stranding. If fragments involve movement into the left periphery with concomitant ellipsis, we expect languages that allow preposition stranding to also allow bare DPs as fragments. This is most clearly illustrated by English: ()

A: To whom did John speak?

B: Mary John spoke to.

English allows preposition stranding and hence the fragment answering the question in () can be just the complement of the preposition to, i.e., Mary. Other languages, like Dutch, display more complex behavior with respect to preposition stranding. Consider the examples in ().10 () a.

Op Anna/Op haar/Daarop wacht Piet het liefst. On Anna/on her/there.on waits Piet the best ‘For Anna/For her/For that Piet enjoys waiting the most.’

b. *Anna/*Haar/Daar wacht Piet het liefst op. Anna/her/there waits Piet the best on ‘Anna/Her/That Piet enjoys waiting for the most.’ c. Waar wacht Piet het liefst op? where waits Piet the best on *(Op) Anna/*(Op) haar/Daar*(op). on Anna/on her/there on ‘What does Piet enjoy waiting for the most? Anna/her/it.’ The examples show that Dutch speakers allow preposition stranding with R-pronouns (see van Riemsdijk ). However, a fragment answer involving just the R-pronoun seems to be ungrammatical. We refer the reader to Kluck () for an analysis of this observation.

10

For some speakers of Dutch, preposition stranding with DPs is possible. Some speakers also allow DP fragments in these preposition-stranding contexts. Merchant would predict that the group of speakers allowing preposition stranding with DPs be identical with the group of speakers allowing bare DP-fragments in sentences like (76c). This is not necessarily the case, as the judgments of one of the reviewers reveals (s/he does not allow preposition stranding with bare DPs but does allow bare DP fragments like the ones in (76c)). More research is needed into the correlation between preposition stranding and fragments in Dutch.



   

Just as in English, fragments are insensitive to certain islands: ()

A:

Willen ze iemand aannemen die syntaxis geeft? want they someone hire that syntax teaches ‘Do they want to hire someone who teaches syntax?’ B: Nee, ze willen iemand aannemen die fonologie no they want someone hire that phonology ‘No, they want to hire someone who teaches phonology.’ B' *Nee, fonologie willen ze iemand aannemen die no phonology want they someone hire that

geeft. teaches geeft. teaches

B'': Nee, fonologie. no phonology ‘No, phonology.’ Movement out of a complex NP is impossible in Dutch; see (B'). The fragment that derives from this underlying structure is grammatical, though; see (B''). Dutch poses some problems for Merchant’s analysis of fragments: negation niet ‘not’ cannot be fronted to the beginning of the sentence; see (B'). This predicts that it cannot be a fragment. This prediction is not borne out, however, as the grammaticality of (B) shows. () A: Komt Jan nu wel of niet comes Jan now AFF or not ‘Is Jan coming to the party or not?’ B: Niet/Wel. not/AFF ‘He is./He isn’t.’ B': *Niet/Wel not/AFF

komt comes

Jan Jan

naar to

het the

naar to

het the

feestje? party

feestje. party

Note, by the way, that an analysis of () as not involving ellipsis at all cannot be correct, since these fragments are sensitive to certain islands, as illustrated with an adjunct island in (). () A:

Is Jan boos omdat Piet wel of niet naar het feestje komt? is Jan angry because Piet AFF or not to the party comes ‘Is Jan angry because Piet is coming to the party or because he is not?’

B: *Niet/*Wel. not/AFF Multiple fragments are also possible in Dutch (see (B)), whereas multiple movement into CP is not (see (B')): () A: Heeft Jan Henk gekust? has Jan Henk kissed ‘Has Jan kissed Henk?’

 B:



Nee, Piet Marie. no Piet Marie ‘No, Piet Marie.’

B' *Nee, no

Piet Piet

Marie Marie

heeft has

gekust. kissed

Merchant (a) also observes this for German and argues that ellipsis in some way repairs the problems posed by multiple fronting. Finally, Dutch allows for embedded fragments (Temmerman ). Consider the examples in () (from Temmerman , her example ()). ()

A:

B:

Wie dacht Carl dat de wedstrijd who thought Carl that the contest ‘Who did Carl think would win the contest?’

Hij he B': %Hij he B'':

had had had had

zou would

winnen? win

gedacht Kim. thought Kim Kim gedacht. Kim thought

Hij dacht Kim. he thought Kim ‘He (had) thought that Kim would win the contest.’

As the answers in B, B', and B'' show, the fragment Kim can be embedded within another clause, either following the verb as in B and B'' or preceding it, as in B'. Temmerman provides several tests to show that embedded fragments are the result of ellipsis rather than, for instance, parenthesis. We will not repeat these tests here, but refer the reader to the original work. For more details and elaborate analysis of fragments in Dutch we refer the reader to Aelbrecht (), Temmerman (), Boone (), and Broekhuis and Corver ().

. N 

.................................................................................................................................. Nominal ellipsis occurs when the noun, possibly in combination with material accompanying the noun, is missing from the noun phrase (see Lobeck ). In (), some illustrations of nominal ellipsis in Dutch are given. These examples clearly illustrate that the nominal head or a larger part of the noun phrase can be missing. () Jan kocht die dure Chinese vaas van porselein en . . . Jan bought that expensive Chinese vase of porcelain and . . . a. Marie kocht die goedkope Japanse vaas van plastic. Marie bought that cheap Japanese vase of plastic ‘Marie bought that cheap Japanese one of plastic.’



    b. Marie kocht die goedkope Japanse vaas Marie bought that cheap Japanese vase ‘Marie bought that cheap Japanese one of plastic.’

van

plastic.

of

plastic

After elision of the noun or a larger nominal projection, one or more remnants are left behind in the elliptical noun phrase. As shown by the examples in () and (), nominal ellipsis in Dutch is only possible with certain types of remnants. For example, the prenominal “satellites” Jans ‘Jan’s’ (possessor) and veel ‘many’ (indefinite quantifier) do not license a nominal gap, whereas groene ‘green’ (attributive adjective), and die ‘that’ (demonstrative) do.11 ()

a. ?* Els Els [

heeft [ Piets has Piet’s Jans auto] Jan’s car b. * Els heeft [ weinig Els has few 12 ] verkocht. auto’s sold cars

auto] verkocht en Marie heeft car sold and Marie has verkocht. sold auto’s] verkocht maar Marie heeft [ cars sold but Marie has

veel many

() a. Els heeft [ een witte auto] verkocht en Marie heeft [ een Els has a white car sold and Marie has a groene auto] verkocht. green car sold ‘Els sold a white car and Marie sold a green one.’ b. Els heeft [ deze auto] verkocht en Marie heeft [ die auto] verkocht. Els has this car sold and Marie has that car sold ‘Els sold this car and Marie sold that one.’

11

Interestingly, the examples in (83) are possible if nominal ellipsis is combined with gapping, as in (i). We leave the analysis of these structures for future research.

(i) Els Els

verkocht [Piets sold Piet’s

auto] en car and

Marie Marie

verkocht sold

[

Jans Jan’s

auto]. car

12

Note that the quantifier veel or the cardinal drie can be followed by a nominal gap (ø) in clausal constructions featuring the quantitative pronoun er (literally: there; paraphrasable as ‘of them’):

(i) Marie

heeft er gisteren [ veel / drie ø] verkocht. Marie has there yesterday many / three sold ‘Mary sold many/three yesterday.’ (e.g., many/three cars)

In Bennis (), the noun phrase internal gap (Ø) is analyzed as a silent pro-form bound (and licensed) by er, which is base-generated in a special clause-internal position. In Coppen (), on the other hand, the gap is taken to be a trace that results from displacement of er out of the noun phrase.





Notice also that postnominal satellites (e.g., PPs and clauses) do not license NPE; see (). Note in passing that the same example shows that a(n in)definite article is not able to license NPE either.13 ()

*Els heeft [ een/de auto met trekhaak] gekocht en Els has a/the car with towbar bought and Marie heeft [ een/de auto zonder trekhaak] gekocht. Marie has a/the without towbar bought car

In the literature on Dutch NPE (see, for instance, Muysken and van Riemsdijk ; Kester b), it has been argued that licensing of the ellipsis site in (a) relates to the presence of the inflectional e-ending on the attributive adjective. In a way, this inflectional ending makes it possible to recover the properties of the elided noun. Evidence in support of this licensing role of -e comes from the fact that certain speakers of Dutch reject NPE when the attributive adjective does not carry this inflection. As shown in (a), this happens when the attributive adjective modifies a singular, indefinite neuter noun. Adding -e after the attributive adjective (as a last resort), as in (b), rescues the NPE structure for those speakers. It should be noted that the pattern in () does not apply to all speakers of Dutch. Some (mostly speakers of Southern Dutch) do allow a morphologically bare adjective like zwart in nominal ellipsis patterns.14 These speakers also find the variant with an e-ending, as in (b), degraded or even completely impossible. The inter-speaker variation attested for these examples is represented by means of %. () a. %Els heeft [ een wit paardNeut.Sg.] verkocht en Marie heeft [ een Els has a white horse sold and Marie has a zwart ] verkocht. paard black sold horse ‘Els sold a white horse and Marie sold a black one.’

Sentence (i) is fine in Dutch, with één ‘one’ as a numeral and quantitative er as a pro-form that is related to the noun phrase internal gap (Ø). (i) Els heeft een auto met trekhaak gekocht en Marie heeft eri Els has a car with towbar bought and Marie has there één øi zonder trekhaak gekocht. one without towbar bought ‘Els has bought a car with a towbar and Marie has bought one without a towbar.’ 13

14 The question arises why (especially) speakers of Southern Dutch permit a morphologically bare adjectival remnant. Corver and van Koppen (2011) suggest that dialectal variation might play a role here. Specifically, many southern varieties of Dutch morphologically distinguish three genders (masculine, feminine, neuter), while northern varieties have a two-gender system (neuter versus non-neuter). This contrast in the morphological gender system possibly plays a role in the different behavior of attributive adjectives in NPE contexts.



    b. %Els heeft [ een wit paardNeut.Sg.] verkocht Els has a white horse sold heeft [ een zwarte paard] verkocht has a black-e horse sold ‘Els sold a white horse and Marie sold a black one.’

en and

Marie Marie

According to the analysis just sketched, -e is an adjectival inflection which licenses the elided noun. Schematically, with ø representing the missing noun: [DP een [NP groen-e [NP ø]]]. In Corver and van Koppen (), it is proposed that -e is not an agreement marker but rather a focus marker, whose Spec-position is the landing site of a displaced focalized AP: [DP een [FocP groeni [Foc' -e [NP ti [NP ø]]]]]. In Corver and van Koppen (), on the other hand, -e is analyzed as a pro-form that spells out N(P). This pronominalization strategy (compare English one in a green one) yields the following structure: [DP een [NP groen [NP -e]]]. According to this last analysis, een groene does not involve nominal ellipsis, which means that the issue about the licensing of the missing noun becomes irrelevant. We refer the reader to these articles for discussion of the pros and cons of these proposals. NP-ellipsis can be triggered by a non-linguistic context or a linguistic context. For example, while looking at two bikes on display in a shop one can say (). The contents of the elided noun can also be recovered from an antecedent noun in the linguistic context, whereby NP-ellipsis normally requires that the overtly realized noun precede the empty one. See also Broekhuis : – for discussion. () Ik ga de groene kopen. I go the green buy ‘I will buy the green one.’ In () an example is given of a sentence in which the NP to which NP-ellipsis has applied precedes the “antecedent” noun phrase. Most speakers of Dutch reject sentences like (), instead preferring the order in which the overt noun precedes the empty one (see Broekhuis :  for discussion). () %Ik I

ga go

een a

witte white

kopen buy

nadat after

ik I

de the

groene green

fiets bike

heb have

verkocht. sold ‘I will buy a white bike after having sold the green bike.’ In this section, we have discussed NPE, especially NPE with adjectival remnants. We will come back to NPE with other remnants like possessors, wh-pronouns, and demonstratives in section ... For more information and in-depth analyses of nominal ellipsis in Dutch we refer the reader to Kester (b), Corver and van Koppen (, ), Kranendonk (), Broekhuis (), and Zwart ().





. C D

.................................................................................................................................. As shown by (), material can be missing in Dutch comparative constructions: () Jan heeft meer vrouwen uitgenodigd dan Jan has more women invited than ‘Jan invited more women than I had invited.’

ik __ I

had had

uitgenodigd. invited

() exemplifies the phenomenon of Comparative Deletion. The compared element vrouwen ‘women’ in the dan ‘than’-clause has been deleted under identity with the element vrouwen ‘women’, which is introduced by the comparative operator meer ‘more’ in the antecedent clause. It is generally assumed that the elided element in () is not simply a bare nominal element (vrouwen ‘women’) but rather a quantified nominal expression (x- vrouwen). Interesting evidence in support of the presence of an underlying quantifier (x-) in the compared constituent comes from an example like (a), which minimally differs from () in having so-called quantitative er (there, ‘of them’) in the comparative clause (Bennis , ). This clitic-like element typically occurs in clausal environments featuring a(n indefinite) direct object noun phrase whose nominal head is empty, possibly as a result of movement, and which is introduced by a cardinal or indefinite quantifier (see (b)). Analogously to the structure in (b), the comparative clause in (a) can be assigned the structure in (c), where QØ represents the silent (i.e., phonetically empty) quantifier x-. () a. Jan heeft meer vrouwen gevraagd dan ik er had gevraagd. Jan has more women invited than I there had invited ‘Jan has invited more women than I had invited.’ b. Jan dacht dat ik eri toen [ drie/veel ti] had gevraagd. Jan thought that I there then three/many had invited ‘Jan thought that I had invited three/many of them then.’ c. . . . dan ik eri [QØ ti] than I there ‘ . . . than I had invited.’

had had

gevraagd. invited

Besides the phenomenon of Comparative Deletion, Dutch also exhibits the phenomenon of Comparative Subdeletion (a) and the pattern in (b), a so-called Phrasal Comparative: ()

a. Jan heeft meer vrouwen uitgenodigd dan Marie __ Jan has more women invited than Marie had uitgenodigd. had invited ‘Jan invited more women than Marie had invited men.’ b. Jan heeft meer vrouwen uitgenodigd Jan has more women invited ‘Jan invited more women than I.’

dan than

ik. I

mannen men



   

In (a), the element that is missing in the comparative clause corresponds to a quantifying element (x-). In (b), there is no (comparative) clause that follows dan ‘than’, at least not at the surface. Instead of a full clause, we find a phrasal constituent (in casu the pronoun ik ‘I’) right after dan. The question, obviously, arises as to whether this phrasal constituent is base-generated as a complement of dan ‘than’ or should be treated as a remnant in an elided comparative clause. In what follows, we will discuss these three phenomena in more detail. Section .. discusses Comparative Deletion, section .. Comparative Subdeletion, and section .. Phrasal Comparatives. Section .. addresses the question to what extent Comparative Deletion and Comparative Subdeletion display coordination-like properties.

.. Comparative Deletion An important property of Comparative Deletion (henceforth CD) is the obligatoriness of elision: the compared quantified expression (x- vrouwen) in () must be deleted. Compare () with (). () *Jan heeft meer vrouwen uitgenodigd Jan has more women invited had uitgenodigd. had invited ‘Jan invited more women than I had.’

dan than

ik I

[veel many

vrouwen] women

As shown by (), CD not only applies to direct object noun phrases but also to noun phrases with other grammatical functions (see (a)) and adjective phrases (including those with an adverbial function). Notice by the way that the two occurrences of er in (a) are expletive subjects and not instances of quantitative er. () a. Er houden meer mannen van voetbal dan er __ van tennis there love more men of soccer than there of tennis houden. love ‘There are more men who are fond of soccer than there are men who are fond of tennis.’ b. Deze weg is tegenwoordig breder dan hij vroeger __ was. this road is nowadays wider than he formerly was ‘This road is wider nowadays than it used to be in the past.’ The elided compared phrase can be part of an embedded clause which is selected by a so-called bridge verb, as in (a). The elided phrase, however, cannot be contained within an island configuration (e.g., a wh-island); see (b). This grammatical behavior of Dutch CD-constructions suggests that a movement operation is involved in the derivation of the comparative clause (Chomsky b; Bennis ). In other words, there are good reasons to assume that the gap in the comparative clause is a trace that results from the application of a movement operation.





() Jan heeft meer romans besproken . . . Jan has more novels discussed ‘Jan has discussed more novels . . . ’ a. dan ik denk [ dat ik ooit __ than I think that I ever ‘than I think I will ever read.’ b. *dan than

ik I

me me

afvraag [ wonder

wanneer when

zal will

lezen]. read

ik __ I

zal will

lezen]. read

In the CD-constructions discussed so far, the comparative clause introduced by dan ‘than’ follows the matrix clause containing the antecedent compared phrase. For example, in (a), the comparative clause dan ik er had uitgenodigd ‘than I had invited’ follows the matrix clause Jan heeft meer vrouwen uitgenodigd ‘Jan has invited more women’. This means that the comparative clause is separated from the compared nominal expression (meer vrouwen ‘more women’) with which it stands in a dependency relation via the quantifier meer. Importantly, as shown in (), the comparative clause can also stand in a subordinate relation to the antecedent noun phrase: () [ Meer complimenten [ dan ik je nu __ gegeven heb]] more compliments than I you now given have zul je nooit meer krijgen. will you never again get ‘You won’t ever get more compliments than the number of compliments I just gave you.’ Notice that the antecedent compared phrase and the comparative clause together occupy the clause-initial position, i.e., the position preceding the finite verb that occupies the second position in Dutch main clauses.

.. Comparative Subdeletion In Comparative Subdeletion (henceforth CsubD) a quantifying element (x-) is missing in the comparative clause. As shown in (), elision of this quantifier is obligatory. Compare () with (a). () *Jan heeft meer Jan has more had uitgenodigd. had invited

vrouwen women

uitgenodigd invited

dan than

Marie [ Marie

veel many

mannen] men

CsubD can apply to nominal expressions with different grammatical functions (a) and to adjectival expressions that function as predicate complements (b) or modifiers:



   

() a. Er houden meer mannen van voetbal dan er [SU __ vrouwen] van there love more men of soccer than there women of tennis houden. tennis love ‘There are more men fond of soccer than there are women fond of tennis.’ b. Deze weg is breder dan hij [ __ lang] is. this road is wider than he long is ‘This road is wider than it is long.’ Compared to the CD-constructions in (), CsubD-constructions in which the deleted quantifier is part of an embedded clause selected by a bridge verb seem to be degraded (see (a)). It is not entirely clear whether the ill-formedness of the patterns in (b,c) is due to a violation of an island constraint—implying that the gap corresponding to the “subdeleted” quantifier results from movement—or a kind of complex phrase constraint on deletion operations, which states that the antecedent of deletion should not be arbitrarily far from the deletion site (see Bresnan ; Chomsky b). Importantly, under the latter interpretation, CD and CsubD involve different mechanisms for creating a gap in the comparative clause (viz. movement versus deletion). () Jan heeft meer romans besproken . . . Jan has more novels discussed ‘Jan discussed more novels . . . ’ a. ?? dan ik denk [ dat ik ooit [ __ than I think that I ever ‘than I think I will ever read poems.’ b. *dan than

ik I

c. *dan than

ik [ I

me me

afvraag [ wonder

iemand someone

wanneer when

die [ __ that

gedichten] poems

ik [ __ I

gedichten] poems

zal will

gedichten] poems gelezen read

lezen]. read zal will

heeft] has

lezen]. read ken. know

For certain speakers of Dutch, CsubD cannot apply when the comparative clause is embedded within a noun phrase headed by the antecedent compared constituent (see ()). There are also speakers who judge these sentences as quite acceptable. For the first group of speakers, there is a strong contrast between CsubD () and CD (). () */? [ Meer complimenten [ dan ik je nu __ more compliments than I you now adviezen gegeven heb]] zul je nooit meer advices given have will you never anymore

krijgen. get

.. CsubD, CD, and coordination In the previous section, we discussed a number of phenomena for which CsubD seems to display different behavior from CD, specifically long-distance behavior ((a,b) versus





(a–c)) and embedding behavior (() versus ()). If these contrasts are real, then different syntactic mechanisms are possibly at the basis of the derivation of these two types of comparative construction. Another contrast between CD and CsubD regards the phenomenon of double comparison (cf. von Stechow ; Corver , ; Rijkhoek ). According to von Stechow, a sentence like () must be interpreted as involving two pairs of compared elements. The first comparison is between the number of girls and the number of boys (i.e., the indirect objects in ()), and the second comparison is between the number of flowers and the number of postcards (i.e., the direct objects). () Jan heeft meer meisjes meer bloemen gestuurd dan hij Jan has more girls more flowers sent than he [ __ jongens] [ __ kaarten] heeft gestuurd. boys postcards has sent ‘Jan sent more flowers to more girls than he sent postcards to boys.’ Double comparison is impossible with CD-constructions: ()

*Meer mannen more men geschreven. written

hebben have

meer more

gedichten poems

gelezen read

dan __ __ than

hebben have

The question, obviously, arises how this contrast between CD and CsubD can be accounted for. A line of analysis that has been explored in the literature regards the nature of the element dan ‘than’, which introduces the comparative clause. It has been proposed that dan ‘than’ in CD-constructions behaves like a subordinator whereas dan ‘than’ in CsubDconstructions behaves more like a coordinator (Corver , ; Hendriks , ). Note that, under this approach, the ill-formedness of () simply follows from the fact that the coordinator dan ‘than’ does not conjoin two phrases of the same categorial and semantic type. That is, meer complimenten ‘more compliments’ in () and ik je nu adviezen gegeven heb ‘I gave you bits of advice now’ are not “like” categories. If a CsubD-construction involves a coordination of the antecedent clause and the comparative clause (with dan ‘than’ as the conjoining element), we would expect other traits that hint at a resemblance between CsubD-constructions and coordinate constructions. One piece of evidence in support of this coordinative relation comes from gapping (see section .), which is typically found in coordinate structures (see (a)). As shown in (b), it is possible to delete the (clause-final) finite verb of the dan ‘than’-clause under identity with the finite verb (in Verb-Second position) of the preceding clause. ()

a. Jan spreekt Romaanse talen en Marie spreekt Germaanse Jan speaks Romance languages and Marie speaks Germanic talen. languages ‘Jan speaks Romance languages and Marie Germanic languages.’



    b. Jan spreekt meer Romaanse talen dan Marie Germaanse Jan speaks more Romance languages than Marie Germanic talen spreekt. languages speaks ‘Jan speaks more Romance languages than Marie Germanic languages.’

Secondly, we saw in section .. that Backward Conjunction Reduction is also typically found in coordinate structures. Notice now the CsubD-constructions in () where bewonderen ‘admire’ has undergone BCR. ()

Ik denk dat meer vrouwen de schilderijen van Picasso bewonderen I think that more women the paintings of Picasso admire dan mannen de tekeningen van Rembrandt bewonderen. than men the drawings of Rembrandt admire ‘I think that more women admire the paintings of Picasso than men the drawings of Rembrandt.’

Although the examples in () and () seem to support the proposal that CsubDconstructions involve a coordination of the antecedent clause and the comparative clause, they also raise a new question. It turns out that both gapping and Backward Conjunction Reduction are also found in CD-constructions, as is exemplified in (a) and (b), respectively. () a. [Jan heeft aan Marie meer boeken gegeven] Jan has to Marie more books given dan [Piet aan Els heeft gegeven]. than Piet to Els has given. ‘Jan has given more books to Marie than Piet to Els.’ b. Ik denk [ dat meer jongens Duits zullen I think that more boys German will dan [ Frans zullen kiezen]. than French will take ‘I think that more boys will take German than French.’

kiezen] take

The possibility of having gapping and BCR in CD-constructions suggests that this type of comparative construction also has coordination-like characteristics. In other words, the earlier statement that dan ‘than’ in CD-constructions behaves like a subordinator needs further investigation. We leave that for future research.

.. Phrasal comparatives Phrasal comparatives have the characteristic property that a single phrase follows the comparative conjunction dan.

 ()

Jan heeft op woensdag meer mannen Jan has on Wednesday more men ‘Jan invited more men on Wednesday . . . ’ dan Piet/vrouwen/op zaterdag. than Piet/women/on Saturday ‘than Piet/women/on Saturday.’



uitgenodigd . . . invited . . .

These structures trigger the following question: Are these phrasal comparatives derived from a clausal source through deletion of identical material, or is the phrase following dan ‘than’ a direct (i.e., base-generated) complement of dan ‘than’? An argument in support of a clause-reduction analysis comes from phrasal comparatives like (), in which the phrase following dan ‘than’ is a pronominal subject carrying nominative case. Under a clausereduction analysis, the nominative case would simply be assigned by the finite verb (i.e., T) before it gets deleted at PF. This argument is weakened, though, by the fact that the default case in Dutch is nominative (unlike in English), as is clear from an example like IkNOM dit probleem oplossen? Geloof je het zelf? ‘Me solve this problem? Do you really believe it?’. Possibly, ik in () is a default case. () Jan heeft op woensdag meer mannen Jan has on Wednesday more men ‘Jan invited more men on Wednesday than I.’

uitgenodigd invited

dan than

ikNOM. I

It should also be noted that in colloquial Dutch one often hears dan mij ‘than me’ instead of dan ik ‘than I’ in a sentence like (). The object form mij ‘me’ suggests that it behaves like a complement of dan ‘than’ and not as the subject of a reduced finite comparative clause. There are arguments that go against a clause reduction analysis of phrasal comparatives (see Hoeksema ; Hendriks ). For example, there are phrasal comparatives for which no clausal source can be found. In (), wereldrecord ‘world record’ cannot possibly be interpreted as an argument of sprong ‘jumped’. ()

Jan sprong hoger dan het wereldrecord Jan jumped higher than the world.record ‘Jan jumped higher than the world record.’

(*sprong). (jumped)

For more information and in-depth analyses about comparative deletion in Dutch we refer the reader to Bennis (), den Besten (), Corver (, , ), Hendriks (, ), and Lechner and Corver ().

. N C A

.................................................................................................................................. The phenomenon of Null Complement Anaphora (NCA) involves the elision of a complete complement after certain verbal predicates (Depiante ). As shown by the following



   

examples, it is generally impossible to elide the complement of a verb in Dutch, even if the contents of the elided material are recoverable on the basis of the linguistic () or discourse context (). () *Op het moment dat Piet at the moment that Piet vermoedde ik het ook. suspected I it too

het it

() A: Realiseer je je wat dat realize you you what that ‘Do you realize what that means?’ B: *Ja, yes

ik I

realiseer realize

me me

heel very

begon began

te to

vermoeden, suspect,

betekent? means

goed well

wat

dat

betekent.

what

that

means

Note in passing that the ill-formed sentences in ()–() are fine if a pronominal element fulfills the role of complement. This is illustrated in () for () and (): ()

a. Op het moment dat Piet het begon te vermoeden, at the moment that Piet it began to suspect, vermoedde ik het ook. suspected I it too ‘At the moment that Piet began to suspect it, I suspected it too.’ b. Ja, ik realiseer me dat heel goed. yes I realize me that very well ‘Yes, I realize that very well.’

It is important to distinguish the phenomenon of Null Complement Anaphora from the phenomenon of Topic drop (see Weerman ). The latter phenomenon is exemplified in (), where ‘Ø’ represents the dropped constituent: ()

A: Jan is ziek. Jan is ill ‘Jan is ill.’

B: Dat/Ø vermoedde ik al. that suspected I already ‘That’s what I suspected.’

Topic drop can only apply to a constituent that occupies the left periphery of the main clause (i.e., specCP). It is generally assumed that the topic ends up in specCP as a result of movement. It has been shown that constructions involving Topic drop are island-sensitive. NCA does not seem to be island-sensitive in view of the well-formedness of an English example like: The teacher told the children that it was time to leave even though they already knew Ø. In this example, the gap (Ø) is contained within an adjunct island introduced by even though. This suggests that NCA involves a base-generated (pronominal) gap (see Depiante ). Although the default rule seems to be that NCA is not attested in Dutch, examples like the following can be found:

 ()

a. Ik had mijn vriend gevraagd een ring voor me I had my friend asked a ring for me maar hij weigerde. but he refused ‘I had asked my friend to buy a ring for me, but he refused.’

te to



kopen buy

b. Ik vroeg Jan om mij  euro te lenen en na enig I asked Jan for me  euro to lend and after some aandringen stemde hij in. insisting agreed he  ‘I asked Jan to lend me  euros and after some insisting he agreed.’ Notice, however, that verbs like weigeren ‘to refuse’ and instemmen ‘to agree’ are simply optionally transitive, just like the verb koken ‘to cook’, as in Jan kookte (de aardappelen) ‘Jan cooked (the potatoes)’. In other words, the examples in () are not true instances of NCA. Consider also the comparative constructions in (); see also Cremers (): ()

Messi had meer doelpunten Messi had more goals ‘Messi scored more goals . . . ’ a. dan than b. dan than

hij __ he hij __ he

gemaakt . . . made . . .

dacht/besefte/vermoedde/hoopte. thought/realized/suspected/hoped had voorspeld/verwacht/beoogd. had predicted/expected/intended

In these examples, the dan ‘than’-clause of a comparative construction contains a verbal predicate whose complement has been elided. The contents of the elided complement are recoverable on the basis of the matrix clause containing the compared constituent. For example, the null complement of dacht ‘thought’ in (a) can be interpreted as: ‘that he would score x-many goals’. The question obviously arises what the nature of the gap is in these examples. Since base-generated null-complements are generally impossible in Dutch (see ()–()), one might hypothesize that the gap results from a movement operation. One approach would be to say that the clause following dan ‘than’ is a free relative clause whose operator (in specCP) is null. This free relative hypothesis is represented schematically in (): ()

[ dan [NP Ø [CP OPi than ‘ . . . than he thought.’

hij ti he

dacht]]] thought

According to his analysis, dan ik dacht ‘than I thought’ is a phrasal comparative with dan ‘than’ as a prepositional element that combines with a nominal complement whose head is empty and which contains a relative clause (see den Besten  for the existence of free relatives as complements of dan ‘than’). For certain speakers, it is possible to have an overt realization of the relative pronoun of the free relative clause, as in (). Other speakers reject such examples. This inter-speaker variation is represented here by means of %.



   

()

De rekening is iets hoger dan the bill is a.bit higher than ‘The bill is a bit higher than I expected.’

(%wati) what

ik I

had ti had

verwacht. expected

Possibly, this free relative analysis extends to constructions like (), which also do not have an overt complement after the transitive verb. ()

Naar (%wat) ik verwacht, kost het ongeveer  euro. after what I expect cost it approximately  euro ‘According to my expectations, it will cost approximately  euros.’

()

[ naar [NP Ø [CP OPi ik ti verwacht]]] after I expect ‘According to what I expect . . . ’

So far, the phenomenon of Null-Complement Anaphora has not been a topic of extensive investigation in research on Dutch ellipsis. For some information, see Cremers ().

. E  D 

.................................................................................................................................. So far, this chapter has focused on standard Dutch. There is, however, an extensive literature on ellipsis phenomena in dialects of Dutch. We want to use this final section to discuss some of these constructions. We will go into clausal ellipsis in subsection .. and nominal ellipsis in subsection ...

.. Clausal ellipsis in Dutch dialects The Dutch dialects show several interesting cases of clausal ellipsis, which are most recently discussed in van Craenenbroeck (, b). In particular, he discusses socalled Spading (i.e. Sluicing Plus a Demonstrative in Non-insular Germanic), Short Do Replies, and conjugated yes and no. We will, very briefly, discuss these ellipsis cases below. First consider the example of spading in () from the Dutch dialect of Wambeek (spoken in the Belgian province of Brabant) (from van Craenenbroeck b: ). ()

a. Jef ei gisteren iemand gezien. Jef has yesterday someone seen ‘Jef saw someone yesterday. Who?’

Wou who

b. Jef eid iemand gezien, mo ik weet Jef has someone seen but I know ‘Jef saw someone, but I don’t know who.’

da? that nie not

wou who

da. that





Spading is impossible in standard Dutch: ()

a. Jef heeft gisteren iemand gezien. Jef has yesterday someone seen ‘Jef saw someone yesterday. Who?’

Wie who

(*dat)? that

b. Jef heeft iemand gezien, maar ik Jef has someone seen but I ‘Jef saw someone, but I don’t know who.’

weet know

niet not

wie who

(*dat). that

Spading is a subtype of sluicing in which the remnant contains both a wh-element, wou ‘who’ in this case, and a demonstrative pronoun, da ‘that’. As van Craenenbroeck shows, spading is not possible in non-sluiced contexts (see (a)) and only involves simplex wh-elements (b) (from van Craenenbroek b: –):15 ()

a. Ik vruig ma af me wou (*da) I ask me PRT with who that ‘I wondered to whom Lewie has talked.’ b. *Welken boek dat? which book that

da that

Lewie Lewie

geklapt talked

eit. has

The structure underlying spading is not simply ellipsis of the IP complement of C (just like regular sluicing), but rather IP-ellipsis of an underlying cleft. The underlying structure in spading can be represented as follows: ()

Jef ei gisteren iemand gezien. Woui da Jef has yesterday someone seen who that ‘Jef saw someone yesterday. Who?’

is ti

da

Jef

gezien

ei?

is

that

Jef

seen

have

Van Craenenbroeck shows that the Southern Dutch dialects have three further ellipsis patterns that are unattested in standard Dutch: Short Do Replies (SDR), da’s nie/da’s wel, and conjugated yes and no. We provide examples from Wambeek Dutch (van Craenenbroeck b: –) in (a, b, and c) respectively. ()

a. Short Do Reply A: Marie zie Pierre Marie sees Pierre ‘Marie loves Pierre.’

geirn. gladly

B: Z’en duut. she.NEG does ‘She doesn’t.’

Van Craenenbroeck (, b) provides several arguments to show that the da ‘that’ in spading is a demonstrative pronoun rather than a complementizer. We will not repeat these arguments here, but we refer the reader to the original work. Furthermore, one might wonder whether the ungrammaticality in (a) is simply a case of haplology. Again we refer the reader to van Craenenbroeck (, b) for arguments against this hypothesis. 15



    b. Da’s wel/nie A: Marie gaat naar de Marie goes to the ‘Marie goes to the movies.’

film. movies

c. Conjugated yes/no A: Kom Marie mergen? comes Marie tomorrow ‘Is Marie coming tomorrow?’

B: Da’s nie. that.is not ‘No, she doesn’t.’ B: Nieje-s. no.she ‘No.’

SDRs contradict the previous declarative statement. The remnant in SDRs consists of the verb to do, a subject clitic, and, if the context requires it, a negative clitic (i.e., if a positive statement is being contradicted). The second set of data, dubbed da’s wel/niet ‘that is so, that is not so’ are argued to be derived from the same structure as SDRs, as are conjugated yes/no in which the polarity marker yes/no can be inflected and/or bear a subject clitic. Again these constructions are ungrammatical in standard Dutch: ()

a. Short Do Reply A: Marie houdt niet van Marie loves not of ‘Marie doesn’t love Pierre.’

Pierre. Pierre

B: *Ze doet. she does ‘She does.’

b. Da’s wel/nie A: Marie gaat naar de film. Marie goes to the movies ‘Marie goes to the movies.’

B: *Da’s nie. that.is not ‘No, she doesn’t.’

c. Conjugated yes/no A: Kom Marie morgen? comes Marie tomorrow ‘Is Marie coming tomorrow?’

B: *Nee-ze. no.she ‘No.’

We refer the reader to van Craenenbroeck (, b) for extensive discussion and analysis of these data.

.. Nominal ellipsis in Dutch dialects The Dutch nominal domain shows a lot of microvariation concerning ellipsis as well. First consider ellipsis in possessive constructions. There are three prenominal possessive constructions in Dutch illustrated in example (): a prenominal, pronominal possessor, possessor doubling, and the Saxon genitive (cf. Corver ; van de Craats, Corver, and van Hout ). ()

a. z’n auto his car ‘his car’ b. Jan z’n Jan his ‘Jan’s car’

auto car

pronominal possessor construction

possessor doubling construction

 c. Jans auto Jan’s car ‘Jan’s car’



Saxon genitive

Standard Dutch only allows NPE in constructions like (a), but not in (b,c): ()

a. *(de) the ‘his’

zijne his

b. * Jan (de) zijne Jan the his ‘Jan’s’ c. *Jans Let us first consider the example in (a). This remnant structure necessarily contains a definite article. This definite article cannot be part of the non-elided version, however: ()

(* de) zijn the his ‘his car’

auto car

Corver and van Koppen () argue that the definite article in these NPE structures is a pronoun which pronominalizes the ‘elided’ part of the possessive DP, similar to the role of one in English NPE with adjectival remnants, i.e., the red one. In particular, they argue that this definite article spells out the DP-structure below the possessive pronoun in the same way that one spells out the DP-structure below adjectives in English pronominalization: ()

a. [DP a [AdjP red [NP car]]] one ‘a red one’ b. [DP [PosP mijn [AdjP [NP auto]]]] de

One argument in favor of this analysis is that de cannot be combined with adjectives or numerals, which are usually assumed to occupy a position above NP and below DP: ()

a. (* de) zijn groene the his green ‘his green ones’ b. (* de) zijn vier the his four ‘his four’



   

Corver and van Koppen argue that the definite article is present in these constructions to recover the gender agreement of the elided possessee, since the possessive pronoun is not capable of doing that. The first person singular possessive pronoun, for instance, always appears in the form mijn ‘my’, independent of the gender and number of the possessee. As a result, the remnant of ellipsis, the possessive pronoun, cannot recover the gender information of the elided possessee. The definite article is marked for gender, and hence this element has to appear in these ellipsis constructions. There are Dutch dialects, however, that do show agreement between the possessor and the possessee, like for instance the dialect of Winterswijk. ()

a. masculine singular: b. feminine singular: c. plural:

mien-en hood mien-e muts mien-e mutse

‘my hat’ ‘my bonnet’ ‘my bonnets’

Corver and van Koppen’s analysis correctly predicts that the definite article does not have to appear in these types of dialects: ()

a. masculine singular: b. feminine singular: c. plural:

(d’n) mienen (de) miene (de) miene

‘mine’ ‘mine’ ‘mine’

There are also dialects that have a pronominalization strategy comparable to standard Dutch, but with a different pronoun: Hindeloopen Dutch, for instance, where ellipsis of the possessee leads to an en-affix or s-affix on the remnant. The en-affix appears with the singular possessive pronouns and the s-affix with the plural possessive pronouns. This is illustrated in (). ()

a. b. c. d.

masculine singular: feminine singular: neuter singular: plural:

mien-en mien-en mien-en mien-en

‘mine’ ‘mine’ ‘mine’ ‘mine’

uze-s uze-s uze-s uze-s

‘ours’ ‘ours’ ‘ours’ ‘ours’

Corver and van Koppen argue that these en- and s-endings are also a pronominal element spelling out part of the possessed DP. Let us now turn to the possessor doubling construction in (b) and (b). As these examples show, standard Dutch does not allow ellipsis in this construction. There are, however, dialects that do allow ellipsis in the possessor doubling construction; see (). These dialects differ among each other as well. In Asten Dutch, (a), the definite article is obligatorily present. In Winterswijk Dutch it is optional, just like in example (), and in Hindeloopen Dutch the definite article is absent (as expected on the basis of example ()), but the invariant form of the possessive pronoun appears: ()

(Over auto’s gesproken / Talking about cars) a. Ik vein Teun *(de) zinnen echt I find Teun the his really ‘I find Teun’s really great.’

geweldig great (Asten Dutch)

 b. Vonnebos (de) Vonnebos (the) ‘Vonnebos’s’ c. Jan (*de) Jan (the) ‘Jan’s’

zien-e ___ his-fem.sg



(Winterswijk Dutch)

sienen his

(Hindeloopen Dutch)

We refer the reader to Corver and van Koppen () for extensive discussion and an analysis of this pattern. Finally, let us look at the pattern in (c) and (c). Again, standard Dutch does not allow ellipsis with the Saxon genitive, but some dialects do. The example in () from the dialect of Katwijk is one example of this. ()

Ik vind Teuns I find Teun’s ‘I love Teun’s.’

(auto) (car)

geweldig. great (Katwijk Dutch)

The pronominalization strategy we have discussed above in () is found in several other NPE contexts as well. For instance, with adjectives (), demonstratives (), and wh-pronouns () (see also Corver and van Koppen , ): ()

Piet ei een vervelend-e/*-en opa en Jan ei een leuk-en. Piet has an annoying-e/-en grandfather and Jan has a nice-en ‘Piet has an annoying grandfather and Jan has a nice one.’ (Zierikzee Dutch)

()

(*Den) deze opa is al oud, maar the this grandfather is already old but ‘This grandfather is already old, but that one isn’t.’

()

a. de the-masc/fem

waffer-en what.for-masc

b. de the-masc/fem

waffer what.for

c. et waffer the-neut what.for ‘what kind of ’

den the

dieje nie. that not (Zierikzee Dutch)

(Oerle Dutch, de Bont )

The example in () shows the pronoun -en that we already saw with possessive pronouns in example (), appearing with adjectives. The en-affix is not simply adjectival inflection, since that is an e-affix as the first part of the example shows. Another argument for the pronominal status of this en-affix, as Corver and van Koppen () show, is that in a sequence of adjectives it only appears on the final one. The other ones show regular adjectival inflection. We also find this en-pronoun with demonstrative pronouns and wh-pronouns in several dialects.



   

Example () provides an example of the pronoun de, which we already saw in the possessive constructions discussed above, but now with demonstrative remnants rather than possessive remnants. The dialects that show this de in elliptical demonstrative constructions are a subset of the dialects with de in elliptical possessive constructions. The same holds for the appearance of de/et in () with elliptical wh-pronouns. The dialects that show the construction in () are again a subset of the dialects that have the construction in (). We refer the reader to Corver and van Koppen (, , , ) for extensive discussion and an analysis of these patterns.

. S

.................................................................................................................................. To summarize, this chapter has discussed the major ellipsis constructions in Dutch. We have discussed the properties of gapping and stripping (section .), predicate ellipsis (VP-ellipsis and pseudogapping) (section .), Conjunction Reduction and Right-Node Raising (section .), sluicing (section .), fragments (section .), nominal ellipsis (section .), Comparative Deletion (section .) and Null Complement Anaphora (section .). Finally, in section . we have discussed some special instances of ellipsis in Dutch dialects.

A We would like to thank three anonymous reviewers as well as the editors of this volume for extensive and useful comments on a previous draft of the chapter.

  ......................................................................................................................

                  ......................................................................................................................

 

. I

.................................................................................................................................. T chapter discusses ellipsis in Finnish Sign Language (FinSL). First, in order to provide a necessary background for the discussion, an overview is given of both the basic grammar of FinSL (section .) and the larger role ellipsis plays in FinSL (section .). After this (sections .–.), several elliptical phenomena discussed in this book (nominal ellipses, conjunction reduction, VP-ellipsis, sluicing, gapping, stripping, and fragment answers) are introduced and described for FinSL, together with other elision phenomena (e.g. argument ellipsis, also known as pro-drop). Not all the phenomena can be discussed in equal depth, nor can all the phenomena presented in the book be addressed, because research into ellipsis in sign languages, FinSL included, is still in its early stages (e.g., Cecchetto et al. ; Jantunen ; the majority of the work carried out so far has dealt almost exclusively with argument ellipsis: see Bahan et al. ; Jantunen ; Koulidobrova ; Wulf et al. ). Finally, the chapter takes a broader view of ellipsis once again and contemplates the role that gesture and mime have in the elliptical phenomena found in signed languages, and perhaps also in spoken languages (section .).

. B  FSL 

.................................................................................................................................. The grammar of FinSL, and consequently ellipsis in the language too, has been investigated from the ontological premises of the functional (communicative–cognitive) approach outlined, for example, in Van Valin and LaPolla (). Methodologically, the research has been based on the general conventions of Basic Linguistic Theory (Dixon ; Dryer ; Jantunen ), a framework which is widely used by, for example, field linguists, and which emphasizes informal description over formal explanation. On the other hand, research into FinSL grammar has also been influenced by one of the guiding assumptions of modern functional typology: that is, that individual languages should be described as much as possible in their own terms with locally defined categories (e.g., Haspelmath a, ; Jantunen ).





.. Major sign classes The word-like units—signs—of FinSL, as of all sign languages, vary in the degree to which their form–meaning connection is fixed (e.g., Jantunen ; Johnston and Schembri ; Liddell ). In many signs this connection is highly conventionalized and categorical, a relatively stable form always conveying a relatively unambiguous meaning. However, in some signs the form–meaning connection is based on looser principles, either the form (or some aspect of it) or the meaning, or both, displaying a relatively high degree of unconventionality and gradience. On the basis of this difference, FinSL signs have been distributed on a sign-type continuum of which (conventional, categorical, and often arbitrary) monomorphemic lexemes form one end and (unconventional, gradient, and typically highly iconic) pantomimic gestures the other (for more on the underlying semiotic view on gesture, see Okrent ). Apart from clear instances of gesture and mime, signs can also be assigned into sign classes (cf. word classes, parts of speech). In FinSL, the major sign classes are the nominal and the verbal (Jantunen ; Rissanen , ). These categories have been defined by semantic and grammatical criteria such as reference (nominals refer to entities, verbals to temporally manifested dimensions of events), the marking of aspect (the markers of aspect and Aktionsart distinctions attach only to verbals), and distribution in clauses (the position of verbals in clauses is more constrained than that of nominals; see section ..). The categories of nominal and verbal are not the same as the categories of noun and verb, for various reasons. First, unlike traditionally defined nouns and verbs, nominals and verbals in FinSL also include signs which are closer to the gesture end of the sign-type continuum. Examples of such nominal signs—that is, nominal signs with a gestural component—are the various pointing signs referring to entities, which in other frameworks have often been analyzed as pronouns but in FinSL research are treated as a subclass of nominals. Examples of verbal signs with a gestural component, on the other hand, are the so-called Type  verbals, which can be meaningfully directed in space, as well as the formationally and semantically complex depicting signs, which in the present framework are called Type  verbals. Second, unlike traditionally defined nouns and verbs, the definitions of FinSL nominals and verbals also cover signs that denote property. Consequently, as characterizing signs are analyzed according to their semantics and grammatical behavior as either marginal nominals or marginal verbals, there is no need for a third major sign class, adjectives, in FinSL (Jantunen ). The typology of verbal signs is relevant in a discussion of ellipsis in FinSL. In general, FinSL verbals are analyzed as belonging to one of three main subcategories: Type , Type , and Type  verbals (Jantunen , , ; for other types of classifications, see Engberg-Pedersen ; Liddell ; Padden ; Rissanen ). Type  verbals are formationally the most fixed type of FinSL verbals, with a relatively straightforward form– meaning connection. Type  and Type  verbals, on the other hand, are groups of signs which include gradient features (i.e., a gestural component) as part of their structure. In Type  verbals (as, for example, in pointings), the gestural component is manifested through the directionality of the movement of the hand: in order to understand the meaning of these signs, the addressee must make a semantic association between the morphological content of the sign and the location toward which the sign is directed (see

  



 . FinSL verbals with the meanings (from left to right) ‘know’ (the finger pads of the open hand touch the forehead twice), ‘teach [someone in front of the signer]’ (the two hands move forward twice in the shown configuration), and ‘an oblong vehicle (e.g. a bicycle) drives forward over a mound-like location’ (the dominant hand articulates an arc-shaped movement over the stationary non-dominant hand). The verbals represent Type , Type , and Type , respectively (see also Figure .) Source: Suomalaisen viittomakielen perussanakirja ()

Liddell ). In Type  verbals, the gestural component refers to the analogous and gradient properties inherently present in the placement, orientation, and movement parameters of the sign structure. Together these features enable Type  verbals to iconically depict various events that have movement taking place in different topographic locations; Type  verbals also contain a nominal classifier morpheme (the handshape), which refers to the entities present in the event. Examples of the three verbal types are given in Figure ..

.. Basic syntax Both nominals and verbals can function as predicates in FinSL clauses (Jantunen , ). Simple declarative (e.g., identificational, characterizing) clauses with a nominal predicate have no copula inbetween the two nominal phrases of which the latter expresses the main predication (Jantunen ). Declarative clauses with a semantically one- or two-place verbal predicate (V) display more variation in their structure: while intransitive clauses follow the SV pattern (S=single participant, prototypically the agent), transitive clauses may be ordered either according to the AVP pattern or the APV pattern (A=active participant, prototypically the agent; P=passive participant, prototypically the patient). The factors affecting the choice of the order are primarily textual (e.g., thematic continuity) and semantic (e.g., avoidance of ambiguity), the AVP order being the more common of the two (Jantunen ). The predicates of the AVP and APV transitive clauses discussed above are either Type  or Type  verbals. Type  verbals form a syntactically separate class as they are considered to be full, well-formed clauses on their own (Jantunen ). The analysis derives on the one hand from the fact that the classifier handshape or handshapes included in these verbals can be analyzed as nominal core arguments of the predicate, and on the other from





the fact that the meaning of these verbals is typically very clause-like: that is, it covers the whole event. Due to the semantics of classifiers, however, there is often some vagueness in the meaning of such verbals/clauses. For this reason, Type  verbals may be preceded by one or two topics, which are nominal phrases that function to set an interpretative framework for the main predication (here, the Type  verbal). In such structures, the Type  verbal is always sentence-final (Jantunen ). Due to their articulatorily bound status as well-formed clauses, Type  verbals are not a direct target of the elliptical phenomena discussed in this chapter. In general, the topic–comment structure of FinSL can be used to encode all types of events (e.g., identificational, intransitive, transitive, etc.) with all kinds of signs. The topic of the topic–comment structure is a clause-external left-detached nominal phrase, whose function is to set a spatial, temporal, or an individual framework for the subsequent comment. Topics in FinSL are marked syntactically/prosodically by their sentence-initial position, topic-final pause, and a layered non-manual feature, “eyes widened and eyebrows raised.” It is also possible to mark topics morphologically, that is, with a topic-final index finger pointing (Jantunen ). Adding a topic in front of the clause makes the structure as a whole more complex. Another way to create complex structures is to bind together two (or more) clauses. Structurally, the resulting complex sentences are traditionally analyzed as being constructed either through coordination or through subordination. To date, only clausal coordination has been investigated in FinSL (Jantunen , ). Concerning its three main types (Haspelmath b), conjunctive coordination (‘and’) in FinSL is primarily asyndetic (i.e., it relies on juxtaposition), whereas adversative (‘but’) and disjunctive (‘or’) coordinations are primarily expressed syndetically (i.e., they rely on attaching coordinators, the signs BUT and OR, respectively, to the second coordinand). Reverse options for all three types are also possible: that is, conjunctive coordination can be expressed syndetically (the signs ALSO and PLUS attaching to the second coordinand function as coordinators) and adversative and disjunctive coordinations can be expressed asyndetically (with the appropriate prosody and other non-manual behavior; see section ..). Because of their relatively infrequent use, the reverse forms are treated in FinSL research as marked. In constructing central functional sentence types other than declarative sentences—that is, negative and interrogative sentences—FinSL employs manual signs together with specific non-manual operators (Rissanen ; Savolainen ). The basic negative operator in FinSL is a headshake. The scope of the headshake can be the whole sentence or a shorter sequence. However, no matter where the headshake begins, it tends to last to the end of the sentence. The interrogative operator in FinSL has two main forms, one for yes/no questions and the other for wh-questions. In the main yes/no question marker, the head tilts forward and the eyebrows are raised, while in the main wh-question marker the forward head tilt is accompanied with furrowed eyebrows. The scope of the non-manual interrogative operator is conditioned similarly to that of the negative operator.

.. Constructed action When situated in the wider cross-linguistic context, the basic grammar of FinSL, and especially its syntax, shares many of the characteristics found in the world’s spoken

  



languages. However, this is not to say that the visual–gestural channel of FinSL—as of all sign languages—does not affect the manifestation of grammatical phenomena: there definitely are modality differences. In addition to signs with gestural components, another salient example of these is the frequent use of constructed action: that is, of gestural enactment (mime) in the creation of linguistic messages. The use of constructed action is not exclusively limited to the domain of sign languages (e.g., Ladewig ) but—as, for example, Liddell (), Lukasczyk (), and Ferrara and Johnston () have shown—sign languages employ it extensively and very similarly on various grammatical levels. In FinSL, constructed action is almost obligatorily linked to the use of Type  and Type  verbals: when these verbals are used in signing, the signer typically also shows via “acting” some aspects of the location, behavior, or motion of the entity participating in the event. Another domain where constructed action is very saliently employed in FinSL is direct speech: to directly report the words, actions, or thoughts of others, signers literally transform their bodily behavior via enactment into that of those others. Moreover, together with other types of use of space (e.g., pointing to spatial locations with the index finger and articulating signs in different spatial locations), constructed action is also an important means to indicate and increase discourse cohesion in FinSL and to maintain discourse reference. On a general level, Ferrara and Johnston () have shown that constructed action combines with all types of signs and clauses to form composite expressions that affect the construction of meaning in sign languages. They have also argued—in line with claims made on FinSL (Jantunen , )—that constructed action can also influence the surrounding grammatical structure. This observation is relevant for the discussion of ellipsis in FinSL and we will return to it in sections .. and ..

. O        FSL

.................................................................................................................................. Let us now turn to elliptical phenomena, and first to their overall role in FinSL grammar. Previous studies (Ala-Sippola ; Jantunen , ) have demonstrated that, in terms of their preferred argument structure, the clauses discussed in section . are frequently realized in a way that can be considered to be incomplete. For example, Ala-Sippola () found that  percent of the clauses in her data ( instances out of a total of  clauses) occurred without the expected S/A argument. There was thus a high incidence of argument ellipsis in Ala-Sippola’s data, but this is fully in line with the numbers presented for other sign languages: for example, in their quantitative study on pro-drop, Wulf et al. () found that American Sign Language expressed only  percent of its subject-like core arguments belonging to what they defined as the class of pronoun signs (see also Bahan et al. ; Koulidobrova ). Similar observations about the frequency of argument ellipsis have also been made regarding Danish Sign Language (Engberg-Pedersen ) and Australian Sign Language (Johnston and Schembri ). New corpus data processed for the purpose of this chapter further demonstrates that the elision of core arguments is indeed a frequent phenomenon in FinSL. In a sample of 





Table . Ellipsis of A and P core arguments in the sample of  transitive clauses containing a Type  or Type  verbal predicate Type of ellipsis A argument ellipsis P argument ellipsis Ellipsis of A or P argument Ellipsis of both A and P arguments

N

%

   

% % % %

structurally annotated intransitive (n=) and transitive clauses (n=) containing a Type  or Type  verbal predicate, collected from ten FinSL signers telling a story about a cartoon, a total of  clauses— percent—were produced without their S/A and/or P arguments. For intransitive clauses, the share of S argument ellipsis was  percent: that is, S was left unexpressed in  clausal instances. The corresponding figures for transitive clauses are given in Table .. As can be seen from Table ., argument ellipsis in transitive clauses of FinSL affects not only the A argument but also the P argument. Another important observation is that only  percent of the transitive clauses were realized with a syntactically full structure (see the row “ellipsis of A or P argument”). It is also worth noting that  percent of the transitive clauses consisted of only a two-place verbal predicate without any nominal elements. On the basis of the high frequency of argument ellipsis, for one thing, Jantunen () has claimed that FinSL is a highly discourse-oriented language (claims of this type have also been presented for other sign languages at least since Friedman , a recent overview being included in Kimmelmann ). Theoretically, Jantunen’s argument is grounded in McShane’s () cross-linguistic work on ellipsis. According to McShane, discourseoriented languages permit more elliptical phenomena than non-discourse-oriented languages, elide especially thematic elements, and use elliptical phenomena as a central means to support the discourse structure. In other words, in discourse-oriented languages the clause is not a unit that is highly governed syntactically, (con)textual and thematic relations being more important than grammatical ones in the creation and interpretation of meanings. The discussion so far has shown that, at least in terms of eliding clausal core arguments, ellipsis is highly permitted in FinSL: when contrasted to the ellipsis rates of non-discourseoriented languages such as Spanish, in which the ellipsis rate has been evaluated, on the basis of speech corpora, to be . percent (Alcántara and Bertomeu ), it is evident that FinSL fulfills the first of McShane’s criteria. With regard to the two other criteria, however, some further explication is needed. Let us first consider the claim that FinSL elides especially the thematic elements. The example in () shows an excerpt from a narrative in which the signer is talking about a journey he made abroad. (Note that in the examples, the signs are transcribed with uppercase glosses presenting the rough meaning of the signs; glosses may include additional information specifying, for example, the approximate location at which the sign is produced/directed; and the apostrophe marks a syntactic–prosodic break and the backslash a pause.) The story begins in line  with a pragmatic discourse particle marking the beginning of the narrative. In line , the signer produces a complex sentence with two conjunctively

  



coordinated intransitive clauses. The clause in line  is a transitive one and it is followed in line  by an embedded nominal locative clause explicating the setting in which the meeting referred to in line  took place. In line  the signer repeats the sentence in line  and goes on with the story. () The signer is talking about his journey abroad: line : SUCH-THINGS / line : ME FLY-THERE:left ’ STAY-THERE:left / line : MEET THERE:left / line : DALLAS AIRPORT THERE:left / line : MEET . . . ‘So, I flew there and stayed there. I met there, there at Dallas airport . . . I met there (the person responsible for international affairs.)’ (adapted from Jantunen : ) In (), there are altogether four verbal-centered clauses, in lines , , and . However, only one of these clauses, the very first intransitive clause in line , is realized as syntactically full, the other three being incomplete in terms of their S or A arguments. Syntactically, the missing elements have different roles (i.e., they are either S or A) but thematically they all represent the same thing: the discourse theme ‘me’, overtly expressed only in the first clause with the nominal pointing sign ME. This type of thematically based argument ellipsis is observed widely in FinSL narratives and is not found in non-discourse-oriented languages. McShane’s third defining criterion of discourse orientation—that ellipsis is not merely an optional grammatical process in discourse-oriented languages but an important means of supporting the discourse structure—can also be exemplified for FinSL with the help of example (). An important dimension in the structuring of discourse is cohesion. When talking about increasing the cohesion of texts or discourse, people often expect to see overt markers such as particular morphemes, a certain word order, or a specific choice of words. However, as Hendriks and Spenader () point out, not saying something aloud often increases cohesion to at least the same extent: for example, by leaving the S argument of the second clause unexpressed in line  of example () the second clause becomes automatically connected to the first clause, because the overtly mentioned theme in the first clause has to be used to retrieve the referent of the missing element. When the non-expression of the theme is repeated over a longer stretch of discourse (in lines  and ), the consequent effect—combined with the coreferential use of space (see section ..)—is that this stretch of discourse also becomes thematically a cohesive whole. This is precisely how ellipsis works in FinSL.

. E      FSL

.................................................................................................................................. This section focuses on elliptical phenomena that involve syntactic constituents headed by nominal signs, that is, nominal phrases. The ellipsis of core arguments, discussed on a general level in the previous section, has already provided one example of such a





phenomenon: in the prototypical case, core arguments are realized as nominal phrases, yet they can also be elided. Because of the frequency and importance of argument ellipsis in FinSL and in other sign languages, the phenomenon will also be further discussed in this section when addressing the contexts that encourage the ellipsis of whole nominal phrases (see sections .. and ..). However, before turning to these, section .. first describes for FinSL some elliptical phenomena that target elements inside the nominal phrase.

.. Nominal phrase-internal ellipsis The ellipsis of elements inside nominal phrases—corresponding to the phenomena of nominal ellipses discussed in Chapter  of this book—has not been thoroughly investigated in FinSL. However, it is known that, within nominal phrases, simple lexical nominals may be left unexpressed in complex sentences if they are preceded by a numerical or a characterizing nominal sign within the nominal phrase. Elicited examples of such instances are given in (); square brackets indicate nominal phrases and the symbol ‘Ø’ the site of ellipsis. () a. BOY BUY [TWO BOOK] ’ GIRL BUY [THREE Ø] ‘The boy bought two books, the girl bought three.’ b. BOY HAVE [RED FLOWER] ’ GIRL HAVE [BLUE Ø] ‘The boy has a red flower, the girl has a blue (one).’ In (), the nominal phrases functioning as P arguments in the second clause are manifested as syntactically incomplete, the referents of the missing (a) numerical and (b) characterizing signs being retrieved on the basis of the information expressed in the corresponding nominal phrases of the first clause. In addition to these elided nominal heads, the verbal signs BUY and HAVE in the second clauses—as well as the sign HAVE in the first clause of (b)—are optional, that is, they could also be omitted. Many languages allow the head of the NP to be elided in the second clause if the head is preceded by a possessive pronoun. A corresponding process is possible also in FinSL, as demonstrated by the complex identificational sentence in () with its several pointing signs referencing object and person. ()

There are two glasses on the table: THAT:left [MY-OWN GLASS] / THAT:right [YOUR-OWN Ø] ‘That (on the left) is my glass and that (on the right) is yours.’

In (), the sign THAT is an index finger pointing which, in the first identificational clause, is directed at the glass located on the signer’s left and in the second at the glass located on the right. The signs MY-OWN and YOUR-OWN also involve gestural pointing: they are produced with an open palm moving in the first case toward the signer’s chest and in the second toward a spot in front of the signer. Again, it is worth noting that the head nominal GLASS can also easily be left out of the predicating nominal phrase of the first clause: all the meaning can easily be constructed on the basis of the directionality of the remaining signs and the situational context.

  



.. Ellipsis of nominal phrases in coordinated two-clause structures A syntactic context that strongly attracts the ellipsis of full NPs in the world’s languages is the two-clause structure of conjunctive coordination. As seen in the previous examples (() and ()), this context also allows the ellipsis of nominal phrases in FinSL. In addition to internal elements in the nominal phrase, the target of the ellipsis in this domain in FinSL is in many cases the whole S/A argument of the second clause, which makes the process similar to the one discussed in Chapter  under the name of conjunction reduction. The phenomenon is exemplified from this perspective in (); transcripts with lines above the glosses indicate various non-manual behaviors and their scope. () a. The signer is telling a story about a snowman: constructed action SNOWMAN GO-IN ’ Ø LOOK-AROUND ’ . . . ‘The snowman goes in and looks around.’ (Jantunen ) b. MAN GO-IN / Ø SEE WOMAN ‘The man goes in and sees the woman.’ (Jantunen : ) In the examples in (), the valency of the verbal predicates LOOK-AROUND and SEE suggests that the S/A argument should be present also in the second clause of the twoclause structures. However, in the examples, no such element is expressed, which together with other similar instances has led Jantunen () to propose, following McShane (), that in FinSL the symmetric structure of conjunctive two-clause coordination makes it very easy for the second clause to manifest itself as syntactically incomplete: the overt expression of the S/A argument in the first clause and the semantic–structural parallelism between the coordinated clauses provide strong cues to recover the referential identity of the elided element. In (a), the constructed action through which the signer mimically with his bodily behavior shows the snowman looking around the premises, layered with the Type  verbal predicate LOOK-AROUND, further reduces the need to lexically express the ‘looker’. Superficially, the examples in () seem to suggest that the referents of the missing elements are recovered on the basis of a pivot constraint similar to that which governs the occurrence of ellipsis in corresponding English sentences (i.e. the elided NP and the NP with which it is coreferential both have to be in S/A function). A pivot constraint controls the way two clauses combine to form a complex sentence; they are language-specific and involve constraints in terms of co-occurrence and omission on the syntactic functions (S/A and P) of an NP that is common to the two clauses (Dixon ). However, example () demonstrates that FinSL behaves differently from English: ()

MAN GO-IN / WOMAN LOOK-AT Ø ‘The man goes in and the woman looks at (the man/him).’ (Jantunen : )





In (), the elided nominal phrase in the second clause has the syntactic function of P, yet it is referentially identical with the S/A nominal phrase referent (‘man’) in the first clause. Consequently, because languages have been shown to prefer either the S/A or S/P paradigm (Dixon ), it cannot be argued that the identity of the referents of the missing elements in (b) and () can be recovered on the basis of any pivot constraint. Instead, drawing on the fact that FinSL is a discourse-oriented language (section .), it is suggested that, in the retrieval of the referential identity of the missing nominal phrases in second coordinands, FinSL simply employs the discourse theme (here, ‘man’). The two-clause coordinated structures—and, in (), also the inherent directionality of the Type  verbal LOOK-AT— are additional means that make it possible to rely very effectively on this recovery method based on the discourse theme.

.. Other contexts licensing the ellipsis of nominal phrases in FinSL In addition to two-clause coordinated structures, the preceding discussion has made it clear that there are also other contexts in FinSL that allow a relatively systematic elision of nominal phrases functioning as clausal core arguments. These include the contexts of constructed action and Type  verbals involved in examples (a) and () above, as well as the topic–comment structure (see ..). Let us now take a look at these, considering first the topic–comment (TOP-COM) structure, of which two examples are given in (). () a. When explaining her daily use of languages: TOP[AT-HOME] / TOP[UNIVERSITY AREA] / COM[Ø SIGN] ‘As for my home, as for the university campus, (I) sign.’ (Jantunen : ) b.

TOP[KEY

THERE:down] / COM[WOMAN Ø TAKE:down] ‘As for the key, the woman picks (it/the key) up.’ (Jantunen : )

The sentence in (a) is a two-topic structure in which the topics function to set a locative framework for the following short comment clause. In the example, the comment clause consists only of the Type  verbal predicate SIGN. The S argument of the clause has been elided and its referent (the first person ‘I’) is inferred on the basis of the discourse context. The sentence in (b) represents a different kind of topic–comment structure, with only one topic and an elided nominal phrase in the role of the P argument. The important thing in (b) is that, in the frameworks of other researchers (e.g., Aarons ; Liddell ; Sandler and Lillo-Martin ), (b) would perhaps not be considered as involving any kind of ellipsis but as the movement of an element (‘key’) from the clause-internal object position to the specific clause-external topic position. In the grammatical study of FinSL, however, the ellipsis analysis has been favored (Jantunen ). One type of justification for this analysis comes from the work done on information structure and semantics: many researchers in this field (e.g., Lambrecht ; see also Chafe ) have claimed that topics in general are not under the influence of the valency of the comment clause predicates, as

  



assumed in the movement analysis. Another, and more concrete, argument for the ellipsis reading of (b) is the fact that it is definitely possible (albeit marked) to produce the example so that the comment clause includes, for instance, a pointing sign (P argument) indicating the same referent as that identified by the clause-external topic (see the symbol ‘Ø’ in the example). If the topic were analyzed as being moved, it would still be considered to have a relationship with the predicate, and no anaphoric pointing could be used in the comment clause. As was mentioned in section ., Type  verbals (such as TAKE in (b)) include a gestural component which allows them to be meaningfully directed and moved in space between referents of the situations they encode. As the referents are often—but not obligatorily (de Beuzeville et al. )—indicated by the forms of the verbals themselves, their overt expression within the clause with lexical elements is unnecessary (although the referents are also frequently expressed lexically in FinSL). The locations of the referents at which the verbals are directed may be physical or they may be grammatically established. In (), displayed also in Figure ., the Type  verbal TEACH (for a context-free variant of this sign, see Figure .) is directed between two abstractly created locations referring to ‘the tall (and presumably older) you’ and to ‘the short (presumably younger) me’. The clause formed around the verbal does not contain an overtly expressed A or P but their referents are retrieved from the context with the knowledge that the verbal TEACH always points toward the patient or undergoer-like referent.

 . FinSL Type  verbal TEACH as used in the elliptical clause (). Note that the signer also employs constructed action to show the imaginary locations of the referents Source: Suomalaisen viittomakielen perussanakirja ()

 ()

 During a story about teaching: constructed action Ø you:TEACH:me Ø ‘(You who are tall/older) teach (me who is short/younger).’ (Jantunen : )

The use of Type  verbals such as TEACH in (), together with the use of Type  verbals (see section .), has been found to be strongly linked to the phenomenon of constructed action (see de Beuzeville et al. ), which is yet another context favoring argument ellipsis in FinSL (see also Wulf et al.  for American Sign Language). For example, in (), the argument structure of the transitive Type  verbal LOOK-AT (see also example ()) is realized incompletely because the nominal P argument—supposed to convey the meaning ‘text (on the computer screen)’—is elided. However, from the point of view of semantic interpretation this is not a problem because the way the signer mimics the person working with the computer (also associated with the verbal predicate WRITE-KEYBOARD of the previous clause) provides all the necessary information for the construction of the correct meaning. In the case of LOOK-AT, the constructed action is manifested through the signer’s body posture, eye gaze, and the downward movement of the hand and fingers, which represent the eyes. Overall, the constructed action explicates the meaning by creating a pantomimic image of the signer actually reading—literally with her eyes—the text from the computer screen. ()

constructed action constructed action COMPUTER / ME WRITE-KEYBOARD / ME LOOK-AT:up-down Ø ‘I was typing with the computer. When I was reading (the text on the screen) . . . ’ (Jantunen : –)

As has been shown in the examples above, the various contexts favoring ellipsis in FinSL are often realized at the same time. This makes the overall effect of elliptical phenomena sometimes appear very strong as it is possible for the discourse to proceed with only a minimal amount of lexically expressed information (see, for example, (), which is a full clause consisting of only one lexical sign). In particular, it can be argued that the gestural and mimical aspects of signs (e.g., signs with gestural components) and signing (e.g., constructed action) contribute strongly to the occurrence of elliptical phenomena in FinSL (see Jantunen ). We will return to this issue with more evidence in section ..

. E    

.................................................................................................................................. Let us now turn to the issue of elliptical phenomena involving verbal predicates in FinSL. First, it should be noted that as yet little is known about these types of issues in sign languages: several instances of ellipsis involving a verbal predicate (e.g., sluicing, gapping, and stripping, often also treated as instances of clausal ellipsis) have been

  



described for FinSL by Jantunen (), but it is only recently that investigation has begun in sign languages of the elliptical phenomenon resembling the VP-ellipsis of spoken languages (for example, Cecchetto et al. ). Accordingly, a lot of additional work is still called for in the whole field of elliptical phenomena in sign languages related to the verbal predicate. The following two sections present on a general level what the traditional elision phenomena of, on the one hand, VP-ellipsis (..) and, on the other, sluicing, gapping, and stripping (..) could be like in FinSL.

.. On the elliptical phenomenon resembling VP-ellipsis In spoken languages, VP-ellipsis—also discussed under the name of predicate ellipsis in Chapter —is a process that typically elides the whole verb phrase constituent, that is, the verb together with its possible complements and adjuncts (McShane ). VP-ellipsis is well attested in English but it is not always clear if it exists as a process in many other languages. Cecchetto et al. ()—working in the formal framework—have argued that Italian Sign Language exhibits an elliptical process that shares common properties with VP-ellipsis in languages like English. They note, like McShane () in the functional framework, that the process is not easily distinguishable from the elliptical process of stripping, for example, but argue that the presence of what they call a modal or an auxiliary-like element together with evidence from, for instance, embedded wh-structures, support the VP-ellipsis reading. One example that Cecchetto et al. provide of this VP-ellipsis-like phenomenon, demonstrating the presence of the modal element MUST, is given in () (for a discussion, see Cecchetto et al. ). ()

GIANNI BOOK BUY MUST. MARIA MUST SAME ‘Gianni must buy a book and also Maria must (buy a book).’ (Cecchetto et al. : ; original notation)

Corresponding examples of what Cecchetto et al. () analyze for Italian Sign Language can also be constructed for FinSL. One that directly resembles () is given in (), which shows that the signs BUY and APPLE that are present in the first clause—perhaps constituting a unit that may be called a verb phrase—are not expressed in the second clause. ()

Telling a story about a boy and a girl who go to the shop together: BOY WANT BUY APPLE ’ GIRL WANT Ø ALSO ‘The boy wants to buy an apple and the girl wants (to buy an apple) also.’

If the phenomenon of verb phrase ellipsis does indeed exist in FinSL, sentence () is most likely an example of it. However, it must be noted that not all FinSL signers consider sentence () to be an entirely natural expression. The hesitation of these signers relates to the presence of the verbal WANT in the second clause: to express () more naturally, the sign WANT would be left out of the second clause. This, on the other hand, would change the analysis of () from verb phrase ellipsis to something more like stripping (Jantunen





). Alternatively, the order of WANT and ALSO could be reversed. However, while some signers say this configuration is possible, others indicate that this kind of ellipsis is perhaps blocked completely, or that at least it feels unnatural. Another type of evidence against analyzing () as verb phrase ellipsis draws on the diagnostics of strict vs sloppy readings. Claims have been made (e.g., Quer and Rossello ) that in VP-ellipsis there is a sloppy reading available for the elliptical clause. More work is needed on this phenomenon in FinSL.

.. Gapping, stripping, and sluicing in FinSL According to Jantunen (), FinSL exhibits the traditional elision phenomena of gapping, stripping, and sluicing. Like the process resembling VP-ellipsis described above (..), the processes of gapping, stripping, and sluicing—often considered to be traditional cases of clausal ellipsis—have not yet been thoroughly investigated in FinSL. Consequently, they too will be only generally outlined in the following. Let us first consider the examples in (). They include gapping which, in the spoken language literature, has been defined as an elliptical process that renders unexpressed the verb and, optionally, other elements of the verb phrase in the latter clause(s) of a clause coordinate or comparative structure (McShane : ). ()

a. GIRL HAS-GOT TWO-PIECES / BOY Ø ONE-PIECE
 ‘The girl has two and the boy (has) one.’ (Jantunen : ) b. The boy and girl go to the shop together: BOY BUY APPLE ’ GIRL Ø ORANGE ‘The boy bought an apple and the girl (bought) an orange.’ (Jantunen : )

The sentences in () are two-clause coordinate structures in which the verbal predicates of the second clauses have been omitted. The meaning of the second clause is interpreted, in both cases, on the basis of the overtly expressed verbal predicate in the first clause—that is, on the basis of the structural and semantic parallelism between the two clauses. This parallelism is quite strong and reliable: for example, in (a), if the second clause were to stand on its own as an isolated clause, the interpretation of its meaning would be ‘one (piece of a) boy’. However, in the actual elliptical clause in (a), there is no danger of such ambiguity. Like gapping, so too is stripping based on the structural parallelism between the directly connected antecedent clause and the ellipsis clause. In stripping, as defined in the spoken language literature, all but one main constituent in the ellipsis clause is stripped away under identity with the antecedent clause; some peripheral constituents, like adverbial elements or negators, can be overt in the stripped clause as well (McShane : ). Cross-linguistically, stripping bears a strong resemblance to VP-ellipsis (see ..); of the two, however, stripping is normally considered the more common.

  



The sentences in () are examples of stripping in FinSL. ()

a. MY FRIENDS VISIT-ME HOSPITAL ’ RELATIVES ALSO ‘My friends came to visit me at the hospital, and (my) relatives (came to visit me at the hospital) too.’ b.

neg BOY WIN COMPETITION ’ FATHER ‘The boy won the competition, the father (did) not (win the competition).’

In (a), the second clause lacks the verbal predicate and the subsequent locative P argument present in the first clause. However, the clause contains as a peripheral element the sign ALSO, belonging to a minor sign class of particles in FinSL, to mark stripping. In (b), the second clause includes only the A argument, followed by the non-manual negative operator (i.e., a headshake, although a reduced one). Note that the non-manual operator is produced after the A argument as an independent element, perhaps providing structural evidence that the ellipsis of manual clausal material has indeed occurred after the A argument; in FinSL, non-manual operators are layered upon manual activity (see ..). Sluicing is an elliptical process that affects interrogative clauses. In sluicing, interrogative clauses are elided in such a way that only their wh-words or phrases are left overt. Sluicing is a slightly different phenomenon from gapping or stripping: in addition to being licensed by the structural parallelism between the antecedent and ellipsis clause, it is also licensed by the preceding wh-word (McShane : ). According to Jantunen (), the sentence in () represents an example of sluicing in FinSL (square brackets indicate the sluiced clause). ()

wh-question ME KNOW [YOU BUY+ALREADY APPLE] ’ BUT WHY [Ø] ‘I know you bought an apple but (I don’t know) why (you bought it)?’ (Jantunen : )

The example in () is a two-sentence complex structure involving both clausal subordination and coordination (see ..). Of the two coordinated sentences, the second one is manifested elliptically in that the content of the complement clause in the first sentence, expected to function as an interrogative clause in the second sentence, is elided. To mark the ellipsis, the second clause includes the question sign WHY accompanied by the non-manual wh-question operator. Note that the operator starts already during the conjunction sign BUT and ends with a hold that relaxes only after the interrogative sign WHY. The hold of the non-manual operator after the interrogative sign may function here as a similar formal marker of ellipsis to that discussed with respect to example ().





. O      FSL

.................................................................................................................................. The previous sections have discussed several major types of elliptical phenomena involving both nominally and verbally headed units in FinSL. This section addresses two more types of elliptical phenomena in FinSL: answer ellipsis (..) and the ellipsis of conjunctions (..).

.. Answer ellipsis Work on FinSL (Jantunen ; see also Caponigro and Davidson  for American Sign Language) has shown that question and answer pairs form yet another discourse– grammatical context that clearly favors ellipsis in FinSL: questions permit reduced answers (cf. fragment answers) because all the necessary information has typically already been included in the question part (Jantunen ; McShane ). A classic example of the phenomenon is the signed question, Are you deaf?, which may be answered only with a head nod without manual signs (cf. () and ()). Another example, extracted from an actual dialogue, is given in (). In (), the reduced answer contains neither A nor P—the latter of which may be analyzed as a full subordinate clausal complement—because their meaning can be directly inferred from the preceding question. () Question: yes/no AT-ALL YOU THINK / HOW MANY DIFFERENT LANGUAGES wh DURING-ONE-DAY YOU USE ‘Have you thought how many different languages you use during a day?’ Answer: YES / Ø OFTEN THINK Ø ‘Yes, (I have) often thought (about it/how many languages I use daily).’ (Jantunen : ) In addition to the very common argument ellipsis present in (), the question and answer context in FinSL creates a specific discourse framework in which verbal predicates, for example, can also be elided relatively freely. Consider the example in (), which is a transitive clause immediately following the answer given in (): ()

The signer is continuing to answer the question of what languages she uses daily: ME Ø SIGN-LANGUAGE / . . . ‘I use sign language.’ (Jantunen : )

  



In (), the overt structure of the transitive clause is similar to the nominal phrase plus nominal phrase structure found in FinSL nominal clauses (see ..). However, as the meaning of the clause in () obviously cannot be ‘I am (in) sign language’, the clause is interpreted as being incomplete in terms of its verbal predicate, the sign USE. The meaning of the missing verbal in this case is inferred on the basis of the preceding context, that is, the question asked by the other signer: in fact, in this stretch of conversation, the verbal USE occurs only in this discourse-initial question and is consistently omitted—similarly to nominal discourse themes—in the rest of the discourse. It should be noted that superficially the process resembles gapping (see ..). However, unlike in gapping, there is no real structural or semantic parallelism with any directly connected sentence in the example that could be referred to in the retrieval of the missing predicate.

.. Ellipsis of conjunctions In spoken languages, the ellipsis of conjunctions is a process that elides elements that function as conjunctions (McShane ). In general, languages differ in the degree to which they allow this type of ellipsis: some languages (e.g., English) prefer to express all conjunctions overtly whereas others (e.g., Russian) permit conjunctions to be omitted relatively freely. Naturally, there are also languages in which conjunctions do not exist, either at all, or for every structure (see Haspelmath b for an overview from the perspective of coordination). In such languages it does not make sense to talk about the ellipsis of conjunctions. As mentioned in section .. and as we have seen in the examples in this chapter, FinSL is a language that typically does not use conjunction signs when coordinating clauses conjunctively (cf. ‘and’ coordination): conjunctions may be present in these structures but, given their low frequency of occurrence and native intuitions, among other things (Jantunen ), such cases are best analyzed as instances of adding a conjunction sign to the structure, rather than treating the other cases, in which they are not used, as instances of leaving the conjunction out. However, in the typical adversative (‘but’) and disjunctive (‘or’) coordination, the conjunction signs (BUT and OR, respectively) are used. An example of this is shown in (), which demonstrates a FinSL disjunctive two-clause structure with the conjunction sign OR in-between the clauses. ()

. . . ME SELF GROW-UP ALREADY ’ OR CULTURE CLIMATE CHANGE . . . ‘(I don’t know if it is) me who has grown up or if the cultural atmosphere has changed . . . ’ (Jantunen )

In (), the disjunctive linkage is typically marked not only by the conjunction sign OR but also by an emphatic change in the head or upper torso position between the two clauses. Interestingly, if the non-manual activity manifesting the prosodic break between the clauses is made very explicit—that is, it is almost exaggerated—the disjunctive linkage can also be expressed without the sign OR (Jantunen ; for American Sign Language, see Davidson ). As the syndetic (with OR) form of the linkage is considered to be the





neutral one in FinSL, the asyndetic instance (without OR) can be analyzed as being a marked result of the ellipsis of the conjunction element. A similar elliptical phenomenon with a non-manual intensification process can also target adversative coordination employing the conjunction sign BUT (Jantunen ).

. O            FSL

.................................................................................................................................. As we have seen in this chapter (section .), elliptical phenomena are widely permitted in FinSL because of the language’s discourse-oriented nature, which emphasizes the coherence of thematic structure over clausal structure. Another general factor that could be linked to the frequent occurrence of elliptical phenomena in FinSL is gesture and mime: because the channel of gesture and mime is naturally available in FinSL (via its visual– gestural modality), the construction of meaning does not have to rely solely on traditional grammatical structures such as full phrases or clauses. This, again, means that these units can manifest themselves relatively freely as incomplete (Jantunen , ). That gesture and mime actively participate in elliptical phenomena in FinSL has been shown in many of the examples in this chapter as well as in the discussion of contexts that especially favor ellipsis in FinSL (Type  verbals and constructed action in particular). Similar observations about the role of gesture and mime in ellipsis have also been made in the research into other sign languages. For example, in their work on the role of constructed action in Australian Sign Language clauses, Ferrara and Johnston () summarized its effect on grammatical structure as being identical to that found in FinSL (CA=constructed action): CA has the potential to influence surrounding grammatical structure. First, signers may use CA instead of manual lexical signs to effect a particular construal of an event and to profile participants and processes. [ . . . ] Secondly, and more generally, [ . . . the] meanings prompted by a CA may cause a signer to [ . . . ] leave an argument or process unexpressed by manual lexical signs, because it is already ‘active’ in the discourse. (Ferrara and Johnston : )

To demonstrate how the constructed action affects full clausal structure from the perspective discussed by Ferrara and Johnston (), let us consider the FinSL example in (), illustrated with images captured from the video in Figure . (cf. example ()). () In telling a story about a boy, dog, and a frog: constructed action BOY LOOK-AT Ø ’ Ø LET-IT-BE ‘The boy watched (the dog barking around) and (thought:) “umh, let it be.” ’ (Jantunen ) The sentence in () contains two conjunctively coordinated main clauses. The first clause is exceptional in that the whole subordinate complement clause in P function has

  



 . Video frames showing the production of the sentence ()

been omitted (‘the dog barking around’) because the activity of the dog that the boy is observing is made clear by the discourse context and enacting. The second clause, on the other hand, expresses overtly only the gestural behavior of the signer that, via constructed action, signals the direct speech of the boy (‘umh, let it be’). This simple gesture may be analyzed as the P argument of a matrix clause that has been omitted from the sentence because the boy’s thinking is already made visible by the signer enacting the boy. The remarkable thing in () is that, regardless of its highly reduced syntactic structure, the meaning of the whole complex sentence can still be understood without difficulty. Linguistic structure is frequently reduced and subsumed under gesture and mimic behavior in sign languages. However, as an important final remark, it must be noted that gesture and mime are not an exclusive property of sign languages but occur also with speech (e.g., Kendon ; McNeil ); this means that they may also play an important role in the elliptical phenomena found in the grammar of spoken language. Ladewig (), for example, has discussed how gestural enactment may fill syntactic gaps of either nouns or verbs in German. One of her examples involves a story describing how a girl is pushed through a window by relatives. The story proceeds in spoken German until the end of the second clause of a two-clause coordinated structure, und wir hinten (‘and we from behind’), at the end of which a two-handed gestural action describing the pushing movement replaces the speech (Ladewig : ). Concerning this replacement, Ladewig argues that the pushing gesture fills a syntactic slot that is normally occupied with a finite verb following the subject wir (‘we’) in the construction. The process described by Ladewig () is intuitively very familiar and is also known to occur in spoken language more generally. Incidentally, the process that originally causes the syntactic gap in Ladewig’s example quite closely resembles the elliptical process of gapping. In traditional accounts of elliptical phenomena in spoken language, the role of gesture and mime has not been the focus of attention. However, the evidence presented here suggests that the gestural dimension should perhaps be taken more widely into account in this field, too.

. C

.................................................................................................................................. This chapter has presented an exposition of ellipsis in FinSL. Within the limits of the grammatical work carried out so far on FinSL, the chapter has discussed both nominal and





verbal ellipsis and has touched upon several other types of elliptical phenomena. The chapter has also discussed the relationship between ellipsis, discourse orientation, and gesture in FinSL. As a contribution to the field, the chapter has suggested that gesture and mime, used extensively in FinSL, may also have a larger role than is normally assumed in many of the elliptical phenomena found in spoken languages.

A The author wishes to thank Eleanor Underwood for checking the English of this chapter. The financial support of the Academy of Finland under grants  and  is gratefully acknowledged.

  ......................................................................................................................

 ......................................................................................................................

 

E phenomena have received, by far, much less attention in French than in English or even Spanish within formal frameworks, in spite of the initial attempt by Zribi-Hertz () to provide a comprehensive analysis of them in terms of null proforms. This is all the more regrettable as French displays constructions that resemble the main types of ellipses reported in the literature but with sometimes diverging properties, and also has pro-forms in contexts where other languages have ellipsis. French thus appears as a large fallow field for those wishing to investigate cross-linguistic accounts of ellipsis and its much debated relationships with anaphora. Today, one of the central issues, in particular for predicate and clausal ellipses, is whether ellipses, as opposed to pronouns, involve abstract syntactic structure: this question will be the central theme of this chapter. I will first examine whether the existence of ellipses following modal verbs challenges the consensus view that French lacks VP-ellipsis, and show that it does not (section .). I will then turn to constructions that linguists have argued display clausal ellipsis: in section ., I will examine sluicing—and to a lesser degree, fragment answers. Section . discusses coordinate ellipses, and I will show that French calls for a nonuniform analysis of clausal ellipses. In the last two sections, I will very briefly point to a few questions raised by French comparative ellipsis and NP-ellipsis.

. P     F

.................................................................................................................................. English-type predicate ellipses are ruled out in present-day French (for Old French, see Miller a), in particular after auxiliaries. In VP-ellipsis, an auxiliary can be followed by a gap: () shows that French auxiliaries, unlike their English counterparts in the translation, cannot. Instead, French uses substitutive pro-forms, cf. (): roughly, agentive vPs are anaphorized by le faire ‘do it’, as in (a); adjectival and passive predicates by le ‘it’, as in (b–c). Clausal complements can be represented by various pronouns (le ‘it’, en ‘of it’, y ‘to it’, ça ‘that’) according to the verb and the context (cf. Zribi-Hertz ; Roussarie and Amsili ).





() *Jean a acheté du vin mais Jean has bought of.the wine but ‘John has bought wine, but Luke has not.’

Luc Luc

n’a 1=has

pas. not

VPE

() a. Jean a acheté du vin mais Luc ne l’a pas fait. VP-anaphor Jean has bought of.the wine but Luc =it=has not done ‘Jean bought wine but Luc didn’t do it.’ b. Jean est content mais Luc Jean is happy but Luke ‘John is happy but Luke is not.’

ne l’est pas / *n’est pas. =it=is not / =is not

c. Le garage sera démoli dès que la maison le sera. The garage be. destroyed as. soon.as the house it=be. ‘The garage will be destroyed as soon as the house will (be).’ In English pseudogapping (PG), some contrastive remnant is present next to the auxiliary, cf. (b). In French, PG also has an anaphoric equivalent, as shown in (a). As (c) shows, it allows adjuncts, but not arguments, to be stranded. () a. Jean offrira du vin à Marie lundi et Luc le fera Jean offer. of.the.wine to Mary Monday and Luc it=do. demain. PG-anaphor tomorrow ‘Jean will offer wine to Mary on Monday and Luc will do it tomorrow.’ b. Jean will offer wine to Mary on Monday and Luc will tomorrow. c. Jean a offert du vin à Marie et Luc Jean has offered of.the.wine to Mary and Luc *du whisky/ *à Bill. of.the.whisky/ to Bill ‘Jean offered wine to Mary and Luc did whisky/to Bill.’

Pseudogapping

l’a fait it=has done

French then seems more suited for the study of competition between different pro-forms than to the investigation of predicate ellipsis. However, there are two contexts where a verbal form can be followed by a gap, as in English: after some lexical verbs, as in (), and after (nonepistemic) modal semi-auxiliaries such as pouvoir ‘can’, devoir ‘must’, and, as shown by Authier (), vouloir ‘want’, falloir ‘have to’, and avoir le droit ‘have the right’; cf. (). () a. Jean voulait participer, mais il n’a John wanted participate, but he==has ‘John wanted to participate, but he didn’t dare.’

pas not

osé. dared

b. John wanted to participate, but he didn’t dare. 1

In all examples, the preverbal part of the discontinuous negation, ne, which does not carry semantic negation per se, is glossed as  (scope marker), after Kayne (). Other abbreviations are  (future),  (perfect),  (past perfect),  (present),  (feminine),  (masculine),  (singular),  (plural). Cliticization is indicated by =.





() a. Jean voulait participer, mais il ne peut pas/ il n’a pas le droit. John wanted participate, but he==can not/ he==has not the right ‘John wanted to participate, but he can’t/he is not allowed to.’ b. John wanted to participate, but he can’t. In English, it has been argued that sentences like (b) and (b) involve different structures, a null pro-form devoid of inner structure in (b), known as Null Complement Anaphora (NCA), and VP-ellipsis in (b), where a fully articulated syntactic structure is generally considered to be present but unpronounced.2 At first sight, (a) resembles English NCA, while (b), named French Modal Ellipsis (FME) by Busquets and Denis (), could resemble verb phrase ellipsis (VPE) or Null Complement Anaphora (NCA), or another structure found in Dutch named Modal Complement Ellipsis (MCE) (see Aelbrecht ; Aelbrecht and Harwood, this volume). I will argue that FME actually differs from all of them. One crucial test to diagnose the presence/absence of inner structure is extraction: if some item can be extracted from a position that would lie inside the gap, it seems reasonable to consider that this gap is syntactically structured though unpronounced. If nothing can be extracted from it, however, the gap can be seen as an opaque null pronoun. On this basis, Authier () concludes that French NCA cases such as (a) involve an opaque null complement. French modal ellipses, on the other hand, differ from NCA in that they pass all the usual tests for diagnosing the presence of internal structure (cf. Authier ; Busquets and Denis ; Dagnac ). In particular, they pass all extraction tests, including, unlike MCE (cf. (c), from Aelbrecht : ), A’-movement tests:3 they can enter Antecedent-Contained Deletion (ACD) constructions, where an elliptical restrictive relative modifies a quantified antecedent and then appears to be contained inside its antecedent (cf. ()), but also free relatives and wh-questions (cf. ()) and A-movement extraction (cf. ()). ()

Léa lit tous les livres qu’elle peut . Lea reads all the books that=she can read 4 ‘Lea reads all the books that she can.’

() a. Il embrasse [WH quii] il peut . he kisses [WH who(ever)i] he can ‘He kisses who(ever) he can.’ b. Je sais quels livres LEA peut lire et je sais aussi quels livresi BEN I know which books LEA can read and I know also which booksi BEN ne peut pas . FME =can not . ‘I know which books LEA can read and I also know which books BEN can’t.’ 2

Zribi-Hertz () considers all ellipses as empty categories. On the basis of their distribution, she considers French NCA and English VPE as allowing the same type of pro-forms, and leaves the question open for English NCA. 3 Authier () further shows that they pass the Missing Antecedent test and the Quantifier Scope test introduced by Hankamer and Sag (). 4 I don’t give the English VPE equivalents to the French examples independently when the translation provides them.



 c. ?*Ik weet niet wie Kaat WOU uitnodigen, maar ik weet wel wie ze MOEST.MCE I know not who Kaat wanted invite but I know  who she must.PST ‘I don’t know who Kaat WANTED to invite, but I do know who she HAD to.’ VPE

()

La fenêtre peut être réparée mais la portei ne peut pas The window can be repaired but the door  can not < ti être réparée ti>. < be repaired> ‘The window can be repaired but the door cannot .’

FME thus patterns with VPE in this respect and qualifies as ellipsis via deletion. FME shares another property with VPE: what can be elided matches what can be topicalized (Johnson b). As Authier () shows, the modal verbs found in FME also allow their complement to topicalize, cf. (), unlike French auxiliaries, NCA-taking verbs, and epistemic modals, cf. (): ()

J’aimerais bien faire des études de médecine, mais disséquer I would-like well to-do some studies of medicine but dissect les cadavres, je peux pas. the corpses I can not ‘I’d like to undertake medical studies, but dissect corpses, I just can’t.’

() a. *Anne voulait manger des moules et [VP mangé des moules]i, elle a ti. Anne wanted to-eat some mussels and eaten some mussels she has ‘Anne wanted to eat mussels, and eat mussels, she did.’ b. *Anne Anne

ne croit =believes

pas not

avoir to-have

fermé shut

les the

rideaux, drapes

mais but

avoir to-have

fermé les fenêtres, elle croit. shut the windows she believes ‘Anne doesn’t think that she closed the drapes, but the windows, she thinks she did.’ c. *[ Arriver d’un moment à l’autre ]i, la police peut ti; alors accélère. to.arrive of a moment to the other the police may so speed-up ‘The police may arrive at any moment, so hurry up.’ (epistemic) Authier () then argues that both English-type VPE (cf. Johnson b) and FME are licensed by movement, and that the possibility for a tensed or auxiliary verb to have its predicate topicalized predicts the possibility of a corresponding predicate ellipsis. What remains to be explained is why the higher copy of a topicalized predicate can be unpronounced and why topicalization (and then potential ellipsis) of the VP is allowed after English auxiliaries but not after French ones. The answer to the latter question may correlate with the size of the elided material. While Zribi-Hertz () and Busquets and Denis () consider that FME is an instance of VPE, Dagnac () argues that FME does not delete an infinitival VP but a full clause. This is





straightforward with constructions involving vouloir ‘want’, where the elided material is a tensed clause, as in ():5 ()

Je lirai tous les livresi que tu voudras . I read. all the books that you want. that I read. ‘I’ll read every book that you [will] want me to read.’

Pouvoir ‘can’ and devoir ‘must’, on the other hand, are subject-raising verbs selecting an infinitival TP (Ruwet : –).6 The hypothesis that FME elides the whole CP/TP accounts for various differences between FME and English VPE. First, in French, if the infinitival predicate has a tense, aspect, or voice auxiliary, as in (a), it cannot be stranded, nor can any constituent situated higher than vP inside the predicate, such as negation; cf. (b). ()

a. Tom peut avoir fini en juin, et Léa peut aussi (*avoir). Tom can have finished in June and Lea can also have ‘Tom can have finished in June, and Lea can have, too.’ b. *Paul peut aller à Madrid mais Léa peut ne pas. Paul can go to Madrid but Lea can not ‘Paul is allowed to go to Madrid but Lea is allowed not to.’

These differences are expected if English modal auxiliaries select an (elidable) predicate VP, while French modal verbs like pouvoir and devoir select an (elidable) clause, as sketched in (): ()

a. Modal < [TP . . . [(NegP) [T . . . [Asp [passive BE [[P [VP . . . .]]]]]]>

(French)

b. Modal (NegP) [Asp [passive BE [P ]]]

(English)

Second, Merchant (a, d) argues that clausal ellipses disallow voice mismatches between the ellipsis site and the antecedent, while smaller ellipses tolerate them: unlike 5

Dagnac’s () initial proposal, based on pouvoir and devoir, is that FME deletes the TP complement selected by the modal. ‘Clausal’ must be qualified as either TP or CP for the other modals argued by Authier () to license FME, since they can select both tensed CPs and infinitives. The nature of their infinitival complements is debated: I will assume here that they are control verbs selecting a CP, too. 6 Monoclausal analyses of pouvoir + infinitive exist. To my knowledge, however, none of them account for the full range of Ruwet’s arguments, in particular the initial observation that complements of the subject can cliticize on the infinitive, as in (i), where en ‘of it’ stands for du problème; this is impossible with monoclausal complex predicates; cf. (ii): (i) La solution peut en être publiée. the solution can of.it be published ‘The solution (of the problem) can be published.’ (ii) a. La solution [en a été publiée]. the solution of.it has been published b. *La solution a [VP en été publiée]. the solution has of.it been published c. *La solution a été [VP en publiée]. the solution has been of.it published ‘The solution (of the problem) has been published.’





VPE, FME disallows voice mismatches (cf. ()); this is expected if the latter, but not the former, elides a clause: ()

*Il faut remplacer l’ampoulei de l’escalier, mais ellei ne It needs replace the bulb of the.staircase but it  peut pas—ellei est coincée. can not it is jammed ‘Someone should replace the bulb in the staircase but it can’t be—it’s jammed.’

The topicalization/ellipsis parallelism pointed out by Authier () can be reformulated in this way: English can topicalize and then delete VPs, French can only topicalize and then delete clauses. ()

a. *Anne voulait manger des moules et [VP mangé des moules ]i, elle a ti. (cf. (a)) b. Anne wanted to eat mussels, and [VP eat mussels]i, she did i.

() a. Anne voulait manger des moules mais [TP manger des moules ]i, elle n’a pas pu ti. b. Anne wanted to eat mussels, but [VP eat mussels]i, she couldn’ti. To sum up, French does display a type of ellipsis produced by deletion after (non-epistemic) modals. However, it is not a predicate ellipsis, but a clausal ellipsis—a consequence, if Authier () is right, of the ability of French to allow clauses, but not VPs, to be topicalized. The particular properties of FME, which differ both from those of English VPE and Dutch MCE, provide comparative material for further investigations into ellipsis licensing. Whether other clausal ellipses in French also involve deletion, though, is an open question.

. S   

.................................................................................................................................. French displays sluicing, cf. ()–(), as well as sprouting, cf. (). In sluicing, a constituent question is missing, except for the wh-phrase. Sluices can be embedded (cf. (), ()), or appear as dialogue fragments (cf. ()): I will refer to them as ‘embedded sluices’ and ‘root sluices’, respectively. Sprouting is a particular instance of sluicing, in which the wh-phrase has no overt correlate in the antecedent clause. ()

Paul a reconnu quelqu’un mais j’ai oublié qui __. Embedded sluice Paul has recognized someone but I have forgotten who. ‘Paul has recognized someone, but I forgot who.’

()

A: Je dois parler avec quelqu’un. B: Avec qui __? I must talk with someone. with whom ‘I must talk with someone. With whom?’

()

Elle a rencontré Paul à Berlin mais j’ai oublié quand __. she has met Paul in Berlin but I have forgotten when ‘She met Paul in Berlin, but I forgot when.’

Root sluice

Sprouting





French also has a construction similar to dialectal Dutch spading (cf. (b), from van Craenenbroeck a: ), where the wh-word is followed by a demonstrative, cf. (a). Unlike Germanic languages, though, it lacks swiping altogether, cf. (): when a wh-PP is sluiced, the wh-word cannot be followed by its preposition, as in the English translation. () a. A: J’ai parlé à un journaliste. B: Qui I.have talked to a journalist. who ‘I talked to a journalist.’ ‘Who?’

ça? that

Spading (French)

b. A: Jef ei gisteren iemand gezien. B: Wou da ? Spading (Wambeek Dutch) Jeff has yesterday someone seen. who that ‘Jeff saw someone yesterday.’ ‘Who?’ ()

* Il a rendez-vous avec quelqu’un, mais j’ai oublié qui avec. Swiping He has appointment with someone, but I have forgotten who with ‘He has an appointment with someone, but I forgot who with.’

French sluicing has been poorly studied so far, and seems to share most of the properties of its English counterpart (cf. Chapter ). Three cases, however, need closer attention: bare sluices (), sluices involving the direct object quoi (), and degree comment sluices (). In standard French, they can be found both in spontaneous sloppy conversation and in careful writing, and can constitute the main utterance, as in the (a) examples, or be embedded, as in the (b) ones. () a. Je dois parler à quelqu’un, mais qui? I must talk to someone but who ‘I must speak to someone, but who?’

Bare sluice

b. Je dois parler à un étudiant, mais je ne sais pas quel étudiant/lequel. I must talk to a student, but I  know not which student/the.which ‘I must talk to a student, but I don’t know which (one).’ () a. Il faut faire quelque chose. Mais quoi? it must do something but what ‘We must do something, but what?’ b. Tu viens de dire quelque chose. you come from say something ‘You just said something. I forgot what.’ () a. – Est-ce que tu me trouves beau?  that you me=find handsome – Oui. Yes

Quoi sluices

J’ai I have

oublié forgotten

quoi. what

Degree comment sluice7

7 ‘Degree comment’ is analyzed in Moline (). The answers given to the sluiced questions in () and () suggest that comment is a degree phrase. Manner interpretations may exist, but this does not bear on the analysis.



 – Mais beau comment? but handsome how – Écoute, Boris, je ne sais pas, moi! Très très listen Boris I =know not me very very (Schreiber, Un silence d’environ une demi-heure, ) ‘Do you find me handsome? – Yes. – But how handsome? – Listen, Boris, I don’t know! Very, very handsome.’

beau. handsome

b. On m’a dit qu’il était grand, mais je sais pas grand comment. one me=has told that.he was tall but I know not tall how ‘I was told that he was tall, but I don’t remember how tall.’ The structures in ()–() all share one property: no grammatical full question exists where what corresponds to the remnant would be fronted, which appears to challenge analyses of sluicing relying on the deletion of a TP-constituent after wh-movement of the remnant. Bare sluices show that standard French, together with Spanish, Brazilian Portuguese (Rodrigues et al. ), Indonesian (Sato b), and Serbo-Croatian (Stjepanović ), forms an exception to the ‘preposition-stranding generalization’, henceforth PSG: 8 while it can’t strand prepositions9 in regular wh-questions (cf. ()), a prepositionless remnant can correspond to a PP-correlate in sluicing: () a. *Mais qui dois-je parler but who must I talk ‘But who should I talk to?’

(à/avec)?10 (to/with)

b. *Je ne sais pas quel étudiant je dois parler (à). I =know not which student I must talk (to). ‘I don’t know which student I must talk to.’ The remnant can correspond to all kinds of wh-phrases, simple or complex, and the correlate PP can have any syntactic function, but it must be overt; as in English (Chung ), bare sluices cannot be sprouted: () Paul est allé au cinéma, mais je ne sais pas *qui / ✓avec qui. Paul has gone to.the movies, but I =know not who / with who. ‘Paul went to the movies, but I don’t know with whom.’ 8 Since Merchant (), French bare sluices have often been presented as ungrammatical (e.g., Sato ). Merchant himself (: , n. ) suggests that judgements may vary. I find his example grammatical, if not perfect: the most natural examples involve a D-linked wh-phrase such as quel NP ‘which NP’, lequel ‘which one’, and/or an indefinite correlate. 9 Varieties of Canadian French allow (some degree of ) preposition stranding (see Poplack et al.  and references therein). We focus here on the standard variety spoken in France. 10 For ease of reading, most of the examples are illustrated by full root interrogative clauses involving subject clitic inversion. Less formal structures without inversion are equally ruled out.





Within mainstream generative grammar, two main types of analyses have been proposed for other non-preposition-stranding languages allowing for bare sluices. The first type derives these sluices from underlying clefts (Rodrigues et al. ), arguably as a last resort (van Craenenbroeck a). The second type considers them as instances of elliptical repair (cf. Sato b), for instance via the deletion of an otherwise illicit trace. Rodrigues et al. () propose that in Spanish and Brazilian Portuguese, exceptions to the PSG are derived from underlying clefts, more specifically a cleft based on a specificational copular sentence, as in (): ()

Juan ha hablado con una chica pero no sé cuál the that has talked Juan ‘John talked to a girl but I don’t know which one.’

It is unclear whether a cleft strategy can be extended to French. For some bare sluices, in particular with a non-partitive correlate (roughly, an antecedent whose referent is not part of a salient entity), no type of cleft seems directly available.11 For instance, for the bare sluice in (a), a standard interrogative cleft is of no help, since it would pied-pipe the preposition (cf. (b–c)); a short cleft as in (d) is not semantically equivalent to (a): unlike the bare sluice, it presupposes that the speaker has some prior knowledge of the individual whose identity is questioned (for instance, he has seen him, but does not know his name). As for a specificational construction, to be well-formed (cf. (e–f )), it requires additional material: if an analysis along these lines is to be pursued, it must account for how this additional material is made compatible with the identity conditions on ellipsis. () a. Paul a Paul has

probablement probably

parlé talked

b. * . . . mais qui c’est avec qui but who it is with who ‘ . . . but who is it he talked to?’

avec with il he

c. . . . mais avec qui c’est qu’ il but with who it is that he ‘but who is it he talked with?’

quelqu’un, someone a has

a has

mais but

qui? who

parlé? talked parlé? talked

d. . . . mais qui c’est ? but who it is ‘but who is it?’ 11 In French, the full equivalent to () is grammatical, while its sluiced version is ungrammatical (it requires the pronominal wh-phrase laquelle):

(i) *Jean a parlé avec une fille mais je ne sais pas quelle Jean has talked with a girl but I =know not which laquelle Jean a parlé>. the-that Jean has talked> ‘John talked to a girl but I don’t know which one.’

< est la fille avec < is the girl with



 e. * . . . mais but f. . . . mais but

qui who qui who

est is est is

le the la the

quelqu’un avec qui il a parlé? someone with who he has talked personne avec laquelle il a parlé? person with which he has talked

Contrary to Spanish (Rodrigues et al. ), French admits bare sluices with ‘else-modifiers’, which cannot occur in clefts (Merchant ): ()

A: Il est assez connu. Il a joué avec Sokhiev. he is quite famous he has played with Sokhiev ‘He’s quite famous. He played with Sokhiev.

B: Et qui d’autre? and who else? ‘And who else?’

() A: On peut le remplir avec de l’eau, de l’huile . . . B: Et quoi d’autre? one can it=fill with of the water, of the oil and what else? ‘One can fill it with water, oil . . . ’ ‘And what else?’ Bare sluices can also involve the inanimate wh-phrase quoi ‘what’ ()–(), which is also incompatible with all types of clefts in standard French, cf. (b–d): ()

Avec quoi l’as-tu rempli? with what it=have=you filled ‘What did you fill it with?’

() On pourrait le remplir avec quelque chose, mais (je ne sais pas) quoi. one could it=fill with some.thing but (I =know not) what ‘One could fill it with something, but I don’t know what.’ ()

a. * . . . (je ne sais pas) quoi c’est avec quoi on pourrait le remplir. I =know not what it is with what one could it=fill b. * . . . (je

ne sais pas) quoi est la chose avec quoi on pourrait le remplir. what is the thing with what one could it=fill

c. * . . . (je ne sais pas)

quoi c’est. what it=is

More generally, quoi, whether a complement of a preposition or a direct object as in (), cannot be fronted by itself in standard European French (cf. (b))—and neither can degree comment (cf. ()).12 In standard French tensed clauses, quoi can only be found in situ: the weak form que is required in root questions, cf. (b), and the corresponding embedded questions take the form ce que, where a demonstrative pronoun is modified by a relative clause, cf. (b).13 12 Quoi can be fronted only when it is conjoined with another wh-word (qui ou quoi ‘who or what’), modified (quoi d’autre ‘what else’), or in infinitival embedded questions (Je ne sais pas quoi faire ‘I don’t know what to do’). It regularly appears after a preposition and the whole PP can be fronted (Sur quoi est-il? ‘On what is he?’). 13 All judgements here are for standard European French. Varieties of Canadian French may allow fronted Adj+comment. Colloquial French may also allow quoi and/or Adj+comment to be clefted in root





() a. Il faut faire quelque chose mais quoi? it must do some thing but what b. . . . mais que/*quoi faut-il faire? but whatweak/whatstrong must it do? c. . . . mais il faut faire quoi? but it must do what ‘We must do something but what (must we do)?’ ()

a. Il faut it must

faire do

quelque some

chose thing

mais but

je I

ne sais =know

pas not

quoi. what

b. . . . mais je ne sais pas ce que/*que/*quoi il faut faire. but I =know not that=what/whatweak/whatstrong it must do ‘We must do something but I don’t know what we must do’ () a. A: Il est grand. B: Grand comment? he is tall tall how b. B: *Grand comment est-il? tall how is he? B: Il est grand comment? c. he is tall how? ‘He is tall.’ ‘How tall (is he)?’ If one wants to maintain a deletion analysis of these sluices, it may be more promising to consider that bare sluices, object quoi sluices, and comment sluices stem from in situ interrogatives, and investigate which repair device can account for their well-formedness.14,15 Non-deletion approaches (see e.g. Chung et al. , ; Ginzburg and Sag ), on the other hand, account straightforwardly for most of these cases. For instance, Chung et al. (, ) propose that the wh-phrases are base-generated in specCP, and followed by a null (TP) category, e, which is replaced at LF by a TP phrase marker copied from the antecedent together with its interpretation. The syntactic structure of the sluice is then similar to (a); at LF, the wh-phrase binds a variable provided by the copy of the indefinite correlate, as roughly illustrated in (b). Sprouting consists of the same mechanism, but with the additional insertion of a copy of the wh-phrase, cf. (). () a. We must do something, but I don’t know [CP what Co [TP e ]]. b. We must do somethingx, but I don’t know [CP whatx Co [ TP we must do somethingx ]].

and/or embedded clauses. The point here is that the corresponding sluices are not limited to such varieties and are completely unmarked. 14

Dependency Grammar, which allows the deletion of non-constituent units such as catenae (Osborne et al. ), could derive root bare sluices from in situ questions. However, standard French does not allow embedded in situ questions, so embedded sluices would remain unexplained. 15 Assuming that the P-stranding generalization holds for French, Sato (b) excludes PF-repair for French on the basis of the respective timing of the syntactic violation and TP-deletion—a point that needs reconsidering. French bare sluices may also involve an instance of P-drop in the sense of Stjepanović (), though whether and why P-drop would be limited to sluicing requires further investigation.





() a. They were firing, but at what was unclear. b. [CP at what C [TP they were firing

]] # c. [CP at what C [TP they were firing at what ]] First, this approach offers a way to deal with the object quoi sluices. According to Sportiche (), the choice between strong forms (here quoi) and weak or clitic forms (here que) is a PF-phenomenon. In a perspective where only bundles of features are merged in the syntactic component, and the choice of the morphological variant occurs at Spell-Out, que is unsurprisingly ruled out in contexts where it has no verb to cliticize onto, which correctly predicts its occurrence in (): () . Syntax On doit faire quelque chose, mais wh-phrase(+DO, +inanimate)? . PF On doit faire quelque chose, mais wh-phrase(+DO, +inanimate)? > strong form selection before Ø . On doit faire quelque chose, mais quoi? ‘We must do something, but what?’ The coindexing mechanism between the wh-phrase and the correlate makes no claim with respect to constituency, hence it also correctly predicts bare sluices to be possible, as in (): () Jim a parlé à un étudiant mais je ne sais pas [CP lequelx [Jim a parlé Jim has talked to a student but I =know not the.which Jim has talked à un étudiantx ]]. to a student ‘Jim spoke with some student but I know not which one’ The Adj+comment sluices, however, require some stipulation. They question the unspecified degree of the antecedent adjective. As in the measure phrase in (), from Abeillé and Godard (), the degree complements comme un jet de flèche and comment in (), are presumably part of the AP: () [AP

Haute [PP de m]], une falaise surplombait high of m a cliff overhung ‘ meters high, a cliff hung over the lake.’

le the

lac. lake

() – C’est qu’ il y a loin du latin au français. it=is that there.is far from.the Latin to.the French – [AP Loin [PP comment]]? far how – Loin comme un jet de flèche. (FT, Lanzmann, La Horde d’or, ) far like a throw of arrow ‘French is far from Latin. – How far? – Far as an arrow’s throw.’ While this AP cannot move to specCP, in French, an LF approach should postulate that it can still be base-generated in specCP in overt syntax, cf. ():

 () Il est grand, mais je ne sais he is tall but I =know ‘He is tall but I don’t know how tall.’

pas [CP not

grand tall



comment C [TP e]]. how

A construction-based approach, in which the sluices are considered as fragments directly inserted into the syntactic structure (see e.g. Ginzburg and Sag ; Sag and Nykel ), sidesteps this problem, since it would presumably treat the AP fragment grand comment as depending directly on the main verb in the syntax. While non-deletion analyses prove too powerful for languages that do comply with the PSG (see Merchant a: – for discussion),16 they come close to making the right predictions for French. An open question is whether a similar mechanism applies to other clausal ellipses. For instance, fragment answers and gapping have both been argued to involve movement of the remnants to the left periphery of the clause, followed by TPdeletion. But French fragment answers, though sharing most properties with their English counterparts, can strand a preposition, cf. (), while this is impossible in gapping, cf. (): () A: Avec qui parlais-tu? B: Paul. (*je with whom were.talking=you Paul ( I ‘Who where you talking with?’ ‘Paul.’ () Léa a parlé avec Jim et Marie Lea has talked with Jim and Mary ‘Lea talked with Jim and Mary with Bob.’

parlais was.talking

avec ti) with)

*(avec) Bob. *(with) Bob

If non-deletion approaches seem able to deal with the specific cases studied here, what remains to be explained is why and how such analyses could be constrained for PSG languages, and whether a non-deletion analysis is a last resort scenario or a regular one for French sluicing.

. C 

.................................................................................................................................. This chapter deals with coordinate constructions like the ones in ()–(). They differ in the number, nature, and syntactic function of the remnants, but all raise a problem for models in which coordination is assumed to operate on: (i) items of the same category and/or semantics; (ii) syntactic constituents.

16

According to Merchant (a, this volume), they predict, contrary to fact, that no case violation would ban bare sluices in languages complying with the PSG; besides, they allow for languages that strand prepositions in full questions but cannot strand them in sluicing—a pattern that seems as yet unattested.

 ()

 ‘Special coordinations’ a. Jim est heureux et Jim is happy and

en in

bonne good

santé. form

b. Jim aime Catherine et est aimé Jim loves Catherine and is loved ‘Jim loves Catherine and is loved by her.’ c. Anne Ann

va goes

au to.the

cinéma movie

le the

Coordination of unlikes

d’elle. of her

lundi Monday

Tensed VP coordination

et and

au to.the

théâtre theater

le mardi. Non-constituent coordination the Tuesday (NCC) ‘Ann goes to the movies on Monday and to the theater on Tuesdays.’ d. Léa adore et Lina déteste le football. Lea loves and Lina hates the soccer ‘Lea loves and Lina hates soccer.’

Right-node–raising (RNR)

() a. Jim viendra, mais pas Léa/ Léa non/ et Jim come. but not Lea/ Lea no / and ‘Jim will come but not Lea/ Lea won’t/ and Lea too.’

Léa aussi. Lea too

b. Léa est pour l’Allemagne et ses amies pour le Brésil. Lea is for the Germany and her friends for the Brazil ‘Lea is for Germany and her friends for Brazil.’

Stripping

Gapping

At first sight, in (a), the conjuncts do not have the same categorial status (AP vs PP); in (b), they differ in voice features (active VP/passive VP); in (c), (d), and (b), au théâtre le mardi ‘to the theater on Tuesdays’, Léa adore ‘Lea loves’, and ses amies pour le Brésil ‘her friends for Brazil’ do not form a constituent. In (a), while the first conjunct is a clause, the second one looks like a simple phrase. Unlike the other ‘coordinate ellipses’, stripping can sometimes be found after subordinating conjunctions in French, as shown in ().17 ()

Paul aime le Baileys alors que Marie non. Paul likes the Baileys whereas Marie no ‘Paul likes Baileys, whereas Mary doesn’t.’

17 The cases in (i)–(ii), which look like instances of gapping with a negative remnant, may be exceptions to the generalization that gapping is licensed by coordination. This, to my knowledge, has never been investigated.

(i)

Son père a été reçu à l’Élysée alors que moi jamais. his father be. received at the Elysée whereas me never ‘His father was received by the president whereas I never was.’ (, consulted on  January )

(ii)

Léa fume  cigarettes par jour tandis que Marie aucune. Lea smokes five cigarettes by day whereas Mary none ‘Lea smokes five cigarettes a day, while Mary smokes none.’





It has been argued that the English equivalents to ()–() all rely on clausal coordination plus deletion; deletion has sometimes been thought to be preceded by movement of the parts that remain overt, the remnants, to the left periphery of the elided clause (see e.g. Merchant a, Sailor and Thoms , and references therein). Under such a view, at the syntactic level, two clauses are conjoined, i.e. two constituents of the same category; cf. () for stripping: ()

[TP Jim speaks passable Korean], and [FP Leai [TP ti speaks passable Korean]], too.

In French, however, the dominant view is that none of these constructions involves abstract syntactic structure: see e.g. Zribi-Hertz () for (a,b,c); Mouret (, ) for (a,c), Abeillé et al. () for gapping and RNR, and Abeillé (, b) for stripping. One major argument is that the set of coordinators that can be found in ()–() does not match the set of coordinators found in clausal coordination but the set found in regular non-clausal coordination. We illustrate this argument for stripping here, though the same holds for the other constructions. For instance, car ‘for’ can conjoin clauses (a), not smaller constituents (b): its incompatibility with ‘elliptical’ coordinations () argues against their clausal status.18 Conversely, ainsi que ‘as well as’, a constituent coordinator, cannot conjoin tensed clauses (cf. (a)), but it is possible with elliptical structures (b), again suggesting that they cannot be clausal: () a. Paul prendra son rhum à h car il prend son Xanax à h. Paul take. his rum at  for he takes his Xanax at  ‘Paul will take his rum at  o’clock for he’s taking his Xanax at .’ b. *Paul a acheté une imprimante car un ordinateur. Paul has bought a printer because a computer Intended: ‘Paul bought a printer because he bought a computer.’ ()

*Paul prendra son rhum à h car Marie aussi/car pas Marie. Paul take. his rum at  for Mary too / for not Mary ‘Paul will take his rum at  o’clock because Mary will too/won’t.’

() Jim aime Jim likes a. *ainsi qu’ as.well.as

Stripping

le whisky . . . the whisky il aime la tequila. he likes the tequila

b. ainsi que la tequila. as well as the tequila ‘Jim likes whisky as well as tequila.’

18

Morphological mismatches are also put forward: in (b), for instance, the missing verb requires plural agreement while the antecedent is singular. The authors arguing for a non-deletion analysis in French consider that such mismatches argue against a deletion analysis in other languages, too (e.g. Abeillé et al. : –).

 ()

 Jim boira un whisky à h ainsi que Jim drink. a whisky at  as.well.as ‘Jim will drink whisky at  and so will Mary.’

Marie. Mary

Stripping

These authors focus on finding a proper way to redefine coordination, so that it can select fragments (stripping, coordination of unlikes) or non-headed strings of two (or more) constituents (gapping, NCC). In this approach, fragments are a proper syntactic unit, which can be found in fragment answers or questions, as well as being conjoined to a clause. Such analyses focus on (a) formalizing these units, (b) investigating which grammatical, morphological, and semantic features condition (i) at the beginning of this section actually subsumes, and (c) defining exactly how each conjunct and their union contribute to external grammatical relations such as case, agreement, extraction, selection, or theta-marking (see detailed proposals in the references mentioned). In parallel, the licensing conditions on ellipsis are restated as licensing conditions on fragmentary coordination, and are argued to rely on discourse relations, for instance S and C for gapping (see, in particular, Abeillé et al. ): () can be interpreted as a symmetrical (semantically reversible) descriptive coordination, but not as an asymmetrical coordination conveying consequence. () a. Han est content et Luc furieux / Luc est furieux et Han content. Han is happy and Luke furious/ Like is furious and Han happy Intended: ‘Han is satisfied and at the same time Luke is furious.’ b. Han est content et (??donc) Luc furieux / #Luc est (donc) furieux et Han is happy and hence Luke furious/ Luke is then furious and Han content. Han happy Intended: ‘Han is satisfied and as a consequence Luke is furious.’ One of the questions raised is whether the labels in ()–(), such as ‘stripping’ or ‘gapping’, describe homogeneous classes of constructions. For instance, (), in which the remnant is preceded by the negative marker pas ‘not’, and (), in which it is followed by the negative pro-form or polarity particle (Authier ) non ‘no’,19 though generally considered as two cases of stripping, have been shown to differ by Abeillé (b) and Morris ():20 19 Non ‘no’ and oui/si ‘yes’ are commonly held to be TP pro-forms; see for instance Rowlett (). They can be embedded under a finite complementizer:

(i)

Je pense que non. I think that no ‘I don't think so.’

(ii)

Si oui, j’irai à la piscine. if yes I will.go to the swimming pool ‘If so, I will go to the swimming pool.’

Authier () argues they are polarity particles heading FinP and licensing the ellipsis of their complement TP. 20 Somewhat confusingly, Morris and Abeillé name the former negative stripping and delayed coordination, and the latter polarity ellipsis and stripping respectively. I name them delayed coordination and polarity ellipsis. The term delayed captures the empirical impression that the remnant is conjoined to its correlate as an addendum.

 ()



Jim a écrit une lettre au directeur, mais/et pas Lea. Jim has written a letter to.the director, but/and not Lea ‘Jim wrote a letter to the director, but/and Lea didn’t.’ negative delayed coordination

() Jim est venu mais/et Jim is come but/and ‘Jim came but Jules didn’t.’

Jules Jules

non. no

polarity ellipsis

Polarity ellipses differ from other cases of ‘stripping’ in that they must follow the antecedent clause (), they can be embedded (), and they require an overt antecedent for the remnant (): () a. ??Jean, John

mais but

Marie Mary

non, no

écrira will.write

une a

lettre letter

au to.the

b. Jean, mais pas Marie, écrira une lettre au John but not Mary will.write a letter to.the ‘John will write a letter to the president but Mary won’t.’ () a. Jim Jim

sera will.be

là, there

mais but

je I

crois think

que/alors que that/whereas

b. *Jim sera là, mais je crois que/alors que Jim will.be there but I think that/whereas ‘Jim will be there but I think that/whereas Jules won’t.’ ()

Paul aime chasser, Paul loves hunt a. *mais le sanglier but the boar

président. president

président. president

Jules Jules

non. no

pas not

Jules. Jules

non. no

b. mais pas le sanglier. but not the boar ‘Paul likes hunting, but he doesn’t like to hunt boars.’ If non is a TP pro-form in (), it already conveys a clausal meaning and Abeillé () argues it involves no syntactic TP. What is to be explained, then, is the source of the remnant. Abeillé () proposes that it is directly adjoined to the pro-form, cf. (a). Morris () and Authier () argue that such structures do involve TP-deletion. According to this approach, the remnant is the left-dislocated component of an elided TP which is dominated by a PolarityP headed by the pro-form, cf. (b). All analyses can account for the fact that such constructions can be extracted from or bound into, as in (): the relative pronoun dont ‘whose’ is interpreted as a complement of souvenir ‘memory’ in both conjuncts. In Morris’s or Authier’s view, the relative is syntactically extracted from the TP, cf. (a); in Abeillé’s view, the relative does not involve movement and the pronoun’s interpretation is integrated into the anaphoric resolution of the proform non, cf. (b).





() a. Paul

viendra

mais [TP [Marie] [TP non]]

b. Paul viendra mais [CP [Mariei] [PolP non [TP ti elle ne viendra pas]] Paul will.come but Mary no she =will.come not ‘Paul will come but Mary won’t.’ ()

Un individu dont l’Histoire n’avait pas gardé souvenir, mais lui oui. a fellow of.which the.history =has not kept memory but him yes ‘a fellow that history didn’t remember but  did.’ (Anne-Marie Garat, Pense à demain, , p. )

() a. Un individu donti l’Histoire n’avait pas gardé souvenir ti, mais [CPlui [PolP oui [TP il avait gardé le souvenir ti]]]. b. Un individu donti l’Histoire n’avait pas gardé souvenir ti, mais [XP lui oui ], (oui: ‘il avait gardé le souvenir de cet individu’) Morris’s proposal additionally accounts for the fact that sprouted remnants and indefinite remnants that do not correspond to a subtopic are ruled out: they would not be salient enough to be left-dislocated. This can be extended to Authier’s analysis. Along these lines, if ‘stripping’ refers to TP-deletion with an XP remnant, polarity ellipses are actual instances of stripping. Note that the main argument against TP-deletion in French stripping does not apply to polarity ellipsis (contrary to negative delayed coordination): polarity ellipsis is, indeed, compatible with clausal coordinators (), and excluded with constituent coordinators (): () a. Marie aime le rum. Or, Pierre Mary likes the rum yet Peter ‘Mary likes rum. Yet, Peter does not.’

non. no.

b. Quand vous avez reçu votre carte mère, était-elle sous blister ( . . . )? when you have received your motherboard was=it under blister car la mienne non.21 for the mine no ‘When you received your motherboard, was it in a blister pack? Because mine wasn’t.’ ()

*Paul est venu ainsi que Marie oui. Paul come.PPF as well as Mary yes Intended: ‘Paul came as well as Mary.’

The uncovering of their distinct properties argues for a fine-grained investigation of other, barely studied structures, such as (a–b), which display a reverse order for the remnant and the negative item: (a) yields a corrective interpretation absent from other cases of stripping; in (b) the negative reinforcement allows an otherwise impossible remnant–negation order. 21

Example from , consulted on  December .

 () a. Jim est venu et Jim is come and ‘J came, not J.’ b. Jim Jim

aime likes

le the

café coffee

non no



Jules. Jules

et/mais and/but

Jules Jules

pas not

*(du tout). *(at all)

That ‘gapping’ also subsumes different constructions is argued by Dagnac (), on the basis of gapped sentences with simple ni ‘nor’ and double ni ‘neither . . . nor’. Ni is a strong Negative Polarity Item (NPI) (cf. de Swart ; Mouret b) that must be c-commanded by a negative expression in overt syntax, which accounts for the grammaticality contrasts in (): in (a), contrary to (b), ni fails to be c-commanded by jamais ‘never’. Double ni is a negative expression (cf. de Swart ; Mouret b) that does not require such a licensing configuration, cf. (). () a. *[Jean ne votera pas/jamais pour Tim] ni [Marie ne votera pour Bob]. John  vote. not/never for Tim nor Mary  vote. for Bob ‘John will not/never vote for Tim nor will Mary vote for Bob.’ b. [CP Jamais [Jean ne votera pour Tim ni Marie ne s’abstiendra]]. never John  vote. for Tim nor Mary  abstain. ‘Never will John vote for Tim nor will Mary abstain from voting.’ ()

Ni Jean ni Marie n’aiment les huîtres. neither John nor Mary =like the oysters ‘Neither John nor Mary like oysters.’

Dagnac () argues that neither a clausal deletion view nor a fragment analysis of gapping can account for French gapped constructions conjoined by simple ni ‘nor’: the grammaticality contrast between the full and gapped sentences in () vs (a) cannot be explained if () is derived from (a). The non-pronunciation of part of the second conjunct should not affect a syntactic licensing configuration occurring outside of it. Likewise, the semantic reconstruction of the fragment Marie pour Bob cannot affect the external licensing of ni. () Jean ne votera pas/jamais pour Tim ni John  vote. not/never for Tim nor ‘John will not/never vote for Tim nor Mary for Bob.’

Marie Mary

pour for

Bob. Bob

A more plausible analysis is to view () as an instance of vP-coordination, instead of clausal coordination, relying on ATB movement of the verb out of the conjoined vPs, rather than an instance of TP-deletion (see Dagnac , and Johnson ,  for detailed argumentation). However, gapped constructions involving French double ni must be clausal. A subject inside the first conjunct can bind into the gapped one when the conjunction is simple ni, cf. (). This is predicted if vPs are conjoined, since the higher subject chaque enfant c-commands the lower one ses parents, which stays in specvP.





However, this is not possible with the double conjunction, cf. (): this, as Centeno (: –) argues for Spanish, is expected if clauses rather than vPs are conjoined, since the higher subject does not c-command the lower one. () Ici, jamais chaque enfanti n’ auraj [vP tj un violoncelle ni here, never each kid  have. a cello nor sesi parents tj les moyens de lui en louer un.] his parents the money of to.him  rent one ‘Here, never will each child have a cello, nor his parents enough money to rent one for him.’ () *Ici, ni chaque enfanti n’ aura (jamais) un violoncelle here, nor each kid  have. (never) a cello [ni sesi parents les moyens de lui en louer un.] nor his parents the money of to.him  rent one ‘Here, neither will each child (ever) have a cello, nor his parents enough money to rent one for him.’ Besides, in French double coordination, the first conjunction must immediately precede the first conjoined item, as in () (cf. Abeillé and Godard ; Mouret ), and this holds for ni . . . ni. While the position of the first coordination clearly argues for a non-clausal analysis of NCC (cf. Mouret ), since in () the first et can only precede whisky, not Jim, it argues in turn for a clausal view of gapping with ni . . . ni (and more generally with double conjunctions): in gapped examples such as (), the only licit position for the first ni is clause-initial.22 ()

a. Jim boira (et un whisky et Jim will.drink and a whisky and ‘Jim will drink both a whisky and a tequila.’

une a

tequila). tequila

b. (Et Jim boira un whisky) (et Marie boira une tequila). and Jim will.drink a whisky and Marie will.drink a tequila ‘Jim will drink a whisky and Mary will drink a tequila.’ () (*Et) Jim boira (et) un whisky à h et une tequila à h. and Jim will.drink (and) a whisky at  and a tequila at 

NCC

() a. Ni Paul ne connait (*ni) un mot d’ anglais ni Jim Gapping nor Paul  knows nor one word of English nor Jim un mot de français. one word of French ‘Paul knows neither a single word of English, nor does Jim know a single word of French.’

22 Note that the main argument against a clausal view of coordinate ellipses in French is not applicable in these cases, either: the coordinating conjunctions that only select either clauses or nontensed constituents, such as car ‘for’, or ‘yet’, and ainsi que ‘as well as’, only exist as single functors.





What the French data suggest so far, then, is that research must now focus on a finer-grained investigation of coordinate ellipses, in particular stripping and gapping, and that the lexical items that surround the remnants may be of greater import to define their syntactic structure than was previously thought—a conclusion also reached by Repp () for gapping in other languages. If conflicting approaches have not evolved into a consensus so far, it may be partly because languages have more than one way to produce coordinate ellipses.

. E  

.................................................................................................................................. Ellipses in comparatives come in two species: an obligatory type, specific to comparatives, illustrated in () and an optional type, consisting of a predicate or coordinate ellipsis occurring inside a comparative clause, as in (). () a. Paul has more cats than we have [x many cats]. b. Paul has more cats than he has [[x many ] kids].

Comparative Deletion Comparative Subdeletion

() Paul is happier than Mary.

Stripping in a comparative

The size of this chapter makes it impossible to do justice to the whole range of questions raised by French elliptical comparatives, all the more so because it has received relatively little attention within formal frameworks (but see Desmets ). I will first focus on stripped and gapped comparatives, which look similar to the English ones. I will then briefly discuss the specific properties of French in the other cases.

.. Stripping and gapping in comparatives The first question raised by these comparatives in French is whether they are all syntactically clausal. In other Romance languages the equivalent of a stripped comparative such as French (a) is formed as a phrasal complement to a preposition-like item, as was sometimes the case in Old French (c). () a. Marie Mary

est is

plus more

belle pretty

b. Maria è più bella Mary is more pretty ‘Mary is prettier than I am.’

que than di of

moi/*je. me/I me/*io me/I

c. N’est pas mains riche de mon  is not less rich of my ‘He is not richer than my father.’

[French] [Italian]

père. father

[Old French]

Though nowadays the remnant seems to be introduced by the complementizer que, it could be a homonymous preposition-like item, as argued for Romanian and Italian by van





Peteghem (). In Romanian, indeed, the complementizer-like word ca can be followed by a pronominal remnant that, though corresponding to a subject antecedent, bears accusative case and not nominative case as expected under a clause deletion analysis, cf. (a). This, however, does not extend to gapping-like comparatives, whose subject remnant bears nominative case, cf. (b): () a. El este la fel de înalt He is the same big ‘He is as big as you are.’

ca as

mine Psg.Acc

/*eu. Psg.Nom

b. Ea lucreaza mai mult acasa decât tu/*tine la serviciu. she works much more at.the.house as you./*you. at work ‘She works much more at home than you do at your office.’ Case can be of no help for French, though: strong pronouns that appear in these constructions are not overtly marked for case. Moi in (a) could equally correspond to the argument of a preposition-like que or to a dislocated subject remnant in a clausal-deletion analysis. However, French displays another contrast between stripping-like and gappinglike constructions in comparatives, suggesting that the former are not instances of ellipsis. Most overt quantifiers select a de-NP, cf. (). So do covert quantifiers, noted as eQ here, found for instance (cf. Müller ) before the indefinite object of a negative clause, cf. (a). Such de-NPs cannot be licensed when preceded by a preposition, cf. (b): ()

Léa a bu beaucoup de champagne. Lea has drunk much of champagne ‘Lea has drunk much champagne.’

() a. Il n’a pas mangé eQ he =has not eaten ‘He didn’t eat any apples.’

de pommes. of apples

b. *Il n’a pas parlé à/avec eQ he =has not talked to/with ‘He did not talk to anyone.’

de gens. of people

A covert quantifier is generally assumed to precede the compared item in full comparatives, cf. ().23 () Léa a bu plus de champagne [que Jim n’a bu eQ de tequila].24 Lea has drunk more of champagne than Jim =has drunk of tequila ‘Lea drank more champagne than Jim drank tequila.’ 23 Its representation and analysis varies; Bresnan () initially marks it as x much, Kennedy () as Op. I mark it as eQ. 24 Comparative constructions trigger the presence of an expletive ne in the than-clause, which I will not analyze—I gloss it as .





De-NPs also occur in comparatives when they have a PP correlate, as in (a). If strippinglike comparatives involve TP-deletion, it is unclear how (a) could be derived from the ungrammatical (b), as argued by Zribi-Hertz (). If they have a monoclausal structure similar to (), where que acts as a phrase functor, this fact finds an explanation: no empty quantifier is required to account for the presence of de femmes; both de-NPs are directly licensed by the comparative quantifier plus ‘more’. ()

a. Il a parlé à/avec plus d’hommes he has talked to/with more of men ‘He talked to/with more men than women.’

que than

de of

femmes. women

b. *Il a parlé à/avec plus d’hommes qu’il n’a parlé à/avec eQde femmes. he has talked to/with more of men than he=hastalked to/with of women ‘He talked to more men than he talked to women.’ () Il a parlé à/avec [QP plus [PP d’hommes he has talked to/with more of men ‘He talked to/with more men than women.’

que de femmes]]. than of women

In putatively gapped comparatives, on the other hand, remnants with PP-correlates are ruled out, just as in full comparatives; cf. (): () *Paul a parlé à/avec plus d’hommes que Marie (à) de femmes. Paul has talked to/with more of men than Mary to of women Intended: ‘Paul talked to more men than Mary women.’ The questions raised by these comparatives are then: if stripped comparatives are monoclausal, how are they mapped to a biclausal meaning? If other French comparatives are embedded clauses (Desmets ), do they actually involve gapping, which is generally viewed as unavailable in embedded contexts—and if so, why and how is gapping licensed only in these contexts?

.. Anaphora in comparatives Comparative deletion and comparatives with predicate ellipses raise another related question in French: they involve an overt pronoun (cf. ()–()), or can alternate between an overt pronoun and a gap (cf. ()). () a. Paul a plus de chats que nous n’en avons (, eQ de chats). Paul has more of cats than we =of-it have ( of cats) ‘Paul has more cats than we have.’ b. Paul est plus heureux que Paul is more happy than ‘Paul is happier than Mary is.’

ne l’est =it=is

Marie Mary

(, eQ (

heureuse). happy)





() a. Marie a parlé à plus d’étudiants que Léa ne l’a fait. b. Marie a parlé à plus d’étudiants à Madrid que Léa ne l’a fait Mary has talked to more of students (in Madrid) than Lea =it=has done à Berlin. (in Berlin) VP-anaphor ‘Mary talked to more students (in Madrid) than Lea did (in Berlin).’ () a. Marie a parlé à plus d’étudiants que je ne pensais NCA b. Marie a parlé à plus d’étudiants que je ne le pensais Clausal pronoun Marie has talked to more of students than I =(it=) thought ‘Mary talked to more students than I thought.’ Modals can be followed by TP-ellipsis (cf. section .), but also by a vP anaphor (le faire ‘do it’) or by a full clausal pro-form (le ‘it’), and all three structures show up in comparatives: () Marie a parlé à plus d’étudiants . . . Marie has talked to more of students a. qu’elle n’aurait dû. than she =must.. b. qu’elle than she

n’aurait dû =must..

le faire. it=do

Modal Ellipsis Predicate anaphor

c. qu’elle ne l’aurait dû. Clausal pro-form than she = it=must.. ‘Mary talked to more students than she should (have (done)).’ What is of concern here is the nature and behavior of the null degree item eQ, which has been extensively debated for other languages (see Kennedy  and references therein). The French data do indeed challenge the view that eQ or a constituent containing eQ is moved (overtly or covertly) to an operator position in specCP. () illustrates (one simplified version of) such an analysis with an adjectival comparison: ()

Paul is more faithful than eQi Mary .

Since pro-forms are reputedly opaque to syntactic extraction, eQ would have no position to move from in (), (b), and (b,c), as well as in (a) if, as argued in section ., it involves an opaque null pronoun.25 For instance, in () illustrating a comparative with a VP-anaphor, eQ should have moved from the syntactically opaque le.

25 Kennedy and Merchant (b) deal with a related problem for English NCA, Bentzen et al. () with a similar one raised by German es können ‘it can’ structures. Whether one of these proposals could extend to all the French cases discussed here has not been investigated yet. As argued by Houser et al. () for the Danish pro-form det found in similar contexts, le can (to a certain extent) be A-extracted from: finding some common property between A-movement and eQ-movement may provide a solution to this problem.

 () Paul est plus fidèle que eQi Paul is more faithful than ‘Paul is more faithful than I thought.’



je ne le pensais. I =it=thought

One may then assume that in comparatives all these cases involve, one way or another, the deletion of a clause. But such a solution must explain why remnants with PP correlates are allowed, cf. (a,c), while they are ruled out in full clauses, cf. (b,d): ()

Paul a écrit à plus de collègues Paul has written to more of colleagues ‘Paul has written to more colleagues . . . ’ a. que Jim ne l’a than Jim =it=has ‘than Jim has done.’

fait. done

b. *que Jim n’a écrit than Jim =has written ‘than Jim has written to.’

à to

eQ x much

de of

collègues. colleagues

c. que Jim ne le pensait. than Jim =it=thought ‘Paul wrote to more colleagues than Jim thought.’ d. *que Jim ne pensait qu’il avait écrit à eQ de collègues. than Jim =thought that he had written to eQ of colleagues ‘Paul wrote to more colleagues than Jim thought he had written to.’ The French data may then argue for a semantic reconstruction of the comparative clause. Predicate anaphors, indeed, cannot be wh-extracted from: (a) shows that, for instance, le faire rules out wh-extraction. However, predicate anaphors can be bound into.26 In (b), the quantifier chaque ‘each’ in the second conjunct must bind into the VP-anaphor for the possessive to be correctly interpreted: ()

a. *Paul devait voir plusieurs collègues, mais je ne sais Paul should see several colleagues, but I =know pas lesquels il l’a fait. not which.ones he it=has done ‘Paul was to meet several colleagues, but I don’t know which ones he did.’

26 They also seem to license a wide-scope reading of quantifiers, cf. (i). This is reminiscent of Japanese null arguments, which also ban overt extraction while allowing binding and QR (Sakamoto forthcoming).

(i) Un médecin examinera chaque victime et un psychologue a doctor examine. each victim and a psychologist ‘A doctor will examine each victim, and a psychologist will, too.’ (8 victim, ∃ doctor/ 8 victim, ∃ psychologist)

le fera aussi. it=do. too



 b. Chaque adultei a déclaré soni identité, et chaque enfantj le fera aussi. each adult has declared his identity, and each child it=do. too ‘Each adulti declared hisi identity, and each childj will (declare itsj identity too), too.’

The exact nature of this coindexing should then be investigated in parallel to that of the eQ item, in order to see whether and how it fits in with the arguments put forward to diagnose the presence of movement in non-pronominal comparatives.

. N   (NPE)

.................................................................................................................................. French has two main kinds of nominal ellipses, indefinite NPE and definite NPE, exemplified in () and () respectively.27 () Des étudiants sont venus. .. students have come. Trois/ certains/ plusieurs sont déjà repartis. three/ some/ several (of these) students are already gone ‘Three students came. Three/some/several (of them) are already gone.’ ()

J’aime bien la robe bleue, I like well the dress blue ‘I like the blue dress,’ a. mais je préfère la < robe > but I prefer the < dress> ‘but I prefer the red one.’

rouge. red

b. mais la < robe> rouge me va but the dress red to.me=goes ‘but the red one fits me better.’

mieux. better

Definite ellipses can be found in any syntactic position. They require the presence of an unmodified adjective or adjective-like constituent sometimes described as classifying (superlatives, color and measure adjectives, ordinals, possessives, and seul ‘only’, autre ‘other’, même ‘same’), cf. ().28 Other adjectives may be found only in explicitly contrastive contexts, cf. (). Additional constituents, such as PPs or relatives, may be present, cf. (). Though much more infrequently, possessive and demonstrative determiners sometimes also allow definite NPE, cf. ().

The properties of French NPE have been most comprehensively studied by Sleeman (, ), but see also Corblin et al. () and references therein, Cabredo-Hofherr (), and Gagnon (b). 28 When a PP or relative modifier is present, in the absence of an adjective the demonstrative pronoun celui must be used. See Cabredo-Hofherr () for a detailed analysis. 27





() a. *Je préfère la de Léa. I prefer the of Lea Intended: ‘I prefer Lea’s.’ b. Je préfère la rouge I prefer the red ‘I prefer Lea’s red one.’

de of

Léa. Léa

() Il y a actuellement deux types d’énergies propres: la solaire et l’éolienne. there are presently two kinds of energies clean: the solar and the aeolian ‘There are presently two kinds of clean energy available: solar energy and wind power.’ () Je préfère la rouge que tu portais la I prefer the red that you wore the ‘I prefer the red one you were wearing last time.’

dernière last

() %J’aime bien ta robe bleue mais je préfère I like well your dress blue but I prefer ‘I like your blue dress but I prefer my/this red one.’

fois. time

ma/cette rouge. my/this red

Indefinite NPE is limited to the subject preverbal position: in postverbal direct object position the nominal content must be expressed either by a noun or by the quantitative clitic en, cf. (). Within PPs a local set-triggering modifier, such as autre ‘other’, d’entre eux ‘of them’ is required, cf. (c). The ellipsis as well as the pronouns can have a partitive or a non-partitive reading, cf. (). () a. Deux sont partis. two have left ‘Two have left.’ b. J’*(en) ai I of.it have ‘I saw two.’

vu seen

deux. two

c. Ils ont parlé à deux *(autres/d’entre eux). they have talked to two others of them ‘They talked to two more/two of them.’ ()

Ils ont arrêté plusieurs étudiants They have arrested several students a. et ils *(en) ont emprisonné and they of.it have jailed ‘They have jailed two of these students.’ ‘They have jailed two (more) students.’ b. Deux ont été emprisonnés. Two have been jailed ‘Two of these students were jailed.’ ‘Two (more) students were jailed.’

deux. two partitive reading non-partitive reading

partitive reading non-partitive reading





Note that in both definite and indefinite NPE, if the nominal content corresponding to the missing N is right-dislocated, it takes the form de N: () a. Je préfère la rouge, *(de) robe. I prefer the red of dress ‘As for dresses, I prefer the red one.’ b. Deux sont venus, *(d’) two have come of ‘As for students, two came.’

étudiants. students

Among the many questions raised by French NPE are the following, none of which have received a clear answer so far. First, do they involve deletion of a nominal constituent (cf. Ronat ), a null pro-form (Gagnon b for definite ellipsis), or, as is argued by Marandin () for definitive ellipses, some other mechanism, such as a DP headed by a morphologically nominalized adjective in cases like () or by a non-N in the other cases? Unlike verbal and clausal ellipses, nominal ellipses lack consensual tests to tell these analyses apart: they mostly rely on the distribution and potential anaphoric range of the remnants. A related question is whether in all French nominal ellipses, an NP is missing. Gagnon (b), for instance, argues that both in French and in English, indefinite ellipses elide an anaphoric Partitive Phrase d’entre eux ‘of them’. This is consistent with the full structure of most of the stranded determiners, which does involve a PartP, cf. (b,c): ()

Ils they

ont have

a. Beaucoup many b. *Beaucoup many

arrêté arrested ont have

plusieurs several été been

étudiants students

étudiants. students

emprisonnés. jailed. ont have

été been

emprisonnés. jailed

c. Beaucoup d’entre eux ont été emprisonnés. many of them have been jailed ‘They have arrested several students. Many (students/of them) have been jailed.’ But this analysis is only consistent with the partitive reading (cf. b); if a PartP is elided, it cannot always take the form d’entre eux ‘of them’. Instead, two underlying forms should be postulated: d’entre eux ‘of them’ (partitive indefinite ellipsis) and de N ‘of N’ (non-partitive indefinite ellipsis). More generally, both the structure of partitive DPs and the structural position of various kinds of adjectives and modifiers or adjuncts and how they can account for their ability to appear in a nounless DP (cf. ()) have not reached a clear consensus so far: the answer is linked to how much structure a given framework is ready to admit for DPs: ()

a. La jaune /??Plusieurs jaunes / Plusieurs des jaunes the yellow several yellow several of.the yellow ‘The yellow one. Several yellow ones. Several of the yellow ones.’ b. Les autres / Plusieurs autres / Plusieurs des autres the others / several others / several of.the others ‘The other ones. Several other ones. Several of the other ones.’





How nounless DPs are licensed is also an open question. Lobeck () and, to some extent, Bouchard () propose that agreement morphology on the (governing) remnants, D (+Adj) or Q, license NPE: this would account for the difference between English () and its French counterpart (): ()

*I want the green.

no agreement morphology

() Je veux les rouges / la I want the. red. / the. ‘I want the red ones / the red one.’

rouge. red

plural/feminine morphology

However, acquisition data (Sleeman and Hulk ) show that in French, NPE is mastered well before adjectival agreement. Besides, gender and number agreement per se fails to license NPE with relational adjectives, as in (). ()

L’énergie est chère, *mais la solaire The energy is expensive but the solar ‘Energy is expensive, but solar energy is promising.’

est is

prometteuse. promising

A more promising view is to consider that NPE is licensed via the partitive semantics of the remnant (Bouchard ; Sleeman ) or contrastive focalization on it (Corver and van Koppen ). Partitive adjectives as well as focalization create a subset of the larger set introduced by the antecedent DP. Explicit contrast (cf. ()) does indeed enable the otherwise relational adjectives, as in (), to create subsets of ‘clean energy’. In Sleeman’s view, classifying adjectives, such as color and measure adjectives, denote subsets, while intéressant ‘interesting’, a non-classifying adjective, in (a) does not: les plus in (b) extracts the upper part of the scale from the contextual set of interesting conferences, then licensing NPE: () a. *Je I

n’ai =have

pas entendu not heard

les the

intéressantes, interesting..

(de conférences). (of conferences)

b. Je n’ai pas entendu les plus intéressantes, (de conférences). I =have not heard the most interesting.. (of conferences) ‘(As for conferences,) I missed the (most) interesting ones.’ A difficulty for assessing syntactic proposals for the licensing of NPE stems from the lack of consensus on what actually counts as NPE. Structures such as (), for instance, have received diverging treatments: ()

Il y a des pommes. it there have of.the apples ‘There are apples.’ a. Quelques-unes sont gâtées. some.ones are bruised ‘Some are bruised.’ b. La mienne / Celle de Léa est gâtée. the mine / that of Lea is bruised ‘Mine/Lea’s is bruised.’



 c. Chacune est gâtée. each.one is bruised ‘Each one is bruised.’

According to their etymology, one can indeed conceive of quelques-uns lit. ‘some ones’, la mienne lit. ‘the mine’, celui lit. ‘this him/it’, chacun lit. ‘each one’ as sequences composed of a determiner and an adjective (or two determiners). In this case, most of them are clear cases of NPE, since a nominal head is missing, and their properties can be compared to those of other cases of NPE, in French or cross-linguistically. But, they can also be analyzed as atomic pronouns in present-day French, as generally assumed in French grammars. In this case, nothing is ‘missing’ in their structure, and () lies outside the empirical body of elliptical phenomena. Nominal ellipses in French provide more questions than answers, so far—including one about the exact range of empirical data to be accounted for under the label ‘nominal ellipsis’. As such, they form part of the many issues for future work on French ellipsis to address, and may shed light on other aspects of the grammar, in particular on the structure of the French DP.

. C

.................................................................................................................................. French provides evidence that ellipsis, as well as its subtypes, is a label that may, crosslinguistically or within one language, encompass various anaphoric devices, some of which are best analyzed by a deletion process while others are not. Coordinate ellipses in particular call for a much finer-grained investigation than has yet been done, in order to account for their syntactic properties, the lexical items they rely on (including the coordinators) and the contextual discourse relations that license them. As sluicing and comparatives in particular point out, our understanding of these phenomena is crucially correlated with our understanding of the way wh-movement, binding, and coindexing can variously interact with ellipsis and anaphora. While within French the licensing of these constructions is still a barely explored domain, the particular and sometimes puzzling properties of the French data can, no doubt, provide a stimulating testing-ground for existing and/or new cross-linguistic theories of ellipsis licensing.

  ......................................................................................................................

 ......................................................................................................................

´ ˊ 

. N 

.................................................................................................................................. A in many other languages, the head noun or the head noun together with one or more modifiers can be unpronounced to the exclusion of a modifier, numeral, or (quantificational) determiner in Hungarian. Ellipsis is strictly only found in non-possessive noun phrases, while in possessed noun phrases, a non-elliptical anaphoric strategy is used. The next two subsections give details of both strategies.

.. Nominal ellipsis in non-possessed nominals In non-possessed nominals, the missing nominals can be understood with reference to an entity in the linguistic or the extra-linguistic context. In the following examples, the part of the noun phrase that is understood to be missing is indicated by __ .1 () a. Ezt a házat régen építették. Azzal a kettő __ -vel this. the house. long.ago built. that. the two - viszont most készült el a kivitelező. on.the.other.hand now got.ready.  the contractor ‘This house was built a long time ago. Those two on the other hand were now made ready by the contractor.’ b. Az új kis házak el takarták a the new small house.  concealed. the ‘The new small houses concealed the old small ones.’

régi- __ -ek-et. old --

Non-standard abbreviations in this chapter are the following: : affirmative particle; : complementizer; : conditional, : reversing response particle; : future auxiliary; : habitual auxiliary; : nominalizer; : participial form; : potential modality; : discourse particle; : preverbal particle; : morpheme marking relative pronouns; : copulative element linking arguments/adjuncts to derived nouns. Past tense and object definiteness agreement on the verb are not glossed throughout. Preverbal particles and verbs are spelled in two words, contrary to the Hungarian orthographic tradition. 1



ˊ  c. [context: Standing in front of a heap of melons at the market] Kérek két nagy__ -ot! ask. two big - ‘I’d like to have two big ones.’

As can be observed, when the noun is missing, the overt number morphemes and case morphemes that normally get spelled out on the noun only appear on the linearly last remnant preceding the missing noun (in case they contain a harmonic vowel, they harmonize with the last remnant, too), as was noted in Bánréti (, ), Kenesei et al. (), Laczkó (), and Saab and Lipták (). The linearly last remnant can also be a clausal modifier: for example, an adjectival participial clause, as shown in the following example, (b): () a. Csak közjegyző által hitelesített iratokat only notary by certify. document.. ‘They accept only documents certified by a notary.’ b. Csak közjegyző által hitelesített __-eket only notary by certify. -. ‘They accept only ones certified by a notary.’

fogadnak accept.

fogadnak accept.

el. 

el. 

Missing nouns or nominal constituents have been analyzed as involving a silent pronominal pro in Bánréti ,  (see also Laczkó ) and as ellipsis in Dékány  and Saab and Lipták .2 Here I follow the latter kind of proposals and assume that the missing nominal is the result of ellipsis. Adopting a structure like () for (unpossessed) nominal constituents (see, e.g., Cinque ), in which adjectives are adjoined to the noun phrase, and the plural marker originates in the NumP projection, the deleted category in Hungarian nominal ellipsis corresponds to an NP. The missing NP may be unmodified or modified (see () and () for examples of both types). ()

[DP (demonstratives) D [NumP (numerals) Num[pl] [NP (AP) [NP N ]]]]

Evidence that the elided category cannot be as big as a numeral phrase (NumP) comes from the observation that numerals and number morphology must always survive the ellipsis (cf. () and ()). () a. Az új ház el takarta the new house  concealed. ‘The new house concealed the old ones.’

a the

régi- __ -ek-et. old --

2 Under either type of analysis, these facts have to be differentiated from nominalized adjectives (see Giannakidou and Stavrou ; Panagiotidis b) such as gazdag-ok ‘rich-pl’ ‘the rich’, beteg-ek ‘sick-pl’ ‘the sick’, fiatal-ok ‘young-pl’ ‘the young’ for at least two reasons: (i) nominalized adjectives, unlike missing nouns like in () or (), can be used without any (linguistic or non-linguistic) antecedent; (ii) nominalized adjectives have a generic [+human] interpretation, while the missing nouns in () and () are not restricted in this way.





b. A két új ház el takarta a három régi- __ -t. the two new house  concealed. the three old - ‘The two new houses concealed the three old ones.’ () Ez a két új ház el takarta azt this the two new house  concealed. that ‘These two new houses concealed those three (new ones).’

a the

hárm- __-at. three -

There are some information-structural criteria that NP ellipsis usually complies with. Preferentially, NP ellipsis contains adjectival remnants that are new and not given in the discourse—in the sense of not having been mentioned yet. Pronouncing given adjectival modifiers gives a slightly awkward, redundant utterance, but they do not count as ungrammatical. In the case of numerals as remnants, givenness is fully tolerated and gives rise to no sense of redundancy. () a. János vett egy kék autót. Mari is vett egy (? kék) __-et. János bought. a blue car. Mari also bought. a blue - ‘János bought a blue car. Mari also bought a blue one.’ b. Levi kivett két almát a kosárból. Én is kivettem kettő__-t. Levi took. two apple. the basket. I also took. two - ‘Levi took two apples from the basket. I also took two.’ It is also important to mention that non-identity between remnants of NP ellipsis and their correlates does not trigger any syntactic marking of contrast via contrastive focusing. Consider the following examples where the adjectival remnant is non-identical to another adjective in the antecedent clause: () János vett egy kék autót. Mari pedig vett egy piros __-at. János bought. a blue car. Mari  bought. a red - ‘János bought a blue car. Mari bought a red one.’ In (), the elliptical noun phrase is in postverbal position and is intonationally unmarked— there is no contrastive stress on the remnant. Both properties are earmarks of constituents that do not distribute as contrastive focus or contrastive topic expressions (with contrast on their adjective/numeral). That being said, elliptical noun phrases can contain contrastively focused remnants, and—in line with the rules of Hungarian—appear in the preverbal focus position as contrastively focused phrases (cf. ()). () János vett egy kék autót. Én ´ __-et vettem volna János bought. a blue car. I white  bought.  a helyében. the place.. ‘János bought a blue car. I would have bought a  one in his place.’



ˊ 

This shows that contrastive focus on the remnant is optional, and not a necessary property of elliptical nouns, and thus cannot be considered the licensing factor for noun phrase ellispis in general (contra Corver and van Koppen ; Eguren ).

.. Anaphoric possessed nominals In contrast to non-possessed noun phrases, possessed noun phrases do not allow for nominal ellipsis. This equally holds for possessives with dative (a) and with nominative possessors (b) (see Szabolcsi  for basic differences between the two). The lack of elliptical possessives stems from the fact that anaphoric possessives make exclusive use of a pronominal strategy that is earmarked by the use of the -é suffix (Bartos b; Laczkó ; Dékány , ), illustrated in (c). () Nádasdynak a könyv-e már Nádasdy. the book- already ‘Nádasdy’s book has already appeared.’ a. * Esterházynak __ Esterházy. b. * Esterházy __ Esterházy. c.

csak only csak only

jövőre next.year jövőre next.year

meg 

kerül get. kerül get.

Esterházy-é csak jövőre kerül Esterházy-´ only next.year get. ‘Esterházy’s will only appear next year.’

jelent. appeared. kiadásra. publication.

kiadásra. publication.

kiadásra. publication.

The precise analysis of the -é suffix is a point of contention in the literature (Bartos b equates it with a functional head that selects the noun, Laczkó  with the proform, and Bartos b, Dékány  with genitive case). What all analyses agree on is that anaphoric noun phrases involve a proform. As can be seen in (a), the anaphoric noun phrase is obligatorily adorned with the possessor agreement morpheme (such agreement morphemes are present with pronominal possessors in Hungarian) and the number morpheme indicating plurality of possession: -tek spells out agreement with a  possessor and -i indicates plural possession. Importantly, as (b) shows, -é can never co-occur with the possessedness morpheme -(j)a/(j)e, neither can it co-occur with the possessed noun (c). () a.

a ti-é-i-tek the you-´-- ‘your() ones’

c. * a ti-é-i-tek the you.-´-- ‘your() book’

b. * a the

ti-je-é-i-tek you.--´--

könyv book

Anaphoric possessed noun phrases furthermore cannot contain any overt numeral or adjectival modifier; see the next example as illustration. In Dékány , the latter property





is explained with reference to the fact that anaphoric possessed noun phrases contain a proform and pronominals cannot be modified in Hungarian. The ungrammaticality of numeral and adjectival remnants in possessed nominals sharply contrasts with nonpossessed nominals, where such remnants are allowed (see () and () again). This difference provides a strong argument to the effect that the missing element in anaphoric possessed noun phrases is a proform, while in non-possessed noun phrases it corresponds to ellipsis of a nominal projection (see Dékány ). ()

Nádasdy(nak az) új könyv-e már meg jelent. Nádasdy. the new book- already  appeared. ‘Nádasdy’s new book is already in print.’ * Esterházy új-é csak jövőre kerül kiadásra. Esterházy new-´ only next.year get. publication. ‘Esterházy’s new one will only appear next year.’

Concerning the “size” of the anaphoric pronoun in possessives, Dékány argues that the anaphoric proform replaces a piece of structure that is bigger than an NP. To understand why, consider the structure of possessives as in () (Dékány ; also É. Kiss ) where there are two functional projections dedicated to marking the possessive relation: the possessor agreement projection (PossP) and the possessedness projection (PossP). These two projections flank the NumP that hosts the plural possession marker -i. Standardly, the possessor is taken to be generated in Spec,PossP and the possessed noun as head of the NP: ()

[DP [PossP [NumP [PossP (possessor) -(j) a/(j)e [NP N ]]]]]

From the fact that in anaphoric possessives, the noun and the possessedness morphemes are never overt (cf. (b,c)), Dékány concludes that the proform must minimally correspond to the Poss’ node, which subsumes the NP projection and the Poss0 head. To finish off the discussion of anaphoric possessives, consider the following example, which at first sight seems to contradict the claim that ellipsis is impossible in possessed noun phrases. The interpretation of the missing noun is preferred to be that of a possessed nominal. ()

Mari régi kabát-a-i mindig tiszták voltak, de az új__-akat Mari old coat-- always clean. were but the new -. nem mossa ki soha. not wash.  never ‘Mari’s old coats were always clean, but she does not ever wash the new ones.’

There are, however, two strong indications that the possessed interpretation is only pragmatically controlled for in cases like this and that we are dealing with an unpossessed nominal undergoing ellipsis here. One indication is provided by the nominal morphology found in the elliptical nominal: the endings are characteristic of non-possessed noun phrases. In possessed noun phrases,



ˊ 

the plurality of the possession is spelled out by the invariable -i morpheme, cf. kabát-a-i (coat--), while in the elliptical új-ak (new-.) the plural marker is the ordinary -k morpheme (together with an epenthetic vowel) that is found on non-possessed nouns. As the ungrammatical forms furthermore illustrate in (), there is no other variant that is acceptable (as noted in Kenesei et al. ). ()

Mari régi kabát-a-i mindig tiszták voltak, de az {új __-ak / Mari old coat-- always clean. were but the new - *új __-a-i / *új __-i } teljesen koszosak. new -- new - completely dirty. ‘Mari’s old coats were always clean, but the new ones are completely dirty.’

The other argument against a possessed NP analysis of these data comes from the observation that the possessor can never be overtly present in the elliptical nominal, either in dative or nominative case (). ()

Mari régi kabát-a-i mindig tiszták voltak, de (*neki) az (*ő) Mari old coat-- always clean. were but . the  új __-ak teljesen koszosak. new - completely dirty.

These two observations jointly confirm that the elliptical noun phrase in () is not a possessed NP, but an unpossessed one, and the possessed reading of the missing nominal must be derived pragmatically.

. P 

.................................................................................................................................. The main predicate of the clause can be missing in Hungarian in two configurations: what can be referred to as auxiliary()-stranding VP-ellipsis and V-stranding VP-ellipsis, following a distinction made by Goldberg (). This section reviews these two types, together with a further reduced variant of V-stranding, the so-called preverb-stranding pattern. This section closes with a short discussion of pseudogapping. Before turning to the specifics of predicate ellipsis, it is important to introduce key assumptions about the clause structure of Hungarian that are going to be made use of in the discussion. In most syntactic accounts, Hungarian clauses are taken to contain an inflectional layer, termed TP in the following (comprising tense, agreement, and mood specifications/projections not distinguished here any further), and a predicate layer, termed vP below. The predicate layer also comprises various subprojections, most notably an aspectual projection (AspP/PredP) and the core lexical predicate, the VP. Lexical verbs in Hungarian often combine with so-called verbal modifiers with aspectual/predicative meaning, which are phrasal constituents (Koopman and Szabolcsi ; den Dikken ; Surányi a) and comprise preverbal particles (or preverbs, PVs for short), incorporated nominals, and PPs of distinct types.

 () Bea fel hívta a Bea  called. the ‘Bea called her parents.’



szüleit. parent...

Verbal modifiers are syntactically independent of the verbal head (for this reason they will be spelled as separate words in this chapter, contrary to rules of Hungarian orthography). Following Piñón (), Olsvay (, ), and Surányi (a), I take verbal modifiers to originate from an AspP/PredP-internal position and to move to specTP in overt syntax (in order to satisfy an EPP property, note that subjects do not raise to specTP). Finite verbs raise to the tense head (Brody ; Kenesei ; Surányi a), and this results in the obligatory adjacency between verbal modifiers and the verbal head that characterizes all clauses without focal material. The structure in (), indicating the position of the verb and the most frequent verbal modifier, the preverbal particle, will be adopted in the discussion in ..–... ()

[TP  verb [vP . . . [Pred/AspP . . . [VP . . . ]]]]

.. -stranding predicate ellipsis -stranding predicate ellipsis data have been described in Bánréti () and Bartos (a) (see also Gyuris ). This is a type of ellipsis that removes a predicate and strands a finite auxiliary or a so-called semi-lexical verb. Hungarian has two frequently used auxiliaries, the habitual auxiliary szokott  and the future auxiliary fog  (Kenesei ). Semi-lexical verbs are verbs like akar ‘want’, szeretne ‘would like’, or modals like kell ‘need’ or lehet ‘may’. -stranding VP-ellipis can occur with all these items, in matrix or embedded contexts (including ACD contexts and comparative clauses as well): ()

a. Péter alszik, és én is Péter sleep. and I also ‘Péter is sleeping and I will, too.’

fogok .

b. Péter alszik, de nekem nem Péter sleep. but . not ‘Péter is sleeping but I don’t need to.’

kell need

aludni. sleep. aludni. sleep.

c. Többet aludtam ma, mint amennyit máskor szoktam aludni. more slept. today than how.much. otherwise . sleep. ‘I have slept more today than I usually do on other days.’ In (a) the tenses of the two clauses are not identical, which provides evidence that the elided category is smaller than a tense phrase. This conclusion dovetails with accounts that place finite auxiliaries in T0, such as Kenesei () and Surányi (a). In line with these accounts, -stranding predicate ellipsis is ellipsis of a vP constituent. While the examples in () contain overt subjects (nominal and dative), other types of phrases can also appear A-bar extracted out of the site of -stranding predicate ellipsis and occupy positions such as that of topic, contrastive topic, or focus/question word phrase before the auxiliary (a,b,c).

 ()

ˊ  a. Nem tudom, kivel beszéljek a problémámól. not know. who. talk.. the problem.. Te kivel szoktál? you who. . ‘I don’t know who to talk to about my problem. Who do you usually talk to (about yours)?’ b. Péterrel beszéltem, de Marival Péter. talked. but Mari. ‘I talked to Péter, but I won’t with Mari.’

nem not

fogok. .

c. János   vett. Mari is   fog. János a bike. bought. Mari also a bike. . ‘János bought a . Mari will also buy a .’ The constituents that line up before the auxiliaries in these examples move to specific positions in the left periphery of Hungarian. Te in (a) is a topic, kivel is a question word, Marival in (b) is a contrastive topic, and   in (c) is a contrastive focus constituent. The order of these items reflects the usual order of left peripheral elements in the language. Hungarian places focus, topic, and quantificational material in various ordered positions in the left periphery (giving rise to the often quoted discourse-configurationality of the language; É. Kiss ). The articulated left periphery, illustrated in (), houses topics (contrastive and non-contrastive, in TopPs), quantifiers (in DistPs as in Szabolcsi  or adjoined to FocP and TP as in Surányi  and É. Kiss ), and contrastive focus and wh-phrases (in a unique FocP, cf. É. Kiss , Brody , and Szabolcsi ).3 ()

[TopP topics / contrastive topic [DistP quantifiers [FocP focus/wh-phrase [TP

]]]]

That we are dealing with A-bar extraction out of predicate ellipsis in () and not (Englishtype) pseudogapping is evidenced by at least three observations. First, as (c) shows, the second remnant need not be contrastive with respect to a correlate in the antecedent (it is the same item egy biciklit ‘a bike’), unlike in English-type pseudogapping. Second, long-distance extraction is possible in examples of this type: see (), where the missing predicate corresponds to akar, hogy fölvegyünk, meaning ‘want that we hire’. Such long-distance dependencies cannot be established in English pseudogapping (Johnson b). ()

Azt tudom, hogy J´  kit akar, hogy föl that. know.  János who. want.   vegyünk. De azt nem tudom, hogy A kit fog. hire.. but that. not know.  Anna who. . lit. ‘I know who János wants us to hire. But I don’t know who A will (want us to hire).’

In one theory of focus placement (Brody ), specFocP harbors a single focused or wh-moved constituent of the clause. The verbal head raises up to the Foc head stranding its verbal modifier and creating obligatory adjacency between the raising head and the focused item. The verb does not raise any further than Foc. In other words, there is no head movement to Dist or Top in Hungarian. 3





Last but not least, examples like () can support a sloppy reading (cf. the interpretation of a), while pseudogapping does not support such a reading (see Johnson b and section ... for illustration).

.. V-stranding predicate ellipsis Predicate ellipsis in Hungarian can also exhibit a pattern of V-stranding (Bánréti ) similar to the one found in Finnish (Holmberg , ), Irish (McCloskey a), or Hebrew (Doron ), among other languages. There are two syntactic contexts in which V-stranding can rear its head in Hungarian: one is polarity contexts, such as yes/no questions and answers or confirmations to declaratives (cf. ), the other is a context with non-emphatic polarity, typically involving an is-phrase (also/too) before the stranded verb (cf. (); Surányi a,b). Interestingly, there is extensive microvariation concerning these two types of V-stranding: one dialect of Hungarian only allows for V-stranding in polarity contexts (call it variant A), another for both types (call it variant B).4 () A: Fel hívta Bea a szüleit  called. Bea the parent... ‘Did Bea call her parents yesterday?’

tegnap? yesterday

B: Fel hívta.  called. ‘She did.’ () Bea fel hívta a szüleit tegnap. Ibi is fel hívta. Bea  called. the parent... yesterday Ibi also  called. ‘Bea called her parents yesterday. Ibi also did.’ A quick comparison with patterns of pro-drop shows that the missing material in these examples can only be due to ellipsis of an entire verb phrase, and not to individual argument ellipsis. As example () shows,  objects (animate and non-animate alike) cannot be dropped in Hungarian. () Bea meg látta a szüleit az utcában. Üdvözölte *(őket). Bea  saw. the parents... the street. greeted. . ‘Bea saw her parents in the street. She greeted them.’

4 The precise geographical spread of this interspeaker variation is unknown. In a small-scale survey involving thirteen informants (dating back to ), the B speakers appeared to be from the Budapest area (see Lipták b for further details). Note that dialectal variation along the same lines is also reported to exist between Capeverdean and Portuguese as described in Costa et al. (): Capeverdean can only resort to V-stranding in polarity contexts, while Portuguese can do so both in polarity and nonpolarity contexts. See Costa et al. () for the facts and an account of the variation.



ˊ 

Based on this consideration as well as others, Lipták (b) argues that facts like (B/) involve predicate ellipsis: V-stranding VP-ellipsis strands the finite verb in T and elides the vP, including all arguments contained in there. Lipták (b) furthermore suggests that the microvariation concerning the availability of the two patterns identified above is due to variation in the licensing of V-stranding ellipsis in the two variants of Hungarian. While in variant B, V-stranding ellipsis can be licensed by finite tense, just like in -stranding predicate ellipsis that strands an auxiliary, in variant A V-stranding ellipsis is licensed by emphatic polarity only. The precise mechanism of licensing by emphatic polarity can be successfully modelled in the theory of ellipsis licensing in Aelbrecht (), where the licensor corresponds to a syntactic head that must c-command and establish an Agree relation with the head whose complement is elided. In polarity contexts, the licensor is a polarity head, Pol0 (similar to Laka’s  Σ) that selects the TP, and whose lexical content amounts to stress on the (preverb+)verb combination. In elliptical configurations, the T head hosts an ellipsis-specific feature [E] that brings about the non-pronunciation of the complement category. Via an agreement process between Pol and T, ellipsis is licensed. In variant B, V-stranding ellipsis is licensed in a local relation by T. ()

PolP

the Agree-based licensing of ellipsis in (22B)

Pol' TP

Pol0licensor PV feli

T' T0

hívtaj [E]

ellipsis of vP vP ti tj a szüleit tegnap

Further evidence for the long-distance Agree mechanism between an ellipsis licensor and the head that triggers ellipsis, can be provided from V-stranding phenomena in sentences containing verbal complexes. When a finite verb is followed by a series of infinitival complements, V-stranding ellipsis can strand any finite or non-finite verbal projection and yield exactly the same meaning. The following example illustrates the various acceptable patterns in B, B, and B.5

5

The descibed optionality in the size of V-stranding ellipsis in verbal complexes is only available under the so-called ‘straight order’ of verbal complexes (Koopman and Szabolcsi ; É. Kiss ). See an explanation of why it is missing in the ‘inverted’ orders in Lipták (b). Note also that the preverb appears left-adjacent to the finite auxiliary in these examples due to an ellipsis-independent process of what is referred to as preverb-climbing in the literature: a preverb belonging to an infinitive obligatorily appears before the finite auxiliary or semi-lexical verb that selects the infinitive. (i) Bea fel fogja hívni a szüleit. Bea  . call. the parent... ‘Bea will call her parents.’

 () A: Bea nem fogja akarni fel Bea not . want.  ‘Bea will not want to call her parents.’

hívni call.

a the

szüleit. parent...

B: De, fel fogja akarni hívni a   . want. call. the ‘That’s not right, she will want to call them.’

szüleit. parent...

B: De, fel fogja akarni hívni   . want. call. ‘That’s not right, she will want to.’

a the

szüleit. parent...

B: De, fel fogja akarni   . want. ‘That’s not right, she will.’

a the

szüleit. parent...

hívni call.



In examples like (B) or (B), the licensing category (Pol0) is non-adjacent to the ellipsis site, as a number of overt verbal projections (akarni hívni in B and akarni in B) intervene (and note that these verbal projections appear in the base-generated order). Such non-adjacency provides a strong argument for a long-distance approach to ellipsis licensing such as that of Aelbrecht ().

.. Preverb-stranding ellipsis Hungarian also exhibits a stranding-type ellipsis that does not strand the entire verb, but rather only the preverbal particle that combines with the verb. With reference to the frequently occurring instance of stranded preverbal particles, this type of ellipsis can be referred to as preverb-stranding ellipsis. Preverb stranding is allowed in both variant A and variant B of Hungarian and is strictly confined to polarity contexts. This type of ellipsis can occur in positive answers to polar questions and in affirmations to declaratives. It cannot occur in non-polarity contexts like () above; cf. (). () A: Fel hívta Bea a szüleit  called. Bea the parent... ‘Did Bea call her parents yesterday?’

tegnap? yesterday

B: Fel.  ‘She did.’ () Bea fel hívta a szüleit tegnap. *Ibi Bea  called. the parent... yesterday Ibi : ‘Bea called her parents yesterday. Ibi also did.’

is also

fel. 

The most straightforward account of preverb stranding would have it that it is structurally identical to V-stranding, except that the verbal head fails to raise out of the ellipsis site, cf. () (as v-to-T-movement is bled by ellipsis, as proposed in Lipták ; see also van Craenenbroeck and Lipták  for other bleeding effects of ellipsis on verb movement).

 ()

ˊ  [PolP [TP fel T0 [vP hívta a szüleit tegnap] ]]

There are, however, indications that () and the account offered in terms of bleeding are unlikely to be on the right track. A closer look at the data reveals that the syntactic distribution of V-stranding and that of preverb stranding is not fully identical. One crucial difference between the two is that while verb stranding can be used as a positive response to a negative yes/no question, preverb stranding cannot:6 () A: Nem hívta fel Bea a not called.  Bea the ‘Did Bea not call her parents?’ B: De, fel hívta.   called. ‘That’s not right, she did.’

szüleit? the parent... B: * De, fel.   : ‘That’s not right, she did.’

One way of explaining this contrast is to say that preverb stranding deletes more structure than just the vP projection, and in fact even more than the entire TP: it deletes the entire PolP that is standardly generated in answers and affirmations to polar questions. With the assumption that preverb stranding elides a full clause including the polarity specification, the ungrammaticality of (B) follows as a failure of identity between the elided constituent and its antecedent. To give a sketch of how this could potentially work, consider the following preliminary analysis. First, assume that PolP is the locus of both negative and positive polarity specifications in the clause and contains interpretable [Neg] and [Aff ] features in negative and positive clauses respectively (see independent evidence for these features in Lipták b). Assume furthermore that preverbs move to a position outside PolP in preverb stranding, a position that will be referred to as FP below.7 The complement of FP is affected

6

De is a sentential answer particle encoding the reverse function that indicates a relative switch of polarity (from negative to positive) with respect to an antecedent. See Farkas () for this element in Hungarian. 7 What motivates the placement of the preverb to the clause-initial FP projection is far from clear. Since the preverb is not a morphosyntactic category that is related to either polarity or affirmation/answerhood, it is unlikely that this movement is driven by a morphosyntactic feature. Note that the movement is sensitive to prosodic structure, specifically to accenting, as preverb–verb combinations have pitch accent on the preverb. See Dvořák () for the role that accenting plays in a similar ellipsis process in Slovenian. FP should be thought of as a position that hosts the sentential answer particles igen ‘yes’ and nem ‘no’ as well (possibly a polarity focus position of some sort). Independent justification for this comes from the similar distribution of igen and preverb-stranding in some contexts: e.g. neither a stranded preverb nor a single igen ‘yes’ can be used as a positive answer to an alternative question. As (i.B) shows, V-stranding can be used in this context. (i)

A: Fel hívta Bea a szüleit tegnap vagy nem hívta fel őket?  called. Bea the parent.... yesterday or not called.  . ‘Did Bea call her parents yesterday or did she not call them?’ B: Fel hívta. verb stranding B: *Fel. preverb stranding B: *Igen.  called.  yes ‘She did.’ ‘She did.’ ‘Yes.’





by deletion in preverb stranding, as shown in (). Importantly, this contrasts with ellipsis in V-stranding which affects a lower category, vP only. ()

a. [FP b.

 [PolP[TP[vP

]]] ]

[PolP [TP  V[vP

ellipsis in preverb stranding ]] ]

ellipsis in verb stranding

Using (), we can explain the pattern in (). V-stranding with a negative antecedent (cf. B) is grammatical, because V-stranding involves vP-deletion and the elided vP in (B) (marked by < >) is strictly identical to the vP in the antecedent (A). () A. [PolP [Neg] nem hívtaj [TP feli ti [vP Bea ti tj a szüleit ]]]? B. De, [PolP [Aff] [TP feli hívtaj ]]. The preverb-stranding answer in (B / B), however, is ill-formed since in this case a larger category is elided. Most importantly, the elided category contains the PolP which is featurally non-identical to its antecedent: it contains an affirmative feature while the antecedent has a negative one. ()

A. B. De, [FP feli
].

If this account is on the right track, the Hungarian facts in this section demonstrate that stranding-type ellipsis does not always elide vP predicates, but also higher projections in the clause. If this is correct, preverb stranding should be classified as a case of clausal ellipsis.

.. Pseudogapping As was mentioned in section .., -stranding predicate ellipsis can occur together with A-bar extraction, resulting in sentences that resemble but in fact do not instantiate Englishtype pseudogapping (see examples (a–c)). There are, however, examples that closely match the syntactic profile of English-type pseudogapping. These feature an auxiliary followed by a focal remnant that is contrastive with respect to a preverbal constituent. The postverbal and the preverbal constituents form an ordered pair whose thematic relation is the reverse of that in the antecedent (small caps stand for emphatic stress): () Balázs szokott nekem küldeni képeslapot. É is szoktam . Balázs . . send. postcard. I also . . ‘Balázs usually sends postcards to me. I also send him postcards usually.’ Note that the post-auxiliary remnant necessarily corresponds to a given constituent that ‘switches’ its argument position with respect to the antecedent. The pre-auxiliary constituent is normally a topic or an also-phrase. This shows that the ordered pair of constituents in the elliptical clause do not form a multiple focus construction (as such constructions always involve the preverbal constituent in focus position, see sections .. and ..). At the



ˊ 

same time, examples like () express pair-wise focus in the sense that the elliptical clause and its antecedent necessarily differ only in the order of the pairing relation between participants. This kind of focus will be referred to as reversing focus below.8 A clear indication that we are dealing with pseudogapping in () comes from (i) the fact that the final remnant is contrastive in the way defined above (see also n. ), (ii) the fact that long-distance extraction akin to () is impossible in these cases. Last but not least, the unavailability of sloppy identity readings in these constructions further supports the suspicion that we are dealing with an instance of English-type pseudogapping here. As Johnson (b) points out (attributing the observation to Chris Kennedy), pseudogapping does not license sloppy identity, while VP ellipsis does: ()

a. Fred gave flowers to his sweetie because Frank had.

(sloppy/strict)

b. Fred gave flowers to his sweetie because Frank had chocolates.

(strict only)

Exactly the same difference is observable in Hungarian. Consider first the case of VP ellipsis in (), which allows for both sloppy and strict readings: () Frici be mutatta Marit a barátnőjének. Tomi is be fogja. Frici  introduced Mari. the girlfriend.. Tomi also  . ✓ sloppy: ‘Frici introduced Mari to his girlfriend. Tomi will introduce Mari to Tomi’s girlfriend, too.’ ✓ strict: ‘Frici introduced Mari to his girlfriend. Tomi will introduce Mari to her, too.’ Pseudogapping on the other hand is incompatible with a sloppy reading, as the following example shows—note that the sloppy reading is unavailable even though this is the reading favored by the context, the strict reading being unlikely: () Ha Frici nem mutatja be a barátnőjének Tomit, if Frici not introduce.  the girlfriend.. Tomi. Tomi sem fogja ő! Tomi also.not . . *sloppy: ‘If Frici does not introduce Tomi to his girlfriend, Tomi will not introduce Frici to Tomi’s girlfriend, either.’ ✓strict: ‘If Frici does not introduce Tomi to his girlfriend, Tomi will not introduce Frici to her, either.’

8

The necessity of the inverse pairing relation is further indicated by the fact that examples like (i), where the post-auxiliary remnant does not repeat an already mentioned participant are judged as degraded by speakers. Some speakers report that they only accept such non-inverse examples if they understand the relation between the participants in the elliptical clause (i.e., me and Tamás), to be similar to the relation between the participants in the antecedent (i.e., me and Balázs). (i)

Balázs szokott nekem küldeni képeslapot. ?(?)Én is szoktam Tamásnak. Balázs . . send. postcard. I also . Tamás. ‘Balázs usually sends postcards to me. I also usually send Tamás postcards.’





Concerning the derivation of Hungarian pseudogapping, only speculations can be given at this point. Clearly, reversing focus never moves to the preverbal focus position (it is not an instance of exhaustive focus), and its position in the postverbal domain is rather free: it can occur in various positions, showing a slight preference for the clause-final one, as the following non-elliptical versions of () show. () Balázs szokott nekem küldeni képeslapot. Én is szoktam Balázs . . send. postcard. I also . {?} küldeni {?} képeslapot {}. . send. . postcard. . ‘Balázs usually sends postcards to me. I also send him postcards usually.’ The elliptical variant in () can be derived assuming that the focal remnant undergoes short A-bar movement to end up right before the infinitive. In that position, the infinitival VP undergoes ellipsis as in (). This account is compatible with analyses of pseudogapping in the literature that postulate a (short) A-bar movement process to the left (Jayaseelan ; Gengel , among others). () Én I

is also

szoktam .

i .

[ küldeni send.

képeslapot ti ] postcard.

Finally, it can be noted that the same type of reversing focus can also survive ellipsis of a predicative constituent out of which V-movement has taken place, i.e. pseudogapping can also take place when accompanied by V-stranding: () Balázs mindig küld nekem képeslapokat. Én is küldökj Balázs always send. . postcard.. I also send. i [ tj képeslapokat ti]. . postcard.. lit. ‘Balázs always sends postcards to me. I also send him (postcards).’

. C 

.................................................................................................................................. Hungarian, like many other languages, exhibits various instances of clausal ellipsis, such as sluicing, stripping, and fragments. All these types involve TP-ellipsis and as such are characterized (and can be identified) by strong tense-identity effects. The tense of the elliptical clause is always identical to that of the antecedent clause in clausal ellipsis. Observe this in the following instance of stripping. ()

Péter táncolt, és Mari is táncolt / * táncol. stripping Péter danced. and Mari also dance.. dance.. ‘Péter danced and Mari, too.’



ˊ 

The following three subsections review single-remnant and multi-remnant clausal ellipsis (including gapping) in Hungarian.

.. Single-remnant clausal ellipsis and the theory of ellipsis licensing Remnants of clausal ellipsis move to the structurally rich left periphery (see ()). As a direct result of Hungarian’s rich left periphery, clausal ellipsis exists with various remnant constituents (Bánréti , ), such as is-phrases in stripping (cf. ()), wh- and focus constituents in sluicing and fragments (cf. ()), and focus constituents in fragments (cf. ()). () A: Péter táncolt tegnap egy Péter danced. yesterday a ‘Péter danced with a girl yesterday.’

lánnyal. girl.

B: Kivel? who. ‘With who?’

sluicing

B: Igen, B´. yes Bea. ‘Yes, with Bea.’

(non-contrastive) fragment

() A: Péter  táncolt Beával. Péter yesterday danced. Bea. ‘It was yesterday that Péter danced with Bea.’ B: Nem, ő. no yesterday.before ‘No, the day before yesterday.’

(contrastive) fragment

Sluicing and fragments with single remnants share identical derivations, including the position targeted by the remnants in them (Lipták ; Griffiths and Lipták )— which is not surprising given that wh-constituents and preverbal focus occupy the same structural position, FocP, in non-elliptical sentences as well. The structural similarity extends to embedded contexts as well. In sluicing contexts (where an ellipsis remnant corresponds to an indefinite in the antecedent clause), TP-ellipsis can have a wh- as well as a focus remnant, as example () shows (Horvath ; van Craenenbroeck and Lipták )—this example can also be treated as a case of a genuine embedded fragment.9

Note that the embedding verb tud ‘know’ used in () cannot introduce embedded fragments in other languages such as English, where only parenthetical embedded fragments are allowed. These can only feature verbs like think, expect, or hope (Morgan ; Temmerman ): 9

(i)

A: Peter danced with someone. B: I {✓ think / ✓ expect / ✓ hope / * know } with Bea.





() János meghívott egy lányt, de nem tudtam, hogy {kit / A´ }. János invited. a girl. but not knew.  who. Anna. ‘János invited a girl, but I didn’t know {who / that it was Anna}.’ This is in line with the fact that wh-questions and focus constructions make use of the same syntax and occupy the same position in the left periphery. They are in complementary distribution in main clauses and trigger the separation of the preverbal particle from the verb in exactly the same way: () A: Kit hívott who. called. ‘Who did Bea call?’

fel 

B: A ̈  the parent... ‘Bea called her .’

Bea? Bea hívta called.

fel 

Bea. Bea

In fact, TP-ellipsis is perfectly well-formed in embedded contexts (just as it is in fragments) with fronted emphatic operator constituents such as universal quantifiers (whose position is just above FocP): () Tudtam, hogy Péter táncolt tegnap egy pár lánnyal, de nem knew. that Péter danced. yesterday a couple girl. but not tudtam, hogy minddel. knew.  all. ‘I knew that Péter danced with some girls yesterday, but I didn’t know he danced with all.’ () and () have important repercussions for the theory of ellipsis as they show that whmovement is not a distinctive trait of sluicing. Rather, ellipsis licensing is sensitive to the type of feature the wh-phrase checks in overt syntax, which in Hungarian happens to be an operator feature. Based on data like () and () in Hungarian and other languages, van Craenenbroeck and Lipták () suggest that sluicing tracks wh-syntax in this sense across languages: the feature content of wh-elements in non-elliptical questions determines what kind of remnants can escape TP-ellipsis in sluicing, referred to as the wh/sluicing correlation:10

10 The []-feature in () should be understood as the specific notation introduced by Merchant (: –, a: –). Merchant argues that ellipsis in sluicing should be implemented by means of a syntactic []-feature, which resides on C0 and has all the relevant properties that distinguish elliptical structures from their non-elliptical counterparts, summarized in (i):

(i)

a. the syntax of []: [uwh*,uQ*] b. the phonology of []: φ ! Ø /  __ c. the semantics of []: [[  ]] = λp : e- (p) [p]

Van Craenenbroeck and Lipták () argue that the syntax of [] can vary across languages.



ˊ   /  (van Craenenbroeck and Lipták ) The syntactic features that the []-feature checks in a certain language are identical to the strong features a wh-phrase checks in a non-elliptical constituent question in that language.

()

The reach of () interestingly extends to the domain of relative clauses as well in Hungarian: relative pronouns, which arguably have operator features just like question words, allow for ellipsis of their complement in free relatives and pronominally headed relatives (Lipták )—contradicting the received opinion that TP-ellipsis stranding a relative operator should be impossible (van Riemsdijk , Lobeck , and Merchant ). () A rovaroknak mindig  pár lábuk van, a százlábúaknak meg the insects. always  pair foot.. is the millipedes.  annyi, [RC amennyi van nekik ]. that.many .how.many is . ‘Insects always have  pairs of feet. Millipedes on the other hand have however many they have.’ Evidence for clausal ellipsis in () is culled from various domains in Lipták , including the observation that the annyi-amennyi pronominal+relative pronoun complex occurs only in syntactic positions where one would expect a relative clause, and the fact that () exhibits characteristic properties of antecedent-contained sluicing (Yoshida ), such as tense and modality mismatches between the elliptical and the antecedent clause. () Pénzügyi válság nélkül nem tartanánk ott, [RC ahol tartunk ] financial crisis without not be... there .where be.. a. ‘Without the financial crisis we would not be where we are.’ b. *‘Without the financial crisis we would not be where we are not.’ Sluicing in this type of relative clause furthermore has a characteristic (cross-linguistically rare) prosodic profile that is unlike the prosodic profile of non-elliptical relatives.11

11 The relative pronouns in examples like () and () must carry stress (in line with Sprouse ; Sáez ). The latter is not possible for relative pronouns in non-elliptical clauses; compare the nonelliptical (i) and the elliptical (ii), where ▾ indicates lexical accent and 0 lack thereof. ▾

▾

megcsókolta ▾azt, kissed. that.

▾ akit megcsókolt. .who. kissed. 0

(i)

Mindenki everyone

(ii)

▾ ▾ Mindenki ▾megcsókolta ▾azt, akit. everyone kissed. that. .who. ‘Everyone kissed whoever he/she kissed.’

The stress pattern in (ii) is independently attested in Hungarian in environments not featuring clausal ellipsis: it is also found in relative clauses that contain no other overt material but the relative pronoun due to pro-drop and copula drop. This is possible in sentences that relativize an AP or NP predicate and where both the subject and the  present tense copula are zero, such as (iii).





.. Multiple-remnant clausal ellipsis and ellipsis repair Just like many languages, Hungarian can also form clausal ellipsis stranding multiple remnants. This is not surprising in the domain of wh-syntax, since it has been known since É. Kiss (, ) that Hungarian allows for multiple wh-movement to the left periphery. In fact, there is evidence from interpretation that multiple sluicing can only be formed via the derivational route of ordinary multiple wh-movement and no other. As Grebenyova (, ), and following her, van Craenenbroeck and Lipták (b) observe, multiple wh-sluicing only allows for interpretations that are also available for multiple wh-movement as well. Hungarian multiple wh-movement only supports a pair-list reading (É. Kiss ), cf. (), and multiple wh-sluices are only possible in contexts that support such a reading, too. Thus they are ruled out in single-pair contexts such as (), but are well-formed in contexts such as () where a distributive reading is established. () Ki mit vett Balázsnak? who what. bought Balázs. ‘Who bought what for Balázs?’ a. ✓Everyone bought something for Balázs. I wonder what each person bought for him. b. *A single person bought something for Balázs. I wonder who the person was and what he bought for him. ()

Valaki vett valamit Balázsnak. #Nem tudom, someone bought. something. Balázs. not know. ki mit. who what. ‘Someone bought something for Balázs. I don’t know who what.’

hogy 

() Mindenki vett valamit Balázsnak. Nem tudom, hogy ki everyone bought. something. Balázs. not know.  who mit. what. ‘Each person bought something for Balázs. I don’t know what each person bought.’ In the domain of focus, however, the derivational options of multiple focus under ellipsis do not run parallel to the choices of overt focus movement in Hungarian: multiple focus remnants are allowed notwithstanding the fact that multiple focus fronting in non-elliptical contexts does not exist. In other words, two adjacent foci are only possible when followed by ellipsis:

(iii) ▾Fogadjuk el ▾olyannak, ▾amilyen Øpro accept..  such. .what.kind he ‘Let’s accept him/her the way he is.’

Øcop. is

 ()

ˊ  Mari azt akarja, hogy adjak valamit valakinek. Mari that. want.  give.. something. someone. Úgy emlékszem, hogy  ̈  ´  (*adjak) so remember.  a book. Bea. give.. ‘Mari wants me to give something to someone. What I remember is that (she wants me to give) a book to Bea.’

One possible way of thinking about this apparent breakdown between ellipsis and nonelliptical syntax is to consider it to be a PF-repair effect of ellipsis, as proposed in van Craenenbroeck and Lipták (b). The proposal is based on the assumption that multiple focus movement in Hungarian triggers a PF-violation, namely that non-final focus constituents are not adjacent to the verb (recall from n.  that there is obligatory adjacency between the lexical verb and the fronted focused constituent). () a. * a b.

̈  book.

´  Bea.

adott gave

János. János

 ̈  adott János  ´ . a book. gave János Bea. : ‘J´  gave a book to B.’

no focus-V adjacency for the initial focus focus-V adjacency observed for the initial focus

If the observed focus-verb adjacency is a phonological requirement, it is possible to attribute the repair effect of ellipsis to the fact that it removes the verb in clausal ellipsis and thus eliminates the confounding factor altogether: if there is no verb, it does not have to be adjacent to anything and multiple focus movement is allowed as in ():12 ()

. . . [FocP  ̈  [FocP ´  [TP adott János  ̈  ´ ]] ]

Finally it must be mentioned that multi-remnant TP-ellipsis can also strand constituents with distinct discourse functions. It can exhibit a combination of topics, quantifiers, and focus constituents, in the order allowed in non-elliptical sentences, i.e., topic < quantifiers < focus. For illustration, consider the following contrastive fragments in (B) and (B). B involves a quantifier followed by a focus, B has a contrastive topic followed by a focus. () A: Mindenkinek ´ telefonáltál? everyone. twice called. ‘Did you call everyone twice?’ B: Nem, mindenkinek ´ . no everyone. three.times lit. ‘No, everyone three times.’ (‘No, I called everyone three times.’) 12

See van Craenenbroeck and Lipták () for the specific technical implementation: multiple focus movement does take place in narrow syntax but it has to spell out the lower copy of Beának in (b). Note also that the movement of the verb from T0 to Foc0 (see n. ) is bled by clausal ellipsis taking place in (), similarly to (), (), (). See Horvath () and van Craenenbroeck and Lipták () for details on bleeding.





B: Nem, a lányoknak ´ . no the girls. three.times lit. ‘No, the girls three times.’ (‘No, I called the girls three times.’)

.. Multiple remnants in gapping As Bánréti (, ) has pointed out, gapping-like elliptical constructions have two basic types in Hungarian: cases in which the final remnant is a contrastively focused constituent and cases in which the final remnant is not focal, but instead the elliptical clause (and its antecedent) corresponds to an information-structurally unmarked sentence. Starting with the former type, consider (), which illustrates gapping with two remnants, a contrastive topic (én) and a preverbal focus (ma). () Péter  táncolt, én pedig Péter yesterday danced I  ‘Péter danced yesterday and I today.’

. today

As the reader can ascertain, this kind of example resembles the cases in () in the sense that it features left peripheral constituents arranged in the expected order. Accordingly, the ellipsis here corresponds to ordinary TP-ellipsis, and thus does not in fact instantiate English-type gapping. One strong reason to think so is that unlike English gapping, the elliptical clause here can be embedded with respect to its antecedent—cf. the following example, adapted from Bánréti () (see Farudi  for similar facts in Farsi.) () Az elsősök ´  akarnak vizsgázni, a the st.year.student. syntax. want. write.exam. the másodikosok meg azt mondják, hogy ők inkább ´ . nd.year.student.  that. say.  they rather morphology. ‘The first year students want an exam in syntax, and the second year students say that they rather (want an exam) in morphology.’ A distinct subtype of gapping with focal remnants is illustrated in (). This contains two focus remnants, corresponding to the two members of a complex multiple focus construction (as defined in Krifka , corresponding to focus on an ordered pair). As reflected by word order in the antecedent clause (and the preference for parallelism in structure), both members correspond to focused entities: the first has a preverbal correlate and the second a postverbal one. Note that this kind of gapping cannot contain the elliptical clause embedded with respect to its antecedent, thus more closely resembling gapping in other languages (b): () a. Az őö választották a , a ´  the st.year.student. chose. the syntax. the nd.year.student. pedig az .  the morphology. ‘It was the  year students who chose , and it was the  year students who chose .’



ˊ  b. *Az őö választották a , a másodikosok the st.year.student. chose. the syntax. the nd.year.student. pedig azt mondták, hogy ő az .  that. said.  they the morphology. ‘It was the  year students who chose , and the  year students said that it was  who chose .’

The derivation of these constructions is arguably the same as that presented for multiple focus remnants in section ... Multiple focus remnants are allowed in the left periphery in these contexts due to elliptical repair: two focal remnants normally cannot both front as they would violate the focus–verb adjacency rule. Under ellipsis, this rule is vacuously satisfied and both constituents can undergo fronting: ()

. . . [FocP  ´  [FocP   [TP  ´  választották  ]]]

The least understood pattern of gapping is gapping with remnants that are non-focal in nature, which are anteceded by an information-structurally unmarked clause. The following examples, adapted from Bánréti , illustrate this pattern. As Bánréti notes, the remnants normally correspond to arguments rather than adjuncts, and the number of remnants can be higher than two. () a. Misi megvitatott egy forgatókönyvet, Robi pedig egy novellát. Misi discussed. a script. Robi  a short.story. ‘Misi discussed a script and Robi a short story.’ b. Misi megvitatott egy forgatókönyvet Erzsivel, Robi pedig egy Misi discussed. a script. Erzsi. Robi  a novellát Valival. short.story. Vali. ‘Misi discussed a script with Erzsi and Robi (discussed) a short story with Vali.’ Since in this construction type the antecedent is an information-structurally unmarked clause, and the remnants do not have any specific discourse functions, one would be inclined to think that the missing finite verb is elided in situ as a result of non-constituent deletion. () Robi Robi

pedig 

megvitatott discussed.

egy a

novellát short.story.

Valival. Vali.

Given, however, that Hungarian does not show evidence for such in-situ deletion in other contexts, the account in () is unlikely.





. R- 

.................................................................................................................................. Right-node raising (RNR) in Hungarian is an attested phenomenon (Bánréti , , ), as the following examples (adapted from Bánréti  and Surányi b respectively) show. The shared pivot is enclosed in brackets in these examples. () Péter a tíz- Mari pedig a húsz-[betűs szavakat kereste meg]. Péter the ten Mari  the twenty-letter word.. searched.  ‘Péter looked up the words with ten letters, and Mari looked up the ones with twenty letters.’ () Neked el küldi, nekem fel hozza [valaki a leveleket you.  send. .  bring. someone the letter.. a portáról]. the reception. ‘Someone sends the letters to you from the reception desk, whereas someone brings them up to me.’ As () shows, RNR need not observe syntactic constituency, which supports the conclusion that this example of RNR is not derived by movement. Importantly, Hungarian only allows for RNR in which the pivot corresponds to nonfocal material. Focal pivots that represent contrastive focus or the answer to a wh-question cannot be formed; consider () and (): () Context: Apparently, Viktor read and Bea translated Crime and Punishment. * Nem, Viktor el olvasta, Bea pedig le fordította [ ´´ ´ ´´]. no Viktor  read. Bea   translated. War and Peace. : ‘No, Viktor read and Bea translated War and Peace.’ () Context: Which book did Tomi read and Bea translate? * Tomi el olvasta, Bea pedig le fordította [ ´ ´ ´  ´ ´]. Tomi  read. Bea   translated. War and Peace. : ‘Tomi read and Bea translated War and Peace.’ The fact that focal pivots are disallowed in Hungarian supports the generalization by Valmala () according to which RNR with focal and RNR with non-focal pivots have distinct information-structural, prosodic, and syntactic properties (see also Hirsch and Wagner ). If, as Valmala claims, focal-pivot RNR is derived via rightward movement of the focal chunk in languages like English, where () and () are grammatical, the impossibility of this kind of RNR in Hungarian can be understood with reference to the fact that Hungarian lacks rightward movement processes for the expression of focus.



ˊ 

. E   

.................................................................................................................................. Hungarian comparative clauses may contain unpronounced elements. The most important distinction between comparative clauses in Hungarian and those in English in this domain is that a degree phrase does not have to be eliminated in comparative clauses via comparative deletion in Hungarian: such a phrase may be covert (a), or overt (b) (Kenesei ). When it is overt, the nominal or adjectival degree expression always contains an overt operator (a relative wh-operator), as shown in (b). () a. Mari Mari

több more

macskát cat.

vett, bought.

mint than

Péter. Péter

b. Mari több macskát vett, mint *(ahány) macskát Péter Mari more cat. bought. than .how.many cat. Péter vett. bought. ‘Mari bought more cats than Péter.’ When overt, the degree expression must appear in the left periphery of the comparative clause, it cannot be left in situ (cf. ()). This follows straightforwardly as Hungarian does not allow in situ relative pronouns, either. () * Mari több macskát vett, mint Péter vett ahány macskát. Mari more cat. bought. than Péter bought. .how.many cat. ‘Mari bought more cats than Péter.’ Bacskai-Atkari (, ) claims that the non-existence of covert degree operators in the left periphery in Hungarian and the lack of an obligatory process of comparative deletion are interrelated. She argues that comparative deletion only affects degree expressions in the left periphery, featuring covert operators, and thus it never applies to Hungarian as in this language the degree operator is overt. She also argues that cases where the entire degree constituent is missing (cf. (a)) are derived by ordinary ellipsis processes that are operative in Hungarian clauses in general, such as predicate ellipsis, sluicing, or gapping (cf. () for the latter two cases). In these examples the degree phrase stays in situ in Bacskai-Atkari’s analysis, and gets eliminated due to it being inside an ellipsis site. The obligatoriness of ellipsis is supported by the observation that in cases where the degree phrase is covert, a given verb cannot be pronounced, as can be observed in (a).13 13 Non-given verbs and verbs expressing a tense distinction with respect to the antecedent can survive the ellipsis.

(i)

Mari több könyvet vesz holnap, mint Péter Mari more book. buy. tomorrow than Péter ‘Mari will buy more books tomorrow than Péter did last year.’

vett tavaly. bought. last.year

 () a. Mari több könyvet vett, mint Mari more book. bought than ‘Mari bought more books than Péter did.’

Péter Péter



(*vett). bought.

b. Mari több könyvet vett a fiának, mint Mari more book. bought. the son.. than a lányának. the daughter.. ‘Mari bought more books for her son than Péter did for her daughter.’

Péter Péter

. N  

.................................................................................................................................. Hungarian null complement anaphora (NCA) are difficult to detect for the reason that many verbs that are known to allow NCA complements in other languages take complements in Hungarian that can be missing as a result of pro-drop. Verbs with nominative or accusative nominal complements or finite clausal complements associated with nominative and accusative sentential pronouns are such: their complements, which are headed or spelled out by definite pronouns can undergo the regular process of subject and object prodrop (see Kenesei ). Null complement anaphora can be evidenced among verbs that normally select prepositional complements, however. These are verbs that take nominal complements marked with oblique case, or finite clause complements associated with oblique sentential pronouns. These types of complements cannot undergo pro-drop in the language, so the lack of an otherwise obligatory complement is telling in these cases. Verbs where such complements can be missing, and which therefore allow for NCA are e.g. vállalkozik ‘volunteer’ (with a sublative PP/finite clause complement), ajánlkozik ‘offer’ (with a sublative PP/finite clause complement), jelentkezik ‘sign up’ (with a sublative PP/finite clause complement), csodálkozik ‘wonder’ (with a superessive PP/finite clause complement), emlékezik ‘remember’ (with a sublative PP/finite clause complement), or egyetért ‘agree’ (with an instrumental PP complement). The following two examples illustrate the use of these verbs without overt complements. () Kérdeztük, ki viszi el a virágot Beának. Balázs vállalkozott. asked. who bring.  the flower. Bea. Balázs volunteered. ‘We asked who might bring the flower to Bea. Balázs volunteered.’ ()

Javasoltuk, hogy az igazgató nyissa meg a konferenciát. suggested.  the director open..  the conference. Mindenki egyetértett. everyone agreed. ‘We suggested that the director should open the conference. Everyone agreed.’

Note that extraction cannot be used as a test for the null complement analysis in these cases, as extraction from the complements of these verbs is disallowed anyway. Extraction



ˊ 

can be used as a test for predicates selecting infinitival complements, verbs such as akar ‘want’, mer ‘dare’, szeretne ‘like’, hajlandó ‘be willing’, etc. on the other hand. Interestingly, this test reveals that the complements of this class of verbs do allow for extraction, which in turn shows that these predicates do not take null anaphors as complements. () Mondd el, melyik filmet akarod meg nézni, és melyiket tell..  which film. want.  watch. and which. nem akarod. not want. lit. ‘Tell me which film you want to watch and which you don’t want (to watch).’

. C

.................................................................................................................................. This chapter has illustrated the major types of elliptical constructions in Hungarian that have been identified and analyzed in some detail in previous works. It was shown that Hungarian has a wealth of elliptical constructions. Many of these constructions are derived by leftward movement of a syntactic constituent and ellipsis of (a subpart of) the domain out of which movement has taken place. The existence of sluicing inside relative clauses and preverb stranding furthermore points to the important role prosody plays in the formation of elliptical clauses.

A I thank the editors and three anonymous reviewers for their spot-on remarks on the previous version of this chapter. All mistakes and discrepancies are my own responsibility. This research was supported by NWO (Netherlands Organization for Scientific Research) in the form of the Vrije Competitie grant ‘Ellipsis licensing beyond syntax’.

  ......................................................................................................................

 ......................................................................................................................

 

. I

.................................................................................................................................. T chapter provides an overview of ellipsis phenomena in Indonesian, a MalayoPolynesian (Austronesian) language. Unlike some other well-studied Malayo-Polynesian languages, such as Tagalog and Malagasy, Indonesian employs SVO constituent order and is nominative-accusative. The chapter is organized as follows. In ., I describe sluicing, the type of Indonesian ellipsis that has received the most attention in the literature to date (Fortin b; Sato ; b). Subsequent sections address fragment answers (.), verb phrase ellipsis (.), pseudogapping (.), gapping (.), stripping (.), comparative deletion (.), right-node raising (.), conjunction reduction (.), null complement anaphora (.), and nominal ellipsis (.). With the possible exceptions of right-node raising and comparative deletion, Indonesian displays all of these phenomena to some extent. A notable feature of Indonesian ellipsis is that it permits prepositions to be omitted in some elliptical contexts, despite preposition stranding being prohibited in non-elliptical contexts. This is unexpected, given the otherwise broadly accurate Preposition Stranding Generalization (henceforth, PSG) (Merchant ), which links a language’s ability to omit a preposition under ellipsis to its ability to strand a preposition via extraction of its complement. Phenomena affected, which include sluicing, fragments, pseudogapping, stripping, and gapping, are considered in .. Section . concludes.

. S

.................................................................................................................................. Indonesian displays sluicing, under which an interrogative clause appears in a phonologically reduced form. In most respects, Indonesian sluicing resembles sluicing in a language like English. The sole remnant—i.e. the phonologically overt material—in the sluiced clause is the wh-phrase, which is taken to reside in a left-peripheral position, such as a specifier within the complementizer phrase (CP) layer. The remainder of the clause—the tense phrase (TP)—is phonologically null under identity with an antecedent.





The wh-remnant may have an overt correlate in the antecedent (), but is not required to (); the latter type of wh-remnants are termed “sprouted” (Chung, Ladusaw, and McCloskey ). The wh-remnant may be an argument () or an adjunct (), and both syntactically complex wh-phrases () and bare wh-phrases, as in the other examples, are permissible. As in other languages, there is a ban on overt complementizers in sluicing (), although the complementizer yang is mandatory in the corresponding non-elliptical question. (The role of yang is addressed in ..) () Saya tahu Ali membeli sesuatu, tapi SG know Ali buy something, but saya tidak tahu apa (*yang) Ali beli. SG NEG know what (*REL) Ali buy ‘I know Ali bought something, but I don’t know what.’1, 2 () Polisi menangkap Dewi, tapi dia tidak tahu police arrest Dewi, but SG NEG know ‘The police arrested Dewi, but she didn’t know why.’

kenapa why

() Ali membeli buku, tapi saya tidak tahu (buku) Ali buy book, but SG NEG know (book) ‘Ali bought a book, but I don’t know which (book).’3

yang mana. which

Additionally, island violations are “ameliorated” under sluicing (Ross b): a wh-remnant can have a correlate within an island, such as a relative clause (), although the corresponding non-elliptical wh-question is ill-formed. () a. Siti lihat orang yang punya banyak anak. Siti see person REL has many child Coba tebak berapa anak! try guess how.many child ‘Siti saw a person who has many kids. Try to guess how many kids!’ b. *Berapa anak yang Siti lihat orang how.many child REL Siti see person ‘*How many kids did Siti see a person that has?’

yang REL

punya? has

For some speakers, multiple sluicing is permitted, although Indonesian is not a multiplewh-fronting language. (An example, with discussion, is provided in ().)

1

Strikeouts indicate the interpretation of the elided material. No claims about the precise content of the ellipsis site are intended. 2 The following abbreviations are used in examples: ACT, active voice; CLASS, classifier; COMP, complementizer; FUT, future tense; INCL, inclusive; NEG, negation; PASS, passive; PL, plural; PROG, progressive; Q, question; REL, relativizer; SG, singular; //, person. 3 Following (e.g.) Merchant , I assume that the optionality of buku ‘book’ involves a second type of ellipsis internal to the DP.





Despite these parallels with sluicing in other languages, Indonesian sluicing also displays characteristics which set it apart. Indonesian does not allow so-called “contrast sluices,” which are frequently attested in other languages (). () Enam orang pelajar pergi ke pesta, tapi saya six CLASS student go to party, but SG tidak tahu berapa orang guru *(yang pergi). NEG know how.many CLASS teacher (REL go) ‘Six students went to the party, but I don’t know how many teachers (went).’ Sluicing is also well-known to be subject to various kinds of morphosyntactic “connectivity” effects, which include morphological case matching and the PSG (). ()

Preposition Stranding Generalization A language L will allow P-stranding under sluicing iff L allows P-stranding under regular wh-movement. (Merchant : )

Indonesian lacks morphological case marking entirely, and diverges from the norm regarding the PSG. When the remnant wh-phrase is the complement of a preposition, and there is a correlate PP in the antecedent, the preposition can be omitted from the remnant (), although preposition stranding is prohibited in non-elliptical questions (). () Ali berbicara dengan seseorang, tapi saya tidak Ali talk with someone but SG NEG ‘Ali spoke with someone, but I don’t know (with) who.’ () a. *Siapa (yang) Ali sedang berbicara who (REL) Ali PROG speak ‘Who is Ali speaking with?’ b. Dengan siapa Ali sedang with who Ali PROG ‘With who is Ali speaking?’

tahu know

(dengan) (with)

siapa. who

dengan? with

berbicara? speak

The acceptability of () is surprising, as the PSG predicts this pattern to be impossible (Fortin b; Sato , , b). I return to this issue in greater detail in section ., and use it to further explore the structure of Indonesian questions and the derivation of sluices.

. F 

.................................................................................................................................. Indonesian permits fragment answers, including determiner phrases (DPs) (), PPs (), and certain inflectional markers, including modal auxiliaries ().

 ()

 Q: Siapa yang menelpon who REL call ‘Who phoned just now?’

tadi? just.now

A: Ali. () Q: Dengan siapa Ali berdansa? with who Ali dance ‘With who did Ali dance?’ A: (Dengan) ibu=nya. (with) mother=SG ‘(With) his mom.’ ()

Q: Boleh=kah saya ke pesta? May=Q SG to party ‘May I go to the party?’ A: (Nggak) boleh. (NEG) may May (not).’

Merchant a (see also Temmerman , and Sato , for Indonesian) proposes fragments are derived via movement of the fragment to a clause-peripheral focus position, followed by deletion of the remainder of the clause. Such an analysis accounts for the connectivity effects observed in Indonesian fragments (Sato ), which include certain island effects, including adjuncts and relative clauses (). ()

Q: Kamu kenal [perempuan yang berbicara SG know [woman REL speak ‘Do you know a woman who speaks Korean?’ A: *Nggak, bahasa NEG language ‘No, Japanese.’

bahasa language

Korea]? Korea]

Jepang. Japan

Certain other island effects are amnestied in fragments. Possessor DPs are legal fragments (), although possessor DPs cannot be extracted (). ()

Q: Jaket siapa yang kamu pinjam? jacket who REL SG borrow ‘Whose jacket did you borrow?’ A: Kakak saya. older.sibling SG ‘My sibling’s.’

()

*Siapa yang kamu pinjam jaket? Who REL SG borrow jacket ‘*Whose did you borrow jacket?’





Under the above analysis, () requires kakak saya ‘my sibling’ to focus-front independently of the DP containing it. Merchant () argues, of sluicing, that “left-branch condition” violations (Ross ) like () are repaired by deletion, in contrast to the island violation resulting from extraction out of a relative clause (), which is not. If so, the Indonesian data are consistent with a focus-fronting analysis of fragments. Alternative analyses of fragments that require neither ellipsis nor a fully clausal source (e.g., Barton and Progovac ; Fortin a) do not immediately capture the connectivity effects.

. V  

.................................................................................................................................. Indonesian displays auxiliary-stranding VPE, under which a constituent containing a verb and any complements/adjuncts is realized as phonologically null under identity with an antecedent. The elided constituent must be preceded by a licensing element, an overt inflectional marker, such as bisa ‘can’ in (). VPE is licensed in coordinate structures (), adjuncts (), and embedded clauses (). ()

Saya tidak bisa pergi ke toko sekarang, SG NEG can go to store now, ‘I can’t go to the store now, but maybe you can.’

() Ali akan mencuci piring karena Siti tidak Ali FUT wash dish because Siti NEG ‘Ali will wash the dishes because Siti cannot.’ ()

Ali tidak bisa makan daging, tapi Ali NEG can eat meat but ‘Ali can’t eat meat, but Dewi said Siti can.’

Dewi Dewi

tapi but

mungkin maybe

kamu SG

bisa. can

bisa. can

bilang say

Siti Siti

bisa. can

Voice alternations are not permitted, whether the antecedent clause is passive and the target of ellipsis is active (), or vice versa. This pattern indicates the ellipsis site contains the voice head (Merchant a). ()

*Rumah=ku belum di-bersihkan, jadi saya harus house=SG not.yet PASS-clean so SG must ‘My house hasn’t yet been cleaned, so I have to *(clean it).’

The question of which inflectional markers license VPE is somewhat complicated. Markers of tense (), and certain markers of modality () and aspect (), can appear string-adjacent to the elided constituent (Fortin b). All root modals—e.g. bisa ‘can’; harus ‘must’—allow this; no epistemic modals—e.g. pasti ‘be certain’; mungkin ‘be possible’—do (Fortin ) ().

 ()

 Q: Siapa yang bakal Who REL FUT ‘Who will help Siti?’ A: Saya (tidak) SG (NEG) ‘I will (not).’

membantu help

Siti? Siti

bakal. FUT

() Ibu=ku sering tertidur sewaktu menonton filem, tapi mother=SG often fall.asleep one.time watch film but tidak pernah. NEG ever ‘My mom often falls asleep watching movies, but I never (do).’ ()

saya SG

*Dia bilang dia tidak capai, tapi dia {pasti/ mungkin}. SG say SG NEG be.tired but SG {must/ might} : ‘She says she’s not tired, but she must/might (be).’

This root/epistemic split parallels one previously observed in Dutch Modal Complement Ellipsis (MCE) (Aelbrecht , ) and Romance MCE (Dagnac ). These languages differ from Indonesian in two important respects. First, root modals are the only possible licensers for ellipsis in these languages, while tense and some aspectual markers in Indonesian can also immediately precede the ellipsis site. Second, the root modals in the European languages have been analyzed as raising verbs (contra Ross’s a influential proposal, under which epistemic modals are raising verbs, and root modals control verbs). Indonesian modals are not, in contrast, raising predicates. One diagnostic involves truth conditions: a passivized complement of a raising predicate is truthconditionally equivalent to its active counterpart. Indonesian modals fail this diagnostic; (a) and (b) are not truth-conditionally equivalent, as (b) is a statement of Ali’s ability to be fired (Paul Kroeger, p.c.), rather than a statement of his boss’s ability to fire him. () a. Bos Ali bisa memecat=nya. Bos Ali can fire=SG ‘Ali’s boss can fire him.’ b. Ali bisa dipecat oleh bos=nya. Ali can PASS-fire by boss=SG ‘Ali can be fired by his boss.’ Consistent co-occurrence patterns of inflectional markers demonstrate that tense, aspect, and modality are not conflated (). () Ali pasti akan bisa membaca buku Ali certain FUT can read book ‘Ali certainly will be able to read that book.’

itu. that





Certain aspectual markers—e.g. the progressive markers sedang and lagi—precede tense, while others—e.g. the completive markers sudah and telah—follow it, indicating that there are two aspect-related projections sandwiching T. Similarly, two subcategories of modality—epistemic versus root, and necessity versus possibility—are encoded syntactically, mapping onto four distinct structural positions. This is schematized in () (Fortin ). ()

ModEpNec>ModEpPoss>Asp>T>Asp>ModRNec>ModRPoss

This ordering sheds light on the root-epistemic split noted above. T, and all markers below T, can appear immediately string-adjacent to the ellipsis site, while no heads higher than T can do so. Assuming the inflectional marker immediately adjacent to the ellipsis site is the licensing element, Indonesian is typologically unusual with respect to its VPE licensers: it is more permissive than Dutch/Romance, and perhaps less permissive than a language like English, some of whose epistemic modals arguably do license VPE.4 Indonesian also differs from closely related Javanese. Sato () argues Kendal Javanese displays MCE, and evidences the dichotomy Ross proposed: root modals, which are control predicates, permit ellipsis of the complement, while epistemic modals, which are raising predicates, do not. Vander Klok () argues Paciran and Semarang Javanese display VPE, and identifies a set of licensers which includes some epistemic and root modals and negation.

. P

.................................................................................................................................. Pseudogapping is often taken to be a subspecies of auxiliary-stranding verb phrase/ predicate ellipsis, with the distinguishing characteristic of an overt VP-internal remnant. The status of pseudogapping, in contrast to that of VPE, is murky. Sato (, b) reports pseudogapping is possible, providing a single example (). () ?Esti harus berdansa dengan Fernando, tapi Fatimah Esti must dance with Fernando, but Fatima berdansa (dengan) Ali. dance (with) Ali ‘Esti must dance with Fernando, but Fatimah CAN (with) Ali.’

4

bisa can

(Sato b: )

As a reviewer notes, the status of English epistemic modals as VPE licensers is controversial. McDowell (), for example, claims epistemic readings are uniformly blocked under VPE, while Gergel () notes some epistemic readings are available, as when have follows the modal (i). (i) My team didn’t win today, but your team might have.





Other speakers deem pseudogapping ill-formed, whether the remnant is an adjunct, such as the PP in (), or a DP argument ().5 () Ali mengagetkan Siti, tapi dia tidak Ali surprise Siti but SG NEG ‘Ali surprised Siti, but he can’t (surprise) Mom.’

bisa can

*(mengagetkan) (surprise)

Ibu. mom

The availability of pseudogapping with a DP remnant may roughly correlate with a speaker’s tolerance for rightward extraposition of heavy DPs in general (); my consultants reject both, while Sato’s permit both. This correlation is unsurprising under Jayaseelan’s () analysis of pseudogapping as rightward extraposition of the remnant DP, followed by ellipsis of the VP. () *Fatimah harus makan ti hari ini [nasi goreng yang Fatimah must eat day this rice fry REL yang di-buat oleh Ibu]i. REL PASS-make by mother ‘Fatimah must eat __ today very spicy fried rice that Mom made.’

pedas spicy

sekali very

Why speaker judgments vary is not yet clear. Sato does not directly address the availability of pseudogapping contexts similar to (). If the correlation noted above is on track, however, we expect speakers who share the judgment reported by Sato to accept ().

. G

.................................................................................................................................. Gapping is attested in Indonesian, with a few qualifications. The first is sociolinguistic: Hernández (: ) notes that while gapping ‘is not allowed in written language . . . it is accepted by most speakers in spoken language’, conjecturing the restriction in written language is prescriptive.6 The facts described herein may, then, be limited to less formal registers. As in other languages, gapping in Indonesian targets a verb (and, optionally, additional material), and is restricted to coordinate structures. Gapping is licensed under identity with an antecedent at an identical level of embedding which linearly precedes it. Gapping in Indonesian is acceptable when the remnant is an adjunct PP () or AdvP (). If the remnant is an argument, either DP or CP, gapping is typically, but not uniformly, deemed acceptable (). Judgments vary widely for gapping that leaves a short DP remnant, perhaps due to prosodic or pragmatic considerations. 5

For some speakers, pseudogapping in comparatives is slightly improved.

Siti bisa membaca lebih banyak buku daripada Ali bisa */??(membaca) majalah. Siti can read more many book than Ali can */??(read) magazine ‘Siti can read more books than Ali can (read) magazines.’ I thank a reviewer for this suggestion. 6 I thank a reviewer for directing me to this work.

(i)

 () Ali berdansa dengan ibu=nya dan Dewi dengan Ali dance with mother=SG and Dewi with ‘Ali danced with his mother and Dewi with her father.’



ayah=nya. father=SG

() Ali makan nasi dengan cepat {sedangkan/*karena} Siti perlahan-lahan. Ali eat rice with quick {while/*because} Siti slowly ‘Ali ate rice quickly and/*because Siti slowly.’ () a. %Ali naik bis, sedangkan Siti kereta. Ali ride bus while Siti train ‘Ali rode the bus, and Siti the train.’ b. Ali yakin Jerman akan menang dan Siti Argentina Ali believe Germany FUT win and Siti Argentina akan menang. FUT win ‘Ali believes Germany will win and Siti Argentina will win.’ Gapping fails if the conjunction is replaced with a subordinating complementizer, such as karena ‘because’ in (). The gap may contain more than a single lexical verb, crossing a clausal boundary (), as Ross () originally noted for English. Although the gapped material does not superficially appear to be a constituent, under some accounts (e.g., Sag ) gapping involves ellipsis of a phrase which the remnants (Siti and sepeda ‘bicycle’) have evacuated, having moved to a focus position. ()

Ali mau membeli mobil dan Siti sepeda. Ali want buy car and Siti bicycle ‘Ali wants to buy a car, and Siti a bike.’

. S

.................................................................................................................................. Stripping elides all material in the second clausal conjunct, save for a single constituent remnant, accompanied by a polarity element such as also, too, or not. Stripping is attested in Indonesian (). () Siti melihat diri=nya di cermin dan Ali juga. Siti see self=SG in mirror and Ali also ‘Siti saw herself in the mirror and Ali too.’ a. Siti saw herself in the mirror, and Siti saw Ali too. b. Siti saw herself in the mirror, and Ali saw Siti too. c. Siti saw herself in the mirror, and Ali saw himself too.





The remnant Ali can be interpreted both as the object of the second conjunct, taking diri=nya ‘oneself ’ as antecedent (a), and as the subject of the second conjunct, taking Siti as antecedent. In the second case, both strict (b) and sloppy (c) readings of diri=nya are available. The remnant and polarity item can also appear in the opposite constituent order (i.e., juga Ali ‘also Ali’). Stripping is also possible with non-DP adjunct remnants. As with DP remnants, the polarity element may either precede () or follow the remnant. ()

Ali berdansa dengan Siti di pesta dan juga Ali dance with Siti in party and also ‘Ali danced with Siti at the party, and with Dewi too.’

dengan with

Dewi. Dewi

Stripping is distinct from ordinary coordination. For DP remnants, the shape of the negative marker serves as a diagnostic. Indonesian has two negative markers in complementary distribution: bukan negates DPs, while tidak negates other types of constituents (). Were () an instance of ordinary DP coordination, bukan would appear; that tidak is required instead indicates that the DP, daging ‘meat’, is not directly negated. () a. Siti {bukan/*tidak} Siti {NEG/*NEG} ‘Siti isn’t a teacher.’

guru. teacher

b. Siti {*bukan/tidak} Siti {*NEG/NEG} ‘Siti doesn’t work.’

bekerja. work

()

Ali makan nasi, tapi {*bukan/tidak} Ali eat rice but {*NEG/NEG} ‘Ali ate fried rice, but not meat.’

daging. meat

Stripping in English and Spanish is limited to matrix coordinate structures (Depiante ). Stripping in Indonesian likewise fails in adjunct () or embedded () clauses, unlike VPE, which, as in other languages, is licit in both contexts ()–(). () Ali mencuci piring, {dan/*karena} juga Siti. Ali wash dish {and/because} also Siti ‘Ali washed the dishes, and/*because Siti too.’ () *Ali makan daging, dan Dewi bilang Ali eat meat and Dewi say ‘Ali eats meat, and Dewi said Siti too.’

juga also

Siti. Siti

Under some proposals, stripping entails focus movement of the remnant to the left periphery, followed by TP-ellipsis (Depiante ; Merchant a). Indonesian is compatible with such an analysis.





. C 

.................................................................................................................................. Indonesian displays “short” comparatives, although ellipsis is not necessarily involved in their derivation. The comparative is signaled by a word of comparison, e.g. lebih ‘more’, and dari(pada) ‘than’. Examples below illustrate comparatives on a subject () and object ()7 and on the object of a preposition (). The compared element may be the entire predicate, as in (), or internal to the predicate, as in the preceding examples. () Ali makan nasi lebih banyak Ali eat rice more much ‘Ali ate more rice than Siti.’

daripada than

Siti. Siti

() Ali makan nasi lebih banyak Ali eat rice more much ‘Ali ate more rice than meat.’

daripada than

daging. meat

() Ibu saya berbicara dengan adik=nya mother SG speak with younger.sibling=SG lebih sering daripada (dengan) saya. more often than (with) SG ‘My mother speaks with her sibling more often than (with) me.’ ()

Siti lebih tinggi daripada Siti more tall than ‘Siti is taller than Ali.’

Ali. Ali

A longer version of the compared clause is generally available, although the abbreviated version is preferred by speakers (). () Siti lebih suka pisang daripada Ali (suka Siti more like banana than Ali (like ‘Siti likes bananas more than Ali (likes bananas).’

pisang). banana)

It is nonetheless likely that a “direct analysis,” akin to Potsdam’s () analysis of Malagasy comparatives, will better account for the properties of comparatives such as (). Under such an analysis, dari(pada) is a preposition, taking a phrasal complement, and no ellipsis occurs; the short form of () is not a truncated version of the long form. The distribution of clitic pronouns, such as the first singular -ku, supports such an analysis. A long comparative is incompatible with a clitic pronoun subject (). 7 As these data suggest, a comparative such as Ali lebih menyukai Siti daripada Dewi ‘Ali likes Siti more than Dewi’ allows two interpretations, as it does in English: Dewi can be construed as either the subject or the object of the compared clause.





() Ali mempunyai lebih banyak saudara Ali have more many sibling ‘Ali has more siblings than me/I (*have).’

daripada=ku than=SG

(*mempunyai). (*have)

More generally, while many Indonesian prepositions host enclitic pronouns (Sneddon, Adelaar, Djenar, and Ewing ), no complementizers do so ().8 () *Siti memutuskan bahwa=nya memasak Siti decide that=SG cook ‘Siti decided that=she would cook tofu.’

tahu. tofu

In both examples, replacing the clitic with the equivalent strong pronoun (aku and dia respectively) results in a grammatical sentence. The dari(pada) in the short comparatives is thus plausibly different from the dari(pada) in the long comparatives—a preposition rather than a complementizer—suggesting that a direct analysis may be on the right track.

. R- 

.................................................................................................................................. In some recent literature (e.g., Wilder ; Barros and Vicente ), right-node raising (RNR) is argued to be a type of “backward” ellipsis, under which a portion of the first conjunct is null under identity with the second conjunct. Whether Indonesian exhibits RNR is somewhat controversial. RNR which targets a DP complement of V () or a VP modifier is not permitted for my consultants. Muadz (: ) reports that RNR which targets a DP complement of P is, however, allowed ().9 () *Siti menghilangi lalu Ali menemukan Siti lose then Ali find ‘Siti lost, then Ali found, her key.’

kunci=nya key=SG

() Allena pulang dari, dan Marie pergi ke Allena go.back from and Marie go to ‘Allena went home from, and Marie went to, school.’

sekolah. school

VPs also resist RNR (), although forward VPE is possible (.). () *Siti boleh sedangkan Ali harus pergi Siti may while Ali must go ‘Siti may, and Ali must, go to the party.’

ke to

pesta party

8 Sneddon, Adelaar, Djenar, and Ewing (: ) state that dari ‘than’ permits a clitic pronoun complement, although the homophonous dari ‘from’ does not. 9 I thank a reviewer for directing me to this example.





For some speakers, examples such as (), which resemble RNR of a clausal complement, are permitted. As ingat ‘remember’ is a verb permitting NCA (.), it is plausible that () involves “backward” NCA, rather than RNR. () Dewi ingat tapi Ali lupa bahwa ada ujian hari Dewi remember but Ali forget COMP exist exam day ‘Dewi remembered, but Ali forgot, that there is an exam today.’

ini. this

Future investigation of RNR is warranted, particularly around the apparent contrast depicted in () and ().

. C 

.................................................................................................................................. I assume, contra Wilder (), that small conjuncts are in principle possible, and there are no constraints on the types of constituents that can be coordinated. I limit my discussion to two cases for which there exists independent motivation for conjunction reduction (CR). I address here “shared subject coordination,” as defined in Wilder (), and postpone discussion of “non-constituent coordination” internal to the VP to .. In Indonesian, the subject of a second conjunct clause may be null under identity with an antecedent in the first conjunct, yielding what may be shared subject coordination. In English, there is an interpretive asymmetry that arises between shared subject sentences (a) and sentences in which the second conjunct’s subject is an overt pronoun (b) (Wilder  and references therein). The first type permits a bound variable interpretation, indicating the second conjunct subject is in the scope of the quantifier in the first conjunct. The second type permits only an “e-type” interpretation (Wilder : ). Wilder uses this asymmetry to argue against an analysis in which the second conjunct’s null subject is a pronoun—namely, the null pronoun pro (Van Valin )—and for an ellipsis analysis. ()

a. Few congressmen admire Kennedy and are very junior. b. Few congressmen admire Kennedy and they are very junior. (Wilder : , ex. )

Null subjects in tensed clauses in Indonesian, particularly when they refer to animate entities, are restricted in their distribution. Given two conjoined clauses, the second clause subject can be null (), and the same interpretive asymmetry arises. () Beberapa mahasiswa belajar bahasa Inggris dan (mereka) some student study language English and (PL) bekerja di perpustakaan. work in library ‘Some students study English and (they) work in the library.’





When the second conjunct subject is null, a bound variable interpretation arises: there are some students who both study English and work in the library. When overt, the e-type interpretation obtains: all of the students who study English work in the library. Adopting Wilder’s analysis, if the second conjunct subject were pro, we would expect—contrary to fact—the interpretation to be identical to the interpretation that arises with mereka ‘they’. The interpretive asymmetry points to a shared subject analysis. A reviewer suggests that () involves argument ellipsis (AE) instead of CR. The restricted distribution of Indonesian null subjects, which require a very local antecedent, renders many typical AE diagnostics of limited use. One involves the availability of sloppy interpretations. Pronouns permit only a strict interpretation, while sloppy interpretations involve ellipsis (Oku ). This must be evaluated separately for each type of null argument. Sato () argues that in Javanese empty objects, but not empty subjects, allow sloppy interpretations, indicating that null subjects do not involve AE. Consider (). The embedded empty subject must refer to the most local antecedent, Dewi. It cannot find an antecedent in the previous utterance, and so although the sloppy interpretation (Dewi’s child) is unavailable, the strict reading is also. ()

a. Siti bilang anak=nya suka Siti say child=SG like ‘Siti said her child likes carrots.’

wortel. carrot

b. Dewi bilang e suka bawang. Dewi say e like onion ‘Dewi said e likes onions.’ Although it is plausible that Indonesian permits AE, additional investigation of the distribution of Indonesian pro is required to fully evaluate these two potential analyses (i.e., AE and CR).

. N  

.................................................................................................................................. Indonesian permits null clausal complements following certain embedding verbs, including tahu ‘know’, tolak ‘refuse’, ingat ‘remember’, and lupa ‘forget’. I take these null clausal complements to be NCA. As in other languages, NCA can be interpreted as a property or a proposition (). () Siti menyuruh Ali untuk mencuci Siti order Ali for wash ‘Siti told Ali to wash the dishes . . . ’

piring, dish

tapi but

i. property interpretation: ‘ . . . but he forgot to wash the dishes.’ ii. proposition interpretation: ‘ . . . but he forgot that she told him to wash the dishes.’

dia SG

lupa. forget





Depiante (: ) analyzes NCA in English, French, and Spanish as a “deep” sentential proform—i.e., a phonologically null proform with no internal syntactic structure— following much earlier work on NCA, including Hankamer and Sag (). Hankamer and Sag distinguish deep anaphors from surface anaphors, which include VPE, on the basis of several characteristics. Although Indonesian NCA resembles NCA in these languages in many respects, it also displays properties which are not anticipated under the null proform analysis. I consider first the characteristics of Indonesian NCA which are amenable to this analysis, and then those which are not. First, as Depiante notes, English displays overt counterparts of this proposed proform, it and so (). (Note that it and NCA do not have the same distribution.) Indonesian does likewise, via the third person singular clitic proform =nya. Example () permits both NCA and the overt proform, and the same interpretation—namely, that Siti refused to read the book—arises. ()

I believe she means business and you’d better believe it too. (Depiante : , ex. )

() Ali menyuruh Siti untuk membaca buku ini, tapi dia menolak(=nya). Ali order Siti for read book this but SG refuse(=SG) ‘Ali told Siti to read this book, but she refused (it).’ Secondly, like English NCA, Indonesian NCA () allows both strict and sloppy identity readings. Under the strict interpretation, Siti forgot to call Ali’s mother; under the sloppy interpretation, Siti forgot to call her own mother. ()

Ali ingat menelpon ibu=nya, tapi Siti lupa. Ali remember call mother=SG but Siti forgot ‘Ali remembered to call his mother, but Siti forgot.’

Lastly, Indonesian NCA may be licensed by a pragmatic antecedent (). () [At a gathering, amid a discussion of who should be next to tell their favorite joke. Silently, I point to Siti. She replies:] Saya tidak mau. SG NEG want ‘I don’t want.’ These data are consistent with an analysis of Indonesian NCA as a deep anaphor. Depiante () further argues that NCA does not permit extraction because it contains no structure, and here, Indonesian NCA behaves differently. The verb tolak ‘refuse’ allows an overt clausal complement (a), as well as NCA (). NCA combined with extraction of a whphrase from within the ellipsis site is also permitted (b). This has the interpretation that Siti’s parents know which book Siti refused to read, which requires NCA. () a. Siti menolak untuk membaca Siti refuse for read ‘Siti refused to read this book.’

buku book

itu. this



 b. Ali tahu buku yang mana yang dibaca oleh Siti, sedangkan Ali know book REL which REL PASS-read by Siti while orangtua Siti tahu buku yang mana yang ditolak oleh Siti. parents Siti know book REL which REL PASS-refuse by Siti ‘Ali knows which book was read by Siti and her parents know which book was refused (to be read) by Siti.’

It is possible that Indonesian NCA is a surface anaphor, as Cyrino () has proposed for Brazilian Portuguese NCA. With the possible exception of the availability of pragmatic antecedents, which specifically point to a deep anaphor analysis (Hankamer and Sag ), the Indonesian data are consistent with a surface anaphor analysis. There is nothing that in principle prevents an overt proform from alternating with a surface anaphor, and Depiante () further observes the availability of sloppy identity readings is not specific to deep anaphors. Further investigating the properties of Indonesian NCA will contribute to our understanding of the typology of NCA cross-linguistically.

. N 

.................................................................................................................................. Indonesian permits certain kinds of ellipsis within nominal constituents under identity with an antecedent. Ellipsis of the head noun following a classifier, or a numeral, is possible (). () a. Berapa ekor ayam yang how.many CLASS chicken REL ‘How many chickens did you buy?’ b. Lima (ekor) five (CLASS) ‘Just five.’

kamu SG

beli? buy

saja. just

Ellipsis of a head noun preceding a relative clause, which is signaled by yang, is likewise permitted, although ellipsis of a head noun preceding an adjectival modifier is prohibited ().10 () Baju bagus ini ada di atas *(yang) jelek shirt beautiful this exist at top REL ugly ‘The beautiful shirt is on top of (the one) that is ugly.’

10

(i)

itu. that

Headless relative clauses are widely attested, and do not require an antecedent (i).

Andi memilih yang besar. Andi choose REL big ‘Andi chose the big one.’ (Sneddon, Adelaar, Djenar, and Ewing : ) The lack of an antecedent in (i) suggests ellipsis is not involved. As the interpretation of () is determined by an antecedent, I assume ellipsis has occurred.





Lexical DP possessor remnants are not permitted, although possessor pronouns can surface as remnants (). Given Indonesian’s lack of morphological case matching, the surface string in the second conjunct—kamu kotor—also means ‘you are dirty’. Although this reading is plausible, it is not salient to speakers in this context; speakers report a preference to interpret the second conjunct as parallel to the first. () Baju saya bersih, tapi (baju) kamu shirt SG clean but (shirt) SG ‘My shirt is clean, but yours is dirty.’

kotor. dirty

Ellipsis within the DP is possible with a variety of remnants, including numerals, classifiers, and relative clause modifiers. If the remnant is a possessor, ellipsis is not uniformly possible. A reviewer suggests that ellipsis leaving a lexical DP possessor is blocked for lack of a licensing head, a role which the pronoun in () fills; this merits further investigation. Additional exploration may shed further light on the structure of the Indonesian DP, which has thus far received little attention in the literature (see Loewen  for one proposal).

. I    PSG

.................................................................................................................................. Indonesian highlights the need to examine the behavior of prepositions under ellipsis, which cuts across several of the phenomena described above. Although Indonesian never permits preposition stranding in non-elliptical contexts, it allows prepositions to be omitted in some elliptical contexts. This is unexpected, given the PSG (Merchant ), which links a language’s ability to omit a preposition under ellipsis to its ability to strand a preposition via extraction of its complement and which is known to be very robustly supported empirically. These contexts are sluicing (Fortin b; Sato , b), fragments (Sato ), pseudogapping (Sato , b), stripping, and possibly RNR (Muadz ). Sato (, b, ) proposes Indonesian prepositions are subject to a constraint that requires them to be linearly adjacent to their complement at PF; violations of this constraint are “repaired” by ellipsis. Prepositions affected by gapping and a type of CR involving apparent non-constituent coordination within the predicate, however, behave as the PSG predicts.

.. Sluicing In Indonesian, as in English, when the remnant wh-phrase is the complement of a preposition, and a correlate PP is found in the antecedent, the preposition can be omitted from the remnant (). Unlike English, Indonesian prohibits preposition stranding in nonelliptical contexts (), so the acceptability of () is surprising: the PSG predicts this pattern to be impossible (Fortin b; Sato , , b). If the PP is sprouted, the preposition is obligatory (). This contrast is consistent with Chung’s Generalization, which states





prepositions cannot be omitted from sprouted wh-PPs, even in P-stranding languages like English (Chung ). () Ali menonton filem — coba tebak Ali watch movie — try guess ‘Ali watched a movie—guess *(with) who!’

*(sama) *(with)

siapa! who

Apparent violations of the PSG occur in a number of languages. For these languages, it is generally argued that the underlying source of the prepositionless remnant is not the corresponding wh-question, but rather a type of wh-cleft, the derivation of which crucially involves no preposition stranding. These include Spanish and Brazilian Portuguese (Rodrigues, Nevins, and Vicente ), Mandarin Chinese (Wang ), Serbo-Croatian (Stjepanović , ), Russian and Polish (Szczegielniak , ; Nykiel ), Greek (Vlachos ), and Romanian (Nicolae ); see also van Craenenbroeck (a). When an alternative source of the prepositionless remnant can be found, the strength and predictive power of the PSG is not challenged. The prepositionless sluices in Indonesian, like those in Emirati Arabic (Leung ), are not, however, amenable to an elliptical wh-cleft analysis. Sato (, b) argues such sluices behave differently from the corresponding wh-clefts, concluding they are parallel in structure to the corresponding wh-questions, and that Indonesian thus is a genuine counterexample to the PSG. Although some of the relevant data are subtle, I review here the two types of empirical evidence Sato provides which are the most convincing: sluicing with multiple potential antecedents (), and sluicing involving multiple remnants (). () Saya tahu seseorang dari Indonesia berdansa dengan seseorang SG know someone from Indonesia dance with someone dari Jepang tapi saya tidak ingat siapa (itu). from Japan but SG NEG remember who (that) ‘I know someone from Indonesia danced with someone from Japan, but I don’t remember who (it was).’ (Sato b: , ex. ) () Saya sering sekali bicara dengan banyak orang-orang di berbagai SG often very talk with many people in various masjid di segala penjuru AS jadi saya tidak ingat (dengan) mosque in all corner US so SG NEG remember (with) siapa, (di) masjid yang mana (*itu). who (in) mosque which (*that) ‘I very often talk with many people in various mosques in all corners of the US, so I don’t remember who in which mosque (*it is).’ (Sato b: –, examples –) Sato reports that, if itu is present in ()—that is, if we have a non-elliptical wh-cleft—there is only one available antecedent for siapa ‘who’: seseorang dari Jepang ‘someone from Japan’. Without itu, an ambiguity arises, as a second possible antecedent is made available: seseorang dari Indonesia ‘someone from Indonesia’. This asymmetry, he argues, indicates that the sluiced variant cannot be derived from the wh-cleft. Of (), he notes that multiple





sluicing with two prepositionless remnants is possible, and that the wh-cleft construction cannot accommodate more than one constituent as pivot; given this incompatibility, () cannot be an elliptical wh-cleft either. Sato (, b, ) and Leung () propose that in Indonesian and EA respectively preposition-stranding violations are properties of PF representations only, and are repaired via ellipsis of the stranded P. Sato (: ) suggests Indonesian prepositions are subject to a language-particular constraint requiring them to be linearly adjacent to their complement at PF. Languages which obey the PSG, in contrast, realize preposition-stranding violations elsewhere, such that ellipsis cannot repair them. Since Ross (b), it has been known that sluicing “ameliorates” island effects. It is often assumed, following Merchant (), that some strong islands are properties of PF representations, which are repaired by deletion of the island.11 Violations of the left-branch condition, for example, which prohibits extraction of determiners, possessives, and adjectives from a DP, are hypothesized to create PF-islands. Not all languages prohibit left-branch extractions in the overt syntax; Latin, for example, permits it (Ross ). Assuming Merchant’s typology of islands is on track, and given there is otherwise cross-linguistic variation in what constitutes an island in the syntax, it is plausible that preposition-stranding violations are realized differently across languages. Sato’s proposal rests on the assumption that garden-variety wh-movement of DPs obtains in Indonesian, and this assumption has been contested in the literature. A wh-DP—like other wh-phrases—may appear in situ (a), or at the left edge of the clause (b). Note that, in (b), the relativizer yang is required, and the verbal prefix meNcannot appear. This prefix optionally appears on transitive, active verbs, and is incompatible with the extraction of DPs it c-commands.12 When the wh-phrase is not a DP, these requirements are reversed: yang cannot appear, while meN- appears freely (b). () a. Ali mem-beli apa? Ali ACT-buy what ‘What did Ali buy?’ b. Apa *(yang) Ali what *(REL) Ali ‘What did Ali buy?’

(*mem-)beli? (*ACT-)buy

() Kepada siapa (*yang) Ali mem-beri to who (*REL) Ali ACT-give ‘Who did Ali give his car to?’

mobil=nya? car=SG

Questions such as () require no special analysis, and involve wh-movement from a predicate-internal position. Questions such as (b), however, are not necessarily amenable

11 Merchant () argues that a second type of strong island violations, which involve extraction from a propositional domain, including extraction out of complex NPs and adjuncts, is never actually incurred in sluicing, and that the derivation of such sluices does not involve island-violating wh-movement. 12 The literature offers many analyses of meN-, and many accounts of this ‘blocking effect’. For a recent survey, see Soh and Nomoto (). For exposition, meN- is glossed here as an active voice marker, following e.g. Sneddon, Adelaar, Djenar, and Ewing ().





to such an analysis. Cole, Hermon, and Aman (forthcoming) argue, for the closely related Malay, that yang is a relativizer, and questions such as (b) have a copular structure. The wh-DP, apa ‘what’, is predicated of a null-headed relative clause, yang Ali beli ‘that Ali bought’. Wh-DPs thus do not undergo movement from a predicate-internal position, as the wh-DP itself is the predicate of the copular clause.13 This type of structure is found in wh-questions in several other Austronesian languages, such as Malagasy (e.g., Paul and Potsdam ). Cole, Hermon, and Aman’s analysis plausibly extends to Indonesian (Fortin ). If Indonesian wh-DPs cannot wh-extract from a predicate-internal position, a P-stranding wh-question is ungenerable. If so—if prepositionless sluices cannot be derived via wh-movement of the complement of P—Sato’s account does not tell the whole story. Speaking of what they call ‘sluicing-like constructions’ in wh-in-situ languages, Paul and Potsdam (: ) note there is more than one “route to a sluicing-like surface representation,” and that languages related to Indonesian, including Malagasy and Javanese, do not have “prototypical sluicing” (Paul and Potsdam : ). Indonesian also lacks prototypical sluicing, regardless of which account proves accurate.

.. Fragment answers Merchant (a) further observed that his () PSG extends to fragment answers. Languages which do not permit P-stranding in non-elliptical contexts also prohibit fragments containing only a DP complement of P. Again, Indonesian breaks the mold: prepositions are easily omitted from fragments. () a. Dengan siapa=kah Ali berdansa? with who=Q Ali dance ‘Who did Ali dance with?’ b. (Dengan) Siti. (with) Siti ‘(With) Siti.’ Sato () addresses this pattern, arguing Merchant’s (a) analysis of fragments (see .) is otherwise empirically supported within Indonesian. Indonesian fragment DPs, for example, exhibit island effects, as this analysis predicts. In the context of (a), the fragment (b), whose antecedent, Budi, lies within an adjunct island, is ungrammatical.

13

Under this account, the ban on preposition stranding manifests as a ban on preposition-stranding movement of the null operator within the relative clause. Similarly, meN- blocks movement of the null operator, not of the wh-DP itself. Further evidence for null operator movement comes from island effects. The relativized NP cannot, for example, originate within a relative clause. (i)

*Siapa [yang Ali suka [perempuan yang fikir tinggal di Singapura]]? who REL Ali like woman REL think live in Singapore ‘*Who does Ali like [the woman that thinks __ lives in Singapore]?’ (Cole, Hermon, and Aman forthcoming: , ex. )

 () a. Esti pulang karena Budi tidak mau berdansa Esti return because Budi NEG want dance ‘Did Esti go home because Budi didn’t want to dance with her?’

dengan with

 dia? SG

b. *Bukan! Fernando! no Fernando ‘No! Fernando!’ (Sato : ) Sato further argues, as with sluicing, the prepositionless fragment cannot be analyzed as the pivot of an elided cleft. The “short” cleft (b) is not an acceptable response to the question posed in (a), and, given the ban on preposition stranding in non-elliptical contexts, there is no grammatically acceptable “long” cleft. () a. Tentang pria yang mana kamu about man REL where SG ‘About which man did you talk?’

bicara? speak

b. (*Itu) Pak Sugiarto. ( DEM) Mr Sugiarto ‘It was Sugiarto.’ As with sluicing, Sato (: ) accounts for this pattern with an Indonesian-particular constraint requiring prepositions to be linearly adjacent to their complement at PF. Violations of this constraint, he argues, are repaired under ellipsis of the constituent containing the preposition.

.. Pseudogapping As noted above, Sato (, b, ) reports pseudogapping is possible in Indonesian, providing an example demonstrating preposition omission (). He invokes this data in support of his analyses of Indonesian sluicing and fragments, noting his analysis predicts P-stranding violations will be repaired by any type of ellipsis which minimally eliminates the offending PP from the PF representation, a scenario which obtains in pseudogapping, Sato reports his consultant freely permits rightward extraposition of DPs (), but not of complements to P (). () *Fatimah harus berdansa dengan ti hari ini [seorang Fatimah must dance with day this a yang kaya]i REL rich *‘Fatimah must dance with __ today a rich man.’ (Sato : )

laki-laki man

For Sato, this supports his claim that violations of the ban on preposition stranding in Indonesian are repaired under ellipsis, as long as the ellipsis site contains the stranded preposition.





.. Stripping In stripping, preposition omission is again possible: the remnant can be the DP complement of P. () Ali berbicara tentang musik dan juga (tentang) Ali talk about music and also (about) ‘Ali talked about music and (about) politics too.’

politik. politics

As noted above, stripping is argued to be focus movement of the remnant to the left periphery, followed by TP-ellipsis (Depiante ; Merchant a). Preposition omission under stripping can be accounted for with Sato’s proposal that violations of the ban on P-stranding in Indonesian can be repaired if the stranded preposition is contained within the ellipsis site.

.. Gapping Gapping offers a departure from the pattern described in the previous subsections. Unlike in English, gapped material cannot contain a preposition to the exclusion of its complement. ()

*Ali berdansa dengan ibu=nya dan Dewi Ali dance with mother=SG and Dewi ‘Ali danced with his mother and Dewi her father.’

ayah=nya. father=SG

Johnson (: ) proposes gapping results from the ellipsis of a constituent containing a verb from which the remnants have scrambled. If so, the derivation of () involves scrambling of the DP ayah=nya ‘her father’ out of the PP headed by dengan ‘with’. Although P would be contained within the ellipsis site, gapping nonetheless fails. It is not yet clear why gapping and pseudogapping behave differently in this regard. As noted above, speakers who permit pseudogapping allow P to be contained within the ellipsis site while its DP complement survives as remnant. If both processes involve A’-movement of the remnant DP, followed by ellipsis of the constituent containing P, this contrast is perhaps noteworthy; there are, however, analyses of gapping (e.g., Johnson ) which do not involve ellipsis, and which may account for this contrast.14

14 It is perhaps worth noting that, as originally observed by Neijt (), prepositions cannot be stranded under gapping in English: the gapped material cannot contain P’s complement to the exclusion of P (i), although gapping otherwise can target discontinuous strings.

(i) *John is confident of a successful outing and Peter is dependent on a successful outing. (Neijt : )





.. Conjunction reduction A certain type of CR perhaps presents a clearer counterexample to the pattern that has been established for Indonesian. Sailor and Thoms () consider cases which, on the surface, resemble coordination of strings which are not constituents. In their terminology, each underlined string contains two “chunks”: the first, the DP complement of a preposition, the second, a PP. () I spoke to John on Thursday and Mary on Friday. (Sailor and Thoms : ) The analogue of () is not well-formed in Indonesian (): dengan ‘with’ is mandatory in the second conjunct. () Ali berdansa dengan Dewi pada hari Jumat dan *(dengan) Ali dance with Dewi on Friday and *(with) Siti pada hari Sabtu. Siti on Saturday ‘Ali danced with Dewi on Friday and (with) Siti on Saturday.’ Under Sailor and Thoms’ analysis, the two conjuncts are full CPs. The two chunks which comprise the second string move independently to a left-peripheral position in the second conjunct, and the TP in the second conjunct deletes. As they note, such an analysis reflects the similarities between these data and stripping, which is also argued to involve focusfronting followed by TP-ellipsis (Depiante ; Merchant a). Sluicing is abstractly similar. Sluicing and stripping permit preposition omission in Indonesian; it is not yet clear why it fails here.

. C

.................................................................................................................................. This preliminary sketch, in conjunction with earlier work (especially by Yosuke Sato) on Indonesian sluicing, fragments, and VPE, indicates that Indonesian ellipsis phenomena could contribute meaningfully to our understanding of the various ways these phenomena manifest cross-linguistically. One issue raised by Indonesian involves the behavior of prepositions under ellipsis, which contradicts the otherwise very robustly supported PSG. Although the anomalous behavior of Indonesian prepositions in ellipsis was first noted in the literature for sluicing, other ellipsis phenomena, including fragments, pseudogapping, and stripping, and possibly RNR, display the same property.15 Gapping and CR do not. Developing a fuller understanding of these ellipsis phenomena is a necessary first Additionally, in short comparatives, preposition omission is possible. As discussed in ., however, it is plausible that these do not involve ellipsis at all. 15





step. Additionally, a deeper understanding of the nature of the possibly unique relationship between Indonesian prepositions and their DP complements, and of the other ways that Indonesian permits “repair” under ellipsis, is required. A second area meriting further exploration is NCA, as its properties are not entirely accounted for under the standard null proform analysis.

A I thank Emma Sunog, Diandra Irawan, two anonymous reviewers, and the editors of this volume. All errors are my own.

  ......................................................................................................................

 ......................................................................................................................

 

. I

.................................................................................................................................. T objective of this chapter is to provide an overview of ellipsis phenomena in Japanese. I will discuss the properties of fragment phenomena, such as stripping, sluicing,1 ellipsis in comparatives, and short answers, as well as Right-Node Raising, null arguments, and N’-deletion, and while doing so, I will also touch upon VP-ellipsis, gapping, and pseudogapping. Before getting into the examination of these phenomena, let us briefly summarize some of the characteristics of Japanese that are relevant to the ensuing discussion. First, Japanese has the basic word order of SOV, but it also allows a relatively free word order (a phenomenon called scrambling; see Saito , , ; Miyagawa , ; Ueyama , , among many others). Second, the thematic relations of noun phrases to the verb are indicated by particles suffixed to the noun phrases, and thus postpositions, rather than prepositions, are attested. In the following discussion, I will call such particles (including postpositions) case markers. Third, Japanese has a phenomenon called case drop, in which case markers on noun phrases can be optionally dropped, especially in casual speech. Fourth, Japanese is a wh-in-situ language, in which a wh-phrase can stay in its base position. Fifth, Japanese allows arguments to be apparently missing, and the contents of the missing arguments are recovered from the context. This chapter is organized as follows. Section . discusses the properties of fragment phenomena in Japanese, such as stripping, sluicing, ellipsis in comparatives, and short answers, especially with respect to Subjacency effects and the availability of sloppy readings. Section . deals with what appears to be comparable to Right-Node Raising in English, mainly from the viewpoint of Subjacency effects. Section . addresses the debate between the ellipsis view and the non-ellipsis view of the null argument phenomenon in

1 Although this construction in Japanese is not analyzed on a par with English sluicing as will be seen in section ..., I will use the term sluicing as a descriptive term to refer to phenomena in Japanese that resemble English sluicing. But see n. .





Japanese. Section . is concerned with so-called N’-deletion, and examines its properties mainly with respect to the availability of the sloppy reading. Section . summarizes the chapter. In the discussions of fragment phenomena, it will be clear that the presence and absence of case markers play an important role in the analyses of ellipsis in Japanese. It will be observed that case-marked fragments are to be analyzed as having a full-fledged structure in the ellipsis site (surface anaphora; cf. Hankamer and Sag ) while non-case-marked fragments are not (deep anaphora; cf. Hankamer and Sag ).2 Since the case-marked and non-case-marked contrast also shows up in other phenomena in Japanese that do not involve ellipsis, such as the topic construction (cf. Saito : ch. ) and the cleft construction (cf. Hoji : ch. ), the study of fragments can also be regarded as part of a larger research project that explores the properties of Japanese.

. F

.................................................................................................................................. I start with a discussion of the properties of four types of fragment phenomena observed in Japanese: stripping, sluicing, ellipsis in comparatives, and short answers. It will be seen that stripping, sluicing, and ellipsis in comparatives exhibit parallel properties and should thus be regarded as a uniform phenomenon in this language while short answers, which appear to be comparable to those three types, depart from them with respect to Subjacency effects.

.. Stripping, sluicing, and ellipsis in comparatives as a uniform phenomenon ... Stripping As Hoji (: ch. ) observes, Japanese has a construction that appears to be analogous to English stripping (cf. Hankamer ; Reinhart , among others), as illustrated in ().3 The underlining indicates the correlate, i.e., the element in the first conjunct that corresponds to the stranded phrase in the second conjunct (let us call this stranded phrase the remnant). Note here that stripping in Japanese is typically accompanied by the copula.

2

Note that this is a somewhat simplified view of the case-marked and non-case-marked distinction. Given that case markers can optionally be dropped in Japanese, there could be cases where a non-casemarked version is taken to be parallel to its case-marked counterpart. Case-marked fragments also have some complications in that there are cases where case-marked fragments appear without a linguistic antecedent. Fragments with and without case markers, however, tend to be taken as instances of surface anaphora and deep anaphora, respectively, and thus what is presented in this chapter represents the analyses of unmarked cases. See section . and Fukaya and Hoji (). 3 What is claimed to be stripping in Japanese is different from stripping (or Elliptic Conjunctions to use the term of Reinhart ) in English in that the former can appear across speakers as in () and also in embedded contexts, whereas the latter typically only occurs in matrix conjunctions.

 ()

A: John-ga Bill-o syootaisita John- Bill- invited ‘John invited Bill (I heard).’ B: Paul(-o)-mo Paul(-)-also ‘Paul too.’

da. 



(tte). (I heard)

(Based on Hoji : ch.  ())

As observed in Hoji (: ch. ), there are two subtypes of stripping in Japanese, as indicated in (): (i) case-marked stripping (cm-stripping), where the remnant is marked with a case particle, and (ii) non-case-marked stripping (non-cm-stripping), where it is not.4 Hoji (: ch. ) points out that although the surface difference between cm-stripping and non-cm-stripping is minimal, they exhibit quite distinct properties.5 Hoji observes that cm-stripping exhibits Subjacency effects while non-cm-stripping does not.6 (B’) is acceptable with the intended reading in the parentheses while (B) is not. () A: [NP[S' proi

huransu ryoori-o tukuru] hito]-ga Yoku French cuisine- make people- often koko-ni kuru. here-to come ‘People who make French cuisine often come here.’ B: *Itariya ryoori-o-mo da. Italian cuisine--also  ‘Italian cuisine, too.’ (‘People who make Italian cuisine also often come here.’) B’: Itariya ryoori-mo da. Italian cuisine-also  (Based on Hoji : ch.  () and ())

Hoji also observes that sloppy readings are available in stripping. Adopting the standard assumption that the availability of the sloppy reading is constrained in the same way as that of bound variable anaphora (cf. Lasnik : appendix and Reinhart a: ch. ), Hoji assumes the necessary conditions for the availability of the sloppy reading along the lines summarized in Fukaya and Hoji () as in ().

4

Note that multiple stripping, which strands more than one phrase, is possible. Thus, (i) is possible as a reply to (A). (i) B’: Susan-ga Mary-o da yo. Susan- Mary-   ‘Susan (invited) Mary.’ One of the editors pointed out to me the possibility of multiple stripping being analyzed as equivalent to gapping in English. I will leave this issue open. 5 In addition to what will be discussed in this section, Hoji observes that cm-stripping requires a linguistic antecedent while non-cm-stripping does not. See Hoji (: ch. ) for details. 6 There is one complication regarding Subjacency effects in cm-stripping in Japanese. Fukaya (: ch. ) observes that if the remnant is not suffixed by -mo ‘also’, Subjacency effects seem not to be observed, even if the remnant is case-marked. See Fukaya (: ch. ) for a detailed discussion.

 ()

 a. The lexical condition: The dependent term must have certain properties such that it can be ‘dependent’ on another expression. (Let us refer to the relevant property as [+β] (cf. Fiengo and May ).) b. The structural condition: The dependent term must be c-commanded by its antecedent (cf. Lasnik : appendix; Reinhart a: ch. ). (Based on Fukaya and Hoji : ())

Personal pronouns in English and so-words in Japanese like so-ko ‘that place’ can be [+β], while names and a-words in Japanese like aso-ko ‘that place over there’ cannot.7 Hoji then observes that the sloppy reading in cm-stripping is constrained by () while that in non-cm-stripping is not. Consider the cm-stripping example in (), where the dependent term is [+β]. () A: John-wa Toyotai-ni [NP pro sokoi-ni hairitagatteita hito]-o John- Toyota- there- wanted.to.join person- syookaisita]. introduced ‘John introduced to Toyotai (the/a) person(s) who wanted to join iti.’ B: Nissan-ni-mo Nissan--also ‘To Nissan, too.’

da. 

(Hoji : ch.  ())

(B) clearly gives rise to the sloppy reading that John introduced to Nissanj (the/a) person (s) who wanted to join itj, in addition to the strict reading that John introduced to Nissan (the/a) person(s) who wanted to join Toyota. As pointed out by Hoji, the sloppy reading becomes unavailable if we replace the [+β] term soko ‘that place/the place’ with the non[+β] term asoko ‘that place’ or Toyota in ().8 If the antecedent does not c-command the dependent term in cm-stripping, as in (), the sloppy reading is unavailable.9

See Ueyama (, especially chapter ), for a detailed discussion of the difference between a-demonstratives and so-demonstratives with regard to indexing possibilities, and for relevant references. 8 As noted in nn.  and  of Hoji (), the effects of so-called Condition C of Binding Theory, which states that an R-expression must be free, are known to be rather weak (cf. Bach and Partee :  n. , Haïk :  n. , and Milner , among others), and they are also known to be very weak in Japanese (cf. Oshima  and Kuno , among others). Some speakers, however, might detect the effects more strongly, as one reviewer does. Such speakers might have difficulty with the tests on the lexical condition to be discussed in sections ...–, ..., and .. 9 Notice that the intended “antecedent” does not precede the dependent term in the utterance of A in (). This type of example is used because the sloppy reading seems to be available if the antecedent precedes the dependent term, even if the former does not c-command the latter. The availability of the sloppy reading in cases where the antecedent precedes but does not c-command the dependent term is as expected if the “two types of dependency” discussed in Ueyama () can underlie not only bound variable anaphora but also coreference and the sloppy reading. See also Fukaya (: chs  and ) for detailed discussion of sloppy readings. 7

 ()



A: John-wa [NP pro sokoi-ni hairitagatteita hito]-ni Toyotai-nituite John- there- wanted.to.join person- Toyota-about tazuneta. asked ‘John asked (the/a) person(s) who wanted to join iti about Toyotai.’ B: Nissan-nituite-mo da. Nissan-about-also  ‘About Nissan, too.’ (Hoji : ch.  (), slightly adapted)

Non-cm-stripping, on the other hand, exhibits distinct properties. As Hoji observes, if the case marker on the remnant is dropped as in Nissan-mo ‘Nissan-also’ in () and (), the sloppy reading is available, and it continues to be so even if the [+β] term soko is replaced by the non-[+β] term asoko. Given these observations, Hoji proposes to analyze cm-stripping as in (a) and non-cmstripping as in (b), whose structure, he claims, is identical to (b’). () a. [CP [TP . . . ] ] NP-CASE(-mo) da. b. pro NP(-mo) da. b’. sore-wa NP(-mo) da. that- NP-also  ‘(lit.) That is NP as well.’

(Hoji : ch.  (), adapted)

Under Hoji’s (: section ..) approach, (B) with the case marker is derived as follows. It is base-generated as in (a). In (A) the correlate is raised and adjoined to the TP that dominates it by the operation Constituent Raising (CR) proposed by Reinhart (), as illustrated in (b). Then, the λ-operator is introduced and the trace of Bill-o ‘Bill-’ is translated into a variable bound by the λ-operator, and we thus obtain (c). The TP that immediately dominates the λ-operator is copied into (a), resulting in (d). Note that this way the remnant is associated with a position within the theta-domain of the verb and the case marker on it can be interpreted. Hoji (: section ..) assumes that “the λ-operator assumes the function of an empty operator” and thus (d) is equivalent to (e), which is a representation Hoji (: section ...) proposes for the cleft construction in which the focus is case-marked (cm-cleft). Hence, cm-stripping is analyzed on a par with cm-clefts.10 () a. [CP [TP . . . ]] Paul-o-mo da. b. [TP Billi-o Bill-

[TP John-ga John-

c. [TP Bill-o Bill-

[TP λx [TP John-ga John-

ti syootaisita]]. invited x syootaisita]]]. invited

10 Fukaya (: ch. , section ...), however, discusses examples that seem to indicate that cmstripping may not be completely equivalent to cm-clefts with respect to the availability of bound variable anaphora. See his examples in ()–().



 d. [CP [TP λx [TP

John-ga John-

e. [CP [C OPi] [TP John-ga John-

x syootaisita]]]. invited ti syootaisita]] invited

Paul-o-mo Paul--also

(no wa) C 

da. 

[NP Paul]i-o-mo Paul--also

da. 

CR is assumed to be subject to Subjacency, following Reinhart (), just as the null operator movement in cm-clefts as illustrated in (e) is; hence, Subjacency effects are observed in cm-stripping, as in the case of cm-clefts.11 To see how the sloppy reading is obtained, let us consider (). CR raises and adjoins the correlate to the matrix TP in (A), resulting in (a). The λ-operator is introduced, and the trace of the raised NP and the dependent term are turned into variables bound by the λ-operator, as in (b). Then, the TP that immediately dominates the λ-operator is copied into the second conjunct in (B), resulting in (c), which yields the sloppy reading. () a. [TP Toyotai-ni [TP John-wa ti Toyota- John- hito]-o syookaisita]]. person- introduced

[NP pro soko-ni there-

b. [TP Toyotai-ni [TP λx [TP John-wa x Toyota- John- syookaisita]]]. introduced

[NP pro x

hairitagatteita wanted.to.join

hairitagatteita wanted.to.join

hito]-o person-

c. [CP [TP λx [TP John-wa x [NP pro x hairitagatteita hito]-o syookaisita]]]. John- wanted.to.join person- introduced Nissan-ni-mo da Nissan--also  Because of the lexical condition in (a), the sloppy reading becomes unavailable if we replace the [+β] dependent term by a non-[+β] term. Due to the structural condition in (b), the sloppy reading is unavailable if the relevant c-command relation does not hold as in (). Let us turn to non-cm-stripping, which Hoji analyzes as in (b). He suggests that this construction is equivalent to English sentences like ‘That/The same thing is true of NP as well’ and observes that ‘that’ or ‘the same thing’ seems to be able to refer to a property available in the pragmatic context, and that this property is predicated of the remnant. (B’), for example, has the structure as in (), and pro seems to refer to “some property of an object such that people who make it come here often” (Hoji : section ..). () pro

Itariya ryoori-mo da. Italian cuisine-also  ‘(lit.) pro is Italian cuisine as well.’

11 Hoji (, : ch. ) identifies another variety of the cleft where the remnant is not case-marked (non-cm-cleft) and proposes that no movement is involved in the derivation of non-cm-cleft; thus, no Subjacency effects are expected.





Since there is no movement involved in its derivation, unlike in cm-stripping as illustrated in (a–d), no Subjacency effects are expected. Turning to the sloppy reading, let us consider the non-cm-version of (B), which is obtained by omitting the case marker. Under Hoji’s proposal, it has the structure in (), and the pro can refer to some property of an object such that John introduced to it (the/a) person(s) who wanted to join it. This property is predicated of Nissan; hence, Nissan is interpreted to have such a property, which results in the “sloppy” reading.12 ()

pro Nissan-mo Nissan-also

da. 

However, since what appears to be the sloppy reading results from the interpretation of the pro, not from a full-fledged structure, unlike in the case of cm-stripping, it is not surprising that the availability of the “sloppy” reading is not contingent upon the lexical and structural conditions in ().

... Sluicing As was first observed by Inoue (, ), Japanese has an ellipsis phenomenon that seems to be similar to what Ross (b) calls sluicing, as in (). The underlining indicates the correlate, and the angled brackets indicate the intended interpretation of the portion that is considered missing.13 ()

John-wa dareka-o suisensita ga, boku-wa [CP dare(-o) John- someone- recommended but I- who- (da) ka] siranai.  Q know.not ‘John recommended someone, but I don’t know who .’ (Inoue : , adapted)

Note that the case marker on the remnant wh-phrase is optional. Let us call sluicing with a case marker on the remnant cm-sluicing and sluicing without a case marker on the remnant non-cm-sluicing.14

The “sloppy” reading here is in fact equivalent to what Hoji () calls the sloppy-like reading; see section .... 13 Note that multiple sluicing, in which more than one wh-phrase appears, is possible as in (i). See Takahashi () and Nishigauchi (), among others, for discussion on this type of sluicing. (i) John-wa dareka-ni nanika-o watasita rasii ga, John- someone- something- handed seem but boku-wa dare-ni nani-o ka wakaranai. I- who- what- Q know.not ‘It seems that John handed something to someone, but I don’t know who what.’ 12

14

See Nakamura () for a view that distinguishes between case-marked NPs and PPs in sluicing. See also Nishiyama et al. () for some relevant discussion.





We will now see that sluicing exhibits quite similar properties to stripping. First, as illustrated in (), a copula can appear after the remnant. Second, cm-sluicing exhibits Subjacency effects while non-cm-sluicing does not. Takahashi () and Fukaya and Hoji () observe that Japanese cm-sluicing exhibits Subjacency effects, but since the effects do not seem to be detected very clearly with their examples, we will examine the effects in contrast sluicing, following Fukaya (, : ch. , ).15 Consider (). ()

boku-wa [CP keisatu-ga [NP (ISLAND) [proi [Tanakagiin]-ni wairo-o okutta] I- police- Rep.Tanaka-to bribe- gave otokoi]-o taihosita no]-wa sitteiru ga, man - arrested that- know but [CP [NP hoka-no dono giin]-ni ka]-wa siranai. other- which Rep.-to Q - know.not ‘I know that the police arrested the man who had given a bribe to Representative Tanaka, but I don’t know to which other Representative(s).’ (Fukaya : ())

() does not give rise to the reading for the second conjunct ‘I don’t know which other Representativej is such that the police arrested the man who had given a bribe to himj’, which the non-elliptical counterpart of () in () yields. ()

. . . [CP keisatu-ga [NP (ISLAND) [prok [hoka-no dono giin]-ni wairo-o police- other- which Rep.-to bribe- okutta] otokok]-o taihosita ka]-wa siranai. gave man - arrested Q - know.not ‘ . . . I don’t know which other Representative is such that the police arrested the man who had given a bribe to him.’ (Fukaya : ())

Fukaya claims that the unavailability of the reading mentioned above indicates that the second conjunct in () cannot have a structure comparable to that of (), where the entire TP containing the complex NP is reconstructed. Fukaya then argues that this is because the derivation of the second conjunct of () involves movement out of the complex NP. Note that in the non-cm-version of (), which is obtained by omitting the case marker on the remnant as in [hoka-no dono giin] ‘which other Representative’, the reading equivalent to () seems to be available. Third, both cm- and non-cm-sluicing give rise to sloppy readings, but the sloppy readings in the former are constrained by the lexical and structural conditions in () while those in the latter are not. Let us consider the cm-sluicing example in (),16 in which the sloppy reading is readily available. It gives rise to the reading ‘I remember that 15

It seems that speakers tend not to detect Subjacency effects as clearly as is claimed by Takahashi () and Fukaya and Hoji () in sluicing examples with an indefinite correlate, presumably because of the availability of what is comparable to the local-movement strategy as proposed by Merchant (). See Fukaya (: ch. , ) for a detailed discussion. 16 Note that () is an example comparable to what Merchant calls contrast sluicing. I use this type so that the correlate (and the remnant, as a result) will be the antecedent of the dependent term, just as in the case of stripping.





Toyota sued Toyota’s attorney last year, but I don’t remember [which other company]i sued itsi attorney last year.’ ()

[TP boku-wa [CP Toyota-ga sakunen [NP soko-no bengosi]-o uttaeta I- Toyota- last.year that.place- attorney- sued no] wa oboeteiru] ga, [TP [CP [NP hoka-no dono kaisya]-ga C  remember but other- which company- ka] wa oboeteinai]. Q  remember.not ‘I remember that Toyota sued its attorney last year, but I don’t remember which other company.’ (Fukaya : ch.  (), slightly adapted)

Replacing [+β] soko ‘that/the place’ by non-[+β] asoko ‘that place’ in () makes the sloppy reading unavailable. The sloppy reading is also unavailable if the c-command condition in (b) is not satisfied, as in (), in which the antecedent USC-o ‘USC-’ does not c-command the dependent term soko ‘that/the place’ in the first conjunct. ()

[TP[NP soko-no sotugyoosei]-ga USC-o suisensita] no wa that.place- graduate- USC- recommended C  oboeteiru ga, [CP [NP hoka-no dono daigaku]-o ka]-wa remember but other- which university- Q - oboeteinai. remember.not ‘I remember that its graduates recommended USC, but I don’t remember which other university.’ (Based on Fukaya : ch.  ())

() does not give rise to the sloppy reading ‘I remember that USC’s graduates recommended USC, but I don’t remember [which other university]i is such that itsi graduates recommended iti.’ The non-cm counterpart of (), which can be obtained by dropping the nominative marker on the remnant as in [hoka-no dono kaisya] ‘which other company’, also gives rise to the sloppy reading. The sloppy reading is available in the non-cm-version of () even if a non-[+β] term, such as asoko ‘that place’ or Toyota, is used as the dependent term. Note also that the non-cm counterpart of (), which can be obtained by omitting the accusative marker on the remnant as in [hoka-no dono daigaku] ‘which other university’, gives rise to the sloppy reading as well. These similarities between sluicing and stripping can be captured if cm-sluicing is analyzed analogously to cm-stripping, which is in turn to be analyzed as parallel to cmclefts, and if non-cm-sluicing is analyzed analogously to non-cm-stripping. This analysis of cm-sluicing is in line with the cleft analysis by Nishiyama et al. (), Shimoyama (), Kizu (b), Kuwabara (), Saito (),17 and Hiraiwa and Ishihara (), 17

As was clarified by one reviewer, Saito () proposes to analyze sluicing in Japanese as a “concealed cleft” with an elided CP subject, thus combining the cleft analysis by Nishiyama et al. () and the argument ellipsis analysis by Oku () and Kim () (see section ...), in order to solve the problem noted by Takahashi () and Nishiyama et al. (), i.e., the copula optionally appearing in sluicing.





among others.18 It is also in line with the analysis proposed by Hiraiwa and Ishihara () and later adopted by Nakao (: ch. ) and Nakamura (), among others, which derives sluicing from what they call the no da in-situ focus construction, because the no da in-situ focus construction is also assumed to underlie the cleft construction. It is also partially in consonance with the proposals by Hasegawa () and Abe (), who claim that Japanese has both cleft-type sluicing and genuine sluicing, the latter of which cannot be reduced to the cleft.19 The analysis of cm-sluicing as analogous to cm-stripping, however, is not compatible with the sluicing analysis by Takahashi (), who proposes to derive sluicing in Japanese in the same way as sluicing in English, or the one by Nishigauchi (, ), who proposes to derive Japanese sluicing with wh-remnants base-generated in the spec of CP, along the lines of Chung et al. ().20

... Ellipsis in comparatives Ellipsis is observed in Japanese comparatives as well (cf. Hoji a, b, ). Hoji () argues that two types of ellipsis can be identified in Japanese comparatives: casemarked comparatives (cm-comparatives) and non-case-marked comparatives (non-cmcomparatives), as exemplified in ().21 The underlining indicates the correlate. () cm- and non-cm-comparatives: sensei-wa [[John(-ni) yorimo] sakini] Bill-ni sono hon-o yom-aseta. teacher- John- than earlier Bill- that book- read-made ‘The teacher made Bill read that book earlier than John.’

18

See also Nakao and Yoshida () for an analysis of what they call Japanese Pronominal Sluicing as in (i) as a specificational pseudocleft. (i) John-ga dareka-ni atta ga, watasi-wa sore-ga dare-ni (da) ka sira-nai. John- someone- met but I- it- who-dat () Q know-not ‘John met someone, but I don’t know who (it is).’ (Nakao and Yoshida : (b), slightly adapted) 19

Hasegawa () distinguishes between the type of sluicing where the remnant resides in the embedded clause (embedded sluicing) and the type of sluicing where the remnant appears in the matrix clause (matrix sluicing), and argues that the former is best analyzed as a truncated cleft sentence while the latter should be analyzed on a par with English sluicing. Drawing on Hasegawa’s proposal, Abe (: ch. ) argues that matrix sluicing can be ambiguous between the cleft type and genuine sluicing and further that there are instances of genuine sluicing even in the embedded context (what he calls partially truncated ellipsis and the contrast type) in addition to the cleft type. See also Takita () for a claim that Japanese has genuine sluicing. Under the assumption that Japanese lacks overt feature-driven movement (see Fukui  and his subsequent works), which is adopted here, however, the possibility of the English type of sluicing is excluded. 20 I cannot provide a critical evaluation of each of the proposals here because it would require substantial discussion of what each analysis assumes and proposes, and what testable predictions are made under each analysis. The reader is referred to Fukaya (: ch. ) for some relevant discussion. 21 Hoji () identifies a third type of comparative, i.e., non-elliptical comparatives, as in (i), which presumably contains the null argument construction in the than-clause. (i)

. . . [[John-ni ec yom-aseru yorimo] sakini] . . . John- read-make than earlier ‘(lit.) . . . earlier than (he) made John read (that book).’





Notice that the NP John in the than-phrase in () can be case-marked or non-case-marked. We will now see that ellipsis in comparatives also exhibits properties quite similar to stripping. First, cm-comparatives exhibit Subjacency effects, as in (). ()

[AdvP

[UCLA-ni yorimo] sakini] sensei-wa [NP USC-ni googakusita UCLA- than earlier teacher- USC- was.accepted gakusei]-to hanasita. student-with talked ‘(lit.) The teacher talked with a student who had been accepted to USC earlier than to UCLA ( ≠ . . . than a student who had been accepted to UCLA).’

() yields the reading that the teacher talked with a student who had been accepted to USC earlier than he had been accepted to UCLA, but crucially, it does not give rise to the intended reading in the parentheses in the translation. That reading, however, is available in the non-cm-version of (), which we can obtain by omitting -ni on UCLA-ni. Now let us examine the availability of the sloppy reading in comparatives, based on Hoji (, : section .). The cm-comparative in () gives rise to both the strict and the sloppy reading. Note that the [+β] term so-ko ‘that/the place’ is used here. ()

[Toyotaj-ni yorimo sakini] seihu-wa Hondai-ni Toyota- than earlier government- Honda- [sokoi/j-no Arizona koozyoo]-o hihans-aseta. it- Arizona factory- criticize-made ‘The government made Honda criticize its (=soko’s) Arizona factory earlier than Toyota.’ (i) (strict) (ii) (sloppy) (Based on Hoji : ())

Substituting a non-[+β] term for the [+β] term makes the sloppy reading unavailable. Thus, if we replace soko ‘that place’ in () with asoko ‘that place over there’ or Honda, the sloppy reading becomes unavailable. On the other hand, in the non-cm-counterpart of (), which can be obtained by eliminating the case marker on the NP within the than-phrase as in Toyota yorimo ‘than Toyota’, both the strict and the sloppy readings seem to be available, irrespective of whether the dependent term is [+β] or not, unlike cm-comparatives. Thus, the sloppy reading continues to be available even if we replace soko-no ‘its’ with asoko-no ‘that place-’ or Honda-no ‘Honda’s’. The sloppy reading is unavailable in cm-comparatives if the antecedent does not c-command the dependent term, as can be seen in (). ()

[Toyotaj-ni yorimo sakini] [sokoi/j-to torihiki-no aru buhin Toyota- than earlier that.place-with dealings- have parts meekaa]-ga Hondai-ni kuzyoo-o itta (rasii). maker- Honda- complaint- said seem ‘The parts maker that has dealings with it complained to Honda earlier than Toyota.’





Soko ‘that place’ can be coreferential with either Toyota or Honda, although the former interpretation seems to be more salient presumably because Toyota precedes soko. Thus, () gives rise to both the reading that the parts maker that has dealings with Toyota complained to Honda earlier than it complained to Toyota and the reading that the parts maker that has dealings with Honda complained to Honda earlier than it complained to Toyota.22 But crucially, () does not give rise to the sloppy reading that the parts maker that has dealings with Honda complained to Honda earlier than the parts maker that has dealings with Toyota complained to Toyota. The non-cm-comparative counterpart of (), which can be obtained by omitting the dative marker -ni on Toyota-ni ‘Toyota-’, on the other hand, can yield the sloppy reading.

... Summary As we have seen in this section, there are striking similarities among stripping, sluicing, and comparatives in Japanese. Their cm-versions are subject to Subjacency and give rise to sloppy readings that are constrained by the conditions in (). Their non-cm-versions do not exhibit Subjacency effects, and they yield sloppy readings that are not contingent upon the conditions in (). It thus seems reasonable to assume that these three “constructions” are a uniform phenomenon, i.e, the cm-versions are derived from cm-clefts and the non-cm-versions have copula structures with a null element in the subject position.

.. Short answers As Nishigauchi () observes, Japanese allows short answers (fragment answers to use the term of Merchant a) to wh-questions, as in (B). Note that short answers are typically followed by a copula. () A: John-wa dare-ni sono hon-o {agemasita/ageta no desu} ka? John- who- that book- gave gave C   ‘Who did John give the book to? / Who is it that John gave the book to?’ B: Mary-ni desu. Mary-  ‘(lit.) It’s to Mary.’ (B) can be regarded as a short form of the non-elliptical answer in (). ()

22

B’: John-wa Mary-ni sono hon-o {agemasita/ageta no desu}. John- Mary- that book- gave gave C  ‘John gave the book to Mary. / It’s to Mary that John gave the book.’

I cannot address the issue of what formally underlies the coreference in () that gives rise to the strict readings just mentioned. See Ueyama (: ch. ) for some related discussion.





Since short answers can be either case-marked (cm-short answers) or non-case-marked (non-cm-short answers), it is suspected that they might also be analyzed along the lines of stripping, sluicing, and ellipsis in comparatives, which we discussed in the previous subsection. We will see, however, that short answers exhibit distinct properties with respect to Subjacency effects. Nishigauchi (: ch. , section .., : section ..) observes that in reply to the question in (A), for example, the answers in (B) and (B) are possible. Note that the wh-phrase is within a complex NP. () A: kimi-wa [NP dare-o egaita hon]-o yonda no? you- who- depicted book- read Q ‘(lit.) Who is such that you read a book that depicts him?’ B: Bill Gates desu. Bill Gates  ‘(lit.) (It’s) Bill Gates.’ B: [NP [NP Bill Gates]-o egaita hon] desu. Bill Gates- depicted book  ‘(lit.) (It’s) the book that depicts Bill Gates.’ (Nishigauchi : ch.  ()) He proposes an LF pied-piping analysis, according to which the entire NP that contains the wh-phrase, i.e., [dare-o egaita hon]-o ‘(lit.) the book that depicts who’, undergoes movement to specCP at LF, and this makes (B) a possible answer to the question in (A). He then claims that the short answer in (B) is derived by a discourse deletion process that deletes constituents recoverable from the context. Nishigauchi () claims that this process is constrained by a rule that states, “If deletion can be applied, apply it maximally; otherwise, do not apply it”;23 thus all the recoverable constituents must be deleted if deletion applies. An important observation here is that short answers seem to be insensitive to islands in Japanese, as indicated in (B). Note that the short answer in (B) can be case-marked as Bill Gates-o desu ‘Bill Gates- .’ Saito () argues that cm-short answers and non-cm-short answers each have two sources. One is the cleft and the other is the copula structure with a null pronoun in the subject position. He assumes, following Hoji (: ch. ) and Murasugi (), that cmcleft does, and non-cm-cleft does not, exhibit Subjacency effects. He also assumes that a phonetically null pronoun pro residing in the subject position in the copula structure gives rise to the meaning equivalent to that of the cleft and that the pro-subject copula structure does not induce Subjacency effects because it does not involve movement. Thus, the apparent island-violating example in (B) can be analyzed as a case of the non-cm-cleft as in (a) or a case of the copula structure as in (b), which is equivalent to (c) with an overt subject sore ‘that’.

23 This constraint is based on Kuno’s () proposal. Kuno examines cases where non-focus constituents are left undeleted as well and proposes a rule called “Ban Against Partial Discourse Deletion.” See also Kuno (a, ) and Kuno and Masunaga () for some related discussion.





() a. [CP [TP pro [pro

egaita hon]-o yonda] no]-wa depicted book- read C - Bill Gates desu. Bill Gates  ‘(lit.) It’s Bill Gates that I read a book that depicts _.’ b. pro Bill Gates desu. Bill Gates  ‘(It’s) Bill Gates.’ c. sore-wa Bill Gates that- Bill Gates ‘It’s Bill Gates.’

desu. 

In the case of cm-short answers, Saito argues that when the remnant in the second conjunct corresponds to an element within an island in the first conjunct, the cm-cleft analysis is not available because a cm-cleft is subject to Subjacency, and that the apparent lack of Subjacency effects is observed only if the pro-subject copula structure is possible. For Saito, the pro-subject in short answers is possible only if its overt counterpart is possible. It is thus expected that the apparent lack of Subjacency effects is observed only if the version with an overt subject as in (c) is acceptable. As an example which exhibits Subjacency effects, Saito cites the multiple cm-short answer example from Nishigauchi () in (). He then argues that this example is unacceptable because neither of the derivations is possible. The pro-subject copula structure is unavailable because the example with an overt pronominal subject as in (a) is impossible, and the cm-cleft as in (b) is ruled out by Subjacency. () A: [NP [TP

dare-ga dare-ni kaita] tegami]-ga mitukarimasita who- who-to wrote letter- found.was ‘(lit.) [A letter [that who wrote to whom]] was found?’

ka? Q

B: *Tanaka-san-ga Nakasone-san-ni desu. Tanaka-Mr- Nakasone-Mr-to  ‘It is [Mr Tanaka to Mr Nakasone].’ (Nishigauchi : ch.  () and (), with his judgment) () a. #sore-wa Tanaka-san-ga Nakasone-san-ni desu. it- Tanaka-Mr- Nakasone-Mr-to  ‘It is [Mr Tanaka to Mr Nakasone].’ (Saito : () with his judgment) b. *[CP Opi Opj [TP [NP [TP ti tj

kaita] tegami]-ga mitukatta] no] -wa wrote letter- found.was C - [Tanaka-san-gai Nakasone-san-nij] desu. Tanaka-Mr- Nakasone-Mr-to  ‘(lit.) It is [Mr Tanaka to Mr Nakasone] that [a letter that wrote] was found.’ (Saito : ())





Though (a) and (b) do seem to be degraded, (B) does not seem to be as bad as is claimed. In addition, the use of sore-wa as in () seems degraded even in a single short answer as in (B), but its corresponding short answer in (B’) is readily acceptable.24 () A: [NP [TP

Tanaka-san-ga dare-ni kaita] tegami]-ga mitukarimasita Tanaka-Mr- who-to wrote letter- found.was ‘(lit.) [A letter [that Mr Tanaka wrote to whom]] was found?’

B: #sore-wa Nakasone-san-ni it- Nakasone-Mr-to ‘It is to Mr Nakasone.’

ka? Q

desu. 

B’: Nakasone-san-ni desu. Nakasone-Mr-to  ‘To Mr Nakasone.’ Thus, it seems difficult to maintain that an apparently Subjacency-violating cm-short answer is possible only if its corresponding copula structure with an overt pronominal subject is possible. Abe (), adopting Kimura’s () in-situ analysis of sluicing, which derives sluicing without moving the wh-remnant from its base position, argues that short answers are derived from the no da in-situ focus construction without involving movement. Under his proposal, (B) (as an answer to (A)), for example, is derived by deletion as illustrated in (). () [FocP [CP

John-wa MARY-ni John- Mary- ‘John gave that book to Mary.’

sono that

hon-o book-

ageta gave

no] that

desu]. 

Notice that this derivation involves non-constituent deletion. Abe (: ) assumes that “a target of deletion is a constituent and yet an actual deletion operation applies to it in the way that a phrase carrying [Focus] evades such an operation.” Thus, the target of deletion (called E (llipsis)-site by Abe ) in () is the CP, but since the NP Mary-ni ‘Mary-’ carries [Focus], it is exempt, and the deletion takes place to the exclusion of this NP. Since there is no movement involved in the derivation, island insensitivity is as expected. Abe also argues that there is evidence that a full-fledged structure containing an island is in fact present in the ellipsis site. Consider (), which is modeled on the structure of Abe’s (). () A: [NP

kanari-no kazu-no nikkei kigyoo]-ga great.deal- number- Japanese company- [NP [TP nani-o mudande kizi-ni sita] zassi]-o what- without.permission article- made magazine- uttaeta no? sued Q ‘(lit.) A large number of Japanese companies sued the magazine that had written an article about what without permission?’

24

I would like to thank Hajime Hoji (p.c., April ) for pointing out the unavailability of (B) to me.



 B: [NP

soko-no sinseihin]-o it- new.product- ‘Its new products.’

desu. 

The bound variable construal seems to be available in (B), i.e., (B) seems to give rise to the reading ‘a large number of Japanese companies are such that [each of them]i sued the magazine that had written an article about itsi new products without permission.’ This indicates that the island insensitivity of short answers is not illusory as Fukaya (: ch. ) claims following Merchant (). Note also that the availability of the bound variable reading indicates that Saito’s proposal with a null pronominal subject cannot be maintained here because there needs to be a structure where the dependent term is bound by the QP. Neither is his cleft analysis available in this case, because the cm-cleft that would yield the reading is not available due to Subjacency. Despite their apparent similarities, cm-short answers exhibit properties distinct from those of cm-versions of stripping, sluicing, and ellipsis in comparatives with respect to Subjacency effects. This indicates that some other ellipsis resolution strategy must be available in this language. Some possibilities would be the discourse deletion analysis by Nishigauchi (), the in-situ analysis proposed by Abe (), and the island repair analysis by Nishigauchi and Fujii (), who follow Merchant (b) and Fox and Lasnik (), among others. The first approach seems to have a problem in dealing with multiple short answers across a clause boundary. Claiming that a multiple short answer is not possible when an overt subject intervenes between the wh-correlates as in (), Abe () suggests that what is at work in multiple short answers is a condition that constrains the relative distance between the two remnants, such as Specified Subject Condition. He then claims that such a syntactic condition cannot be employed in the discourse deletion analysis.25 () A: dare-ga [CP Akira-ga doko-de sono syasin-o totta to] itteita no? who- Akira- where that picture- took C said Q ‘Who said that Akira had taken that picture where?’ B: *Anna-ga Mosukuwa-de Anna- Moscow-in ‘Anna in Moscow.’

desu. 

(Abe : ())

With the second and the third approaches, we would need to address the issue of why ellipsis phenomena, such as stripping, sluicing, and ellipsis in comparatives, exhibit Subjacency effects, because if an analysis that evades or nullifies Subjacency effects is available for short answers in the grammar, it is expected to be applicable to other fragments as well, unless some other constraints are at work. I leave this issue for future research.

25 The unacceptability status of (B), however, does not seem to be as clear as Abe claims it to be. To the extent that examples like (B) are not clearly unacceptable, the discourse deletion analysis cannot be ruled out entirely.





.. Summary In this section, we have seen the properties of four types of fragment phenomena in Japanese: stripping, sluicing, ellipsis in comparatives, and short answers. It was observed that stripping, sluicing, and ellipsis in comparatives behave in a strikingly similar way with respect to Subjacency effects and the availability of sloppy readings. In addition, we saw that the presence and absence of the case marker on the remnant yields significant differences in the same way across these phenomena. Given such similarities, it was concluded that these three types of fragments can reasonably be assumed to be a uniform phenomenon in Japanese. In contrast, it was observed that cm-short answers are not subject to Subjacency, unlike the cm-versions of stripping, sluicing, and ellipsis in comparatives. This led us to conclude that short answers are in fact a different type of fragment phenomenon from stripping, sluicing, and ellipsis in comparatives. This divergence raises an issue for the analysis of ellipsis in Japanese and also for the general theory of ellipsis. Unless some factors hinder its application, the analysis responsible for the lack of Subjacency effects in cm-short answers should also be applicable to the cm-versions of stripping, sluicing, and ellipsis in comparatives, and this possibility would circumvent Subjacency effects in these phenomena as well, contrary to fact. This issue needs to be addressed in future research.

. R-N R: A-- , ,   ?

.................................................................................................................................. As Kuno (b) observes, Japanese has an elliptical construction that is comparable to Right-Node Raising, “which extracts the rightmost common constituent from conjuncts” (Kuno b: section ..). He assumes that (b), for example, is derived from (a).26 () a. Taroo ga Amerika-ni ik-i, Hanako ga Huransu-ni Taro- America-to go-ing Hanako- France-to ‘Taro went to America, and Hanako went to France.’

itta. went

b. Taroo ga Amerika-ni, Hanako-ga Huransu-ni, itta Taro- America-to Hanako- France-to went ‘Taro went to America, and Hanako, to France.’ (Kuno b: section . ()) Note that in contrast to Right-Node Raising in English, the relevant right node is a verb in Japanese, an SOV language.

26

In the ensuing discussion, I will use Right-Node Raising as a descriptive term without committing myself to a particular theory.





There are two major strains for the analyses of Right-Node Raising in the literature: (i) one that involves across-the-board movement of the rightmost common constituents (Ross ; Maling ; Postal , ; Bresnan ; Hudson b; Hankamer , among others) and (ii) one that involves deletion in the first conjunct (Neijt ; Wexler and Culicover ; Banfield ; Levine ; Kayne ; Wilder ; Hartmann , among others).27 For the similar phenomenon in Japanese illustrated in (), three types of analyses have been proposed: (i) the across-the-board movement analysis, which corresponds to the first strain above, (ii) the gapping analysis, and (iii) the String Deletion analysis, which corresponds to the second strain. In this section, we will examine those three types of analyses, especially with respect to Subjacency effects. Kuno (b: ) and Saito (), among others, propose to analyze Right-Node Raising as being derived from across-the-board movement of the nodes at the right edge of conjoined clauses. Saito argues that in (), for example, the dative- and the accusative-marked NPs in the first and the second conjuncts raise and adjoin to TP, resulting in (a), as illustrated by Abe and Hoshi (: section ). Then, the underlined lowest TP undergoes across-the-board rightward movement to adjoin to the highest TP, as in (b). ()

John-ni hana-o sosite Bill-ni tyoko-o John-to flower- and Bill-to chocolate- Mary-ga okutta. Mary- sent ‘(lit.) To John _ flowers, and Mary sent chocolates to Bill. (= Mary sent flowers to John, and Mary sent chocolates to Bill.’ (Abe and Hoshi (: ())

() a. [TP [TP John-nii [TP hana-oj [TP Mary-ga [VP ti [V’ tj okut]] ta]]], sosite [TP Bill-nik [TP tyoko-om [TP Mary-ga [VP tk [V’ tm okut]] ta]]]]. b. [TP [TP [TP John-ni [TP hana-o tTP ]], sosite [TP Bill-ni [TP tyoko-o tTP ]]] [TP Mary-ga [VP t [V’ t okut]] ta]].

(Based on Abe and Hoshi (: ())

Note that the raising of the NPs in the first conjunct (let us call them the remnants) and the NPs in the second conjunct (let us call them the correlates) is necessary; otherwise, there would be no constituents in the first and the second conjuncts that are parallel to each other.28 Abe and Hoshi () propose to analyze Right-Node Raising in Japanese as analogous to gapping. Slightly modifying Jayaseelan’s () analysis of English gapping, they propose

27 28

See Chapter  for more references. Under the copy theory of movement, however, the TPs are not parallel to each other.





that Right-Node Raising as in (a) is to be analyzed as (b), as illustrated by Abe and Nakao (). ()

a. John-ga ringo-o, (sosite) Mary-ga mikan-o tabeta. John- apple- and Mary- orange- ate ‘(lit.) John _ an apple, and Mary ate an orange. (= John ate an apple, and Mary ate an orange.)’ (Abe and Nakao : ()) b. [TP John-ga John- [TP Mary-ga Mary-

[T’ ringo-oi [T’ ti tabeta]]], soshite apple- ate and [T’ mikan-oj [T’ tj tabeta]]]. orange- ate (Abe and Nakao : ())

The first conjunct is base-generated with an empty T’ to which the accusative-marked NP is adjoined, as in [TP John-ga [T’ ringo-oi [T’ Ø ]]]. In the second conjunct, the accusativemarked NP undergoes leftward movement and adjoins to T’, resulting in the second conjunct in (b). Then, the underlined T’ in the second conjunct is copied onto the empty T’ in the first conjunct, resulting in the first conjunct in (b). Mukai () argues that what she calls String Deletion is responsible for Right-Node Raising, based upon the fact that Right-Node Raising is insensitive to islands as indicated in (). () Mike-ga raion-ni, Tom-ga [NP kuma-ni osowareta Mike- lion- Tom- bear- was.being.attacked otoko]-o tasuketa. man- saved ‘Mike saved the man who was being attacked by a lion, and Tom saved the man who was being attacked by a bear.’ (Mukai : ()) Note that the underlined correlate kuma-ni ‘bear-’ resides within the complex NP [NP kuma-ni osowareta otoko]-o ‘the man who was being attacked by a bear’. String Deletion is an operation that deletes a phonetic string under PF identity, regardless of its constituency (cf. Mukai : section .). It applies to the phonetic string osowareta otoko-o tasuketa ‘saved the man who was being attacked’ as in (), resulting in (). ()

Mike-ga [raion-ni osowareta otoko]-o tasuketa, Tom-ga [kuma-ni osowareta otoko]-o tasuketa.

(Mukai : ())

Note that the phonetic string subject to String Deletion is not a syntactic constituent. Mukai argues that the across-the-board movement analysis and the gapping analysis, both of which involve movement, would fail to account for the fact that Right-Node Raising is insensitive to islands. Abe and Nakao () argue against the String Deletion analysis, claiming that RightNode Raising is in fact sensitive to islands as in ().





() *John-wa kuma-ni, sosite Mary-wa John- bear-by and Mary- koto-o minna-ni itta. fact- everyone- told

raion-ni lion-by

Bill-ga Bill-

osowareta attacked

‘(lit.) John _ by a bear, and Mary told everyone that Bill was attacked by a lion. (= John told everyone that Bill was attacked by a bear, and Mary told everyone that Bill was attacked by a lion.)’ (Abe and Nakao : () with their judgment) They argue that the unacceptability of () is captured under the gapping analysis because the remnant undergoes movement out of the complex NP to adjoin to the T’, as illustrated in ().29, 30 () *[TP John-wa [T’ kuma-nii [T’ [ Bill-ga ti John- bear-by Bill- osowareta koto]-o minna-ni itta]]]. attacked fact - everyone- told (Abe and Nakao : (), with their judgment) They claim that under the String Deletion analysis, on the other hand, the relevant movement can take place within the island as indicated in () and hence no island violation should be detected. () John-wa [ John- itta. told

kuma-nii bear-by

Bill-ga ti Bill-

osowareta attacked

koto]-o fact-

minna-ni everyone-

(Abe and Nakao : ())

Abe and Nakao argue that the movement of the remnant out of the island takes place even in cases like () but that the relevant movement is string-vacuous, i.e., it does not affect the word order, and that “string-vacuous movement is always ‘covert’ in that the bottom copy of a movement chain is pronounced” (Abe and Nakao : section .) and is not islandsensitive, following Abe and Hornstein (). Abe and Nakao provide () as a case in which non-string-vacuous movement is involved. Although they claim that it is unacceptable, the judgment does not seem to be as clear as is reported. To examine island effects more clearly, let us turn to complex NPs with a relative clause, which are known to exhibit clearer island effects in Japanese than complex NPs with a CP complement. Consider ().

29 Note that Abe and Nakao () present their proposal under the deletion analysis, unlike Abe and Hoshi (). The choice between copying and deletion does not affect the argument here. 30 Another type of data Abe and Nakao provide to argue against the String Deletion analysis is cases where the remnant and the correlate are adjuncts, as illustrated in their (). Although they claim that Subjacency effects are observed in such cases, the judgment does not seem to be as clear as is reported there.





() John-wa hana-o, (sosite) Bill-wa [NP tyokoi-o John- flower- and Bill- chocolate- Tom-ga ti ageta hito]-o sagasiteiru. Tom- gave person- is.looking.for ‘(lit.) John _ flowers, and Bill is looking for the person who Tom gave chocolates to. (= John is looking for the person who Tom gave flowers to and Bill is looking for the person who Tom gave chocolates to.)’ This example seems to be acceptable, although we should expect stronger island effects here than in (). Thus, it seems reasonable to consider that the relevant movement takes place within the islands in the first and the second conjuncts in () and () in a fashion comparable to (), and that String Deletion applies to a non-constituent, resulting in the Right-Node Raising construction, along the lines of Mukai (). Both the across-the-board movement analysis and the gapping analysis, on the other hand, involve movement out of an island in examples like (), (), and (); hence, the lack of Subjacency effects cannot be captured in these approaches. In this section I have surveyed three different proposals for Right-Node Raising in Japanese: the across-the-board movement analysis, the gapping analysis, and the String Deletion analysis. We have seen that Right-Node Raising does not exhibit Subjacency effects and concluded that the String Deletion analysis can best capture this fact.

. N  :   -?

.................................................................................................................................. In Japanese apparently missing arguments are allowed, as in (). () Taroo-ga ec tabeta. Taro- ate ‘(lit.) Taro ate ec.’ Several analyses have been proposed for such null elements. Kuroda () regards the empty object as a phonetically null pronoun. Hasegawa (/), following Huang (), claims that the null object is a variable bound by a phonetically unrealized topic phrase. Otani and Whitman () argue that it is a manifestation of a null VP, adapting Huang’s (a, ) analysis of the null object in Chinese. Kim () and Oku () maintain that arguments in Japanese and Korean can be elided. This position has been further defended by Saito (, ), Takahashi (, a, b, ), and Takita (), among others. The VP-ellipsis analysis of the null object construction is challenged by Hoji (), who argues that the null object construction in Japanese cannot be analyzed as being





analogous to VP-ellipsis in English. In this section, I will discuss two main types of analyses of the null argument construction: the ellipsis analysis (which subsumes the VP-ellipsis analysis and the argument ellipsis analysis) and the non-ellipsis analysis.

.. Analyses of the null argument construction in Japanese ... The ellipsis analysis 1: The VP ellipsis analysis As summarized by Murasugi (: –), citing Hinds () and Kuno (b), Japanese does not have VP-ellipsis of the type found in English, as illustrated in () and (). () *Taroo-ga [VP kaetta]-node Taro- left -because ‘I left because Taro did.’

watasi-mo I-also

kaetta. left (Murasugi : –)

() Taroo-wa [VP huransugo-ga hanas]-eru ga, Taro- French- speak-can but *Ziroo-wa [VP huransugo-ga hanas]-e-nai. Jiro- French- speak-can-not ‘Taro can speak French, but Jiro cannot.’ The lack of an equivalent to English do-support in Japanese makes ellipsis in a context like () impossible. () is unacceptable because auxiliaries in Japanese are bound morphemes that have to be attached to verb stems, in contrast to English auxiliaries, which are free morphemes that can stand alone, as pointed out in Kuno (b). VP-ellipsis as observed in English is thus not attested in Japanese, and neither is pseudogapping, which involves VP-ellipsis. Otani and Whitman (), however, argue that the null object construction in Japanese can be analyzed as an instance of a null VP, along the lines of Huang (a, ). They claim that (), for instance, can have the structure represented in (a) after the overt raising of the V to T and that the null VP is equivalent to VP-ellipsis in English as in (b). The content of the null VP in (a) is obtained by copying a derived VP available in the discourse (cf. Williams b). () a. [TP Taroo-ga [VP ec Taro- ‘Taro ate _.’

tv] [V tabe]-[T ta]]. eat-

b. John ate an apple, and Bill did [VP ec], too. One crucial argument they put forth supporting their proposal is that the null object construction in Japanese allows the sloppy reading, as in (), just like English VP-ellipsis. (b) allows the sloppy reading in (b-i), in addition to the strict identity reading in (b-ii).





() a. John-wa [zibun-no tegami]-o suteta. John- self- letter- discarded ‘Johni threw out selfi’s letters.’ b. Mary-mo ec suteta. Mary-also discarded (i) ‘Maryi also threw out selfi’s letters.’ (ii) ‘Mary also threw out John’s letters.’

(Otani and Whitman : ())

... The non-ellipsis analysis Hoji () argues that the sloppy reading available in the null object construction in Japanese as seen in () is not a genuine sloppy reading but the “sloppy-like” reading, which can arise pragmatically, and that a more stringent type of sloppy reading, which arises only based on a full-fledged structure, is unavailable in this construction. Consider one of his examples in (). He claims that (B) does not yield the reading indicated in (), unlike its VP-ellipsis counterpart in English in (). It only yields the reading that every American couple recommended a/some student(s). () A: [Subete-no nihonzin huuhu]-ga [betubetu-no all- Japanese couple- different- suisensita. recommended ‘Every Japanese couple recommended different students.’ B: [Subete-no amerikazin huuhu]-mo ec suisensita. all- American couple-also recommended ‘(lit.) Every American couple recommended ec, too.’

gakusei]-o student-

(Hoji : ())

()

For each of the American couples the student(s) the husband recommended is/are different from the student(s) the wife recommended, as well.

()

Every Japanese couple recommended different students; and every American couple did too.

If (B) could be analyzed on a par with VP-ellipsis in English, it would be able to yield the interpretation in (). This is in contrast to cm-comparatives, which Hoji (b) claims to be analogous to VP-ellipsis in English and readily give rise to the reading in question (see Hoji’s b discussion on his (b)). Assuming that null elements reside in argument positions in this construction (thus, no ellipsis is involved), Hoji (:) suggests that “the content of the N head of the null argument is supplied by the context of discourse” and that the content “is most likely a feature bundle, excluding phonological features.” He then identifies two





possible sources for the sloppy-like reading: the referential use and the concept use of the null argument.31 If the content of the N-head supplied by the context of discourse (“supplied N head”) is a name, it can be coreferential with another name. In (), for example, the “supplied N head” can be (a feature bundle corresponding to) the name Bill, in which case the null argument can be coreferential with Bill in the same clause. () A: John-ga zibunzisin-o suisensita. John- self- recommended ‘John recommended himself.’ B: Billi-mo eci suisensita. Bill-also recommended ‘Billi also recommended eci.’

(Hoji : ())

Turning to the second source, if the “supplied N head” is (a feature bundle corresponding to) a bare nominal, the null argument can give rise to various readings that a bare nominal can give rise to.32 Hoji maintains that the null argument in (B), for example, is more analogous to booru-o ‘ball-’ than to soitu-no booru-o ‘that.guy- ball-’. () A: [subete-no itinensei]i-ga [soitui-no all- first.year.student- that.guy- ‘Every first-year studenti kicked hisi or heri ball.’ B: [subete-no ninensei]-mo ec ketta. all- second.year.student-also kicked ‘Every second-year student also kicked ec.’

booru]-o ball-

ketta. kicked

(Hoji : ())

To confirm that the sloppy-like reading is not a genuine sloppy reading, which is considered to be based on bound variable anaphora under the standard assumption, Hoji (: section ) presents four sets of observations regarding the sloppy-like reading. Here I will take up the two that I consider to be of greater relevance to the discussion in this chapter. First, Hoji observes that the sloppy-like reading in the null object construction holds, even if the dependent term is a name. Even if zibun-no tegami-o ‘self ’s letter’ is replaced by John-no tegami-o ‘John’s letter’ in (), the reading in (b-i) continues to be available. Second, the sloppy-like reading in the null object construction is available even if the antecedent does not c-command the dependent term, as in ().33

31 Tomioka () argues that the null argument in Japanese denotes a contextually salient property and demonstrates how various interpretations are obtained semantically and pragmatically, basically in line with Hoji’s proposal. 32 As Hoji (: ) notes, the bare nominal booru ‘ball’, for example, can give rise to readings that correspond to ‘a ball’, ‘ball’, ‘the ball’, ‘balls’, etc. See Kuroda (: ch. ). 33 But see n. .

 () A: [NP [ec

mukasi Johni-o osieta] sensei]-ga karei years ago John- taught teacher- him homete iru. is.praising ‘[The teacher who taught Johni years ago] is praising himi.’



(no koto)-o (about)-

B: [NP [ec

mukasi Billj-o osieta] sensei]-mo ecj hometeiru. years.ago Bill- taught teacher-also is.praising ‘[The teacher who taught Billj years ago] too is praising ecj.’ (Hoji : ())

This contrasts sharply with a genuine sloppy reading based on bound variable anaphora, which is not available if the relevant c-command relation is missing.34 These facts are unexpected if the null object construction is derived analogously to VP-ellipsis in English and cm-comparatives in Japanese and the “sloppy reading” observed in this construction is a genuine sloppy reading based on bound variable anaphora. Under Hoji’s proposal, however, it is not clearly stated how the various interpretations of the null argument can be derived except that “the content of the N head of the null argument is supplied by the context of discourse” (Hoji : ) and that the “supplied N head” can be (a feature bundle corresponding to) a name or a bare nominal. As we will see in the next subsection, the null argument seems to give rise to an interpretation beyond what a bare nominal seems to yield. It is thus an issue for this account to determine how the interpretation beyond the “supplied N head” can be retrieved from the context (if it is meant to give rise to a rigorously testable prediction). An attempt to do so, however, might not lead to an analysis that makes definite predictions because, according to Hoji’s proposal, the interpretation of the null argument is not structurally based but pragmatically controlled and such aspects of interpretation, in principle, cannot be stated in syntactic terms clearly enough to yield definite predictions.35

... The ellipsis analysis 2: The argument ellipsis analysis The argument ellipsis analysis has been developed out of the VP-ellipsis analysis by Otani and Whitman () to account for a wider range of facts than can be covered by the

34 In this connection, Hoji (: –) notes the English VP-ellipsis example in (i), which Fiengo and May (: –) cite from Dalrymple et al. (), and which is originally attributed to Michael Wescoat.

(i) The policeman who arrested John read him his rights and the one who arrested Bill did too. Although the relevant c-command relation is absent in (i), the sloppy reading is available. Hoji suggests that the relevant reading is indeed the sloppy-like reading and points out that it continues to be available for many speakers even if him is substituted by John, acknowledging that how VP-ellipsis in English gives rise to the sloppy-like reading needs to be accounted for. 35 And that seems to be what Hoji () intends his proposal in question to be; see Hoji (: preface, .).





VP-ellipsis analysis. Oku () observes that the sloppy reading is possible with a null subject as well, as in ().36 ()

Mary-wa [[zibun-no teian]-ga saiyosareru Mary- self- proposal- is.accepted ‘Mary thinks that her proposal will be accepted.’ John-mo [ ec saiyosareru· to] omotteiru. John-also is.accepted C think ‘(lit.) John also thinks that ec will be accepted.’

to] C

omotteiru. think

(Oku : )

Oku (: ) points out that () yields both the sloppy reading ‘John also thinks that John’s proposal will be accepted’ and the strict reading ‘John also thinks that Mary’s proposal will be accepted.’ He then argues that the availability of the sloppy reading is not contingent upon the availability of VP-ellipsis. Oku also points out that it is difficult or impossible to get the interpretation of (b) where the VP-adverb teineini ‘carefully’ is included. () a. Bill-wa kuruma-o teineini Bill- car- carefully ‘Bill washed the car carefully.’

aratta. washed

b. John-wa ec arawa-nakat-ta. John- wash-not- ‘(lit.) John didn’t wash ec.’

(Oku : )

(b) does not give rise to the interpretation ‘John didn’t wash the car carefully’. This suggests, Oku argues, that it is not VP-ellipsis that brings about the sloppy readings observed in Otani and Whitman. In order to account for these facts in addition to those observed in Otani and Whitman, Oku proposes an argument ellipsis analysis.37 He argues that a sentence in which an argument is apparently missing in fact does not contain the argument in overt syntax and that the relevant argument in the antecedent clause is copied into the

36

Regarding (), one reviewer suggested that non-passive examples be used in order to ensure that the null subject is indeed situated outside the VP. Consider the non-passive example in (i). (i) Mary-wa [[zibun-no teian]-ga sairyoo da Mary- self- proposal- best  ‘Mary thinks that her proposal is the best.’

to] omotteiru. C think

John-mo [ ec sairyoo da to] omotteiru. John-also best  C think ‘(lit.) John also thinks that ec is the best.’ The sloppy reading is available in this case as well. 37 See Kim () for a similar proposal based on Korean data.





missing argument position in covert syntax.38 Thus, under his theory, (b), for example, is base-generated as (a), and the object [zibun-no tegami]-o ‘self ’s letter’ in the antecedent is copied into it, resulting in (b). ()

a. [TP Mary-mo Mary-also

[VP sute]-ta]. discard-

b. [TP Mary-mo [VP [NP zibun-no tegami]-o Mary-also self- letter- ‘Maryj also threw out selfj’s letters.’

sute]-ta]. discard-

Saito () provides examples where the null argument receives an interpretation that a pronoun cannot yield, thereby arguing against the analysis of the null argument as pro in favor of the argument ellipsis analysis. Saito points out that “if pro is simply a pronoun without phonetic content, we would expect it to be definite in interpretation” (Saito : section ).39 He also presents examples where the indefinite interpretation of pro does not yield the correct interpretation. Consider (). () a. Sensei-wa [subete-no itinensei] -ni [zibuni-no booru]-o teacher- all- first.grader- self- ball- ‘The teacher let all first-graders kick their own balls.’

ker-aseta. kick-made

b. Demo, ninensei -ni -wa ec ker-ase-nakatta. but second.grader-- kick-make-did.not ‘But she/he did not let the second-graders kick their own balls.’ (Saito : ()) Saito (: section ) claims that (b) “is appropriate in the situation where the teacher did let the second-graders kick balls but just did not allow them to use their own” and that an example with an overt indefinite booru-o ‘ball-’ in the place of the null object means that the teacher did not let the second-graders kick balls. In addition, Shinohara () cited in Saito (), Takahashi (a,b), and Takita () provide examples in which the null argument is interpreted as a quantified NP (e.g., okyaku-o san-kumi-izyoo ‘more than three groups of guests’ and taitei-no NP ‘most NP’) and a

38 Oku attempts to explain why LF copying of a missing argument is allowed in Japanese by postulating à la Bošković and Takahashi () that the θ-features are weak in this language, thereby relating the availability of scrambling with the availability of null arguments that allow the sloppy reading; see Oku : ch. . Saito () provides an alternative view of the availability of LF copying of a missing argument in Japanese, claiming that it is allowed because Japanese lacks obligatory agreement as proposed by Kuroda () and supported by Takahashi (, ). Takahashi () examines the proposals with data from Chinese and Turkish, and concludes that Saito’s analysis better captures the facts. I would like to thank a reviewer for clarifying the theoretical relations mentioned above. See also Takahashi (a) for a related discussion. 39 Hoji (), in fact, does not claim that null arguments are “pronouns.” It has been clear since Hoji (a: section , especially n. ), that he believes that Japanese lacks elements, overt or covert, that are [+pronominal, anaphor] in binding-theoretic terms, i.e., that Japanese does not have “pronouns” of the sort that are observed in English.





negative polarity item (e.g., nanimo ‘anything’ and NP-sika ‘only NP’), which is not possible with pronouns. In order to investigate whether a full-fledged structure is indeed present in the null argument position, let us examine the availability of a more stringent type of sloppy reading, based on the discussion in Hoji (: section ). As is observed by Hoji (b: section , : section .), cm-comparatives in Japanese exhibit the same pattern of judgments as VP-ellipsis in English with regard to the availability of mixed readings, which were originally observed by Dahl () and subsequently discussed by Sag (a), Dalrymple et al. (), Fiengo and May (: ch. ), and Fox (: ch. ), among others. Hoji claims that () and () exhibit the pattern of judgments indicated in () and (), respectively. ()

[[Bill-ni yorimo] saki-ni] sensei-wa John-ni [CP kare-ga Bill- than earlier teacher- John- he- ruumumeeto]-o butta to] iw-aseta. roommate- hit C say-made ‘The teacher made John say he hit his roommate earlier than Bill.’

[kare-no he-

() [[Bill-ni yorimo] saki-ni] sensei-wa John-ni [CP [kare-no Bill- than earlier teacher- John- he- ruumumeeto]-ga kare-o butta to] iw-aseta. roommate - he- hit C say-made ‘The teacher made John say his roommate hit him earlier than Bill.’ ()

The teacher made Johni say hei hit hisi roommate earlier than . . . a. the teacher made Billj say hei hit hisi roommate. b. the teacher made Billj say hej hit hisj roommate. c. the teacher made Billj say hej hit hisi roommate. d. *the teacher made Billj say hei hit hisj roommate.

()

The teacher made Johni say hisi roommate hit himi earlier than . . . a. the teacher made Billj say hisi roommate hit himi. b. the teacher made Billj say hisj roommate hit himj. c. the teacher made Billj say hisj roommate hit himi. d. the teacher made Billj say hisi roommate hit himj.

Notice that the first kare ‘he’ c-commands the second in (), while the first kare does not c-command the second in (). The across-the-board strict readings in (a) and (a) and the across-the-board sloppy readings in (b) and (b) are available in () and (), respectively. The Bill-Bill-John readings in (c) and (c) (let us call it Mix  reading, following Hoji b) are also available in () and (), respectively. However, the BillJohn-Bill readings in (d) and (d) (let us call it Mix  reading, also following Hoji b) are not available if the first kare c-commands the second as in (), while it is available if the first kare does not c-command the second as in (). This pattern of judgments is exactly what is observed in English VP-ellipsis.





If the null argument construction is indeed an instance of argument ellipsis and has a full-fledged structure at LF, it is then reasonable to expect that it exhibits the same properties as cm-comparatives. Let us consider () and (). () A: John-wa [NP [kare-ga [kare-no ruumumeeto]-o John- he- he- roommate- uwasa]-o hiteisita. rumor- denied ‘John denied the rumor that he hit his roommate.’

butta hit

toyuu] that

butta hit

toyuu] that

B: Bill-mo ec hiteisita. Bill-also denied ‘(lit.) Bill also denied.’ () A. John-wa [NP [[kare-no ruumumeeto]-ga kare-o John- he- roommate- he- uwasa]-o hiteisita. rumor denied ‘John denied the rumor that his roommate hit him.’ B: Bill-mo ec hiteisita. Bill-also denied ‘Bill also denied.’ In (B) and (B), both the across-the-board strict reading and the across-the-board sloppy reading are available, while mixed readings seem to be unavailable.40, 41 It then seems reasonable to assume that the null argument construction cannot have a full-fledged structure at LF required to give rise to the mixed-reading paradigm, in contrast to VP-ellipsis in English and cm-comparatives in Japanese. Note that this is in contradiction to both the VP-ellipsis analysis and the argument ellipsis analysis because both analyses allow a full-fledged syntactic structure for the ellipsis site.

40 Some speakers might find mixed readings to be available in the null argument construction, as one reviewer does. I suspect that, even for those speakers, we do not observe the pattern of judgments, including the contrast in the availability of Mix  readings mentioned in the main text. Ultimately, what is crucial is whether or not we obtain the pattern of judgments for the mixed-reading paradigm, rather than whether or not mixed readings are available. In order to justify this claim, we must (i) have hypotheses that give rise to the predicted pattern of judgments in the mixed-reading paradigms and (ii) obtain experimental results in accordance with the prediction. But we are not yet in a position to do either of these. See n.  and the surrounding text. 41 Under the argument ellipsis analysis, it would be expected that the mixed readings are also available in non-ellipitical comparatives mentioned in n.  because they presumably contain the null argument construction in the than-clause. As Hoji (: section .) claims, however, non-elliptical comparatives do not seem to give rise to mixed readings, and thus a full-fledged structure does not seem to be available in the than-clause. This is an indication that, like the non-cm-versions of stripping, sluicing, and ellipsis in comparatives, the null argument construction is an instance of deep anaphora, which is pragmatically controlled and hence renders it impossible to make definite predictions in syntactic terms. See also section ..





.. Summary This section has addressed the issue of how the null argument construction in Japanese should be analyzed. We have examined two main types of approaches: the ellipsis analysis (the VP-ellipsis analysis and the argument ellipsis analysis) and the non-ellipsis analysis. We have seen that some predictions the VP-ellipsis analysis makes regarding the availability of the sloppy reading have been disconfirmed (cf. Hoji ) and also that the unavailability of the couple-internal reading (cf. Hoji ), the VP-adverb (cf. Oku ), and the mixed readings cast doubt on the plausibility of the analysis. These facts indicate that the VP-ellipsis analysis cannot be maintained. We have also observed that the non-ellipsis and the argument ellipsis analyses both have shortcomings. In the non-ellipsis analysis, it cannot be clearly stated, for principled reasons, how much interpretation can be retrieved from the context for the “supplied N head” because pragmatically controlled aspects of interpretation cannot be stated in syntactic terms clearly enough to yield definite predictions. The argument ellipsis analysis too has been claimed to have difficulty with cases of sloppy readings of a more stringent type, i.e., mixed readings. When considering both the advantages and disadvantages of these two approaches, the question of which approach is more plausible cannot be settled at this point. It certainly warrants further research.

. N’-

.................................................................................................................................. Saito and Murasugi (b) observe that, as illustrated in (a), Japanese has a phenomenon comparable to what is called N’-deletion in English as in (b).42, 43 () a. [gakubusei-no sensei-e-no izon]-wa yurusareru ga, undergraduate- teacher-on- reliance - tolerate.can though [insei-no _ ]-wa yurusarenai. grad.student-- tolerate.cannot ‘I can tolerate undergraduates’ reliance on the faculty, but not the graduate students’.’ (Saito and Murasugi b: (a)) b. Bill’s story about Sue may be amazing, but Max’s _ is virtually incredible. (Jackendoff : ()) In these examples the NP in the second conjunct is elided to the exclusion of the genitive-marked NP. In this section, I will survey the characteristics of this construction 42 Although N’-deletion is reanalyzed as NP-deletion in Saito and Murasugi (b), I will continue to use the term N’-deletion to refer to the phenomenon in this chapter. 43 Kamio () observes that no can be a proform for a concrete noun but not for an abstract noun. In order to exclude the possibility that no is a proform of a noun, which is comparable to one in English, rather than the genitive marker, Saito and Murasugi (b) use abstract nouns like izon ‘reliance’ as the head noun in their discussion of this phenomenon.





in Japanese, mainly based on Saito and Murasugi (b), Murasugi (), and Saito et al. ().44 One of the characteristics of N’-deletion in Japanese observed by Saito and Murasugi (b) and Saito et al. () is that there seems to be an argument–adjunct asymmetry in the possibility of deletion.45 Consider the examples in () and (). The no-marked phrase is the semantic object of the head noun in () and N’-deletion is possible, whereas the nomarked phrase in (), which acts as a modifier, is regarded as an adjunct and N’-deletion is not possible. () [Rooma no hakai] -wa Rome no destruction- [Kyooto no hakai] –yorimo hisan datta. Kyoto no destruction-than miserable was ‘Rome’s destruction was more miserable than Kyoto’s.’ (Saito and Murasugi b: (b)) () *saikin-wa [hare no hi] -ga recently- clear no day- [ame no hi] -yorimo ooi. rain no day -than plentiful ‘Recently, there have been more clear days than rainy days.’ (Saito and Murasugi b: (a)) Saito et al. () observe that no appears in a wider range of examples than the possessive marker in English. They claim that the head noun cannot be elided in the numeral+classifier+no+noun sequence as in (), for example. () *Taroo-wa iti-niti-ni [san-satu no hon]-o yomu ga, Taro- one-day-in three-CL no book- read though Hanako-wa [go-satu no hon]-o yomu. Hanako- five-CL no book- read ‘Taro reads three books in a day, but Hanako reads five.’ (Saito et al. : (b)) In order to account for the contrast between () and (), Saito and Murasugi (b) propose that, just as in the case of English, Japanese has a DP projection above the NP and

44

Miyamoto () contains a more detailed survey of this phenomenon. This dichotomy, however, is not without a problem. Consider (i), in which an adjunct seems to be able to “license” N’-deletion. 45

(i) [yuki-no siizun]-wa warito mizikai kedo, [ame-no]-wa zuibunto nagai. snow-no season- relatively short though rain-no- considerably long ‘Though the snow season is relatively short, the rainy season is very long.’ (Hoji : ()) If an adjunct can license N’-deletion, as in (i), the DP analysis of N’-deletion will be difficult to maintain. See also Kinsui () for a related discussion.





that a genitive-marked phrase needs to be in the spec of DP in order to license the deletion of the complement NP. They also propose that the argument and adjunct genitive phrases are both base-generated within the NP projection in the complement of DP and that an argument genitive phrase can move to the spec of DP while such movement of an adjunct genitive phrase is not possible, as indicated in (). They claim that this is due to the generalization that “non-expletives without a θ-role are in general unable to undergo NP movement” (Saito and Murasugi b: ). () a. . . . [DP[inseii-no] graduate- [NP ti sensei-e-no izon] D ]-wa . . . teacher-on- reliance - ‘ . . . graduate students’ reliance on the faculty . . . ’ b. . . . *[DP [amei-no] [NP ti hi] D ] -yorimo . . . rain- day -than ‘ . . . than rainy days’ In the case of argument genitive phrases, the NP in the complement of DP can be deleted after the movement, stranding the genitive phrase in the spec of DP. In the case of adjunct genitive phrases, on the other hand, the NP in the complement of DP cannot be deleted to the exclusion of the genitive phrase because the phrase cannot move out of the NP. The unacceptability of (), Saito et al. () argue, indicates that the numeral+classifier sequence is also an adjunct and cannot undergo movement to the spec of DP.46 If N’-deletion is an ellipsis phenomenon that has a full-fledged structure in the ellipsis site at LF, we expect that it exhibits properties that we observed in cm-stripping, cmsluicing, and cm-comparatives. In what follows, I will examine how N’-deletion behaves with respect to the availability of sloppy readings. First, the sloppy reading is expected to be available in N’-deletion. Consider (). ()

[A-sya-no [soko-to torihiki-ga aru buhin meekaa]-e-no A-company- that.place-with dealings- have parts maker-to- yookyuu]-wa B-sya-no yori kibisii rasii. demand- B-company- than severe seem (i) ‘I heard Company Ai’s demand on a parts maker that has dealings with iti is more severe than Company Bj’s demand on a parts maker that has dealings with iti.’ (ii) ‘I heard Company Ai’s demand on a parts maker that has dealings with iti is more severe than Company Bj’s demand on a parts maker that has dealings with itj.’

() gives rise to both the strict and the sloppy reading in (.i) and (.ii), respectively, although the sloppy reading is more salient than the strict reading because the latter is 46

Saito et al. () assume, following Kitagawa and Ross (), that no is a modifying marker that is inserted by the Mod-insertion rule in (i). (i) Mod-Insertion: [NP . . . XP Nα ] -> [NP XP Mod Nα], where Mod=no (Saito et al. : (), adapted)





pragmatically rather odd. The availability of the sloppy reading is as expected if the ellipsis site has the structure in () prior to the deletion of the NP. ()

. . . [DP

B-syai-no [NP ti [sokoi-to B-company- that.place-with meekaa]-e-no yookyuu] D] yori . . . maker-to- demand than

torihiki-ga dealings-

aru have

buhin parts

More importantly, if the sloppy reading is solely based on this structure, the analysis predicts that substituting the [+β] term by a non-[+β] term makes the sloppy reading unavailable. Substituting soko ‘that place’ by asoko ‘that place’ or A-sya ‘Company A’ in (), however, does not seem to make the sloppy reading unavailable.47 This leads us to conclude that the “sloppy” reading available in () is not a genuine sloppy reading but a sloppy-like reading and that the reading is not based on the full-fledged structure in (). Another test would be the mixed-reading test. Consider (). () a. John-no [NP [CP kare-ga kare-no ruumumeeto-o tasuketa toyuu] John- he- he- roommate- helped that syutyoo]-wa sinzirareru ga, Bill-no-wa sinzirarenai. claim- believe.can but Bill-- believe.cannot ‘We can believe John’s claim that he helped his roommate, but we cannot believe Bill’s.’ b. John-no [NP [CP kare-no ruumumeeto-ga kare-o tasuketa toyuu] John- he- roommate- he- helped that syutyoo]-wa sinzirareru ga, Bill-no-wa sinzirarenai. claim- believe.can but Bill-- believe.cannot ‘We can believe John’s claim that his roommate helped him, but we cannot believe Bill’s.’ The mixed-reading paradigm seems to be observed in (). That is, (a) gives rise to the readings in (a–c) but not (d), whereas (b) yields all the readings in (). ()

47

We can believe Johni’s claim that hei helped hisi roommate, but we cannot believe . . . a. Billj’s claim that hei helped hisi roommate. (ATB strict) b. Billj’s claim that hej helped hisj roommate. (ATB sloppy) c. Billj’s claim that hej helped hisi roommate. (Mix ) d. *Billj’s claim that hei helped hisj roommate. (Mix )

Another prediction it makes would be that the sloppy reading is unavailable if the antecedent does not c-command the dependent term, but since we should check cases where the antecedent does not precede the dependent term because of the complications alluded to in n. , we cannot construct appropriate examples of N’-deletion to test this.

 ()

 We can believe Johni’s claim that hisi roommate helped himi, but we cannot believe . . . a. Billj’s claim that hisi roommate helped himi. (ATB strict) b. Billj’s claim that hisj roommate helped himj. (ATB sloppy) c. Billj’s claim that hisj roommate helped himi. (Mix ) d. Billj’s claim that hisi roommate helped himj. (Mix )

This fact indicates that there can be a full-fledged structure in the ellipsis site. Given this and the conclusion that the “sloppy” reading in () is not based on a full-fledged structure, we are led to conclude that the N’-deletion construction can be ambiguous between two structural possibilities. One has a full-fledged structure in the complement of DP at LF, as was proposed by Saito and Murasugi (b). The other is the non-elliptical structure in (a), along the lines of Kadowaki (), which has no full-fledged structure in the complement of DP at LF. This parallels the structure with the overt expression sore ‘that’ in that position, as in (b). () a. [DP NP- [NP ec ] D] b. [DP NP- [NP sore] D] I wish to suggest that the structure in (a) makes the sloppy-like reading possible through the pragmatic interpretation of the null element (hence, even without satisfying the lexical requirement) and also that a full-fledged structure, as proposed by Saito and Murasugi, results in the availability of the mixed-reading paradigm. In this section we have surveyed some properties of the construction in Japanese that is comparable to so-called N’-deletion in English, based on Saito and Murasugi (b), Murasugi (), and Saito et al. (). We have also observed that the sloppy reading is available in this construction even if the lexical condition on the dependent term is not satisfied, which indicates that the “sloppy reading” available in such cases is in fact the sloppy-like reading. Also observing that the mixed-reading paradigm is detectable, I have suggested a possibility that the N’-deletion construction is structurally ambiguous between the structure with deletion, which has a full-fledged structure in the complement of DP at LF, and the non-elliptical structure, which lacks such a full-fledged structure at LF.

. S

.................................................................................................................................. In this chapter, I have surveyed various types of elliptical phenomena in Japanese. In section ., we observed that stripping, sluicing, and ellipsis in comparatives exhibit parallel properties and suggested that they can reasonably be assumed to be a uniform phenomenon in this language. We then concluded that their properties can be best captured if their cm-versions are analyzed on a par with cm-clefts and their non-cmversions are analyzed as a copula structure with a null subject. We also saw in section . that short answers behave differently from the other three types of fragments with respect to Subjacency effects and concluded that despite their surface similarity to the other types





of fragments, they should be regarded as a distinct phenomenon. Section . investigated the properties of Right-Node Raising in terms of Subjacency effects, and it was claimed that this phenomenon can be best captured by the non-constituent String Deletion analysis. In section ., we examined the ellipsis analysis and the non-ellipsis analysis of the null argument construction, and observed that both types of approaches have problems that need further attention. In section . some properties of N’-deletion were examined, and it was suggested that in order to capture the facts regarding the availability of the sloppy reading, we need a non-ellipsis analysis in addition to the ellipsis analysis. One significant aspect of the ellipsis phenomena in Japanese is that the presence and absence of a case marker on the remnant plays a crucial role. We have seen that casemarked fragments, such as cm-stripping, cm-sluicing, and cm-comparatives, are to be analyzed as instances of surface anaphora, which have a full-fledged structure in the ellipsis site, while non-case-marked fragments, such as non-cm-stripping, non-cm-sluicing, and non-cm-comparatives, are to be analyzed as instances of deep anaphora, which are controlled pragmatically rather than syntactically, as was originally proposed by Hoji (: ch. ). Although non-cm-versions can be instances of surface anaphora as briefly mentioned in n. , it is not possible to make definite predictions with non-cm-versions because they can also be instances of deep anaphora. Thus, it is vital to focus on cm-versions to make definite predictions. The use of cm-versions, however, does not guarantee rigorous testability because of the complication alluded to in n. . One way to pursue rigorous testability with ellipsis phenomena is to follow the methodology outlined in Hoji () (see also Hoji : preface), which would not only help us resolve some remaining issues pointed to in this chapter but might also suggest profitable general research directions for the future.48

A I would like to express my deep gratitude to Hajime Hoji for his extensive and very helpful feedback at various stages of the writing of this chapter, from which I benefitted greatly. I would also like to thank Takeo Kurafuji for his comments on a portion of the chapter. I am also indebted to three anonymous reviewers and the editors of this volume for their critical and constructive comments and suggestions, which led to considerable improvement of the chapter. I would also like to thank Mark Freiermuth for his help with the style. All remaining errors are of course my own.

48

Due to the space considerations, I did not directly address testability issues in this chapter. See Hoji (: preface, section .) for the difficulty one faces in pursuing rigorous testability in line with Hoji () when dealing with ellipsis-related phenomena. See also Hoji ( preface, section ) for the difference between testability-seeking and compatibility-seeking research.

  ......................................................................................................................

   ......................................................................................................................

́      

. I

.................................................................................................................................. T chapter investigates elliptical constructions in two closely related Bantu languages: Kiswahili and Shingazidja. To our knowledge, it is the first systematic study of ellipsis in any Bantu language. As we shall see, while Kiswahili and Shingazidja share many properties and grammatical features, they differ on some key aspects that are of importance for the analysis of ellipsis (e.g., the presence of the so-called “augment” in Shingazidja vs its absence in Kiswahili). The two languages also differ in the elliptical constructions they allow (e.g., right-node raising— section .), and the properties of elliptical constructions (e.g., gapping—section ..). The chapter is organized as follows. Section . gives a brief overview of Kiswahili and Shingazidja, and summarizes previous discussions of ellipsis in the literature on Bantu languages. V-stranding VP-ellipsis, a structure that has been discussed in detail in the literature on ellipsis in Bantu, is presented in section ., which also addresses the issue of object drop. Section . focuses on N-deletion, and section . deals with right-node raising. We explore gapping and related phenomena, such as stripping, in section .. In section ., we briefly examine null complement anaphora, and section . deals with sluicing. Section ., which examines fragments, is followed by a discussion of the absence of predicate ellipsis in G. languages (section .). Section . provides a brief conclusion.

. B

.................................................................................................................................. In this section, we provide some background on the languages of the study and previous discussions of ellipsis in Bantu.

  



.. Kiswahili and Shingazidja Kiswahili and Shingazidja are Bantu languages spoken in Eastern Africa. Kiswahili, referred to as G. in Guthrie’s (–) classification, is spoken in the coastal regions of Kenya and Tanzania, including Zanzibar. Shingazidja is one of the five Comorian languages (along with Shindzuani (Gb), Shimwali (Gc), Shimaore (Gd), and Shikombani). It is referred to as G.a in Guthrie’s (–) classification and is spoken on Grande Comore (or Ngazidja), the largest island of the Comoros. Shingazidja was formerly considered to be a dialect of Kiswahili, but they are now classified as distinct languages (Rombi and Alexandre ). The Kiswahili data for this chapter were mostly gathered and recorded in Tanzania by Sophie Manus in July–August ,1 but additional data, from a speaker of Kenyan Kiswahili,2 were collected in Paris in May . The Shingazidja data were recorded by Cédric Patin in Lille, France, between July  and February . One main speaker was consulted and recorded for each of the languages, and additional speakers provided input on particular or important points, but were not recorded. Like many Bantu languages, Kishawili and Shingazidja have a large number of noun classes, with each class marked by a prefix on the noun. Class prefixes trigger agreement markers that appear on determiners, the verb, etc. (). () Shingazidja3 a. nɗ=e=zí-fuɓa zi-ɾaɾu zí-nu =8=-chest -three -  ‘These three chests were beautiful.’ b. nɗ=e=mí-hono mi-ɾaɾu í-nu =4=-arm -three -  ‘These three arms were beautiful.’

zá-djisa .-be beautiful yá-djisa .-be beautiful

The basic word order is SVO (a), and the indirect object (benefactive, recipient, etc.) typically precedes the direct object (b).

1

Funded by the ANR-DFG project BantuPsyn / SynPhonI and Laboratoire CNRS DDL (Dynamique du Langage). 2 Examples collected from the speaker of the Kenyan variety of Kiswahili are identified as such in the text. 3 The following abbreviations are used in the glosses (numbers refer to agreement classes):  = augment;  = applicative;  = copula;   = distant past;  = demonstrative (  = demonstrative medial;   = demonstrative proximate);  = focus marker;  = future;  = negative;  = locative;  = object marker;  = passive;  = past;  = perfective;  = plural;  = possessive;  = present;  = pronoun; Q = question marker;  = reciprocal;  = relative marker;  = singular;  = stabilizer;  = subjunctive. In this study, a tonebearing unit is underlined. In (), for instance, the first two vowels of the word zádjisa ‘they are beautiful’ are underlined, meaning that the word has the following underlying form: /zádjísa/.



  

() Shingazidja a. ye=mw-aná ha-li ḿ̩ -kaʈe 1=-child .-eat -bread ‘The child ate (some) bread.’ b. ha-niká ye=m-limadji ɲ-úmɓa .-give 1=-farmer -house ‘He gave the farmer a house.’ Any word order is possible, however, when an object marker is prefixed to the verb (e.g., ())—but some orders are rare (e.g., SOV, VSO). ()

Shingazidja ze=m-ɓeɾé tsi-zi-ɾéŋge. 10=-ring .-10-take ‘The rings, I took them.’

The Shingazidja basic noun phrase linear order is briefly summarized in (a), and illustrated in (b) and (c). ()

Shingazidja a. Augment – Noun – Adjective (Adverb) (Adjective) – Demonstrative – Possessive – Quantifier – Relative b. ze=m-ɓuɗa m-ɓíli n-jeu 10=-stick -two -white ‘His two very white sticks.’ c. ze=m-ɓuɗa ḿ-titi 10=-stick -small ‘All these small sticks.’

hálisi very

zi-nu - 

z-a=háhe -of=1

pía all

While Kiswahili and Shingazidja share most of their grammatical features, they differ in some important ways. First, a definite noun in Shingazidja is preceded by the so-called “augment” (see examples ()–()), a category that is absent from Kiswahili. While it is a matter of controversy in the Bantu literature whether the augment is conditioned by syntactic/pragmatic contexts like Focus/Negation (Hyman and Katamba ) or whether its function is similar to articles (Buell ), its behavior in Shingazidja is clearly reminiscent of definite articles in Romance languages (). ()

Shingazidja a. m̩ -ɓuú ‘(a) baobab’ -baobab b. i-ɲáma ‘(an) animal’ -animal

vs

wo=m-ɓuú 3=-baobab

‘the baobab’

vs

ʃe=i-ɲáma 7=-animal

‘the animal’

  



Second, Shingazidja has retained a complex tone system (Cassimjee and Kisseberth ; Patin , among others), while Kiswahili is a toneless language that has regular penultimate group-final stress, marked by lengthening of the penult and a high pitch (Ashton , and many others). A final difference between the two languages is the fact that an animate object obligatorily triggers the presence of an object marker on the verb in Standard Kiswahili (a), while this is not the case in Shingazidja (b). ()

a. Kiswahili i. Juma a-li-mw-ona Juma --1-see ‘Juma saw Fatima.’

Fatima Fatima

ii. *Juma a-li-ona Fatima Juma --see Fatima Intended: ‘Juma saw Fatima.’ b. Shingazidja Djumwá ha-wono Juma .-see ‘Juma saw Fatima.’

Fátima Fatima

.. Previous studies on ellipsis in Bantu languages To our knowledge, very little attention has been paid to ellipsis in the literature on Bantu languages, and we are not aware of any systematic study of ellipsis in a Bantu language.4 A few examples, however, are discussed in the literature. Bresnan and Mchombo (), for instance, provide an interesting Chicheŵa (N.—Malawi) example of N-deletion5 in their well-known paper on the “lexical integrity principle”; the example is reproduced in (). ()

Chicheŵa (Bresnan and Mchombo :  (a))6 A-nyamǎta a-na-vín-á njerero pa bwaló lá -boy - -dance- name of dance  courtyard  mfúmú Kapanga ndí pá (bwaló) lá mfúmú Kapatuka. chief K. and  (courtyard)  chief K. ‘The boys danced the njerero dance on chief Kapanga’s courtyard and on chief Kapatuka’s (courtyard).’

In (), the second occurrence of the noun bwaló ‘courtyard’ is absent. Interestingly, the connective (or ‘associative’) -á, along with the agreement prefix - (cl. ) that refers to the elided noun, is maintained. 4 Khamisi () briefly discusses ellipsis in Kiswahili in a very short paper, which we were not able to consult. 5 They refer to it as “gapping.” 6 The glosses in this example follow the originals (numbers refer to classes):  = associative;  = subject;   = recent past.



  

Another example of ellipsis is provided by Halpert (), who briefly discusses an example of right-node raising7 in Zulu (S.—South Africa)—(). ()

Zulu (Halpert :  ())8 ngi-buk-e:la futhi (ngi-phinde) -watch- and -again ‘I watch and I (also) play soccer.’

ngi-dlale -play.

ibho:la .soccer

Crucially, two authors have provided extensive, theoretical discussions of specific elliptical structures in various Bantu languages. Paul R. Bassong extensively examines two elliptical constructions in Basàá (A.a—Cameroon) in his PhD thesis (Bassong ), namely sluicing and fragments;9 we will briefly discuss a few of his examples in the sections on these structures, i.e., section . and section . respectively. Deo Ngonyani discusses Kiswahili and Ndendeule (P.—Tanzania) in two papers (Ngonyani , ), and treats Chingoni (N.—Tanzania) and Kikuyu (E.—Kenya) in an article with Peter Githinji (Ngonyani and Githinji ). Taken together, these works propose an in-depth analysis of what Ngonyani refers to as VP-ellipsis, which will be discussed in the following section.

. V- VP-

.................................................................................................................................. This is an elliptical structure of Bantu, usually referred to as VP-ellipsis (but see below), that has been discussed in detail in the literature, in particular by Ngonyani (Ngonyani , ; Ngonyani and Githinji ). Consider the following sentences (), where an object is missing. () a. Ndendeule10 (Ngonyani —from Goldberg :  ())11 Joni a-ki-hemé nyumba na Malia a-ki-hemé John --buy .house and Mary --buy ‘John bought a house, and Mary bought (a house) also.’ b. Kikuyu12 (Ngonyani and Githinji :  (a)) Juma nĩ-a-ra-gũra nyumba ona Amina Juma ---buy house and Amina ‘Juma is buying a house and Amina is too.’13

helahe. also

nĩ-a-ra-gũr-a ---buy-

We discuss right-node raising in section .. The glosses in this example follow the originals (numbers refer to classes):  = first person singular;  = applicative;  = augment;  = subjunctive (?—the gloss is absent from Halpert’s  list of abbreviations). Halpert’s dissertation was published in  (Halpert ), but we were unfortunately unable to access this updated version of her research. 9 We thank an anonymous reviewer for pointing out this research to us. 10 P.—Tanzania. 11 The glosses in this example follow the originals (except for the so-called “final vowel” of the verb, which was separated from the root)—numbers refer to classes:  = focus marker;  = past;  = present tense;  = progressive aspect;  = subject marker. 12 E.—Kenya. 13 We kept the original translation here (and in (d)). A more literal translation, highlighting the ellipsis, would be: ‘ . . . and Amina is buying (a house)’. 7 8

   c. Chingoni14 (Ngonyani and Githinji :  (a)) Juma i-gula nyumba na Amina i-gula Juma -buy house and Amina -buy ‘Juma is buying a house and Amina is too.’



mewa. also

It is reasonable to wonder why, when reading these examples in which only the object of the second conjunct is elided, they are considered to be instances of VP-ellipsis rather than, e.g., N(P)-ellipsis. In Ngonyani’s analysis, however, the verb of the second clause in these examples moves to the I position, and the deletion of the object at the Phonological Form level results from an ellipsis that occurs at the VP level. The ellipsis in () differs from English-like VP-ellipsis in that it is not introduced by an auxiliary (as we shall see in section ., English-like VP-ellipsis is impossible in G. languages), but rather by a verb in a finite form. As a consequence, the elliptical structure in () should be labeled as V-stranding VP-ellipsis (henceforth VVPE—Goldberg ), rather than as VP-ellipsis. This type of elliptical construction is possible in G. languages: () a. Kiswahili Juma a-na-penda m-tindi (lakini) Saidi ha-pend-i Juma --like -yogurt (but) Saidi .-like- ‘Juma likes yogurt, (but) Saidi doesn’t like [EL15].’ b. Shingazidja Maɾ(i)=yé h(w)-andza̩ fromaʒí, ʃ(a)=Álii=yé k-andzá Mary=1 -like cheese but=Ali= 1 .-like ‘Mary likes cheese, but Ali does not like [EL].’ In both Kiswahili (a) and Shingazidja (b), the verb of the second CP is not followed by the expected complement NP (mtindi ‘yogurt’ in the Kiswahili example, fromáʒi16 ‘cheese’ in the Shingazidja example). The presence of the finite verb in sentences such as () and () makes it necessary to rule out an alternative analysis of these constructions that would involve Object Drop (henceforth OD). Building upon Ngonyani’s data and analysis, Goldberg () argued against such an analysis. In order to understand her point, consider the key Kiswahili examples in (). () Kiswahili (Goldberg : ,  (, )—Ngonyani ) a. Kamau a-li-m-beba m-toto Kamau ---carry -child ‘Kamau carried a child.’ b. Kamau a-li-m-beba Kamau ---carry ‘Kamau carried him (the child).’ 14

N.—Tanzania. We indicate the location of a possible ellipsis site with this neutral abbreviation (preferred to representations such as ‘yogurt’, which imply deletion). 16 When a Shingazidja word appears in the text outside the examples, it is transcribed as it is in isolation. 15



   c. *Kamau a-li-beba Kamau --carry ‘Kamau carried (a child).’ d.

Juma a-li-beba17 m-toto na Kamau a-li-beba Juma --carry -child and Kamau --carry ‘Juma carried a child, and Kamau carried (a child) too.’

pia too

(c) demonstrates that the absence of an argument after the verb in (b) is only possible thanks to the agreement class  object marker -m- that is prefixed to the verb root. As a consequence, the grammaticality of (d), where the verb of the second conjunct is neither followed by an object NP nor accompanied by an object marker, signals that examples such as (), (), and (d) are instances of ellipsis rather than cases of OD. As convincing as this argument may seem, it does not suffice to demonstrate the validity of the analysis. Indeed, in G. languages, and especially in a language such as Shingazidja where an animate object does not trigger the presence of an object marker on the verb, most of the verb forms can emerge in the intransitive form (i.e., without the presence of an overt object or an object marker); a sentence such as Saidi hapendi ‘Saidi does not like’ (see (a)) is acceptable in Kiswahili, for instance.18 Moreover, the rare verb forms that do require the presence of an object (marker) frequently disallow ellipsis—compare () and () with (). ()

Shingazidja a. Djumwá ka-dja-ka̩ya Juma NEG.-PER.NEG-be ‘Juma was not a good man.’

m-nɖu -person

w-á=heɾi -of=.good

b. *Djumwá ka-djá-kaya Juma NEG.-PER.NEG-be Intended: ‘Juma was not.’ c. *Alií ha-kaya m-nɖu w-á=heɾi ʃa=Djúmwa Ali .PER-be -person -of=.good but=Juma Intended: ‘Ali was a good man but Juma was not.’ ()

Shingazidja a. Djumwá ha-ʋumɓua Juma .PER-speak ‘Juma spoke to Ali.’

ka-djá-ka̩ya NEG.-PER.NEG-be

Álii Ali

17 Sic. The object marker is mandatory in such sentences in Standard Swahili, since the object of the verb is an animate (but it can be absent in casual speech in Kenyan Swahili, for instance). Note that (c) is to our knowledge ungrammatical in any form of Kiswahili. 18 Additionally, the rejection of some verb forms lacking an object—this is for instance the case for Alií ye kandzá ‘Ali does not like’ in Shingazidja (see (b))—may more likely result from problems of interpretability rather than from their potential ungrammaticality.

  



b. *Djumwá ha-ʋumɓúa Juma .PER-speak Intended: ‘Juma spoke to.’ c. *Djumwá ha-ʋumɓua Álii ʃa=Fátima ka-dja-ʋúmɓua Juma .PER-speak Ali but=Fatima NEG.-PER.NEG-be Intended: ‘Juma spoke to Ali but Fatima did not speak to.’ Ngonyani () provides several other arguments that favor an ellipsis analysis of examples such as (d) in Kiswahili over an OD analyis. First, he explains that there “is one fundamental distributional difference between the two gaps. [OD] is always an object gap and no constituent, other than objects, can be marked on the verb and leave a postverbal gap. In contrast, [VVPE] can be associated with constituents other than objects” (Ngonyani : )—see (a) and the corresponding Shingazidja example (b). ()

a. Kiswahili (Ngonyani :  ()19) m-kurugenzi a-li-taka ku-tembelea ki-wanda ch-ote na wa-kuu -director --want -visit -factory -all and -boss wa idara wa-li-taka ku-tembelea ki-wanda ch-ote pia of .department --want -visit -factory -all too ‘The director wanted to visit the entire factory and heads of department did too.’ b. Shingazidja ye=modiɾú ha-ka-handza yá-zuɾu le=ʃiɾika pi̩á 1=.director .--want ..-visit 5=.factory all n’=e=ðámana w-a=hé m-bawa ha-ka-handzá βa-(y)é and=1=.boss -of=1 -branch .--want - (= too) ‘The director wanted to visit the entire factory and the head of department wanted [EL] too.’

In the Kiswahili example (a), an infinitival complement clause is elided; in the corresponding Shingazidja example (b), the elided clause does not include a verb in the infinitival form, but in the relative form. Shingazidja differs from Kiswahili in that the word for ‘too’ is a complex pronoun in Shingazidja, βá-(y)e ‘also/too’, whose second part agrees with the subject of the second clause.20 This complex pronoun may also precede the verb, according to our informants, but this configuration is less natural. Another argument raised by Ngonyani in favor of a VVPE analysis of examples such as () relies on differences in the interpretation of examples that include similar structures, on the one hand, and sentences where the object is prefixed to the verb root, on the other hand. Consider ().

19

The glosses given by Ngonyani, in examples ()–(), were adapted to ours for purposes of clarity, but we kept his original translations. Crossing out the words that have undergone ellipsis in (a) is Ngonyani’s choice. 20 If the subject was in class , the form of this complex pronoun would be βá-(w)o.

 ()

   a. Kiswahili (Ngonyani :  ()) Juma a-li-beba mi-zigo y-ake na Juma --carry -luggage -his and ‘Juma carried his luggage and Jamila did too.’ i. = Jamila carried Juma’s luggage ii = Jamila carried Jamila’s luggage

Jamila Jamila

a-li-beb-a --carry

pia too

b. Shingazidja Djumwá ha-ɓaliyá ye=valizi y-á=hahe na=Fatimá ha-ɓaliyá Juma .-carry 9=.suitcase -of=1 and=Fatima .-carry βa-(y)é - ‘Juma carried his luggage and Fatima carried [EL] too.’ i. = ?Fatima carried Juma’s luggage ii = Fatima carried Fatima’s luggage The nature of the missing object in () is ambiguous:21 it may be the luggage of the subject of the first clause, Juma, or the luggage of the subject of the second clause, Jamila/Fatima. This ambiguous interpretation disappears when an object marker is prefixed to the verb of the second clause (). () a. Kiswahili (Ngonyani :  ()) Juma a-li-beba mi-zigo y-ake na Jamila a-li-i-beb-a pia Juma --carry -luggage -his and Jamila ---carry too ‘Juma carried his luggage and Jamila did too.’ i. = Jamila carried Juma’s luggage ii = *Jamila carried Jamila’s luggage b. Shingazidja Djumwá ha-ɓaliyá ye=valizi y-á=hahe Juma .-carry 9=.suitcase -of=1 ha-i-ɓáliya βá-(y)e .--carry - ‘Juma carried his luggage and Fatima carried it too.’ i. = Fatima carried Juma’s luggage ii = *Fatima carried Fatima’s luggage

na=Fatimá and=Fatima

In (), the luggage has to belong to Juma, the subject of the first clause in the sentence. A final difference between VVPE and OD is the fact that the postverbal NP of idiom chunks cannot be object-marked on the verb (()–()).

21

The situation is less clear in the Shingazidja example (b). We have chosen to translate Ngonyani’s Kiswahili example, and this had consequences for the choice of the Shingazidja example, but it has to be noted that true ambiguity was obtained for lookalike sentences in Shingazidja, such as ‘Djumwa has read his book and Fatima has read [EL] too’.

   ()



Kiswahili (Ngonyani :  (c–d)) a. Mumbi a-li-ku-la ki-apo Mumbi ---eat -oath ‘Mumbi took the oath.’ b. *Mumbi a-li-ki-la Mumbi ---eat Intended: ‘Mumbi took the oath.’

() Shingazidja a. u-henɗa ɗa̩hó=ni -walk .house=in ‘To do the Ãda (a rite of passage).’ b.

ha-l(i)-énɗe .--walk ‘He has been in it (a house).’ #‘He has done it (the Ãda).’

The noun kiapo ‘oath’ in the expression ‘took the oath’ cannot be pronominalized in Kiswahili (b), and this is also the case for the sequence ɗahó=ni ‘in the house’ of the expression ‘to do the Ãda’ (b). Both kiapo and ɗahó=ni can be subject to VVPE, however; see (). () a. Kiswahili (Ngonyani :  (b)) Mumbi a-li-ku-la ki-apo na Nioroge Mumbi ---eat -oath and Nioroge ‘Mumbi took the oath and Nioroge did too.’

a-li-ku-la ---eat

b. Shingazidja i. Djumwá h-enɗé ɗaho=ní na=Alií h-enɗé Juma .-walk .house=in and=Ali .-walk ‘Juma has done the Ãda and Ali did [EL] too.’

pia too

βa-(y)é -

ii. h-enɗé .-walk ‘He has done [EL].’ (answering the question: ‘Has Ali done the Ãda?’) From Ngonyani’s tests on Kiswahili and their application to Shingazidja, we conclude that examples such as () involve ellipsis, and are not instances of OD. Not all the complements of a verb can be elided, however. Ngonyani and Githinji (), in their discussion of the Bantu languages Kikuyu and Chingoni, show that the indirect object of constructions where the verb is in the applicative form cannot be the target of ellipsis. In the Shingazidja example (), only the direct object of the verb in the applicative form (the root -ndziʃía ‘write to’, in the applicative form, derives from -ndzíha ‘write’) can be elided.

 ()

   Shingazidja a. Djumwá ha-ndziʃia Álii Juma .-write. Ali ‘Juma wrote a letter to Ali . . . ’ b. i.

ɓaɾuá . . . .letter

. . . na=Saidí ha-ndziʃia Fátima and=Saidi .-write. Fatima ‘ . . . and Saidi wrote a letter to Fatima too.’

ii. . . . na=Saidí ha-ndziʃia Fátima and=Saidi .-write. Fatima ‘ . . . and Saidi wrote [EL] to Fatima too.’

ɓaɾuá .letter

βa-(y)é -

βá-(y)e -

iii. * . . . na=Saidí ha-ndziʃiá ɓaɾuá βa-(y)é and=Saidi .-write. .letter - Intended: ‘ . . . and Saidi wrote a letter to [EL] too.’ iv. * . . . na=Saidí ha-ndziʃiá βa-(y)é and=Saidi .-write. - Intended: ‘ . . . and Saidi wrote (lit. ‘wrote to’) [EL] too.’ v. . . . na=Saidí βa-(y)é and=Saidi - ‘ . . . and Saidi [EL] too.’ () shows that the indirect object of a verb in the applicative form cannot be elided under identity in Shingazidja, unless the verb in the applicative form is part of the ellipsis site ((b.v)—a case of stripping, see section ..). Ellipsis of the indirect object is indeed impossible if the direct object remains (b.iii), and even when both objects are absent from the second clause (b.iv). As for this latter case, Shingazidja differs from the languages studied by Ngonyani and Githinji (), which allow for such a construction.

. N-

.................................................................................................................................. When two DPs are coordinated, the noun of the second DP can be elided under identity conditions (i.e., when the elided noun is identical to the noun of the first DP). This type of noun ellipsis, N-deletion, is possible in both Kiswahili (a) and Shingazidja (b). ()

a. Kiswahili darasa-ni, m-toto w-a kw-anza na w-a mw-isho .classroom- -child -of -start and -of -end wa-na-fahamiana --understand.. ‘In the classroom, the first child and the last [EL] get along with each other.’

  



b. Shingazidja ho=klasi=ní,̩ e=m-naʃ(i)óni w-a=mw-a̩nɗo n’=é w-a=mw-isó 17=class=in 1=-student -of=-first and=1 -of=-last wo u-yeléwa 2 -understand e.o. ‘In the classroom, the first student and the last [EL] get along with each other.’ In (), the second occurrence of the word mtoto ‘child’/ mnaʃióni ‘student’ is elided. The elliptical nature of () is highlighted by the class  agreement morpheme w- prefixed to the connective -a that agrees with the elided noun.22 In his recent thesis dedicated to the noun in Shingazidja, Kassim Mohamed-Soyir () discusses similar cases in the numeral system. Examples of what he refers to as “l’accord elliptique des numéraux additifs” (the elliptical concord of additive numerals) are provided in (). ()

Shingazidja (Mohamed-Soyir :  (a–b)) a. wa-nɖu kume na wa-raru23 -person ten and -three ‘Thirteen persons’ (lit. ‘persons ten and [EL] three’) b. mi-ri kume na mi-ne -tree ten and -four ‘Fourteen trees’ (lit. ‘trees ten and [EL] four’)

In (a), there is no visible agreement between the word wánɖu ‘persons’ and the invariable word kúme ‘ten’, which is analyzed as noun-like in the sense of Corbett () by Mohamed-Soyir. The other numerals in (), on the contrary, agree with the head nouns wánɖu ‘persons’ and mirí ‘trees’.24 Note that the numeral adjectives that follow kume are in the ‘default’ class  when ‘thirteen’ or ‘fourteen’ is produced in isolation:

22 Two reviewers asked us if the second conjuncts in (b) could be seen as examples of free relatives. Such a hypothesis cannot be retained because the class  subject prefix of a relative verb (in the perfective as well as in the imperfective) is ya- (see (b), for instance), not wa-. For details on the morphosyntax of relatives in Shingazidja, see Patin (). 23 Mohamed-Soyir () and Ahmed-Chamanga () do not indicate the tones. The glosses have been slightly modified in (), and were added in () and (). 24 The corresponding structures in Kiswahili may or may not involve concord, depending on speech style. Compare (i.a)—formal speech, where there is no concord on ‘two’—and (i.b)—casual speech, where ‘two’ agrees with ‘persons’:

(i)

a. wa-tu kumi -persons ten b. wa-tu kumi -persons ten ‘Twelve persons’

na and na and

m-bili -two wa-wili -two

(formal speech) (casual speech)

 ()

   Shingazidja (Mohamed-Soyir :  (c–d)) a. kume na n-ɖaru ten and -three ‘thirteen’ b. kume na ten and ‘fourteen’

n-ne -four

There are two main competing analyses of the examples in (). The first one, sketched in Mohamed-Soyir (), assumes that two noun phrases are conjoined in (), and that there is an ellipsis of a noun—let us call this hypothesis H. The second one25 is that there is concord on the whole conjoined numeral, which spells out only on the numerals that trigger concord (i.e., -ráru ‘three’ vs kume ‘ten’); under such an analysis, the conjunction na links two adjectival phrases—let us call this hypothesis H. In our opinion, Mohamed-Soyir () provides a solid argument in favor of H. Consider the forms in (): ()

Shingazidja (Ahmed-Chamanga : ) a. ma-gari m-engo mi-raru na ma-raru -car -ten -three and -three ‘Thirty-three cars.’ b. zi-ri m-engo -chair -ten ‘Fifty-six chairs.’

mi-tsanu -five

na and

zi-ranɗaru -six

As can be observed in (), cardinal numbers such as ‘twenty’, ‘thirty’, etc. in Shingazidja consist of the class  noun méŋgo ‘ten, decade’ accompanied by an adjective that agrees in class with this noun. Additionally, the final adjective in these examples agrees with the head noun of the sequence, not with méŋgo ‘ten, decade’. If H were correct, the adjective in () would be joined to a noun phrase. However, the conjunction na can only join structures of the same nature in Shingazidja (Mohamed-Soyir : ). As a consequence, we consider that na does not join two adjectival phrases in (), but rather two noun phrases, with elision of the head of the second.

. R- 

.................................................................................................................................. Example () provides an apparent example of right-node raising (henceforth RNR) in Shingazidja. In this example accepted by all our speakers, the complement Zanzibar seems to be elided in the first conjunct (but see below).

25

We thank a reviewer for suggesting this alternative to us.

   ()



Shingazidja ha-ĩʃí ʃa=k̬a-dja-fa̩ɲá hazi Zanzibaɾ .-stay but =.-.-do .work Zanzibar ‘He has stayed [EL?] but not worked in Zanzibar.’

The possibility of similar examples in Kiswahili remains unclear. While our main speaker accepts the equivalent of () in the language, i.e. (), others strongly reject it as a possible Kiswahili sentence (and such is the case for other examples involving RNR). ()

Kiswahili Khamisi a-me-kaa lakini ha-ku-fanya kazi Khamisi --stay but .-.-do .work Intended: ‘Khamisi has stayed but not worked in Zanzibar.’

?

Zanzibar Zanzibar

There is no clear evidence in favor of an RNR analysis of () over an analysis involving object-drop. In order to distinguish RNR from sentences involving the coordination of two verbs where one has a null complement, RNR must be tested with verbs that trigger an obligatory complement (for instance, a sentence such as ‘Mary cooked and John ate the bananas’ is potentially ambiguous). However, the first verb of the sentence, haḭ ʃi, ̱ ́ can be pronounced in isolation, raising doubts as to the nature of the ellipsis in (); yet note that, in such cases, it has a different meaning: ‘he lives’. The difference in acceptance between () and () by our Kiswahili and Shingazidja speakers may, in fact, be linked to the possibility of dropping an object in the two languages. Consider (), for instance: () a. Kiswahili ? Saidi a-me-tengeneza jana, na a-me-endesha leo, pikipiki Saidi --repair yesterday and --drive today .motorbike y-ake m-pya ni-li-ye-mw-uzia - -new - -1-1-sell. Intended: ‘Saidi repaired yesterday and drove today the new motorbike I sold him.’ b. Shingazidja Saidí ha-ɾeŋgezá djana ha-ɓaliyá leo wo=moto Saidi .-repair yesterday .-drive today =.motorbike w-a=ɲúmeni na-m-hulíza -of=.new .--sell. ‘Saidi repaired yesterday and drove today the new motorbike I sold him.’ The Kiswahili example (a), as was the case for (), is accepted by our main speaker while others strongly reject it. On the contrary, none of our Shingazidja speakers rejects (b). Now compare () with ().



  

() a.

b.

Kiswahili *Saidi a-me-teŋgeneza Saidi --repair Intended: ‘Saidi repaired.’ Shingazidja Saidí ha-ɾeŋgéza Saidi .-repair ‘Saidi repaired.’

(a) is considered ungrammatical by all our Kiswahili speakers, including the speaker who accepts () and (a). (b), on the other hand, is possible according to the Shingazidja speakers. Thus, the acceptability of examples () and (b) may be due to the fact that Shingazidja allows dropping of the object. As a consequence, these examples may not be examples of RNR, but rather of object-drop. There are, however, two arguments in favor of an RNR analysis of () and (b). First, compare (b) with (), where the object precedes the adverb in the second clause. ()

Shingazidja Saidí ha-ɾeŋgezá djana, ha-ɓaliyá wo=moto Saidi .-repair yesterday .-drive =.motorbike w-a=ɲúmeni na-m-hulizá leo -of=.new .--sell. today ‘Saidi repaired [EL?] yesterday and drove the new motorbike I sold him today.’

While () can be seen as an alternative to (b), there is a clear difference in the interpretation of the two sentences according to our main informant. Indeed, it is clear in (b) that Saidi has repaired and driven the same motorbike, while there are doubts about whether he repaired a bike in ()—he may have repaired something else. This difference in the semantics of (b) and (), alongside the fact that the structure in (b), where the object follows the adverb, obeys the right-edge condition on RNR (), advocates in favor of an RNR analysis of (b). ()

Right-Edge Condition on RNR (Abels ) In a configuration of the form [XP1 . . . Y . . . ] conj [XP2 . . . Y . . . ], Y must be the rightmost element within XP and within XP before RNR may apply.

A second argument in favor of an RNR analysis of () and (b) comes from prosody. It has been observed in different languages that the shared material in RNR constructions is prosodically reduced: see for instance Féry and Hartmann () for German, and Kentner et al. () for English. This prosodic reduction has been observed in Kiswahili, Shingazidja, and Símákonde, a P. language spoken in Zanzibar (Manus and Patin ). Consider, for instance, the figure in (), which corresponds to ().

   ()



Pitch (Hz)

160

100

50 ∫a

haĩ∫í H

kadjafa̩nya ˇ H

hazi

Zanzibar

2.183

0 Time (s)

From the F0 curve, one can see that the word Zánzibaɾ, which should have a high tone on the first syllable according to the tone rules of the language, has no surface tone. Additionally, the word displays several properties that are typical of the shared material of RNR sentences, such as the fact that its pitch register26 and intensity are reduced more than for other sentence-final words in our Shingazidja data. Based upon these arguments, we assume that Shingazidja does have RNR structures. However, further research is required in order to demonstrate more decisively that examples such as () are not instances of object-drop. Still related to RNR in Shingazidja, two final points remain to be made. First, it seems that RNR is possible with an adjunct, as opposed to a direct object.27 ()

Shingazidja Djumwá ye u-siwuhá Aminá ye u-faɲa házi βo=ʈási Juma 1 -sleep Amina 1 -do .work 16=.morning ‘Juma (usually) sleeps [EL], (and) Amina (usually) works, in the morning.’

Second, the RNR structure is said by our informants to be infrequent with respect to corresponding sentences where the object only emerges after the first verb—compare () with ().28 () Shingazidja ha-ı̃ʃí Zanzibáɾ ʃa k̬a-dja-fa̩ɲá hazi .-stay Zanzibar but .-.-do .work ‘He has stayed in Zanzibar but not worked [EL].’

26

Note that the shared material does not exhibit the lowering of F0 that characterizes the word házi. We thank a reviewer for suggesting this possibility to us. 28 The repetition of ‘Zanzibar’ at the end of the sentence seems to be possible only in cases of repetition following a misunderstanding. 27



  

. G   

.................................................................................................................................. Clausal ellipsis has rarely been addressed in the literature on Bantu. In this section, we will discuss several cases of clausal ellipsis, starting with gapping constructions in section ... In section .., we extend our analysis to other types of clausal ellipsis, namely stripping and argument cluster coordination, focusing on structures where the verb and (one of) its complement(s) are elided.

.. Gapping In Shingazidja (a), as in other Bantu languages such as Símákonde (b), gapping is possible. In (), the potential second occurrence of the verb is elided, and no material is inserted to compensate for the ellipsis of the verb. () a. Shingazidja Djumwá hw-anza n-dróvi Saidí yembe Juma -like -banana Saidi .mango ‘Juma likes bananas, Saidi [EL] mangoes.’ b. Símákonde Júúma ankulóóta di-ndiísi Saíídi má-ngéela Juma ..want -banana Saidi -mango ‘Juma wants/likes bananas, Saidi [EL] mangoes.’ While there are a few exceptions in our data,29 a phonological phrase boundary generally follows the subject of the second conjunct, indicating that the prosodic break that separates the subject NP from the VP is maintained even though the verb is elided. Crucially, the entire verb, rather than just a part, must undergo elision (see also section .). In (), the verb in the second conjunct is in a compound tense (Ahmed-Chamanga ). In such a case, the verb of the second conjunct may or may not be gapped, but the ellipsis of the root lalíka ‘invite’ by itself is impossible (c), indicating that Shingazidja does not allow English-type VP-ellipsis or English-type pseudogapping. ()

Shingazidja a. Saidí ha-lalika Fátima Alií ŋgu-djo Saidi -invite Fatima Ali .- ‘Saidi invited Fatima, Ali will invite Mary.’

lalika invite

Máɾi Mary

b. Saidí ha-lalika Fátima, Alií Maɾí Saidi -invite Fatima Ali Mary ‘Saidi invited Fatima, Ali [EL] Mary.’

29

Mostly when the sentences are produced at a high speech rate.

  



c. *Saidí ha-lalika Fátima, Alií ŋgu-djo Maɾí Saidi -invite Fatima Ali .- Mary Intended: ‘Saidi invited Fatima, Ali will [EL] Mary.’ In Kiswahili, contrary to what is observed in (), the ellipsis of the verb is generally associated with the addition of new material. As we can see in (), the subject of the second conjunct in gapped structures (b–c) is optionally followed by the so-called comitative nae, a reduced form of na yeye ‘and/with him/her’ that is prosodically associated with the preceding noun, and which is absent from non-gapped sentences.30 Gapped sentences where the comitative is absent, such as (b), seem to be infrequent with respect to gapped sentences where it is present, such as (c). ()

Kiswahili a. Juma a-ta-kuja kesho, Saidi a-ta-kuja kesho Juma --come .tomorrow, Saidi --come .tomorrow ‘Juma will come tomorrow, Saidi will come the day after tomorrow.’

kutwa hours

b. Juma a-ta-kuja kesho, Saidi kesho kutwa Juma --come .tomorrow, Saidi .tomorrow hours ‘Juma will come tomorrow, Saidi [EL] the day after tomorrow.’ c. Juma a-ta-kuja kesho, Saidi na-e kesho kutwa Juma --come .tomorrow Saidi and- .tomorrow hours ‘Juma will come tomorrow, Saidi [and him] [EL] the day after tomorrow.’ In some contexts, the comitative element is obligatory. Examples such as (c), where no comitative element occurs, are rejected by our speakers. They put forth the idea that a sentence such as (c) would only be used and accepted by young people in contexts that allow the use of familiar language and slang. (c) would then be interpreted a joke, where Saidi Maembe refers to a person with Maembe as his last name. ()

Kiswahili a. Juma a-na-penda n-dizi, Saidi a-na-penda Juma --like -banana Saidi --like ‘Juma likes bananas, Saidi likes mangoes.’

ma-embe -mango

b. Juma a-na-penda n-dizi, Saidi na-e ma-embe Juma --like -banana Saidi and- -mango ‘Juma likes bananas, Saidi [as for him] [EL] mangoes.’ c.

*Juma a-na-penda n-dizi, Saidi ma-embe Juma --like -banana Saidi -mango

When this type of ambiguous interpretation is not possible, the comitative is optional, as illustrated in (). 30

This is the case for broad-focus sentences. The emergence of nae depends on parameters related to information structure.



  

Ellipsis of the verb is also possible in Kiswahili when the verb is associated with several objects, as in (). In other words, gapping with multiple remnants is allowed in Kiswahili (vs English, see for instance Reeve : ). ()

Kiswahili Juma a-na-uza samaki kwa wa-swahili Asha na-e ma-ziwa kwa wa-masai Juma --sell .fish to -Swahili Asha and- -milk to -Maasai ‘Juma sells fish to the Swahilis, Asha (as for her) [EL] milk to the Maasais.’

Note that the sentence in () is also possible without the direct object, indicating that the ellipsis of a verb may extend to its complements ()—see also section .. for a discussion of stripping. In such cases, it is understood that Asha also ‘sells fish’. () Kiswahili Juma a-na-uza samaki kwa wa-swahili Asha na-e kwa Juma --sell .fish to -Swahili Asha and- to ‘Juma sells fish to the Swahilis, Asha (as for her) [EL] to the Maasais.’

wa-masai -Maasai

Interestingly, a sentence such as () is unacceptable in Shingazidja, according to our speakers (b). () Shingazidja a. Djumwá ŋgu-(hu)liz’=o wa-sawahíli n-fí Juma .-sell to=2 -Swahili -fish Aiʃá ŋgu-(hu)liz’=o wa-búʃi dzíwa Aisha .-sell to=2 -Madagascan .milk ‘Juma sells fish to the Swahilis, Aisha sells milk to the Madagascans.’ b. *Djumwá ŋgu-(hu)liz’=o wa-sawahíli n-fí Juma .-sell to=2 -Swahili -fish Aiʃá (nɗ ’)31 o=wa-búʃi dzíwa Aisha  2=-Madagascan .milk Intended: ‘Juma sells fish to the Swahilis, Aisha [it is] milk to the Madagascans.’ However, a sentence where both the verb and the direct object are elided is accepted (). () Shingazidja Djumwá ŋgu-(hu)liz’=o wa-sawahíli n-fí halafw Aíʃa Juma .-sell to=2 -Swahili -fish then Aisha o=wa-buʃí ̥ 2=-Madagascan ‘Juma sells fish to the Swahilis, and Aisha [EL] to the Madagascans.’

The so-called “stabilizer” emerges in gapped sentences when the object is in the definite form—see the discussion of (). 31

  



One may conclude from the comparison of () with () that Shingazidja has a two-remnant restriction on gapping. However, the presence of a third remnant is possible as soon as the direct object immediately follows the subject, as in ()—compare this example with (b), where there is no inversion.32 () Shingazidja Djumwá ŋgu-(hu)liz’=o wa-sawahíli n-fí halafw Aíʃa dzíw’ Juma .-sell to=2 -Swahili -fish then Aisha .milk o=wabuʃí ̥ 2=.Madagascan ‘Juma sells fish to the Swahilis, and Aisha [EL] milk to the Madagascans.’ This latter configuration is highly interesting, given that the canonical word order in this language is S V IO DO, as explained in section ... It seems that no overt material, i.e. a prosodic word, can separate the subject from the direct object in gapping structures.33 Note that there are other interesting cases of inversion in gapping constructions in Shingazidja. In many cases, it is possible to invert the positions of the subject and the complement around the ellipsis ().34 ()

Shingazidja Djumwá ŋgu-djó-(h)u-djá mauɗu halaf ’ ú-ʃao Saídi 35 Juma .-- -go tomorrow then .day after tomorrow Saidi ‘Juma will come tomorrow, Saidi [EL] the day after tomorrow [lit. ‘day after tomorrow Saidi’].’

At this point in our research, it remains unclear under what circumstances this option is selected, but it may be due to the fact that contrastive topics usually emerge in sentenceinitial position.36 As for gapping in Shingazidja, two final points remain to be made. The first one relates to definiteness (recall from section .. that a definite noun is preceded by an augment in this language). When ellipsis is followed by an augment or a demonstrative, a so-called “stabilizer”—an “emphasis particle” (Ahmed-Chamanga : ) that introduces clefts and presentative sentences—generally appears after the elided material ().

32 A reviewer has suggested that the difference between () and () may lie in a difference in structure: the DO-IO canonical order would correspond to a true double object, while the reversed order would correspond to a prepositional dative. 33 One editor wonders whether we have any hypotheses that would explain this point. The challenge here is to develop an account that explains the difference between Shingazidja and Kiswahili, since the latter is not characterized by such a restriction. We think the difference may originate in the prosody of the two languages: the tone rules of Shingazidja would render opaque the relationships between the various remnants of the second clause of the sentence. Further research is needed to test this hypothesis. 34 Note that gapping is also possible without this inversion. 35 A monosyllabic root is preceded by the marker for class . 36 We thank a reviewer for suggesting this hypothesis to us.

 ()

   Shingazidja a. Djumwá ŋgu-djó-(h)u-djá mauɗu Saidí ŋgu-djó-(h)u-djá Juma .---go tomorrow Saidi .---go o=u-ʃáo 11=-day after tomorrow ‘Juma will come tomorrow, Saidi will come the day after tomorrow.’ b. Djumwá ŋgu-djó-(h)u-djá mauɗu Saidí Juma .---go tomorrow Saidi nɗ’=o=ú-ʃao =11=-day after tomorrow ‘Juma will come tomorrow, Saidi [EL] [it’s] the day after tomorrow.’

Thus, there is a difference in the structure of gapped sentences as a function of the indefiniteness (‘Paul gave apples, Mary bananas’) or definiteness (‘Paul gave the apples, Mary it is the bananas’) of the element that follows the ellipsis. A final property of gapping in Shingazidja is its potential to occur in an embedded clause ().37 ()

Shingazidja Djumwá hw-anza n-dróvi ŋgam-fiki̩ɾi Juma -like -banana .-think ‘Juma likes bananas, I think Saidi [EL] mangoes.’

Saidí Saidi

yembe .mango

.. Other cases of clausal ellipsis In this subsection, we will briefly discuss further examples involving clausal ellipsis. We will examine cases of stripping, argument cluster coordination, and comparative ellipsis. Stripping, generally defined as “a rule that deletes everything in a clause under identity with corresponding parts of the preceding clause, except for one constituent” (Hankamer and Sag : ), is widely considered to be closely related to gapping. As was the case for gapping, stripping is possible in G. languages ()–(). ()

Shingazidja Saidí ha-onehá haɾumwá e=ʃuhulí Saidi .-see. inside 10=.party ‘Saidi has been seen at the party, Juma [EL] too.’

Djumwá Juma

βa=(h)é 16-1

() a. Kiswahili (Kenya) i. Wa-tu w=engi wa-li-m-salimia Ali na Amina -person =many --1-greet. Ali and Amina a-li-m-salimia pia --1-greet. too ‘Many people have greeted Ali, and Amina greeted him too.’ 37

We thank an editor who suggested testing this structure. Unfortunately, we do not have Kiswahili data relating to this point.

   ii. Wa-tu w=engi wa-li-m-salimia Ali na -person =many --1-greet. Ali and ‘Many people have greeted Ali, and Amina [EL] too.’ iii. Wa-tu w=engi wa-li-wa-salimia Ali na -person =many --2-greet. Ali and ‘Many people have greeted Ali and Amina.’

Amina Amina



pia too

Amina Amina

b. Shingazidja i. Maɾ(i) (h)á-lauliya Álii, n(a)=Aminá βa-(h)é Mary .-greet Ali and=Amina 16- 1 ‘Mary greeted Ali, and Amina [EL] too.’ #‘Mary greeted Ali, and [EL] Amina (too).’ ii. Maɾ(i) (h)á-lauliya Álii n(a)=Amína Mary .-greet Ali and=Amina ‘Mary greeted Ali and Amina.’ Note that prosody helps to disambiguate the structure. A sentence such as (b.i), for example, cannot receive an interpretation where Mary greeted Amina. In sentences such as () or (b.i), the expansion of the register (pitch-ranged expansion) of the second conjunct, along with its intensity, is slightly reduced. This is not the case for the last word in (b.ii), for instance. In the preceding section, it was shown that ellipsis of the verb may be extended to a complement. (), which exemplifies argument cluster coordination,38 arguably demonstrates that the ellipsis may include another argument of the verb, its subject. () a. Kiswahili Saidi a-na-m-pa kaka y-ake vi-tabu na dada Saidi --1-give .older brother - -book and .older sister y-ake n-guo - -clothes ‘Saidi gives books to his older brother, and [EL] clothes to his older sister.’ b. Shingazidja Saidí ŋgu-nik-o Álii z-iyó Fatimá ŋ-guwo Saidi .-give- Ali -books Fatima -clothes ‘Saidi gives books to Ali, and [EL] clothes to Fatima.’ Note that inversion of the indirect object and the direct object in (b) is possible in Shingazidja, depending on the information status that is associated with the word ŋgúwo ‘clothes’. However, such inversion is impossible in Kiswahili: the animate object has to precede the inanimate. What is here called “argument cluster coordination” (henceforth ACC) may arguably be considered to be a form of gapping. According to Mouret’s () analysis of this structure in French, however, ACC differs from gapping in several aspects. For instance, “[ACC] may occur within NP, AP or PP with the same restriction, i.e. long-distance ACC is excluded [while gapping] only occur[s] in the sentential domain” (Mouret : ). 38



  

Another interesting case that involves clausal ellipsis is comparative ellipsis, which is illustrated in ().39 () a. Kiswahili Juma a-na-ku-la zaidi kuliko40 Juma ---eat more than ‘Juma eats more than Saidi [EL].’

Saidi Saidi

b. Shingazidja Djumwá hu-lá z-indjí ha=na Juma -eat -many for=with ‘Juma eats more than Saidi [EL].’

Saídi Saidi

Corresponding examples where the verb of the second conjunct is realized are ungrammatical—compare (a) and (b). ()

Shingazidja a. Djumwá ʋu-nwá dziwa naɗí ha=na Amína Juma -drink .milk often for=with Amina ‘Juma drinks milk more often than Amina [EL].’ b. *Djumwá ʋu-nwá dziwa naɗí ha=na Aminá ʋu-nwá Juma -drink .milk often for=with Amina -drink Intended: ‘Juma drinks milk more often than Amina drinks [EL].’

Structures where the subject and the complement of the second conjunct surround the elided verb, i.e. cases of gapping in comparative contexts, such as comparative subdeletion, are also rejected (). ()

Shingazidja *Djumwá ha-hulu má-gaɾi y-a=ãlí ha=na Alií ɲ-umɓá Juma .-buy -car -of=expensive for=with Ali -house Intended: ‘Juma bought more expensive cars than Ali [EL] houses.’

In order to express the idea in (), Shingazidja must adopt an alternative strategy: the use of possession, as illustrated in ().

39 One of the editors has suggested an alternative analysis of (a) as a “true” phrasal comparative, with kuliko as a preposition-like element. Despite our efforts, we were not able to obtain any evidence in favor of either of the two hypotheses. Further research is required to resolve this issue. 40 Interestingly, the informant speaking Kenyan Swahili proposed a version of this sentence lacking Saidi, with kumliko (i.e., kuliko with a class  object marker: ‘than him’) in the final position of the sentence. Such a configuration seems to be impossible in Standard Kiswahili.

   ()



Shingazidja Djumwá ha-hulu má-gaɾi y-a=ãlí ha=na (ye=)ɲ-umɓa Juma .-buy -car -of=expensive for=with (10)=-house y-[a]=Álii -of=Ali ‘Juma bought more expensive cars than Ali [EL] (the) houses [lit. ‘than (the) houses of Ali’]’.

In (), the buyer Ali emerges as a complement of the noun ɲumɓá ‘houses’. Another example of subdeletion is provided in (). This example demonstrates that comparative sentences where two verbs occur are possible, so long as the structure of the second conjunct does not mirror the structure of the first. () Shingazidja ŋgwa-hulá-o ɲ-umɓá z-indjí ha=na=(y)é ma-gaɾi ɾí-hula-o .-buy- -houses -many for=with=6 -car .-buy- ‘They buy more houses than we buy cars [lit. ‘cars that we buy’].’ In (), the noun of the second object is the head of a relative clause41—the verb of the second conjunct is in the relative form (the non-relative form is ŋgarihuláo). Both example () and () are impossible in Kiswahili.

. N  

.................................................................................................................................. Another case of clausal ellipsis is provided by the Shingazidja example in (). ()

Shingazidja Djumwá ŋgu-fikiɾi Á̰iʃa ng-e fuɾahá=ni ʃá=Saidí Juma .-think Aisha - happiness= but=Saidi ka-tsw-ámini .-.-believe ‘Juma thinks Aisha is happy, but Saidi doesn’t believe [EL].’

() is a case of null complement anaphora (henceforth NCA), where the clausal complement of a verb, here A̰ iʃa nge furahá=ni ‘Aisha is happy’, is elided. As in other languages, the elided complement is often an infinitival clause, as in ().42 41 Evidence comes from the shift of the tone of magáɾi ‘cars’ to the verb in the relative form. See Patin () for details on the morphosyntax and prosody of relatives in Shingazidja. 42 A reviewer asked whether we had evidence in favor of CP-ellipsis as opposed to DP-ellipsis, since infinitives are nominals in Bantu; (s)he suggested object marking as a possible test. Indeed, in (b), for instance, utsaβuhá goma ‘playing/to play drums’ can be replaced by u-li-tsáβuha ‘playing it (e.g. the guitar—class )’, demonstrating that utsaβúha is not a DP in (b). Additional evidence in favor of this

 ()

   a. Kiswahili Juma a-na-penda ku-cheza ngoma, Saidi hapendi Juma --like -play .drums Saidi -like- ‘Juma likes dancing (lit. = ‘[to] play drums’), Saidi does not like [EL].’ b. Shingazidja Djumwá=(yé) hw-an(d)zá u-tsaβuha go̩má Said(i)=yé ka-ndza Juma=1 -like -play .drum Saidi=1 .-like ‘Juma likes playing drums, Saidi does not like [EL].’

Structural identity between the (infinitival) clauses is not a requirement for ellipsis to occur. In (b), the verb hudjá ‘to come’ can be deleted, even if the verb in the antecedent clause is in the subjunctive form. () Shingazidja a. Saidí ha-haɾayá hu-djá ʃa=Á̰ lii ha-kubáli (hu-djá) Saidi .-refuse -come but=Ali .-accept (-come) ‘Saidi refused to come but Ali agreed (to come).’ b. wa-n-ámbiya ni-djé ʃa=tsi-háɾaya (hu-djá) .--tell ..-come. but=.-refuse (-come) ‘They told me to come but I refused (to come).’

. S

.................................................................................................................................. Sluicing is possible in both Kiswahili and Shingazidja. In (), the entire constituent question is elided, with the exception of the wh-phrase. () a. Kiswahili Juma a-na-ku-la na m-tu si-ju-i na Juma ---eat with -person .-know- with ‘Juma is eating with someone, I don’t know with whom [EL].’

nani who

b. Shingazidja Djumwá ŋgu-l-w=(y)é̩ na=m-nɗru ŋgam-djuzis-o Juma .-eat-= and=-person .-wonder- ná=nɗo(=βí)43 with=who= ‘Juma is eating with someone, I wonder with whom [EL].’

analysis comes from the shift of the tone: the tone of hwándza ‘he likes’ does not shift to utsaβúha ‘playing/to play’, while this would be the case if the latter word were a DP. 43 The role of this optional focus marker is unclear. According to Ahmed-Chamanga (: ), it is used when additional information is required, or when the answer was not properly heard when the question was first asked.

  



In Bantu G., as shown in (), no agreement marker links the remnant to its antecedent in the first clause. In another Bantu language spoken in Cameroon, Basàá (A.), whose sluicing constructions have been extensively discussed in Bassong (: ch. ), the sluiced operator44 agrees in class with its antecedent (). () Basàá (Bassong : )45 a. Mɛ n-nɔ́ɣ lɛ́ Ewas a-n-sɔ́mb míntómbá, mɛ ḿ-ɓatɓá I -hear that .Ewas .--buy .sheep I -wonder lɛ́ tɔɔ́ mímbɛ̂ mî that whether/if .which ones . ‘I heard that Ewas has bought sheep, I wonder which ones.’ b. Bíkûn a-bí-lɔná líwándá jé jádá makebla yaaní, mɛ .Bikun .--bring .friend .his .one .presents yesterday I ḿ-ɓatɓá lɛ́ tɔɔ́ límbɛ̂ lî -wonder that whether/if .which one . ‘Bikun brought the presents to one of his friends, I wonder which ones [sic].’ In (a), the question word -mbɛ̂ agrees in class  with the antecedent míntómbá ‘sheep’, while it agrees in class  with the antecedent líwándá ‘friend’ in (b). In (), the wh-phrase has an overt correlate in the antecedent clause: ‘person’ (mtu in Kiswahili, ḿnɗru in Shingazidja)—this construction is sometimes referred to as the merger type of sluicing following Chung et al. (). As in other languages such as Basàá (Bassong : ff.), ellipsis is also possible when the wh-phrase has no direct correlate (sprouting). () a. Kiswahili Juma a-me-m-kamata punda na-ji-uliza na ki-tu gani Juma ---catch .donkey --wonder with -thing which ‘Juma caught [the] donkey, I wonder with what [EL].’ b. Shingazidja Djumwá ha-zingaɾ=é m-punɗrá ŋgam-djuzis̩ -w (h)á=(hi)ndri=ní Juma .-catch=1 -donkey .-wonder- with=what= ‘Juma caught the donkey, I wonder with what [EL].’ Recall from section . that a verb in the applicative form must be followed by a complement in Shingazidja. In the sluicing example (), however, where the wh-phrase nɗo=βí ‘who’ has no overt correlate in the antecedent clause, the verb in the applicative form in the first conjunct46 is not followed by a complement.

This is accompanied to its right by a “definitizer, which is a specific morpheme with an identificational/specificational reading and which always occurs on the right of the question operator” (Bassong : ). Bassong also refers to this morpheme as an “evidential marker” (Bassong : ). 45 The transcription and glosses follow the original:  = evidential(ity),  = present,  = past,  = subject marker. 46 Note that () would be ungrammatical if the verb of the first conjunct was not in the applicative form. 44



  

() Shingazidja Alií ŋgw-andzi-ʃi-o ʃá=ntsu-djuwá nɗo=βí Ali .-read-- but=.-know who= ‘Ali is reading to [EL?] but I don’t know [to] who [EL].’ But now consider the sentence in (), where the order of the clauses is reversed. () Shingazidja ntsu-djuwá nɗo=βí ʃ[a]=Alií ŋgw-andzi-ʃi-ó .-know who= but=Ali .-read-- ‘I don’t know to who [EL], but Ali is reading to someone.’

*(m-nɗru) -person

In (), the verb in the applicative form has to be followed by the word ḿnɗru ‘person’, indicating that the absence of an overt correlate in () is only possible thanks to the sluice that follows. Other examples of sluicing are given in ().47 ()

Shingazidja a. ŋgu-djo-hú-dja ʃá ntsu-djuwa sa̩ya ŋgá=βi .---come but .-know hour Q=48 ‘He will come, but I don’t know when [lit. ‘which hour’] [EL].’ b. Alií ŋgu-l-ó ʃá=ntsu-djuwa (hi)ndrí=ni Ali .-eat- but=.-know what= ‘Ali is eating, but I don’t know what [EL].’ c. Alií ha-faɲa ma-djailiya̩nó ʃ[a]=ntsu-djuwá uhuswiya[na] na=(hi)ndrí=ni Ali .-do -talk but=.-know about and=what= ‘Ali gave a lecture, but I don’t know about what [EL].’ d. Alií ha-nika Fátima pitʃa ya-li-fa̩ɲá ʃa=ntsú-djuw[a] Ali .-give Fatima .picture ..--do but=.-know [h]a=(hi)ndrí=ni with=what= ‘Ali gave Fatima a picture he made, but I don’t know with what [EL].’ e. Fatimá ŋgu-djuw-o m-nɗr[u] u-ló-na m-ɓwa tsa̩nú, Fatima .-know- -person ..-.-have -dog five ʃa=ntsú-djuw[a] ma-βaha ma-ŋgá=βi but=.-know -cat -Q= ‘Fatima knows a person who has five dogs, but I don’t know how many cats [EL].’

Contrary to the English example on which (c) is based (van Craenenbroeck and Merchant :  (c)), no inversion of the canonical order of the preposition and the

(b–e) are inspired by similar examples in van Craenenbroeck and Merchant (: , ). Considering the similar structure in (e), it is probable that the underlying structure of this word is /N-nga βí/ (-Q ). The class prefix would be deleted in order to avoid a succession of nasals. 47 48

  



wh-phrase is observed in (c), indicating that there is no swiping, a subtype of sluicing where the order of the wh-remnant and the preposition is reversed (see Merchant ), in Shingazidja. Finally, () and () demonstrate that sluicing may occur in matrix questions. () a. Kiswahili i. Ali a-ta-ku-ja Ali ---come ‘Ali will come.’ ii. Lini? when ‘When?’ b. Shingazidja i. Alií ŋgu-djo-ká hu-djá Ali .--be -come ‘Ali will come.’ ii. nɗí? / nɗí=ni? when when= ‘When?’ () a. Kiswahili (Kenya) i. Djirani w=angu a-na-enda49 .neighbor =mine --go ‘My neighbor goes on holidays.’ ii. Wapi? where ‘Where?’

likizo holidays

b. Shingazidja i. ye=djiɾani w-á=hangú ŋgw-enɗ-o 1=.neighbor -of=mine .-go- ‘My neighbor goes on holidays.’ ii. nɗa=hûː? Q= ‘Where?’

vakã́si holidays

. F

.................................................................................................................................. Fragments exist in both Kiswahili and Shingazidja. Ellipsis is for example possible in (short) answers, as in () and ().

In standard Kiswahili, a-na-kw-enda—with a class  prefix preceding the root ‘go’—would be used instead. 49



  

() a. Kiswahili (Kenya) i. A-me-mw-ona nani? ---see who ‘Who did he see?’ ii. Ali / Ali ‘Ali.’

Amemwona ---see ‘He saw Ali.’

Ali Ali

b. Shingazidja i. ha-o̩no nɗó? .-see who ‘Who did he see?’ ii. Alíi / Ali ‘Ali.’

haonw’ Álii --see Ali ‘He saw Ali.’

() a. Kiswahili (Kenya) i. na-weza ku-la nini? .-can -eat what ‘What can I eat?’ (lit. ‘I can eat what?’) ii. nyama / .meat ‘Meat.’ u-na-weza ku-la nyama --can -eat .meat ‘You can eat (some) meat.’ b. Shingazidja i. ye tsí-ɓaki na=hindrí ʃahulá? Q .-stay with=what .food ‘What is left for me to eat?’ (lit: ‘I stayed with what food’?) ii. mba djandzé / some .crab ‘Some crab.’ βa-báki mba djandzé .-stay some .crab ‘There’s [lit. ‘it stays’] some crab (left).’ In both languages, using the long answer is possible, but our informants stress that the fragment answer is the more common answer. Shingazidja fragments differ from those in languages such as English in the way they deal with connectivity effects.50 Merchant (a: ) notes that a name in a fragment cannot

50

The situation is unclear in Kiswahili. Further research is required.

  



“corefer with a c-commanding name or pronoun” in the question, which is a Principle C effect—cf. (). () English (Merchant a: ). a. Where is hei staying? b. i. * In Johni’s apartment. ii. * Hei is staying in Johni’s apartment. It has been shown by Bassong (: ff.) that the Bantu language Basàá exhibits connectivity effects in fragment answers (). ()

Basàá (Bassong : )51 a. Q: njɛ́ɛ́ Ewasi a-ŋ́-gwês .who .Ewas .--like ‘Who does Ewasi like?’ b. A: proi ai-ŋ́-gwês nyɛ́-mɛ́dɛ́i /*nyɛ́i pro .--like him- /him ‘He likes himselfi /*himi.’ c. A: nyɛ́-mɛ́dɛ́i /* nyɛ́i him- /him ‘Himselfi /*himi.’

In (b–c), “only the anaphor nyɛ́-mɛ́dɛ́i ‘himself ’ is licensed as a felicitous answer to the question under discussion [ . . . ] even though there appears to be no visible antecedent DP/ NP to bind this anaphor in [(c)]” (Bassong : ). The Shingazidja version of (), however, is grammatical—cf. (). () Shingazidja a. (y)é ŋgui-some-ó nɗa=hûː? Q PRES.i-study.- Q= ‘Where does hei study?’ b. (nɗ=ó) harimwá ye=ɲumɓa y-a=Álii. = inside =.house -of=Alii ‘(It is) In(side) Alii’s house.’ Fragment DP answers to questions in which the wh-phrase is preceded by a preposition have an interesting characteristic in Shingazidja. Merchant (a) explains that “[i]n languages like English and the Scandinavian languages, which allow preposition stranding wh-movement in questions, ‘bare’ DP answers to such questions are permissible” (Merchant a: —e.g. (a)), while “[i]n non-preposition-stranding languages such as Greek,

The transcription and glosses follow the original:  = evidential(ity),  = present,  = reflexive marker,  = subject marker. 51



  

German, Yiddish, Czech, Russian, Bulgarian, and Hebrew [ . . . ], such ‘bare’ DP answers are impossible” (Merchant a: —e.g. (b)). () a. English (Merchant a: ) i. Who was Peter talking with? ii. Mary. b. Greek (Merchant a: ) i. Me pjon milise i with whom spoke the ii. Me iii. with

ton *Ton the

Anna? Anna?

Kosta. Kosta. Kostas

In Shingazidja, a question such as (a.i) is generally expressed by leaving the interrogative marker in situ (a). In such cases, “bare” DP answers are permissible. This is also true for answers that are introduced by the preposition na ‘with/and’ (b). () Shingazidja a. Yé Alií ha-ka-(h)u-ɾóngowa nɗa=é na=nɗó Q Ali ---talk =1 and=who ‘Who was Ali talking with?’ (lit. ‘Ali was talking it him with who?’) b. Djummwá Juma ‘Juma’

/

Na=Djummwá and=Juma ‘With Juma’

However, Shingazidja also allows wh-movement with preposition stranding in questions, if the position formerly occupied by ‘who’ is filled with the default question marker yé (a). In such cases, “bare” DP answers are also possible, confirming the distribution that was stated in Merchant (a). This is again the case for answers that are introduced by na ‘with/and’ (b). ()

Shingazidja a. Yé nɗo=βí Alií ha-ka-(h)uɾóngowa nɗa=é na=yé Q who= Ali ---talk =1 and=Q ‘Who (exactly) was Ali talking with?’ (lit. ‘Who Ali was talking it him with?’) b. Djummwá Juma ‘Juma’

/

Na=Djummwá and=Juma ‘With Juma’

There is yet another configuration in which the question marker nɗó ‘who’ is fronted in Shingazidja. In this alternative construction, no preposition emerges at the end of the question, but a so-called “applicative extension” is suffixed to/modifies the verb. See (a) where the stem -ɾongówa ‘talk’ is transformed into the stem -ɾongóza ‘talk to’. In these cases, only the bare answer is permissible.

   ()



Shingazidja a. Ye̩ nɗó uwo52 Alií ha-ka-(h)u-ɾóngoza Q who . Ali ---talk. (= talk to) ‘Who (exactly) was Ali talking with?’ (lit. ‘Who this Ali was talking to?’) b. Djummwá Juma ‘Juma’

/

*Na=Djummwá and=Juma Intended: ‘With Juma’

. I E- VPE ?

.................................................................................................................................. In section ., we briefly discussed situations where the verb is in a compound tense. Consider now example (): () Shingazidja a. tsi-ka ts-énɗe Paɾí, ha-ka hw-énɗe .-be .-go Paris .-be .-go ‘I had gone [to] Paris, he had gone [to] Marseille.’

Maɾsɛ́j Marseille

b. tsi-ka ts-énɗe Paɾí, hw-enɗe Máɾsɛj .-be .-go Paris .-go Marseille #‘I had gone [to] Paris, he had [EL] [to] Marseille.’ ‘I had gone to Paris, he went to Marseille.’ c. ??tsi-ka ts-énɗe Paɾí, ha-ka Máɾsɛj .-be .-go Paris .-be Marseille #‘I had gone [to] Paris, he had [EL] [to] Marseille.’ Intended: ‘I had gone [to] Paris, he had to Marseille.’ In (a), two clauses that include a verb in a compound tense, the remote past, are conjoined. In (b), however, hwenɗé is not preceded by haká in the second conjunct. In such a case, as shown by the translation, a speaker of Shingazidja will not understand the sentences as involving an elided element haká, i.e., a verb in the remote past where the first part is absent, but rather as a sentence that includes in its second conjunct a complete verb in the perfective tense. The deletion of hwenɗé (c) would lead to an uninterpretable sentence that would mean ‘I had gone to Paris, he was Marseille’. In (), the compound tenses involve two compounds that exhibit inflection. One may wonder whether ellipsis is possible when the second part of a verb in a compound tense is in the infinitival form. In other words: is predicate ellipsis, such as English-like VP-ellipsis or English-like pseudogapping, possible in G. languages? As can be observed in (), where the verb is in the imperfective tense, this is not the case. 52

This set of demonstratives can refer to an element that is semi-proximate or to a referent that has already been mentioned in the discourse. Here, it seems to be used as a focus particle. According to our main informant, it is not obligatory, but at the same time its absence would cause the sentence to be judged not fully grammatical.



  

() Shingazidja a. tsi-ká hu-la yémɓe, Alií ha-ká .-be -eat .mango Ali .-be ‘I used to eat mangoes, Ali used to eat oranges.’ b.

hu-la -eat

ma-ɾúnɗa -oranges

??

tsi-ká hu-la yémɓe, Alií ha-ka ma-ɾúnɗa .-be -eat .mango Ali .-be -oranges #‘I used to eat mangoes, Ali used [EL] oranges.’ Intended: ‘I used to eat mangoes, for Ali it was oranges.’

All our attempts to produce sentences that are similar in nature to examples of Englishlike pseudogapping (and/or of English-like VP-ellipsis) were rejected by our speakers. The closest types of structures we were able to obtain were antecedent-contained deletion examples such as () in Kiswahili. ()

Kiswahili Juma a-me-soma kila a-li-ch-o-ki-weza Juma --read each - ---7-can ‘Juma read everything he could [EL].’

Examples that are similar in nature to () were rejected by our Shingazidja speakers. What is possible in Shingazidja, however, is the deletion of elements of lexical constructions such as hufaɲá hazi ‘to work’. In (), for example, the word házi ‘work’ is deleted in the second conjoint.53 ()

Shingazidja ŋgu-faɲ-ó hazi leó ʃa=(u)wó ká-tsu-faɲa .-do- .work today but= 1 -.-do ‘Hei works today but [this one] hei/j does not [EL] tomorrow.’

má̩uɗu tomorrow

It seems that only a prosodic word can be deleted under ellipsis in G. languages. For instance, we failed in all our attempts to obtain sentences where a class prefix is elided. It has been reported that class prefixes may be dropped in some Bantu languages (e.g., prefixes of classes , , , and  in Sesotho,54 according to Machobane ), mostly in casual speech situations. To our knowledge, this possibility is not related to ellipsis. The ellipsis of a morpheme is also rejected in the context of a correction ():55 () Kiswahili a. a-na-soma ki-tabu - -read -book ‘He is reading a book.’ 53

A reviewer has indicated that this pattern is reminiscent of v-stranding VP-ellipsis as it emerges in Persian (see Toosarvandani , ). 54 S.—South Africa and Lesotho. 55 Testing the form in () was suggested by a reviewer.

  



b. *hapana, a-li no - Intended: ‘No, he HAS [EL]!’ The agglutinative nature of the verb forms in Bantu, along with this constraint against the deletion of what is not a prosodic word, may result in the absence of predicate ellipsis in these languages.

. C

.................................................................................................................................. In this chapter, we have provided a detailed discussion of ellipsis in Kiswahili and Shingazidja, two languages that belong to the G. group of Bantu languages. We have demonstrated in sections . through . that several kinds of ellipsis are possible in these languages: clausal ellipses such as gapping, sluicing, and fragments, as well as comparative ellipsis and N-deletion were all judged to be perfectly acceptable by our speakers. Predicate ellipsis (e.g., English-like VP-ellipsis, English-like pseudogapping), however, does not seem to be possible in Kiswahili and Shingazidja, with the exception of V-stranding VP-ellipsis. The two languages differ in the way they deal with some of the elliptical structures that were explored in this chapter. Shingazidja accepts right-node raising, for instance, while several of our Kiswahili speakers reject this possibility. Gapping structures in Shingazidja also differ from those of Kiswahili, the latter optionally marking gapping via the presence of a comitative morpheme that is absent from the non-gapped sentences, and which is absent from Shingazidja. Kiswahili also allows gapping with multiple remnants, contrary to Shingazidja, unless the respective positions of the direct object and the indirect object are inverted. The origin of these differences remains unclear at this point in our research, but we suspect that the differences in the prosody of the two languages (Shingazidja has a complex tone system, while Kiswahili has phrasal stress) may explain some of them. Further research is required in order to explore this issue.

A We would like to thank all our informants for their contributions. We are also grateful to Kathleen M. O’Connor, Gérard Philippson, two anonymous reviewers, and the editors of the volume for helpful remarks on this chapter. Parts of this chapter were presented at Topics in the Typology of Elliptical Constructions in Paris in June , and at Bntu in Berlin in September . We are grateful to audiences at these conferences and to everybody in the TUL project ‘Approches typologiques des constructions elliptiques’ (directed by Anne Abeillé and François Mouret) for helpful questions and suggestions. None of the above necessarily endorses any of the content in this chapter; all errors and omissions are the sole responsibility of the authors.

  ......................................................................................................................

 ......................................................................................................................

 

P has several types of ellipsis familiar from other languages. These fall into three categories. First, there are elliptical constructions that occur exclusively in coordinations: gapping (a), stripping (b), and right-node raising (c). (The missing material is indicated by Δ.)1 () a. Gapping Râdmehr diruz gusht xord Rodmehr yesterday meat eat.. ‘Rodmehr ate meat yesterday, and Giti fish.’

va Giti and Giti

b. Stripping Farnâz qorme sabzi dorost kard=e Farnaz qorme sabzi correct do.=be.. ‘Did Farnaz make qorme sabzi or Maryam?’

Δ mâhi. fish (Farudi : ) yâ Maryam Δ? or Maryam

c. Right-node raising Ali dâstân=e kutâ Δ va Maryam român mi-xun-e. Ali story=ez short and Maryam novel -read.- ‘Ali is reading a short story, and Maryam is reading a novel.’ (Shabani a:) Second, there are elliptical constructions that can occur outside of coordination structures and whose antecedents can freely be located across sentence boundaries: verb phrase ellipsis (a), sluicing (b), and noun phrase ellipsis (c).

1

Any data not attributed to any source comes from two native speakers of Persian. Whether original or taken from elsewhere, all examples come from the variety of Persian spoken in Iran, which is called Farsi in the language itself. Other dialects are spoken in Afghanistan (Dari) and Tajikistan (Tajiki). Examples from other sources have been retranscribed, reglossed, and retranslated for uniformity. The abbreviations used in the interlinear glosses are:  = accusative,  = adjectivizer,  = classifier,  = comparative,  = ezafe,  = imperative,  = imperfective,  = indefinite,  = negative,  = past,  = present,  = participle,  = singular,  = subjunctive,  = superlative.

 () a. Verb phrase ellipsis Sohrâb piran=â=ro out na-zad Sohrab shirt=pl= iron -hit.. ‘Sohrab didn’t iron the shirts, but Rostam did.’ b. Sluicing Râmin ye chiz=i xarid. Hads Ramin one thing= buy.. guess ‘Ramin bought something. Guess what.’

vali but



Rostam Δ zad. Rostam hit.. (Toosarvandani : )

be-zan chi Δ. -hit what (Toosarvandani : )

c. Noun phrase ellipsis Q: Xodkâr=e sabz dâr-i? = green have.- ‘Do you have a green pen?’ A: Δ Sabz na-dâr-am. Δ Âbi mi-xâ-y? green -have.- blue -want.- ‘I don’t have a green one. Do you want a blue one?’ (Ghaniabadi : ) Finally, there are fragment answers, which are found in a different utterance from their antecedent. () Fragment answer Q: Ali chi xarid? Ali what buy.. ‘What did Ali buy?’ A: Δ Shalvâr Δ. pants ‘Pants.’

(Shabani b: )

In what follows, each of these constructions in Persian will be examined in turn. Several of them—namely, gapping and stripping, fragment answers, verb phrase ellipsis, and sluicing—turn out to have properties quite different from their counterparts in betterstudied languages. There are a few other elliptical constructions, which either do not exist in Persian or cannot easily be distinguished from other kinds of null anaphora. Comparative deletion and subdeletion fall into this first category. While the language has phrasal comparatives (a), any type of clausal comparative is strictly ungrammatical (b–c). The requisite structure is simply not available: the preposition az ‘from’ does not take clausal complements; cf. Hindi (Bhatt and Takahashi ). () a. Sohrâb az Râmin boland-tar=e. Sohrab from Ramin tall-=be.. ‘Sohrab is taller than Ramin.’ b. * Sohrâb az Râmin Δ=e Sohrab from Ramin =be.. Intended: ‘Sohrab is taller than Ramin is.’

boland-tar=e. tall-=be..



 c. *Sohrâb az mâshin derâz Δ=e boland-tar=e. Sohrab from car long =be.. tall-=be.. Intended: ‘Sohrab is taller than the car is long.’

On the other hand, conjunction reduction and null complement anaphora look like run-of-the-mill null arguments. The subject in a coordination () or a clausal complement () can go missing. () Sohrâb mi-r-e farânse yâ Δ Sohrab -go.- France or ‘Will Sohrab go to France or stay in Iran?’

irân mi-mun-e? Iran -stay.-

() Q: Sohrâb=o Râmin mi-dun-an ke emtehân emruz=e? Sohrab=and Ramin -know.- that exam today=be.. ‘Do Sohrab and Ramin know that the exam is today?’ A: Râmin mi-dun-e Ramin -know.- ‘Ramin knows.’

Δ.

But in general, any argument can be null in Persian, including subjects (a), direct objects (b), and indirect objects (a). () a. Dotâ xanum be man âb dâd-an. Δ Δ Nun=ham dâd-an. Two woman to me water give.- bread=also give.- ‘Two women gave me water. They gave me bread, too.’ b. Râmin român=e engelisi xarid. Parhâm Δ barâ=sh Ramin novel= English buy.. Parham for= kard. do.. ‘Ramin bought an English novel. Parham translated it for him.’

tarjome translation

It may not be impossible to isolate conjunction reduction and null complement anaphora in Persian as distinct elliptical constructions, but this challenging work remains to be done.

. G  

.................................................................................................................................. In English, gapping removes the finite verb and possibly more in the second and subsequent coordinates of a coordination, leaving behind two remnants. Persian has an elliptical construction like this: just the main verb goes missing in (a), and it goes missing with additional material in (b).

 () a. Ânâhitâ mâhi xord va Annahita fish eat.. and ‘Annahita ate fish, and Rod meat.’

Râd Rod



gusht Δ. meat

b. Râtâ bastani=ro be Pari dâd va Mehrân Δ be Farmehr Δ. Rata ice.cream= to Pari give.. and Mehran to Farmehr ‘Rata gave ice cream to Pari, and Mehran to Farmehr.’ (Farudi : –) As discussed in section .., it is possible to strand a single remnant, a construction often called stripping. But leaving behind three remnants is degraded (), likely for the same reasons it is in English. () ??Ârtur be Giti pul dâd va Pari be Ânâhitâ kârt Δ. Arthur to Giti money give.. and Pari to Annahita card ‘Arthur gave money to Giti, and Pari (gave) a card to Annahita.’ (Farudi : ) In English, gapping can remove just the finite auxiliary, leaving the main verb behind in a non-finite form, e.g., Kim can play bingo, and Sandy Δ stay at home. This is not possible in Persian. () a. *Sâra be Giti pul dâde bud va Mahin az Mâzyâr ketâb Sara to Giti money give. be.. and Mahin from Maziar book gerefte Δ. take. Intended: ‘Sara had given money to Giti, and Mahin taken a book from Maziar.’ (Farudi : ) b. *Sohrâb be bâshgâh xâh-ad raf va Râmin xune Δ mund. Sohrab to gym want.- go and Ramin home stay Intended: ‘Sohrab will go to the gym, and Ramin stay home.’ Neither the perfect aspect auxiliary budan ‘be’ (a) nor the future tense auxiliary xâstan ‘want’ (b) can go missing in gapping all by itself.

.. Three properties of gapping Moving beyond this surface characterization, Johnson () identifies three properties of gapping that distinguish it from other elliptical operations in English, such as verb phrase ellipsis. First, gapping is restricted to coordinate structures, a property that gapping in Persian shares. It appears with both simple coordinators, such as va ‘and’ (a), and complex ones, such as na . . . na (b), cf. either . . . or (Schwarz ). But gapping is not permitted in a subordinate clause, such as a temporal adjunct (a)2 or the antecedent of a conditional (b). The clausal complements of prepositions, such as az ‘from’, must be introduced by the determiner in ‘this’. 2

 ()

 a. Man mâhi xord-am I fish eat.- ‘I ate fish, and Giti meat.’

va and

Giti Giti

gusht Δ. meat (Farudi : )

b. Na Farnâz qorme sabzi dorost kard=e na Maryam qeyme.  Farnaz qorme sabzi correct do.=be..  Maryam qeyme ‘Neither Farnaz made qorme sabzi, nor Maryam qeyme.’ ()

a. *Ânâhitâ mâhi xord [bad az in Annahita fish eat.. after from this Intended: ‘Annahita ate fish after Giti ate meat.’

ke that

Giti Giti

gusht Δ]. meat

b. *Dâryush be Giti pul mi-d-e [agar Râtâ be Ânâhitâ Δ]. Daryush to Giti money -give.- if Rata to Annahita Intended: ‘Daryush will give money to Giti if Rata will to Annahita.’ (Farudi : ) Gapping in Persian does not, however, share the other two properties that Johnson () identifies. In English, the material that goes missing in gapping cannot be contained within an embedded clause, e.g., *Some had eaten mussels, and she claims that others shrimp (Johnson : ). In Persian, Farudi (: –) observes that this is possible. ()

Mahsâ in ketâb=ro dust dâr-e vali hichkas bâvar ne-mikon-e Mahsa this book= friend have.- but no.one belief -do.- [mâmân=esh un ketâb=ro Δ]. mother= that book= ‘Mahsa likes this book, but nobody believes that her mother likes that book.’

The antecedent of the missing material also cannot be contained in an embedded clause in English, e.g., *She’s said Peter has eaten his peas, and Sally her green beans, so now we can have dessert (Johnson : ). But again, Farudi demonstrates (: –) that this is possible in Persian. (The antecedent material is bolded in the following example.) ()

Fekr mi-kon-am [ke Ânâhitâ châyi=ro xord] vali Giti thought -do.- that Annahita tea= eat.. but Giti qahva=ro Δ. coffee= ‘I think that Annahita drank tea, but Giti drank coffee.’ ‘I think that Annahita drank tea but that Giti drank coffee.’ (Farudi : )

Crucially, in () the antecedent is embedded in the first coordinate only under the first interpretation. The second interpretation, where the entire coordination structure is embedded, is irrelevant. This confound is avoided by embedding the gap and the antecedent separately in each coordinate.

 ()



Ajib nist [ke Râdmehr mâhi=ro xord=e] vali strange .be.. that Rodmehr fish= eat.=be.. but ajib=e [ke Ânâhitâ gusht=ro Δ]. strange=be.. that Annahita meat= ‘It’s not strange that Rodmehr ate fish, but it is strange that Annahita ate meat.’3 (Farudi : )

This contrasts strikingly with the parallel gapping sentence in English—*It’s not strange that Rodmehr ate fish, but it is strange that Annahita Δ meat—which is decidedly ungrammatical.

.. Towards an analysis The derivation of gapping in Persian can use some of the same ingredients that it does in English. Since gapping does not have to remove a constituent, e.g. (b), it is usually derived through movement of the remnants, followed by either deletion (Jayaseelan : –; Sag a: –, a.o.) or across-the-board movement (Johnson a). There are two arguments that the remnants in gapping also undergo movement in Persian. First, as Farudi observes (: –), a remnant cannot originate inside an island. (The islands in these examples are bracketed.) () a. Complex NP Constraint *Giti [mard=hâ=yi ke futbâl bâzi mi-kon-an] dust dâr-e Giti man== that soccer game -do.- friend have.- va Sârâ Δ tenis Δ. and Sara tennis Intended: ‘Giti likes men who play soccer, and Sara likes men who play tennis.’ b. Coordinate Structure Constraint *Râmin [kabâb xord va ruznâme xund] va Sohrâb Δ Ramin kebab eat.. and newspaper read.. and Sohrab român Δ. novel Intended: ‘Ramin ate kebab and read the newspaper, and Sohrab ate kebab and read a novel.’ c. Adjunct Constraint *Moallem [chon Râmin sar=e kelâs dir umad] az=ash teacher since Ramin head= class late come.. from= nomre kam kard va man Δ Sohrâb Δ. grade little do.. and I Sohrab Intended: ‘The teacher took off points because Ramin was late coming to class, and I took off points because Sohrab was late coming to class.’

Farudi assigns ‘?’ as the judgment for (), but both speakers I consulted with judged it fully grammatical. 3





Second, remnants exhibit the same case marking they would have if they had occupied a case position at some point in the derivation. ()

Ânâhitâ mâhi=ro xord va Râd Annahita fish= eat.. and Rod ‘Annahita ate the fish, and Rad the meat.’

gusht(=ro) Δ. meat= (Farudi : )

While the accusative case marker is only optionally present in (), the fact that it is possible at all suggests that the remnant originates as the complement of a verb that has gone missing.

.. No low coordination in Persian If the remnants in gapping undergo movement in Persian, where do they move to? The answer to this question depends in part on how large the coordination structure in gapping is. Building on earlier work by Siegel (), Johnson (a, ) proposes for English that gapping always involves low coordination: a single T head is shared by more than one vP coordinate. There is thus never a finite verbal element inside the second and subsequent coordinates. Additional material is removed through across-the-board movement, or possibly deletion (Coppock ; Lin ). Johnson (: –) argues that the syntax of low coordination accounts for at least two of the properties of gapping (see also Toosarvandani ). The missing material cannot be embedded because a single T head cannot be shared both with the first coordinate and a vP that is embedded inside a coordinate. Similarly, the antecedent cannot be embedded because a single T head cannot be shared both with the second vP coordinate and a vP embedded inside the first coordinate. But gapping in Persian does not exhibit these two properties. This is plausibly, as Farudi argues (: –), because it does not use low coordination. In English, negation can take wide scope over conjunction in gapping, e.g., Ward can’t eat caviar, and his guest Δ beans (Siegel : ). In Persian, however, this wide-scope interpretation is never possible. ()

Bizhan xâvyâr ne-mi-xor-e va Maryam noxod Δ. Bijan caviar --eat.- and Maryam chickpea ‘Bijan doesn’t eat caviar, and Maryam doesn’t eat chickpeas.’ ¬p∧¬q Not possible: ‘It’s not the case Bijan eats caviar and Maryam eats chickpeas.’ ¬(p∧q) (Farudi : )

Since negation can only take narrow scope within each conjunct, gapping in Persian must coordinate full clauses. In this clausal coordination structure, Farudi proposes that the remnants move to leftperipheral positions dedicated to topic and focus (see section .. for discussion of these positions). As a consequence, across-the-board movement cannot be responsible for removing material in Persian, as Johnson (a, ) proposes for English. Instead, TP in the second coordinate must be deleted.

 ()

TopP

= (8a)

TopP

Ânâhitâ

&P

Top

TopP

&

Top′

DP1



TP t1 mâhi xord

va

Top′

DP2 Râd

Top

FocP DP3 gusht

Foc′ Foc

TP t2 t3 xord

This derivation for gapping in Persian accounts for why it allows both the gap () and its antecedent () to be embedded: T is not shared across coordinates. In addition, it accounts for why finite auxiliaries cannot go missing all by themselves (a–b). If they are located in T and cannot move, they will always be contained in the elided constituent. Since gapping in Persian does not involve low coordination, one property—the restriction to coordinations—remains unexplained. Farudi conjectures (: –) that it derives from the inventory of heads available in Persian to license ellipsis. Coordinators, but not subordinators, possess a version of Merchant’s () E-feature that can trigger deletion of TP. If this is correct, then it must be possible for a head to license ellipsis of a phrase that is not its complement, pace Lobeck () and others. This might happen through agreement (Aelbrecht ), or as Farudi suggests, through movement of TP into the specifier of the coordinator that licenses ellipsis.

.. When there is just one remnant It is also possible for just one remnant to be left behind in Persian. Sometimes this elliptical construction is called stripping, though it exhibits many of the same properties as gapping. To start, stripping is available in coordination structures with simple (a) and complex (b) coordinators.4

4 There may be some interspeaker variation in the coordinators that are permitted with stripping. In certain contexts, one reviewer did not allow stripping with va ‘and’, though other speakers I consulted with found it perfectly fine.





() a. Farnâz qorme sabzi dorost kard=e yâ Maryam Δ? Farnaz qorme sabzi correct do.=be.. or Maryam ‘Did Farnaz make qorme sabzi or Maryam?’ b. Ham Farnâz qorme sabzi dorost kard=e ham Maryam Δ. also Farnaz qorme sabzi correct do.=be.. also Maryam ‘Both Farnaz made qorme sabzi and Maryam did.’ Stripping is ungrammatical in temporal adjuncts (a) and in the antecedents of conditionals (b). ()

a. *Ânâhitâ mâhi xord [baad az in ke Giti Δ]. Annahita fish eat.. after from this that Giti ‘Annahita ate fish after Giti ate fish.’ b. *Dâryush be Giti pul mi-d-e [age Râtâ Δ]. Daryush to Giti money -give.- if Rata Intended: ‘Daryush will give money to Giti if Rata will give money to her.’

In addition, the gap can be embedded (), as can its antecedent in the first coordinate (). Both the gap and the antecedent also can be embedded separately in each coordinate (). () Mâmân savâr=e charxfalak shod vali hichkas mom aboard= roller.coaster become.. but no.one ne-mi-kon-e ke bâbâ=ham. --do.- that dad=also ‘Mom got on a roller coaster, but nobody believes that Dad did, too.’

fekr thought

() Shenid-am [ke pedar mâdar=et irân raft-an] vali xâhar=et na Δ. hear.- that father mother= Iran go.- but sister=  ‘I heard that your parents went to Iran, but your sister didn’t.’ ‘I heard that your parents went to Iran but that your sister didn’t.’ () Sam fekr mi-kon-e [ke espânyâ World Cup=ro Sam thought -do.- that Spain World Cup= va John fekr mi-kon-e [ke porteqâl Δ] and John thought -do.- that Portugal ‘Sam thinks that Spain will win the World Cup, and John thinks win the World Cup.’

be-bar-e] -win-

that Portugal will (Farudi : )

Farudi’s account of gapping can be extended straightforwardly to stripping. The sole remnant moves to a left-peripheral position, while the rest of the clause is elided. There is some evidence for this movement: the remnant can bear the accusative case marker when it is the direct object.





() Pardis piran=â=ro out zad=e vali na Δ shalvâr=â=ro Δ. Pardis shirt== iron hit.=be.. but  pant== ‘Pardis ironed the shirts, but not the pants.’ Stripping does not, however, exhibit the same sensitivity to islands that gapping does. This is in fact expected if stripping involves deletion of TP in Persian. Sluicing, another elliptical operation that removes full clauses, is similarly island-insensitive (Ross b: –). () a. Complex NP Constraint Giti [mard=â=yi ke futbâl bâzi mi-kon-an] dust dâr-e Giti man== that soccer game -do.- friend have.- yâ Δ tenis Δ or tennis ‘Does Giti like men who play soccer or tennis?’ b. Coordinate Structure Constraint Moqe=ye nâhâr harruz Râmin [kabâb mi-xor-e va ruznâme time= lunch every.day Ramin kebab -eat.- and newspaper mi-xun-e] yâ Δ român Δ. -read.- or novel ‘Every day for lunch, Ramin eats kebab and reads the newspaper, or he eats kebab and reads a novel.’ c. Adjunct Constraint Moallem [chon Râmin sar=e kelâs dir umad] asabâni teacher since Ramin head= class late come.. angry shod yâ Δ Sohrâb Δ? become.. or Sohrab ‘Did the teacher get angry because Ramin came to class late, or did she get angry because Sohrab came to class late?’ It is the island sensitivity of gapping in Persian that is more surprising (a–c), if it has the same derivational source. It may be tempting to attribute this to an independent constraint on remnants straddling a finite clause boundary, e.g., *Charles decided that  boys are coming along, and Harrie  girls (Neijt : ). But remnants in gapping also cannot originate inside a coordinate structure (b), where there is no clause boundary to separate them. It remains an open question why leaving behind two remnants, as opposed to just one remnant, is sensitive to islands.

. F 

.................................................................................................................................. Fragment answers in Persian involve ellipsis, as in other languages (Merchant a). Shabani (b: –) shows that they occur in an argument position at some level of representation. A direct object fragment, for instance, bears accusative case.





() Q: Diruz chi=ro xarid-i? yesterday what= buy.- ‘What did you buy yesterday?’ A: Mashin*(=o) Δ. car= ‘The car.’

(Shabani : )

In addition, fragment answers interact with the binding principles—Condition A (a), Condition B (b), and Condition C (c)—as if they have occupied an argument position. () a. Q: Ali ki=ro dust dâr-e? Ali who= friend have.- ‘Who does Ali like?’ A: Xod=esh=o Δ. self== ‘Himself.’ b. Q: Ali sa’y mi-kon-ad ki=ro gij kon-ad? Ali trying -do.- who= confused do.- ‘Who is Ali trying to confuse?’ A: *U=ro Δ. him= Intended: ‘Him.’ c. Q: U kojâ zendegi mi-kon-ad? he where life -do.- ‘Where does he live?’ A: *Dar xâne=ye Ali Δ. in house= Ali Intended: ‘In Ali’s house.’

(Shabani b: )

Following Merchant (a), Shabani proposes (b: –) that fragments in Persian move to a position in the left periphery before TP is elided, though this is not the same position as in English. There is a striking difference between fragment answers in Persian and other languages. While fragments are sensitive to islands in English (), they are not in Persian (a–b). () Q: Does Abbey speak [the same Balkan language that Ben speaks]? A: *No, Charlie Δ. (Merchant a: –) () a. Complex NP Constraint Q: [Har âdam=i ke che kas=i=sh kotak be-zan-ad] every person= that what person== beating -hit- divâne ast? crazy be.. ‘Whoi is every person who beats ti crazy?’

 A: Zan=esh=o Δ. wife== ‘Their wife.’



(Shabani b: )

b. Adjunct Constraint Q: Har âdam=i xeyli azyat mi-sh-e [vaqt=i every person= very annoyed -become.- time= kojâ=sh dard mi-gir-e]? where= pain -get.- ‘Where does every person become annoyed when it hurts?’ A: Qalb=esh Δ. heart= ‘Their heart.’ (Shabani b: ) For English, Merchant (a) argues that the remnant raises to a focus projection located above CP. When it originates inside an island, the traces it leaves behind are marked as uninterpretable at PF: they crash the derivation unless they are deleted (see also Fox and Lasnik  and Merchant b). The fragment answer in () is ungrammatical, then, because the trace that the remnant leaves as it moves cyclically through specCP is not deleted. () *

FocP Foc′

DP1 Charlie

= (29)

Foc

CP C′

t1* C

TP T′

DP Abbey

T

vP speaks

the same Balkan language that t1* speaks

For Persian, Shabani (b) proposes that the remnant only raises as high as a focus position that is located immediately above TP (see section .. for the properties of this





position). This derives the island insensitivity of fragment answers: when the TP is deleted, all traces of the remnant are also deleted. ()

FocP

= (30a)

DP1 zan=esh=o

Foc′ Foc

TP

DP

T′ T

har âdam=ike t1 * kotak bezanad

vP divâne ast

This derivation for fragment answers parallels the derivation of stripping in Persian, which accounts for their shared insensitivity to island constraints. But, while stripping is restricted to coordination structures, fragment answers appear in root clauses. There must be a different inflectional head that licenses ellipsis of TP in this syntactic context.

. R- 

.................................................................................................................................. There are three main theories of how material goes missing in right-node raising: movement (Ross  and many others), ellipsis (Wilder  and many others), and multidominance (McCawley  and many others). Recently, Barros and Vicente () propose that right-node raising may actually be a heterogenous construction that uses more than one of these mechanisms. They provide evidence that at least ellipsis and multidominance are involved in English. Whether or not movement also participates depends on whether the pivot can be clearly shown to be located outside the coordination structure (Postal : –; Sabbagh ). There is evidence that both ellipsis and multidominance are involved in deriving rightnode raising in Persian (Shabani a, ). On the one hand, it allows for vehicle change. In (), if the pivot were syntactically present in the first coordinate, the sentence would violate Condition C. ()

Un ne-mi-tars-e Δ vali man mi-tars-am he --be.scared.- but I -be.scared.- Trump entexâbât=ro be-bar-e. Trump elections= -win- ‘He isn’t afraid, but I am afraid, that Trump will win the election.’

ke that





This is expected if right-node raising can involve ellipsis, which allows for the substitution of coreferent noun phrases to avoid violation of the binding principles (Fiengo and May : ). On the other hand, right-node raising in Persian can also involve multidominance. As Shabani (: –) observes, it allows for cumulative agreement. In (), the verb exhibits plural agreement with the subjects of the two coordinates, as if it were present in both coordinates simultaneously. () Tim=e espânyâ bâ piruzi bar âlmân Δ, va tim=e holand bâ team= Spain with victory on Germany and team= Holland with piruzi bar urugvây be finâl=e jâm=e jahân=i=ye  râh victory on Uruguay to final= cup= world== path yâft-and/*yâft. find.-/find.. ‘The Spanish team, through victory over Germany, and the Dutch team, through victory over Uruguay, made it to the finals of the  World Cup.’ (Shabani : ) In addition, a relational adjective in the pivot can have an internal reading. In (), motefâvet ‘different’ conveys that Sara and Maryam benefit from distinct approaches. ()

Sârâ dar zamine=ye âmuzesh Δ, va Maryam dar zamine=ye pazhohesh Sara in field= education and Maryam in field= research az râhkâr=hâ=ye motefâvet=i bahre mi-gir-and. from approach== different= profit -get.- ‘Sara in education and Maryam in research will benefit from different approaches.’ (Shabani : )

If right-node raising used only ellipsis, the relational adjective would be present once inside each coordinate, and neither occurrence would have a plural subject, which is needed to license the internal reading. From Shabani’s perspective, it is surprising that inflectional mismatches are never allowed between the pivot and the gap in the first coordinate in Persian. In (), the verb mixune ‘read’ is in the imperfective aspect, satisfying the selectional requirements of the progressive auxiliary dâshtan ‘have’, but not the verb xâstan ‘want’, which selects for an embedded verb in the subjunctive mood; cf. (). Person–number agreement also has to match between the two coordinates () (Farudi : –). () *Maryam tu=ye dâneshgâh mi-xâ-d Δ va Râmin alân dâr-e Maryam tu= university -want.- and Ramin now have.- fizik mi-xun-e. physics -read.- ‘Maryam wants to study physics at university, but Ramin is doing so now.’





() *Man mâhi Δ va Giti gusht xord. I fish and Giti meat eat.. Intended: ‘I ate fish, and Giti ate meat.’

(Farudi : –)

If both ellipsis and multidominance are possible sources for right-node raising in Persian, then such mismatches should be possible. Ellipsis does not usually require identity in inflectional form between the material that goes missing and its antecedent. There has not been a lot of cross-linguistic investigation of right-node raising. But there is some evidence that different languages may use different strategies for deriving right-node raising. Irish has been argued not to use movement to derive right-node raising (McCloskey ), while Tagalog does (Sabbagh ). Sabbagh () observes that this cross-linguistic variation is not surprising if multiple strategies can coexist alongside one another in a single language. This may be the case in Persian, just as it is in English.

. V  

.................................................................................................................................. Persian does not have verb phrase ellipsis of the same kind found in English. It is not possible to elide the entire verb phrase, leaving behind an auxiliary (). But part of a complex predicate (or light verb construction) can go missing. In (), the direct object piran=â=ro ‘the shirts’ and the nominal non-verbal element otu ‘iron’ are elided, leaving the light verb zadan ‘hit’ behind.5 () *Rostam be dandun-pezeshk rafte bud va Sohrâb=ham Δ bud. Rostam to tooth-doctor go. be.. and Sohrab=also be.. Intended: ‘Rostam had gone to the dentist, and Sohrab had, too.’ () Sohrâb piran=â=ro out na-zad vali Rostam Δ zad. Sohrab shirt== iron -hit.. but Rostam hit.. ‘Sohrab didn’t iron the shirts, but Rostam did.’ (Toosarvandani : ) According to one theory of complex predicates (Folli et al. ), the light verb is the overt realization of v, selecting for the external argument (see also Megerdoomian ; Pantcheva ). The non-verbal element, which selects for any internal arguments, heads its complement. It is this phrase, Toosarvandani () proposes, that is elided.

5

Complex predicates are formed productively in Persian. The class of simple verbs, which numbers some  members, is mostly closed (Mohammad and Karimi : ).

 ()

Sohrâb piran=â=ro otu nazad vali

TP

 = (39)

T′

DP1 Rostam

T

vP

t1

v′ v

NP DP piran=â=ro

N

zad

otu

In English, verb phrase ellipsis gives rise to pseudogapping when it leaves behind a remnant. This is not possible with ellipsis in Persian complex predicates (see also Farudi : ). ()

a. Rostam mâshin=esh=o be man neshun dâd va Sohrâb=ham Δ Rostam car== to me showing give.. and Sohrab=also dâd. give.. ‘Rostam showed his car to me, and Sohrab did, too.’ b. *Rostam mâshin=esh=o be man neshun dâd va Sohrâb Rostam car== to me showing give.. and Sohrab motor=esh=o Δ dâd. motorcycle== give.. Intended: ‘Rostam showed his car to me, and Sohrab did his motorcycle.’ c. *Rostam mâshin=esh=o be man neshun dâd va Sohrâb be Rostam car== to me showing give.. and Sohrab to dust=emun Δ dâd. friend=  give.. Intended: ‘Rostam showed his car to me, and Sohrab did to our friend.’

In (a), the non-verbal element neshun ‘showing’ is elided along with the direct and indirect objects. Neither of the internal arguments can be left behind (b–c).

.. Not just a null argument Since DPs and PPs go missing freely in Persian (a–b), the missing non-verbal element and internal argument in () might also just be null arguments. There are two arguments that





seem initially to preclude this conclusion. First, the missing object in (a) can have a strict reading (it is Ramin’s car that Sohrab showed me) or a sloppy reading (it is his own car). This ambiguity is a well-known property of elliptical operations, such as verb phrase ellipsis (Ross : ). Second, the missing non-verbal element in () receives an indefinite interpretation: it describes a novel ironing event of which Rostam is the agent. In English, an elided verb phrase has the same indefinite interpretation. But as Shafiei (: ) observes, null arguments in Persian also exhibit the strict–sloppy ambiguity, a property it shares with Chinese, Japanese, and Korean (Otani and Whitman ). In (), the null object can refer to either Rostam’s or Ramin’s car. () Rostam mâshin=esh=ro be Sohrâb neshun dâd, vali Râmin Δ Rostam car== to Sohrab showing give.. but Ramin neshun na-dâd. showing -give.. ‘Rostam showed his car to Sohrab, but Ramin didn’t show his/ car to him.’ (Shafiei : ) Shafiei also observes that null arguments can have a quantificational interpretation in Persian, like their counterparts in Japanese (Takahashi b). In () the missing object is interpreted as an indefinite, since the sentence is compatible with Sohrab showing the speaker two or fewer cars. () Râmin be man se-tâ mâshin neshun dâd ammâ Sohrâb be man Δ Ramin to me three- car showing give.. but Sohrab to me neshun na-dâd. showing -give.. ‘Ramin showed me three cars, but Sohrab didn’t show me three cars.’ Since these properties are shared by verb phrase ellipsis and null arguments, they are not useful for settling whether the ellipsis in complex predicates can be reduced to one or more null arguments. But there are some more decisive arguments (Toosarvandani : –). In Persian, null arguments exhibit properties of deep anaphora (Hankamer and Sag ). They can have an antecedent that is present solely in the non-linguistic context (a). It is also not possible to extract out of them (b). () a. Context: A child picks up a broom to sweep the carpet. Her mother says: Movâzeb bâsh xub (farsh=o) jâru be-zan-i! careful be.. good carpet= broom -hit- ‘Be careful to sweep the carpet well!’ b. *Rostam qasam xord ke piran=o otu zad=e, va Rostam oath eat.. that shirt= iron hit. =be.. and shalvâr=o=ham qasam xord Δ. pants==also oath eat.. Intended: ‘Rostam swore he ironed the shirt, and the pants, too.’





By contrast, the ellipsis in complex predicates exhibits the properties of a surface anaphor. Like verb phrase ellipsis (Hankamer and Sag : ), it cannot have a non-linguistic antecedent (a). It is also possible to extract out of the ellipsis site (b), again just like verb phrase ellipsis (Schuyler ). () a. Context: A child picks up a broom to sweep the carpet. Her mother says: Motmaen bâsh xub #(farsh-o jâru) be-zan-i! Sure be.. well carpet= broom -hit- ‘Be sure to sweep the carpet well!’ (Toosarvandani : ) b. Rostam piran=o otu na-zad=e vali shalvâr=o Rostam shirt= iron -hit.=be.. but pants= mi-dun-am ke Δ zad=e. -know.- that hit.=be.. ‘Rostam didn’t iron the shirt, but I know that he ironed the pants.’ (Toosarvandani : ) When the non-verbal element and internal arguments go missing in a complex predicate, they are not simply null arguments. They are removed by an operation like verb phrase ellipsis.

.. Evidence for low ellipsis On the surface, this ellipsis in complex predicates resembles verb phrase ellipsis in languages with verb raising. In Hebrew, Irish, and Swahili, V raises to T, so that when vP is deleted, the main verb is stranded (Doron ; Goldberg ; McCloskey a; Ngonyani ). This raises the possibility that the light verb raises to T and the entire vP is elided in complex predicates. It is difficult to demonstrate where Persian has verb raising since the language is verb-final: head movement of the verb to a (right-headed) T would be string-vacuous. The interpretation of certain adverbs, however, shows that just the complement of v goes missing. The adverb dobâre ‘again’ is semantically ambiguous when it modifies a transitive causative verb, like its counterpart in English. Under the repetitive reading, it presupposes a previous occurrence of the entire causative event, while under the restitutive reading, it presupposes only a previous occurrence of the result state. Under one analysis, these readings correspond to different adjunction sites for the adverb in the extended verbal projection (Rapp and von Stechow ; von Stechow ). As Johnson (b: –) shows for verb phrase ellipsis in English, this ambiguity can be used to probe the size of a constituent that has gone missing. In the complex predicate pâk kardan ‘clean’, the adjective pâk ‘clean’ encodes the result state of being clean. If just the constituent headed by this adjective is elided, then dobâre ‘again’ should be able to survive ellipsis and have a restitutive reading. () Dishab âshpazxune pâk bud. Leylâ umad kasif=esh kard. last.night kitchen clean be.. Leila come.. dirty= do..





Kes=i na-raft pâk=esh bo-kon-e. Emshab mi-xâ-m person= -go.. clean= -do- tonight -want.- dobâre Δ bo-kon-am. again -do- ‘Last night, the kitchen was clean. Leila came and dirtied it. Nobody went to clean it. Tonight, I will clean it again.’ (Toosarvandani : ) Indeed, the sentence in () has the restitutive reading: the preceding context only entails that the kitchen was clean earlier, not that the speaker cleaned it before. In Persian complex predicates, ellipsis removes the complement of v, a constituent smaller than the vP that is removed by verb phrase ellipsis in English (Johnson b; Merchant d). At the same time, both operations appear to be licensed locally by an inflection-bearing verbal element. In English, this is an auxiliary or modal (Lobeck : ; Merchant : ), while in Persian it can be a light verb, which bears tense, aspect, and mood morphology, in addition to agreement.

.. Stranding a simple verb While Persian may not have verb raising all the way to T, V presumably raises to v outside of complex predicates. If the complement of v can be elided, it might be possible to strand a simple verb, as in Hebrew, Irish, and Swahili. But in Persian, it would be stranded in v. ()

TP T′

DP1

T

vP v′

t1

v

VP t2

v

V2

Sailor (: –) proposes that the answer in () has precisely this derivation (see also Rasekhi ). The simple verb raises to v before VP is elided. Just the internal argument goes missing. () Q: Naysan ketâb=o xund? Naysan book= read.. ‘Did Naysan read the book?’

 A: Na, Δ na-xund. No -read.. ‘No, he didn’t read it.’



(Sailor : )

Of course, since Persian has null arguments, the answer in () has another derivation in which the direct object is simply not pronounced. Sailor contends that there are some contexts where only the derivation with ellipsis is available. When there is a manner adverbial like bâ deqqat ‘carefully’ in the antecedent clause, it seems to be interpreted inside the site of ellipsis as well. () Naysan ketâb=ro bâ deqqat xund. Nasim=ham Δ xund. Naysan book= with attention read.. Nasim=also read.. ‘Naysan read the book carefully. Nasim also read it carefully.’ (Sailor : ) But this sentence would be true in a situation where Nasim reads the book carefully, even if the adverbial was not contained in the elided constituent. It would have the weaker entailment that Nasim reads the book in any manner. To eliminate this confound, the elided constituent must occur in a downward-entailing environment. () Context: Maysam read the book carefully; Nasim read it, too, but very quickly. Maysam ketâb=o bâ deqqat xund ammâ Nasim Δ Maysam book= with attention read.. but Nasim na-xund. -read.. ‘Maysam read the book carefully, but Nasim didn’t read it carefully.’ The sentence in () is true in a situation where Nasim reads the book but does not do so carefully. Assuming that adverbs cannot simply be null, this interpretation must arise through ellipsis of a constituent that contains the adverb. In Persian, not only can verb phrase ellipsis remove the phrase headed by the non-verbal element in a complex predicate, it can also elide a VP, stranding a simple verb.

. S

.................................................................................................................................. Persian has a sluicing construction that superficially resembles its counterpart in other languages. In () everything in a constituent question is elided except for the wh-phrase. ()

Râmin ye chiz=i xarid. Ramin one thing= buy.. ‘Ramin bought something. Guess what.’

Hads guess

be-zan chi Δ. -hit. what (Toosarvandani : )





In English, sluicing is a product of obligatory wh-movement, which strands the remnant outside the constituent that is deleted (Merchant ). But Persian is a wh-in-situ language: in a fully pronounced constituent question, the wh-phrase does not have to move. () Râmin ye chiz=i xarid. Hads be-zan Râmin chi xarid. Ramin one thing= buy.. guess -hit. Ramin what buy.. ‘Ramin bought something. Guess what Ramin bought.’ (Toosarvandani : ) But wh-phrases are not completely immobile in Persian. There are several scrambling operations that have different information-structural consequences. One of these—focus fronting—moves a phrase into a left-peripheral position. ()

Râmin ye chiz=i xarid. Hads be-zan chi Râmin t xarid. Ramin one thing= buy.. guess -hit. what Ramin buy.. ‘Ramin bought something. Guess what Ramin bought.’

Toosarvandani () proposes that focus fronting allows the wh-phrase to escape deletion, so that the sentence in () is the underlying source for the sluice in ().

.. Deriving sluicing in Persian When a phrase undergoes focus fronting, Karimi (: –) argues that it moves into the specifier of a focus projection that is located above TP but below CP. Under Toosarvandani’s account of sluicing, this is the position the wh-phrase moves into in order to escape ellipsis of TP. ()

Râmin ye chiz=i xarid. Hads bezan

CP C

= (51) FocP

DP2

chi

Foc′ Foc

TP DP1

T′ vP

Râmin t1

T v′

VP t2

v V xarid





While focus fronting is optional—compare () and ()—it must apply obligatorily when TP is elided. This follows, Toosarvandani argues (p. ), if ellipsis is licensed by Merchant’s () E-feature, located on Foc. If this head is only present in the extended verbal projection when it projects a specifier, there will always be something that undergoes focus fronting in sluicing. In languages where the wh-phrase in sluicing occupies specCP, it is not usually possible to strand a complementizer (Merchant : –). But in Persian, it is possible to leave one behind. ()

Mahin mi-xâ-d ye chiz=I be-xar-e vali yâd=esh Mahin -want.- one thing= -buy- but memory= ne-mi-yâ-d ke chi. --come.. that what ‘Mahin wants to buy something, but she doesn’t remember what.’ (Toosarvandani : )

This follows from Toosarvandani’s proposal: the complementizer is located outside the constituent that is elided. Similarly, topics can also be stranded in sluicing, as in the first answer in (). As the second answer shows, they scramble to a position above FocP (Karimi : –). () Q: Sohrâb ketâb=o film=â=ro be kes=i Sohrab book=and movie== to person= ‘Sohrab gave the books and movies to someone.’ A: Mi-dun-am ke ketâb=â=ro -know.- that book== ‘I know who Sohrab gave the books to.’

be to

dâd. give..

ki Δ. who

A: Mi-dun-am ke ketâb=â=ro be Râmin Sohrâb t t dâd. -know.- that book== to Ramin Sohrab give.. ‘I know that Sohrab gave the books to Ramin.’ All by itself, this account cannot explain why the remnant in sluicing is always a wh-phrase. Non-interrogative phrases are able to undergo focus fronting, but they cannot be stranded by ellipsis.6 () Kes=i dar mi-zan-e. *Fekr mi-kon-am bâbâ Δ. person= door -hit.- thought -do.- dad Intended: ‘Someone is knocking. I think it’s Dad.’ To prevent this overgeneration, Toosarvandani proposes (pp. –) that the E-feature only appears on a Foc head that requires a wh-phrase in its specifier (Foc[+Q]). It is not clear whether this restriction can be motivated or whether it must remain a stipulation. 6 Persian appears to violate van Craenenbroeck and Lipták’s () generalization that all and only the phrases that undergo a movement operation can serve as the remnant in a sluicing construction derived through that operation.





.. The properties of focus fronting If the wh-remnant in sluicing undergoes focus fronting, it should exhibit the prosodic and semantic effects of this process. According to Karimi (: –), a constituent that has been focus-fronted is prosodically prominent and has a contrastive interpretation. () Man divân=e ˆ = barâ Kimeâ t I Divan= Hafez= for Kimea ‘It was the Divan of Hafez that I bought for Kimea.’

xarid-am. buy.- (Karimi : )

For wh-phrase remnants in sluicing, Toosarvandani (: ) observes that “[j]ust like the nonelliptical examples of focus fronting [ . . . they] all bear a pitch accent,” pointing to sentences like (). () Faqat mi-dun-am kojâ Sohrâb dustdoxtar=esh=o did. only -know.- where Sohrab girlfriend== see.. Ne-mi-dun-am . --know.- when ‘I only know where Sohrab saw his girlfriend. I don’t know when.’ (Toosarvandani : ) In terms of semantics, focus fronting a wh-phrase is clearly not information-structurally neutral, since it is infelicitous in an out-of-the-blue context. () # Sohrâb t âvord? what Sohrab bring.. ‘What did Sohrab bring?’

(Toosarvandani : )

But it is challenging to develop a more precise characterization of the contrastive interpretation that wh-phrases receive when they undergo focus fronting. According to Karimi (: ), a question like () is only felicitous when there is a set of things under discussion and the speaker is asking which of those things Sohrab brought. In other words, focus fronting the wh-phrase evokes a salient set of alternatives that make up the possible answers to the question. In (), however, the speaker and hearer need not already be discussing the set of things that Ramin might have bought. Instead, building on earlier work by Romero (: –), Toosarvandani (: –) proposes that when a wh-phrase undergoes focus fronting, it evokes a salient set of alternatives to the wh-phrase itself. For the remnant chi ‘what’ in (), this might include wh-phrases that vary in their sortal restriction, e.g., ki ‘who’ and kojâ ‘where’. Or, it might include noun phrases that vary in the identity of the quantificational determiner, such as the indefinite noun phrase ye chiz=i ‘something’. With this semantics for focus fronting, the remnant in () would stand in a contrastive relationship to its correlate in the antecedent clause. There are types of sluicing, however, where the correlate is not an overt indefinite. Like English, Persian allows the correlate to be definite () or implicit (a–b).





() Sohrâb âlmâni balad=e vali yâd=am nist Sohrab German knowledgeable=be.. but memory= .be.. che zabun=â=ye dige=yi. what language== other= ‘Sohrab knows German, but I don’t remember what other languages.’ () a. Gorbe dâsht mi-xord. Mi-dun-i cat have.. -eat.. -know.- ‘The cat was eating. Do you know what?’

chi? what

b. Rostam mâshin=o tamir kard=e vali malum nist Rostam car= repair do.=be.. but evident .be.. kojâ. where ‘Rostam repaired the car, but it’s not clear where.’ (Toosarvandani : ) Toosarvandani’s semantics for focus fronting would probably extend to the contrastive relationship between a wh-phrase remnant and a definite correlate. But this seems less plausible for implicit correlates, since there is no overt linguistic expression for the wh-phrase to contrast with.

.. The island (in)sensitivity of sluicing Sluicing in English is famously insensitive to islands. In Persian, too, when the correlate is indefinite, the wh-remnant can start out inside a relative clause (a) or an adjunct clause (b). (Again, islands are bracketed, while correlates are bolded.) () a. Complex NP Constraint Unâ mi-xâ-n [ye nafar=ro ke yek=i az they -want.- one person= that one= from zabun=â=ye urupâyi=ro balad bâsh-e] estexdâm language== European= knowledgeable be.- hiring kon-an. Yâd=et mi-yâ-d kodum zabun Δ? do.- memory= -come.- which language ‘They want to hire someone who knows one of the European languages. Do you remember which language?’ b. Adjunct Constraint Râmin [chon ye doxtar=i=ro dust dâr-e] raft gol Ramin since one girl== friend have.- go.. flower be-xar-e. Be mâ na-goft kodum doxtar Δ. -buy- to us -say.. which girl ‘Ramin went to buy flowers since he likes a girl. He didn’t tell us which girl.’ (Toosarvandani : )





When the correlate is implicit, sluicing in Persian is sensitive to islands (a–b). It shares this property with sluicing in other languages (Chung et al. : –), which may arise from the identity condition on ellipsis (Chung ). () a. Complex NP Constraint *Unâ mi-xâ-n [ye nafar=ro ke âlmâni balad they -want.- one person= that German knowledgeable bâsh-e] estexdâm kon-an. Yâd=et mi-yâ-d az be.- hiring do.- memory= -come.- from kojâ Δ? where Intended: ‘They want to hire someone who knows German. Do you remember where they want that person to know it from?’ b. Adjunct Constraint *Moallem [chon Râmin sar=e kelâs dir umad] az=ash nomre teacher since Ramin head= class late come.. from= grade kam kard. Yâd=et mi-yâ-d cherâ Δ? little do.. memory= -come.- why ‘The teacher took off points because Ramin came late to class. Do you remember why Ramin came late to class and because of that the teacher took off points?’ But in general, ellipsis of TP remedies island violations when a remnant undergoes movement to a focus projection in sluicing, just as it does in stripping and fragment answers.

. N  

.................................................................................................................................. Persian has noun phrase ellipsis: the head noun, and possibly more material, can go missing inside DPs. () Q: Xodkâr=e sabz dâr-i? pen= green have.- ‘Do you have a green pen?’ A: [DP Δ Sabz ] na-dâr-am. [DP Δ Âbi ] mi-xâ-y? green -have.- blue -want.- ‘I don’t have a green one. Do you want a blue one?’ (Ghaniabadi : ) Ghaniabadi () identifies a wide range of elements that are able to license noun phrase ellipsis. This includes various prenominal expressions, including demonstrative determiners (a), interrogative determiners (b), and superlatives (c).7 7

The antecedent of the ellipsis in (c) is the noun kâr ‘work’.





() a. Q: Kudum ketâb gerun-tar=e? which book expensive-=be.. ‘Which book is more expensive?’ Un Δ] gerun-tar=e. that expensive-=be.. ‘That one is more expensive.’

A: [DP

b. Kodum ketâb barâ man=e; [DP kodum Δ ] barâ shomâ? which book for me=be.. which for you ‘Which book is mine, and which is yours?’ (Ghaniabadi : ) c. Gâh=i to fekr mi-kon-I kâr=i ke anjâm time= you thought -give.- work= that completed mi-d-i [DP dorost-tar-in Δ]=e. -give.- correct-- =be.. ‘Sometimes you think that whatever you do is the best.’ (Ghaniabadi : ) But not all determiners are able to license noun phrase ellipsis. In particular, with the indefinite determiner ye ‘a’, the head noun cannot go missing. () Q: Piran mi-xâ-y? shirt -want.- ‘Do you want a shirt?’ A: Âre, [DP ye yeah a ‘I want a shirt.’

*(piran)] shirt

mi-xâ-m. -want.-

While determiners and other functional elements precede the noun, modifiers and arguments follow. Some of these can license noun phrase ellipsis without the presence of a prenominal element. This is most clear for attributive adjectives, such as qermez ‘red’. (The dependents of a noun are linked to other elements within the DP by the ezafe clitic =e; Samiian .) () Q: Che piran=i mi-xa-y? what dress= -want.- ‘What dress do you want?’ A: [DP Δ Qermez] mi-xâ-m. red -want.- ‘I want a red one.’ This is not simply a pronominal form of the adjective, since multiple adjectives can be stranded. () Q: Kudum piran=o mi-xâ-y? which dress= -want.- ‘Which dress do you want?’





A:

[DP Un Δ qermez=e gerun]=o mi-xâ-m. that red= expensive= -want.- ‘I want that red expensive one.’

In addition, the missing material can be a constituent that is larger than just a noun, such as a noun and a modifying adjective. () Sinâ kif=e charm=i=ye kuchik=o bâ [DP Δ bozorg ] avaz kard. Sina bag= leather== small= with big exchange do.. ‘Sina exchanged the small leather bag for a big one.’ (=a big leather bag) (Ghaniabadi : ) Ghaniabadi claims (: –) that several types of nominal dependents cannot license noun phrase ellipsis. There is some variation in judgments here, since for another speaker modifying nouns () and prepositional phrases () can be stranded. ()

Q: Che kif=i tu=ye irân gerun=e? which bag= in= Iran expensive=be.. ‘Which bags are expensive in Iran?’ A: [DP (Kif=e) charm] gerun=e. bag= leather expensive=be.. ‘Leather ones are expensive.’

() Q: Kudum kafsh=â qashang-tar=e? which shoe= beautiful-=be.. ‘Which shoes are prettier?’ A: [DP (Kafsh=â=ye) tu=ye vitrin] qashang-tar=e. shoe== in= window beautiful-=be.. ‘The ones in the window are prettier.’ By contrast, an internally complex adjective phrase () or possessor () can never license noun phrase ellipsis. () Baz=i az keshvar=â negarân=e afzâyesh=e qeymat=e naft=and. some= from country= worried= increase= price= oil=be.. *[DP (Keshvar=â=ye) negarân=e afzâyesh=e qeymat=e naft] ettelâ’iyye=i country== worried= increase= price= oil statement= sâder kard-and. out do.- ‘Some countries are worried about the increase in the price of oil. The countries worried about the increase in the price of oil issued a statement.’ () Q: Kif=e ki peydâ shod=e? bag= who found become.=be.. ‘Whose bag has been found?’

 *(Kif=e) un khânom] peydâ bag= that woman found ‘That woman’s bag has been found.’

A: [DP



shod=e. become.=be..

It remains to be seen how these data from Persian are relevant, if at all, for a more general theory of noun phrase ellipsis.

. S

.................................................................................................................................. Persian has several elliptical constructions that resemble their counterparts in betterdescribed languages. Many of these turn out to have surprisingly different properties with different syntactic derivations. Stripping and gapping, which allow for the elided material and the antecedent to be embedded, involve the coordination of full clauses. Further, in both these constructions, a full TP is elided, accounting for the island insensitivity of stripping. Fragment answers are also insensitive to islands, because the remnant raises to a focus position located immediately above the elided TP. This focus position is also implicated in the derivation of sluicing in Persian, which otherwise does not have obligatory wh-movement. Finally, verb phrase ellipsis removes the complement of v: either the constituent headed by the non-verbal element in a complex predicate or a VP out of which the simple verb has escaped.

A I am indebted to Hayedeh Nasser and Abbas Toosarvandani, who generously provided the data for this chapter. It was also greatly improved by the comments of Annahita Farudi, Hayedeh Nasser, and two anonymous reviewers.

  ......................................................................................................................

 ......................................................................................................................

 

. I

.................................................................................................................................. T chapter overviews Polish elliptical constructions and their characteristics, including some that challenge existing theories of ellipsis. I briefly define each construction in a separate (sub)subsection before describing its main features. The sections are ordered as follows: sluicing, fragment answers, stripping, predicate ellipsis, and other elliptical constructions. The constructions vary in terms of the amount of discussion they receive, reflecting the amount of attention they have received in the linguistic literature. A separate section addresses the possibility of dropping prepositions from remnants. The picture that emerges from this chapter is that many elliptical constructions in Polish exhibit characteristics consistent with existing theories of ellipsis. At the same time, a particularly interesting fact about Polish is evidence against the PF-deletion approach to ellipsis that cuts across a range of similar constructions, that is, sluicing, fragment answers, and stripping.

. S

.................................................................................................................................. Sluicing is a construction where an interrogative phrase (a remnant) is stranded in a context in which an interrogative clause would normally appear. () and () illustrate. () Ktoś chce kibicować Niemcom w niedzielę, ale nie wiem somebody.NOM wants root for Germany on Sunday but not I know kto. who.NOM ‘Somebody wants to root for Germany on Sunday, but I don’t know who.’ () Adrian bierze na wakacje jedno dziecko, ale nie wiem które. Adrian takes on vacation one kid.ACC but not I know which.ACC ‘Adrian is taking one kid on vacation, but I don’t know which.’





The stranded interrogative phrases in () and () are embedded in the clauses following the antecedents, but they needn’t be, as () and () show. () A: Ktoś chce kibicować Niemcom w niedzielę. somebody.NOM wants root for Germany on Sunday B: Kto? who.NOM A: ‘Somebody wants to root for Germany on Sunday.’ B: ‘Who?’ () A: Adrian Adrian

bierze takes

na on

wakacje vacation

jedno one

dziecko. kid.ACC

B: Które? which.ACC A: ‘Adrian is taking one kid on vacation.’ B: ‘Which?’ Polish sluicing shows case-matching effects. This can be seen in the examples above: in () and () the remnant and the correlate (ktoś ‘somebody’) are nominative, and in () and (), the remnant and its correlate (jedno dziecko ‘one kid’) are accusative. These patterns are consistent with the PF-deletion approach (Barros b; Merchant ) and the direct interpretation approach (Ginzburg and Sag ), since both predict case identity between a remnant and its counterpart in the antecedent for languages with overt case marking. Sluicing contrasts with overt interrogative clauses in that it disobeys island constraints, as first noticed by Ross (b) for English. Example (), showing extraction from an adjunct island, illustrates this pattern for Polish. Extraction of the interrogative phrase from the overt interrogative clause in (a) leads to ungrammaticality, but the corresponding sluice is fully grammatical (b) (though, as we will see shortly, the picture is somewhat more complex; see also Abels, this volume and Vicente, this volume). () a. *Niemcy strzelili  goli zanim Adrian zdążył przeczytać dwa Germany scored  goals before Adrian managed read two emaile, ale nie pamiętam czyje (dwa emaile) Niemcy emails.ACC but not I remember whose.ACC (two emails) Germany strzelili  goli zanim Adrian zdążył przeczytać. scored  goals before Adrian managed read ‘*Germany scored  goals before Adrian managed to read two emails, but I don’t remember whose (two emails) Germany scored  goals before Adrian managed to read.’ b. Niemcy strzelili  Germany scored  emaile, ale nie emails.ACC but not ‘Germany scored  goals remember whose.’

goli zanim Adrian zdążył przeczytać dwa goals before Adrian managed read two pamiętam czyje. I remember whose.ACC before Adrian managed to read two emails, but I don’t





Sluicing remains well-formed in the following examples, in which extraction from relative clauses (), sentential subjects (), and coordinate structures () also violates island constraints. () Naj popularniejsze są dania, które przygotowują sławni kucharze, most popular are dishes that prepare famous chefs.NOM ale nie pamiętam którzy. but not I remember which.NOM ‘The most popular dishes are those that are prepared by famous chefs, but I don’t remember which (chefs).’ () To, że dwa kraje będą organizować Mistrzostwa Świata w  this that two countries.NOM will host World Cup in  roku zostało podane w wiadomościach, ale nie pamiętam które. year was reported in news but not I remember which.NOM ‘That two countries will be hosting the  World Cup was reported on the news, but I don’t remember which (countries).’ () A: Messi chciał Messi wanted B: Jaką? what.ACC

zdobyć win

puchar Cup.ACC

I and

jeszcze yet

inną another

nagrodę. award.ACC

A: ‘Messi wanted to win the World Cup and another award.’ B: ‘What (award)?’ Two exceptions to the ability of sluicing to disobey island constraints are known from research on English. One involves contrastive remnants, as shown in (). () ?Naj popularniejsze są dania, które kucharze przygotowują SŁAWNYM most popular are dishes which chefs.NOM prepare famous klientom na zamówienie, ale nie wiem KTÓRYM jeszcze. customers.DAT on order but not I know which.DAT other ‘The most popular dishes are those that chefs prepare for famous customers on request, but I don’t know which other (customers).’ Here, the remnant is degraded, unlike the remnants we have seen so far. Griffiths and Lipták () associate such degradation with remnants and correlates bearing contrastive focus.1 They argue that such remnants violate the requirement of scopal parallelism between the remnant and the correlate: the correlate can’t scope out of the island, while the remnant takes high scope. Recent research suggests further that contrastive remnants are subject to information-structural constraints that enforce structural parallelism 1

Griffiths and Lipták () observe that sentence-final phrases can always serve as correlates for contrastive and non-contrastive remnants. This pattern has been attributed to the sentence-final position representing the canonical informational focus position in English (Carlson, Walsh Dickey, Frazier, and Clifton a). Since the canonical informational focus position is also sentence-final in Polish, the same appears to hold there (though judgments are delicate).





independently of ellipsis (Clifton and Frazier ; Kertz ). Further, Schmeh, Culicover, Hartmann and Winkler () report lower acceptability ratings for English contrastive fragment answers located within island structures compared to their non-contrastive counterparts,2 which they interpret as evidence for processing difficulty associated with contrastive focus. If this is correct, then the behavior of contrastive remnants follows from independent constraints on contrastive focus. The second exception concerns sprouting, a construction minimally different from sluicing that features implicit correlates for remnants.3 In () the remnant has an implicit indirect object correlate and appears to be degraded in Polish. () ?Naj popularniejsze są dania, które przygotowują sławni kucharze, most popular are dishes that prepare famous chefs.NOM ale nie wiem komu. but not I know who.DAT ‘The most popular dishes are those that famous chefs prepare, but I don’t know for who.’ Griffiths and Lipták () connect such data in English to violation of scopal parallelism between remnants and implicit correlates (the remnant takes high scope but the correlate takes low scope). Island insensitivity of sluicing is entirely expected on the direct interpretation approach (there is no underlying structure and hence no islands to violate), while island sensitivity could be attributed to extra-grammatical factors. On the PF-deletion approach, it is attributed to ellipsis being able to repair offending structures through non-pronunciation (Merchant b; Ross b). Other possibilities include deriving remnants from () subparts of antecedents (“short sources”) rather than from entire island-violating structures (‘Everybody loves dishes famous chefs make, but I don’t know which chefs make them.’), or () copular clauses (‘Everybody loves dishes famous chefs make, but I don’t know which chefs it is.’) (Barros, Elliot, and Thoms ; Griffiths and Lipták ; Merchant ). Only the first possibility is consistent with case-matching effects that arise in Polish; the second possibility requires every remnant to be marked for nominative, contrary to fact.

2

Contrastive fragment answers embedded within islands are illustrated in (i) and non-contrastive ones in (ii), from Schmeh, Culicover, Hartmann, and Winkler (). The presence of an island degraded acceptability ratings in addition to the degradation caused by contrastive focus. (i)

A: Did Rufus call the friendly painter that painted a big room of his mother’s house? B: No, the spacious kitchen.

(ii) A: Did Rufus call the friendly painter that painted a big room of his mother’s house? B: Yes, the spacious kitchen For more discussion of fragment answers, see section .. 3 Evidence for island effects under English sprouting is found in Chung, Ladusaw, and McCloskey (). However, Culicover and Jackendoff (: n. ) cite island-disobeying examples of sprouting, such as (i). Culicover and Jackendoff suggest that the conflicting judgments found in the literature reflect extra-grammatical factors influencing the acceptability of island violations under sprouting. (i) Bob found a plumber who fixed the sink, but I’m not sure with what.





A further characteristic of Polish sluicing is its ability to appear discourse-initially. Below are examples from naturally occurring speech. ()

[Context: A is in a cab, approaching the airport] Cab driver to A: Który terminal? which terminal.NOM ‘Which terminal?’

()

[Context: A and B get in an elevator] A to B: Które piętro? which floor.NOM ‘Which floor?’

Such examples are accepted as sluicing and included in the analyses of Ginzburg and Sag (), Stainton (b), and Merchant (b, b). However, they are rare compared to sluicing with linguistic antecedents. One explanation for their rareness might come from Miller and Pullum’s () work on discourse-initial uses of English verb phrase ellipsis (VPE), which argues that such uses are subject to discourse conditions not easily satisfiable by non-linguistic antecedents. Assuming that sluicing resembles Subject Focus verb phrase ellipsis (see Kertz ),4 as suggested to me by Philip Miller (p.c.), both require the surrounding discourse to provide a salient open proposition. For example, in (), the open proposition is A go to terminal X, given that every airport has a limited number of terminals and A is headed for one of them. The direct interpretation approach and the PF-deletion approach can accommodate such examples, but the former does so more easily because it posits no underlying structure for remnants. As far as I’ve been able to determine, Polish discourse-initial sluices are marked for nominative case, and a plausible underlying source must license this case. A copular construction, shown in () for the remnant in (), which Merchant (b) proposes as a source for English discourse-initial sluicing, and longer versions of it (long clefts) are unavailable for Polish discourse-initial sluicing. ()

Który terminal which terminal.NOM ‘Which terminal is it?’

to it

jest? is

() is appropriate as a request for identification of the terminal but does not fit the open proposition evoked by the non-linguistic context in (). Long clefts (*Który terminal to jest na który jedziesz? ‘Which terminal is it you’re going to?’) are ungrammatical in Polish.5 4

(i)

Subject Focus VPE is illustrated in (i). Jim Parsons could win an Emmy Award and Mayim Bialik could too. (open proposition: X could win an Emmy Award)

5

It remains unclear why it is nominative case morphology that discourse-initial remnants surface with. One possibility is that non-linguistic antecedents impose weaker identity conditions on remnants than do linguistic antecedents, resulting in remnants being marked for default case, which is nominative in Polish (see Progovac  and Paesani  for discussion of remnants surfacing with default case morphology).





Finally, Polish sluicing is sensitive to argument structure mismatch, a pattern known from research on English. Voice mismatch or the benefactive alternation are disallowed, as shown in () and (). ()

*Ciasto zostało zrobione przez kogoś, ale nie wiem kto. cake was made by someone.ACC but not I know who.NOM ‘*The cake was made by someone but I don’t know who .’

()

*Adrian zrobił komuś ciasto, ale nie wiem dla kogo. Adrian made someone.DAT cake but not I know for who.GEN ‘*Adrian made someone a cake, but I don’t know for who .’

However, Polish allows more flexibility than English with respect to the benefactive alternation when remnants have implicit correlates. The antecedent in () is fine with either a PP or NP as a remnant, a pattern arguably due to the flexible nature of the Polish word order. Because the orders Adrian zrobił ciasto komuś ‘*Adrian made a cake someone’ and Adrian zrobił komuś ciasto ‘Adrian made someone a cake’ are both possible, the clause Adrian zrobił ciasto ‘Adrian made a cake’ doesn’t commit us to the V NP PP structure Adrian zrobił ciasto dla kogoś ‘Adrian made a cake for someone’ (the way it does in English), to the exclusion of a V NP NP structure. Hence the NP remnant komu ‘who’ may still be used. () Adrian zrobił ciasto, ale nie wiem dla kogo/ komu. Adrian made cake but not I know for who.GEN/ who.DAT ‘Adrian made a cake but I don’t know for who/*who .’ Based on the unavailability of structural mismatch in English, scholars have posited syntactic identity conditions on sluicing within the PF-deletion approach (Chung ; Merchant d). The direct interpretation approach blocks structural mismatch by requiring that remnants and their correlates correspond in terms of syntactic category and case (Culicover and Jackendoff , ; Ginzburg and Sag ).

.. Multiple sluicing Multiple sluicing is the phenomenon where several wh-phrases occur as remnants. In Polish, its availability corresponds to the availability of multiple wh-fronting in interrogative clauses. Multiple wh-fronting is illustrated in ()–(), and corresponding multiple sluicing is illustrated in ()–(). ()

Kto kogo zauważył who.NOM who.ACC saw ‘Who saw who in the street?’

()

Kogo gdzie wysłała who.ACC where sent ‘Who did Eliza send where?’

Eliza? Eliza

na on

ulicy? street

 ()

 Ktoś zauważył kogoś na ulicy, ale nie wiem kto someone.NOM saw someone.ACC on street but not I know who.NOM kogo. who.ACC ‘Someone saw someone in the street, but I don’t know who saw who.’

() Eliza wysłała kogoś gdzieś, ale nie wiem kogo gdzie. Eliza sent someone.ACC somewhere but not I know who.ACC where ‘Eliza sent someone somewhere but I don’t know who where.’ Stjepanović (, ) notes that multiple sluicing in Serbo-Croatian differs from multiple wh-fronting by showing superiority effects. The linear order of remnants must be such that the one that’s syntactically superior to another remnant precedes it (but see Bošković , ,  for arguments that superiority effects show up in Serbo-Croatian interrogatives, as well). The behavior of Polish is consistent with these patterns. The order of the wh-phrases can be reversed in ()–() with no loss of acceptability, but the multiple sluicing examples in ()–() show superiority effects, that is, a severe degradation in acceptability for the reversed orders of the wh-phrases, as in ()–(). ()

*Ktoś zauważył kogoś na ulicy, ale nie wiem kogo someone.NOM saw someone.ACC on street but not I know who.ACC kto. who.NOM

() *Eliza wysłała kogoś gdzieś, ale nie wiem gdzie kogo. Eliza sent someone.ACC somewhere but not I know where who.ACC Such a discrepancy between multiple sluicing and interrogative clauses can be problematic for the PF-deletion account more than the direct interpretation approach, which doesn’t expect correspondences between (multiple) sluicing and interrogative clauses. Stjepanović (, ) addresses this problem by positing different structures for multiple sluicing and multiple fronting in interrogative clauses. The latter are argued to undergo fronting not via overt wh-movement, but via focus movement, which voids superiority violations. But multiple sluicing is subject to Superiority, since it features overt wh-movement of remnants. A different line of argument is adopted in Grebenyova () for Russian. She attributes the behavior of sluicing to an independent requirement of scopal parallelism (cf. the discussion of island amelioration by sluicing in section .) between the antecedent clause and the elliptical one hosting the remnants.6 This requirement captures the Polish facts as well: reversing the order of the correlates in ()–(), produces acceptable remnants, as in ()–().

6

For a detailed account of why PF-deletion can’t rescue superiority violations while it rescues island violations, see also Bošković ().





() Kogoś zauważył ktoś na ulicy, ale nie wiem kogo someone.ACC saw someone.NOM on street but not I know who.ACC kto. who.NOM () Eliza wysłała gdzieś kogoś, ale nie wiem gdzie kogo. Eliza sent somewhere someone.ACC but not I know where who.ACC These data also fit in with the idea articulated in Hawkins () that structurally parallel antecedents and ellipsis sites have a processing advantage over non-parallel ones. Another characteristic of multiple sluicing to be discussed here is the ability of remnants to have correlates in separate tensed clauses. Merchant () points out that English multiple sluicing lacks this ability, as shown in () from Lasnik (: ). ()

*One of the students said that Mary spoke to one of the professors, but I don’t know which student to which professor.

Lasnik (), however, argues that such multiple sluicing can reach into separate clauses in Serbo-Croatian. Polish multiple sluicing follows the pattern of English: example (), in which the remnants have correlates in separate clauses, is ill-formed. () *Jakiś klient twierdzi, że Eliza some customer.NOM says that Eliza ale nie wiem który klient but not I know which customer.NOM

wysyła sends które which

do niego jakieś emaile, to him some emails.ACC emaile emails.ACC

Interestingly, the corresponding interrogative clause in () seems much better than (). () Który klient które emaile twierdzi, że Eliza do niego wysyła? which. customer.NOM which emails.ACC says that Eliza to him sends ‘Which customer says that Eliza sends what emails to him?’ These patterns are problematic for the PF-deletion approach, since wh-phrases should be able to move across clause boundaries both in interrogative clauses and in sluicing.7

. F 

.................................................................................................................................. Typical fragment answers are elliptical responses to interrogative clauses. They are known to share several features with sluicing, and hence accounts of fragment answers and sluicing proceed along similar lines (Ginzburg and Sag ; Griffiths and Lipták ; 7

See, however, Lasnik () for a rightward movement account of English apparent multiple sluicing, which avoids this difficulty.





Merchant a). Shared features in Polish include case-matching effects, discourse-initial use, resistance to argument structure mismatches, and island insensitivity. Example () shows the remnant and its correlate obligatorily sharing accusative case. () A: Kogo who.ACC

Eliza Eliza

widziała saw

na on

ulicy? street

B: Swojego trenera/ *Swój trener. Her coach.ACC/ *her coach.NOM A: ‘Who did Eliza see in the street?’ B: ‘Her coach.’ Examples ()–() illustrate discourse-initial fragments. () [Context: A is looking for a book in B’s office] B to A: Górna półka. top shelf.NOM ‘Top shelf.’ () [Context: A sees his five-year-old child eating his food with his fingers] A to child:

Noż i knife.NOM and ‘Knife and fork!’

widelec!/ fork.NOM/

Nożem knife.INSTR

i and

widelcem! fork.INSTR

The case occurring on discourse-initial fragments is usually nominative, although example () shows that instrumental is also possible.8 As is the case with sluicing remnants, a copular construction and long clefts are unavailable here, challenging the PF-deletion approach. The direct interpretation approach can accommodate nominative- and instrumental-marked fragments (as well as sluicing remnants) via the dialogue gameboard (DGB), which records current questions under discussion and the focus-establishing constituents (FECs) within them (see Ginzburg  for more detail). FECs are phrases to which fragments refer that combine semantic, syntactic, and phonological information, as needed. Like sluicing, fragment answers don’t tolerate argument structure mismatches, such as those involving the benefactive alternation or voice. Examples ()–() illustrate. ()

A: Komu who.DAT

Eliza Eliza

kupiła bought

prezent? present.ACC

B: *Dla siebie for herself.ACC A: ‘Who did Eliza buy the present?’ B: ‘*For herself .’

8 To the best of my knowledge, there is as yet no empirical data on whether there are acceptability/ frequency differences between the instrumental and nominative case in such examples or whether the cases lead to differential processing costs.

 () A: Przez By

kogo who.ACC

prezent present

został was



kupiony? bought

B: *Eliza. Eliza.NOM A: ‘Who was the present bought by?’ B: ‘*Eliza .’ As for island insensitivity, fragment answers closely track sluicing. Although Merchant (a) argued that fragment answers obey island constraints, later research has shown this to be partially incorrect. Griffiths and Lipták () demonstrate that English fragment answers violate island constraints except when fragments and their correlates located in non-sentence-final position are contrastive or when fragments refer back to implicit correlates in a configuration parallel to sprouting. () illustrates the first case and () the second; () seems more degraded than (). ()

A: Eliza Eliza od from

zna osoby, które mówią, że lubią muzykę klasyczną knows people who say that they.like music.ACC classical zawsze. forever

B: *Nie, jazz. no jazz.ACC A: ‘Eliza knows people who say they’ve liked classical music since time immemorial.’ B: *‘No, jazz.’ () A: Eliza Eliza B: ?Tak, yeah

wyszła left na at

z from

siebie each

pokoju, room

bo because

wszyscy everybody

krzyczeli. yelled

nawzajem. other.ACC

A: ‘Eliza left the room because everybody was yelling.’ B: ‘Yeah, at each other.’ Examples ()–() contrast with the non-contrastive fragment in (), where the island violation causes no degradation in acceptability. ()

A: Eliza Eliza od from B: Tak, yeah

zna osoby, które mówią, że lubią muzykę klasyczną knows people who say that they.like music.ACC classical zawsze. forever Mozarta. Mozart.ACC

A: ‘Eliza knows people who say they’ve liked classical music since time immemorial.’ B: ‘Yeah, Mozart.’ Fragment answers can be embedded. That genuine embedding is involved can be tested by applying Rooryck’s () criteria (which Temmerman () also applies to Dutch embedded fragment answers). Consider example ().





() A: Kto who.NOM B: (a) Myślę, I think

się zawsze spóźnia REFL always is late że Eliza. that Eliza.NOM

(b) Wszyscy everybody

myślą, thinks

(c) Wszyscy everybody

myśleli, thought

że that

do for

pracy? work

Eliza. Eliza.NOM

że that

Eliza. Eliza.NOM

(d) Naprawdę myślę, że Eliza. really I think that Eliza.NOM (e) Nikt nie myśli, że Eliza. nobody not thinks that Eliza.NOM The B-replies reveal all the characteristics of non-parentheticals identified by Rooryck (): the subject can vary ((a) and (b)), the tense can vary ((b) and (c)), adverbs can be inserted (d), and negation of the main clause is possible (e). Note further that an overt complementizer (że ‘that’) is required in such fragments, just as it is required in embedded clauses.

.. Split questions Split questions are constructions in which antecedent clauses are questions followed by remnants providing tentative answers to these questions, both uttered by the same speaker (Arregi ). We find in them characteristics similar to those we have found in fragment answers, that is case-matching effects, resistance to argument structure mismatches, and island insensitivity. Examples of these characteristics are shown in ()–(). To test for island sensitivity in (), I use an in situ wh-question. () Kto się zawsze spóźnia who.NOM REFL always is ‘Who is always late for work, Eliza?’ ()

*Kto odnalazł Elizę, who.NOM found Eliza.ACC ‘*Who found Eliza, by her coach?’

do late

przez by

pracy, for work

jej her

Eliza? Eliza.NOM

trenera? coach.ACC

() Eliza zna osoby, które mówią, że lubią co, Eliza knows people who say that they like what.ACC klasyczną? classical ‘Eliza knows people who say they like what, classical music?’

muzykę music.ACC





. S

.................................................................................................................................. In stripping (as first introduced in Hankamer and Sag ), a single DP remnant is left behind that has the semantic content of a clause. I adopt the broader definition of stripping (or Bare Argument Ellipsis) found in Culicover and Jackendoff (), where stripping is not limited to declarative and coordinate clauses (for the narrow definition of stripping, see Lobeck , Merchant a, and Wurmbrand , among others). This construction occurs quite freely in Polish, both in embedded and non-embedded contexts, and it can violate island constraints, as in (). () Eliza chodzi na koncerty, ale niezbyt Eliza goes to concerts but not too ‘Eliza goes to concerts, but not too often.’ ()

często. often

A: Eliza chodzi na koncerty. B: Często? Eliza goes to concerts often A: ‘Eliza goes to concerts.’ B: ‘Often?’

() Eliza była na koncercie i ja też. Eliza was at concert and me too ‘Eliza was at a concert and me too.’ () A: Eliza była na koncercie. Eliza was at concert.LOC A: ‘Eliza was at a concert.’

B: Długim? long.LOC B: ‘A long one?’

() Eliza poszła na koncert zaraz jak wróciła z podróży Eliza went to concert.LOC right when she returned from trip.GEN służbowej, nie wiem czy długiej. business not I know if long.GEN ‘Eliza went to a concert when she returned from a business trip, and I don’t know if the trip was long.’ Note that stripping can be embedded in an interrogative clause, following the interrogative complementizer, as in (). The case marked on the remnant corresponds to the case marked on its correlate, ruling out the possibility that a copular clause (I don’t know whether it was long) underlies such remnants, in which case the remnant would always be nominative. There are discourse-initial uses of stripping in Polish. Consider, for instance, the examples in () and (), similar to English examples reported in Hankamer and Sag (: n. ). () [Context: A bends down to kiss his son] Son to A: Nie przy ludziach! not in front of people ‘Not in front of people!’





() [Context: A is about to put her feet up on the table] B to A: Nie na stole! not on table ‘Not on the table!’ A cautionary note is in order here. We could alternatively define these examples as discourseinitial fragment answers on the view that stripping appears only in coordinate clauses.

. O  

.................................................................................................................................. An interesting fact about Polish is that it provides no support for the so-called PrepositionStranding Generalization (PSG). The PSG, given in (), was first formulated by Merchant (: ) as a way to capture an apparently cross-linguistic correspondence between overt interrogative clauses and sluicing remnants. It constitutes one of the strands of evidence supporting the PF-deletion approach to ellipsis (Ross b; Merchant , a) in that it suggests that sluicing remnants have full sentential sources, which are unpronounced. () Preposition-Stranding Generalization A language L will allow preposition stranding under sluicing iff L allows preposition stranding under regular wh-movement.9 The PSG predicts that preposition omission under sluicing is impossible in Polish, because it disallows preposition stranding in overt interrogative clauses. Although Merchant () lists Polish among the languages supporting the PSG, later research shows that it doesn’t behave as predicted (Dadan ; Nykiel ; Sag and Nykiel ; Szczegielniak ). Preposition omission under sluicing is illustrated in (), and the corresponding interrogative clause, which disallows preposition stranding, is given in (). While () is ungrammatical, () is not. () A: Premier powołał Prime minister referred B: Jakie (badania)? what.ACC (research)

się REFL

na to

jakieś some

badania research.ACC

naukowe. scientific.ACC

A: ‘The Prime Minister referred to some scientific research.’ B: ‘What (research)?’ () *Jakie badania naukowe powołał się premier na? what.ACC research.ACC scientific.ACC referred REFL Prime Minister to ‘What scientific research did the Prime Minister refer to?’ 9

In referring to the absence of prepositions from remnants, I use preposition omission as a more theory-independent term that avoids suggesting that remnants contain underlying structure and separates missing prepositions from stranded ones, a phenomenon known as swiping and found in some Germanic languages (Merchant ).





Szczegielniak () is the first to notice that examples like () are well-formed in Polish. He contrasts them with examples in which bare wh-phrases are used instead of which/ what-NP phrases, as in (), which he finds unacceptable. () A: Premier Prime Minister

powołał referred

się REFL

na to

coś. something.ACC

B: Co? what.ACC A: ‘The Prime Minister referred to something.’ B: ‘What?’ Examples () and () do differ in acceptability (() is worse than ()), though not in the way reported in Szczegielniak (). Adopting the PF-deletion approach to sluicing, Szczegielniak attributes the acceptability difference to different underlying sources for the remnants: the underlying source for (B) is an interrogative clause, but the underlying source for (B) is a cleft structure. These sources are shown in () and (). The PFdeletion approach assumes that what is to become a remnant moves to clause-initial position and the rest of the clause deletes. ()

*Co

() Jakie ? what.ACC ‘What scientific research was it that the Prime Minister referred to?’ The derivation in () violates the constraint that bans prepositional objects from moving out of PPs in Polish, and hence is ill-formed. Szczegielniak () argues that the derivation in () avoids this problem, because only part of the prepositional object (the wh-pronoun) moves out of the PP in this type of cleft structure. However, as Nykiel () notes, there is an immediate problem with structures like (): they incorrectly prevent the derivation of remnants with overt head nouns, that is, jakie badania ‘what research’ vs just jakie ‘what’ (cf. the remnant in ()), given that head nouns stay in situ. An experimental investigation of data like () and () reveals that neither of them is as unacceptable as preposition stranding in interrogative clauses in () and () (Nykiel ). The acceptability of both () and () is also significantly higher than that of the cleft structure in (). These findings show that if sluicing remnants in Polish derive from full underlying sources, these sources are not clefts, challenging the ability of the PSG to capture the behavior of prepositions in sluicing. At the same time, Nykiel () finds that prepositionless remnants are overall less acceptable than remnants with prepositions. They are also less frequent in corpus data, accounting for . percent in a sample of  sluicing remnants and fragments (Nykiel ). Omission of prepositions is found outside sluicing in Polish. Examples of prepositionless fragments are given in ()–(), including split questions ().

 ()

 A: Z którym with which B: Swoim. hers.INSTR

trenerem coach.INSTR

się REFL

Eliza Eliza

widziała? met

A: ‘Which coach did Eliza meet with?’ B: ‘Hers.’ () A: Do to

którego which

sławnego famous

aktora actor.GEN

jestem am I

podobny? similar

B: Toma Cruise’a. Tom Cruise.GEN A: ‘Which famous actor do I look like?’ B: ‘Tom Cruise.’ ()

Z kim Eliza się widziała, with who.INSTR Eliza REFL met ‘Who did Eliza meet with, her coach?’

swoim her

trenerem? coach.INSTR

Examples () and (), repeated here as () and (), illustrate omission of prepositions under stripping. () A: Eliza Eliza

była was

na at

koncercie. concert.LOC

A: ‘Eliza was at a concert.’

B: Długim? long.LOC B: ‘A long one?’

() Eliza poszła na koncert zaraz jak wróciła z podróży Eliza went to concert.LOC right when she returned from trip.GEN służbowej, nie wiem czy długiej. business not I know if long.GEN ‘Eliza went to a concert when she returned from a business trip, and I don’t know if the trip was long.’ Multiple sluicing also appears to allow prepositionless remnants if they precede remnants with prepositions, as in ().10 () Eliza pisze do kogoś z jakiegoś komputera, ale nie Eliza writes to someone.GEN from some computer.GEN but not wiem kogo z którego komputera. I know who.GEN from which computer.GEN ‘Eliza writes someone on some computer but I don’t know who on which computer.’ Stjepanović (: ) and Dadan () argue that Serbo-Croatian and Polish, respectively, disallow preposition omission under multiple sluicing. Unlike in (), in their examples only the final remnant has a corresponding PP in the antecedent. Their data thus confirm that the final remnant can’t be prepositionless but don’t rule out the possibility that the first remnant can be. See also Lasnik () for evidence that English multiple sluicing is better if final remnants are PPs instead of NPs. 10





The reverse order of remnants is degraded, as in (). () *Eliza pisze do kogoś z jakiegoś komputera, ale nie Eliza writes to someone.GEN from some computer.GEN but not wiem do kogo którego komputera. I know to who.GEN which computer.GEN ‘Eliza writes someone on some computer but I don’t know who on which computer.’ Omission of prepositions is also found in examples that appear to involve nonsentential antecedents. Like sluicing remnants, the remnants in ()–() are prepositionless wh-phrases matching the case of their correlates.11 () A: Jak how

tam there

z with

pracą? work.INSTR

B: Jaką pracą? what work.INSTR A: ‘How’s the work situation?’ B: ‘What work situation?’ () A: Co what

z with

B: Którą which

naszą our

wycieczką? trip.INSTR

wycieczką? trip.INSTR

A: ‘What about our trip?’ B: ‘Which trip?’ () A: Ja I

do to

dyrektora. manager.GEN

B: Którego? which.GEN A: ‘I’d like to see the manager.’ B: ‘Which?’ () A: Dzięki thanks B: Jaką what

za for

pomoc. help.ACC

pomoc? help.ACC

A: ‘Thanks for your help.’ B: ‘What help?’ That the antecedents are nonsentential is suggested by their temporal reference. Each of them can only refer to the present or the time of utterance (a characteristic of nonsententials; see Progovac ), while a full sentential source would exhibit the usual three-way tense contrast (past, present, or future) in Polish. One could argue that the PSG doesn’t Examples () and () are reprise utterances (cf. Ginzburg and Sag ). By using such utterances, speakers ask for clarification of parts of the antecedent. 11





apply to examples like these. But doing so doesn’t bring us any closer to understanding why these remnants closely track the behavior of the sluicing remnants discussed in section .. There is one context where omission of prepositions is banned. These are remnants corresponding to covert phrases, including both sprouting and fragment answers. For example, () hosts a remnant that is mandatorily a PP. The case marked on the remnant corresponds to the one marked on the prepositional object of the covert PP selected for by the verb in the antecedent clause. () A: Musimy porozmawiać. we must talk A: ‘We must talk.’

B: *(O) czym? about what.INSTR B: ‘What about?’

Similarly, the fragment in () corresponds to a covert phrase and must be realized as a PP. () A: Musimy porozmawiać. we must talk A: ‘We must talk.’

B: Tak, *(o) nowym yeah about new B: ‘Yes, about a new car.’

samochodzie. car.INSTR

The unavailability of preposition omission for such remnants is fully general crosslinguistically (Chung ) and has been argued, like the impossibility of argument structure mismatches (cf. section .), to speak in favor of identity conditions on remnants and their correlates that involve syntactic category (Chung ). Polish is not unique in its behavior with respect to omission of prepositions. Languages without preposition stranding like Russian, Czech, Serbo-Croatian, Greek, Bahasa Indonesian, and Emirati Arabic allow preposition omission in sluicing, and appropriate underlying structures from which to derive prepositionless remnants have not been reported for them (Caha ; Fortin b; Kluck ; Leung ; Philippova ; Stjepanović , ). Bošković (b) proposes, contra Merchant (), that prepositionless remnants are entirely expected, because sluicing can repair preposition-stranding violations occurring in interrogative clauses. He admits, however, that the proposed repair mechanism seems to overgenerate. This is certainly so in Polish, given the lower acceptability and frequency of prepositionless remnants compared to remnants with prepositions. Nykiel (, ) addresses such differences in acceptability and frequency by articulating a processing-based account of the availability of preposition omission in sluicing and fragment answers. Given that the types of remnant discussed here require corresponding PPs in the antecedents, remnants with prepositions match the syntactic category of the corresponding PPs. This option facilitates retrieval of the corresponding phrases and reflects a general preference for structural parallelism in ellipsis (Arregui, Clifton, Frazier, and Moulton ; Frazier and Clifton ; Kertz ; Merchant ). Prepositionless remnants can be seen as reduced versions of remnants with prepositions.12 Form reduction is constrained by independent factors, such as the semantic and syntactic content of the corresponding PPs in the antecedents (see Nykiel ; Sag and Nykiel ), which play a role in anaphora resolution (Ariel ; Karimi, In this regard, see also the P-drop rule of Stjepanović (, ), which can be interpreted as form reduction. Stjepanović proposes this rule in Serbo-Croatian as a rule unique to simple sluicing. 12





Fukumura, Ferreira, and Pickering ) and more generally in efficient language processing (Hawkins , ). Nykiel’s () proposal also offers an explanation in terms of structural persistence (i.e., reuse of syntactic structure) for why languages without preposition stranding drop prepositions from remnants less than languages with preposition stranding.13 In sum, Polish remains problematic for the PF-deletion approach so long as the PSG is assumed to be correct. But if we permit the grammar to place no constraints on the availability of preposition omission and employ independent processing principles to account for it, these principles can constrain the generation of prepositionless remnants. This is the line of argumentation originally proposed in Sag and Nykiel (), and the proposals of Nykiel (, ) further develop this idea. It is also worth noting that processing principles can be accommodated on Bošković’s (b) PF-deletion account as a means to constrain it.

. P 

.................................................................................................................................. In this section, I discuss three elliptical constructions: VPE, verb-stranding verb phrase ellipsis (VVPE), and polarity ellipsis.14 These are the types of predicate ellipsis Gribanova (a) lists for Russian. The constructions differ in terms of which part of the VP survives ellipsis: VPE leaves behind an auxiliary verb (), VVPE leaves behind the main verb (), and polarity ellipsis leaves behind a negative polarity head (). () Eliza będzie wychodzić z domu po  wieczorem, i ja też będę. Eliza will leave from house after  p.m. and I also will ‘Eliza will be leaving the house after  p.m. and I will too.’ () Eliza przeniosła półki na książki do większego pokoju i kanapę do Eliza relocated shelves for books to bigger room and couch to mniejszego, a Adrian nie przeniósł. smaller but Adrian not relocated ‘Eliza relocated book shelves to a bigger room and a couch to a smaller room, but Adrian didn’t.’ () Eliza będzie wychodzić z domu po  wieczorem, a ja nie. Eliza will leave from house after  p.m. but I not ‘Eliza will be leaving the house after  p.m., but I won’t.’ According to Gribanova (a), VVPE can be posited in Russian for contexts where antecedents host two coordinated VPs as a simpler analysis than argument drop. For 13

Present-day English corpus data show a higher frequency of prepositionless remnants (. percent) than Polish, although in earlier English prepositionless remnants were outnumbered by remnants with prepositions (Nykiel , ). 14 Another type of predicate ellipsis, pseudogapping, is unavailable in Polish.





examples like (), an argument drop analysis would require elision of coordinated non-constituents or of pairs of NPs and PPs, which is problematic. I take coordination of VPs as a reasonable diagnostic of VVPE in Polish, although diagnostics of VVPE and argument drop continue to be debated (see Bošković ; Goldberg ; Runić ; Sener and Takahashi ). Aspectual mismatch is possible for VVPE. VVPE in general imposes strict identity requirements on the remnant, such that the stem of the remnant verb must match the stem of the verb hosted by the antecedent clause,15 while tense, number, person, and aspect may differ. () illustrates.16 () Eliza przeniosła książki do większego pokoju i kanapę do mniejszego, Eliza relocated. books to bigger room and couch to smaller {*a Adrian przemieści jutro/ a Adrian przeniesie jutro/ a and Adrian will.move tomorrow/and Adrian will.relocate. tomorrow/and pozostali przeniosą jutro/ a pozostali przenoszą teraz}. others will.relocate. tomorrow/and others relocate. now Example () further shows that voice can mismatch under VVPE. () A: Książki powinny być przeniesione do większego pokoju i kanapę books should be relocated to bigger room and couch do mniejszego. to smaller B: Zobaczymy we will see

czy if

ktoś someone

przeniesie. will.relocate

A: ‘Books should be relocated to a bigger room and a couch to a smaller room.’ B: ‘Let’s see if someone does.’ Aspectual mismatch is impossible under VPE. The auxiliary Polish employs in VPE is imperfective and requires an imperfective antecedent verb, as shown in ().17 Voice mismatch appears to be unavailable under VPE, as well. ()

Eliza wychodzi/ *wyszła z domu po  wieczorem, to ja też Eliza leaves./ *left. from house after  p.m. so I also będę. will. ‘Eliza left the house after  p.m., so I will too.’

15 For evidence for this requirement in various languages, see Doron () and Goldberg () for Hebrew, McCloskey (a, , ) and Goldberg () for Irish, and Gribanova (a,b) for Russian. 16 In () X. stands for imperfective and X. stands for perfective. 17 General unavailability of aspectual mismatch has also been reported for Serbian VPE. See Todorović () for discussion and a phase-based account of the Serbian facts.





Wh-extraction out of VPE sites is degraded in simple sentences, as in (). () *Eliza będzie pamiętać kogoś znającego języki bałkańskie, Eliza will remember someone speaking languages Balkan ale nie wiem kogo będzie. but not I know who she will ‘*Eliza will remember someone speaking Balkan languages, but I don’t know who she will.’ But wh-extraction improves when the antecedent and the elliptical clause contrast, as in (). () Wiem kogo Eliza będzie zwalniać i wiem kogo I know who Eliza will fire and I know who ‘I know who Eliza will be firing and I know who she won’t.’

nie not

będzie. she will

Wh-extraction is also degraded in island contexts, as in (). () *Eliza będzie zatrudniać osoby, które znają jakiś język bałkański, Eliza will hire people who know a language Balkan ale nie wiem który będzie. but not I know which she will ‘*Eliza will be hiring people who speak a Balkan language, but I don’t know which she will.’ The wh-extraction facts show that Polish VPE contrasts with sluicing, which would be fine in both () and (), closely tracking the patterns familiar from English (see Fox and Lasnik ; Merchant b). Merchant (b) accounts for this contrast by formulating the condition of MaxElide that ensures that ellipsis targets a maximally large constituent if wh-extraction is involved.

. O  

..................................................................................................................................

.. Gapping Gapping is a construction in which a verb, sometimes with an object, is elided in a coordinated construction. Examples ()–() illustrate. () Eliza kupi karczochy, a Adrian Eliza will buy artichokes.ACC and Adrian ‘Eliza will buy artichokes and Adrian zucchini.’

cukinie. zucchini.ACC

() Eliza czyta książki w bibliotece, a Adrian w pobliskiej kafejce. Eliza reads books in library.LOC and Adrian at local café.LOC ‘Eliza reads book in the library and Adrian at a local café.’





Johnson () notes, drawing on Hankamer’s () observation, that the elided material must include the highest verb in its own clause and must match the highest verb in the antecedent clause. Examples () and () are ill-formed in Polish. In () the intended interpretations include the highest verb in the antecedent clause but not the highest verb in the clause containing the elided material. In () the highest verb in the clause containing the elided material mismatches the highest verb in the antecedent clause. () *Eliza kupi karczochy, a wydaje mi się, że Adrian Eliza will buy artichokes.ACC and seems to me REFL that Adrian cukinie. zucchini.ACC : ‘Eliza will buy artichokes and it seems to me that Adrian will buy zucchini.’ () *Wydaje mi się, że Eliza kupi karczochy, a Adrian seems to me REFL that Eliza will buy artichokes.ACC and Adrian cukinie. zucchini.ACC : ‘It seems to me that Eliza will buy artichokes and Adrian will buy zucchini.’ Gapping remnants are known to behave like constituents that have undergone movement (Neijt ). Polish gapping is consistent with this observation. For example, gapping can’t reach into adjunct clauses () or tensed clauses (). However, () improves when the matrix subject is identical to the embedded subject in both clauses (). () *Eliza wyszła z pokoju, bo wszyscy śpiewali hymny, a Adrian Eliza left from room because everybody sang hymns.ACC and Adrian piosenki.

songs.ACC : ‘Eliza left the room because everybody sang hymns and Adrian because everybody sang songs.’ () *Eliza mówi, że Monika lubi muzykę klasyczną, a Adrian jazz.ACC : ‘Eliza says Monika likes classical music and Adrian says she likes jazz.’ ()

Eliza Eliza

mówi, że lubi muzykę klasyczną, a Adrian ACC ‘Eliza says she likes classical music and Adrian says he likes jazz.’ Example () also improves if the complementizer że ‘that’ is left overt, as in (). () Eliza mówi, że Monika lubi muzykę klasyczną, a Adrian , że Eliza says that Monika likes music.ACC classical and Adrian that





jazz. jazz.ACC ‘Eliza says Monika likes classical music and Adrian says she likes jazz.’ It’s possible that a combination of gapping and stripping is involved in sentences like (). Gapping elides the main verb mówić ‘say’, leaving behind the clausal complement, and stripping elides the material inside the clause following the complementizer.

.. Right-node raising Right-node raising (RNR) is a term referring to constructions like (). () Eliza musiała , a Adrian Eliza had and Adrian nową ustawę new act.ACC ‘Eliza had to and Adrian wanted to support the new act.’

chciał wanted

poprzeć support

I discuss it here because one of the theoretical accounts of RNR assumes that material in the first conjunct undergoes deletion and is present overtly on the right edge of the second conjunct (Bošković b; Ha b; Wilder ). This account is supported by the availability of sloppy and strict identity readings for examples like (), a pattern also found in VPE. () Eliza nie jeździ

a but

ja I

jeżdżę ride

na on

However, cumulative agreement, shown in (), challenges the ellipsis account by having plural agreement marking on the second verb in the right conjunct instead of the expected singular (see Bachrach and Katzir ; Grosz ; McCawley  for multidominance accounts). () Eliza się cieszy, że Sonia, a Adrian się martwi, że Ela Eliza REFL is happy that Sonia but Adrian REFL worries that Ela znalazły nową pracę. found.PL new job.ACC ‘Eliza is happy that Sonia, but Adrian is worried that Ela, have found a new job.’ Below are further examples of RNR. If the two conjuncts share an object, as in (), the case marked on that object is the case assigned by the right conjunct. Such examples support the ellipsis account better than the multidominance account (but see Citko  for a defense of a multidominance account of Polish RNR). () Eliza poparła , a Adrian był przeciw nowej ustawie. Eliza supported but Adrian was against new act.DAT ‘Eliza supported and Adrian was against the new act.’





On the other hand, RNR allows sharing of material that can’t be elided under any known types of ellipsis. For instance, a prepositional object can be shared, if both conjuncts host (contrasting) prepositions (). ()

Adrian znalazł telefon pod a Eliza znalazła telefon Adrian found phone.ACC under and Eliza found phone na stole. on table.LOC ‘Adrian found the phone under and Eliza found the phone on the table.’

The shared material can be a non-constituent, as in (), but non-constituents can’t be elided. () Eliza kupuje a Adrian piecze ciasta Eliza buys but Adrian bakes cakes.ACC dla dzieci. for kids.GEN ‘Eliza buys, but Adrian bakes, cakes for kids.’

.. Non-constituent coordination Non-constituent coordination (NCC) refers to omission of material located on the left edge of the second conjunct. The material that remains doesn’t form a syntactic constituent. In (), the elements of the second conjunct correspond to the direct object and the PPadjunct of the embedded clause located in the first conjunct. () Eliza powiedziała Adrianowi, że złapała pokemony w czwartek, a Eliza told Adrian that she caught Pokemon on Thursday and

piłki we wtorek. balls on Tuesday ‘Eliza told Adrian that she’d caught Pokemon on Thursday and balls on Tuesday.’ Ellipsis analyses of NCC include a deletion operation removing the material enclosed in angular brackets and hence avoid having to countenance coordination of nonconstituents (Beavers and Sag ; Crysmann ; Sailor and Thoms ; Wilder , ). Sailor and Thoms () argue for a combination of movement and deletion, such that each phrase in the second conjunct moves separately, and deletion of the remaining material follows. This mechanism is intended to prevent deletion of nonconstituents.

.. Comparative deletion Polish has two standard markers, one associated with clausal comparatives () and the other with phrasal comparatives (), the latter showing preposition-like behavior by assigning genitive case to its object. Here, the standard contrasts with the object of the comparative predicate.

 () Eliza zebrała więcej grzybów niż Adrian Eliza picked up more mushrooms than Adrian.NOM .

‘Eliza picked up more mushrooms than Adrian picked up.’ ()



(zebrał) (picked up)

Eliza zebrała więcej grzybów od Adriana. Eliza picked up more mushrooms than Adrian.GEN ‘Eliza picked up more mushrooms than Adrian.’

When the standard contrasts with the subject of a comparative predicate, phrasal comparatives () are degraded compared to clausal comparatives () (Pancheva ). () Więcej pokemonów siedzi w parkach niż more Pokemon sit in parks than ‘More Pokemon sit in parks than sit in apartments.’ () ?Więcej pokemonów siedzi w parkach od more Pokemon sit in parks than ‘More Pokemon sit in parks than in apartments.’

w in

mieszkanich. apartments.LOC

mieszkań. apartments.GEN

Pancheva (, ) interprets examples ()–() as evidence against the Direct Analysis of Slavic comparatives such that phrasal standard markers take DP complements (with no ellipsis involved), while clausal standard markers take CP complements undergoing ellipsis (Bhatt and Takahashi ; Kennedy , ). Instead, she argues for a version of the Reduction Analysis of comparatives where both phrasal and clausal standard markers involve ellipsis (Bresnan ; Lechner , ; Merchant ; Ross ).

.. Antecedent-contained deletion Antecedent-contained deletion (ACD) refers to VPE contained within its antecedent. An example appears in (). () Eliza będzie łapać wszystkie te pokemony co/które Adrian będzie. Eliza will catch all those pokemon that Adrian will ‘Eliza will catch all the Pokemon that Adrian will.’ The incompleteness of the clause following the relativizer is explained by positing an underlying VP, and a covert movement operation (Quantifier Raising) is evoked to relate to each other the antecedent and the ellipsis site within it (Fiengo and May ; Fox ; Kennedy ; May ; Merchant ). In addition, Polish allows omission of the auxiliary under ACD, as in ().18 18 Absence of the auxiliary has been argued to lead to a degradation in acceptability for examples hosting the relative marker który ‘which’ as opposed to the relative marker co ‘what’ (Szczegielniak , ), a judgment I don’t share with regard to example ().





() Eliza będzie łapać wszystkie te pokemony Eliza will catch all those Pokemon ‘Eliza will catch all the Pokemon that Adrian will.’

co/które that

Adrian. Adrian

Although examples like () are analyzed as instances of VPE by Szczegielniak (, ), they are analyzed as instances of relative deletion by van Craenenbroeck and Lipták (), a phenomenon they classify as a type of sluicing.

.. Noun phrase ellipsis Noun phrase ellipsis (NPE) is evoked to account for the absence of the head noun within an NP. NPE is licensed in Polish by determiners and adjectives, which exhibit morphological agreement with the antecedents. ()–() illustrate. () A: Które which

pytanie question.ACC

wybierasz you choose

najpierw? first

B: Trudne. hard.ACC A: ‘Which question do you want to answer first?’ B: ‘A hard one.’ () Jeśli Eliza przyprowadzi jednego psa, if Eliza will bring one dog.ACC ‘If Eliza brings one dog, I’ll bring two.’

ja I

przyprowadzę will bring

dwa. two.ACC

NPE in languages showing inflectional morphology on adjectives and determiners is associated with deletion of the head noun, while NPE in languages without such morphology is associated with pro-forms, e.g., the English pronoun one (cf. Kester b; Lobeck , ). NPE is used with a wider range of licensers in Polish than in English (see ()) for reasons that remain unclear, but see Bošković (a) for some discussion of NPE in Serbo-Croatian and English.

. S

.................................................................................................................................. This chapter has provided an overview of Polish elliptical constructions and their main characteristics. Some of these constructions are better known than others and hence have been discussed in greater detail, but there are also constructions I haven’t discussed for reasons of space (e.g., null complement anaphora, Hankamer and Sag ; or antecedent-contained sluicing, Yoshida ). I wish to stress the role played by empirical results achieved through, for example, psycholinguistic experimentation or corpus studies, in linguistic theorizing. Empirical methods are both a tool for testing the accuracy of theoretical claims and a way to advance our understanding of phenomena that are more complex than they appear to be. My hope is that section ., in particular, has made these points clear.





What emerges from this chapter is that Polish elliptical constructions behave in ways that are largely consistent with existing theories of ellipsis. However, Polish, along with several other languages, poses a challenge for those versions of the PF-deletion approach that assume that ellipsis can’t repair preposition-stranding violations. As discussed in section ., independent processing principles are likely candidates for motivating omission of prepositions in any language, a scenario that can be easily implemented on the direct interpretation approach.

A This chapter has benefited from the helpful comments of two anonymous referees. I am also grateful for comments and feedback from the editors Tanja Temmerman and Jeroen van Craenenbroeck. All remaining errors are mine.

  ......................................................................................................................

 ......................................................................................................................

     

. I

.................................................................................................................................. R employs a wide range of elliptical devices in both its written and spoken registers. In this chapter, we review the major sorts of elliptical constructions that Russian displays, with particular emphasis on those that present typologically challenging puzzles for the analysis of ellipsis cross-linguistically. As we will see, however, when properly analyzed, there are not so many such puzzles—Russian displays the usual kinds of phrasal ellipsis (predicate ellipsis, clausal ellipsis, and argument ellipsis), with the various additional devices that allow gapping and similar constructions to appear, and has a fairly typical set of restrictions on the licensing of elliptical constructions. We review these constructions in the chapter, emphasizing various Russian-specific properties, before turning at the end to what are possibly the most controversial puzzles in Russian (apparent) ellipsis constructions, namely the status of the missing material in verb-stranding constructions of the kind that languages like English do not allow at all. The chapter is organized as follows. In section . we discuss NP-ellipsis in Russian. In section ., we present non-controversial cases of clausal ellipsis (sluicing, sprouting, polarity ellipsis), followed by a discussion of predicate/vP-ellipsis, including gapping in section .. We then turn to the controversial verb-stranding constructions in Russian in section ., and in section . briefly discuss comparative deletion, Right-Node Raising, and fragment answers before ending with a conclusion.

. N-  R

.................................................................................................................................. In this section we present basic facts about NP-ellipsis (NPE) phenomena in Russian.1 Following Pereltsvaig (, ) and Bailyn (), we assume the functional structure of 1

In this survey of NPE, we do not consider various theoretical possibilities, such as the pronominalization strategy discussed in Corver and van Koppen (), the potential absence of an NP complement (Werner ), and licensing by focus (Corver and van Koppen ) or by classifiers (Alexiadou and Gengel ) which, to our knowledge, haven’t been examined with respect to Russian.





nominal phrases such as () in Russian is something like that shown in (), where quantifiers and demonstratives occupy specDP position, possessive pronouns are in D, and adjectives and PP modifiers are adjuncts to NP, and the adnominal genitive is generated as the complement of the noun. () Oni vyvesili každuju moju krasivuju fotografiju Moskvy v they put up every my beautiful photo of Moscow in širokoj ramke. wide frame ‘They put up every [one of] my beautiful photo[s] of Moscow in a wide frame.’ ()

Functional structure of Russian DP:2 DP

kazˇduju D' NP

D moju AP

NP

krasivuju NP

PP N'

N

P DP

DP

v ˇsirokoj ramke

fotografiju Moskvy In contrast to English, Russian lacks articles, and it also doesn’t have a nominal proform like English ‘one’, but there is extensive evidence in favor of the DP projection (Pereltsvaig , ). Russian has rich agreement morphology which appears on demonstratives, quantifiers, possessive pronouns, and adjectives. All NP-elements that show agreement (gender, number, case) can serve as remnants in NPE constructions in Russian.3 Thus, agreeing demonstratives (), possessives (), and adjectives () can license NP-ellipsis:

2 We also assume that numerals in Russian are not head nouns taking nominal complements (see Babby ; Bailyn ; Pereltsvaig ), but rather head an intermediate projection NumP between NP and DP (Bailyn ), which is absent in nominal phrases without numerals. Numerals in Russian have peculiar morphosyntactic behavior (see Babby ; Bailyn ; Pereltsvaig , among others). We do not discuss the interaction of numeral phrases and ellipsis here. 3 It’s not obvious whether quantifiers, which also show agreement with the noun in Russian, can be remnants of NPE, because it is difficult to find appropriate antecedents and pragmatically plausible contexts for them. But the following example suggests that they, just like demonstratives, are generally possible NPE remnants:

(i) Posle togo, kak Vasja kupil každuju knigu [ètogo avtora], after that how Vasja bought every book [this author]Gen on možet podarit’ tebe ljubuju [NP knigu [ètogo avtora]. he can give you anyFemSg bookFemSg [this author]Gen ‘After Vasja bought every book of this author, he can give you any [book of this author].’



  

() My kupili tu knigu po matematike posle togo, kak we bought that book about math after that how on kupil ètu [NP knigu po matematike]. he bought thisFemSg bookFemSg about math ‘We bought that book about math after he bought this [one].’ () Saša pročital moju stat’ju, a Dima pročital tvoju [NP statju]. Sasha read my article and Dima read yourFemSg articleFemSg ‘Sasha read my article and Dima read your[s].’ () Petja uvidel belogo zajca posle togo, kak Vasja uvidel černogo [NP zajca]. Petja saw white hare after that how Vasja saw blackMascSg hareMascSg ‘Petja saw a white hare after Vasja saw a black [one].’ Finally, bare nouns (Ns) in Russian can never be elided irrespective of whether other NP-material is retained (for example, possessive pronouns (), adjectives ()): if the complement of the noun (adnominal genitive) is present, the noun cannot be omitted (cf. (a) and (b), (a) and (b)): () a. *Vasja uvidel moju fotografiju [novogo prezidenta] posle togo, kak Vasja saw my photo [new president]Gen after that how Petja kupil tvoju fotografiju [starogo prezidenta]. Petja bought your photo [former president]Gen Intended reading: ‘Vasja saw my photo of the new president after Petja bought your photo of the former president.’ b. Vasja uvidel moju [fotografiju [novogo prezidenta]] posle togo, kak Vasja saw my [photo [new president]]Gen after that how Petja kupil tvoju [fotografiju [novogo prezidenta]]. [new president]]Gen Petja bought your [photo ‘Vasja saw my photo of the new president after Petja bought your[s] photo of the new president.’ () a. *Odna novost’ vyzvala iskrennee udivlenie studentov, a one news evoked sincere astonishment studentGenPl and drugaja novost’ vyzvala pritvornoe udivlenie professorov. other news evoked feigned astonishment professorGenPL Intended reading: ‘One piece of news evoked students’ sincere astonishment, and the other piece of news evoked the professors’ feigned astonishment.’ b. Odna novost’ vyzvala iskrennee [udivlenie studentov], a one news evoked sincere [astonishment studentGenPl] and drugaja novost’ vyzvala pritvornoe [udivlenie studentov]. other news evoked feigned [astonishment studentGenPl] ‘One piece of news evoked students’ sincere astonishment, and the other piece of news evoked the students’ feigned astonishment.’ To sum up, various NP-elements can become remnants in NPE constructions in Russian: elements in specDP (demonstratives), in D (possessive pronouns), in Num (numerals),





and modifiers to NP (adjectives). Whether adjunct PPs can serve as NPE remnants is not clear; perhaps, such constructions should be attributed to gapping.4 Bare nouns in Russian cannot be elided. We assume that agreement licensing is a promising explanation for Russian NPE (see n.  for a list of possible analyses), but further investigation is required for a full analysis.

. C   R

..................................................................................................................................

.. Sluicing Russian demonstrates sluicing constructions of various kinds. Any argument or adjunct can become a sluiced remnant: Subjects: () Kto-to s”el vse pečen’ e, no ja ne znaju kto [ ]. somebody ate all biscuits but I not know who [ ] ‘Somebody ate all the biscuits, but I don’t know who ate all the biscuits.’ Direct Objects: () Marina uvidela čto-to neobyčnoe, no ja ne znaju čto Marina saw something strange but I not know what ‘Marina saw something strange, but I don’t know what Marina saw.’

[ [

]. ]

4 Adjunct PPs can be retained as remnants only in clauses with coordination (ii), which is not required in sentences with agreeing NP remnants. PP remnants (unlike, for example, adjectives (iii), ()) cannot appear within subordinate clauses (i):

(i) *Nadja kupila knigi v tverdyx obložkax posle togo, kak Nadja bought book in hard covers after that how my kupili knigi v mjagkix obložkax. we bought books in paper covers Intended reading: ‘Nadja bought books in hardback after we bought [books] in paperback.’ (ii)

Nadja kupila knigi v tverdyx obložkax, a Nadja bought book in hard covers CONJ my kupili knigi v mjagkix obložkax. we bought books in paper covers ‘Nadja bought books in hardback, and we bought [books] in paperback.’

(iii) Nadja kupila starye knigi posle togo, kak Nadja bought old books after that how my kupili novye knigi. we bought newPl bookPl ‘Nadja bought old books after we bought new [ones].’ The fact that adjunct PP remnants are possible only in constructions with coordination might indicate that sentences like (ii) with stranded PP adjuncts are actually instances of gapping, not NPE. In addition, PP adjuncts in the construction at hand are always contrastive foci, just like the remnants in gapped clauses (Kazenin ). All other NP remnants are possible in sentences with subordination, even numerals that don’t agree with the noun.



  

Adjuncts: () Mitja gde-to budet prazdnovat’ den’ roždenija, no ja ne znaju gde [ ]. Mitja somewhere will celebrate day of birth but I not know where [ ] ‘Mitja will celebrate his birthday somewhere, but I don’t know where Mitja will celebrate his birthday.’ Sluicing in Russian can occur both in subordinate contexts (as in examples ()–()) and in matrix contexts: Direct Question—Argument: () A: Kto-to s’’el vse pečen’je. somebody ate all biscuits ‘Somebody ate all the biscuits.’

B: Kto [ ]? who [ ] ‘Who ate all the biscuits?’

Direct Question—Adjunct: () A: Vova kogda-to pridet. Vova sometime will come ‘Vova will come sometime.’

B: Kogda [ ]? when [ ] ‘When will Vova come?’

The sluiced material can either precede () or follow () the antecedent: ()

Marina uvidela čto-to, no ja ne znaju čto Marina saw something but I not know what ‘Marina saw something, but I don’t know what Marina saw.’

()

Ja ne znaju čto [ ], no čto-to Marina uvidela. I not know what [ ] but something Marina saw ‘I don’t know what Marina saw, but Marina saw something.’

[ [

]. ]

In other words, sluicing in Russian doesn’t obey the Backward Anaphora Constraint (Langacker ), unlike some other types of Russian predicate ellipsis (see section .. for further discussion). Russian is a language that allows multiple wh-fronting, so it allows multiple sluicing () and multiple coordinated sluicing () as well: Multiple Sluicing: () Maša komu-to čto-to podarila, no ja ne pomnju komu čto [ ]. Masha to.somebody something gave but I not remember to.whom what [ ] ‘Masha gave something to somebody, but I don’t remember what Masha gave to whom.’ Multiple Coordinated Sluicing: () Kto-to čto-to pel, no ja ne rasslyšal kto i čto [ ]. somebody something sang but I not get who and what [ ] ‘Somebody sang something, but I didn’t get who [it was] and what [they] sang.’ In multiple sluicing () we see two wh-remnants without conjunction, while multiple coordinated sluicing has two wh-remnants which are linked by i ‘and’ conjunction.





Theoretical analyses of Russian sluicing (Grebenyova , c, ; Scott ) all presuppose that in sluiced constructions wh-fronting takes place to the same positions as in wh-questions, followed by TP-ellipsis. However, since there are different approaches to whquestions in Russian, approaches to sluicing constructions differ accordingly. There are two crucial aspects of wh-behavior that have been the source of disagreement. First, there is the issue of the nature of wh-fronting: whether Russian displays true wh-movement or is a wh-insitu language which fronts wh-phrases for a different reason than English (and other whmovement languages) and to another place in the structure. Second, there is a question of whether Russian exhibits superiority effects in structures with multiple wh-phrases, and if it does, what syntactic or semantic principles underlie this phenomenon. While superiority effects are crucial for the analyses of both Grebenyova () and Scott (), they disagree on such basic empirical facts as whether there are superiority effects in embedded wh-questions and matrix sluiced constructions. Due to this discrepancy, their analyses differ significantly. Grebenyova (c, ), following Stjepanović () and Stepanov (), analyzes Russian as a wh-in-situ language, attributing wh-fronting to contrastive focalization. Stepanov () and Grebenyova (c, ) claim that wh-phrases in Russian are fronted to a focus position below CP, the specifier of a Foc projection. Grebenyova (c) argues that it is the strong [+focus] feature on Foco that licenses TP-elision5 in Russian sluiced constructions () and that the focus-licensing is the right analysis for sluicing in general (e.g., for English as well). ()

Ivan kupil čto-to, no ja ne pomnju [FocP čto [TP Ivan kupil]]. Ivan bought something but I not remember what ‘Ivan bought something, but I don’t remember what.’ (p. )

This approach has several weaknesses that are worth mentioning. First, it presupposes the existence of a Foc projection below CP in Russian, which was extensively argued against in Bailyn (), Kučerová and Neeleman (), and Titov ().6 Second, it complicates the analysis of sluicing for languages like English, because it introduces at least two stipulations: (i) that there is a weak [+ focus] feature on Co in English which does not attract wh-phrases to its specifier, (ii) that wh-movement in English (still driven by a [+wh] feature on Co) simply happens to create the necessary configuration for TP-ellipsis to take place (when the specifier of C with a [+focus] feature is filled, the TP is deleted). What is more, as we will see, this approach cannot easily account for the superiority effects observed in Russian, which, contra Grebenyova (c, ), appear not only in sluiced constructions, but in non-elliptical wh-questions as well. According to Scott (), wh-fronting in Russian is a genuine case of wh-movement which, just like in English, relocates wh-phrases from their base positions to the specifier(s) of CP. In 5 Note that Russian doesn’t exhibit TP-deletion with non-interrogative focused remnants in embedded clauses:

(i) *Petja uvidel kogo-to, i Vasja skazal, čto MAŠU. Petja saw someone and Vasja said that Masha Intended: ‘Petja saw someone, and Vasja said that MASHAi [Petja saw ti].’ This is unexpected under Grebenyova’s (c, ) approach. 6 Bailyn (), Kučerová and Neeleman (), and Titov () have claimed that there is no Focus projection in the left periphery in Russian. Bailyn () argues that Russian doesn’t have a position for focused elements at all, while Kučerová and Neeleman () and Titov () claim that all focused constituents in Russian share an underlying clause-final position.



  

addition, Russian exhibits multiple wh-movement and, as Scott () claims, it has superiority effects. In other words, wh-behavior in Russian is considered to be similar to that of Bulgarian. The only thing that distinguishes Russian from Bulgarian on this account is the presence of one more functional projection in the matrix clauses of the former—the higher operator phrase (HOP). The existence of this projection in matrix clauses and its absence in embedded clauses explains the following paradigm (adapted from Scott ): ()

Gde ty sejčas rabotaeš’? where you now work ‘Where do you work now?’ (p. )

()

Ty gde sejčas rabotaeš’? you where now work ‘Where do you work now?’ (p. )

() Ja ne uveren, gde ty sejčas rabotaeš’. I not sure where you now work ‘I’m not sure where you work now.’ (p. ) ()

*Ja ne uveren, ty gde sejčas rabotaeš’. I not sure you where now work Intended reading: ‘I’m not sure where you work now.’ (p. )

In matrix clauses, where the HOP is present, there is one additional position at the left periphery (specHOP); and, as we can see from (), a non-wh-phrase (ty ‘you’) can move there. In embedded questions, however, there is no such position () and nothing can precede the wh-phrase(s). In sentences with sluicing wh-fronting works the same way: the wh-remnant moves to specCP, and then the TP is elided (Scott : ). Russian sluicing is thus identical to sluicing in English. () a. Deti s kem-to podralis’, no mne [VPne važno [CP s kem [TP ]]]. kids with smbd. fought but me not important with who [ ] ‘The kids fought with somebody but I don’t care with who the kids fought.’ (p. ) b. Sluiced construction with a single wh-remnant (adapted from Scott : )

VP ne vazˇno

CP

s kem [+wh]

C' TP e

The presence of the special functional projection (HOP) is justified by a range of phenomena (Scott : –, –, –); below we will show how it can neatly derive the multiplewh-behavior and predict the contexts where superiority effects arise in Russian.





Multiple wh-fronting and the (non)existence of superiority effects in different contexts constitute the essential data for both Grebenyova’s and Scott’s analysis. According to Grebenyova (c), in Russian there are no superiority effects in wh-questions (both matrix ()–() and embedded ()–()), but there are superiority effects in sluicing (both matrix () and embedded ()–()) (examples from Grebenyova c): () Kto1 kogo2 [ t1 who whom ‘Who loves whom?’

ljubit loves

t2]?

() Kogo2 kto1 [ t1 ljubit whom who loves ‘Who loves whom?’

t2]?

() Ja ne znaju [kto I not know who ‘I don’t know who loves who.’

kogo whom

() Ja ne znaju [kogo kto I not know whom who ‘I don’t know who loves who.’

ljubit] loves

ljubit] loves

() A: Každyj priglasil kogo-to na tanec. everyone invited someone to dance ‘Everyone invited someone to dance.’ B: Kto who

kogo? whom

/ /

*Kogo whom

kto? who

() Každyj priglasil kogo-to na tanec, no ja ne pomnju kto kogo [ ]. everyone invited someone to dance but I not remember who whom [ ] ‘Everybody invited someone to a dance, but I don’t remember who invited whom.’ () *Každyj priglasil kogo-to na tanec, no ja ne pomnju kogo kto [ ]. everyone invited someone to dance but I not remember whom who [ ] Grebenyova (c), who considers superiority a diagnostic of movement, concludes from these data that there is no wh-movement in Russian and that something else should be responsible for the peculiar restrictions on the word order of wh-remnants in sluiced constructions ()–(). She proposes that it is semantic parallelism in the sense of Fiengo and May () and that it requires the surface order of multiple sluices to be parallel to the order of quantifiers in the antecedent. On the other hand, according to Scott (), superiority effects are present in both wh-questions and sluiced constructions, but due to the existence of the HOP projection we do not observe them in matrix clauses like ()–(), whereas they emerge in matrix clauses where the movement of a non-wh-phrase to specHOP has taken place ()–() and in embedded clauses ()–() (examples adapted from Scott ):



  

() Kto čto posovetoval Darii? whoNOM whatACC advised DariaDAT ‘Who advised what to Daria?’ (p. ) ()

Čto kto posovetoval Darii? whatACC whoNOM advised DariaDAT ‘Who advised what to Daria?’ (p. )

() Darii kto čto posovetoval? DariaDAT whoNOM whatACC advised ‘Who advised what to Daria?’ (p. ) ()

*Darii čto kto posovetoval? DariaDAT whatACC whoNOM advised ‘Who advised what to Daria?’ (p. )

() Maria sprosila, kto čto posovetoval Maria asked whoNOM whatACC advised ‘Maria asked who advised what to Daria.’ (p. ) ()

*Maria sprosila, čto kto posovetoval Maria asked whatACC whoNOM advised ‘Maria asked who advised what to Daria.’ (p. )

Darii. DariaDAT

Darii. DariaDAT

() ?*Komu-to kto-to zvonit s utra, no ja ne somebodyDAT someoneNOM calls since morning but I not znaju, komu kto. know whoDAT whoNOM Intended reading: ‘Somebody has been calling someone since the morning, but I don’t know who [is calling] whom.’ (p. ) To sum up, the two leading approaches to multiple sluicing in Russian (Grebenyova c and Scott ) are based on conflicting sets of data. The data presented in Scott () seem more extensive and reliable: it was collected from seventy-six participants. But even if there are no superiority effects in embedded wh-questions and there are superiority effects in matrix sluicing constructions, as Grebenyova (c, ) claims, her analysis still has the problem of explaining the data in ()–(). If the principle of semantic parallelism is responsible for the superiority effects in the sluiced sentences, then the example in () should be grammatical, because it exhibits the required parallelism in variable binding. But as we see, it is not the case: the word order ‘wh.DAT > wh.NOM’ is ungrammatical in sluiced clauses irrespective of the order of the quantifiers in the antecedent (). And if the wh-fronting in Russian is movement to the specifier of a focus projection (or, alternatively, if it is a TP-adjunction (Bošković , ; Gribanova c), then the emergence of superiority effects in ()–() is very unexpected: why would scrambling of a non-wh-phrase to the left periphery evoke superiority effects in a matrix wh-question, where no semantic parallelism can take place? In conclusion, it seems that analyzing wh-fronting as wh-movement and a purely syntactic approach to sluicing give better coverage of the observed wh-behavior.





Multiple coordinated wh-sluicing (as well as multiple coordinated wh-questions) is possible with wh-adjuncts and wh-arguments as well as with a wh-argument plus whadjunct combination: () Kto-to gde-to poet, no ja ne znaju, kto i gde. someone somewhere sings but I not know who and where ‘Someone is singing somewhere, but I don’t know who [is singing] and where.’ () Kto i gde budet spat’? who and where will sleep ‘Who will sleep and where?’ (p. )

(adapted from Scott )

In sluicing constructions with multiple coordinated wh-arguments, superiority effects emerge: () Kogo-to kto-to parodiruet, no ja ne znaju, someone.ACC somebody.NOM parodies but I not know kto i kogo. who.NOM and who.ACC ‘Somebody is parodying someone, but I don’t know who [is parodying] whom.’ () *Kogo-to kto-to parodiruet, no ja ne znaju, someone.ACC somebody.NOM parodies but I not know kogo i kto. who.ACC and who.NOM ‘Somebody is parodying someone, but I don’t know whom [is parodying] who.’ ()

Kto i kogo obmanyvaet? who.NOM and who.ACC cheats ‘Who is cheating whom?’ (p. )

(adapted from Scott )

() *Kogo i kto obmanyvaet? who.ACC and who.NOM cheats ‘Who is cheating whom?’ (p. ) Note that the facts in ()–() cannot be explained by semantic parallelism, because the grammatical sentence in () violates the proposed semantic constraint, and the ungrammatical sentence in () obeys it. As we see from the examples above, wh-fronting in multiple coordinated sluicing constructions ((), ()–()) behaves the same way as wh-fronting in multiple coordinated wh-questions ((), ()–()), suggesting that the analyses of the two should be the same. In Kazenin  and Gribanova c a uniform monoclausal approach to coordinated multiple-wh-constructions in Russian is argued for, according to which there is only one CP in the structure and the coordination happens between two wh-phrases: ()

[CP [&P wh and wh] [TP t1 . . . t2]]



  

Scott (), on the other hand, claims that a uniform approach to multiple coordinated wh-fronting is on the wrong track, and proposes a separate analysis for multiple coordinated wh-adjuncts, i.e., backwards sluicing (see, for example, Giannakidou and Merchant ; Camacho ), while maintaining the monoclausal approach for multiple coordinated wh-arguments and the “mixed type” (coordination of wh-arguments and whadjuncts). However, none of the theoretical analyses give a convincing explanation of why multiple wh-fronting is impossible when the two wh-phrases are adjuncts,7 which remains an issue for further investigation.8 7

Multiple wh-fronting in Russian in both questions (i–ii) and sluiced constructions (iii–iv) is ungrammatical when both wh-phrases are adjuncts (for a similar phenomenon in Croatian see Citko and Gračanin-Yuksek ): (i)

*Kogda gde vy vystupaete? when where you perform ‘When do you perform where?’ (p. )

(ii) *?Gde kogda vy vystupaete? Where when you perform ‘Where do you perform when?’ (p. )

(adapted from Scott )

(adapted from Scott )

(iii) *My gde-to kogda-to budem sdavat’ èkzamen po matematike, we somewhere sometime will have exam in math no ja ne znaju, gde kogda. but I not know where when ‘We will have an exam in math somewhere sometime, but I don’t know where when.’ (iv) *My gde-to kogda-to budem sdavat’ èkzamen po matematike, we somewhere sometime will have exam in math no ja ne znaju, kogda gde. but I not know when where ‘We will have an exam in math somewhere sometime, but I don’t know when where.’ Multiple wh-adjuncts can be fronted only if they are coordinated (v–viii). (v)

Kogda i gde vy vystupaete? when and where you perform ‘When do you perform and where?’ (p. )

(adapted from Scott )

(vi)

Gde i kogda vy vystupaete? where and when you perform ‘Where do you perform and when?’ (p. )

(adapted from Scott )

(vii) My gde-to kogda-to budem sdavat’ èkzamen po matematike, We somewhere sometime will have exam in math no ja ne znaju, gde i kogda. but I not know where and when ‘We will have an exam in math somewhere sometime, but I don’t know where and when.’ (viii) ?My gde-to kogda-to budem sdavat’ èkzamen po matematike, we somewhere sometime will have exam in math no ja ne znaju, kogda i gde. but I not know when and where ‘We will have an exam in math somewhere sometime, but I don’t know when and where.’ We leave the explanation for this restriction open. 8 See the suggestion in Scott (: –), where this restriction is explained by the lack of proper government of adjunct traces.





.. Sprouting Sprouting is a widespread phenomenon in Russian: it is a subtype of sluicing in which the sluiced wh-phrase has no overt correlate in the antecedent clause (van Craenenbroeck and Merchant ; Chung et al. , among others): () Marina čto-to pela, no ja ne znaju čto [ Marina something sang but I not know what [ ‘Marina sang something, but I don’t know what Marina sang.’ () Marina pela, no ja ne znaju čto [ ]. Marina sang but I not know what [ ] ‘Marina sang, but I don’t know what Marina sang.’

]. ]

(Sluicing)

(Sprouting)

The sentence in () differs from () in that the antecedent for the wh-phrase (‘something’) is absent. The former is sluicing, the latter sprouting. Sprouting is mostly widespread with adjunct remnants ((), ()) but is also possible with some argument remnants: sprouted remnants can be direct objects () and indirect objects (): () [Ja ne znaju kogda [ ] ], no Vova pridet. I not know when [ ] but Vova will.come ‘I don’t know when Vova will come, but Vova will come.’

(adjunct sprouting)

() Mitja budet otmečat’ den’ roždenija, [no ja ne znaju gde [ ]]. Mitja will celebrate day of birth but I not know where [ ] ‘Mitja will celebrate his birthday, but I don’t know where Mitja will celebrate his birthday.’ () Marina napisala pis’mo, [no ja ne znaju komu [ ]]. (IO sprouting) Marina wrote letter but I not know whom [ ] ‘Marina wrote a letter, but I don’t know whom Marina wrote a letter.’ In Russian, sprouting can never have a subject as a wh-remnant, possibly because subject drop is generally ungrammatical in matrix clauses (): compare the sluiced subject in () with the ungrammatical sprouted one in (): () Tam kto-to smeetsja, no ja ne znaju kto [ ]. (subject sluicing) there somebody is.laughing but I not know who [ ] ‘Somebody is laughing there, but I don’t know who is laughing.’ () *Tam there ()

smeetsja. is.laughing

*Tam smeetsja, no ja ne znaju kto [ ]. there is.laughing but I not know who [ ] *‘There is laughing, but I don’t know who is / are laughing.’

(*subject sprouting)



  

Sprouting shares many properties with sluicing: it can also occur in both direct and indirect questions, the clause with the sprouted material can either precede () or follow () the antecedent clause, multiple () and multiple coordinated () sprouting are also present: () Marina otpravila, [no ja ne znaju čto komu9 [ ]]. (multiple sprouting) Marina sent but I not know what whom [ ] ‘Marina sent [something] [to somebody], but I don’t know whom Marina sent what.’ ()

My budem sdavat’ ekzamen po matematike, [no ja ne znaju, we will have exam in math but I not know gde i kogda [ ]]. (multiple coordinated sprouting) where and when [ ] ‘We will have an exam in math, but I don’t know where and when we will have an exam in math.’

To sum up, there are no significant differences in Russian sluicing and sprouting vs their English equivalents with the obvious exception of the availability of multiple wh-movement in Russian and the consequent availability of multiple sluicing and multiple sprouting.

.. Polarity ellipsis in Russian Kazenin () labels Russian constructions with the polarity markers da (positive polarity marker) and net (negative polarity marker)—da/net constructions. We will call such constructions “polarity ellipsis.” Examples are given in () and (B): () Petja v Moskvu priexal, a Vasja v Peterburg net [ ]. (polarity ellipsis) Petja in Moscow arrived but Vasja in St Petersburg no [ ] ‘Petja arrived in Moscow, but Vasja did not [arrive] in St Petersburg.’ ()

A: Ty pogovoril i s Vasej, i s Petej? you talked CONJ with Vasja CONJ with Petja ‘Have you talked both to Vasja and to Petja?’ B: S Vasej da, a s Petej net [ ]. (polarity ellipsis) with Vasja yes but with Petja no ‘I’ve talked to Vasja, but I have not talked to Petja.’ (p. )

The remnants can originate both in the matrix clause as in () and () and in embedded infinitival clauses as in ():

9

Note that this sentence with two wh-phrases can have only a Pair-List (PL) reading, so it is possible only under the following scenario: Marina has sent several letters to different people, the speaker knows that Marina has sent these letters, but he doesn’t know what letter has Marina sent to what person. The Single-Pair (SP) reading is not available: this sentence can’t be uttered if there was only one letter and one addressee.





() V Moskvu ja xoču exat’, a v Peterburg net [ ]. to Moscow I want to.go but to St Petersburg no [ ] ‘I want to go to Moscow, but to St Petersburg I do not want to go.’ (p. ) Remnants cannot originate in adjunct clauses () or relative clauses (), which are barriers for syntactic movement in Russian: () *Kogda Petja prišel, ja obradovalsja, a Vasja net [ ].10 when Petja came I was.glad but Vasja no ‘I was glad when Petja came, but I was not glad when Vasja came.’ (p. ) () *Ja videl mašinu, kotoruju kupil Petja, a Vasja net [ ].11 I saw car which bought Petja but Vasja no ‘I saw the car which Petja bought but I did not see the car which Vasja bought.’ These restrictions can be easily explained if we assume that remnants undergo movement from the TP they are generated in before the elision. Kazenin () argues that da/net constructions are cases of ellipsis. First of all, they are not restricted to coordinate constructions: () Do Peti moe pis’mo došlo, poètomu stranno, čto do Koli ešče net [ ]. to Petja my letter reached therefore strange that to Kolja still no [ ] ‘My letter has reached Petja, therefore it is strange that it still has not [reached] Kolja.’ (p. ) Second, Kazenin observers that like some instances of English VP-ellipsis ()–(), Russian polarity ellipsis ()–() obeys the Backward Anaphora Constraint (initially proposed by Langacker  for pronouns): backward ellipsis is typically impossible in coordinate structures and in matrix clauses if the antecedent is in the subordinate clause: ()

*Sue didn’t [e] but John ate meat.(p. )

()

*John didn’t [e] because Sue ate meat. (p. )

() *Petja net [ ], a Kolja poedet v Peterburg. Petja no [ ] but Kolja will.go to Petersburg lit. ‘Petja will not go to Petersburg, but Kolja will go to Petersburg.’ (p. ) () *Do Peti moe pis’mo da [ ], poètomu stranno, čto do Koli ešče ne došlo. to Petja my letter yes [ ] therefore strange that to Kolja still not reached lit. ‘My letter [has reached] Petja, therefore it is strange that it still has not reached Kolja.’ (p. ) 10

This sentence is grammatical under the reading where Vasja is the subject of the matrix clause: ‘I was glad when Petja came, but Vasja was not glad when Petja came’. 11 This sentence is grammatical under the reading where Vasja is the subject of the matrix clause: ‘I saw the car which Petja bought, but Vasja did not see the car which Petja bought.’



  

Based on this similarity to English VP-ellipsis, Kazenin argues that Russian da/net constructions also involve ellipsis. However, note that sluicing constructions in Russian, which were claimed to involve TP-ellipsis (Scott ; Grebenyova c), do not obey the BAC and can operate backwards in coordinate structures (as was shown in section ..), which may indicate that not all types of ellipsis obey the BAC. Finally, there is another parallelism between English VP-ellipsis and Russian polarity constructions: both can have a non-linguistic antecedent ()–(), in contrast to constructions with gapping (): () [context: the waiter takes out dessert menus] Ja da [e], a on net [e]. I yes but he no lit. ‘I will [have dessert], but he won’t [have dessert].’ ()

I will [e], if you do [e].

()

*I [e] apples, and you [e] bananas.

Sentences () and () are possible when the inferred event is recoverable from the situation, but the sentence in () is ungrammatical even if the situation allows the unambiguous recovery of the verb as ‘eat’. From these facts Kazenin draws the conclusion that da/net constructions involve ellipsis. Kazenin () argues that the polarity items are always focused in da/net constructions and that the remnant phrases (s Vasej and s Petej in (B)) are always contrastive topics, which he claims undergo obligatory syntactic movement in Russian.12 He argues that the polarity items in this construction are heads of their own functional projection (∑P)13 and that they license the ellipsis of their complement TP14 (p. ): ()

Polarity ellipsis construction in Russian ΣP

XP

Σ'

YP

ΣP Σ

12

TP

It is controversial whether contrastively topicalized elements must always front in Russian, as Kazenin assumes they do. See Bailyn () and Kučerová and Neeleman () for relevant discussion. 13 Note that Kazenin’s () view is at odds with the view expressed in Merchant (c): Merchant argues that words like Russian net ‘no’ are phrasal negative adverbs and not heads. For the arguments in favor of viewing polarity items net and da as heads, see Kazenin (: –). 14 Kazenin assumes split-Infl for Russian, so what elides in sentences like (B) is actually AgrsP—a projection which is above TP and which hosts the subject in its specifier.





Kazenin suggests that all remnants (contrastive topics) are adjoined to ∑P and that ∑ licenses TP-ellipsis. Polarity ellipsis is common across languages, and involves TP-ellipsis with a higher licensing head. Nothing about the construction in Russian is particularly unusual, except that there are some distinctions between the possibilities for positive and negative polarity ellipsis.

.. Stripping ... Types of stripping in Russian Stripping is an elliptical construction with just one remnant left behind; standard cases of stripping are presented in ()–(): () Ja dolžen Miše pozvonit’ ili ty [ ]? I should Misha call or you [ ] ‘Should I call Misha or should you call Misha?’ () Ona poprosila Mišu ostat’sja i Vasju tože [ she asked Misha to.stay and Vasja too [ ‘She asked Misha to stay and she asked Vasja to stay too.’

]. ]

Like many other languages, Russian also has not-stripping: () Ja okazalsja prav, ne Mitja [ ]. I turned.out right not Mitja [ ] ‘I turned out to be right, not Mitja turned out to be right.’

(not-stripping)

()

(not-stripping)

Ona menja pozvala guljat’, ne she me invited go.for.a.walk not ‘She invited me to go for a walk, not you.’

tebja you

[ [

]. ]

The elided material can be discontinuous in Russian stripping constructions, as is shown by (): () Storonniki followers protivniki opponents

ètoj thisGEN [ [

] ]

teorii theoryGEN

tože too

[ [

často often

ssylajutsja cite

na on

ètu this

statju article

i and

]. ]

‘This theory’s followers often cite this article and the opponents of this theory often cite this article too.’ (Testelets ) In () the elided material consists of the complement of the subject NP and of the finite clause excluding the subject, which do not form a constituent. Note also that in a nonelliptical clause tože (‘too’) precedes the tensed verb:



  

() Protivniki ètoj teorii tože často ssylajutsja opponents thisGEN theoryGEN too often cite ‘Opponents of this theory often cite this article too.’

na on

ètu this

stat’ju. article

In stripping constructions both the remnant and its correlate in the first conjunct can be embedded: () [Vanja skazal, čto Vika poedet Vanja said that Vika will.go dumaju, čto Maša tože [ ]]. think that Masha too

v to

Peterburg], St Petersburg

i and

[ja I

‘Vanja said that Vika will go to St Petersburg, and I think that Masha will go to St Petersburg too.’ () [Vanja skazal, čto Vika tebja pozvala guljat’], a [ja dumal, Vanja said that Vika you invited go.for.a.walk but I thought čto menja [ ]]. that me [] ‘Vanja said that Vika invited you to go for a walk, but I thought that Vika invited me to go for a walk.’

... Stripping as clausal ellipsis Merchant (c) argues that the conjunction involved in stripping is clausal conjunction and not DP conjunction. That is to say, in examples like () there is clausal ellipsis and not merely a movement of the string and Ben from a position next to Abby or vice versa. ()

Abby speaks passable Dutch, and Ben, too.

Merchant (c) provides several arguments for why stripping examples like () in English are cases of clausal ellipsis. Some of his arguments apply to Russian stripping () as well. () Olja xorošo govorit po-nemecki, i Katja [ ] tože. Olja well speaks German and Katja [ ] too ‘Olja speaks German well, and Katja speaks German well, too.’ First of all, sloppy identity is possible in Russian stripping () and “the grammatical form of dependent elements in the pronounced clause are consistent only with singular, not plural, antecedents” (Merchant c: ): () Olja uexala iz svojej strany v junom vozraste, i Katja [ ] tože. Olja left from her country in young age and Katja [ ] too ‘Oljai left heri country at a young age, and Katjak left herk country at a young age too.’





() Olja i Katja uexali iz svoix stran v junom vozraste. Olja and Katja went.away(Pl) from their countries in young age ‘Olja and Katja left their (respective) countries at a young age.’ () *Olja uexali iz svoix stran v junom vozraste i Katja [ ] tože. Olja went.away(Pl) from their countries in young age and Katja [ ] too Secondly, group predicates like ‘meet’ can have conjoined DPs as their subject, but these predicates cannot occur in stripping (Merchant c; Depiante ): ()

Olja i Katja vstretilis’ v Olja and Katja met in ‘Olja and Katja met in a shop.’

() *Olja Olja

vstretilis’ met

v in

magazine shop

magazine. shop

i and

Katja Katja

[ [

]. ]

So, it seems at least plausible that stripping in Russian is a case of clausal ellipsis. Beyond these similarities, there are some minor differences between Russian and English stripping. For example, in English negation cannot be included in the stripped material, while in Russian it can:15 ()

*Fred didn’t ask Susan out, and Bill too.

() Kirill ne pozval Ljubu v kino, i Vasja tože [ ]. Kirill not invite Ljuba to cinema and Vasja too [ ] ‘Kirill didn’t invite Ljuba to the cinema and Vasja didn’t invite Ljuba to the cinema either.’ This contrast might be due to the fact that English negative clauses in all (not just elliptical) contexts are incompatible with too, while Russian tože can freely occur in negative clauses: () Vasja tože ne pozval Ljubu v kino. Vasja too not invited Ljuba to cinema ‘Vasja didn’t invite Ljuba to the cinema either.’

15

Note that in Russian negation can be included in gapped material too (which contrasts with English: ‘*Fred didn’t ask Susan out, and Susan Fred’): (i)

Miša ne pozval Dimu v banju, a Dima Mišu. Misha not invite Dima to bathhouse and Dima Misha ‘Misha didn’t invite Dima to the bathhouse and Dima didn’t invite Misha to the bathhouse.’



  

To sum up, there are three main kinds of clausal ellipsis in Russian: sluicing and sprouting constructions, polarity ellipsis with focused particles ‘yes’ and ‘no’, and stripping which often goes with the ‘too’ particle. It may turn out that stripping and polarity ellipsis have a similar or even identical structure: both these forms of TP-ellipsis have a focused particle and contrastivity in them; comparing them is an important area for further research.

. P 

.................................................................................................................................. Kazenin () argues that VP-ellipsis (which we assume is in fact vP-ellipsis) is present in Russian, but it is not easily identified as such. For example, the following sentences are sentences with elided vPs: () On skazal, čto Katja budet čitat’ ètu lekciju, a na samom dele he said that Katja AUX give this lecture but on right affair Sereža budet [ ]. Sereža AUX [ ] ‘He said that Katja will give this lecture, but actually Sereža will give this lecture.’ () a. Andrej sčitaet, čto nikto ne budet delat’ èto domašnee zadanie, Andrej thinks that no one not AUX to.do this home work no ja budu [ ]. but I AUX [ ] ‘Andrej thinks that no one will do this homework, but I will do this homework.’ b. vP-ellipsis construction in Russian

TP DP ja

T' T

vP

budu

e

Kazenin () assumes that the future auxiliary byt’ ‘be’ in Russian is in T in () (contra, for example, Harves ). This view is shared by Bailyn (), who observes that adverbs can both precede and follow the future auxiliary (), while they can only precede the tensed verbs () which do not raise to T in Russian (Bailyn ; Kallestinova ; Gribanova ): () Ivan (objazatel’no) budet (objazatel’no) Ivan certainly will certainly ‘Ivan (certainly) will (certainly) watch TV.’

smotret’ watch

() Ivan (objazatel’no) napišet (???objazatel’no) Ivan certainly will.write certainly ‘Ivan will certainly write a new song.’

televizor. TV

novuju new

pesnju. song





Kazenin () argues that in Russian constructions with vP-ellipsis the remnants can either be presentational or contrastive foci. For example, in () the phrase ‘only Petja’ is interpreted as a presentational focus: () A: Kto budet ezdit’ v Moskvu? who AUX travel to Moscow ‘Who will travel to Moscow?’ B: Tol’ko Petja budet [ ]. only Petja AUX [ ] ‘Only Petja will travel to Moscow.’

(presentationally focused subject) (p. )

It turns out that only subjects and some temporal/locative adjuncts can occur in this construction as presentational foci, whereas objects and PPs cannot: ()

A: Kodga ty budeš’ čitat’ kurs tipologii? when you AUX give course of.typology ‘When will you give the course in typology?’ B: (Ja) v ètom godu (budu) [ ]. (presentationally focused temporal adjunct) I in this year AUX [ ] ‘(I) (will) give the course in typology this year.’ (p. )

() A: Kakoj kurs ty budeš’ čitat’ v ètom which course you AUX give in this ‘What course will you give in this semester?’

semestre? semester

B: Kurs tipologii (*budu) [ ]. (presentationally focused object—ungrammatical) course of.typology AUX [ ] ‘[I will give] the course in typology.’ (p. ) () A: Kuda Vasja bol’še ne budet ezdit’? where Vasja further not AUX go ‘Where will Vasja not travel anymore?’ B: V Moskvu (*ne budet) [ ]. to Moscow not AUX [ ] ‘[He will not travel] to Moscow.’ (p. ) (presentationally focused PP—ungrammatical) According to Kazenin (), phrases don’t need to move anywhere to be interpreted as presentational foci. The fact that the elements that are situated inside VP (he mentions objects and PPs) cannot be present in this construction as presentational foci, while subjects and some temporal/locative adjuncts can,16 suggests that this is a construction with vP-ellipsis: objects and PPs are elided if they are not moved out of vP, hence the ungrammaticality of (B) and (B). 16 Note that under Kazenin’s () analysis, subjects move out of vP in order to receive case, not in order to receive a presentational focus interpretation.



  

The structure of sentences with presentational foci is given in (): as we see, vP-ellipsis doesn’t affect subjects in specTP and adjuncts to T. ()

vP-ellipsis construction with presentational focus (adapted from Kazenin : )17,18 TP

DPsubj

T'

Adv T' T AUX

vP e

Furthermore, it is possible in Russian to extract elements from the vP-ellipsis site, in cases of contrastive focus (), as is known to be possible with English VP-ellipsis as well (): ()

St Petersburg, John [loves t], but Moscow, he doesn’t [love t].

() A: Komu ty budeš’ whom you AUX ‘Who will you help?

pomogat’? help (Indirect Object as contrastive focus)

B: Ja vsem budu pomogat’, daže Pete I everyone will help even Petja ‘I will help everyone, I will [help] even Petja.’ () A: Kakoj jazyk ty budeš’ what language you AUX ‘What language will you study?’

budu AUX

[ [

]. ]

učit’? study (Direct Object as contrastive focus)

B: Anglijskij ja ne budu učit’, tol’ko francuzskij English I not will study only French ‘I won’t study English, I will study only French.’

budu AUX

[ [

]. ]

Sentences (B) and (B) are sentences with contrastively focused remnants (daže Pete, tol’ko francuzskij). Kazenin adopts the minimalist approach of Lopez and Winkler () and assumes that contrastive focus is a syntactic feature which selects for TP. If TP is present in (B) and (B) and what is elided is a vP, then it explains why contrastive focusing is possible in such sentences. The following structure is proposed for the second clause of (B):

17 Kazenin assumes that in () and other cases of vP-ellipsis there is a proform in the place of vP— proPRED. We don’t share his assumption because it gives rise to several problems including case and theta-role assignment for the remnants and their moving out of the ellipsis site. 18 Kazenin () assumes that subjects in Russian actually move not into the specifier of TP but into the specifier of AgrSP—a projection above TP responsible for subject agreement. This adjustment doesn’t affect the argumentation.

 ()



vP-ellipsis construction with contrastive focus (adapted from Kazenin : ) FP

tol’ko francuzskij

TP

(ja) T budu

T' vP e

To sum up, true vP-ellipsis is found in Russian in constructions where the auxiliary is present and remnants are either presentationally focused ((B), (B)) or contrastively focused ((B), (B)) phrases.19 Russian doesn’t seem to differ from English in licensing vP-ellipsis: in both languages T is the licensor in these constructions.

.. Gapping Gapping is another type of predicational ellipsis attested in Russian. Some canonical examples are presented below:

19

There are other sentences in Russian which retain the auxiliary as constructions with vP-ellipsis do (Kazenin ): (i)

A: Kakoj kurs ty budeš’ čitat’ v ètom semestre? which course you AUX give in this semester ‘What course will you give this semester?’ (polarity ellipsis with auxiliary) B: Kurs tipologii budu, a kurs sintaksisa ne budu [ ]. course of.typology AUX but course of.syntax not AUX [ ] ‘I will give a course in typology, but I will not give a course in syntax.’ (p. )

Despite their resemblance with vP-ellipsis constructions, Kazenin () argues that sentences like (i.B) are actually cases of clausal ellipsis and have almost the same structure as sentences with da/net constructions (B). Both constructions have contrastively topicalized remnants (kurs tipologii and kurs sintaksisa in (i.B)) and focused auxiliaries. On Kazenin’s () analysis, the only difference between the two is that in sentences like (i.B) the auxiliary verb (budu) head-adjoins to ∑ (where the polarity marker ne resides). The head-adjunction of the focused auxiliary to ∑ explains why backgrounded remnants (Petja in (ii.B)) are not possible in such sentences: (ii) A: Kakie kursy Petja budet what courses Petja AUX ‘Which courses will Petja give?’

čitat’? give

B: (*Petja) Kurs tipologii (*Petja) budet, a kurs sintaksisa ne budet [ ]. Petja course of.typology Petja AUX but course of.syntax not will [ ] ‘(Petja) will (give) the course of typology, but will not give the course of syntax.’ (p. ) If (ii.B) was an instance of vP-ellipsis, then the ungrammaticality of the backgrounded subject would be unexplained. Under the clausal ellipsis analysis, however, it is predictable: the subject position belongs inside the ellipsis site (TP), so if the subject hasn’t moved to some higher position, we don’t expect to see it.

 ()

   Mitja pil čai, a Nadja [ ] Mitja drank tea CONJ Nadja [ ] ‘Mitja drank tea and Nadja drank coffee.’

kofe. coffee

() Petja perestal vyxodit’ pokurit’ po utram, a Vova Petja stopped going out to.smoke on mornings CONJ Vova [ ] po večeram. [ ] on evenings ‘Petja stopped going out to smoke mornings, and Vova stopped going out to smoke evenings.’ There are a number of similarities and differences between gapping in Russian and in English, some of which will be presented below. First of all, in both languages gapping is restricted to coordinate constructions; subordinate clauses disallow gapping (Kazenin ): () *Some had eaten mussels because others ate shrimp. ()

*Ja poedu v Moskvu, esli Petja [ ] v Peterburg. I will.go to Moscow if Petja [ ] to St Petersburg ‘I will go to Moscow if Petja [goes] to St Petersburg.’

Second, gapping is impossible when the second conjunct is embedded under another predicate (Kazenin ): () *Some had eaten mussels and she claims [that others [

] shrimp].

() *Vasja poedet v Moskvu, i ja dumaju, [čto Petja [ ] v Peterburg]. Vasja will.go to Moscow and I think that Petja [ ] to Petersburg ‘Vasja will go to Moscow, and I think that Petja [will go] to Petersburg.’ Third, when gapping affects multiple verbs, it has to affect the uppermost one (Kazenin ): ()

Pete wants to begin to study French, and Bill wants to begin to study German.

()

*Pete wants to begin to study French, and Bill tries to begin to study German.

()

Petja xočet načat’ učit’ francuzskij, a Vasja [ ] nemeckij. Petja wants to.begin to.study French CONJ Vasja [ ] German ‘Petja wants to begin to study French, and Vasja wants to begin to study German.’

() *Petja xočet načat’ učit’ francuzskij, a Vasja Petja wants to.begin to.study French CONJ Vasja pytaetsja [ ] nemeckij. tries [ ] German ‘Petja wants to begin to study French, and Vasja tries [to begin to study] German.’





Fourth, the gap in both languages can be discontinuous: ()

Will John greet Mary first, or will Mary greet John first?

()

Mitja predložil Kate pojti v kino, a Katja [ ] Mite Mitja suggested to.Katja to.go to cinema CONJ Katja [ ] to.Mitja [ ] v muzej. [ ] to museum ‘Mitja suggested to Katja to go to the cinema, and Katja [suggested] to Mitja [to go] to the museum.’

Finally, gapping in both languages can’t operate backwards:20 () ()

*John [drank] tea, and Mary drank coffee. *Nadja [ ] kofe, a Mitja pil Nadja [ ] coffee CONJ Mitja drank ‘Nadja [drank] coffee, and Mitja drank tea.’

čaj. tea

However, there are also a few differences between gapping in the two languages. For example, in English gapping can’t cut into a major constituent—the gap can’t consist of a verb and a part of a noun phrase, while in Russian it can: ()

*I read the story about elves, and you [read the story] about dwarves.

()

Ja čital istoriju pro èl’fov, a ty [ ] pro I read story about elves CONJ you [ ] about ‘I read a story about elves, and you [read a story] about dwarves.’

gnomov. dwarves

Another difference is that, unlike in English (Siegel ; Johnson ), negation on the antecedent predicate cannot take wide scope over the coordinate structure in Russian (adapted from Kazenin ): ()

()

Mrs Smith can’t dance and Mr Smith [ ] sing. A: *‘Mrs Smith can’t dance and Mr Smith can’t dance.’ B: ‘It can’t be the case that Mrs Smith dances and Mr Smith sings.’

(p. )

Petja ne možet polučat’ $, a ego zamestitel’ [ ] $. Petja not can earn $, and his deputy $. A: ‘Petja cannot earn $, and his deputy cannot earn $.’ B: *‘It can’t be the case that Petja earns $, and his deputy earns $.’ (p. )

20 The apparent “backward-gapping” cases that have been found in languages have been argued to involve some other process than gapping (for example, Citko  claims that “backward-gapping” is best analyzed as Right-Node Raising).



  

While in English the conjunction of two events is negated under gapping, in Russian two separately negated events are conjoined. In order to express the wide-scope reading in Russian one has to extract the verb overtly in syntax: ()

Ne možet Petja polučat’ $, a ego zamestitel’ [ ] $. not can Petja earn $, and his deputy $. A: *‘Petja cannot earn $, and his deputy cannot earn $.’ B: ‘It can’t be the case that Petja earns $, and his deputy earns $.’ (p. )

In Johnson () the wide scope of negation is accounted for via ATB movement of the gapped modal, after which the modal takes scope over the conjunction. But as we see from (), the same line of reasoning cannot be adopted in the case of Russian gapping. In other words, if gapping in Russian is the result of ATB, it remains unexplained why the wide scope of negation in examples like () is impossible. Kazenin mentions that another possible approach to Russian gapping is vP/VP-ellipsis with evacuation of retained material out of the vP/VP. For example, Jayaseelan’s () approach could be applied. Jayaseelan argues that gapping is VP-ellipsis that was preceded by rightward extraction of the VP-internal material that is retained. That would mean that for the gapped clause in () we would get a structure in () (Kazenin ): ()

Biznesmen polučil dva ranenija, a ego voditel’ tri Businessman got two injuries CONJ his driver three ‘The businessman got two injuries, and his driver got three injuries.’

()

Ego His

voditel’ driver

[VP

polučil got

ti] [

tri three

ranenija. injuries

ranenija]i injuries

Treating Russian gapping as vP/VP-ellipsis plus rightward extraction would explain the ungrammaticality of examples such as () (Kazenin ):21 ()

*Biznesmen polučil dva ranenija, a ego voditel’ Businessman got two injuries CONJ his driver [VP [VP1 [ ] tri ranenija] i [VP2 byl dostavlen v bol’nicu]]. [ ] three injuries and was delivered to hospital ‘The businessman got two injuries, and his driver got three injuries and was delivered to the hospital.’

If Russian gapping is vP-ellipsis plus rightward extraction and in the second conjunct of () we have conjoined VPs rather than sentences, Kazenin argues that () is ungrammatical because of the CSC violation: the rightward extraction of the phrase tri ranenija from the first of the conjoined VPs violates the CSC. Of course, in order to pursue

21

As has been noted by a reviewer, the grammaticality of () seems to improve when the parallel conjunct is added into the antecedent clause. We don’t have an explanation for this fact.





this line of argumentation that treats gapping in Russian as a case of vP/VP-ellipsis seriously, one has to explain, for example, why this type of vP/VP-ellipsis is restricted to coordinate constructions, and other distributional differences. To our knowledge, such explanations have not been offered for Russian gapping. To conclude, predicate ellipsis in Russian comes in two varieties: vP-ellipsis and gapping. Whether the two are really distinct phenomena or whether gapping in Russian represents just a subtype of vP-ellipsis (as speculated in Kazenin ), remains an unsettled question.

. V- 

.................................................................................................................................. Russian is well-known for allowing normally transitive or ditransitive verbs to appear without any expressed internal arguments. We refer to this phenomenon as “verb stranding”: ()

A: Ty poznakomil Mašu s Petej? you introduced Masha with Petya ‘Did you introduce Masha to Petya?’ B: Konečno, poznakomil [ ] [ ]. of course introduced [ ] [ ] ‘Of course, I introduced [Masha to Petya].’

That this phenomenon probably involves ellipsis and not discourse drop was first argued by Gribanova (b), who shows that the phenomenon is unavailable when two factors coincide: (i) there is no linguistic antecedent, (ii) the stranded verb is found within a syntactic island. The former is a known diagnostic for ellipsis (see Gribanova b for discussion), the latter context rules out successful discourse-licensed argument drop. ()

Situation: A man with ripped jeans enters the room A. Ne volnujsja, babuška zaš’et [ ]. not worry grandma will.sew [ ] ‘Don’t worry, grandma will sew [them].’ (=argument drop; no linguistic antecedent) B. *Ne volnujsja, sejčas pridjet [čelovek, kotoryj zaš’et [ ] ]. (*AD) not worry now will.come [man who will.sew [ ] ] (*vPE) ‘Don’t worry, [someone who will sew [them]] is coming.’ (ex. from Gribanova ), (no linguistic antecedent and within an island) C. Menja volnuet, čto nikto ne zašil me worries that no one not sewed ‘It worries me that no one has sewed these jeans.’

džinsy. jeans

D. Ne volnujsja, sejčas pridet [čelovek, kotoryj zaš’et [] not worry now will.come [man who will.sew [ ] ‘Don’t worry, [someone who will sew [them]] is coming.’ (ex. () from Gribanova b) (same island as (B) but OK with linguistic antecedent)

]. (ellipsis) ]



  

Thus, it appears that ellipsis of some kind is involved in Russian verb stranding. However, there is disagreement in the current literature about what kind of ellipsis is involved. Gribanova (a, , b) argues that such constructions, at least within islands, are the result of vP-ellipsis, where the verb raises to an Asp head outside of vP. Bailyn (, ) argues that the ellipsis involved must be something more akin to argument ellipsis, for a variety of syntactic reasons. Two such arguments are that the conditions on strict and sloppy identity appear to be different in verb stranding from standard vP-ellipsis, and that there is an unusual requirement that the subject also be elided in verb stranding but not in vP-ellipsis (), as shown here ()–(): ()

a. Čto Saša dumaet pro lingvistiku? what Sasha thinks about linguistics.ACC ‘What does Sasha think about linguistics?’ b. Ja uveren, čto (*on) nenavidit I sure that (*he) hates ‘I am sure that (he) hates [it].’

[ [

]. ]

c. Menja udivljaet tot fakt, čto (*on) nenavidit [ me surprises that fact that (*he) hates [ ‘The fact that (he) hates [it] surprises me.’

(V-stranding)

]. ]

(V-stranding)

()

*Ja dumal, čto nikto ne znaet etu pesnju, (V-stranding) I thought that no one not know this song a na samom dele Maša spela [ ]. but on right affair Masha sang [ ]. ‘I thought that no one knows this song, but Masha actually sang [this song].’

()

Ja dumal, čto Saša budet pet’ etu pesnju, (vP-ellipsis) I thought that Sasha will sing this song a na samom dele Maša budet [vP ]. but on right affair Masha will [ ] ‘I thought that Sasha will sing this song, but Masha actually will sing this song.’

Other details of the debate will not be given here; the interested reader is referred to the works cited. Regardless of the correct characterization of the kind of ellipsis involved, the question naturally arises as to why languages like English strongly disallow verb stranding of this kind. After all, neither language is a V → T raising language (a point agreed upon by both Gribanova and Bailyn), and therefore standard (extended) vP-ellipsis after verb raising to T cannot be invoked for such cases as it is for similar constructions in languages that do allow such raising (see Goldberg ). Neither Gribanova nor Bailyn provides a solution for this typological puzzle, though Gribanova suggests it is related to the Russian aspectual system and the existence of additional functional structure between vP and TP (AspP) where the raised verb can be stranded after vP-ellipsis, but still remain lower than TP. The typological consequences of this claim for Russian are not explored by Gribanova.





. S       R

.................................................................................................................................. In this section we’ll briefly discuss three other types of ellipsis present in Russian— comparative deletion, fragment answers, and Right-Node Raising.

.. Comparative deletion There are two types of comparatives in Russian that involve comparative deletion (Pancheva ; Ionin and Matushansky ). In () we see an example of a clausal comparative and in () an example of a phrasal one: ()

Georgij byl sil’nee čem Boris [ Georgij was stronger than BorisNOM [ ‘Georgij was stronger than Boris was strong.’

()

Georgij byl sil’nee Borisa [ ]. Georgij was stronger BorisGEN [ ] ‘Georgij was stronger than Boris was strong.’

]. ]

(clausal comparative)

(phrasal comparative)

The comparatives in () and () differ in many important ways. First of all, while clausal comparatives can be used either with synthetic () or analytic forms (), phrasal comparatives can be formed only with synthetic ones (): ()

Georgij byl bolee sil’nym čem Boris Georgij was more strong than BorisNOM ‘Georgij was stronger than Boris was strong.’

()

*Georgij Georgij

byl was

bolee more

sil’nym strong

Borisa BorisGEN

[ [

[ [

]. ]

]. ]

Secondly, the remnant in clausal comparatives can have different case markings which correspond to different arguments ((), ()), while in phrasal comparatives the only possible case marking is genitive (): ()

Ja ljublju Mišu bol’še čem [ I love Misha more than [ ‘I love Misha more than I love Petya.’

()

Ja ljublju Mišu bol’še čem Petya I love Misha more than PetyaNOM ‘I love Misha more than Petya does.’

] ]

Pet’u. PetyaACC

[ [

]. ]

 ()

   Ja ljublju Mišu bol’še Pet’i. I love Misha more PetyaGEN ‘I love Misha more than I love Petya.’ or ‘I love Misha more than Petya does.’

Thirdly, clause-level elements such as temporal adjuncts and tensed verbs can occur only in clausal comparatives ((), ()), but not in phrasal ones ((), ()): ()

Katja probežala segodnja stometrovku bystree, [čem Katja ran today  meters faster than ‘Today Katja ran  meters faster than Lena did yesterday.’

()

*Katja Katja

()

Miša srisovyval kartinku akkuratnee, [čem srisovyval Misha copied picture more.accurately than copied ‘Misha copied the picture more accurately than Sasha did.’

()

*Miša Misha

probežala ran

srisovyval copied

segodnja today

kartinku picture

stometrovku  meters

bystree faster

akkuratnee more.accurately

Lena Lena

[Leny LenaGEN

[srisovyval copied

včera ]. yesterday

včera] yesterday Saša] SashaNOM

Saši] SashaGEN

Finally, binding also shows that the amount of structure in two comparatives is different: while in clausal comparatives remnants’ possessors are expressed by pronouns (), in phrasal comparatives they are expressed by reflexives () (Pancheva ): ()

Germank byl sil’nee, čem egok / *svojk protivnik German was stronger than his / hisREFL adversary ‘German was stronger than his adversary was strong.’

[ [

()

Germank byl sil’nee *egok /svojegok protivnika German was stronger his /hisREFL adversaryGEN ‘German was stronger than his adversary was strong.’

]. ]

[ [

]. ]

The difference between () and () suggests that clausal comparatives have more structure than phrasal ones. Pancheva () argues that phrasal and clausal comparatives shouldn’t be analyzed as having the same underlying structure. While clausal comparatives take CP-complements (), phrasal comparatives take small clause complements (): ()

Clausal comparatives in Russian: CP-complements a. Miša vyše čem Boris [ ]. Misha is taller than BorisNOM [ ] ‘Misha is taller than Boris is tall.’





DegP

b. Dego

PP

er (ee) Po

CP

than (∅) wh1 (čem)

TP

Boris is d1-tall ()

Phrasal comparatives in Russian: SC-complements a. Miša vyše Borisa [ ]. Misha is taller BorisGEN [ ] ‘Misha is taller than Boris is tall.’

DegP

b. Dego

PP

er (ee) Po

SC

than (∅) Borisa (GEN)

AP d-tall

In clausal comparatives the remnant DP moves out of the TP, and then either the whole TP or just the part below the tensed verb is elided (Pancheva ): ()

[CP wh1 Boris is d1-tall] → [CP wh1 [TP t2 is d1-tall] [DP Boris]2]

In phrasal comparatives it seems like we have a case of AP-ellipsis: ()

[SC Borisa [AP d-tall]] → [SC Borisa [AP d-tall]]

To sum up, there are two different constructions in Russian which display comparative deletion. One is best analyzed as an instance of clausal ellipsis, while in the other the elision of the complement of a small clause (AP) takes place.

.. Right-Node Raising and fragment answers Russian exhibits two other kinds of ellipsis: Right-Node Raising and fragment answers.

... Right-Node Raising ()

Andrej napisal [ ], a Maša otpravila pis’mo. (Right-Node Raising) Andrej wrote [ ] CONJ Masha sent letter ‘Andrej wrote the letter, and Masha sent the letter.’



  

An interesting feature of Right-Node Raising is that it is ungrammatical if the two predicates mark their arguments differently: compare (), where the case of the elided noun phrase (ACC) is the same as the case of the overt one, with (), where two noun phrases receive different cases. ()

Učitel’nica xvalit [ ACC], a mama rugaet teacher praises [ ACC] CONJ mother criticizes ‘The teacher praises and the mother criticizes Masha.’

()

*Učitel’nica teacher

xvalit praises

[ [

ACC], ACC]

a CONJ

mama mother

Mašu. MashaACC

gorditsja is.proud.of

Mašej. MashaINSTR

... Fragment answers As for fragment answers, as in (), they display puzzling behavior with respect to island constraints. ()

A: Kto napisal èto pis’mo? who wrote this letter ‘Who wrote this letter?’ B: Andrej [ ]. Andrej ‘Andrej wrote this letter.’

For example, () and () suggest that fragment answers obey island constraints (in these particular examples, the Complex NP Constraint and the Factive Island Constraint respectively), but the grammatical example in () (Adjunct Island) suggests that in some cases island effects do not obtain. ()

A: Ljuba pytaetsja vyučit’ tot že papuasskij jazyk, čto i Katja? Ljuba tries to.learn the same Papuan language that CONJ Katya ‘Does Ljuba try to learn the same Papuan language that Katya does?’ B: *Net, Maša [ ]. no Masha [ ] ‘No, Ljuba tries to learn the same Papuan language as Masha.’

()

A: On vspomnil, čto Lena uže podpisala pismo? he recall that Lena already signed letter ‘Did he recall that Lena has already signed the letter?’ B: *Net, [ ] zajavlenije. no [ ] application ‘No, he recalled that Lena has already signed the application.

()

A: Saša ušel posle togo kak Lena podpisala zajavlenije? Sasha left -------after------ Lena signed application ‘Did Sasha leave after Lena signed the application?’





B: Net, [ ] dogovor. no [ ] contract ‘No, Sasha left after Lena signed the contract.’ Both fragment answers and RNR in Russian haven’t received much attention and require further research.

. C

.................................................................................................................................. In this chapter, we have reviewed the major elliptical constructions in Russian. Much work remains to be done to determine to what degree various of these construction types share the same properties with those of other languages, but the overall range of Russian ellipsis types should be clear from what we have seen here as should their remarkable similarity, with several notable exceptions such as verb-stranding constructions, to their English counterparts.

  ......................................................................................................................

                ......................................................................................................................

 

. I

.................................................................................................................................. I this chapter I describe how certain ellipsis phenomena differ across varieties of English. The discussion here is mainly focused on VP-ellipsis1 because, insofar as I know, there is little to no description in the literature of dialectal variation in the domain of clausal ellipsis, nominal ellipsis, or other such elliptical constructions.

. T   VP-

.................................................................................................................................. To begin with, let us consider dialectal variation in which verbs may occur as the sole surviving verbal head in the context of VP-ellipsis. The retained verb is typically described as the “licensor” of VP-ellipsis (Lobeck ; Johnson b), since it is clearly required for ellipsis to work,2 and the standard description of English VP-ellipsis is that it is only possible if at least one auxiliary verb is retained as a licensor. This auxiliary may be be, have, a modal, and if none of these options is available, the dummy verb do must appear. ()

a. b. c. d.

John has sang, and Bill has too. John is singing, and Bill is too. John should sing, and Bill should too. John sang, and Bill *(did) too.

Throughout I use the term ‘VP-ellipsis’ following tradition, with no strong commitment to whether or not each case involves deletion of a node identifiable as a VP. Miller and Pullum () argue that Sag’s (a) term ‘post-auxiliary ellipsis’ is more accurate, but it is not clear that that would apply to all the cases discussed here. 2 For different theories of ellipsis licensing, see Lobeck (), Merchant (), Aelbrecht (), and Thoms (b), among many others. 1

  



Dialect data is useful for making this description more precise because there is substantial dialectal variation in English in the domain of auxiliary verbs. In what follows I describe a few different sources of variation and how they interact with VP-ellipsis licensing, and I discuss how this allows us to make our generalizations about licensors more precise.

.. VP-ellipsis and auxiliary-like verbs First let us consider variation in whether or not a given verb shows the morphosyntactic behaviour of an auxiliary verb. One case of interest is ‘main verb’ have used to express possession: this may survive VP-ellipsis in at least British and Australian varieties of English, whereas this is impossible in all3 standard English varieties in the USA (thus requiring do-support). (I henceforth refer to these varieties as US English, USE.) () John has a copy of that book, and Mary has too.

BrE OK, *USE

The difference between the dialects shown by () is perhaps unsurprising, since it is known that have retains the last vestiges of auxiliary verb syntax in the British and Australian ones but behaves fully like a lexical verb in the American ones. The auxiliary-like status of have in these dialects is indicated by cliticization to the subject, raising to C in questions and raising past negation. These tests do not all pattern together uniformly though, as many standard British varieties (i.e. from southeast England) only allow (a), in others inversion is possible in questions but not with negation, and in some varieties (northern England) raising past negation in declaratives is either impossible or restricted only to cases where the possessum is indefinite (dialects in the southwest of Scotland). Broadly, the tendency in the UK seems to be that have tends to retain more raising properties the further north you go, and the dialects which allow raising past negation also allow raising to C (but not vice versa). All these dialects contrast with North American dialects, in which all of () are unacceptable (just like with ()). () a. b. c. d.

I’ve a copy of that book with me. Have you any money? I haven’t any money. I haven’t that book with me.

BrE OK OK in most British varieties OK in many northern British varieties OK in e.g. northeast Scotland

The correlation between auxiliary-like behaviour and the ability to survive ellipsis can also be observed with other uses of have. Consumption have (as in have steak) shows a

3

A note on the contextual restriction of quantification: throughout this chapter I will often make claims about “all dialects” of some kind or another having one characteristic or another. Obviously this is overstating things, as there is still a lot that is not known about just how much some of the phenomena in question vary within dialect communities: for instance, I know little about whether or not have-raising is possible in the southern American English dialects which share a number of relic features with the British dialects which they are closely related to historically. Thus throughout, the set of dialects in question should be understood as those which I have been able to access data on directly or indirectly.





much more lexical verb-like syntax in most British varieties, failing all the raising tests, and it also fails to license ellipsis (much like possessive have for American dialects). The same pattern shows up when we consider other main verb uses of have, such as its causative use. ()

a. *I haven’t steak for dinner every night. b. *John has steak for dinner every night, and Mary has too.

()

a. *I haven’t my engine checked by a mechanic regularly. b. *I have my engine checked by a mechanic regularly, and Mary has too.

The same correlation between raising and ellipsis even seems to be attested across different instantiations of possessive have in single dialects. For many speakers of have-raising dialects, raising is much better with present tense forms of have; for instance, in my own dialect the examples in () are degraded, even though (a)–(b) are wholly acceptable.4 Quite why this would be is not clear, but what’s of interest is that these judgments are again tracked by ellipsis judgments, in that possessive have is not a good VP-ellipsis licensor in its past form. ()

a. ??I’d a copy of that book when I was younger. b. ??Had you any idea what you were letting yourself in for? c. ??I hadn’t any money with me at the time.

()

??John had a copy of that book when he was younger, and Mary had too.

This indicates that it is not some idiosyncratic property of British dialects that they allow all uses of have to be conflated with auxiliary have; rather, it is the syntax of the verb which determines whether or not a given have may survive ellipsis, and it seems there is a correlation between whether a verb displays auxiliary-like properties and whether it can survive ellipsis. Interestingly, it is not the case that auxiliary-like verbs always survive as ellipsis remnants. First, consider the need passive, a non-standard passive construction which is available in various East Coast American dialects5 and in all Scottish English dialects (see Edelstein  and references cited therein). In this construction, a passive VP is embedded under the verb need, which functions as a passive auxiliary of sorts; however need does not show auxiliary-like properties like verb-raising to T or precluding do-support. () shows that need does not license ellipsis (Thoms and Walkden ), thus indicating that not all auxiliary-like verbal elements license ellipsis.

Preliminary results in fieldwork for the Scots Syntax Atlas indicate that this is a general pattern, in that in all eighty-three locations sampled at the time of writing the present tense forms of have are rated higher than the past tense forms. 5 For details see the Yale Grammatical Diversity Project’s page on this phenomenon: , accessed on  June . 4

   () a. b. c. d. e. f.

This car needs washed. He needs spoken to. He doesn’t need spoken to. *He needs not spoken to. Does he need spoken to? *Needs he spoken to?

() a. *This doesn’t need washed, but that needs. b. A: So your car needs washed? B: *Yeah it needs.



Scottish English

Scottish English

The same thing is seen with habitual be in African American English (AAE).6 In this construction, an uninflected form of be encodes a habitual interpretation of the predicate it takes as its complement (see e.g. Green , ). In terms of our diagnostics, habitual be does not behave like an auxiliary verb, in that it does not raise and it triggers do-support in various raising contexts (Green ). Like passive need, habitual be is also unable to license ellipsis (Green ).7 () You know John be at the mall on Wednesdays. ‘You know John is usually at the mall on Wednesdays.’ ()

()

a. b. c. d. e.

Becky be watching the basketball games. Do Becky be watching the basketball games? *Be Becky watching the basketball games? Becky don’t be watching the basketball games. *Becky be not/ben’t watching the basketball games.

*Bruce be singing, and I be, too.

AAE

AAE AAE

These facts allow us to narrow our VP-ellipsis generalization further: only those auxiliaries which can occur in T can serve as licensors.

.. Null verbs Another aspect of VP-ellipsis which emerges from looking at dialect data is that the surviving auxiliary-like element must be overt. For instance, finite non-past instances of auxiliary and copular be may optionally be null in African American English (AAE), as (a) shows. Conner () observes (following up on observations in Labov ) that this optionality disappears when the auxiliary is the licensor of VP-ellipsis.

6 7

Thanks to Tracy Conner for bringing these facts to my attention. A similar pattern holds for the aspect marker BIN, as discussed in Green ().

 ()

 a. Joe (is) so fast he (is) gonna get a ticket. b. If Vikki (is) sipping vino on the porch, then I’m sure uncle Marco *(is).

AAE AAE

One further relevant observation to add is that VP-ellipsis is possible with the null auxiliary construction in AAE when an uncontracted not is present at the ellipsis site. ()

Errbody on instagram lookin like they mad rich but they not.8

One might speculate that the same effect is seen in infinitives in standard English. In a discussion of VP-ellipsis in infinitives Johnson (b) notes that while VP-ellipsis is impossible in all of the environments in () (data from Zwicky  and Zagona a,b), in all cases VP-ellipsis is ameliorated by inserting not to the left of the nonfinite auxiliary to. I add here the observation that the same effect is not seen when negation follows to. ()

a. *Mag Wildwood came to introduce the barkeep, and I came to as well. b. *You should play with rifles because to is dangerous. c. ??Ron wanted to wear a tuxedo to the party, but Caspar couldn’t decide whether to.

()

a. Mag Wildwood came to introduce the barkeep but I came (precisely) not to. b. You should unload rifles because not to is dangerous. c. If Ron knows whether to wear a tuxedo, and Caspar knows whether not to, do they know different things?

()

a. *Mag Wildwood came to introduce the barkeep but I came (precisely) to not. b. *You should unload rifles because to not is dangerous. c. *?If Ron knows whether to wear a tuxedo, and Caspar knows whether to not, do they know different things?

Zwicky () and Zagona (a,b) argue that the problem with () is that non-finite to is a clitic which must attach to a host to its left, and they submit that this cliticization process is impossible when to is separated from its potential host by an island boundary. Assuming this is correct, the effect of adding negation in () and the absence of any such effect in () would then follow if it were the case that a preceding not may provide a host for encliticization of to. In both () and (), then, licensing becomes possible when some additional prosodic constraint on the licensing element is satisfied. Similar effects can be seen in English varieties in some parts of Scotland. Smith () reports that Buckie English optionally allows for null forms of finite non-past do and auxiliary have when the inflection would be zero (e.g. first person), so long as they co-occur

This example is taken from the song ‘Errbody’ by Yo Gotti. Tracy Conner (p.c.) has confirmed for me that it is grammatical in AAE. 8

  



with negation, as shown in ().9 The elicited data in () shows that VP-ellipsis is unacceptable when the null auxiliary is the licensor. ()

()

a. I {na/dinna} mine fa come in. ‘I don’t remember who came in.’ b. I {na/havena} been there in a while. ‘I haven’t been there in a while.’ a. She likes bobbies but I {*na/dinna}. ‘She likes policemen but I don’t. b. You’ve been tae France but I {*na/havena} ‘You’ve been to France but I haven’t.’

Buckie English (Smith : ) Buckie English

Buckie English Buckie English

One point to note regarding the Buckie case is that the negation which occurs with the null auxiliary is the contracted form -na (-nae in western Scottish varieties), which is similar but not equivalent to StE -n’t (see Smith , Weir , and Thoms et al.  for discussion). This contracted negation forms a phonological word with the preceding subject, which is typically pronominal, so it is likely that ellipsis in this context is not possible for the same reason that contracted auxiliaries may not occur adjacent to ellipsis sites or gap sites. ()

John’s waiting, and {*I’m / I am} too.

()

I wonder who {*that’s / that is}.

Going further, it could be the case that the overtness requirement for ellipsis licensors is due to some general prosodic property of structures of this kind too, whereby auxiliary verbs at the edge of a gap must be stressed and hence cannot be null or reduced. Interestingly, there is another corner of Scots dialects which indicates that this might not be adequate either. Thoms et al. () note that in certain Scots dialects from the central belt of Scotland, contraction is possible in constructions like the following, which are typically used in contexts where the speaker has discovered something. ()

a. Here it’s! b. There it’s!

Speakers who accept and produce examples like these nevertheless reject ()–() just like other English speakers. What these examples suggest is that a fully prosodic account of auxiliary contraction is not viable, and that there must be some irreducible syntactic component to the contraction restriction. A complete analysis of these facts should prove instructive when it comes to assessing the extent to which restrictions on ellipsis licensing can be reduced to prosodic constraints.

9

These are based upon naturally occurring examples. Further examples of these have been found in data from other communities in the northeast of Scotland as well.





. VP-  - 

..................................................................................................................................

.. Optionality in ellipsis of non-finite auxiliaries The preceding discussion of ellipsis licensors focuses on simple cases where there is just one auxiliary. However, an interesting property of English VP-ellipsis is that there is often optionality in the size of the ellipsis site in clauses with multiple verbal heads, with some heads having the option of either being included or excluded in the ellipsis site. This is most widely available with the non-finite form of auxiliary be. In ()–(), we see that non-finite be may optionally be included in the ellipsis site when the auxiliary immediately above is a finite modal or finite have. In (), we see that the same optionality is attested when be occurs under non-finite have in clauses with three auxiliaries. ()

a. Molly could be sent home today, and Astrid could (be) sent home today, too. b. Molly might be chasing cats, and Astrid might (be) chasing cats, too.

()

a. Molly has been sent home, and Astrid has (been) sent home today, too. b. Molly has been chasing cats, and Astrid has (been) chasing cats, too.

()

a. Molly could have been sent home today, and Astrid could have (been) sent home today, too. b. Molly might have been chasing cats, and Astrid might have (been) chasing cats, too.

The same pattern is observed with be used as the copula, even though this is often analysed as being the ‘main verb’ of copular clauses. ()

a. Brian has been a student for two years, and Caitlin has (been) a student for two years, too. b. Brian might be a student just now, and Caitlin might (be) a student just now, too. c. Brian might have been a student for two years, and Caitlin might have (been) a student for two years, too.

However this optionality is not always observed with all auxiliaries, and much discussion of the nature of VP-ellipsis licensing has focused on cases where this optionality fails (e.g. Akmajian and Wasow ; Sag a; Thoms b; Sailor a, ; Aelbrecht and Harwood ). One such case is the failure of VP-ellipsis licensing when the auxiliary adjacent to the ellipsis site is in the -ing form (Sag a; Johnson b); I’ll call this the -ing constraint. ()

a. *John isn’t being spied on, but Mary is being. b. *John isn’t being spied on now, but he will have been being in the s.

  



We see the opposite kind of failure of optionality with non-finite forms of have. As noted by Sailor (a, ), Harwood (), and Aelbrecht and Harwood (), even though non-finite be may be optionally elided when it follows a modal auxiliary, the same does not hold with non-finite have: for many speakers, ellipsis of non-finite have is completely impossible. ()

a. *Molly could have been sent home today, and Astrid could have been sent home today, too. b. *Molly might have been chasing cats, and Astrid might have been chasing cats, too.

These patterns seem to show that it is not just the feature specification of T which is relevant to whether or not ellipsis is licensed, but rather non-finite auxiliaries have a role to play as well. These facts are relevant in the present context because there is substantial dialectal variation with respect to the ability of non-finite auxiliaries to be retained in ellipsis contexts. We turn to this next.

.. Non-finite auxiliaries and raising An important fact about VP-ellipsis licensing in English is that, for the most part, a verbal element’s ability to function as a licensor of ellipsis in its finite form tracks its ability to survive ellipsis as an additional non-finite auxiliary in multiple auxiliary constructions of the kind seen in section ... Thus the patterns of successful and failed VP-ellipsis licensing demonstrated in section . can be recreated with their non-finite counterparts. The dialectal variation seen with possessive have can be replicated for cases where it’s nonfinite, while non-finite versions of the other main verb haves still fail to survive ellipsis. () a. John will have a copy of that book, and Keir might have too. OK BrE, *USE b. A: Do you think Newton had a copy of Copernicus’ De revolutionibus? A: I think he must have had. OK BrE, *USE () a. John might have steak for dinner every night, and Mary might (*have) too. b. Mr Burns would have The Rolling Stones killed, and Smithers would (*have) too.

BrE, USE BrE, USE

Habitual be and the need of the need-passive fail to be retained in VP-ellipsis contexts even when there’s a licensor in T: ()

A: You know John be at the mall on Wednesdays. B: No he don’t (*be).

AAE

() John’s car needs washed, and his bike might (*need) too. e.g. Standard Scottish English





The generalization, then, is that only verbs which show the morphosyntactic properties of auxiliary verbs may survive ellipsis, be it the licensor or a non-finite hanger-on.

.. Have-deletion Let’s go back now to the have and -ing restrictions mentioned at the outset of section ... First, consider have: while it is reported by Sailor (a) and Harwood () that havedeletion is unacceptable, the data in () are somewhat controversial, since many speakers accept such examples readily. Examples where have is deleted are given by authors from various different dialect regions: see Lasnik (a) for USE, Thoms (b) for a Scottish variety,10 and Aelbrecht and Harwood () report that a significant proportion of their northern England-based informants accept these examples. It seems likely, then, that this is a matter of interspeaker variation, rather than regional variation. Harwood (: ) (picking up on suggestions in Kayne ) offers an intriguing speculation regarding this variability. He proposes that in general have cannot be deleted by VP-ellipsis and that when have does go missing this is due to some independent rule which allows modals to embed a participial phrase with a perfect interpretation but without the perfect head have; in other words, there is an independent means for removing have for these speakers. As Harwood notes, such rules seem to be available for some Scandinavian languages, and his speculation is lent further plausibility by the fact that have-deletion is possible in a number of environments in some Scottish dialects. For instance, Macafee () has noted naturally occurring examples like () in Scottish varieties, and fieldwork in the SCOSYA project has found that such examples are rated as acceptable quite widely in Scotland (see also the discussion of Smith  in ..). ()

a. I would only Δ been about two at the time. b. I would rather Δ made dinner myself.

What seems to be crucial for such structures is the presence of some other non-verbal element in the area which would normally host have (or indeed the phonologically reduced enclitic schwa which it is often realized as), so it is viable that this is a phonologically conditioned deletion rule which may also apply adjacent to an ellipsis site. This kind of analysis also makes it easier to understand the variability of the data, since a phonological rule for deleting have (or indeed its highly reduced schwa form) seems more likely to be acquired variably by speakers with broadly similar grammars than a syntactic rule which allows for ellipsis of different sizes of constituent in a limited set of syntactic contexts (i.e. those where have is dominated by a modal). Indeed it is relevant that very few other ellipsis-related rules which are normally analysed syntactically are as subject to ‘in-the-room variation’11 across 10 Sailor introduces a number of controls for ensuring that what is elided in these cases really is a larger constituent rather than some smaller VP, as he claims that ambiguities muddy the waters somewhat. My own judgment for a Scottish variety is that have-ellipsis is still possible when these controls are put in place. 11 This term is due to Jim Wood. Wood identifies transitive control promise as an example of a grammatical phenomenon in English which is subject to unpredictable in-the-room variation.

  



varieties as this one; for instance, I have never encountered any such variability in accepting modals as licensors of VP-ellipsis, nor have I encountered variability with British speakers with respect to judgments like () for ellipsis licensing with possessive have. This kind of argument is somewhat indirect, and unlikely to prove convincing in the absence of further supporting evidence, but nevertheless it shows another way in which microcomparative evidence may be brought to bear on the analysis of ellipsis and other such grammatical phenomena. A full assessment of Harwood’s have-deletion analysis would require us to give greater consideration to the other factors which condition its availability. For instance, havedeletion is typically only acceptable when the preceding modal auxiliary is might or should, and examples with must are typically rejected even by the most liberal speakers.

.. The anti -ing constraint Now consider the anti -ing constraint, which was identified by Sailor (a) and Harwood () as a strong constraint which is not subject to the same kind of variation which we see with have. It turns out that there are in fact very many English speakers who allow VPellipsis in the presence of non-finite being, as the following examples from Google searches indicate: () a. He uses this then to tell me that I’m being aggressive. I honestly don’t think I am being, but there’s no doubt he makes me angry enough to become very strident.12 b. A: Rory, be careful with her. B: I am being though!13 c. A: Hey Joe, please be patient. [ . . . ] B: I am being but its not that nice being ignored [ . . . ]14 Although NAmE speakers consistently reject these kinds of examples,15 I have found that they are readily accepted by British English speakers from all around the country, although not all speakers accept them. In general this seems to only occur when there is no progressive auxiliary in the antecedent, and it may be relevant that many speakers reject

12 , accessed on  June . . 13 , accessed on  March . 14 , accessed on  March . 15 One point of potential relevance is that examples of VP-preposing which strand auxiliary being are reported as only marginal by Kyle Johnson in his survey article (Johnson b), even though he finds basic VP-ellipsis examples like () unacceptable.

(i) ?Madame Spanella claimed that discussed widely, Holly is being. Given that the licensing conditions on VP-fronting track those on VP-ellipsis consistently otherwise, this could be taken as a further indication that some NAmE speakers may marginally allow licensing with retention of being in the right discourse conditions.





ellipsis of -ing forms when there is no progressive aspect in the antecedent, as Lasnik (a) reports that examples like the following are unacceptable:16 ()

*John won’t enter the competition, but Peter is entering the competition.

It is possible, then, that the pressure to retain being to avoid violation of the constraint exemplified by () may thus force being to survive ellipsis exceptionally in cases like (). This option also seems to be conditioned by prosody to some extent, as those speakers who do accept examples like () acknowledge that this requires stressing the preceding finite auxiliary. Clearly, the anti -ing condition is not a hard constraint on English VP-ellipsis. But if examples like () are exceptional, in that they are wholly unacceptable for many speakers, they should prove useful in identifying the general rule for retaining additional auxiliaries, and perhaps for VP-ellipsis licensing generally.

.. Additional auxiliaries in tag questions and retorts Most of the variation discussed so far has concerned differences between British and North American dialects, with USE and Canadian English patterning together against the British varieties. However there is at least one case that I know of where the judgments of Canadian English speakers pattern with British dialects and contrast with those of USE speakers, in Sailor’s () discussion of the retention of additional auxiliaries in instances of ‘high VPE’. The starting point for this is the observation that voice mismatches under VP-ellipsis are typically only acceptable when the elided VP is in a subordinate clause, for instance an adverbial; in most other configurations, such as coordinate structures, emphatic retorts, or tag questions, voice mismatches are impossible. () The janitor must remove the trash whenever it is apparent that it should be. (Merchant d) () a. *The janitor removed the trash, but the recycling wasn’t. b. Your car should have been fixed by the mechanic last week{, /.}. (i) *shouldn’t he have? (ii) *No he shouldn’t have! (iii) *He should  have!

16

Again this judgment seems to be subject to some variation, with some speakers accepting other examples with similar configurations. Bronwyn Bjorkman (p.c.) reports that the following example (attributed to David Pesetsky) is relatively acceptable for some speakers: (i)

In the past, Mary hasn’t taken medication for her condition, but she is now.

And while examples like (i) are not always perfect for all speakers, Rooryck and Schoorlemmer have noted that there is even less objection to examples like (ii), where the antecedent is an infinitive. Rooryck and Schoorlemmer argue that the important difference between the good cases and the bad cases is that the bad ones involve mismatches in Mood specification, but this would leave examples like (ii) unaccounted for. (ii)

Emma intends to write a novel, and Anna already is writing a novel.

  



Sailor shows that the latter class of VP-ellipsis environments behave differently from subordinate clause structures with respect to a number of diagnostics, and he explains these asymmetries in terms of the size of the antecedent for ellipsis: with subordinate clauses the antecedent may be a small chunk of the lower part of the inflectional domain, one that excludes VoiceP, whereas with the other environments the antecedent must be a higher constituent which includes VoiceP. Sailor dubs the environments in () ‘high VPE environments’ and the point of relevance here is that he describes intriguing dialectal variation with respect to the retention of additional auxiliaries in high VPE contexts. In outline, the dialectal difference is that in USE, VP-ellipsis in clauses with multiple auxiliaries strongly favours retention of all additional auxiliaries, whereas in Canadian, British, and Australian Englishes (clubbed together as ‘Commonwealth English’, CE), ellipsis of as many auxiliaries as possible is favoured. This manifests itself in some quite sharp differences in judgments when it comes to clauses with several auxiliaries. First, consider cases where been can be elided: USE speakers find retaining all auxiliaries (other than being) acceptable, whereas for CE speakers this is highly marked or worse; rather, the preference is to delete at least been. Sailor notes that this difference is minimal in the case of coordination, although for some CE speakers there is still a markedness to retaining been in this case. () The paper should have been accepted, shouldn’t it have been?

✓USE, ??/*CE

() A: It should have been accepted. B: No it shouldn’t have been!

✓USE, ??/*CE

() A: It should have been accepted. B: It should  have been!

✓USE, ??/*CE

()

This paper should have been accepted, and that one should have been. ✓USE, ?/??CE

The following examples show the other side of this variation: in the same high VPE environments, eliding a much larger chunk of the clause, including non-finite have, is typically acceptable for CE speakers but reported as unacceptable for USE speakers. Note that for coordination Sailor uses an idiom which requires a perfect auxiliary in order to ensure the ellipsis site contains have, although as before the contrasts here are still quite slight. () The paper should have been accepted, shouldn’t it?

??/*USE, ✓CE

() A: It should have been accepted. B: No it shouldn’t!

??/*USE, ✓CE

() A: It should have been accepted. B: It should !

??/*USE, ✓CE

() John will have been to Paris by then, and his wife will, too.

✓/?/??USE, ✓CE





The challenge for making sense of this striking observation is to pin down other dialectal properties which unite CE varieties as distinct from USE and which might therefore be used to explain the variation we see.

.. Interim summary What we’ve seen in this section is that there is variation across dialects with respect to the optionality of retaining additional non-finite auxiliaries in VP-ellipsis contexts. Some of these conditions recall those discussed in the previous section on licensors and keyed to the morphosyntax of auxiliary verbs. But others, such as the –ing restriction, seem distinct and more likely to be the result of interface conditions (at least in part), and so it’s not yet clear whether a unified account is viable. If anything, a unified account is likely to lead us in the direction of reassessing whether the more syntactic conditions discussed in section . can be reduced to interface conditions, with the right model of the syntax–phonology interface, as reducing the dialectal variation seen with -ing to syntactic conditions seems unviable.

. B 

.................................................................................................................................. ‘British do’ is the phenomenon whereby a superfluous non-finite form of do occurs at the right edge of a cluster of auxiliaries in VP-ellipsis contexts bearing the morphological ending that would normally occur on the finite verb (Haddican ; Baltin , ; Aelbrecht ; Thoms a).17 The superfluous do typically occurs after modals and have, but it may also be retained after progressive be in the same circumstances where being is retained in examples like () (with similar variability in judgments18). It cannot occur alongside passive be.19 () a. A: Are you going tonight? B: I might do. b. I haven’t brought wine but I now realize I should have done.

17

Following others, I generalize this do to all British dialects, as I have not identified any specific syntactic differences between dialects with respect to the properties described below. It is possible that do is used more commonly in southeast dialects, but this needs to be investigated more thoroughly. Note also that ‘British dialects’ is (innocently) taken to include all dialects from the British Isles, including those in Ireland, and as with a lot of British dialect phenomena it is also found in Australian English dialects. 18 My impression is that retention of doing is particularly common in southeastern dialects, although Craig Sailor (p.c.) reports that it is also common in the speech of Yorkshire-born Top Gear presenter Jeremy Clarkson. 19 Thoms and Sailor () point out that this is not some incompatibility with passivization as such, since British do occurs in other passive-like structures such as the Scottish need-passive discussed in section .. and in get-passives: (i)

a. The car doesn’t need washed now, but it will do later on. b. Rab will get fired eventually, and Tam will do too.

   c. Rab has left, and Morag has done, too. d. A: Why don’t you sit quietly? B: I AM doing.



(Quirk et al. : )

() *The steak was eaten by Bill, and the fish was done too. British do has attracted a reasonable amount of attention in the literature in recent years, and one of the major issues is whether or not it is best analysed as an elliptical structure or a verbal proform. This discussion has mainly focused on some curious extraction asymmetries between British do environments and their more standard VP-ellipsis counterparts. British do is analysed as a proform akin to do so by Haddican (), and it is easy to see why this is appealing: on the surface, the only difference between the two is the so, and these putative proforms may not co-occur with the VPs which they seem to be replacing. ()

a. *John will do run the race. b. *John will do so run the race.

Haddican strengthens the proform analysis on the basis of intriguing facts from extraction. Building on Baltin (), Haddican reports that various A0 -extractions—wh-extraction, QR of objects for scope inversion with subjects, comparative extraction, topicalization, and pseudogapping—are all incompatible with British do; in this respect it patterns with do so rather than VP-ellipsis, which is compatible with all of these extractions. ()

a.

Although I don’t know which book Fred will read, I do know which book Tom will. b. *Although I don’t know which book Fred will read, I do know which book Tom will {do/do so}.

() a. Some man will read every book and some woman will too. b. Some man will read every book and some woman will {do/do so} too. ()

a. Although she won’t eat pasta she will pizza. b. *Although she won’t eat pasta she will {do/do so} pizza.

()

a. Bart can eat more than Homer can. b. *Bart can eat more than Homer can {do/do so}.

()

a. Hazelnuts I like, peanuts I don’t. b. *Hazelnuts I like, peanuts I don’t {do/do so}.

∀>∃ *∀>∃

Noting that do so is also incompatible with passivization, Haddican therefore arrives at the generalization that both British do and do so disallow extraction, and he proposes that this follows from an analysis where they are both verbal proforms with no internal content. The argument from extraction for the proform analysis has since been contested by Baltin (, ), Aelbrecht (), Thoms (a), and Abels (), who point out that





it is not quite right that all extractions from British do are impossible. First, Baltin notes that A-extraction is possible with unaccusative and raising predicates, seeming to pattern against do so which forbids these.20 Second, Thoms shows that while QR for scope inversion of subjects and objects is not possible, QR to scope an object over negation is as acceptable with do as it is with VP-ellipsis, but impossible with do so.21 Third, Abels () reports that relativization and topicalization are both possible with British do for many speakers, the latter somewhat marginally. No such leeway is found with do so. ()

a. John might seem to enjoy that, and Fred might (do), too. b. ??John might seem to enjoy that, and Fred might do so, too.

()

a. The river will freeze solid, and the lake will (do), too. b. ??The river will freeze solid, and the lake will do so, too.

() a. Rab won’t finish two thirds of the exam. Morag won’t (do) either. b. Rab won’t finish two thirds of the exam. Morag won’t do so either.

/ >¬ */ >¬

() a. A man who steals does not incur the same measure of public reprobation which he would have (done) in the past. (Baker ) b. *A man who steals does not incur the same measure of public reprobation which he would have done so in the past. ()

a. Hazelnuts, he won’t eat, but almonds he might (?do). b. *Hazelnuts, he won’t eat, but almonds he might do so.

Thoms notes (citing Hallman  and p.c. from Carson Schütze) that it is not quite right to say that do so cannot appear with raising predicates, as the effect in examples like (b) is to be attributed to a general restriction which requires that do so must appear with non-stative volitional predicates. When this is controlled for, do so may take an unaccusative as its antecedent. 20

(i)

John appeared from out of nowhere, and Mary did so too.

Similar comments may apply with passivization, as Bruening () notes that many speakers seem to accept passives with do so as well, citing numerous examples like (ii) from Google searches (most of which are perfectly acceptable to me). (ii)

And I think everyone can agree that some of the most beautiful music ever written was done so in the name of God or gods.

This may provide a further argument for providing do so with an ellipsis analysis, as suggested below. 21 Baltin () provides the following example as evidence against QR of objects out of British do sites and over negation.

(i)

*John couldn’t read many books, and Bill couldn’t do either, but the many books that they could read were classics.

The contrast with VP-ellipsis is far from clear for me though, and it seems likely that the degradation of this example might be independent of scope, since couldn’t do is somewhat marked in general (for reasons that are not clear).

  



Finally,22 Baltin () notes a set of facts which seem to point to the conclusion that it might not even be right to say that do so does not involve ellipsis, undermining a key premise of Haddican’s argument. Baltin notes that while only VP-ellipsis (of the three constructions) allows scope inversion of subjects and objects when this would be derived by QR (i.e. where the object is a strong quantifier), scope inversion is possible when the object is an indefinite which would be able to scope out of the putative ellipsis site by in-situ scoping mechanisms; to this, we can add the observation that inverse scope is also possible with British do in these configurations as well. Crucially, this is not possible in cases with do it. (The following examples are adapted versions of Baltin’s, which he attributes to an anonymous reviewer.) () a. b. c. d.

Many men will read five books, and many women will as well. Many men will read five books, and many women will do as well. Many men will read five books, and many women will do so as well. Many men will read five books, and many women will do it as well.

 > many  > many  > many * > many

This difference between do it and the other constructions indicates that the former is the only true verbal proform, with the other constructions being derived by ellipsis, as it is hard to see why do so and do it would differ in their ability to take VPs containing specific indefinites as their antecedents. This should be a welcome conclusion, as we know from Hankamer and Sag () that do it behaves differently from do so and other elliptical constructions in other respects; that is, it seems to be deep anaphor, since it does not require a linguistic antecedent (see also Bentzen et al. ). Accepting this conclusion, the desiderata for an ellipsis analysis of do so is to account for the fact that it systematically blocks A0 -extraction but not A-extraction, and that it requires a non-stative volitional antecedent. It seems likely that the correct analysis will be one where the so part of do so is some kind of focus-related operator which occupies a peripheral position in the vP and acts as an A0 -intervener, but the matter warrants more attention than I can grant it here. The challenge with British do, then, is to work out an analysis which will derive the extraction restrictions, the ellipsis-like licensing conditions, and of course the presence of Thoms also notes a set of contrasts from Condition B (building on Kennedy ), where there is an apparent Condition B effect with VP-ellipsis and British do but not with do so. 22

(i) a. *Kim takes care of himi because hei won’t. b. *Kim takes care of himi because hei won’t do. c. Kim takes care of himi because hei won’t do so. These contrasts are subtle and subject to some variation, but the majority of speakers I have consulted share them. However, it strikes me as plausible that the contrast here is not due to some fundamental difference in the structure of the putative ellipsis sites, but rather to differences in how the three structures may realize focus on the subject: it is much more natural to put emphatic stress on the subject in the context of do so than with British do, and with VP-ellipsis focusing the subject seems to require emphatically focusing the following negation as well, whereas this isn’t the case with do and do so. There may also be interactions between focus and the height of attachment of the reason adjunct which bring antecedent containment and the problems that that poses into the picture as well. And as with many of the other phenomena discussed here, the variability of the data is highly relevant and so one might hope that a full account of these contrasts will provide the right kind of wriggle room to allow for such variation.





the superfluous do just in ellipsis contexts, and it needs to relate to some minor parameter which differs between British dialects and others. No existing account is without its problems. Baltin (, ) and Aelbrecht () fail to predict that some A0 -extractions are possible, and they both resort to stipulation to derive the fact that do occurs only in ellipsis contexts: for both, do is a realization of v which is stipulated as only being possible in the context of ellipsis. Thoms (a) fares better on the latter point: he also assumes that the do is a spell-out of v, but he states a rule for this realization which generalizes to standard do-support, and the Britishness is somewhat loosely tied to the general British propensity for retaining additional auxiliaries. Thoms does not offer a fully worked-out account of the extraction pattern, although he does succeed in pinning down the concrete generalization for when extraction is possible. Thoms begins with an apparent contradiction with respect to QR: while () seems to show that British do prohibits reconstruction, () shows that QR can sometimes escape this environment. He argues that this contradiction goes away if the effect in () is not due to the inability of the object to QR, but rather the subject’s inability to reconstruct back into its base position in the vP. This would follow since it is known independently that scope inversion of the subject and the object requires subject reconstruction as well as QR (Hornstein b; Johnson and Tomioka ; Fox ; Nevins and Anand ). Thoms shows that this analysis is supported by the fact that in cases where subject reconstruction would invert the scope of the subject and some other non-argument, the scope inversion is possible with VP-ellipsis and even with do so, but it is substantially degraded with British do. () a. Every boy won’t finished the exam, and every girl won’t, either. b. Every boy won’t finished the exam, and every girl won’t do, either. c. Every boy won’t finished the exam, and every girl won’t do so, either.

¬>∀ ??¬ > ∀ ¬>∀

Thoms thus argues that in those cases where extraction from British do fails, what is going wrong is reconstruction of the extracted constituent, and he argues that this accounts for the distribution of extraction possibilities: for instance, wh-movement from British do sites is strictly prohibited since wh-movement shows obligatory reconstruction (Chomsky b), whereas relativization and topicalization do not show obligatory reconstruction effects in many situations and may be analysed as being derived as null operator constructions (Lasnik and Stowell ). Abels () shows that this account makes the right predictions when we come to consider relativization a bit more closely: when we look at structures which would require obligatory reconstruction, British do becomes much worse than it is with regular relativization. This is demonstrated for amount relatives, free relatives, and relatives headed by idioms, all of which are known to require reconstruction of the relativized category (Bianchi ). ()

a. I put in my pocket all the money I could. b. ??I put in my pocket all the money I could do.23

While Abels marks this as only somewhat degraded, I and my informants find it completely impossible. 23

   ()

a. He buys what he can. b. *He buys what he can do.

()

a.



It’s easy to spend your time regretting that you haven’t taken advantage of every opportunity that you should have.

b. ??It’s easy to spend your time regretting that you haven’t taken advantage of every opportunity that you should have done. Since the operator movement analysis is not readily available for these relative types, they are derived by raising and hence require reconstruction, in violation of the restriction against reconstructing into British do. As for comparatives, subsequent work has since revealed a number of exceptions to the claim that comparative extraction from do is prohibited:24 () a. He ate more than he should have done. b. He’ll eat more than he should do.

(Abels ) (Thoms and Sailor )

c. The government are acting much more carelessly than they would do if there was an election on the horizon. (Thoms and Walkden ) It seems the main difference between these examples and Haddican’s (b) is that in the former the subjects of the comparative are pronouns, whereas in (b) it is a focused full DP. Haddican discusses a number of other facts about British do which indicate that its distribution is sensitive to prosodic properties of its host; for instance, it is impossible when the finite auxiliary is contracted. ()

*Sarah will arrive on time, and Tom’ll do too.

The generalization seems to be that the finite verb must be sufficiently strong in its prosodic form to host the clitic do, so one possible analysis for (b) is that placing strong accent on the embedded subject for focus reasons precludes giving strong enough accent to the finite auxiliary in order for it to host do (see also n. ). Putting this to one side, the acceptability of () requires an explanation. One possibility, explored in more detail in Thoms and Sailor (), is to unify comparatives with topicalization and matching relatives and treat them all as involving operator movement (Chomsky a; see Kennedy  for an operator movement analysis of comparatives). Thoms’ reconstruction-based generalization seems to stand up well, then. This is still far from a satisfying result, though, since it is hard to see what would actually explain a generalization of this kind. This is the heart of the matter with British do: on the one hand, all that we seem to have is a superfluous little do which is the reflex of some prosodic 24

The do in (a) is allowed not just for British speakers, but for some North Americans as well. Tim Stowell (p.c. to Craig Sailor) reports that he finds such examples acceptable, and has found agreement among other older Canadian English speakers, although for the most part USE speakers reject (a). It would be intriguing to explore whether this is related to the only other Canadian English/USE dialect split discussed in this chapter, namely the variation with respect to retorts and tags discussed in section ...





property of British dialects to retain additional non-finite auxiliaries, and yet there seems to be a hard syntactic restriction at play, one which crucially implicates an LF-phenomenon which is not expected to interact with prosody normally.

. C

.................................................................................................................................. In this chapter I have reviewed several cases of dialectal variation with respect to VP-ellipsis phenomena in English and shown that they allow us to refine generalizations about the licensing conditions on ellipsis and other such matters. The description here is only scratching the surface, and is based on a very limited understanding of the extant variation, but hopefully the recent upsurge in microcomparative syntax research will continue apace and bring with it further advances in our understanding of ellipsis and the language faculty more generally.

R

.....................................

Aarons, Debra (). Topic and topicalization in American Sign Language, Stellenbosch Papers in Linguistics : –. Abbott, Barbara (). Right node raising as a test for constituency, Linguistic Inquiry (): –. Abdulkarim, Lamya and Tom Roeper (). Economy of representation: Ellipsis and NP reconstructions, in Antonella Sorace, Caroline Heycock, and Richard Shillcock (eds), Proceedings of GALA  Conference on Language Acquisition. Edinburgh: University of Edinburgh. Abe, Jun (). Antecedent-contained deletion in Japanese: Support for the VP-ellipsis analysis. Ms. Tohoku Gakuin University. Abe, Jun (). The in-situ approach to sluicing. Amsterdam and Philadelphia: John Benjamins. Abe, Jun (). Make short answers shorter: Support for the in-situ approach, Syntax (): –. Abe, Jun and Norbert Hornstein (). ‘Lasnik-effects’ and string-vacuous ATB movement, in Myriam Uribe-Etxebarria and Vidal Valmala (eds), Ways of structure building. Oxford: Oxford University Press, –. Abe, Jun and Hiroto Hoshi (). Gapping and P-stranding, Journal of East Asian Linguistics : –. Abe, Jun and Chizuru Nakao (). String-vacuity under Japanese right node raising, in Matthew A. Tucker, Anie Thompson, Oliver Northrup, and Ryan Bennett (eds), Proceedings of FAJL . MIT Working Papers in Linguistics . Cambridge, MA: MITPWL, –. Abe, Jun and Christopher Tancredi (). Non-constituent deaccenting and deletion: A phase-based approach. Ms. Tohoku Gakuin University and Keio University. Abeillé, Anne (). Les syntagmes conjoints et leurs fonctions syntaxiques, Langages : –. Abeillé, Anne (a). In defense of lexical coordination, in Olivier Bonami and Patricia Cabredo Hofherr (eds), Empirical issues in formal syntax and semantics. Paris: CNRS, –. Abeillé, Anne (b). Stripping constructions in French. Paper presented at the Coordination and Ellipsis Conference. Université Paris ,  June. Abeillé, Anne, Gabriela Bîlbîie, and François Mouret (). A Romance perspective on gapping constructions, in Hans Boas and Francisco Gonzálvez Garcia (eds), Romance perspectives on Construction Grammar. Amsterdam and Philadelphia: John Benjamins, –. Abeillé, Anne, Berthold Crysmann, and Aoi Shiraishi (). Syntactic mismatches in French peripheral ellipsis. Presentation at CSSP (Colloque de Syntaxe et Sémantique à Paris), Université Paris . Abeillé, Anne and Danièle Godard (). Varieties of ‘esse’ in Romance languages, in Dan Flickinger and Andreas Kathol (eds), Proceedings of the th International HPSG Conference. Stanford: CSLI Publications, –. Abeillé, Anne and Danièle Godard (). The syntactic flexibility of French degree adverbs, in Stefan Müller (ed.), Proceedings of the th International HPSG Conference. Stanford: CSLI Publications, –. Abels, Klaus (). Successive cyclicity, anti-locality, and adposition stranding. PhD thesis, University of Connecticut at Storrs. Abels, Klaus (). Right Node Raising: Ellipsis or across the board movement? in Keir Moulton and Matthew Wolf (eds), NELS : Proceedings of the th annual meeting of the North East Linguistic Society. Amherst, MA: Graduate Linguistic Student Association, –.





Abels, Klaus (). Don’t repair that island! It ain’t broke. Paper presented at the University of the Basque Country, Vitoria/Gasteiz. Abels, Klaus (). Don’t fix that island! It ain’t broke. Talk at the Islands in Contemporary Syntactic Theory workshop, University of the Basque Country. Abels, Klaus (). Phases: An essay on cyclicity in syntax. Berlin: Mouton de Gruyter. Abels, Klaus (a). Displacement in syntax, in Mark Aronoff (ed.), Oxford research encyclopedia of linguistics. New York: Oxford University Press. DOI: ./acrefore/... Abels, Klaus (b). On the interaction of P-stranding and sluicing in Bulgarian. Ms. UCL. Abels, Klaus and Veneeta Dayal (). On the syntax of multiple sluicing. Talk presented at the th annual meeting of North East Linguistic Society (NELS ), UMass, Amherst, MA. Abels, Klaus and Gary Thoms (). Resumption and island repair under sluicing. Talk presented at the LAGB (Linguistics Association of Great Britain) Annual Meeting, Oxford. Abney, Steven (). The English noun phrase in its sentential aspect. PhD thesis, Massachusetts Institute of Technology. Abrusán, Márta (). Weak island semantics. Oxford: Oxford University Press. Ackema, Peter and Kriszta Szendrői (). Determiner sharing as an instance of dependent ellipsis, Journal of Comparative Germanic Linguistics (): –. Adams, Perng Wang and Satoshi Tomioka (). Sluicing in Mandarin Chinese: An instance of pseudo-sluicing, in Jason Merchant and Andrew Simpson (eds), Sluicing: Cross-linguistic perspectives. Oxford: Oxford University Press, –. Adliene, Egle (). Multiple sluicing in Lithuanian. Master’s thesis, UCL. Aelbrecht, Lobke (a). A movement account of Dutch gapping. Talk presented at Linguistics in the Netherlands Day, Utrecht University,  February. Aelbrecht, Lobke (b). A movement account of Dutch gapping. Ms. Catholic University of Brussels. Aelbrecht, Lobke (). You have the right to remain silent: The syntactic licensing of ellipsis. PhD thesis, Catholic University of Brussels. Aelbrecht, Lobke (). The syntactic licensing of ellipsis. Amsterdam: John Benjamins. Aelbrecht, Lobke (). Modal complement ellipsis: VP ellipsis in Dutch?, in Peter Ackema, Rhona Alcorn, Caroline Heycock, Dany Jaspers, Jeroen van Craenenbroeck, and Guido Vanden Wyngaerd (eds), Comparative Germanic syntax: The state of the art. Amsterdam and Philadelphia: John Benjamins, –. Aelbrecht, Lobke (). Ellipsis, in Artemis Alexiadou and Tibor Kiss (eds), Syntax: Theory and analysis. An international handbook. Berlin: Mouton de Gruyter, –. Aelbrecht, Lobke and Will Harwood (). To be or not be elided: VP ellipsis revisited, Lingua : –. Agüero-Bautista, Calixto (). Diagnosing cyclicity in sluicing, Linguistic Inquiry : –. Ahmed-Chamanga, Mohamed (). Introduction à la grammaire structurale du comorien. Volume : Le shiNgazidja. Paris: KomEdit. Akmajian, Adrian (). The role of focus in the interpretation of anaphoric expressions, in Stephen R. Anderson and Paul Kiparsky (eds), A Festschrift for Morris Halle. New York: Holt, –. Akmajian, Adrian (). Sentence types and the form–function fit. Natural Language and Linguistic Theory : –. Akmajian, Adrian, Susan Steele, and Thomas Wasow (). The category AUX in universal grammar, Linguistic Inquiry (): –. Akmajian, Adrian and Thomas Wasow (). The constituent structure of VP and Aux and the position of the verb be, Linguistic Analysis : –. Ala-Sippola, Sanna (). Agentin ilmaiseminen suomalaisella viittomakielellä tuotetussa asiatekstissa [Expressing agent in a formal FinSL narrative]. MA thesis in General Linguistics, Department of Modern Languages, University of Helsinki. Albert, Chris (). Sluicing and weak islands. Ms. UC, Santa Cruz.





Alcántara, Manuel and Nuria Bertomeu (). Ellipsis in spontaneous spoken language, in Jennifer Spenander and Petra Hendriks (eds), Proceedings of the cross-modular approaches to ellipsis. Edinburgh: Heriot-Watt University, –. Alexiadou, Artemis and Kirsten Gengel (). NP-ellipsis without focus movement/projection: The role of classifiers. Talk presented at the Workshop on Interface-Based Approaches to Information Structure, University College London. Alexiadou, Artemis and Kirsten Gengel (). NP-ellipsis without focus movement/projections: The role of classifiers, in Ivona Kučerová and Ad Neeleman (eds), Contrast and positions in information structure. Cambridge: Cambridge University Press, –. Algryani, Ali (). The syntax of ellipsis in Libyan Arabic: A generative analysis of sluicing, VP ellipsis, stripping and negative contrast. PhD thesis, Newcastle University. Algueró, Anna, Anna Gavarró, Maria Teresa Guasti, Laurice Tuller, Philippe Prévost, Adriana Belletti, Luca Cilibrasi, Hélène Delage, and Mirta Vernice (). The acquisition of partitive clitics in Romance five-year-olds, IBERIA: An International Journal of Theoretical Linguistics (): –. Almeida, Diogo A. de A. and Masaya Yoshida (). A problem for the preposition stranding generalization, Linguistic Inquiry (): –. Alonso-Ovalle, Luis (). Disjunction in alternative semantics. PhD thesis, University of Massachusetts. Alrenga, Peter, Chris Kennedy, and Jason Merchant (). A new standard of comparison, in Nathan Arnett and Ryan Bennet (eds), Proceedings of the th West Coast Conference on Formal Linguistics. Somerville, MA: Cascadilla Press, –. Amaral, Luiz, Marcus Maia, Andrew Nevins, and Tom Roeper (). Recursion across domains. Cambridge: Cambridge University Press. Amaral, Luiz and Tom Roeper (). Multiple grammars and second language representation, Second Language Research (): –. Amaral, Patricia, Craige Roberts, and E. Allyn Smith (). Review of The Logic of Conventional Implicatures, Linguistics and Philosophy : –. An, Duk-Ho (). Syntax at the PF interface: Prosodic mapping, linear order, and deletion. PhD thesis, University of Connecticut at Storrs. Anand, Pranav and James McCloskey (). Annotating the implicit content of sluices, in Adam Meyers, Ines Rehbein, and Heike Zinsmeister (eds), Proceedings of the th Linguistic Annotation Workshop (LAW-IX). Red Hook, NY: Curran Associates, –. AnderBois, Scott (). Sluicing as anaphora to issues, in Nan Li and David Lutz (eds), SALT : Proceedings of the th Semantics and Linguistic Theory Conference. Ithaca, NY: CLC Publications, –. AnderBois, Scott (). Issues and alternatives. PhD thesis, UC Santa Cruz. AnderBois, Scott (a). Focus and uninformativity in Yucatec Maya questions, Natural Language Semantics (): –. AnderBois, Scott (b). Indefiniteness and the typology of implicit arguments, in Nathan Arnett and Ryan Bennett (eds), Proceedings of the th West Coast Conference on Formal Linguistics (WCCFL ). Somerville, MA: Cascadilla Press, –. AnderBois, Scott (). The semantics of sluicing: Beyond truth-conditions, Language (): –. AnderBois, Scott, Adrian Brasoveanu, and Robert Henderson (). At-issue proposals and appositive impositions in discourse, Journal of Semantics (): –. Anderson, Stephen R. (). How to get ‘even’, Language : –. Ariel, Mira (). Accessing Noun Phrase antecedents. London and New York: Routledge. Arnold, Doug and Robert D. Borsley (). Non-restrictive relative clauses, ellipsis and anaphora, in Stefan Müller (ed.), Proceedings of the th International Conference on Head-Driven Phrase Structure Grammar. Stanford: CSLI Publications, –.





Arnold, Doug and Robert D. Borsley (). Auxiliary-stranding relative clauses, in Stefan Müller (ed.), Proceedings of the th International Conference on Head-Driven Phrase Structure Grammar. Stanford: CSLI Publications, –. Arnold, Doug and Andrew Spencer (). A constructional analysis for the skeptical, in Stefan Müller (ed.), Proceedings of the nd International Conference on Head-Driven Phrase Structure Grammar. Stanford: CSLI Publications, –. Arnon, Inbal, Bruno Estigarribia, Philip Hofmeister, Florian Jaeger, Jeanette Pettibone, Ivan A. Sag, and Neal Snider (). Rethinking superiority effects: A processing model. Poster presented at the CUNY Sentence Processing Conference, University of Arizona. Arregi, Karlos (). Ellipsis in split questions, Natural Language and Linguistic Theory (): –. Arregi, Karlos and Naiara Centeno (). Determiner sharing and cyclicity in wh-movement, in Randall Gess and Edward J. Rubin (eds), Theoretical and experimental approaches to Romance Linguistics. Amsterdam and Philadelphia: John Benjamins, –. Arregui, Ana, Jr., Charles Clifton, Lyn Frazier, and Keir Moulton (). Processing elided verb phrases with flawed antecedents: The recycling hypothesis, Journal of Memory and Language : –. Asher, Nicholas (). Reference to abstract objects in English. Dordrecht: Kluwer Academic Publishers. Asher, Nicholas, Daniel Hardt, and Joan Busquets (). Discourse parallelism, ellipsis, and ambiguity, Journal of Semantics (): –. Asher, Nicholas and Alex Lascarides (). Logics of conversation. Cambridge: Cambridge University Press. Ashton, Ethel O. (). Swahili grammar, including intonation. London: Longman. Authier, Jean-Marc (). A movement analysis of French modal ellipsis, Probus : –. Authier, Jean-Marc (). Ellipsis as movement and silence: Evidence from French, University of Pennsylvania Working Papers in Linguistics (): –. Authier, Jean-Marc (). Phase-edge features and the syntax of polarity particles, Linguistic Inquiry (): –. Babby, Leonard H. (). Case, prequantifiers, and discontinuous agreement in Russian, Natural Language and Linguistic Theory (): –. Bach, Emmon and Barbara H. Partee (). Anaphora and semantic structure, in Jody Kreiman and Almerindo E. Ojeda (eds), Papers from the Parasesion on Pronouns and Anaphora. Chicago: University of Chicago, –. Bachrach, Asaf and Roni Katzir (). Right-node raising and delayed spellout, in Kleanthes K. Grohmann (ed.), InterPhases: Phase-theoretic investigations of linguistic interfaces. Oxford: Oxford University Press, –. Bacskai-Atkari, Julia (). Parametric variation and comparative deletion, The Even Yearbook : –. Bacskai-Atkari, Julia (). The syntax of comparative constructions: Operators, ellipsis phenomena and functional left peripheries. PhD dissertation, Universität Potsdam. Bahan, Benjamin, Judy Kegl, Robert G. Lee, Dawn MacLaughlin, and Carol Neidle (). The licensing of null arguments in American Sign Language, Linguistic Inquiry (): –. Bailyn, John Frederick (). A configurational approach to Russian ‘free’ word order. PhD thesis, Cornell University. Bailyn, John Frederick (). Generalized inversion, Natural Language and Linguistic Theory (): –. Bailyn, John Frederick (). Kak tebe lingvistika? Nenavižu! How (not) to analyze Russian verb-stranding constructions. Paper presented at Formal Approaches to Slavic Linguistics . MIT, Cambridge, MA. Bailyn, John Frederick (). The syntax of Russian. Cambridge: Cambridge University Press.





Bailyn, John Frederick (). The case of Q, in Olga Arnaudova, Wayles Browne, Maria Luisa Rivero, and Danijela Stojanović (eds), Formal Approaches to Slavic Linguistics : The Ottawa Meeting. Ann Arbor, MI: Michigan Slavic Publications, –. Bailyn, John Frederick (). Against a VP ellipsis account of Russian verb-stranding constructions, in Alexander Vovin (ed.), Studies in Japanese and Korean linguistics and beyond. Boston: Brill, –. Bailyn, John Frederick and Tanya Scott (). Asymmetries in Russian multiple sluicing: A purely syntactic account. Abstract for a talk at FASL (Formal Approaches to Slavic Linguistics) . Baker, C. L. (). Two observations on British English do, Linguistic Inquiry (): –. Baker, C. L. and Michael Brame (). ‘Global rules’: A rejoinder, Language (): –. Baker, Mark, Kyle Johnson, and Ian Roberts (). Passive arguments raised, Linguistic Inquiry (): –. Bakker, Dik (). Samentrekking in Nederlandse syntactische groepen. PhD thesis, Leiden University. Baltin, Mark (). Negation and clause structure. Ms. New York University. Baltin, Mark (). The position of adverbials. Ms. New York University. Baltin, Mark (). The nonunity of VP-preposing, Language (): –. Baltin, Mark (). Deletion versus pro-forms: A false dichotomy? Ms. New York University. Baltin, Mark (). Deletion versus pro-forms: An overly simple dichotomy? Natural Language and Linguistic Theory : –. Banfield, Ann (). Stylistic deletion in coordinate structures, Linguistic Analysis : –. Bánréti, Zoltán (). A mellérendelés, in Ferenc Kiefer (ed.), Strukturális magyar nyelvtan. Mondattan. Budapest: Akadémiai Kiadó, –. Bánréti, Zoltán (). Multiple lexical selection and parallelism in Hungarian VP-ellipsis, Acta Linguistica Hungarica : –. Bánréti, Zoltán (). A mellérendelés és az ellipszis nyelvtana a magyarban. Budapest: Tinta Könyvki-adó. Barbiers, Sjef (). The right-periphery in SOV-languages, in Peter Svenonius (ed.), The derivation of VO and OV. Amsterdam and Philadelphia: John Benjamins, –. Barker, Chris (). Partitives, double genitives and anti-uniqueness, Natural Language and Linguistic Theory (): –. Barker, Chris (). Parasitic scope. Linguistics and Philosophy . : –. Barker, Chris (). Scopability and sluicing, Linguistics and Philosophy : –. Barros, Matthew (a). Sluiced fragment answers: Another puzzle involving islands and ellipsis, Snippets : –. Barros, Matthew (b). Else-modification as a diagnosis for pseudosluicing. Ms. Rutgers University. Barros, Matthew (). Harmonic sluicing: Which remnant/correlate pairs work and why, in SALT : Proceedings of the rd Semantics and Linguistic Theory Conference. Ithaca, NY: CLC Publications, –. Barros, Matthew (a). A non-repair approach to island sensitivity in contrastive TP ellipsis, in Proceedings from the th annual meeting of the Chicago Linguistic Society. Chicago: Chicago Linguistic Society, –. Barros, Matthew (b). Sluicing and identity in ellipsis. PhD thesis, Rutgers University, New Brunswick, NJ. Barros, Matthew and Jeroen van Craenenbroeck (). Tag questions and ellipsis. Paper presented at Deutsche Gesellschaft für Sprachwissenschaft, Theme session: Parentheses and ellipsis: Crosslinguistic and theoretical perspectives. Barros, Matthew, Patrick Elliott, and Gary Thoms (). There is no island repair. Ms., Rutgers University, University College London, and University of Edinburgh. Barros, Matthew, Patrick Elliott, and Gary Thoms (). More variation in island repair: The clausal/non-clausal island distinction, in Proceedings of the th annual meeting of the Chicago Linguistic Society, Chicago: Chicago Linguistic Society, ch. .





Barros, Matthew and Luis Vicente (). Right node raising requires both ellipsis and multidomination, University of Pennsylvania Working Papers in Linguistics (): –. Barros, Matthew and Luis Vicente (). A remnant condition for ellipsis, in Kyeong-min Kim, Pocholo Umbal, Trevor Block, Queenie Chan, Tanie Cheng, Kelli Finney, Mara Katz, Sophie Nickel-Thompson, and Lisa Shorten (eds), Proceedings of the rd West Coast Conference on Formal Linguistics. Somerville, MA: Cascadilla Proceedings Project, –. Barss, Andrew (). Chains and anaphoric dependence: On reconstruction and its implications. PhD thesis, Massachusetts Institute of Technology. Bartlett, Laura and Kay González-Vilbazo (). Classifiers in Spanish-Taiwanese code-switching, in Proceedings of the th annual meeting of the Chicago Linguistic Society. Chicago: Chicago Linguistic Society, –. Bartlett, Laura and Kay González-Vilbazo (). The structure of the Taiwanese DP in Taiwanese– Spanish bilinguals: Evidence from code-switching, Journal of East Asian Linguistics (): –. Barton, Ellen (). Nonsentential constituents: A theory of grammatical structure and pragmatic interpretation. Amsterdam and Philadelphia: John Benjamins. Barton, Ellen (). Toward a nonsentential analysis in generative grammar, in Ljiljana Progovac, Kate Paesani, Eugenia Casielles, and Ellen Barton (eds), The syntax of nonsententials: Multidisciplinary perspectives. Amsterdam and Philadelphia: John Benjamins, –. Barton, Ellen and Ljiljana Progovac (). Nonsententials in Minimalism, in Reinaldo Elugardo and Robert J. Stainton (eds), Ellipsis and nonsentential speech. Amsterdam and Philadelphia: Springer, –. Bartos, Huba (a). Az inflexiós jelenségek szintaktikai háttere, in Ferenc Kiefer (ed.), Strukturális magyar nyelvtan : Morfológia. Budapest: Akadémiai Kiadó, –. Bartos, Huba (b). VP-ellipsis and verbal inflection in Hungarian, Acta Linguistica Hungarica : –. Bartos, Huba (a). A mutató névmási módosítók a magyarban: Egyeztetés vagy osztozás? in Marianne Barkó-Nagy, Zoltán Bánréti, and Katalin É. Kiss (eds), Újabb tanulmányok a strukturális magyar nyelvtan és a nyelvtörtenet köréből. Budapest: Osiris, –. Bartos, Huba (b). Sound–form non-insertion and the direction of ellipsis, Acta Linguistica Hungarica (): –. Bartsch, Renate and Theo Vennemann (). Semantic structures. Frankfurt am Main: Athenaeum. Barwise, John and John Etchemendy (). The liar. New York: Oxford University Press. Barwise, John and John Perry (). Situations and attitudes. Cambridge, MA: MIT Press. Bassong, Paul Roger (). Information structure and the Basa’a left peripheral. PhD thesis, University of Yaounde I. Beaver, David I. (). The optimization of discourse anaphora, Linguistics and Philosophy (): –. Beaver, David I. and Brady C. Clark (). Sense and sensitivity: How focus determines meaning. Malden, MA: Wiley-Blackwell. Beavers, John and Ivan A. Sag (). Coordinate ellipsis and apparent non-constituent coordination, in Stefan Müller (ed.), Proceedings of the th International Conference on Head-Driven Phrase Structure Grammar. Stanford: CSLI Publications, –. Bech, Gunnar (). Studien über das deutsche Verbum infinitum. Tübingen: M. Niemeyer. Beck, Sigrid (). Comparison constructions, in Claudia Maienborn, Klaus von Heusinger, and Paul Portner (eds), Semantics: An international handbook of natural language meaning. Berlin and New York: Mouton de Gruyter, –. Beck, Sigrid, Svetlana Krasikova, Daniel Fleischer, Remus Gergel, Christiane Savelsberg, John Vanderelst, and Elisabeth Villalta (). Crosslinguistic variation in comparison constructions, Linguistic Variation Yearbook : –. Beck, Sigrid and Uli Sauerland (). Cumulation is needed: A reply to Winter (), Natural Language Semantics (): –.





Beckman, Mary E. (). The parsing of prosody, Language and Cognitive Processes : –. Beckman, Mary E. and Gayle E. Elam (). Guidelines for ToBI labeling, Version . Ms. Ohio State University. Beckman, Mary E., Julia Hirschberg, and Stefanie Shattuck-Hufnagel (). The original ToBI system and the evolution of the ToBI framework, in Sun-Ah Jun (ed.), Prosodic typology: The phonology of intonation and phrasing. Oxford: Oxford University Press, –. Beckman, Mary E. and Janet B. Pierrehumbert (). Intonational structure in Japanese and English, Phonology Yearbook : –. Beckman, Mary E. and Jennifer J. Venditti (). Intonation, in John Goldsmith, Jason Riggle, and Alan C. L. Yu (eds), Handbook of Phonological Theory. Malden, MA: Blackwell Publishing, –. Beecher, Henry (). Pragmatic inference in the interpretation of sluiced prepositional phrases, San Diego Linguistics Papers : –. Belazi, Hedi M., Edward J. Rubin, and Almeida Jacqueline Toribio (). Code switching and X-bar theory: The functional head constraint, Linguistic Inquiry (): –. Belletti, Adriana (). Aspects of the low IP area, in Luigi Rizzi (ed.), The Structure of CP and IP. Oxford: Oxford University Press, –. Belletti, Adriana and Ur Shlonsky (). The order of verbal complements: A comparative study, Natural Language and Linguistic Theory (): –. Bennett, Ryan, Emily Elfner, and James McCloskey (). Prosody, focus, and ellipsis in Irish. Ms. Yale University, University of British Columbia, and University of California, Santa Cruz. Bennis, Hans (). Het kwantitatieve er in komparatiefconstructies, Spektator : –. Bennis, Hans (). Comparative deletion is subdeletion. Master’s thesis, University of Amsterdam. Bennis, Hans (). Gaps and dummies. Dordrecht: Foris Publications. Bentzen, Kristine, Jason Merchant, and Peter Svenonius (). Deep properties of surface pronouns: Pronominal predicate anaphors in Norwegian and German, Journal of Comparative Germanic Linguistics (): –. Bergsma, Shane, David Yarowsky, and Kenneth Church (). Using large monolingual and bilingual corpora to improve coordination disambiguation, in Proceedings of the th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies, Volume . Stroudsburg, PA: Association for Computational Linguistics, –. Bernolet, Sarah, Robert Hartsuiker, and Simona Collina (). The persistence of syntactic priming revisited, Journal of Memory and Language : –. Bernstein, Judy (). The syntactic role of word markers in null nominal constructions, Probus : –. Bertomeu, Nuria and Valia Kordoni (). Integrating pragmatic information in grammar: An analysis of intersentential ellipsis, in Stefan Müller (ed.), Proceedings of the th International Conference on Head-Driven Phrase Structure Grammar. Stanford: CSLI Publications, –. den Besten, Hans (). On the presence and absence of wh-elements in Dutch comparatives, Linguistic Inquiry (): –. de Beuzeville, Louise, Trevor Johnston, and Adam Schembri (). The use of space with indicating verbs in Auslan: A corpus-based investigation, Sign Language and Linguistics : –. Bever, Thomas G. (). The cognitive basis for linguistic structure, in John R. Hayes (ed.), Cognition and the development of language. New York: John Wiley, –. Bezuidenhout, Anne (). Structuring silence versus the structure of silance, in Laurence Goldstein (ed.), Brevity. Oxford: Oxford University Press, –. Bhat, D. N. Shankar (). The indefinite-interrogative puzzle, Linguistic Typology : –. Bhatt, Rajesh (). Relative clauses and correlatives, in Artemis Alexiadou and Tibor Kiss (eds), Syntax: Theory and Analysis. An International Handbook. Berlin: Mouton de Gruyter, –. Bhatt, Rajesh and Roumyana Pancheva (). Late merger of degree clauses, Linguistic Inquiry (): –.





Bhatt, Rajesh and Shoichi Takahashi (). Direct comparisons: Resurrecting the direct analysis of phrasal comparatives, in Masayuki Gibson and Tova Friedman (eds), SALT : Proceedings of the th Semantics and Linguistic Theory Conference. Ithaca, NY: CLC Publications, –. Bhatt, Rajesh and Shoichi Takahashi (). Reduced and unreduced phrasal comparatives, Natural Language and Linguistic Theory (): –. Bhattacharya, Tanmoy and Andrew Simpson (). Sluicing in Indo-Aryan: An investigation of Bangla and Hindi, in Jason Merchant and Andrew Simpson (eds), Sluicing: Cross-linguistic perspectives. Oxford: Oxford University Press, –. Bianchi, Valentina (). Resumptive relatives and LF chains, in Luigi Rizzi (ed.), The Structure of CP and IP. Oxford: Oxford University Press, –. Biezma, Maria (). Multiple focus strategies in pro-drop languages: Evidence from ellipsis in Spanish, Syntax : –. Bîlbîie, Gabriela (). Grammaire des constructions elliptiques: Une étude comparative des phrases sans verbe en roumain et en français. PhD thesis, Université Paris Diderot. Bîlbîie, Gabriela (). A quantitative study on right node raising in the Penn treebank. Presentation at the International Congress of Linguists, Geneva. Bjorkman, Bronwyn (). BE-ing default: The morphosyntax of auxiliaries. PhD thesis, MIT. Blackburn, Patrick and Wilfried Meyer-Viol (). Linguistics, logic and finite trees, Logic Journal of the Interest Group of Pure and Applied Logics (): –. Blom, A. (). Against conjunction reduction, in A. Kraak (ed.), Linguistics in the Netherlands –. Amsterdam: Van Gorcum, –. Blom, Elma, Daniela Polišenská, and Fred Weerman (). Articles, adjectives and age of onset: The acquisition of Dutch grammatical gender, Second Language Research (): –. Boas, H. C. (ed.) (). Contrastive studies in construction grammar, vol. . Amsterdam: John Benjamins. Bobaljik, Jonathan and Cynthia Zocca (). Gender markedness: The anatomy of a counterexample, Morphology : –. Bochnak, Michael (). Cross-linguistic variation in the semantics of comparatives. PhD dissertation, University of Chicago. Bock, Kathryn and Zenzi M. Griffin (). The persistence of structural priming: Transient activation or implicit learning? Journal of Experimental Psychology (): –. Bolinger, Dwight (). Asking more than one thing at a time, in Henry Hiz (ed.), Questions. Dordrecht: D. Reidel, –. Bonami, Olivier and Danièle Godard (). On the syntax of direct quotation in French, in Stefan Müller (ed.), Proceedings of the th International Conference on Head-Driven Phrase Structure Grammar. Stanford: CSLI Publications, –. de Bont, A. P. (). Dialekt van Kempenland. Meer in het bijzonder d’Oerse taol. Deel . Klank- en vormleer en enige syntactische bijzonderheden. Assen: Van Gorcum. Booij, Geert (). Coordination reduction in complex words: A case for prosodic phonology, in Harry van der Hulst and N. Smith (eds), Advances in Nonlinear Phonology. Dordrecht: Foris Publications, –. Boone, Enrico (). The syntax and licensing of gapping and fragments. PhD thesis, Leiden University. Borsley, Robert D. (). Against ConjP, Lingua : –. Bos, Johan and Jennifer Spenader (). An annotated corpus for the analysis of VP ellipsis, Language Resources and Evaluation (): –. Bošković, Željko (). D-structure, Theta-Criterion and movement into theta-positions, Linguistic Analysis : –. Bošković, Željko (). Fronting wh-phrases in Serbo-Croatian, in Martina Lindseth and Steven Franks (eds), Formal Approaches to Slavic Linguistics: The Indiana meeting. Ann Arbor, MI: Michigan Slavic Publications, –.





Bošković, Željko (). Wh-movement and wh-phrases in Slavic. Position paper presented at the Comparative Slavic Morphosyntax Workshop, Spencer, IN. Bošković, Željko (). On multiple feature checking: Multiple wh-fronting and multiple head movement, in Samuel Epstein and Norbert Hornstein (eds), Working minimalism. Cambridge, MA: MIT Press, –. Bošković, Željko (). Sometimes in [Spec, CP], sometimes in situ, in Roger Martin, David Michaels, and Juan Uriagereka (eds), Step by step. Cambridge, MA: MIT Press, –. Bošković, Željko (). On multiple wh-fronting, Linguistic Inquiry (): –. Bošković, Željko (). On the interpretation of multiple questions, Linguistic Variation Yearbook : –. Bošković, Željko (a). Be careful where you float your quantifiers, Natural Language and Linguistic Theory (): –. Bošković, Željko (b). Two notes on Right Node Raising, University of Connecticut Working Papers in Linguistics : –. Bošković, Željko (). What will you have, DP or NP? in Emily Elfner and Martin Walkow (eds), NELS : Proceedings of the th annual meeting of the North East Linguistic Society. Amherst, MA: GLSA, –. Bošković, Željko (). Rescue by PF deletion, traces as (non)interveners, and the that-trace effect, Linguistic Inquiry (): –. Bošković, Željko (a). Adjectival escapades, in Steven Franks, Markus Dickinson, George Fowler, Melissa Witcombe, and Ksenia Zanon (eds), Proceedings of Formal Approaches to Slavic Linguistics . Ann Arbor, MI: Michigan Slavic Publications, –. Bošković, Željko (b). Traces do not head islands: What can deletion rescue? in Yoichi Miyamoto, Daiko Takahashi, Hideki Maki, Masao Ochi, Koji Sugisaki, and Asako Uchibori (eds), Deep insights, broad perspectives: Essays in honor of Mamuro Saito. Tokyo: Kaitakusha, –. Bošković, Željko (). Now I’m a phase, now I’m not a phase: On the variability of phases with extraction and ellipsis, Linguistic Inquiry (): –. Bošković, Željko (). On pronouns, clitic doubling and argument ellipsis: Argument ellipsis as predicate ellipsis. Ms. University of Connecticut, Storrs. Bošković, Željko (). On clitic doubling and argument ellipsis, in Aaron Kaplan, Abby Kaplan, Miranda K. McCarvel, and Edward J. Rubin (eds), Proceedings of the th West Coast Conference on Formal Linguistics. Somerville, MA: Cascadilla Proceedings Project, –. Bošković, Željko and Steven Franks (). Across-the-board movement and LF, Syntax (): –. Bošković, Željko and Daiko Takahashi (). Scrambling and last resort, Linguistic Inquiry (): –. Bosque, Ignacio (). Negación y ellipsis, Estudios de Lingüística : –. Bouchard, Denis (). Adjectives, number and interfaces: Why languages vary. Oxford: Elsevier Science. Bouton, L. F. (). Antecedent contained pro-forms, in Papers from the Sixth Regional Meeting of the Chicago Linguistic Society. Chicago: Chicago Linguistic Society, –. Bowers, John (). Transitivity, Linguistic Inquiry (): –. Brame, Michael (). Ungrammatical notes : Smarter than me, Linguistic Analysis (): –. Branco, António and Francisco Costa (). Noun ellipsis without empty categories, in Stefan Müller (ed.), Proceedings of the th International Conference on Head-Driven Phrase Structure Grammar. Stanford: CSLI Publications, –. Bresnan, Joan (). A note on the notion ‘identity of sense anaphora’, Linguistic Inquiry (): –. Bresnan, Joan (). Syntax of the comparative clause construction in English, Linguistic Inquiry (): –. Bresnan, Joan (). The position of certain clause-particles in phrase structure, Linguistic Inquiry (): –.





Bresnan, Joan (). Comparative deletion and constraints on transformations, Linguistic Analysis (): –. Bresnan, Joan (). On the form and functioning of transformations, Linguistic Inquiry (): –. Bresnan, Joan and Sam A. Mchombo (). The lexical integrity principle: Evidence from Bantu, Natural Language and Linguistic Theory (): –. Brody, Mihály (). Remarks on the order of elements in the Hungarian focus field, in István Kenesei (ed.), Approaches to Hungarian . Szeged: JATE Press, –. Brody, Mihály (). Focus and checking theory, in István Kenesei (ed.), Levels and structures. Approaches to Hungarian . Szeged: JATE Press, –. Broekhuis, Hans (). Syntax of Dutch: Adjectives and adjective phrases. Amsterdam: Amsterdam University Press. Broekhuis, Hans and Norbert Corver (). Syntax of Dutch: Verbs and verb phrases, vol. II. Amsterdam: Amsterdam University Press. Brown, Roger (). How shall a thing be called? Psychological Review (): –. Brown-Schmidt, Sarah, Christine Gunlogson, and Michael K. Tanenhaus (). Addressees distinguish shared from private information when interpreting questions during interactive conversation, Cognition (): –. Brucart, José María (). La elisión sintáctica en español. Bellaterra: Publicacions de la Universitat Autonoma de Barcelona. Brucart, José María (). La ellipsis, in Ignacio Bosque and Violeta Demonte (eds), Gramática descriptiva de la lengua española. Madrid: Espasa Calpe, –. Brucart, José María and Jonathan E. MacDonald (). Empty categories and ellipsis, in José Ignacio Hualde, Antxon Olarrea, and Erin O’Rourke (eds), The handbook of Hispanic linguistics. Malden, MA: John Wiley & Sons. Bruening, Benjamin (). Passive do so. Ms. University of Delaware. Brugos, Alenja, Stefanie Shattuck-Hufnagel, and Nanette Veilleux (). Transcribing prosodic structure of spoken utterances with ToBI. Tutorial through MIT OpenCourseWare. Bruner, Jerome Seymour (). On perceptual readiness, Psychological Review : –. Buell, Leston (). Issues in Zulu verbal morphosyntax. PhD thesis, UCLA. Büring, Daniel (). On D-trees, beans, and B-accents, Linguistics and Philosophy : –. Büring, Daniel (). Binding theory. Cambridge: Cambridge University Press. Büring, Daniel and Katharina Hartmann (). Asymmetrische Koordination, Linguistische Berichte : –. Burton, Strang and Jane Grimshaw (). Coordination and VP-internal subjects, Linguistic Inquiry (): –. Busquets, Joan and Pascal Denis (). L’ellipse modale en français: Le cas de pouvoir et devoir, Cahiers de Grammaire : –. Bybee, Joan, Revere Perkins, and William Pagliuca (). The evolution of grammar: Tense, aspect, and modality in the languages of the world. Chicago: University of Chicago Press. Byrant, Doreen (). Koordinationsellipsen im Spracherwerb: Die Verarbitung potentieller GappingStrukturen. Berlin: Walter de Gruyter. Cabredo Hofherr, Patricia (). Pronouns, determiners and N-ellipsis in Spanish, French, and German, in Christopher Davis, Amy Rose Deal, and Youri Zabbal (eds), NELS : Proceedings of the th annual meeting of the North East Linguistic Society. Amherst, MA: GLSA, –. Caha, Pavel (). The nanosyntax of case. PhD thesis, University of Tromsø. Caha, Pavel (). Case in adpositional phrases. Ms. CASTL, Tromsø. Cai, Zhenguang G., Martin J. Pickering, and Patrick Sturt (). Processing verb-phrase ellipsis in Mandarin Chinese: Evidence against the syntactic account, Language and Cognitive Processes : –.





Camacho, José (). The structure of coordination: Conjunction and agreement phenomena in Spanish and other languages. Dordrecht: Kluwer Academic Publishers. Cann, Ronnie (). Towards an account of the English auxiliary system: Building interpretations incrementally, in Ruth Kempson, Eleni Gregoromichelaki, and Christine Howes (eds), Dynamics of lexical interfaces. Chicago: CSLI Press, –. Cann, Ronnie, Ruth Kempson, and Lutz Marten (). The dynamics of language. Oxford: Elsevier. Cann, Ronnie, Ruth Kempson, and Matthew Purver (). Context and well-formedness: The dynamics of ellipsis, Research on Language and Computation (): –. Cantone, Katja Francesca (). Code-switching in bilingual children. Berlin: Springer. Cantor, Sara (). Ungrammatical double-island sluicing as a diagnostic of left-branch positioning. Master’s thesis, University of California at Santa Cruz. Caplan, David and Gloria S. Waters (). Verbal working memory and sentence comprehension, Behavioral and Brain Sciences : –. Caponigro, Ivano and Kathryn Davidson (). Ask and tell as well: Clausal question-answer pairs in ASL, Natural Language Semantics : –. Caramazza, Alfonso (). The logic of neuropsychological research and the problem of patient classification in aphasia, Brain and Language : –. Caramazza, Alfonso and Edgar Zurif (). Dissociations of algorithmic and heuristic processes in sentence comprehension: Evidence from aphasia, Brain and Language : –. Carberry, Sandra (). A pragmatics-based approach to ellipsis resolution, Computational Linguistics (): –. Cardinaletti, Anna (). On the internal structure of pronominal DPs, The Linguistic Review (–): –. Carlson, Greg (). Same and different: Some consequences for syntax and semantics, Linguistics and Philosophy (): –. Carlson, Katy (). The effects of parallelism and prosody on the processing of gapping structures, Language and Speech : –. Carlson, Katy (). Parallelism and prosody in the processing of ellipsis sentences. New York: Routledge. Carlson, Katy, Michael Walsh Dickey, Lyn Frazier, and Charles Clifton Jr (a). Information structure expectations in sentence comprehension, Quarterly Journal of Experimental Psychology : –. Carlson, Katy, Michael Walsh Dickey, and Christopher Kennedy (). Structural economy in the processing and representation of gapping sentences, Syntax : –. Carlson, Katy, Lyn Frazier, and Charles Clifton, Jr. (b). How prosody constrains comprehension: A limited effect of prosodic packaging, Lingua (): –. Carroll, John M., Thomas G. Bever, and Chava R. Pollack (). The non-uniqueness of linguistic intuitions, Language (): –. Casielles, Eugenia (). Big questions, small answers, in Ljiljana Progovac, Kate Paesani, Eugenia Casielles, and Ellen Barton (eds), The syntax of nonsententials: Multidisciplinary perspectives. Amsterdam and Philadelphia: John Benjamins, –. Cassimjee, Farida and Charles W. Kisseberth (). Optimal Domains Theory and Bantu tonology, in Larry M. Hyman and Charles W. Kisseberth (eds), Theoretical aspects of Bantu tone. Stanford: CSLI, –. Cassimjee, Farida and Charles W. Kisseberth (in preparation). The Shingazidja lexicon exemplified. Ms. Columbia University and Tel Aviv University. Cattell, Ray (). Constraints on movement rules, Language : –. Cecchetto, Carlo, Alessandra Cecchetto, Carlo Geraci, Mirko Santoro, and Sandro Zucchi (). The syntax of predicate ellipsis in Italian Sign Language (LIS), Lingua : –.





Cecchetto, Carlo and Orin Percus (). When we do that and when we don’t: A contrastive analysis of VP-ellipsis and VP-anaphora, in Mara Frascarelli (ed.), Phases of interpretation. Berlin and New York: Mouton de Gruyter, –. Centeno, Naiara (). Gapping and determiner sharing in Spanish. PhD thesis, University of the Basque Country, Vitoria-Gasteiz. Chae, Jeongmin, Younghee Jung, Taemin Lee, Soonyoung Jung, Chan Huh, Gilhan Kim, Hyeoncheoi Kim, and Heungbum Oh (). Identifying non-elliptical entity mentions in a coordinated NP with ellipses, Journal of Biomedical Informatics : –. Chafe, Wallace L. (). Givenness, contrastiveness, definiteness, subjects, topics, and point of view, in Charles N. Li (ed.), Subject and topic. New York: Academic Press, –. Chalcraft, Faye (). Do-doubling in West Yorkshire English. Paper presented at the workshop Syntactic Doubling in European Dialects. Chao, Wynn (). On ellipsis. New York and London: Garland Press. Charlow, Simon (). Free and bound pro-verbs: A unified treatment of anaphora, in Tova Friedman and Satoshi Ito (eds), SALT : Proceedings of the Eighth Annual Conference on Semantics and Linguistic Theory. Ithaca, NY: CLC Publications, –. Charlow, Simon (). Cross-categorial donkeys, in Maria Aloni, Vadim Kimmelman, Floris Roelofsen, Galit W. Sassoon, Katrin Schulz, and Matthijs Westera (eds), Logic, language, and meaning. Berlin: Springer, –. Charlow, Simon (). On the semantics of exceptional scope. PhD thesis, NYU. Chaves, Rui P. (). Coordination of unlikes without inlike categories, in Stefan Müller (ed.), Proceedings of the th International Conference on Head-Driven Phrase Structure Grammar. Stanford: CSLI Publications, –. Chaves, Rui P. (). Coordinate structures: constraint-based syntax-semantics. PhD thesis, University of Lisbon. Chaves, Rui P. (). Linearization-based word-part ellipsis, Linguistics and Philosophy : –. Chaves, Rui P. (). A linearization-based approach to gapping, in James Rogers (ed.), Proceedings of FG-MoL  The th Conference on Formal Grammar and the th Meeting on Mathematics of Language. Stanford: CSLI Publications, –. Chaves, Rui P. (). Conjunction, cumulation and respectively readings, Journal of Linguistics : –. Chaves, Rui P. (). On the disunity of right-node phenomena: Extraposition, ellipsis, and deletion, Language : –. Chien, Yu-Chin and Kenneth Wexler (). Children’s knowledge of locality conditions as evidence for the modularity of syntax and pragmatics, Language Acquisition : –. Chomsky, Noam (). The logical structure of linguistic theory. Ms. Harvard University and MIT. (Revised  version published in part by New York: Plenum, ; Chicago: University of Chicago Press, ). Chomsky, Noam (). Syntactic structures. The Hague: Mouton & Co. Chomsky, Noam (). Current issues in linguistic theory. The Hague: Mouton & Co. Chomsky, Noam (). Aspects of the theory of syntax. Cambridge, MA: MIT Press. Chomsky, Noam (). Deep structure, surface structure and semantic interpretaion, in Danny Steinberg and Leon Jacobovits (eds), Semantics. London: Cambridge University Press, –. Chomsky, Noam (a). Some empirical issues in the theory of transformational grammar, in Stanley Peters (ed.), The goals of linguistic theory. Englewood Cliffs, NJ: Prentice-Hall, –. Chomsky, Noam (b). Studies on semantics in generative grammar. The Hague: Mouton. Chomsky, Noam (). Conditions on transformations, in Stephen Anderson and Paul Kiparsky (eds), Festschrift for Morris Halle. London: Rinehart and Winston, –. Chomsky, Noam (a). Conditions on transformations, in Essays on form and interpretation. New York: Elsevier North-Holland, –.





Chomsky, Noam (b). On wh-movement, in P. Culicover, T. Wasow, and A. Akmajian (eds), Formal syntax. New York: Academic Press, –. Chomsky, Noam (). Lectures on government and binding. Dordrecht: Foris Publications. Chomsky, Noam (). Some concepts and consequences of the theory of government and binding. Cambridge, MA: MIT Press. Chomsky, Noam (). Barriers. Cambridge, MA: MIT Press. Chomsky, Noam (). Some notes on economy of derivation and representation, in Robert Freidin (ed.), Principles and parameters in comparative grammar. Cambridge, MA: MIT Press, –. Chomsky, Noam (a). Bare phrase structure, in Gert Webelhuth (ed.), Government binding theory and the minimalist program. Oxford: Oxford University Press, –. Chomsky, Noam (b). The minimalist program. Cambridge, MA: MIT Press. Chomsky, Noam (). Minimalist inquiries: The framework, in Roger Martin, David Michaels, and Juan Uriagereka (eds), Step by step: Essays on minimalist syntax in honor of Howard Lasnik. Cambridge, MA: MIT Press, –. Chomsky, Noam (). Derivation by phase, in Michael Kenstowicz (ed.), Ken Hale: A Life in Linguistics. Cambridge, MA: MIT Press, –. Chomsky, Noam (). Beyond explanatory adequacy, in Adriana Belletti (ed.), Structures and beyond: The cartography of syntactic structures, vol. . Oxford: Oxford University Press, –. Chomsky, Noam (). Three factors in language design, Linguistic Inquiry (): –. Chomsky, Noam (). On phases, in Robert Freidin, Carlos Otero, and Maria Luisa Zubizarreta (eds), Foundational issues in linguistic theory. Cambridge, MA: MIT Press, –. Chomsky, Noam (). Problems of projection, Lingua : –. Chomsky, Noam (). Minimal recursion: Exploring the prospects, in Tom Roeper and Margaret Speas (eds), Recursion: Complexity in cognition. Berlin: Springer, –. Chomsky, Noam and Howard Lasnik (). The theory of Principles and Parameters, in Arnim von Stechow, W. Sternefeld and T. Vennemann (eds), Syntax: An international handbook of contemporary research. Berlin: Walter de Gruyter, –. Chung, Sandra (). Sluicing and the lexicon: The point of no return, in Rebecca T. Cover and Yuni Kim (eds), Proceedings of the annual meeting of the Berkeley Linguistics Society, vol. . Berkeley, CA: Berkeley Linguistics Society, –. Chung, Sandra (). Syntactic identity in sluicing: How much and why, Linguistic Inquiry (): –. Chung, Sandra, William A. Ladusaw, and James McCloskey (). Sluicing and Logical Form, Natural Language Semantics (): –. Chung, Sandra, William A. Ladusaw and James McCloskey (). Sluicing(:) Between structure and inference, in Rodrígo Gutiérrez-Bravo, Line Mikkelsen, and Eric Potsdam (eds), Representing language: Essays in honor of Judith Aissen. Santa Cruz: Linguistics Research Center, –. Ciardelli, Ivano (). Inquisitive semantics and intermediate logics. Master’s thesis, University of Amsterdam. Ciardelli, Ivano, Jeroen Groenendijk, and Floris Roelofsen (). Attention! Might in Inquisitive Semantics, in Ed Cormany, Satoshi Ito, and David Lutz (eds), SALT : Proceedings of the th Semantics and Linguistic Theory Conference. Ithaca, NY: CLC Publications, –. Ciardelli, Ivano, Jeroen Groenendijk, and Floris Roelofsen (). Inquisitive semantics: A new notion of meaning, Language and Linguistics Compass (): –. Cinque, Guglielmo (). Types of A0 -dependencies. Cambridge, MA: MIT Press. Cinque, Guglielmo (). ‘Restructuring’ and functional structure, in Adriana Belletti (ed.), Structures and beyond: The cartography of syntactic structures, vol. . Oxford: Oxford University Press, –. Cinque, Guglielmo (). Deriving Greenberg’s universal  and its exceptions, Linguistic Inquiry (): –. Cinque, Guglielmo (). Restructuring and functional heads: The cartography of syntactic structures, vol. . New York: Oxford University Press.





Citko, Barbara (). On the nature of merge: External merge, internal merge, and parallel merge, Linguistic Inquiry (): –. Citko, Barbara (). Determiner sharing from a crosslinguistic perspective, in Pierre Pica (ed.), Linguistic variation yearbook . Amsterdam and Philadelphia: John Benjamins, –. Citko, Barbara (). Symmetry in syntax. Cambridge: Cambridge University Press. Citko, Barbara (). To gap or to Right Node Raise. Presentation at Formal Approaches to Slavic Linguistics , New York. Citko, Barbara and Martina Gračanin-Yuksek (). Towards a new typology of coordinated wh-questions, Journal of Linguistics (): –. Clark, Herbert (). Using language. Cambridge: Cambridge University Press. Clark, Herbert and Jean Fox Tree (). Using uh and um in spontaneous speaking, Cognition : –. Clifton, Charles, Jr (). The (non)effect of discourse relations on comprehending verb phrase ellipsis. Paper presented at Architectures and Mechanisms of Language Processing. Clifton, Charles, Jr and Lyn Frazier (). Discourse integration guided by Question Under Discussion, Cognitive Psychology : –. Clifton, Charles, Jr and Lyn Frazier (). Focus in corrective exchanges: Effects of pitch accent and syntactic form, Language and Speech (): –. Clifton, Charles, Jr and Lyn Frazier (in progress). Direct and indirect answers to questions. Ms. University of Massachusetts at Amherst. Colbert, Stephen (). I am America (and so can you!). New York: Grand Central Publishing. Cole, Peter, Gabriella Hermon, and Norhaida Aman (to appear). Clefted questions in Malay, in David Gil and James Collins (eds), Malay/Indonesian linguistics. London: Curzon Press. Collins, James N. (). The discourse potential of narrow scope indefinites in Samoan. Ms. Stanford. Collins, James N., Daria Popova, Ivan A. Sag, and Thomas Wasow (). Sluicing and the inquisitive potential of appositives. . Conner, Tracy (). Heads must be overt: Overtness and ellipsis licensing, in Marlies Kluck and Mark de Vries (eds), Parenthesis and ellipsis. Berlin: Mouton de Gruyter, –. Conner, Tracy (). Ellipsis licensing: Testing overtness. Ms. University of Massachusetts at Amherst. Conroy, Anastasia, Jeffrey Lidz, and Julien Musolino (). The Fleeting Isomorphism Effect, Language Acquisition : –. Cooper, Robin (). Austinian truth, attitudes and type theory, Research on Language and Computation (): –. Cooper, Robin (). Type theory and semantics in flux, in Ruth Kempson, Nicholas Asher, and Tim Fernando (eds), Handbook of the philosophy of science. Amsterdam: Elsevier, –. Cooper, Robin and Jonathan Ginzburg (). Type theory with records for natural language semantics, in Chris Fox and Shalom Lappin (eds), Handbook of Contemporary Semantic Theory. Oxford: Blackwell, –. Coppen, Peter-Arno (). Specifying the Noun Phrase. PhD thesis, Catholic University Nijmegen. Coppock, Elizabeth (). Gapping: In defense of deletion, in Mary Andronis, Christoper Ball, Heidi Elston, and Sylvain Neuvel (eds), Proceedings of the th annual meeting of the Chicago Linguistic Society. Chicago: Chicago Linguistic Society, –. Coquand, Thierry, Randy Pollack, and Makoto Takeyama (). A logical framework with dependent types, Fundamenta Informaticae : –. Corbett, Greville (). Universals in the syntax of cardinal numbers, Lingua : –. Corblin, Francis, Jean-Marie Marandin, and Petra Sleeman (). Nounless determiners, in Francis Corblin and Henriette de Swart (eds), Handbook of French semantics. Stanford: CSLI Publications, –. Cormack, A. (). VP anaphora: Variables and scope, in F. Landman and F. Veltman (eds), Varieties of formal semantics. Dordrecht: Foris Publications, –.





Cornish, Francis (). Anaphora, discourse, and understanding: Evidence from French and English. Oxford: Oxford University Press. Corver, Norbert (). The syntax of left branch extractions. PhD thesis, Katholieke Universiteit Brabant. Corver, Norbert (). A note on subcomparatives, Linguistic Inquiry (): –. Corver, Norbert (). Much-support as a last resort, Linguistic Inquiry (): –. Corver, Norbert (). Comparative deletion and subdeletion, in Martin Everaert and Henk van Riemsdijk (eds), The Blackwell companion to syntax, vol. . Malden, MA and Oxford: Blackwell, –. Corver, Norbert (). Freezing effects. Ms. University of Utrecht. Corver, Norbert and Marjo van Koppen (). Let’s focus on NP-ellpsis, Groninger Arbeiten zur Germanistischen Linguistik : –. Corver, Norbert and Marjo van Koppen (). Ellipsis in Dutch possessive noun phrases: A microcomparative approach, Journal of Comparative Germanic Linguistics (): –. Corver, Norbert and Marjo van Koppen (). NP-ellipsis with adjectival remnants: A microcomparative perspective, Natural Language and Linguistic Theory : –. Corver, Norbert and Marjo van Koppen (). Pronominalization and variation in Dutch demonstrative expressions. Ms. Utrecht University. Forthcoming in Tonjes Veenstra (ed.), Demonstratives. Amsterdam and Philadelphia: John Benjamins. Costa, João, Ana Maria Martins, and Fernanda Pratas (). VP ellipsis: New evidence from Capeverdean, in Irene Franco, Sara Lusini, and Andrés Saab (eds), Romance languages and linguistic theory . Amsterdam and Philadelphia: John Benjamins, –. van de Craats, Ineke, Norbert Corver, and Roeland van Hout (). Conservation of grammatical knowledge: On the acquisition of possessive noun phrases by Turkish and Moroccan learners of Dutch, Linguistics (): –. van Craenenbroeck, Jeroen (). Ellipsis in Dutch dialects. PhD thesis, Leiden University. van Craenenbroeck, Jeroen (). Ellipsis and accommodation: The (morphological) case of sluicing. Paper presented at MIT. van Craenenbroeck, Jeroen (a). Invisible last resort: A note on clefts as the underlying source for sluicing, Lingua (): –. van Craenenbroeck, Jeroen (b). The syntax of ellipsis: Evidence from Dutch dialects. New York: Oxford University Press. van Craenenbroeck, Jeroen (a). How do you sluice when there is more than one CP? in Jason Merchant and Andrew Simpson (eds), Sluicing: Cross-linguistic perspectives. Oxford: Oxford University Press, –. van Craenenbroeck, Jeroen (b). Ellipsis, identity and accommodation. Ms. KU Leuven. van Craenenbroeck, Jeroen (). VP-ellipsis, in Martin Everaert and Henk van Riemsdijk (eds), The Wiley Blackwell companion to syntax, nd edition. Hoboken: Wiley-Blackwell, –. van Craenenbroeck, Jeroen and Marcel den Dikken (). Ellipsis and EPP repair, Linguistic Inquiry (): –. van Craenenbroeck, Jeroen and Anikó Lipták (). Ellipsis in Hungarian and the typology of sluicing, in Kiyong Choi and Changguk Kim (eds), Ellipsis in minimalism: Proceedings of the th Seoul International Conference on Generative Grammar. Seoul: Hankook, –. van Craenenbroeck, Jeroen and Anikó Lipták (). The crosslinguistic syntax of sluicing: Evidence from Hungarian relatives, Syntax (): –. van Craenenbroeck, Jeroen and Anikó Lipták (). On the interaction between verb movement and ellipsis: New evidence from Hungarian, in Charles B. Chang and Hannah J. Haynie (eds), Proceedings of the th West Coast Conference on Formal Linguistics. Somerville, MA: Cascadilla Press, –. van Craenenbroeck, Jeroen and Anikó Lipták (). What sluicing can do, what it can’t and in which language: On the cross-linguistic syntax of ellipsis, in Lisa L.-S. Cheng and Norbert Corver (eds), Diagnosing syntax. New York: Oxford University Press, –.





van Craenenbroeck, Jeroen and Jason Merchant (). Ellipsis phenomena, in Marcel den Dikken (ed.), The Cambridge handbook of generative syntax. Cambridge: Cambridge University Press, –. Cremers, Crit (). Syntax of Dutch: Ellipsis and grammar. Ms. Leiden University. Cresswell, Max J. (). The semantics of degree, in Barbara Partee (ed.), Montague grammar. New York: Academic Press, –. Croft, William (). Radical Construction Grammar: Syntactic theory in typological perspective. New York: Oxford University Press. Cruse, Alan (). The pragmatics of lexical specificity, Journal of Linguistics : –. Crysmann, Berthold (). An asymmetric theory of peripheral sharing in HPSG, in Gerald Penn (ed.), Proceedings of FG Vienna: The th Conference on Formal Grammar. Stanford: CSLI Publications, –. Culicover, Peter (a). English (zero-)relatives and the competence-performance distinction, International Review of Pragmatics : –. Culicover, Peter (b). Grammar and complexity. Oxford: Oxford University Press. Culicover, Peter and Ray Jackendoff (). Simpler Syntax. Oxford: Oxford University Press. Culicover, Peter and Ray Jackendoff (). A domain-general cognitive relation and how language expresses it, Language : –. Culicover, Peter and Paul Postal (eds) (). Parasitic gaps. Cambridge, MA: MIT Press. Culicover, Peter and Susanne Winkler (). English focus inversion, Journal of Linguistics (): –. Cutler, Anne and Jerry A. Fodor (). Semantic focus and sentence comprehension, Cognition : –. Cyrino, Sonia (). On the existence of Null Complement Anaphora in Brazilian Portuguese, Revista Letras : –. Cyrino, Sonia and Gabriela Matos (). VP-ellipsis in European and Brazilian Portuguese: A comparative analysis, Journal of Portuguese Linguistics : –. Cyrino, Sonia and Gabriela Matos (). Local licensers and recovering in VP ellipsis, Journal of Portuguese Linguistics : –. Cyrino, Sonia and Gabriela Matos (). Null Complement Anaphora in Romance: Deep or surface anaphora? in Jenny Doetjes and Paz González (eds), Romance Languages and Linguistic Theory . Amsterdam and Philadelphia: John Benjamins, –. Dadan, Marcin (). Preposition omission in sluicing: Teasing apart LF-copying and PF-deletion, in Christopher Hammerly and Brandon Prickett (eds), NELS : Proceedings of the th annual meeting of the North East Linguistic Society. Amherst, MA: GLSA, –. Dagnac, Anne (). Modal ellipsis in French, Spanish and Italian: Evidence for a TP-deletion analysis, in Karlos Arregi, Zsuzsanna Fagyal, Silvina Montrul, and Mireille Tremblay (eds), Romance linguistics : Interactions in Romance. Amsterdam and Philadelphia: John Benjamins, –. Dagnac, Anne (). Gapping needs vP-coordination: An argument from French NPI licensing, The Linguistic Review (): –. Dahl, Östen (). On so-called sloppy identity, Synthese : –. Dahl, Östen (). How to open a sentence: Abstraction in natural language, Logical Grammar Reports . University of Götenburg. Dalrymple, Mary (). Against reconstruction in ellipsis. Unpublished ms. Xerox-PARC and CSLI. Dalrymple, Mary, Stuart M. Shieber, and Fernando C. N. Pereira (). Ellipsis and higher-order unification, Linguistics and Philosophy (): –. Davidson, Kathryn (). ‘And’ and ‘or’: General use coordination in ASL, Semantics and Pragmatics : –. Davies, Mark (). The Corpus of Contemporary American English:  million words, –present. Available online at .





Dayal, Veneeta and Roger Schwarzschild (). Definite inner antecedents and wh-correlates in sluices, Rutgers Working Papers in Linguistics : –. Dékány, Éva (). A profile of the Hungarian DP: The interaction of lexicalization, agreement and linearization with the functional sequence. PhD thesis, University of Tromsø. Dékány, Éva (). The syntax of anaphoric possessives in Hungarian, Natural Language and Linguistic Theory (): –. Depiante, Marcela (). The syntax of deep and surface anaphora: A study of null complement anaphora and stripping/bare argument ellipsis. PhD thesis, University of Connecticut at Storrs. Depiante, Marcela (). On null complement anaphora in Spanish and Italian, Probus (): –. Depiante, Marcela and Jorge Hankamer (). Non-constituent ellipsis. Talk presented at the Identity in Ellipsis workshop, University of California, Berkeley. Depiante, Marcela and Pascual Masullo (). Género y número en la elipsis nominal: Consecuencias para la hipótesis lexicalista. Talk given at the I Encuentro de gramática Generativa, National University of Comahue. Desmets, Marianne (). Les typages de la phrase en HPSG: Le cas des phrases en ‘omme’. PhD thesis, University of Paris . Desmets, Marianne (). Ellipses dans les constructions comparatives en ‘comme’, Linx : –. Dickey, Michael Walsh and Ann C. Bunger (). Comprehension of elided structure: Evidence from sluicing, Language and Cognitive Processes (): –. Dik, Simon (). Coordination: Its implications for the theory of general linguistics. Amsterdam: North-Holland. den Dikken, Marcel (). Agreement and ‘clause union’, in Katalin É. Kiss and Henk van Riemsdijk (eds), Verb clusters: A study of Hungarian, German and Dutch. Amsterdam and Philadelphia: John Benjamins, –. den Dikken, Marcel (). Either-float and the syntax of co-or-dination, Natural Language and Linguistic Theory (): –. den Dikken, Marcel (). The distributed morphology of codeswitching. Talk given at the UIC Bilingualism Forum, University of Illinois at Chicago. den Dikken, Marcel and Shoba Bando-Rao (). On v as pivot in codeswitches, and the nature of the ban on word-internal switches, in Jeff MacSwan (ed.), Grammatical theory and bilingual codeswitching. Cambridge, MA: MIT Press, –. den Dikken, Marcel, André Meinunger, and Chris Wilder (). Pseudoclefts and ellipsis, Studia Linguistica : –. Dillon, Brian, Alan Mishler, Shayne Sloggett, and Colin Phillips (). Contrasting intrusion profiles for agreement and anaphora: Experimental and modeling evidence, Journal of Memory and Language : –. Dixon, Robert M. W. (). Ergativity. Cambridge: Cambridge University Press. Dixon, Robert M. W. (). Basic linguistic theory, vol. : Methodology. Oxford: Oxford University Press. Dobrovie-Sorin, Carmen (). The syntax of Romanian. The Hague: Mouton. Donati, Caterina (). Comparative clauses as free relatives: A raising analysis, Probus (): –. Doron, Edit (). V-movement and VP ellipsis, in Shalom Lappin and Elabbas Benmamoun (eds), Studies in ellipsis and gapping. New York: Oxford University Press, –. Dougherty, Ray C. (). A grammar of coordinate conjoined structures, I, Language (): –. Dougherty, Ray C. (). A grammar of coordinate conjoined structures, II, Language (): –. Dowty, David (). Type raising, functional composition, and non-constituent conjunction, in Richard Oehrle, Emmon Bach, and Deirdre Wheeler (eds), Categorial grammars and natural language structures. Dordrecht: D. Reidel, –. Dowty, David, Robert Wall, and Stanley Peters (). Introduction to Montague grammar. Dordrecht: D. Reidel.





Drai, Dan and Yosef Grodzinsky (a). A new empirical angle on the variability debate: Quantitative neurosyntactic analyses of a large data set from Broca’s aphasia, Brain and Language : –. Drai, Dan and Yosef Grodzinsky (b). The variability debate: More statistics, more linguistics, Brain and Language : –. Drummond, Alex (). Dahl’s paradigm: In defense of the crossover analysis. Poster presented at NELS : the th annual meeting of the North East Linguistic Society. University of Connecticut. Drummond, Alex, Norbert Hornstein, and Howard Lasnik (). A puzzle about P-stranding and a possible solution, Linguistic Inquiry (): –. Dryer, Matthew S. (). Descriptive theories, explanatory theories, and Basic Linguistic Theory, in Felix K. Ameka, Alan Dench, and Nicholas Evans (eds), Catching language: Issues in grammar writing. Berlin: Mouton de Gruyter, –. Duffield, Nigel and Ayumi Matsuo (). A comparative study of ellipsis and anaphora in L acquisition, in Anna H.-J. Do, Laura Domínguez, and Aimee Johansen (eds), Proceedings of the th Boston University Conference on Child Language Development. Somerville, MA: Cascadilla Press, –. Dvořák, Bostjan (). Slovenian clitic pronouns and what is so special about them, Slovene Linguistic Studies : –. É. Kiss, Katalin (). Configurationality in Hungarian. Budapest: Akadémiai Kiadó. É. Kiss, Katalin (). Wh-movement and specificity, Natural Language and Linguistic Theory : –. É. Kiss, Katalin (ed.) (). Discourse configurational languages. Oxford: Oxford University Press. É. Kiss, Katalin (). Identificational focus and information focus, Language : –. É. Kiss, Katalin (). The syntax of Hungarian. Cambridge: Cambridge University Press. É. Kiss, Katalin (). Free word order, (non)configurationality, and phases, Linguistic Inquiry (): –. É. Kiss, Katalin (). An adjunction analysis of quantifiers and adverbials in the Hungarian sentence, Lingua (): –. Edelstein, Elspeth (). This syntax needs studied, in Lawrence Horn and Rafaella Zanuttini (eds), Microsyntactic variation in North American English. Oxford: Oxford University Press, –. Edwards, John (). Foundations of bilingualism, in Tej K. Bhatia and William C. Ritchie (eds), Handbook of bilingualism. Malden, MA: Blackwell, –. Egg, Markus and Katrin Erk (). A compositional account of VP-ellipsis, in Frank van Eynde, Lars Hellan, and Dorothee Beerman (eds), Proceedings of the th International Conference on HeadDriven Phrase Structure Grammar. Stanford: CSLI, –. Eguren, Luis (). Contrastive focus and nominal ellipsis in Spanish, Lingua (): –. Eichler, Nadine (). Code-Switching bei bilingual aufwachsenden Kindern: Eine Analyse der gemischtsprachlichen Nominalphrasen unter besonderer Berücksichtigung des Genus. Tübingen: Narr Verlag. Eichler, Nadine and Natascha Müller (). The derivation of mixed DPs: Mixing of functional categories in bilingual children and in second language learners, in Marzena Watorek, Sandra Benazzo, and Maya Hickmann (eds), Comparative perspectives on language acquisition: A tribute to Clive Perdue. Bristol: Multilingual Matters, –. Eisenberg, Peter (). A note on ‘identity of constituents’, Linguistic Inquiry (): –. Elbourne, Paul (). E-type anaphora as NP deletion, Natural Language Semantics (): –. Elbourne, Paul (). Situations and individuals. Cambridge, MA: MIT Press. Elbourne, Paul (). Ellipsis sites as definite descriptions, Linguistic Inquiry (): –. Elbourne, Paul (). Definite descriptions. Oxford: Oxford University Press. Elliot, Patrick, Andreea Nicolae, and Yasutada Sudo (). VP ellipsis without parallel binding: A QUD approach. Poster presented at SALT (Semantics and Linguistic Theory Conference) . Embick, David and Alec Marantz (). Architecture and blocking, Linguistic Inquiry (): –.





Emonds, Joseph E. (). The verbal complex V’-V in French, Linguistic Inquiry : –. Emonds, Joseph E. (). A unified theory of syntactic categories. Dordrecht: Foris Publications. Engberg-Pedersen, Elisabeth (). Space in Danish Sign Language: The space and morphosyntax of the use of space in a visual language. Hamburg: Signum Verlag. Engberg-Pedersen, Elisabeth (). Grammatical relations in Danish Sign Language, in Anneli Pajunen (ed.), Mimesis, sign, and the evolution of language. Turku: University of Turku Press, –. Engdahl, Elisabet (). Parasitic gaps, Linguistics and Philosophy (): –. Engdahl, Elisabet (). Constituent questions. Dordrecht: D. Reidel. Engel, Ulrich (). Syntax der deutschen Gegenwartssprache. Berlin: Erich Schmidt. Epstein, Samuel (). A derivational approach to syntactic relations. Unpublished manuscript, Harvard University. Eroms, Hans-Werner (). Syntax der deutschen Sprache. Berlin: Walter de Gruyter. Erschler, David (). On embedded gapping. Talk presented at IATL : The nd annual meeting of the Israel Association for Theoretical Linguistics. Erteschik, Naomi (). On the nature of island constraints. PhD dissertation, MIT. Erteschik-Shir, Nomi (). The syntax-focus interface, in Peter W. Culicover and Louise McNally (eds), The limits of syntax. San Diego: Academic Press, –. Erteschik-Shir, Nomi (). The syntax-discourse interface. Oxford: Oxford University Press. Erteschik-Shir, Nomi, Lena Ibn-Bari, and Sharon Taube (). Object ellipsis as topic drop. Ms. Ben Gurion University of the Negev. Erteschik-Shir, Nomi, Lena Ibn-Bari, and Sharon Taube (). Missing objects as topic drop, Lingua : –. Erteschik-Shir, Nomi and Shalom Lappin (). Dominance and the functional explanation of island phenomena, Theoretical Linguistics : –. Eshghi, Arash, Christine Howes, Eleni Gregoromichelaki, Julian Hough, and Matthew Purver (). Feedback in conversation as incremental semantic update, in Proceedings of the th International Conference on Computational Semantics. Stroudsburg, PA: Association for Computational Linguistics, –. Evans, Frederic (). Binding into anaphoric verb phrases, in Joyce Powers and Kenneth de Jong (eds), Proceedings of the th Annual Eastern States Conference on Linguistics. Ohio: ESCOL Publications Committee, Linguistics Department, Ohio State University, –. Evans, N. and S. C. Levinson (). The myth of language universals: Language diversity and its importance for cognitive science, Behavioral and Brain Sciences (), –. Fanselow, Gisbert and Stefan Frisch (). Effects of processing difficulty on judgments of acceptability, in Gisbert Fanselow, Caroline Féry, Ralf Vogel, and Matthias Schlesewsky (eds), Gradience in grammar. Oxford: Oxford University Press, –. Farkas, Donka (). Polarity particles in Hungarian, in Marcel den Dikken and Robert Vago (eds), Approaches to Hungarian . Amsterdam and Philadelphia: John Benjamins, –. Farkas, Donka and Floris Roelofsen (forthcoming). Polar initiatives and polarity particle responses in an inquisitive discourse model. Language. Accepted subject to minor revisions. Farudi, Annahita (). Gapping in Farsi: A cross-linguistic investigation. PhD thesis, University of Massachusetts at Amherst. Fernández, Raquel (). Non-sentential utterances in dialogue: Classification, resolution and use. PhD thesis, King’s College. Fernández, Raquel and Jonathan Ginzburg (). Non-sentential utterances: A corpus study, Traitement automatique des languages (): –. Fernández, Raquel, Jonathan Ginzburg, and Shalom Lappin (). Classifying ellipsis in dialogue: A machine learning approach, in COLING ’: Proceedings of the th International Conference on Computational Linguistics. Stroudsburg, PA: Association for Computational Linguistics, –. Fernández, Raquel, Jonathan Ginzburg, and Shalom Lappin (). Classifying non-sentential utterances in dialogue: A machine learning approach, Computational Linguistics (): –.





Ferrara, Lindsay and Trevor Johnston (). Elaborating who’s what: A study of constructed action and clause structure in Auslan (Australian Sign Language), Australian Journal of Linguistics : –. Féry, Caroline (). Prosodische und tonale Faktoren bei der Disambiguierung syntaktischer Strukturen, in Karl H. Ramers, Heinz Vater, and Henning Wode (eds), Universale phonologische Strukturen und Prozesse. Tübingen: Niemeyer, –. Féry, Caroline and Katharina Hartmann (). The focus and prosodic structure of German Right Node Raising and gapping, The Linguistic Review (): –. Féry, Caroline and Shinichiro Ishihara (eds) (). The Oxford Handbook of Information Structure. Oxford and New York: Oxford University Press. Fiengo, Robert (). Surface structure. Cambridge, MA: Harvard University Press. Fiengo, Robert and Robert May (). Indices and identity. Cambridge, MA: MIT Press. Fillmore, Charles (). Types of lexical information, in Ferenc Kiefer (ed.), Studies in syntax and semantics. Dordrecht: D. Reidel, –. Fillmore, Charles (). Pragmatically controlled zero anaphora, in Proceedings of the Twelfth Annual Meeting of the Berkeley Linguistics Society. Berkeley: Berkeley Linguistics Society, –. Fillmore, Charles, Paul Kay, and Mary Catherine O’Connor (). Regularity and idiomaticity in grammatical constructions: The case of ‘let alone’, Language (): –. Filtopoulos, Rebecca (). Dialogical analysis of theatre: A corpus study of non-sentential utterances in Waiting for Godot and Endgame. Master’s thesis, Université Paris Diderot. von Fintel, Kai (). NPI licensing, Strawson entailment and context dependency, Journal of Semantics (): –. Fischer, Klaus (). German-English verb valency: A contrastive analysis. Tübingen: Narr Verlag. Fitzpatrick, Justin M. (). Deletion through movement, Natural Language and Linguistic Theory (): –. Fletcher, Charles (). Levels of representation in memory for discourse, in Morton Ann Gernsbacher (ed.), Handbook of psycholinguistics. San Diego: Academic Press, –. Fodor, Janet Dean and Ivan Sag (). Referential and quantificational indefinites, Linguistics and Philosophy : –. Fodor, Jerry A., Thomas G. Bever, and Merrill F. Garrett (). The psychobiology of language. New York: McGraw-Hill Publishers. Foley, Claire, Zelmira Núñez del Prado, Isabella Barbier, and Barbara Lust (). Knowledge of variable binding in VP-ellipsis: Language acquisition research and theory converge, Syntax (): –. Folli, Raffaella, Heidi Harley, and Simin Karimi (). Determinants of event type in Persian complex predicates, Lingua : –. Fortin, Catherine (a). Some (not all) nonsententials are only a phase, Lingua (): –. Fortin, Catherine (b). Indonesian sluicing and verb phrase ellipsis: Description and explanation in the Minimalist framework. PhD thesis, University of Michigan. Fortin, Catherine (). We need LF-copying: A few good reasons why, in Mary Byram Washburn, Sarah Ouwayda, Chuoying Ouyang, Bin Yin, Canan Ipek, Lisa Marston, and Aaron Walker (eds), Proceedings of the th West Coast Conference on Formal Linguistics. Somerville, MA: Cascadilla Press, –. Fortin, Catherine (). Indonesian sluicing is LF copying. Ms. Carleton College. Fortin, Catherine (). The distribution of [E]: A view from Indonesian verb phrase ellipsis, in Joseph Sabbagh (ed.), Proceedings of the th Meeting of the Austronesian Formal Linguistics Association (AFLA ). London: University of Western Ontario. Fox, Danny (). Economy and scope, Natural Language Semantics (): –. Fox, Danny (). Reconstruction, binding theory, and the interpretation of chains, Linguistic Inquiry (): –.





Fox, Danny (). Economy and semantic interpretation. Cambridge, MA: MIT Press. Fox, Danny (). Antecedent-contained deletion and the copy theory of movement, Linguistic Inquiry (): –. Fox, Danny and Howard Lasnik (). Successive-cyclic movement and island repair: The difference between sluicing and VP-ellipsis, Linguistic Inquiry (): –. Fox, Danny, and John Nissenbaum (). Extraposition and scope: A case for overt QR. Paper presented at WCCFL : The th West Coast Conference on Formal Linguistics. Fraser, Bruce (). Vice versa, Linguistic Inquiry (): –. Frauenfelder, Uli and Caroline Floccia (). Das Erkennen gesprochener Wörter, in Angela D. Friederici (ed.), Sprachrezeption. Göttingen: Hofgrefe, –. Frazier, Lyn (). Processing ellipsis: A processing solution to the undergeneration problem? in Charles B. Chang and Hannah J. Haynie (eds), Proceedings of the th West Coast Conference on Formal Linguistics. Somerville, MA: Cascadilla Press, –. Frazier, Lyn (). Two interpretive systems for natural language? Journal of Psycholinguistic Research (): –. Frazier, Lyn and Charles Clifton, Jr (). Comprehension of sluiced sentences, Language and Cognitive Processes : –. Frazier, Lyn and Charles Clifton, Jr (). On bound-variable interpretations: The LF-only hypothesis, Journal of Psycholinguistic Research : –. Frazier, Lyn and Charles Clifton, Jr (). Parsing coordinates and ellipsis: Copy a, Syntax (): –. Frazier, Lyn and Charles Clifton, Jr (). The syntax-discourse divide: Processing ellipsis, Syntax (): –. Frazier, Lyn and Charles Clifton Jr (). Ellipsis and discourse coherence, Linguistics and Philosophy : –. Frazier, Lyn and Charles Clifton Jr (). Dynamic interpretations: Finding an antecedent for VPE, in Jesse A. Harris and Margaret Grant (eds), University of Massachusetts occasional papers in linguistics: Processing linguistic structure. Amherst, MA: GLSA, –. Frazier, Lyn and Charles Clifton Jr (). Without his shirt off he saved the child from almost drowning: Intepreting uncertain input, Language, Cognition and Neuroscience (): –. Frazier, Lyn, Charles Clifton Jr, and Katy Carlson (). Focus and VP ellipsis, Language and Speech : –. Frazier, Lyn, Lori Taft, Tom Roeper, Charles Clifton Jr, and Kate Erlich (). Parallel structure: A source of facilitation in sentence comprehension, Memory and Cognition : –. Frederking, Robert E. (). Understanding dialogue ellipsis. Pittsburgh: Center for Machine Translation, Carnegie Mellon University. Fried, Mirjam (). Construction grammar as a tool for diachronic analysis, Constructions and Frames (): –. Friederici, Angela D. (). Towards a neural basis of auditory sentence processing, Trends in Cognitive Sciences : –. Friedman, Lynn (). The manifestation of subject, object and topic in American Sign Language, in Charles N. Li (ed.), Subject and topic. New York: Academic Press, –. Fu, Jingqi, Tom Roeper, and Hagit Borer (). The VP within process nominals: Evidence from adverbs and the VP anaphor do so, Natural Language and Linguistic Theory (): –. Fukaya, Teruhiko (). Island (in)sensitivity in Japanese sluicing and stripping, in Gina Garding and Mimu Tsujimura (eds), Proceedings from the th West Coast Conference on Formal Linguistics. Somerville, MA: Cascadilla Press, –. Fukaya, Teruhiko (). Sluicing and stripping in Japanese and some implications. PhD thesis, USC. Fukaya, Teruhiko (). Island-sensitivity in Japanese sluicing and some implications, in Jason Merchant and Andrew Simpson (eds), Sluicing: Cross-linguistic perspectives. Oxford: Oxford University Press, –.





Fukaya, Teruhiko and Hajime Hoji (). Stripping and sluicing in Japanese and some implications, in S. Bird, A. Carnie, J. Haugen, and P. Norquest (eds), Proceedings of the th West Coast Conference on Formal Linguistics. Somerville, MA: Cascadilla Press, –. Fukui, Naoki (). A theory of category projection and its application. PhD thesis, Massachusetts Institute of Technology. Fukui, Naoki and Margaret Speas (). Specifiers and projection, in Naoki Fukui, Tova R. Rapoport, and Elizabeth Sagey (eds), Papers in theoretical linguistics: MIT working papers in linguistics . Cambridge, MA: MIT Working Papers in Linguistics, –. Funakoshi, Kenshi (). On headless XP-movement/ellipsis, Linguistic Inquiry : –. Funakoshi, Kenshi (). Syntactic head movement and its consequences. PhD thesis, University of Maryland, College Park. Funakoshi, Kenshi (). Verb-stranding verb phrase ellipsis in Japanese, Journal of East Asian Linguistics : –. Gagnon, Michaël (a). Anaphors and the missing link. PhD thesis, University of Maryland, College Park. Gagnon, Michaël (b). Part and parcel of eliding partitives, SALT : Proceedings of the rd Semantics and Linguistic Theory Conference. Ithaca, NY: CLC Publications, –. Gajewski, Jon (). L-analyticity and natural language. Ms. MIT. Gärdenfors, Peter (). The geometry of meaning: Semantics based on conceptual spaces. Cambridge, MA: MIT Press. Gardent, Claire (). Gapping and VP ellipsis in a unification-based grammar. PhD thesis, University of Edinburgh. Gardent, Claire and Michael Kohlhase (). Higher-order coloured unification and natural language semantics, in Proceedings of the th Annual Meeting on Association of Computational Linguistics. Stroudsburg, PA: Association for Computational Linguistics, –. Gardent, Claire and Michael Kohlhase (). Computing parallelism in discourse, IJCAI’: Proceedings of the Fifteenth International Joint Conference on Artifical Intelligence. San Francisco: Morgan Kaufmann Publishers, –. Gardent, Claire and Michael Kohlhase (). Higher-order colored unification: A linguistic application, Technique et Science Informatique : –. Gargett, Andrew, Eleni Gregoromichelaki, Chris Howes, and Yo Sato (). Dialogue-grammar correspondence in Dynamic Syntax, in Jonathan Ginzburg, Pat Healey, and Yo Sato (eds), Proceedings of the th Workshop on the Semantics and Pragmatics of Dialogue (SEMDIAL), –. Gargett, Andrew, Eleni Gregoromichelaki, Ruth Kempson, Matthew Purver, and Yo Sato (). Grammar resources for modeling dialogue dynamically, Cognitive Neurodynamics (): –. Gawron, Jean Mark (). Comparatives, superlatives, and resolution, Linguistics and Philosophy (): –. Gawron, Jean Mark and Andrew Kehler (). The semantics of respective readings, conjunction, and filler-gap dependencies, Linguistics and Philosophy (): –. Gawron, Jean Mark and Stanley Peters (). Anaphora and quantification in situation semantics. Stanford University: Center for the Study of Language and Information. van Gelderen, Elly and Jeff MacSwan (). Interface conditions and code-switching: Pronouns, lexical DPs, and checking theory, Lingua (): –. Gengel, Kirsten (). Focus and ellipsis: A generative analysis of pseudogapping and other elliptical structures. PhD thesis, University of Stuttgart. Gengel, Kirsten (). Pseudogapping and ellipsis. Oxford: Oxford University Press. Gennari, Silvia P. and Maryellen C. MacDonald (). Acquisition of negation and quantification: Insights from adult production and comprehension, Language Acquisition (): –. Gergel, Remus (). Short-distance reanalysis of Middle English modals: Evidence from ellipsis, Studia Linguistica (): –.





Gergel, Remus (). Interpretable features in vP-ellipsis: On the licensing head, in S. Blaho, E. Schoorlemmer, and L. Vicente (eds), Proceedings of ConSOLE XIV. Leiden: SOLE, –. Gergel, Remus (). Modality and ellipsis: Diachronic and synchronic evidence. Berlin and New York: Mouton de Gruyter. Gergel, Remus, Kirsten Gengel, and Susanne Winkler (). Ellipsis and inversion: A feature-based account, in Kerstin Schwabe and Susanne Winkler (eds), On information structure, meaning and form. Amsterdam and Philadelphia: John Benjamins, –. Ghaniabadi, Saeed (). The empty noun construction in Persian. PhD thesis, University of Manitoba. Giannakidou, Anastasia and Jason Merchant (). Reverse sluicing in English and Greek, The Linguistic Review : –. Giannakidou, Anastasia and Melita Stavrou (). Nominalization and ellipsis in the Greek DP, The Linguistic Review : –. Gibson, Edward, Pauline Jacobson, Peter Graff, Kyle Mahowald, Evelina Fedorenko, and Steven T. Piantadosi (). A pragmatic account of complexity in definite antecedent-contained-deletion relative clauses, Journal of Semantics (): –. Ginzburg, Jonathan (). Questions, queries and facts: A semantics and pragmatics for interrogatives. PhD thesis, Stanford University, Stanford. Ginzburg, Jonathan (). An update semantics for dialogue, in Harry Bunt (ed.), Proceedings of the st International Workshop on Computational Linguistics. Tilburg: ITK. Ginzburg, Jonathan (). Resolving questions, I, Linguistics and Philosophy (): –. Ginzburg, Jonathan (). Semantically-based ellipsis resolution with syntactic presuppositions, in Harry Bunt and Reinhard Muskens (eds), Computing meaning. Springer, –. Ginzburg, Jonathan (). The interactive stance: Meaning for conversation. Oxford: Oxford University Press. Ginzburg, Jonathan and Robin Cooper (). Clarification, ellipsis, and the nature of contextual updates in dialogue, Linguistics and Philosophy (): –. Ginzburg, Jonathan and Robin Cooper (). Quotation via dialogical interaction, Journal of Logic, Language, and Information (): –. Ginzburg, Jonathan, Robin Cooper, and Tim Fernando (a). Propositions, questions, and adjectives: A rich type-theoretic approach, in Proceedings of the EACL  Workshop on Type Theory and Natural Language Semantics (TTNLS). Gothenburg: Association for Computational Linguistics, –. Ginzburg, Jonathan, Robin Cooper, Julian Hough, and David Schlangen (). Incrementality and clarification/sluicing potential, in Proceedings of Sinn und Bedeutung . Edinburgh. Ginzburg, Jonathan and Raquel Fernández (). Computational models of dialogue, in Alex Clark, Chris Fox, and Shalom Lappin (eds), Handbook of computational linguistics and natural language processing. Oxford: Blackwell, –. Ginzburg, Jonathan, Raquel Fernández, and David Schlangen (b). Disfluencies as intra-utterance dialogue moves, Semantics and Pragmatics (): –. Ginzburg, Jonathan and Dimitra Kolliakou (). Answers without questions: The emergence of fragments in child language, Journal of Linguistics : –. Ginzburg, Jonathan and Massimo Poesio (). Grammar is a system that characterizes talk in interaction, Frontiers in Psychology : –. Ginzburg, Jonathan and Ivan Sag (). Interrogative investigations. Stanford: CSLI Publications. Gleitman, Lila (). Coordinating conjunctions in English, Language : –. Göksun, Tilbe, Tom Roeper, Kathy Hirsh-Pasek, and Roberta M. Golinkoff (). From noun phrase ellipsis to verb phrase ellipsis: The acquisition path from context to abstract reconstruction, in Jassie A. Harris and Margaret Grant (eds), Processing linguistic structure. Amherst, MA: GLSA, –.





Goldberg, Adele E. (). Constructions: A Construction Grammar approach to argument structure. Chicago: University of Chicago Press. Goldberg, Adele E. (). Surface generalizations: An alternative to derivations, Cognitive Linguistics : –. Goldberg, Adele E. (). Backgrounded constituents cannot be ‘extracted’, in Jon Sprouse and Norbert Hornstein (eds), Experimental syntax and island effects. Cambridge: Cambridge University Press, –. Goldberg, Lotus Madelyn (). Verb stranding VP-ellipsis: A cross-linguistic study. PhD thesis, McGill University. Goldsmith, John (). A principled exception to the Coordinate Structure Constraint, in William H. Eilfort, Paul D. Kroeber, and Karen L. Peterson (eds), Proceedings of the st Annual Meeting of the Chicago Linguistic Society. Chicago: Chicago Linguistic Society, –. Goldsmith, John (). Autosegmental and metrical phonology. Oxford: Blackwell. González-Vilbazo, Kay (). Die Syntax des Code-Switching. PhD thesis, University of Cologne. González-Vilbazo, Kay, Laura Bartlett, Sarah Downey, Shane Ebert, Jeanne Heil, Brad Hoot, Bryan Koronkiewicz, and Sergio Ramos (). Methodological considerations in code-switching research, Studies in Hispanic and Lusophone Linguistics (): –. González-Vilbazo, Kay and Bryan Koronkiewicz (). Tú y yo can codeswitch, nosotros cannot: Pronouns in Spanish-English codeswitching, in Rosa E. Guzzardo Tamargo, Catherine M. Mazak, and M. Carmen Parafita Couto (eds), Spanish-English codeswitching in the Caribbean and the US. Amsterdam and Philadelphia: John Benjamins, –. González-Vilbazo, Kay and Luis López (). Some properties of light verbs in codeswitching, Lingua (): –. González-Vilbazo, Kay and Luis López (). Little v and parametric variation, Natural Language and Linguistic Theory (): –. González-Vilbazo, Kay and Sergio Ramos (). A morphosyntactic condition on sluicing: Evidence from Spanish/German code-switching. Ms. University of Illinois at Chicago. González-Vilbazo, Kay and Volker Struckmeier (). Asymmetrien im code-switching: Eine DM-Lösung zur Partizipselektion, in Eva-Maria Remberger and Guido Menshing (eds), Romanistische Sprachwissenschaft–minimalistisch. Tübingen: Narr Verlag, –. Goodall, Grant (). Parallel structures in syntax. Cambridge: Cambridge University Press. Goodglass, Harold and Edith Kaplan (). The assessment of aphasia and related disorders. Philadelphia: Lea & Febiger. Gračanin-Yüksek, Martina (). About sharing. PhD thesis, MIT. Grant, Margaret (). The parsing and interpretation of comparatives: More than meets the eye. PhD thesis, University of Massachusetts at Amherst. Grant, Margaret, Charles Clifton Jr, and Lyn Frazier (). The role of non-actuality implicatures in processing elided constituents, Journal of Memory and Language (): –. Grebenyova, Lydia (). Interpretation of Slavic multiple wh-questions, in Proceedings of the Annual Workshop on Formal Approaches to Slavic Linguistics (FASL) . Grebenyova, Lydia (a). Multiple interrogatives: Syntax, semantics, and learnability. PhD thesis, University of Maryland, College Park. Grebenyova, Lydia (b). Multiple left branch extraction under sluicing, in Proceedings of the st Regional Meeting of the Chicago Linguistic Society. Chicago: Chicago Linguistic Society. Grebenyova, Lydia (c). Sluicing puzzles in Russian, in Hana Filip, Steven Franks, James Lavine, and Mila Tasseva-Kurktchieva (eds), Formal approaches to Slavic linguistics: The Princeton meeting. Ann Arbor: Michigan Slavic Publications, –. Grebenyova, Lydia (). Sluicing in Slavic, Journal of Slavic Linguistics (): –. Grebenyova, Lydia (). Sluicing and multiple wh-fronting, in Nguyen Chi Duy Khuong and Richa Samar Sinha (eds), Proceedings of GLOW in Asia . New Delhi: Central Institute of Indian Languages, –.





Grebenyova, Lydia (). Syntax, semantics and acquisition of multiple interrogatives. Amsterdam and Philadelphia: John Benjamins. Green, Lisa (). Aspectual BE-type constructions and coercion in African American English, Natural Language Semantics (): –. Green, Lisa (). African American English: A linguistic introduction. Cambridge: Cambridge University Press. Green, Lisa (). Language and the African American child. Cambridge: Cambridge University Press. Gregoromichelaki, Eleni (). Conditionals: A Dynamic Syntax account. PhD thesis, King’s College London. Gregoromichelaki, Eleni (). Review of J. Ginzburg (), The interactive stance, Folia Linguistica : –. Gregoromichelaki, Eleni (). Clitic left dislocation and clitic doubling: A dynamic perspective on leftright asymmetries in Greek, in Gert Webelhuth, Manfred Sailer, and Heike Walker (eds), Rightward movement in a comparative perspective. Amsterdam and Philadelphia: John Benjamins, –. Gregoromichelaki, Eleni (). The case of non-sentential speech. Ms. University of Osnabrueck. Gregoromichelaki, Eleni (). Quotation in dialogue, in Paul Saka and Michael Johnson (eds), Semantic and pragmatic aspects of quotation. Berlin: Springer, –. Gregoromichelaki, Eleni, Ronnie Cann, and Ruth Kempson (). On coordination in dialogue: Subsentential talk and its implications, in Laurence Goldstein (ed.), On brevity. Oxford: Oxford University Press, –. Gregoromichelaki, Eleni, Ruth Kempson, Matthew Purver, Greg J. Mills, Ronnie Cann, Wilfried Meyer-Viol, and Pat G. T. Healey (). Incrementality and intention-recognition in utterance processing, Dialogue and Discourse (): –. Gregoromichelaki, Eleni, Yo Sato, Ruth Kempson, Andrew Gargett, and Christine Howes (). Dialogue modelling and the remit of core grammar, in Harry Bunt, Volha Petukhova, and Sander Wubben (eds), Proceedings of the Eighth International Conference on Computational Semantics (IWCS-). Tilburg, –. Gregory, Howard and Shalom Lappin (). Antecedent contained ellipsis in HPSG, in Gert Webelhuth, Jean-Pierre Koenig, and Andreas Kathol (eds), Lexical and constructional aspects of linguistic explanation. Stanford: CSLI, –. Gribanova, Vera (). Exposing the Russian verbal complex via evidence from prefixation and verb phrase ellipsis. Paper presented at the th Annual Workshop on Formal Approaches to Slavic Linguistics, Yale University. Gribanova, Vera (a). Ellipsis and the syntax of verbs in Russian. Talk given at NYU. Gribanova, Vera (b). Prefixation, verb phrase ellipsis and the structure of the Russian verbal complex. Ms. UCSC. Gribanova, Vera (c). Structural adjacency and the typology of interrogative interpretations, Linguistic Inquiry (): –. Gribanova, Vera (). On diagnosing ellipsis and argument drop: The view from Russian. Talk given at MIT. Gribanova, Vera (a). A new argument for verb-stranding verb phrase ellipsis, Linguistic Inquiry (): –. Gribanova, Vera (b). Verb-stranding verb phrase ellipsis and the structure of the Russian verbal complex, Natural Language and Linguistic Theory (): –. Gribanova, Vera (c). Copular clauses, clefts, and putative sluicing in Uzbek, Language : –. Gribanova, Vera and Emily Manetta (). Ellipsis in wh-in-situ languages: Deriving apparent sluicing in Hindi-Urdu and Uzbek, Linguistic Inquiry (): –. Grice, Martine, Stefan Baumann, and Ralf Benzmüller (). German intonation in autosegmentalmetrical phonology, in Sun-Ah Jun (ed.), Prosodic typology: The phonology of intonation and phrasing. Oxford: Oxford University Press, –. Grice, Paul (). Logic and conversation, in Peter Cole and Jerry Morgan (eds), Syntax and Semantics, vol. . New York: Academic Press, –.





Griffiths, James (). On appositives. PhD thesis, University of Groningen. Griffiths, James and Anikó Lipták (). Contrast and island sensitivity in clausal ellipsis, Syntax : –. Grimshaw, Jane (). Complement selection and the lexicon, Linguistic Inquiry : –. Grimshaw, Jane (). Extended projection. Unpublished manuscript, Brandeis University. Grinder, John and Paul M. Postal (). Missing antecedents, Linguistic Inquiry (): –. Grodzinsky, Yosef (). Language deficits and linguistic theory. PhD thesis, Brandeis University. Grodzinsky, Yosef (). Language deficits and the theory of syntax, Brain and Language : –. Grodzinsky, Yosef (). The neurology of syntax, Behavioral and Brain Sciences : –. Grodzinsky, Yosef (). Ellipsis in deficient language users: Parallels (and divergences) in parallelism. Plenary talk, Generative Approaches to Language Acquisition (GALA), Siena. Grodzinsky, Yosef (). Two puzzles in experimental syntax and semantics, in Proceedings of the th annual meeting of the Chicago Linguistic Society. Chicago: Chicago Linguistic Society, –. Grodzinsky, Yosef and Tanya Reinhart (). The innateness of binding and coreference, Linguistic Inquiry (): –. Grodzinsky, Yosef and Andrea Santi (). The battle for Broca’s region, Trends in Cognitive Sciences (): –. Groenendijk, Jeroen (). Inquisitive Semantics: Two possibilities for disjunction, in Peter Bosch, David Gabelaia, and Jérôme Lang (eds), Logic, language, and computation. Heidelberg: Springer, –. Groenendijk, Jeroen (). Erotetic languages and the inquisitive hierarchy, in J. van der Does and C. Dutilh Novaes (eds), This is not a Festschrift: A Festschrift for Martin Stokhof (online). Groenendijk, Jeroen and Floris Roelofsen (). Inquisitive semantics and pragmatics, in Jesús M. Larrazabal and Larraitz Zubeldia (eds), SPR-: Proceedings of the ILCLI International Workshop on Semantics, Pragmatics and Rhetoric. San Sebastián: Universidad del País Vasco. Groenendijk, Jeroen and Martin Stokhof (). The semantics of questions and the pragmatics of answers. PhD thesis, University of Amsterdam. Grootveld, Marjan Jeannette (). Parsing coordination generatively. PhD thesis, Leiden University. Grosjean, François (). Life with two languages: An introduction to bilingualism. Cambridge, MA: Harvard University Press. Grosjean, François (). Studying bilinguals: Methodological and conceptual issues, Bilingualism: Language and Cognition (): –. Groß, Thomas (). Theoretical foundations of dependency syntax. Munich: Ludicium. Groß, Thomas and Timothy Osborne (). Toward a practical DG theory of discontinuities, SKY Journal of Linguistics : –. Groß, Thomas and Timothy Osborne (). Katena und Konstruktion: Ein Vorschlag zu einer dependen-ziellen Konstruktionsgrammatik, Zeitschrift für Sprachwissenschaft (): –. Grosu, Alexander (). On the nonunitary nature of the Coordinate Structure Constraint, Linguistic Inquiry (): –. Grosu, Alexander (). Subcategorization and parallelism, Theoretical Linguistics : –. Grosu, Alexander (). On acceptable violations of parallelism constraints, in René Dirven and Vilém Fried (eds), Functionalism in linguistics. Amsterdam and Philadelphia: John Benjamins, –. Grosu, Alexander (). Three studies in locality and case. London: Routledge. Grosz, Barbara, Aravind Joshi, and Scott Weinstein (). Centering: A framework for modeling the local coherence of discourse, Computational Linguistics : –. Grosz, Patrick (). Movement and agreement in Right-Node Raising constructions, Syntax : –. Grover, Claire, Chris Brew, Suresh Manandhar, Marc Moens, and Andreas Schöter (). Priority union and generalization in discourse grammar, Edinburgh Working Papers in Cognitive Science : –.





Gualmini, Andrea, Sarah Hulsey, Valentine Hacquard, and Danny Fox (). The question-answer requirement for scope assignment, Natural Language Semantics : –. Guasti, Maria Teresa (). Language acquisition: The growth of grammar. Cambridge, MA: MIT Press. Guida, Flore (). Les phrases sans verbes. Ms. Université Paris-Diderot. Gullifer, Jason (). Processing reverse sluicing: A contrast with processing filler-gap dependences, in Keir Moulton and Matthew Wolf (eds), NELS : Proceedings of the th annual meeting of the North East Linguistic Society. Amherst, MA: GLSA, –. Gumperz, John (). The sociolinguistic significance of conversational code-switching. Papers on language and context, Working paper , University of California, Berkeley, –. Gundel, Jeanette, Nancy Hedberg, and Zacharski (). Cognitive status and the form of referring expressions in discourse, Language (): –. Guthrie, Malcolm (). Comparative Bantu: An introduction to the comparative linguistics and prehistory of the Bantu languages. Farnborough: Gregg Press. Gyuris, Béata (). Semantic licensing of VP-ellipsis, Acta Linguistica Hungarica (–): –. Ha, Seungwan (a). Contrastive focus: Licensor for Right Node Raising, in Emily Elfner and Martin Walkow (eds), NELS : Proceedings of the th annual meeting of the North East Linguistic Society. Amherst, MA: GLSA, –. Ha, Seungwan (b). Ellipsis, Right Node Raising, and across-the-board constructions. PhD thesis, Boston University. Ha, Seungwan (c). On ellipsis features and Right Node Raising, in Proceedings of ConSOLE XV. Leiden: SOLE, –. Hackl, Martin, Jorie Koster-Hale, and Jason Varvoutis (). Quantification and ACD: Evidence from real time sentence processing, Journal of Semantics : –. Haddican, Bill (). On egin: do-support and VP focus in Central and Western Basque, Natural Language and Linguistic Theory (): –. Haddican, Bill (). The structural deficiency of verbal pro-forms, Linguistic Inquiry (): –. Haeseryn, W., K. Romijn, G. Geerts, J. de Rooij, and M. C. van den Toorn (eds) (). Algemene Nederlandse Spraakkunst. Groningen / Deurne: Martinus Nijhoff / Wolters Plantyn. Haida, Andreas (). The indefiniteness and focusing of wh-words. PhD thesis, HumboldtUniversität zu Berlin. Haïk, Isabelle (). Indirect binding, Linguistic Inquiry (): –. Haïk, Isabelle (). Bound VPs that need to be, Linguistics and Philosophy (): –. Halliday, Michael (). Notes on transitivity and theme in English, part , Journal of Linguistics : –. Hallman, Peter (). Symmetry in structure building, Syntax (): –. Halpert, Claire (). Argument licensing and agreement in Zulu. PhD thesis, MIT. Halpert, Claire (). Argument licensing and agreement. Oxford: Oxford University Press. Hamblin, Charles (). Questions in Montague grammar, Foundations of Language (): –. Hankamer, Jorge (). Constraints on deletion in syntax. PhD thesis, Yale University. Hankamer, Jorge (a). Unacceptable ambiguity, Linguistic Inquiry (): –. Hankamer, Jorge (b). Why there are two than’s in English, in Claudia Corum, T. Cedric SmithStark, and Ann Weiser (eds), Proceedings of the th annual meeting of the Chicago Linguistic Society. Chicago: Chicago Linguistic Society, –. Hankamer, Jorge (). On the nontransformational derivation of some null VP anaphors, Linguistic Inquiry (): –. Hankamer, Jorge (). Deletion in coordinate structures. New York: Garland Publishing. Hankamer, Jorge and Ivan Sag (). Deep and surface anaphora, Linguistic Inquiry (): –. Hardt, Daniel (). A corpus-based survey of VP ellipsis. Ms. University of Pennsylvania. Hardt, Daniel (a). An algorithm for VP ellipsis, in Proceedings of the th annual meeting of the Association for Computational Linguistics. Stroudsburg, PA: Association for Computational Linguistics, –.





Hardt, Daniel (b). VP ellipsis and contextual interpretation, in COLING-: Proceedings of the th International Conference on Computational Linguistics. Stroudsburg, PA: Association for Computational Linguistics, –. Hardt, Daniel (c). VP ellipsis and semantic identity, in Chris Barker and David Dowty (eds), SALT : Proceedings of the nd Semantics and Linguistic Theory Conference. Ithaca, NY: CLC Publications, –. Hardt, Daniel (). Verb phrase ellipsis: Form, meaning and processing. PhD thesis, University of Pennsylvania. Hardt, Daniel (). Sense and reference in dynamic semantics, in P. J. E. Dekker and M. J. B. Stockhof (eds) Proceedings of the th Amsterdam Colloquium. Amsterdam: ILLC, –. Hardt, Daniel (a). An empirical approach to VP ellipsis, Computational Linguistics (): –. Hardt, Daniel (b). VPE as a proform: Some consequences for binding. Paper presented at the nd Colloque de Syntaxe et Sémantique à Paris (CSSP). Hardt, Daniel (). Dynamic interpretation of verb phrase ellipsis, Linguistics and Philosophy (): –. Hardt, Daniel and Owen Rambow (). Generation of VP ellipsis: A corpus-based approach, in Proceedings of the th Annual Meeting of the Association for Computational Linguistics. Stroudsburg, PA: Association for Computational Linguistics, –. Hardt, Daniel and Maribel Romero (). Ellipsis and the structure of discourse, Journal of Semantics (): –. Hardt, Daniel, Nicholas Asher, and Julie Hunter (). VP ellipsis without indices, in SALT : Proceedings of the rd Semantics and Linguistic Theory Conference, Ithaca, NY: CLC Publications, –. Harley, Heidi (). One-replacement, unaccusativity, acategorial roots, and bare phrase structure, in Slava Gorbachov and Andrew Nevins (eds), Harvard Working Papers in Linguistics, vol. . Cambridge, MA: Harvard University, –. Harley, Heidi (). Compounding in Distributed Morphology, in Rochelle Lieber and Pavel Stekauer (eds), Oxford Handbook of Compounding. Oxford: Oxford University Press, –. Harris, Jesse (). Processing ‘let alone’ coordination in silent reading, Lingua : –. Harris, Jesse and Katy Carlson (). Keep it local (and final): Remnant preferences for ‘let alone’ ellipsis, Quarterly Journal of Experimental Psychology (): –. Hartman, Jeremy (). Focus, deletion, and identity: Investigations of ellipsis in English. BA thesis, Harvard University. Hartman, Jeremy and Ruizi Ressi Ai (). A focus account of swiping, in Kleanthes K. Grohmann and Phoevos Panagiotidis (eds), Selected papers from the  Cyprus Syntaxfest. Cambridge: Cambridge Scholars Publishing, –. Hartmann, Katharina (). Right Node Raising and gapping. PhD thesis, Goethe-Universität zu Frankfurt am Main, Frankfurt. Hartmann, Katharina (). Right Node Raising and gapping: Interface conditions on prosodic deletion. Amsterdam and Philadelphia: John Benjamins. Hartmann, Katharina (). Background matching in Right Node Raising constructions, in Kerstin Schwabe and Susanne Winkler (eds), The interfaces: Deriving and interpreting omitted structures. Amsterdam and Philadelphia: John Benjamins, –. Hartmann, Katharina (). Right node raising and gapping: Interface conditions on prosodic deletion. Amsterdam: John Benjamins. Hartmann, Katharina and Viola Schmitt (). Violations of the Right Edge Constraint in Right Node Raising, Snippets : –. Harves, Stephanie (). Genitive of negation and the syntax of scope, in Marjo van Koppen, Erica Thrift, Erik-Jan van der Torre, and Malte Zimmerman (eds), Proceedings of ConSOLE IX. Leiden: SOLE, –.





Harwood, Will (). Being progressive is just a phase: Dividing the functional hierarchy. PhD thesis, Ghent University. Harwood, Will (). Being progressive is just a phase: Celebrating the uniqueness of progressive aspect under a phase-based analysis, Natural Language and Linguistic Theory : –. Hasegawa, Nobuko (/). On the so-called ‘empty pronouns’ in Japanese, The Linguistic Review : –. Hasegawa, Nobuko (). Wh-movement in Japanese: Matrix sluicing is different from embedded sluicing, in Cedric Boeckx and Suleyman Ulutas (eds), MIT Working Papers in Linguistics : Proceedings of the th Workshop on Formal Altaic Linguistics (WAFL-). Cambridge, MA: MITWPL, –. Haspelmath, Martin (). Indefinite Pronouns. Oxford: Oxford University Press. Haspelmath, Martin (a). Coordination, in Timothy Shopen (ed.), Language typology and syntactic description: Complex constructions, vol. . Cambridge: Cambridge University Press, –. Haspelmath, Martin (b). Pre-established categories don’t exist: Consequences for language description and typology, Linguistic Typology : –. Haspelmath, Martin (). Parametric versus functional explanations of syntactic universals, in T. Biberauer (ed.), The Limits of Syntactic Variation. Amsterdam: John Benjamins Publishing Company, –. Haspelmath, Martin (). Comparative concepts and descriptive categories in cross-linguistic studies, Language : –. Hausser, Roland and Dietmar Zaefferer (). Questions and answers in a context dependent Montague grammar, in Franz Guenther and Max Schmidt (eds), Formal semantics and pragmatics for natural languages. Dordrecht: D. Reidel, –. Hawkins, John A. (). Efficiency and complexity in grammars. Oxford: Oxford University Press. Hawkins, John A. (). Cross-linguistic variation and efficiency. Oxford: Oxford University Press. Hayashi, Shintaro (). Head movement in an agglutinative SOV language. PhD thesis, Yokohama National University. Haynie, Hannah J. (). What is Null Complement Anaphora? A syntactic account can explain. Ms. University of California at Berkeley. van der Heijden, Emmeken (). Tussen nevenschikking en onderschikking: een onderzoek naar verschillende vormen van verbinding in het Nederlands. PhD thesis, Radboud University Nijmegen. van der Heijden, Emmeken and Maarten Klein (). Rekenkundige voegwoorden: De ‘logica’ van samentrekkingen, De Nieuwe Taalgids : –. Heim, Irene (). The semantics of definite and indefinite noun phrases. PhD thesis, University of Massachusetts, Amherst. Heim, Irene (). Notes on comparatives and related matters. Unpublished manuscript, University of Texas-Austin. Heim, Irene (). Anaphora and semantic interpretation: A reinterpretation of Reinhart’s approach, in Orin Percus and Uli Sauerland (eds), The interpretive tract. Cambridge, MA: MIT Working Papers in Linguistics, –. Heim, Irene (). Degree operators and scope, in Brendan Jackson and Tanya Matthews (eds), SALT : Proceedings of the th Semantics and Linguistic Theory Conference. Ithaca, NY: CLC Publications, –. Heim, Irene (). Remarks on comparative clauses as generalized quantifiers. Ms. MIT. Heim, Irene (). Forks on the road to Rule I, in Muhammad Abdurrahman, Anisa Schardi, and Martin Walkow (eds), NELS : Proceedings of the th annual meeting of the North East Linguistic Society. Amherst, MA: GLSA, –. Heim, Irene and Angelika Kratzer (). Semantics in generative grammar. Malden, MA: Blackwell. Heine, Lena (). Non-coordination-based ellipsis from a Construction Grammar perspective: The case of the coffee construction, Cognitive Linguistics (): –.





Hendriks, Petra (). The coordination-like structure of comparatives, in Frank Drijkoningen and Ans van Kemenade (eds), Linguistics in the Netherlands . Amsterdam and Philadelphia: John Benjamins, –. Hendriks, Petra (). Comparatives and categorial grammar. PhD thesis, University of Groningen. Hendriks, Petra (). Coherence relations, ellipsis and contrastive topics, Journal of Semantics : –. Hendriks, Petra and Jennifer Spenader (). Why be silent? Some functions of ellipsis in natural language, in Jennifer Spenader and Petra Hendriks (eds), Proceedings of the Cross-modular approaches to ellipsis (ESSLLI), –. Hennig, Mathilde (ed.) (). Die Ellipse: Neue Perspektive auf ein altes Phänomen. Berlin: De Gruyter. Heringer, Hans (). Deutsche Syntax dependentiell. Tübingen: Stauffenberg. Hernández, Ana Carrera (). Gapping as a syntactic dependency, Lingua (): –. Hernanz, María Luisa and José María Brucart (). La sintaxis I. Barcelona: Crítica. Hestvik, Arild (). Reflexives and ellipsis. Natural Langauge Semantics (): –. Herring, Jon, M. Carmen Parafita Couto, Moro Quintanilla, and Mónica Moro Quintanilla (). ‘When I went to Canada, I saw the madre’: Evaluating two theories’ predictions about codeswitching between determiners and nouns using Spanish-English and Welsh-English bilingual corpora, International Journal of Bilingual Education and Bilingualism (): –. Heycock, Caroline and Anthony Kroch (). Verb movement and coordination in a dynamic theory of licensing, The Linguistic Review (–): –. Hilbert, David and Paul Bernays (). Grundlagen der Mathematik II. Berlin: Julius Springer. Hinds, John (). On the status of the VP node in Japanese, Language Research : –. Hiraiwa, Ken and Shinichiro Ishihara (). Missing links: Clefts, sluicing and ‘no da’ construction in Japanese, in Tania Ionin, Heejeong Ko, and Andrew Nevins (eds), The MIT Working Papers in Linguistics : Proceedings of the nd HUMIT Student Conference in Language Research (HUMIT ). Cambridge, MA: MITWPL, –. Hiraiwa, Ken and Shinichiro Ishihara (). Syntactic metamorphosis: Clefts, sluicings, and in-situ focus in Japanese, Syntax (): –. Hirsch, Aron (). A case for conjunction reduction. Ms. MIT. Hirsch, Aron and Michael Wagner (). Right Node Raising, scope, and plurality, in Thomas Brochhagen, Floris Roelofsen and Nadine Theiler (eds), Proceedings of the th Amsterdam Colloquium. Amsterdam: ILLC, –. Hirschberg, Julia (). Pragmatics and intonation, in Laurence R. Horn and Gregory Ward (eds), The Handbook of Pragmatics. Malden, MA: Blackwell Publishing, –. Hirschberg, Julia and Gregory Ward (). Accent and bound anaphora, Cognitive Linguistics (): –. Hirschbuhler, Paul (). VP deletion and across-the-board quantifier scope, in James Pustejovsky and Peter Sells (eds), NELS : Proceedings of the th annual meeting of the North East Linguistic Society. Amherst, MA: GLSA, –. Hirst, Graeme (). Discourse-oriented anaphora resolution in natural language understanding: A review, American Journal of Computational Linguistics (): –. Hobbs, Jerry (). Literature and cognition. Technical report, CLSI Lecture Notes . Hobbs, Jerry and Andrew Kehler (). A theory of parallelism and the case of VP ellipsis, in Philip R. Cohen and Wolfgang Wahlster (eds), Proceedings of the th annual meeting of the Association for Computational Linguistics. Stroudsburg, PA: Association for Computational Linguistics, –. Hoeks, John C. J., Gisela Redeker, and Petra Hendriks (). Fill the gap! Combining pragmatic and prosodic information to make gapping easy, Journal of Psycholinguistic Research : –. Hoeks, John C. J., Gisela Redeker, and Louisa J. Zijlstra (). The predominance of non-structural factors in the processing of gapping sentences, in Ron Sun (ed.), Proceedings of the th Annual Conference of the Cognitive Science Society. Philadelphia: Taylor & Francis, –.





Hoeksema, Jack (). Negative polarity and the comparative, Natural Language and Linguistic Theory (): –. Hoeksema, Jack (). Pseudogapping: Its syntactic analysis and cumulative effects on its acceptability, Research on Language and Computation (): –. Hofmann, Ronald T. (). Past tense replacement and the modal system, in James D. McCawley (ed.), Notes from the linguistic underground. New York: Academic Press, –. Hofmeister, Philip (). A linearization accout of either . . . or constructions, Natural Language and Linguistic Theory (): –. Hofmeister, Philip and Ivan A. Sag (). Cognitive constraints and island effects, Language (): –. Hofmeister, Philip, Laura Staum Casasanto, and Ivan A. Sag (). Islands in the grammar? Standards of evidence, in Jon Sprouse and Norbert Hornstein (eds), Experimental syntax and island effects. Cambridge: Cambridge University Press, –. Hofmeister, Philip, T. Florian Jaeger, Ivan A. Sag, Inbal Arnon, and Neal Snider (). Locality and accessibility in wh-questions, in Sam Featherston and Wolfgang Sternefeld (eds), Roots: Linguistics in search of its evidential base. Berlin: Mouton de Gruyter, –. Hogue, Alan (). VP ellipsis exhibits structural priming. Poster presented at the th Annual Meeting of the Linguistic Society of America. Minneapolis, MN. Höhle, Tilman N. (). On reconstruction and coordination, in Hubert Haider and Klaus Netter (eds), Representation and derivation in the theory of grammar, vol. . Dordrecht: Kluwer Academic Publishers, –. Hoji, Hajime (). Japanese clefts and reconstruction/chain binding effects. Talk presented at WCCFL VI: Sixth West Coast Conference on Formal Linguistics held at University of Arizona. Hoji, Hajime (a). Demonstrative binding and Principle B, in Jill Beckman (ed.), NELS : Proceedings of the annual meeting of the North East Linguistic Society. Amherst, MA: GLSA, –. Hoji, Hajime (b). Theories of anaphora and aspects of Japanese syntax. Ms. USC. Hoji, Hajime (a). Sloppy identity and formal dependency, in Brian Agbanyani and Sze-Wing Tang (eds), Proceedings from the th West Coast Conference on Formal Linguistics. Somerville, MA: Cascadilla Press, –. Hoji, Hajime (b). Sloppy identity and principle B, in Hans Bennis, Pierre Pica, and Johan Rooryck (eds), Atomism and binding. Dordrecht: Foris Publications, –. Hoji, Hajime (). Null object and sloppy identity in Japanese, Linguistic Inquiry (): –. Hoji, Hajime (). Surface and deep anaphora, sloppy identity, and experiments in syntax, in Andrew Barss (ed.), Anaphora: A reference guide. Malden, MA: Oxford University Press, –. Hoji, Hajime (). On Saito et al.  (and Saito and Murasugi ). Class handout at USC. Hoji, Hajime (). Gengo kagaku o mezasite [Aiming at linguistic science]: Issues on anaphora in Japanese. Shiga: Oshumi Shoten. Hoji, Hajime (). Language faculty science. Cambridge: Cambridge University Press. Hoji, Hajime (). Issues on anaphora in Japanese. Shiga: Oshumi Shoten. Hoji, Hajime, Satoshi Kinsui, Yukinori Takubo, and Aymui Ueyama (). Demonstratives, bound variables, and reconstruction effects, in Proceedings of the Nanzan GLOW. Nagoya: Nanzan University, –. Holmberg, Anders (). Word order and syntactic features. PhD thesis, Stockholm. Holmberg, Anders (). The syntax of yes and no in Finnish, Studia Linguistica : –. Holmberg, Anders (). The syntax of yes and no. Oxford: Oxford University Press. Hopper, P. and Elizabeth Closs Traugott (). Grammaticalization. Cambridge: Cambridge University Press. Hornstein, Norbert (a). Logical Form: From GB to Minimalism. Cambridge, MA: Basil Blackwell. Hornstein, Norbert (b). Putting truth into Universal Grammar, Linguistics and Philosophy (): –.





Hornstein, Norbert, Howard Lasnik, and Juan Uriagereka (). The dynamics of islands: Speculations on the locality of movement, Linguistic Analysis : –. Horvath, Julia (). Is focus movement driven by stress? Approaches to Hungarian : –. Hough, Julian (). Incremental semantics driven natural language generation with self-repairing capability, in Proceedings of the Student Research Workshop associated with RANLP . Hissar, Bulgaria, –. Hough, Julian (). Modelling incremental self-repair processing in dialogue, PhD thesis, Queen Mary University of London. Hough, Julian and Matthew Purver (). Processing self-repairs in an incremental type-theoretic dialogue system, in Proceedings of the th SemDial Workshop on the Semantics and Pragmatics of Dialogue (SeineDial). France, –. Houser, Michael (). The syntax and semantics of do so anaphora. PhD thesis, University of California at Berkeley. Houser, Michael, Line Mikkelsen, and Maziar Toosarvandani (). Verb phrase pronominalization in Danish: Deep or surface anaphora?, in Erin Brainbridge and Brian Agbanyani (eds), Proceedings of the th Western Conference on Linguistics. Fresno, CA: Department of Linguistics, –. Howes, Christine, Matthew Purver, Pat G. T. Healey, Greg J. Mills, and Eleni Gregoromichelaki (). On incrementality in dialogue: Evidence from compound contributions, Dialogue and Discourse (): –. Hoyt, Frederick and Alexandra Teodorescu (). How many types of sluicing, and why: Single and multiple sluicing in Romanian, English, and Japanese, in Jason Merchant and Andrews Simpson (eds), Sluicing: Cross-linguistic perspectives. Oxford: Oxford University Press, –. Huang, C.-T. James (). Logical relations in Chinese and the theory of grammar. PhD thesis, Massachusetts Institute of Technology. Huang, C.-T. James (). On the distribution and reference of empty pronouns, Linguistic Inquiry : –. Huang, C.-T. James (a). Comments on Hasegawa’s paper, in Tawa Wako and Mineharu Nakayama (eds), Proceedings of Japanese Syntax Workshop Issues on Empty Categories. New London: Connecticut College, –. Huang, C.-T. James (b). Wo pao de kuai and Chinese phrase structure, Language : –. Huang, C.-T. James (). Remarks on the status of the null object, in Robert Freidin (ed.), Principles and parameters in comparative grammar. Cambridge, MA: MIT Press, –. Hubert, Anja (). The acquisition of ellipsis and the structure of the noun phrase. Master’s thesis, University of Potsdam. Huddleston, Rodney (). Some theoretical issues in the description of the English verb, Lingua : –. Huddleston, Rodney and Geoffrey K. Pullum (). The Cambridge grammar of the English language. Cambridge: Cambridge University Press. Hudson, Richard A. (a). Arguments for a non-transformational grammar. Chicago: University of Chicago Press. Hudson, Richard A. (b). Conjunction reduction, gapping and right-node raising, Language (): –. Hudson, Richard A. (). Gerunds without phrase structure, Natural Language and Linguistic Theory : –. Hudson, Richard A. (). Language networks: The new Word Grammar. Oxford: Oxford University Press. Hudson, Richard A. (). An introduction to Word Grammar. Cambridge: Cambridge University Press. Huet, Gérard P. (). A unification algorithm for typed λ-calculus, Theoretical Computer Science (): –.





Hulsey, Sarah (). Focus sensitive coordination. PhD thesis, MIT. Hulsey, Sarah and Uli Sauerland (). Sorting out relative clauses: A reply to Bhatt, Natural Langauge Semantics (): –. Hurford, J. R. (). The origins of grammar: Language in the light of evolution. Volume II: The origins of grammar. Oxford: Oxford University Press. Hyman, Larry M. and Francis X. Katamba (). The argument in Luganda: Syntax or pragmatics?, in Sam Mchombo (ed.), Theoretical aspects of Bantu grammar. Stanford: CSLI, –. Ince, Atakan (). Dimensions of ellipsis: Investigations in Turkish. PhD thesis, University of Maryland, College Park. Ince, Atakan (). Sluicing in Turkish, in Jason Merchant and Andrew Simpson (eds), Sluicing: Crosslinguistic perspectives. Oxford: Oxford University Press, –. Inoue, Kazuko (). Henkei bunpoo to nihongo [Transformational grammar and Japanese]. Tokyo: Taishukan. Inoue, Kazuko (). Nihongo-no bunpoo kisoku [Grammatical rules in Japanese]. Tokyo: Taishukan. Ionin, Tania (). The one girl who was kissed by every boy: Scope, scrambling, and discourse function in Russian, in Marjo van Koppen, Joanna Sio, and Mark de Vos (eds), Proceedings of ConSOLE X. Leiden: SOLE, –. Ionin, Tania and Ora Matushansky (). More than one comparative in more than one Slavic language: An experimental investigation, in Proceedings of the th Meeting of Formal Approaches to Slavic Linguistics (FASL ). Ann Arbor: Michigan Slavic Publications. Ionin, Tania, Ora Matushansky, and Eddy G. Ruys (). Parts of speech: Towards a unified semantics for partitives, in Christopher Davis, Amy Rose Deal, and Youri Zabbal (eds), NELS : Proceedings of the th annual meeting of the North East Linguistic Society. Amherst, MA: GLSA, –. Iseda, Takako (). On focused NPs with nominative case particle -ga in Japanese, in M. Kelepir and B. Öztürk (eds), MIT Working Papers in Linguistics : Proceedings of the nd Workshop on Altaic Formal Linguistics (WAFL ). Cambridge, MA: MITWPL, –. Ishihara, Shinichiro (). The clausemate condition in Japanese multiple clefts, in Matt Tucker, Anie Thompson, Olver Northrup, and Ryan Bennett (eds), Proceedings of FAJL : Formal Approaches to Japanese Linguistics. Cambridge, MA: MITWPL, –. Iwakura, Kunihiro (). The auxiliary system in English, Linguistic Analysis : –. Izvorski, Roumyana (). A solution to the subcomparative paradox, in José Camacho, Lina Choueiri, and Maki Watanabe (eds), Proceedings from the th West Coast Conference on Formal Linguistics. Stanford: CSLI, –. Jackendoff, Ray (). Gapping and related rules, Linguistic Inquiry (): –. Jackendoff, Ray (). Semantic interpretation in Generative Grammar. Cambridge, MA: MIT Press. Jackendoff, Ray (). X0 syntax. Cambridge, MA: MIT Press. Jackendoff, Ray (). Semantic structures. Cambridge, MA: MIT Press. Jackendoff, Ray (). Foundations of language. Oxford: Oxford University Press. Jacobson, Pauline (a). Antecedent contained deletion in a variable-free semantics, in Chris Barker and David Dowty (eds), SALT : Proceedings of the nd Semantics and Linguistic Theory Conference. Ithaca, NY: CLC Publications, –. Jacobson, Pauline (b). Flexible categorial grammars: Questions and prospects, in Robert D. Levine (ed.), Formal grammar: Theory and implementation. Oxford: Oxford University Press, –. Jacobson, Pauline (c). Raising without movement, in Richard K. Larson, Sabine Iatridou, Utpal Lahiri, and James Higginbotham (eds.), Control and Grammar. Dordrecht: Kluwer Academic Publishers, –. Jacobson, Pauline (). Binding connectivity in copular sentence, in Mandy Harvey and Lynn Santelmann (eds), SALT : Proceedings of the th Semantics and Linguistic Theory Conference. Ithaca, NY: CLC Publications, –.





Jacobson, Pauline (). Towards a variable-free semantics, Linguistics and Philosophy (): –. Jacobson, Pauline (a). Paycheck pronouns, Bach-Peters sentences, and variable-free semantics, Natural Language Semantics (): –. Jacobson, Pauline (b). Paychecks, stress, and variable free semantics, in T. Matthews and B. Jackson (eds), SALT : Proceedings of the th Semantics and Linguistic Theory Conference. Ithaca, NY: CLC Publications, –. Jacobson, Pauline (). Binding without pronouns (and pronouns without binding), in G.-J. Kruiff and Richard Oehrle (eds), Binding and resource sensitivity. Dordrecht: Kluwer Academic Publishers, –. Jacobson, Pauline (). Direct compositionality and variable-free semantics: The case of Antecedent Contained Deletion, in Kyle Johnson (ed.), Topics in ellipsis. Cambridge: Cambridge University Press, –. Jacobson, Pauline (). Compositional semantics: An introduction to the syntax-semantics interface. Oxford: Oxford University Press. Jacobson, Pauline (a). Online appendix to: The short answer: Implications for direct compositionality and vice versa, Language : s–s. Jacobson, Pauline (b). The short answer: Implications for direct compositionality and vice versa, Language : –. Jäger, Lena A., Lena Benz, Jens Roeser, Brian Dillon, and Shravan Vasishth (). Teasing apart retrieval and encoding interference in the processing of anaphors, Frontiers in Psychology , . Jaeger, T. Florian and Harry Tily (). On language ‘utility’: Processing complexity and communicative efficiency, Wiley Interdisciplinary Reviews Cognitive Science : –. Jake, Janice L. and Carol Myers-Scotton (). Which language? Participation potentials across lexical categories in codeswitching, in Ludmila Isurin, Donald Winford, and Kees de Bot (eds), Multidisciplinary approaches to code-switching. Amsterdam and Philadelphia: John Benjamins, –. Jake, Janice L., Carol Myers-Scotton, and Steven Gross (). Making a minimalist approach to codeswitching work: Adding the matrix language, Bilingualism: Language and Cognition (): –. Jake, Janice L., Carol Myers-Scotton, and Steven Gross (). A response to MacSwan (): Keeping the matrix language, Bilingualism: Language and Cognition (): –. James, William (). The principles of psychology, vol. I. New York: Henry Holt and Co. Jantunen, Tommi (). The equative sentence in Finnish Sign Language, Sign Language and Linguistics : –. Jantunen, Tommi (). Fixed and free: Order of the verbal predicate and its core arguments in declarative transitive clauses in Finnish Sign Language, SKY Journal of Linguistics : –. Jantunen, Tommi (). Tavu ja lause: Tutkimuksia kahden sekventiaalisen perusyksikön olemuksesta suomalaisessa viittomakielessä [Syllable and sentence: Studies on the nature of two sequential basic units in FinSL]. Jyväskylä: University of Jyväskylä. Jantunen, Tommi (). Suomalaisen viittomakielen pääsanaluokat [The main parts of speech in FinSL], in Tommi Jantunen (ed.), Näkökulmia viittomaan ja viittomistoon [Perspectives on sign and lexicon]. Jyväskylä: University of Jyväskylä, –. Jantunen, Tommi (). Ellipsis in Finnish Sign Language, Nordic Journal of Linguistics : –. Jantunen, Tommi (). Rinnastuksen prosodiaa suomalaisessa viittomakielessä [The prosody of clausal coordination in FinSL], in Mona Lehtinen and Unto K. Laine (eds), XXIX Fonetiikan päivät, Espoo . .., Julkaisut—Papers [Proceedings of the XXIX Phonetics Symposium]. Helsinki: Aalto University, –. Jantunen, Tommi (). Clausal coordination in Finnish Sign Language, Studies in Language (): –.





Jayaseelan, Karattuparambil A. (). Incomplete VP deletion and gapping, Linguistic Analysis (–): –. Jayaseelan, Karattuparambil A. (). IP-internal topic and focus phrases, Studia Linguistica (): –. Jensen, Britta and Rosalind Thornton (). Fragments of child grammar, in M. João Freitas and Anna Gavarró Alguéro (eds), Language acquisition and development: Papers from GALA . Cambridge: Cambridge Scholars Publishing, –. Johannessen, Janne Bondi (). Coordination. Oxford: Oxford University Press. Johnson, Kyle (). When verb phrases go missing, GLOT (): –. Johnson, Kyle (a). Few dogs eat Whiskers or cats Alpo, in Kiyomi Kusumoto and Elisabeth Villalta (eds), University of Massachusetts Occasional Papers . Amherst, MA: UMOP, –. Johnson, Kyle (b). Gapping determiners, in Kerstin Schwabe and Nina Zhang (eds), Ellipsis in conjunction. Tübingen: Niemeyer, –. Johnson, Kyle (a). Sluicing and constraints on quantifier scope, GLOT International (): –. Johnson, Kyle (b). What VP ellipsis can do, and what it can’t, but not why, in Mark Baltin and Chris Collins (eds), The handbook of contemporary syntactic theory. Oxford: Blackwell Publishers, –. Johnson, Kyle (). Restoring exotic coordinations to normalcy, Linguistic Inquiry (): –. Johnson, Kyle (a). In search of the English middle field. Ms. University of Massachusetts, Amherst. Johnson, Kyle (b). How to be quiet, in Nikki Adams, Adam Cooper, Fey Parrill, and Thomas Wier (eds), Proceedings from the th annual meeting of the Chicago Linguistic Society. Chicago: Chicago Linguistic Society, –. Johnson, Kyle (). Gapping, in M. Evereart, H. C. van Reimsdijk, R. Goedemans, and B. Holledbrandse (eds), The Blackwell companion to syntax, vol. II. Malden, MA and Oxford: Blackwell, –. Johnson, Kyle (a). Introduction, in Kyle Johnson (ed.), Topics in ellipsis. Cambridge: Cambridge University Press, –. Johnson, Kyle (ed.) (b). Topics in ellipsis. Cambridge: Cambridge University Press. Johnson, Kyle (). Gapping is not (VP-)ellipsis, Linguistic Inquiry (): –. Johnson, Kyle (). Gapping. Ms. University of Massachusetts at Amherst. Johnson, Kyle and Satoshi Tomioka (). Lowering and mid-size clauses, in Graham Katz, ShinSook Kim, and Winhart Haike (eds), Proceedings of the  Tübingen workshop on reconstruction. Tübingen, Germany: Sprachteoretische Grundlagen für die Computer Linguistik, –. Johnson-Laird, Philip N. (). Mental models. Cambridge, MA: Harvard University Press. Johnston, Trevor and Adam Schembri (). On defining lexeme in a signed language, Sign Language and Linguistics : –. Johnston, Trevor and Adam Schembri (). Australian Sign Language: An introduction to sign language linguistics. Cambridge: Cambridge University Press. Joshi, Aravind (). Processing of sentences with intrasentential code switching, in David R. Dowty, Lauri Karttunen, and Arnold Zwicky (eds), Natural language parsing: Psychological, computational, and theoretical perspectives. Cambridge: Cambridge University Press, –. Jun, Sun-Ah (). Korean intonational phonology and prosodic transcription, in Sun-Ah Jun (ed.), Prosodic typology: The phonology of intonation and phrasing. Oxford: Oxford University Press, –. Kaan, Edith, Carlie Overfelt, Do Tromp, and Frank Wijnen (). Processing gapped verbs, Journal of Psycholinguistic Research : –. Kaan, Edith, Frank Wijnen, and Tamara Y. Swaab (). Gapping: Electrophysiological evidence for immediate processing of ‘missing’ verbs in sentence comprehension, Brain and Language (): –.





Kadowaki, Makoto (). N’-ellipsis reconsidered, in Yukio Otsu (ed.), Proceedings of the th Tokyo Conference on Psycholinguistics. Tokyo: Hituzi Syobo, –. Kallestinova, Elena (). Aspects of word order in Russian. PhD thesis, University of Iowa. Kameyama, Megumi (). Zero anaphora: The case of Japanese. PhD thesis, Stanford University. Kameyama, Megumi (). Stressed and unstressed pronouns: Complementary preferences, in Peter Bosch and Rob van der Sandt (eds), Focus: Linguistic, cognitive and computational perspectives. Cambridge: Cambridge University Press, –. Kamio, Akio (). Meisiku-no koozoo [Noun phrase structure], in Kazuko Inoue (ed.), Nihongo-no kihon koozoo [Basic structures in Japanese]. Tokyo: Sanseido, –. Kaplan, David (). Demonstratives: An essay on the semantics, logic, metaphysics, and epistemology of demonstratives and other indexicals, in Joseph Almog, John Perry, and Howard Wettstein (eds), Themes from Kaplan. Oxford: Oxford University Press, –. Karimi, Hossein, Kumiko Fukukura, Fernanda Ferreira, and Martin J. Pickering (). The effect of noun phrase length on the form of referring expressions, Memory and Cognition : –. Karimi, Simin (). A minimalist approach to scrambling: Evidence from Persian. Berlin: Mouton de Gruyter. Katz, Jerrold J. and Paul Postal (). An integrated theory of linguistic descriptions. Cambridge, MA: MIT Press. Kay, Paul (). Contextual operators: respective, respectively, and vice versa, in Kira Hall, Michael Meacham, and Richard Shapiro (eds), Proceedings of the th annual meeting of the Berkeley Linguistics Society. Berkeley: Berkeley Linguistics Society, –. Kay, Paul (). English subjectless tagged sentences, Language (): –. Kayne, Richard S. (). Connectedness, Linguistic Inquiry (): –. Kayne, Richard S. (). Connectedness and binary branching. Dordrecht: Foris Publications. Kayne, Richard S. (). The antisymmetry of syntax. Cambridge, MA: MIT Press. Kayne, Richard S. (). The English complementizer of, The Journal of Comparative Germanic Linguistics (): –. Kayne, Richard S. (). One and ones as complex determiners. Ms. New York University. Kazanina, Nina, Ellen Lau, Moti Lieberman, Masaya Yoshida, and Colin Phillips (). The effect of syntactic constraints on the processing of backwards anaphora, Journal of Memory and Language : –. Kazenin, Konstantin (). On coordination of wh-phrases in Russian. Ms. University of Tuebingen. Kazenin, Konstantin (). Polarity in Russian and typology of predicate ellipsis. Ms. Moscow State University. Kazenin, Konstantin (). O nekotoryx ograničenijax na ellipsis v russkom jazyke [On some restrictions on ellipsis in Russian], Voprosy jazykoznanija : –. Kazenin, Konstantin (). Russian gapping: Against ATB, in Gerhild Zybatow, Philip Dudchuk, Serge Minor, and Ekaterina Pshehotskaya (eds), Formal studies in Slavic linguistics. Bern: Peter Lang, –. Keenan, Edward (). Names, quantifiers, and a solution to the sloppy identity problem, Papers in Linguistics (): –. Kehler, Andrew (a). A discourse copying algorithm for ellipsis and anaphora resolution, in Proceedings of the Sixth European Chapter of the Association for Computational Linguistics. Stroudsburg, PA: Association for Computational Linguistics, –. Kehler, Andrew (b). The effect of establishing coherence in ellipsis and anaphora resolution, in Proceedings of the st Conference of the Association for Computational Linguistics. Stroudsburg, PA: Association for Computational Linguistics, –. Kehler, Andrew (). Common topics and coherent situations: Interpreting ellipsis in the context of discourse inference. In Proceedings of the nd Conference of the Association for Computational Linguistics. Stroudsburg, PA: Association for Computational Linguistics, –.





Kehler, Andrew (). Current theories of centering for pronoun interpretation: A critical evaluation, Computational Linguistics : –. Kehler, Andrew (). Coherence and the resolution of ellipsis, Linguistics and Philosophy (): –. Kehler, Andrew (). Coherence, reference, and the theory of grammar. Stanford: CSLI Publications. Kehler, Andrew (). Contrastive topics and illusory sloppy interpretation in VP-ellipsis. Paper presented at the Annual Meeting of the Linguistic Society of America. Kehler, Andrew (). On QUD-based licensing of strict and sloppy ambiguities, in Sarah D’Antonio, Mary Moroney, and Carol-Rose Little (eds), Semantics and linguistic theory, vol. . Washington DC: LSA and CLC Publications, –. Kehler, Andrew and Daniel Büring (). Be bound or be disjoint!, in NELS : Proceedings of the th annual meeting of the North East Linguistic Society. Amherst, MA: GLSA, . Kehler, Andrew and Stuart Shieber (). Anaphoric dependencies in ellipsis, Computational Linguistics (): –. Kehler, Andrew and Gregory Ward (). Event reference and morphological transparency, in Proceedings of the th Western Conference on Linguistics (WECOL-). Department of Linguistics, California State University, Fresno, –. Kempson, Ruth, Ronnie Cann, Arash Eshghi, Eleni Gregoromichelaki, and Matthew Purver (). Ellipsis, in Shalom Lappin and Chris Fox (eds), Handbook of contemporary semantic theory. Hoboken: Wiley-Blackwell, –. Kempson, Ruth, Andrew Gargett, and Eleni Gregoromichelaki (). Clarification requests: An incremental account, in Ron Artstein and Laure Vieu (eds), Decalog : Proceedings of the th Workshop on the Semantics and Pragmatics of Dialogue (SemDial ). Kempson, Ruth, Wilfried Meyer-Viol, and Dov Gabbay (). VP ellipsis: Toward a dynamic, structural account, in Shalom Lappin and Elabbas Benmamoun (eds), Fragments: Studies in ellipsis and gapping. Oxford: Oxford University Press, –. Kempson, Ruth, Wilfried Meyer-Viol, and Dov Gabbay (). Dynamic Syntax: The flow of language understanding. Hoboken: Wiley-Blackwell. Kendon, Adam (). Gesture: Visible action as utterance. Cambridge: Cambridge University Press. Kenesei, István (). On Hungarian complementizers, in István Kenesei and Csaba Pléh (eds), Approaches to Hungarian . Szeged: JATE Press, –. Kenesei, István (). Subordinate clauses, in Ferenc Kiefer and Katalin É. Kiss (eds), Syntax and Semantics . New York: Academic Press, –. Kenesei, István (). Adjuncts and arguments in VP focus in Hungarian, Acta Linguistica Hungarica : –. Kenesei, István (). Criteria for auxiliaries in Hungarian, in István Kenesei (ed.), Argument structure in Hungarian. Budapest: Akadémiai Kiadó, –. Kenesei, István, Anna Fenyvesi, and Robert M. Vago (). Hungarian. Oxford: Routledge. Kennedy, Christopher (). Antecedent contained deletion and the syntax of quantification, Linguistic Inquiry (): –. Kennedy, Christopher (). Projecting the adjective: The syntax and semantics of gradability and comparison. New York: Garland Press. Kennedy, Christopher (). Comparative (sub)deletion and ranked, violable constraints in syntax, in Masako Hirotani, Andries Coetzee, Nancy Hall, and Ji-yung Kim (eds), NELS : Proceedings of the th annual meeting of the North East Linguistic Society. Amherst, MA: GLSA, –. Kennedy, Christopher (). Comparative deletion and optimality in syntax, Natural Language and Linguistic Theory (): –. Kennedy, Christopher (). Ellipsis and syntactic representation, in Kerstin Schwabe and Susanne Winkler (eds), The Interfaces: Deriving and interpreting omitted structures. Amsterdam and Philadelphia: John Benjamins, –.





Kennedy, Christopher (). Standards of comparison. Paper presented at the Colloque de Syntaxe et Sémantique de Paris,  October. Kennedy, Christopher (). Argument contained ellipsis, in Kyle Johnson (ed.), Topics in ellipsis. Cambridge: Cambridge University Press, –. Kennedy, Christopher (). Modes of comparison, in Malcolm Elliott, James Kirby, Osamu Sawada, Eleni Staraki, and Suwon Yoon (eds), Proceedings of the rd annual meeting of the Chicago Linguistic Society. Chicago: Chicago Linguistic Society, –. Kennedy, Christopher and Jeffrey Lidz (). A (covert) long distance anaphor in English, in Karine Megerdoomian and Leora Anne Bar-el (eds), Proceedings of the th West Coast Conference in Formal Linguistics. Somerville, MA: Cascadilla Press, –. Kennedy, Christopher and Jason Merchant (a). Attributive comparative deletion, Natural Language and Linguistic Theory (): –. Kennedy, Christopher and Jason Merchant (b). The case of the missing CP and the secret case, in Sandra Chung, James McCloskey, and Nathan Sanders (eds), Jorge Hankamer Webfest. UCSC. . Kentner, Gerrit, Caroline Féry, and Kai Alter (). Prosody in speech production and perception: The case of Right Node Raising in English, in Anita Steube (ed.), The discourse potential of underspecified structures: Event structures and information structures. Berlin: Mouton de Gruyter, –. Kerstens, Johan (). Over samentrekking, De Nieuwe Taalgids : –. Kerstens, Johan (). Bestaat gapping eigenlijk wel?, Spektator : –. Kertz, Laura (). Focus structure and acceptability in verb phrase ellipsis, in Natasha Abner and Jason Bishop (eds), Proceedings of the th West Coast Conference on Formal Linguistics. Somerville, MA: Cascadilla Press, –. Kertz, Laura (). Ellipsis reconsidered. PhD thesis, University of California, San Diego. Kertz, Laura (). Verb phrase ellipsis: The view from information structure, Language (): –. Keshet, Ezra (). Sloppy identity unbound, in Todd Snider (ed.), Semantics and linguistic theory . Ithaca: CLC Publications, –. Kester, Ellen-Petra (a). Adjectival inflection and the licensing of empty categories in DP, Journal of Linguistics : –. Kester, Ellen-Petra (b). The nature of adjectival inflection. Utrecht: LEd. Khamisi, Abdu M. (). A typology of gaps in Kiswahili sentences, Kiswahili : –. Kiguchi, Hirohisa and Rosalind Thornton (). Binding principles and ACD constructions in child grammars, Syntax (): –. Kim, Christina, Gregory M. Kobele, Jeffrey Runner, and John T. Hale (). The acceptability cline in VP ellipsis, Syntax : –. Kim, Christina and Jeffrey Runner (). Strict identity, coherence, and parallelism in VP ellipsis, in Ed Cormany, Satoshi Ito, and David Lutz (eds), Proceedings of the th Semantics and Linguistic Theory Conference. Ithaca, NY: CLC Publications, –. Kim, Christina and Jeffrey Runner (). Discourse structure and syntacic parallelism in VP ellipsis, UMass Occasional Papers in Linguistics : –. Kim, Jong-Bok (). Interactions between constructions and constraints in VP ellipsis and VP fronting, in Charles Chang et al. (eds), Proceedings of the th Annual Meeting of the Berkeley Linguistics Society. Berkeley: Berkeley Linguistics Society, –. Kim, Jong-Bok (). Similarities and differences between English VP ellipsis and VP fronting: An HPSG analysis, Studies in Generative Grammar (): –. Kim, Jong-Bok (). Syntactic and semantic identity in Korean sluicing: A direct interpretation approach, Lingua : –. Kim, Jong-Bok and Peter Sells (). English syntax: An introduction. Stanford: CSLI. Kim, Jeong-Seok (). Syntactic focus movement and ellipsis: A minimalist approach. PhD thesis, University of Connecticut.





Kim, Soowon (). Sloppy/strict identity, empty objects, and NP ellipsis, Journal of East Asian Linguistics (): –. Kimmelman, Vadim (). Information structure in Russian Sign Language and Sign Language of the Netherlands. PhD thesis, University of Amsterdam. Kimura, Hiroko (). A wh-in-situ strategy for sluicing, English Linguistics : –. King, Harold V. (). On blocking rules for contraction in English, Linguistic Inquiry (): –. King, Paul John (). A logical formalism for Head Driven Phrase Structure Grammar. PhD thesis, University of Manchester. King, Paul John (). An expanded logical formalism for Head Driven Phrase Structure Grammar. Technical report, Working Papers of the SFB . Kinsui, Satoshi (). Nihongo-no iwayuru N’ sakuzyo-ni tuite [On so-called N’-deletion in Japanese], in Yasuaki Abe (ed.), The Reports for the Third Nanzan University International Symposium on Japanese Education and Linguistics. Nagoya: Nanzan University, –. Kirchner, Jesse (). Sluicing and quasisluicing in Mandarin. Ms. UCSC. Kitagawa, Chisato and Claudia Ross (). Prenominal modification in Chinese and Japanese, Linguistic Analysis : –. Kitagawa, Yoshihisa (). Subjects in Japanese and English. PhD thesis, University of Massachusetts, Amherst. Kitagawa, Yoshihisa (). Copying identity, Natural Language and Linguistic Theory (): –. Kitahara, Hisatsugu (). Eliminating * as a feature (of traces), in Samuel Epstein and Norbert Hornstein (eds), Working minimalism. Cambridge, MA: MIT Press, –. Kizu, Mika (a). A note on sluicing in wh-in-situ languages, MIT Working Papers in Linguistics : –. Kizu, Mika (b). Sluicing in wh-in-situ languages, in Kora Singer, Randall Eggert, and Gregory Anderson (eds), Proceedings of the Thirty-Third Regional Meeting of the Chicago Linguistic Society. Chicago: Chicago Linguistics Society, –. Klavans, Judith (). The syntax of code-switching: Spanish and English, in Larry D. King and Catherine A. Maley (eds), Selected papers from the XIIIth Linguistic Symposium on Romance Language. Amsterdam and Philadelphia: John Benjamins, –. Klein, Ewan (). A semantics for positive and comparative adjectives, Linguistics and Philosophy (): –. Klein, Ewan (). Comparatives, in Arnim von Stechow and Dieter Wunderlich (eds), Semantik/ Semantics: Ein internationales Handbuch der zeitgenössischen Forschung/An International Handbook of Contemporary Research. Berlin and New York: Mouton de Gruyter, –. Klein, Elaine C. (). Toward second language acquistion: A study of Null Prep. Dordrecht: Kluwer Academic Publishers. Klein, Udo, Dolgor Guntsetseg, and Klaus von Heusinger (). Case in conflict: Embedded subjects in Mongolian, in Monique Lamers and Peter de Swart (eds), Case, word order, and prominence. Dordrecht: Springer, –. Klein, Wolfgang (). Ellipse, Fokusgliederung und thematischer Stand, in Reinhard MeyerHermann and Hannes Rieser (eds), Ellipsen und fragmentarische Ausdrücke. Tübingen: Niemeyer, –. Klein, Wolfgang (). Ellipse, in Joachim Jacobs, Arnim von Stechow, Wolfgang Sternefeld, and Theo Vennemann (eds), Syntax: An international handbook of contemporary research. Berlin and New York: Mouton de Gruyter, –. Klein, Wolfgang and Christiane von Stutterheim (). Quaestio und referentielle Bewegung in Erzählungen, Linguistische Berichte : –. Vander Klok, Jozina (). Diagnosing VP-ellipsis in Javanese: Evidence for a non-movement account, in Henrison Hsieh (ed.), AFLA : Proceedings of the nd meeting of the Austronesian Formal Linguistics Association. Canberra: Australian National University, –.





Kluck, Marlies (). Good neighbors or far friends. Matching and proximity effects in Dutch Right Node Raising, Groninger Arbeiten zur Germanistischen Linguistik : –. Kluck, Marlies (). Sentence amalgamation. PhD thesis, University of Groningen. Kluck, Marlies (). Merchant’s wrinkle: The ban on Dutch bare R-pronoun ellipsis remnants. Paper presented at the West Coast Conference on Formal Linguistics  (WCCFL). Kluck, Marlies and Mark de Vries (). Cumulative rightward processes, in Gert Webelhuth, Manfred Sailer, and Heike Walker (eds), Rightward movement in a comparative perspective. Amsterdam and Philadelphia: John Benjamins, –. Kluender, Robert (). On the distinction between strong and weak islands: A processing perspective, in Peter W. Culicover and Louise McNally (eds), Syntax and semantics : The limits of syntax. New York: Academic Press, –. Kluender, Robert and Simone Gieselman (). What’s negative about negative islands? A reevaluation of extraction from weak island contexts, in Jon Sprouse and Norbert Hornstein (eds), Experimental syntax and island effects. Cambridge: Cambridge University Press, –. Kobele, Gregory M. (). Generating copies: An investigation into structural identity in language and grammar. PhD thesis, University of California, Los Angeles. Kobele, Gregory M. (a). Eliding the derivation: A minimalist formalization of ellipsis, in Stefan Müller (ed.), Proceedings of the th International Conference on Head-Driven Phrase Structure Grammar. Stanford: CSLI, –. Kobele, Gregory M. (b). Ellipsis: Computation of, WIREs Cognitive Science (): –. Kobele, Gregory M. (). LF-copying without LF, Lingua : –. Koeneman, Olaf, Sergio Baauw, and Frank Wijnen (). Reconstruction in VP-ellipsis: Reflexive vs. nonreflexive predicates. Poster presented at the th Annual Conference on Human Sentence Processing. Koizumi, Masatoshi (). Phrase structure in Minimalist syntax. PhD thesis, Massachusetts Institute of Technology, Cambridge, MA. Kolokonte, Marina (). Bare Argument Ellipsis and information structure. PhD thesis, Newcastle University. Konietzko, Andreas (). The syntax and information structure of Bare Argument Ellipsis in English and German: Experimental and theoretical evidence. PhD thesis, Eberhard-Karls-Universität Tübingen. Konietzko, Andreas, Janina Radó, and Susanne Winkler (forthcoming). Focus constraints on relative clause antecedents in sluicing, in Sam Featherston, Robin Hörnig, Sophia von Wietersheim, and Susanne Winkler (eds), Information structure and semantic processing. Berlin and New York: Mouton de Gruyter. Konietzko, Andreas and Susanne Winkler (). Contrastive ellipsis: Mapping between syntax and information structure, Lingua : –. Koopman, Hilda Judith and Dominique Sportiche (). The position of subjects, Lingua : –. Koopman, Hilda Judith and Anna Szabolcsi (). Verbal complexes. Cambridge, MA: MIT Press. Kornfeld, Laura and Andrés Saab (). Nominal ellipsis and morphological structure in Spanish, in Reineke Bok-Benneman, Bart Hollebrandse, Brigitte Kampers-Manhe, and Petra Sleeman (eds), Romance languages and linguistic theory . Amsterdam and Philadelphia: John Benjamins, –. Kornfeld, Laura and Andrés Saab (). Hacía una tipología de las anáforas nominales en español. Talk given at the III Encuentro de Gramática Generativa. Koster, Jan (). Dutch as an SOV language, Linguistic Analysis : –. Koster, Jan (). Domains and dynasties. Dordrecht: Foris Publications. Koulidobrova, Elena (). When the quiet surfaces: ‘Transfer’ of argument omission in the speech of ASL-English bilinguals. PhD thesis, University of Connecticut. Koutsoudas, Andreas (). Gapping, conjunction reduction, and coordinate deletion, Foundations of Language (): –.





Kraak, Albert (). Presuppositions and the analysis of adverbs. Ms. MIT. Krahmer, Emiel and Reinhard Muskens (). Negation and disjunction in discourse representation theory, Journal of Semantics : –. Kranendonk, Huib (). Quantificational constructions in the Dutch nominal domain: Facets of Dutch microvariation. PhD thesis, Utrecht University. Kratzer, Angelika (). Severing the external argument from its verb, in Johan Rooryck and Laurie Zaring (eds), Phrase structure and the lexicon. Dordrecht: Kluwer Academic Publishers, –. Kratzer, Angelika and Junko Shimoyama (). Indeterminate pronouns: The view from Japanese, in Yukio Otsu (ed.), Proceedings of the Tokyo conference on psycholinguistics, vol. . Tokyo: Hituzi Syobo, –. Krifka, Manfred (). A compositional semantics for multiple focus constructions, in Steven Moore and Adam Zachary Wyner (eds), Proceedings of the First Semantics and Linguistic Theory Conference. Ithaca, NY: Cornell University, –. Krifka, Manfred (). Association with focus phrases, in V. Molnár and Susanne Winkler (eds), The architecture of focus. Studies in Generative Grammar . Berlin and New York: Mouton de Gruyter, –. Krifka, Manfred (). Basic notions of information structure, Acta Linguistica Hungarica (–): –. Kubota, Yusuke (). Nonconstituent coordination in Japanese as constituent coordination: An analysis in Hybrid Type-Logical Grammar, Linguistic Inquiry : –. Kubota, Yusuke and Robert Levine (). Gapping as like-category coordination, in Denis Bechet and Alexander Dikovsky (eds), Logical aspects of computational linguistics. Lecture notes in computer science . Berlin: Springer, –. Kubota, Yusuke and Robert Levine (). Gapping as hypothetical reasoning, Natural Language and Linguistic Theory : –. Kubota, Yusuke and Robert Levine (). Pseudogapping as pseudo-VP-ellipsis, Linguistic Inquiry (): –. Kučerová, I. and A. Neeleman (eds) (). Contrasts and positions in information structure. Cambridge: Cambridge University Press. Kuno, Susumu (). Conditions for verb phrase deletion, Foundations of Language : –. Kuno, Susumu (). Gapping: A functional analysis, Linguistic Inquiry (): –. Kuno, Susumu (a). Danwa-no bunpoo. [Discourse grammar]. Tokyo: Taishukan. Kuno, Susumu (b). Japanese: A characteristic OV language, in Winfred Lehmann (ed.), Syntactic typology. Austin: University of Texas Press, –. Kuno, Susumu (). Newness of information and order of deletion, Cahiers Charles V: Recherches de l’Institut d’Anglais Charles : –. Kuno, Susumu (). Discourse deletion, in Susumu Kuno (ed.), Harvard Studies in Syntax and Semantics . Cambridge, MA: Harvard University, –. Kuno, Susumu (). Principles of discourse deletion: Case studies from English, Russian, and Japanese, Journal of Semantics : –. Kuno, Susumu (). Anaphora in Japanese, in Sige-Yuki Kuroda (ed.), Working Papers from the First SDF Workshop in Japanese Syntax. La Jolla: USCD, –. Kuno, Susumu and Kiyoko Masunaga (). Questions with wh-phrases in islands, University of Massachusetts Occasional Papers in Linguistics : –. Kunze, Jürgen (). Abhängigkeitsgrammatik. Studia Grammatica XII. Berlin: Akademie-Verlag. Kuroda, Sige-Yuki (). Generative grammatical studies in the Japanese language. PhD thesis, MIT. Kuroda, Sige-Yuki (). Whether we agree or not: A comparative syntax of English and Japanese, Linguisticae Investigationes : –. Kuroda, Sige-Yuki (). Japanese syntax and semantics. Dordrecht: Kluwer Academic Publishers. Kush, Dave (). Notes on gapping in Hindi-Urdu: Conjunct size and focus-parallelism, Linguistic Analysis (–): –.





Kush, Dave, Jeffrey Lidz, and Colin Phillips (). Relation-sensitive retrieval: Evidence from bound variable pronouns, Journal of Memory and Language : –. Kuwabara, Kazuki (). Multiple wh-phrases in elliptical clauses and some aspects of clefts with multiple foci, in Masatoshi Koizumi, Masayuki Oishi, and Uli Sauerland (eds), MIT Working Papers in Linguistics : Formal Approaches to Japanese Linguistics . Cambridge, MA: MITWPL, –. Kuwabara, Kazuki (). On the properties of truncated clauses in Japanese, in Kazuko Inoue (ed.), Researching and verifying an advanced theory of human language. Chiba: Kanda University of International Studies, –. Labov, William (). Contraction, deletion and the inherent variability of the English copula, Language (): –. Labov, William (). The notion of system in Creole studies, in Dell Hymes (ed.), Pidginization and creolization of language. Cambridge: Cambridge University Press, –. Laczkó, Tibor (). On elliptical noun phrases in Hungarian, in Miriam Butt and Tracy Holloway King (eds), Proceedings of the LFG ’ Conference. Stanford: CSLI, –. Ladd, D. Robert (). Intonational phonology. Cambridge: Cambridge University Press. Ladewig, Silva (). Creating multimodal utterances: The linear integration of gestures into speech, in Cornelia Müller, Alan Cienki, Ellen Fricke, Silva Ladewig, David McNeill, and Jana Bressem (eds), Body—Language—Communication. Berlin: Mouton de Gruyter, –. Laka, Itziar (). Negation in syntax: On the nature of functional categories and their projections. PhD thesis, Massachusetts Institute of Technology. Lakoff, George (). Pronouns and reference, in James D. McCawley (ed.), Syntax and semantics : Notes from the linguistic underground. New York: Academic Press, –. Lakoff, George (). Global rules, Language : –. Lakoff, George (). The arbitrary basis of transformational grammar, Language (): –. Lakoff, George (). Syntactic amalgams, in Michael Galy, Robert Fox, and Anthony Bruck (eds), Papers from the th regional meeting of the Chicago Linguistic Society. Chicago: University of Chicago, –. Lakoff, George (). Frame semantic control of the Coordinate Structure Constraint, in Anne M. Farley, Peter T. Farley, and Karl-Erik McCullough (eds), Proceedings of the nd annual meeting of the Chicago Linguistic Society . Chicago: Chicago Linguistic Society, –. Lakoff, George and Stanley Peters (). Phrasal conjunction and symmetric predicates, in David A. Reibel and Sandford A. Schane (eds), Modern studies in English: Readings in transformational grammar. Englewood Cliffs, NJ: Prentice Hall, –. Lakoff, Robin T. (). Ifs, ands, and buts about conjunction, in Charles J. Fillmore and D. Terence Langendoen (eds), Studies in linguistic semantics. New York: Holt, Rinehart and Winston, –. Lambrecht, Knud (). ‘What, me worry?’—‘Mad magazine sentences’ revisited, in Proceedings of the Annual Meeting of the Berkeley Linguistics Society, vol. . Berkeley: Berkeley Linguistics Society, –. Lambrecht, Knud (). Information structure and sentence form: Topic, focus, and the mental representations of discourse referents. Cambridge: Cambridge University Press. Lance, David (). Spanish-English code-switching, in Eduardo Hernández-Chávez, Andrew D. Cohen, and Anthony F. Beltramo (eds), El lenguaje de los chicanos. Arlington, VA: Center for Applied Linguistics, –. Landau, Idan (). Control in generative grammar: A research companion. Cambridge: Cambridge University Press. Langacker, Ronald W. (). On pronominalization and the chain of command, in David A. Reibel and Sanford A. Schane (eds), Modern Studies in English: Readings in Transformational Grammar. Englewood Cliffs, NJ: Prentice Hall, –. Lappin, Shalom (). The syntactic basis of ellipsis resolution, in Steve Berman and Arild Hestvik (eds), Proceedings of the Stuttgart Ellipsis Workshop, Stuttgart: Universität Stuttgart.





Lappin, Shalom (). The interpretation of ellipsis, in Shalom Lappin (ed.), The handbook of contemporary semantic theory. Oxford: Blackwell, –. Lappin, Shalom (). An HPSG account of antecedent-contained ellipsis, in Shalom Lappin and Elabbas Benmamoun (eds), Fragments: Studies in ellipsis and gapping. Oxford: Oxford University Press, –. Lappin, Shalom (). A sequenced model of anaphora and ellipsis resolution, in António Branco, Tony McEnery, and Ruslan Mitkov (eds), Anaphora processing: Linguistic, cognitive, and computational modelling. Amsterdam and Philadelphia: John Benjamins, –. Lappin, Shalom and Michael McCord (). Anaphora resolution in slot grammar, Computational Linguistics (): –. Larson, Bradley (). A dilemma with accounts of Right Node Raising, Linguistic Inquiry : –. Larson, Bradley (). The syntax of non-syntactic dependencies. PhD thesis, University of Maryland. Larson, Bradley (). Unelided basic remnants in Germanic ellipsis or: ÜBRIGE arguments, University of Pennsylvania Working Papers in Linguistics (): article . Larson, Richard (a). On the double object construction, Linguistic Inquiry (): –. Larson, Richard (b). Scope and comparatives, Linguistics and Philosophy (): –. Larson, Richard (). Double objects revisited: Reply to Jackendoff, Linguistic Inquiry (): –. Larson, Richard and Hiroko Yamakido (). A new form of nominal ellipsis in Japanese, in Patricia M. Clancy (ed.), Japanese/Korean Linguistics . Stanford: CSLI Publications, –. Larsson, Staffan (). Issue-based dialogue management. PhD thesis, Gothenburg University. Larsson, Staffan (). Formal semantics for perceptual classification, Journal of Logic and Computation (): –. Lasnik, Howard (). Remarks on coreference, Linguistic Analysis : –. Lasnik, Howard (a). Case and expletives revisited: On greed and other human failings, Linguistic Inquiry (): –. Lasnik, Howard (b). Verbal morphology: Syntactic structures meets the minimalist program, in Hector Campos and Paula Kempchinsky (eds), Evolution and revolution in linguistic theory. Washington, DC: Georgetown University Press, –. Lasnik, Howard (). A gap in an ellipsis paradigm: Some theoretical implications, Linguistic Analysis : –. Lasnik, Howard (a). A note on pseudogapping, in Howard Lasnik (ed.), Minimalist analysis. Oxford: Blackwell Publishers, –. Lasnik, Howard (b). Pseudogapping puzzles, in Shlom Lappin and Elabbas Benmamoun (eds), Fragments: Studies in ellipsis and gapping. Oxford: Oxford University Press, –. Lasnik, Howard (). When can you save a structure by destroying it?, in Minjoo Kim and Uri Strauss (eds), NELS : Proceedings of the st annual meeting of the North East Linguistic Society. Amherst, MA: GLSA, –. Lasnik, Howard (). On feature strength: Three Minimalist approaches to overt movement, Minimalist Investigations in Linguistic Theory : –. Lasnik, Howard (). Review of The syntax of silence, by Jason Merchant, Language (): –. Lasnik, Howard (). On ellipsis: Is material that is phonetically absent but semantically present present or absent syntactically? in H. Götzche (ed.), Memory, mind and language. Newcastle upon Tyne: Cambridge Scholars Publishing, –. Lasnik, Howard (). Ellipsis and island violation repair. Handout for LING , University of Maryland. Lasnik, Howard (). Multiple sluicing in English? Syntax : –. Lasnik, Howard and Myung-Kwan Park (). The EPP and the Subject Condition under sluicing, Linguistic Inquiry (): –.





Lasnik, Howard and Myung-Kwan Park (). Locality and MaxElide in extraction out of elided VP, in Yoichi Miyamoto, Daiko Takahashi, Hideki Maki, Masao Ochi, Koji Sugisaki, and Asako Uchibori (eds), Deep insights, broad perspectives: Essays in honor of Mamuro Saito. Tokyo: Kaitakusha, –. Lasnik, Howard and Mamoru Saito (). On the nature of proper government, Linguistic Inquiry (): –. Lasnik, Howard and Mamoru Saito (). On the subject of infinitives, in Lise M. Dobrin, Lynn Nichols, and Rosa M. Rodriguez (eds), Proceedings of the th annual meeting of the Chicago Linguistic Society. Chicago: Chicago Linguistic Society, –. Lasnik, Howard and Mamoru Saito (). Move α. Cambridge, MA: MIT Press. Lasnik, Howard and Tim Stowell (). Weakest crossover, Linguistic Inquiry (): –. Lebeaux, David (). Relative clauses, licensing, and the nature of the derivation, in Juli Carter, Dechaine Rose-Marie, Bill Philip, and Tim Sherer (eds), NELS : Proceedings of the th annual meeting of the North East Linguistic Society. Amherst, MA: Graduate Linguistic Student Association, –. Lebeaux, David (). Where does the Binding Theory apply? Cambridge, MA: MIT Press. Lechner, Winfried (). Some remarks on phrasal comparatives. Paper presented at the International Syntax Roundtable, Ypps/Danube, – July . Lechner, Winfried (). Comparative deletion. PhD thesis, University of Massachusetts at Amherst. Lechner, Winfried (). Reduced and phrasal comparatives, Natural Language and Linguistic Theory (): –. Lechner, Winfried (). Ellipsis in comparatives. Berlin and New York: Mouton de Gruyter. Lechner, Winfried (). Zwei bemerkungen zur Verbbewegung. Paper presented at the University of Hamburg. Lechner, Winfried (). On the typology of phrasal comparatives. Paper presented at the University of Vienna. Lechner, Winfried (forthcoming). Clausal vs. phrasal comparatives, in Daniel Gutzmann, Lisa Matthewson, Cécile Meier, Hotze Rullmann, and Thomas Ede Zimmermann (eds), The Blackwell Companion to Semantics. Malden: Blackwell. Lechner, Winfried and Norbert Corver (). Comparative deletion and comparative subdeletion, in Martin Everaert and Henk van Riemsdijk (eds), The Wiley Blackwell companion to syntax. nd edition. Hoboken: Wiley-Blackwell, –. Lee, Yong-hun (). A unified approach to VP-ellipsis and VP-anaphora, in Stefan Müller (ed.), Proceedings of the th International Conference on Head-Driven Phrase Structure Grammar. Stanford: CSLI, –. Lees, Robert (). The English comparative construction, Word : –. Leonetti, Manuel (). El artículo, in Ignacio Bosque and Violeta Demonte (eds), Gramática descriptiva de la lengua española. Madrid: Espasa Calpe, –. Lerner, Jan and Manfred Pinkal (). Comparative ellipsis and variable binding, in Mandy Simons and Tim Galloway (eds), Semantics and linguistic theory. Ithaca, NY: Cornell University Press, –. Leung, Tommi (). The preposition stranding generalization and conditions on sluicing: Evidence from Emirati Arabic, Linguistic Inquiry : –. Levelt, Willem J. (). Monitoring and self-repair in speech, Cognition (): –. Levin, Lori (). Sluicing: A lexical interpretation procedure, in Joan Bresnan (ed.), The mental representation of grammatical relations. Cambridge, MA: MIT Press, –. Levin, Nancy S. (). Some identity-of-sense deletions puzzle me, in D. Farkas, W. Jacobsen, and K. Todrys (eds), Papers from the th regional meeting of the Chicago Linguistic Society. Chicago: Chicago Linguistic Society, –. Levin, Nancy S. (). Main verb ellipsis in spoken English, in Arnold M. Zwicky, Clitics and ellipsis. Columbus: Ohio State University, –.





Levin, Nancy S. (). Main-verb ellipsis in spoken English. New York: Garland Publishing. Levin, Nancy S. and Ellen F. Prince (). Gapping and clausal implicature, Papers in Linguistics : –. Levine, Robert D. (). Right Node (non)-Raising, Linguistic Inquiry (): –. Levine, Robert D. (). The extraction riddle: Just what are we missing? Journal of Linguistics : –. Levine, Robert D. (). Linearization and its discontents, in Stefan Müller (ed.), Proceedings of the th International Conference on Head-Driven Phrase Structure Grammar. Stanford: CSLI, –. Levine, Robert D. and Yusuke Kubota (). Gapping as hypothetical reasoning. Ms. Ohio State University. Lewis, David (). Score keeping in a language game, in Rainer Bauerle, Urs Egli, and Arnim von Stechow (eds), Semantics from different points of view. Dordrecht: Springer, –. Lewis, Richard L., Shravan Vasishth, and Julie A. Van Dyke (). Computational principles of working memory in sentence comprehension, Trends in Cognitive Sciences : –. Liceras, Juana M., Raquel Fernández Fuertes, Susana Perales, Rocío Pérez-Tattam, and Kenton T. Spradlin (). Gender and gender agreement in bilingual native and non-native grammars: A view from child and adult functional-lexical mixings, Lingua (): –. Liddell, Scott K. (). American Sign Language syntax. The Hague: Mouton. Liddell, Scott K. (). Grammar, gesture, and meaning in American Sign Language. Cambridge: Cambridge University Press. Lin, Jo-Wang (). Chinese comparatives and their implicational parameters, Natural Language Semantics (): –. Lin, Vivian (). Determiner sharing, MIT Working Papers in Linguistics : –. Lin, Vivian (). A way to undo A movement, in Karine Megerdoomian and Leora Anne Bar-el (eds), Proceedings of the th West Coast Conference on Formal Linguistics. Somerville, MA: Cascadilla Press, –. Lin, Vivian (). Coordination and sharing at the interfaces. PhD thesis, Massachusetts Institute of Technology. Lindenbergh, Charlotte, Angeliek van Hout, and Bart Hollebrandse (). Extending ellipsis research: The acquisition of sluicing in Dutch. Poster presented at the th Boston University Child Language Development Conference (BUCLD ). Lipták, Anikó (). A fragmentumok mondattana a magyarban, Általános nyelvészeti tanulmányok : –. Lipták, Anikó (). Verb-stranding ellipsis and verbal identity: The role of polarity focus, Linguistics in the Netherlands : –. Lipták, Anikó (a). A note on overt case marking as evidence against pseudo-sluicing. Unpublished manuscript, University of Leiden. Lipták, Anikó (b). The syntax of positive polarity in Hungarian: Evidence from ellipsis, Lingua : –. Lipták, Anikó (). Relative pronouns as sluicing remnants, in Éva Dékány, Katalin É. Kiss, and Balázs Surányi (eds), Approaches to Hungarian . Amsterdam and Philadelphia: John Benjamins, –. Lipták, Anikó and Andrés Saab (). Nominal ellipsis and the stranded affix filter. Paper presented at Nijmegen University. Lipták, Anikó and Andrés Saab (). No N-raising out of NPs in Spanish: Ellipsis as a diagnostic of head movement, Natural Language and Linguistic Theory : –. Llombart-Huesca, Amàlia (). Anaphoric one and NP-ellipsis, Studia Linguistica (): –. Llombart-Huesca, Amàlia (). La elipsis nominal y los sintagmas introducidos por de, Cuadernos de la ALFAL : –. Lobeck, Anne (). Syntactic constraints on ellipsis. PhD thesis, University of Washington, Seattle.





Lobeck, Anne (). Functional heads as proper governors, in Juli Carter (ed.), NELS : Proceedings of the th annual meeting of the North East Linguistic Society. Amherst, MA: GLSA, –. Lobeck, Anne (). The phrase structure of ellipsis, in Susan Rothstein (ed.), Perspectives on phrase structure. San Diego: Academic Press, –. Lobeck, Anne (). Licensing and identification of ellipted categories in English, in Steve Berman and Arild Hestvik (eds), Proceedings of the Stuttgart Ellipsis Workshop, Stuttgart. Lobeck, Anne (). Strong agreement and identification: Evidence from ellipsis in English, Linguistics : –. Lobeck, Anne (). Ellipsis: Functional heads, licensing, and identification. New York: Oxford University Press. Lobeck, Anne (). VP ellipsis and the Minimalist Program: Some speculations and proposals, in Shalom Lappin and Elabbas Benmamoun (eds), Fragments: Studies in ellipsis and gapping. Oxford: Oxford University Press, –. Lobeck, Anne (). Ellipsis in DP, in Martin Everaert and Henk van Riemsdijk (eds), The Blackwell companion to syntax. Malden, MA and Oxford: Blackwell, –. Loewen, Gena (). The syntactic structure of noun phrases in Indonesian. Master’s thesis, University of Manitoba. Longobardi, Giuseppe (). Reference and proper names, Linguistic Inquiry (): –. Lopes, Ruth E. Vasconcellos and Ana Lúcia Santos (). VP-ellipsis comprehension in European and Brazilian Portuguese, in João Costa, Alexandra Fiés, M. João Freitas, Maria Lobo, and Ana Lúcia Santos (eds), New directions in the acquisition of Romance languages. Newcastle upon Tyne: Cambridge Scholars Publishing, –. López, Luis (). Polarity and predicate anaphora. PhD thesis, Cornell University. López, Luis (). Ellipsis and discourse linking, Lingua : –. López, Luis and Susanne Winkler (). Focus and topic in VP-anaphora constructions, Linguistics (): –. Ludlow, Peter (). A note on alleged cases of nonsentential assertion, in Reinaldo Elugardo and Robert J. Stainton (eds), Ellipsis and nonsentential speech. Dordrecht: Kluwer Academic Publishers, –. Lukasczyk, Ulrika (). Sanottua, ajateltua ja tehtyä: Referointi kolmessa suomalaisella viittomakielellä tuotetussa fiktiivisessä kertomuksessa [Referencing in three fictive FinSL narratives]. Master’s thesis, University of Jyväskylä. Macafee, Caroline (). Characteristics of non-standard grammar in Scotland. Ms. University of Aberdeen. Machobane, Malillo (). Variation in Bantu DP-structure: Evidence from Sesotho, Malawian Journal of Linguistics : –. MacSwan, Jeff (). A Minimalist approach to intrasentential code-switching. New York: Garland Publishing. MacSwan, Jeff (). The architecture of the bilingual language faculty: Evidence from intrasentential code switching, Bilingualism: Language and Cognition (): –. MacSwan, Jeff (a). Codeswitching and generative grammar: A critique of the MLF model and some remarks on ‘modified minimalism’, Bilingualism: Language and Cognition (): –. MacSwan, Jeff (b). Remarks on Jake, Myers-Scotton and Gross’s response: There is no ‘matrix language’, Bilingualism: Language and Cognition (): –. MacSwan, Jeff (). Code-switching and grammatical theory, in Tej K. Bhatia and William C. Ritchie (eds), The handbook of bilingualism and multilingualism. Hoboken: Wiley-Blackwell, –. MacSwan, Jeff (). Grammatical theory and bilingual codeswitching. Cambridge: Cambridge University Press. MacSwan, Jeff and Kara T. McAlister (). Naturalistic and elicited data in grammatical studies of codeswitching, Studies in Hispanic and Lusophone Linguistics (): –.





MacWhinney, Brian (). The CHILDES project: Tools for analyzing talk. Mahwah, NJ: Lawrence Erlbaum Associates. Maeda, Masako and Daiko Takahashi (). NP-ellipsis in the Naghasaki dialect of Japanese. Paper presented at the rd Japanese/Korean Linguistics Conference, MIT. Mahootian, Shahrzad (). A null theory of code-switching. PhD thesis, Northwestern University. Makuuchi, Michiru, Yosef Grodzinsky, Katrin Amunts, Andrea Santi, and Angela D. Friederici (). Processing noncanonical sentences in Broca’s region: Reflections of movement distance and type, Cerebral Cortex : –. Maling, Joan M. (). On ‘gapping and the order of constituents’, Linguistic Inquiry (): –. Malouf, Robert (). Mixed categories in the hierarchical lexicon. PhD thesis, Stanford University. Manning, Christopher and Ivan Sag (). Argument structure, valence, and binding, Nordic Journal of Linguistics : –. Manus, Sophie and Cédric Patin (). The prosodic structure of elliptical constructions in Bantu. Paper presented at the th International Conference on Bantu Languages (Bntu). Manzini, Maria Rita (). Locality, vol. . Cambridge, MA: MIT Press. Marandin, Jean-Marie (). Pas d’identité sans identité: L’analyse des groupes nominaux en Det + A, in Bernard Fradin and Jean-Marie Marandin (eds), Mot et grammaire. Paris: Didier Erudition, –. Marchena, Oscar Garcia (). Phrases averbales et fragments de l’espagnol oral. PhD thesis, Université Paris Diderot. Martí, Luisa (). Grammar vs. pragmatics: Carving nature at the joints, Mind & Language : –. Martin, Andrea E. and Brian McElree (). A content-addressable pointer mechanism underlies comprehension of verb-phrase ellipsis, Journal of Memory and Language : –. Martin, Andrea E. and Brian McElree (). Memory operations that support language comprehension: Evidence from verb-phrase ellipsis, Journal of Experimental Psychology: Learning Memory and Cognition : –. Martin, Andrea E. and Brian McElree (). Direct-access retrieval during sentence comprehension: Evidence from sluicing, Journal of Memory and Language : –. Martin, Roger (). On the distribution and case features of PRO. Ms. University of Connecticut. Martin, Roger (). A Minimalist theory of PRO and control. PhD thesis, University of Connecticut. Martin, Roger (). Null case and the distribution of PRO, Linguistic Inquiry (): –. Martins, Ana-Maria (). Enclisis, VP-deletion and the nature of sigma, Probus (/): –. Marušič, Franc and Rok Žaucer (). A note on sluicing and island repair, in Steven Franks (ed.), Annual Workshop on Formal Approaches to Slavic Linguistics: The Third Indiana Meeting . Ann Arbor: Michigan Slavic Publications, –. Matos, Gabriela (). Construções de elipse do predicado em Português: SV nulo e despojamento. PhD thesis, University of Lisbon. Matsuo, Ayumi (). Differing interpretations of empty categories in English and Japanese VP ellipsis contexts, Language Acquisition (): –. Matthews, Peter (). Syntax. Cambridge: Cambridge University Press. May, Robert (). Logical Form: Its structure and derivation. Cambridge, MA: MIT Press. Mayr, Clemens (). Focusing bound pronouns, Natural Language Semantics (): –. McCawley, James D. (). On the applicability of vice versa, Linguistic Inquiry (): –. McCawley, James D. (). Parentheticals and discontinuous constituent structure, Linguistic Inquiry (): –. McCawley, James D. (). Some additional evidence for discontinuity, in Geoffrey Huck and Almerindo E. Ojeda (eds), Syntax and semantics, vol. : Discontinuous constituency. Orlando: Academic Press, –. McCawley, James D. (). Gapping with shared operators, in David A. Peterson (ed.), Proceedings of the annual meeting of the Berkeley Linguistics Society. Berkeley: Berkeley Linguistics Society, –.





McCawley, James D. (). The syntactic phenomena of English. Chicago: University of Chicago. McCloskey, James (). Right node raising and preposition stranding, Linguistic Inquiry : –. McCloskey, James (). Resumptive pronouns, A’-binding, and levels of representation in Irish, in Randall Hendrick (ed.), The syntax of the Modern Celtic languages. San Diego: Academic Press, –. McCloskey, James (a). Clause structure, ellipsis, and proper government in Irish, Lingua : –. McCloskey, James (b). There, it, and agreement, Linguistic Inquiry : –. McCloskey, James (). A note on predicates and heads in Irish clausal syntax, in Andrew Carnie, Heidi Harley, and Sheila Ann Dooley (eds), Verb first: On the syntax of verb-initial languages. Amsterdam and Philadelphia: John Benjamins, –. McCloskey, James (). A language at the edge: Irish and the theory of grammar. Paper presented at the University of North Carolina, Chapel Hill. McCloskey, James (). Subjects, ellipsis, and the EPP in a VSO language. Paper presented at USC. McClure, Erica. F. (). Aspects of code-switching in the discourse of bilingual Mexican-American children. Technical Report . Urbana: Center for the Study of Reading, Technical Report. McConnell-Ginet, Sally (). Comparative constructions in English: A syntactic and semantic analysis. PhD thesis, University of Rochester. McDowell, Joyce P. (). Assertion and modality. PhD thesis, University of Southern California. McElree, Brian and Thomas G. Bever (). The psychological reality of linguistically defined gaps, Journal of Psycholinguistic Research : –. McKillen, Alanah (). On the interpretation of reflexive pronouns. PhD dissertation, McGill University, Montreal. McKoon, Gail, Roger Ratcliff, and Gregory Ward (). Testing theories of language processing: An empirical investigation of the on-line lexical decision task, Journal of Experimental Psychology: Learning Memory and Cognition : –. McNally, Louise (). VP-coordination and the VP-internal subject hypothesis, Linguistic Inquiry : –. McNeill, David (). Hand and mind: What gestures reveal about thought. Chicago: University of Chicago Press. McShane, Marjorie (). A theory of ellipsis. New York: Oxford University Press. Megerdoomian, Karin (). Beyond words and phrases: A unified theory of predicate composition. PhD thesis, University of Southern California. Mehler, Jacques (). Some effects of grammatical transformations on the recall of English sentences, Journal of Verbal Learning and Verbal Behavior : –. Mel’čuk, Igor (). Dependency syntax: Theory and practice. Albany, NY: State University of New York Press. Mel’čuk, Igor (). Levels of dependency description: Concepts and problems, in Vilmos Ágel, Ludwig M. Eichinger, Hans-Werner Eroms, Peter Hellwig, Hans Jürgen Heringer, and Henning Lobin (eds), Dependency and valency: An international handbook of contemporary research. Berlin: Walter de Gruyter, –. Melnik, Nurit (). Extending partial pro-drop in modern Hebrew: A comprehensive analysis, in Stefan Müller (ed.), Proceedings of the th International Conference on Head-Driven Phrase Structure Grammar. Stanford: CSLI Publications, –. Merchant, Jason (). ‘Pseudosluicing’: Elliptical clefts in Japanese and English, ZAS Working Papers in Linguistics : –. Merchant, Jason (). The syntax of silence: Sluicing, islands, and identity in ellipsis. PhD thesis, University of California, Santa Cruz. Merchant, Jason (). Antecedent-contained deletion in negative polarity items, Syntax (): –.





Merchant, Jason (). The syntax of silence: Sluicing, islands, and the theory of ellipsis. Oxford: Oxford University Press. Merchant, Jason (). Swiping in Germanic, in C. Jan-Wouter Zwart and Werner Abraham (eds), Studies in comparative Germanic syntax. Amsterdam: John Benjamins, –. Merchant, Jason (a). Remarks on stripping. Ms. University of Chicago. Merchant, Jason (b). Subject-auxiliary inversion in comparatives and PF output constraints, in Kerstin Schwabe and Susanne Winkler (eds), The interfaces: Deriving and interpreting omitted structures. Amsterdam and Philadelphia: John Benjamins, –. Merchant, Jason (c). Section on stripping excised from submitted version of ‘Fragments and ellipsis’, Linguistics and Philosophy (): –. Merchant, Jason (a). Fragments and ellipsis, Linguistics and Philosophy (): –. Merchant, Jason (b). Resumptivity and non-movement, Studies in Greek Linguistics : –. Merchant, Jason (). Revisiting syntactic identity conditions. Paper presented at the Workshop on Ellipsis, University of California, Berkeley. Merchant, Jason (a). Sluicing, in Martin Everaert and Henk van Riemsdijk (eds), The Blackwell companion to syntax. Maldon, MA and Oxford: Blackwell, –. Merchant, Jason (b). Small structures: A sententialist perspective, in Ljiljana Progovac, Kate Paesani, Eugenia Casielles, and Ellen Barton (eds), The syntax of nonsententials: Multidisciplinary perspectives. Amsterdam and Philadelphia: John Benjamins, –. Merchant, Jason (c). Why no(t)?, Style (–): –. Merchant, Jason (a). An asymmetry in voice mismatches in VP-ellipsis and pseudogapping, Linguistic Inquiry (): –. Merchant, Jason (b). Variable island repair under ellipsis, in Kyle Johnson (ed.), Topics in ellipsis. Cambridge: Cambridge University Press, –. Merchant, Jason (). Phrasal and clausal comparatives in Greek and the abstractness of syntax, Journal of Greek Linguistics : –. Merchant, Jason (). Three types of ellipsis, in François Recanati, Isidora Stojanovic, and Neftali Villanueva (eds), Context-dependence, perspective and relativity. Berlin and New York: Mouton de Gruyter, –. Merchant, Jason (). Ellipsis: A survey of analytical approaches. Ms. University of Chicago. Merchant, Jason (a). Diagnosing ellipsis, in Lisa Lai-Shen Cheng and Norbert Corver (eds), Diagnosing syntax. Oxford: Oxford University Press, –. Merchant, Jason (b). Polarity items under ellipsis, in Lisa Lai-Shen Cheng and Norbert Corver (eds), Diagnosing syntax. Oxford: Oxford University Press, –. Merchant, Jason (c). Revisiting conditions on predicate anaphora, in Philip Hofmeister and Elisabeth Norcliffe (eds), The core and the periphery: Data-driven perspectives on syntax inspired by Ivan A. Sag. Stanford: CSLI Publications, –. Merchant, Jason (d). Voice and ellipsis, Linguistic Inquiry (): –. Merchant, Jason (). An embarrassment of riches? Cutting up the elliptical pie, Theoretical Linguistics (–): –. Merchant, Jason (a). Gender mismatches under nominal ellipsis, Lingua : –. Merchant, Jason (b). Greek-English bilingual code-switching under ellipsis. Paper presented at the th Annual Midwest Greek Linguistics Workshop,  April, University of Chicago. Merchant, Jason (a). How much context is enough? Two cases of span-conditioned stem allomorphy, Linguistic Inquiry : –. Merchant, Jason (b). On ineffable predicates: Bilingual Greek-English code-switching under ellipsis, Lingua : –. Merchant, Jason and Andrew Simpson (eds) (). Sluicing: Cross-linguistic perspectives. Oxford Studies in Theoretical Linguistics . Oxford: Oxford University Press.





Merchant, Jason, Lyn Frazier, Charles Jr. Clifton, and Thomas Weskott (). Fragment answers to questions: A case of inaudible syntax, in Laurence Goldstein (ed.), Brevity. Oxford: Oxford University Press, –. Messick, Troy and Gary Thoms (). Ellipsis, economy, and the (non)uniformity of traces, Linguistic Inquiry : –. Messick, Troy, Andrés Saab, and Luis Vicente (). Deep properties of a surface anaphor: Contextual restrictions of sluicing sites. Ms. University of Connecticut, University of Buenos Aires, and University of Potsdam. Miller, Philip (). Pseudogapping and do so substitution, in Michaela Ziolkowski, Manuela Noske, and Karen Deaton (eds), CLS : Papers from the th Regional Meeting of the Chicago Linguistic Society. Chicago: Chicago Linguistic Society, –. Miller, Philip (a). Auxiliary verbs in Old and Middle French. A diachronic study of substitutive ‘fair’ and a comparison with the Modern English auxiliaries, in Ans van Kemenade and Nigel Vincent (eds), Parameters of morphosyntactic change. Cambridge: Cambridge University Press, –. Miller, Philip (b). Les morphèmes zéro à l’épreuve du rasoir d’Occam, in Jean Chuquet and Marc Fryd (eds), Travaux linguistiques du Cerlico. Rennes: Presses Universitaires de Rennes, –. Miller, Philip (a). The choice between verbal anaphors in discourse, in Ilse Hendrickx, Sobha Lalitha Devi, António Branco, and Ruslan Mitkov (eds), Discourse anaphora and anaphor resolution colloquium. Heidelberg: Springer, –. Miller, Philip (b). The use of the English verbal anaphor ‘do so’ in discourse: Descriptive analysis and theoretical consequences. Paper presented at Typology and Language Universals Workshop, University of Paris . Miller, Philip (). A corpus study of pseudo-gapping and its theoretical consequences, in Christopher Piñón (ed.), Empirical issues in syntax and semantics . Paris: CSSP, –. Miller, Philip and Barbara Hemforth. (a). Verb phrase ellipsis with nominal antecedents. Ms. Université Paris, Diderot. Miller, Philip and Barbara Hemforth (b). VP ellipsis beyond syntactic identity: The case of nominal antecedents. Linguistic Evidence. DOI: ./.... Miller, Philip and Geoffrey K. Pullum (). Exophoric VP-ellipsis, in Philip Hofmeister and Elisabeth Norcliffe (eds), The core and the periphery: Data-driven perspectives on syntax inspired by Ivan A. Sag. Stanford: CSLI Publications, –. Miller, Philip and Ivan A. Sag (). French clitic movement without clitics or movement, Natural Language and Linguistic Theory (): –. Milner, Jean-Claude (). Some remarks on Principle C, in Anna Maria di Sciullio and Anne Rochette (eds), Binding in Romance. Ottawa: Canadian Linguistic Association, –. Miyagawa, Shigeru (). Against optional scrambling, Linguistic Inquiry (): –. Miyagawa, Shigeru (). On the ‘undoing’ property of scrambling: A response to Bošković, Linguistic Inquiry (): –. Miyagawa, Shigeru (). Why Agree? Why Move? Unifying agreement-based and discourseconfigurational languages. Cambridge, MA: MIT Press. Miyamoto, Yoichi (). Meisikunai-no syooryaku [Ellipsis within noun phrases], in Keiko Murasugi, Mamoru Saito, Yoichi Miyamoto, and Kensuke Takita (eds), Nihongo bunpoo handobukku [A handbook of Japanese grammar]. Tokyo: Kaitakusha, –. Mohamed-Soyir, Kassim (). Le nom en shingazidja (Ga): Morphologie, phonologie, sémantique et syntaxe. PhD thesis, Université Paris Diderot. Mohammad, Jan and Simin Karimi (). ‘Light’ verbs are taking over: Complex verbs in Persian, in Joel Ashmore Nevis and Vida Samiian (eds), Proceedings of the Western Conference on Linguistics (WECOL ). Fresno, CA: Department of Linguistics, –. Moline, Estelle (). ‘Comme’ et ‘comment’ en français contemporain: Spécificités et recouvrements, in Antoine Gautier and Thomas Verjans (eds), ‘Comme’, ‘comment’, ‘combien’: Concurrences et complémentarité. Paris: L’Harmattan, –.





Molnár, Valeria (). On different kinds of contrast, in Valeria Molnár and Susanne Winkler (eds), The architecture of focus. Berlin and New York: Mouton de Gruyter, –. Molnár, Valeria and Susanne Winkler (). Edges and gaps: Contrast at the interfaces, Lingua : –. Moltmann, Friederike (). Coordination and comparatives. PhD thesis, MIT. Montague, Richard (). English as a formal language, in Bruno Visentine (ed.), Linguaggi nella società e nella tecnica. Milan: Edizioni di Comunità, –. Montague, Richard (). The proper treatment of quantification in ordinary English, in Richmond H. Thomason (ed.), Formal philosophy: Selected papers of Richard Montague. New Haven: Yale University Press, –. Montali, Emanuele (). Sluicing and adjectives. Term paper, UCL. Morgan, Jerry (). Sentence fragments and the notion ‘sentence’, in Brai Kachru, Robert Lees, Yakov Malkiel, Angelina Pietrangeli, and Sol Saporta (eds), Issues in linguistics: Papers in honor of Henry and Renée Kahan. Urbana: University of Illinois Press, –. Morris, Amanda (). Polarity ellipsis and negative stripping. Master’s thesis, UCSC. Morzycki, Marcin (). Modification. Key topics in semantics and pragmatics. Cambridge: Cambridge University Press. Moulton, Keir (). Small antecedents: Syntax or pragmatics, in Emily Elfner and Martin Walkow (eds), NELS : Proceedings of the th annual meeting of the North East Linguistics Society. Amherst, MA: GLSA, –. Moulton, Keir (). Small antecedents. Ms. Simon Fraser University. Mouret, François (). A phrase structure approach to argument cluster coordination, in Stefan Müller (ed.), Proceedings of the th International Conference on Head-Driven Phrase Structure Grammar. Stanford: CSLI, –. Mouret, François (a). Grammaire des constructions coordonnées: Coordinations simples et coordinations à redoublement en français contemporain. PhD thesis, Université Paris Diderot. Mouret, François (b). Syntaxe et sémantique des construction en ‘ni’, Faits de langues : –. Mouret, François (). Les coordinations de terms dissemblables sont-elles elliptiques? in Jacques Durand, Benoît Habert, and Bernard Laks (eds), Actes du Congrés Mondial de Linguistique française. Paris: CMLF Online Publications. Mouret, Francois and Anne Abeille (). On the rule of Right Node Raising in French. Handout at the Colloque international ‘Topics in the Typology of Elliptical Constructions’. University of Chicago Center, Paris, June. Muadz, H. (). Coordinate structures: A planar representation. PhD thesis, University of Arizona, Tucson. Mukai, Emi (). On verbless conjunction in Japanese, in Makoto Kadowaki and Shigeto Kawahara (eds), NELS : Proceedings of the rd Annual Meeting of the North East Linguistic Society. Amherst, MA: GLSA, –. Müller, Claude (). ‘De’ partitif et la négation, in Danielle Forget, Paul Hirschbühler, France Martineau, and Maria-Luisa Rivero (eds), Negation and polarity. Amsterdam and Philadelphia: John Benjamins, –. Müller, Gereon (). Incomplete category fronting: A derivational approach to remnant movement in German. Dordrecht: Kluwer Academic Publishers. Müller, Gereon (). Constraints on displacement: A phase-based approach. Amsterdam and Philadelphia: John Benjamins. Müller, Stefan (a). Grammatical theory: From transformational grammar to constraint-based approaches. Berlin: Language Science Press. Müller, Stefan (b). HPSG—A synopsis, in Tibor Kiss and Artemis Alexiadou (eds), Syntax— Theory and analysis: An international handbook. Berlin: Mouton de Gruyter, –. Munn, Alan Boag (). Topics in the syntax and semantics of coordinate structures. PhD thesis, University of Maryland.





Murasugi, Keiko (). Noun phrases in Japanese and English: A study in syntax, acquisition and learnability. PhD thesis, University of Connecticut. Murasugi, Keiko (). Sakuzyo [Deletion], in Masatake Arimoto and Keiko Murasugi (eds), Sobukaku to sakuzyo [Binding and deletion]. Tokyo: Kenkyusha, –. Murphy, Gregory (). Processes of understanding anaphora, Journal of Memory and Language : –. Musolino, Julien and Jeffrey Lidz (). The scope of isomorphism: Turning adults into children, Language Acquisition : –. Muysken, Pieter (). Code-switching and grammatical theory, in Lesley Milroy and Pieter Muysken (eds), One speaker, two languages. Cambridge: Cambridge University Press, –. Muysken, Pieter and Henk van Riemsdijk (). Projecting features and featuring projections, in Pieter Muysken and Henk van Riemsdijk (eds), Features and projections. Dordrecht: Foris Publications, –. Myers-Scotton, Carol (). Dueling languages: Grammatical structure in codeswitching. Oxford: Clarendon Press. Myers-Scotton, Carol (). The matrix language frame model: Developments and responses, in Rodolfo Jacobson (ed.), Codeswitching worldwide II. Berlin and New York: Mouton de Gruyter, –. Myers-Scotton, Carol and Janice L. Jake (). Explaining aspects of codeswitching and their implications, in Janet L. Nicol (ed.), One mind, two languages: Bilingual language processing. Malden: Blackwell, –. Myers-Scotton, Carol and Janice L. Jake (). Revisiting the -M model: Codeswitching and morpheme election at the abstract level, International Journal of Bilingualism (): –. Nagata, Hiroshi (). Effect of repetition on grammaticality judgments under objective and subjective self-awareness conditions, Journal of Psycholinguistic Research (): –. Nakamura, Masanori (). Case morphology and island repair, in Jason Merchant and Andrew Simpson (eds), Sluicing: Cross-linguistic perspectives. Oxford: Oxford University Press, –. Nakao, Chizuru (). Island repair and non-repair by PF strategies. PhD thesis, University of Maryland. Nakao, Chizuru and Masaya Yoshida (). Japanese sluicing as a specificational pseudo-cleft, in Yukio Otsu (ed.), Proceedings of the Sixth Tokyo Conference on Psycholinguistics. Tokyo: Hituzi Syobo, –. Nakao, Chizuru and Masaya Yoshida (). ‘Not-so-propositional’ islands and their implications for swiping, in Erin Bainbridge and Brian Agbanyani (eds), Proceedings of the Western Conference on Linguistics (WECOL ). Fresno, CA: University of California Press, –. Napoli, Donna Jo (). Initial material deletion in English, Glossa : –. Napoli, Donna Jo (a). Comparative ellipsis: A phrase structure analysis, Linguistic Inquiry (): –. Napoli, Donna Jo (b). Missing complement sentences in English: A base analysis of Null Complement Anaphora, Linguistic Analysis : –. Napoli, Donna Jo (). Verb phrase deletion in English: A base-generated analysis, Journal of Linguistics : –. Nee, Julia (). Spanish/Zapotec code-switching and the theory of ellipsis. BA thesis, University of Chicago. Neeleman, Ad and Hans van de Koot (). The configurational matrix, Linguistic Inquiry (): –. Neeleman, Ad, Elena Titov, Hans van de Koot, and Reiko Vermeulen (). A syntactic typology of topic, focus, and contrast, in Jeroen van Craenenbroeck (ed.), Alternatives to cartography. Berlin: Mouton de Gruyter, –. Neijt, Anneke (). Gapping: A contribution to sentence grammar. Dordrecht: Foris Publications.





Neijt, Anneke (a). Gapping and remnants: Sentence grammar aspects of gapping, Linguistic Analysis : –. Neijt, Anneke (b). Gapping bestaat, Spektator : –. Nerbonne, John, Masayo Iida, and William Ladusaw (). Running on empty: Null heads in HeadDriven Grammar, in Jane Fee and Katherine Hunt (eds), Proceedings of the Eighth West Coast Conference on Formal Linguistics. Stanford: CSLI Publications, –. Nerbonne, John, Masayo Iida, and William Ladusaw (). Semantics of common Noun Phrase anaphora, in Aaron L. Halpern (ed.), The Proceedings of the Ninth West Coast Conference on Formal Linguistics. Stanford: CSLI Publications, –. Nerbonne, John and Tony Mullen (). Null-headed nominals in German and English, in Frank van Eynde, Ineke Schuurman, and Ness Schelkens (eds), Proceedings of Computational Linguistics in the Netherlands: Selected Papers from the Ninth CLIN Meeting. Amsterdam and Philadelphia: John Benjamins, –. Nevins, Andrew and Pranav Anand (). Some AGREEment matters, in M. Tsujimura (ed.), Proceedings of the West Coast Conference on Formal Linguistics . Somerville, MA: Cascadilla Press, –. Ngonyani, Deo (). VP ellipsis in Ndenduele and Swahili applicatives, in Edward Garrett and Felicia Lee (eds), Syntax at sunset. UCLA Working Papers in Syntax and Semantics . Los Angeles: University of California Los Angeles, –. Ngonyani, Deo (). V-to-I movement in Kiswahili, Afrikanistische Arbeitspapiere : –. Ngonyani, Deo and Peter Githinji (). The asymmetric nature of Bantu applicative constructions, Lingua : –. Nicol, Janet, Janet Dean Fodor, and David Swinney (). Using cross-modal lexical decision tasks to investigate sentence processing, Journal of Experimental Psychology: Learning Memory and Cognition : –. Nicol, Janet and David Swinney (). The role of structure in coreference assignment during sentence comprehension, Journal of Psycholinguistic Research (): –. Nicol, Janet and David Swinney (). The psycholinguistics of anaphora, in Andrew Barss (ed.), Anaphora: A reference guide. Oxford: Oxford University Press, –. Nicolae, Alexandru (). P-stranding in a language without P-stranding? The case of sluicing in Romanian. Paper presented at the th annual meeting of the Societas Linguistica Europaea, University of Stockholm. Nielsen, Leif Arda (). A corpus-based study of verb phrase ellipsis identification and resolution. Ms. King’s College London. . Nishigauchi, Taisuke (). Quantification in the theory of grammar, vol. . Dordrecht: Kluwer Academic Publishers. Nishigauchi, Taisuke (). ‘Multiple sluicing’ in Japanese and the functional nature of wh-phrases, Journal of East Asian Linguistics (): –. Nishigauchi, Taisuke (). Ronri koozoo to bunpoo riron: Nitieigo no WH gensyoo [Logical structure and the theory of grammar: WH phenomena in Japanese and English]. Tokyo: Kurioso. Nishigauchi, Taisuke and Tomohiro Fujii (). Short answers: ellipsis, connectivity, and island repair. Ms. Kobe Shoin Graduate School and University of Maryland. Nishiyama, Kunio, John Whitman, and Eun-Young Yi (). Syntactic movement of overt whphrases in Japanese and Korean, in Noriko Akatsuka, Shoichi Iwasaki, and Susan Strauss (eds), Japanese and Korean linguistics. Stanford: CSLI Publications, –. Nissenbaum, Jon (). Derived predicates and the interpretation of parasitic gaps, in Kimary Shahin, Susan Blake, and Eun-Sook Kim (eds), Proceedings of the th West Coast Conference on Formal Linguistics. Stanford: CSLI Publications, –. Nunes, Jairo (). Linearization of chains and sideward movement. Linguistic Inquiry Monographs. Cambridge, MA: MIT Press.





Nunes, Jairo and Cynthia Zocca (). Lack of morphological identity and ellipsis resolution in Brazilian Portuguese, in Jairo Nunes (ed.), Minimalist essays on Brazilian Portuguese syntax. Amsterdam and Philadelphia: John Benjamins, –. Nykiel, Joanna (). Clefts and preposition omission under sluicing, Lingua : –. Nykiel, Joanna (). Constraints on ellipsis alternation: A view from the history of English, Language Variation and Change : –. Nykiel, Joanna (). Preposition stranding and ellipsis alternation, English Language and Linguistics (): –. Nykiel, Joanna and Ivan A. Sag (). Sluicing and stranding. Ms. University of Silesia and Stanford University. Obdeijn, Margreet (). Discourse anaphora in children and the role of ‘er’ [there] in Dutch. BA thesis, Universiteit van Utrecht. Oehrle, Richard T. (). Boolean properties in the analysis of gapping, in Geoffrey J. Huck and Almerindo E. Ojeda (eds), Syntax and semantics: Discontinuous constituency. San Diego, CA: Academic Press, –. Oehrle, Richard T. (). Categorial frameworks, coordination, and extraction, in Aaron L. Halpern (ed.), Proceedings of the Ninth West Coast Conference on Formal Linguistics. Stanford, CA: Center for the Study of Language and Information, –. O’Grady, William (). The syntax of idioms, Natural Language and Linguistic Theory : –. Oirsouw, Robert R. van (). The syntax of coordination. New York: Croom Helm. Okrent, Anrika (). A modality-free notion of gesture and how it can help us with the morpheme vs. gesture question in sign language linguistics (or at least give us some criteria to work with), in Richard P. Meier, Kearsy Cornier, and David Quinto-Pozos (eds), Modality and structure in signed and spoken languages. Cambridge: Cambridge University Press, –. Oku, Satoshi (). A theory of selection and reconstruction in the Minimalist perspective. PhD thesis, University of Connecticut. Olsvay, Csaba (). Formális jegyek egyeztetése a magyar nemsemleges mondatokban, in László Büky and Márta Maleczki (eds), A mai magyar nyelv leírásának újabb módszerei. Szeged: Szegedi Tudományegyetem, –. Olsvay, Csaba (). The Hungarian verbal complex: An alternative approach, in Katalin É. Kiss and Henk van Riemsdijk (eds), Verb clusters: A study of Hungarian, German and Dutch. Amsterdam and Philadelphia: John Benjamins, –. Ortega-Santos, Iván, Masaya Yoshida, and Chizuru Nakao (). On ellipsis structures involving a wh-remnant and a non-wh-remnant simultaneously, Lingua : –. Osborne, Timothy (). Beyond the constituent: A Dependency Grammar analysis of chains, Folia Linguistica (–): –. Osborne, Timothy (). Shared material and grammar: A dependency grammar theory of nongapping coordination, Zeitschrift für Sprachwissenschaft : –. Osborne, Timothy (). Major constituents and two dependency grammar constraints on sharing coordination, Linguistics : –. Osborne, Timothy (). Comparative coordination vs. comparative subordination, Natural Language and Linguistic Theory : –. Osborne, Timothy (). Edge features, catenae, and dependency-based Minimalism, Linguistic Analysis (–): –. Osborne, Timothy (). Dependency grammar, in Andrew Carnie, Yosuke Sato, and Daniel Saddiqi (eds), The Routledge handbook of syntax. London: Routledge, –. Osborne, Timothy and Thomas Groß (). Constructions are catenae: Construction Grammar meets Dependency Grammar, Cognitive Linguistics (): –. Osborne, Timothy, Michael Putnam, and Thomas Groß (). Bare phrase structure, label-less trees, and specifier-less syntax: Is Minimalism becoming a dependency grammar? The Linguistic Review : –.





Osborne, Timothy, Michael Putnam, and Thomas Groß (). Catenae: Introducing a novel unit of syntactic analysis, Syntax (): –. Oshima, Shin (). Conditions on rules: Anaphora in Japanese, in George Bedell, Eichi Kobayashi, and Masatake Muraki (eds), Explorations in linguistics: Papers in honor of Kazuko Inoue. Tokyo: Kenkyusha, –. Otani, Kazuyo and John Whitman (). V-raising and VP-ellipsis, Linguistic Inquiry (): –. Ott, Dennis (). An ellipsis approach to contrastive left-dislocation, Linguistic Inquiry : –. Ott, Dennis and Volker Struckmeier (). Deletion of non-constituents in clausal ellipsis: Remnants in the middle field. Ms. Humboldt University of Berlin and University of Cologne. Ott, Dennis and Mark de Vries (). Right-dislocation as deletion, Natural Language and Linguistic Theory (): –. Paape, Dario (). A reactivation advantage for sluicing antecedents in German. Poster presented at the th Annual Conference on Human Sentence Processing. University of Southern California. Padden, Carol (). The relation between space and grammar in ASL verb morphology, in Christopher Lucas (ed.), Sign language research: Theoretical issues. Washington DC: Gallaudet Press, –. Paesani, Kate (). Extending the nonsentential analysis: The case of special registers, in Ljiljana Progovac, Kate Paesani, Eugenia Casielles, and Ellen Barton (eds), The syntax of nonsententials: Multidisciplinary perspectives. Amsterdam and Philadelphia: John Benjamins, –. Panagiotidis, Phoevos (). Pronouns, clitics and empty nouns. Amsterdam and Philadelphia: John Benjamins. Panagiotidis, Phoevos (a). One, empty nouns and y-assignment, Linguistic Inquiry (): –. Panagiotidis, Phoevos (b). Empty nouns, Natural Language and Linguistic Theory (): –. Pancheva, Roumyana (). Phrasal and clausal comparatives in Slavic, in James Lavine, Steven Franks, Mila Tasseva-Kurktchieva, and Hana Filip (eds), Formal Approaches to Slavic Linguistics . Ann Arbor: Michigan Slavic Publications, –. Pancheva, Roumyana (). More students attended FASL than CONSOLE, in Wayles Browne, Adam Cooper, Alison Fisher, Esra Kesici, Nikola Predolac, and Draga Zec (eds), Formal Approaches to Slavic Linguistics : The Cornell Meeting, . Ann Arbor: Michigan Slavic Publications, –. Pancheva, Roumyana (). Syntax and semantics of comparatives and superlatives. Class handouts, EGG  summer school,  July– August. Pantcheva, Marina (). First phase syntax of Persian complex predicates: Argument structure and telicity, Journal of South East Asian Linguistics : –. Park, Myung-Kwan (). The syntax of VP ellipsis in Korean, Language Research (): –. Park, Myung-Kwan (). The syntax-semantics interface of pair-list interpretation in whconstructions and their sluicing counterparts, Studies in Generative Grammar (): –. Park, Myung-Kwan and Jung-Min Kang (). Multiple sluicing in English, Studies in Generative Grammar (): –. Park, Myung-Kwan and Jaejun Kim (). English Heavy NP Shift as clause-internal fragmenting, The Journal of Studies in Language (): –. Park-Johnson, Sung (). The acquisition of wh-questions by Korean-English bilingual children: The role of crosslinguistic influence. Master’s thesis, Purdue University. Park-Johnson, Sunny (). The effect of subject person in auxiliary movement acquisition by Korean-English bilingual children, in LSA Annual Meeting Extended Abstracts, vol. , :–. Parrott, Jeffrey K. (). Distributed morphological mechanisms of pronoun-case variation, University of Pennsylvania Working Papers in Linguistics (): –. Parrott, Jeffrey K. (). Danish vestigial case and the acquisition of vocabulary in Distributed Morphology, Biolinguistics (–): –.





Parsons, Terence (). Events in the semantics of English: A study in subatomic semantics. Current Studies in Linguistics . Cambridge, MA: MIT Press. Partee, Barbara Hall and Mats Rooth (). Generalized conjunction and type ambiguity, in R. Bäuerle, C. Schwarze, and Arnim von Stechow (eds), Meaning, use and interpretation of language. Berlin: de Gruyter, –. Patin, Cédric (). La tonologie du shingazidja, langue bantu (Ga) de la Grande Comore: Nature, formalisation, interfaces. PhD thesis, Université Paris . Patin, Cédric (). The prosody of Shingazidja relatives, ZAS Working Papers in Linguistics : –. Paul, Hermann (). Principles of the history of language. New York: MacMillan & Co. Paul, Ileana and Eric Potsdam (). Sluicing without wh-movement in Malagasy, in Jason Merchant and Andrew Simpson (eds), Sluicing: Cross-linguistic perspectives. Amsterdam and Philadelphia: John Benjamins, –. Paul, Waltraud (). Verb gapping in Chinese: A case of verb raising, Lingua : –. Payne, John, Geoffrey K. Pullum, Barbara C. Scholz, and Eva Berlage (). Anaphoric one and its implications, Language (): –. Penn, Gerald (). The algebraic structure of attributed type signatures. PhD thesis, Carnegie Mellon University. Pereltsvaig, Asya (). Small nominals, Natural Language and Linguistic Theory (): –. Pereltsvaig, Asya (). On the universality of DP: A view from Russian, Studia Linguistica (): –. Perez-Leroux, Ana, Mihaela Pirvulescu, and Yves Roberge (). Children’s interpretation of null objects under the scope of negation, in Susie Jones (ed.), Proceedings of the  annual conference of the Canadian Linguistic Association. Vancouver: University of British Columbia, –. Perez-Leroux, Ana, Mihaela Pirvulescu, and Yves Roberge (). Direct objects and language acquisition. Cambridge: Cambridge University Press. Perlmutter, David M. and John R. Ross (). Relative clauses with split antecedents, Linguistic Inquiry (): . Pesetsky, David (). Paths and categories. PhD thesis, Massachusetts Institute of Technology. Pesetsky, David (). Wh-in-situ: Movement and unselective binding, in Eric J. Reuland and Alice G. B. ter Meulen (eds), The representation of (in)definiteness. Cambridge, MA: MIT Press, –. Pesetsky, David (). Zero syntax. Cambridge, MA: MIT Press. Phillips, Colin and Dan Parker (). The psycholinguistics of ellipsis, Lingua : –. van Peteghem, Marleen (). Sur le subordonnant comparatif dans les langues romanes, Langages : –. Pfaff, Carol W. (). Constraints on language mixing: Intrasentential code-switching and borrowing in Spanish/English, Language : –. Philip, William (). Acquiring knowledge of universal quantification, in Jill de Villiers and Tom Roeper (eds), Handbook of generative approaches to language acquisition. Amsterdam: Springer, –. Philipp, Mareike (). German: Anaphor binding in sluiced sentences. Ms. University of Potsdam. Philippova, Tatiana (). P-omission under sluicing, [P clitic] and the nature of P-stranding, in Martin Kohlberger, Kate Bellamy, and Eleanor Dutton (eds), Proceedings of ConSOLE XXII. Leiden: SOLE, –. Phillips, Colin (). Order and structure. PhD thesis, Massachusetts Institute of Technology, Cambridge, MA. Phillips, Colin (). Linear order and constituency, Linguistic Inquiry (): –. Phillips, Colin (). The real-time status of island phenomena, Language : –. Phillips, Colin and Shevaun Lewis (). Derivational order in syntax: Evidence and architectural consequences, Studies in Linguistics : –. Piñón, Christopher (). Around the progressive in Hungarian, in István Kenesei (ed.), Approaches to Hungarian . Szeged: JATE Press, –.





Piantadosi, Steven T., Harry Tily, and Edward Gibson (). The communicative function of ambiguity in language, Cognition : –. Pickering, Martin J. and Guy D. Barry (). Sentence processing without empty categories, Language and Cognitive Processes : –. Pickering, Martin J. and Simon Garrod (). An integrated theory of language production and comprehension, Behavioral and Brain Sciences (): –. Pierrehumbert, Janet (). The phonology and phonetics of English intonation. PhD thesis, MIT. Pierrehumbert, Janet and Julia Hirschberg (). The meaning of intonational contours in the interpretation of discourse, in Cohenm Philip R., Jerry L. Morgan, and Martha E. Pollack (eds), Intentions in communication. Cambridge, MA: MIT Press, –. Pieters, Lody (). Lees maar! Er staat (niet!) wat er staat. Over de beschrijving van samentrekking en andere onvolledige taaluitingen in Nederlandse grammatica’s na . PhD thesis, Radboud University Nijmegen. Pinkham, Jessie E. (). The formation of comparative clauses in English and French. Bloomington, IN: Indiana University Linguistics Club. Plewnia, Albrecht (). Sätze, denen nichts fehlt: Eine dependenzgrammatische Untersuchung elliptischer Konstruktionen. Germanistische Linguistik Monographien . Hildesheim et al.: Olms. Plewnia, Albrecht (). Moglichkeiten und Wahrscheinlichkeiten: Was man in der Dependenzgrammatik mit syntaktischen Leerstellen tun kann. In Mathilde Hennig (eds), Die Ellipsis: Neue Perspectiven auf ein altes Phänomen. Berlin: Walter de Gruyter, –. Poesio, Massimo and Hannes Rieser (). Completions, coordination, and alignment in dialogue, Dialogue and Discourse : –. Poirier, Josée, Lewis Shapiro, Tracy Love, and Yosef Grodzinsky (). The on-line processing of verb phrase ellipsis in aphasia, Journal of Psycholinguistic Research : –. Poirier, Josée, Katie Wolfinger, Lisa Spellman, and Lewis P. Shapiro (). The real-time processing of sluiced sentences, Journal of Psycholinguistic Research : –. Pollard, Carl Jesse and Ivan A. Sag (). Information-based syntax and semantics. Sanford, CA: Center for the Study of Language and Information. Pollard, Carl Jesse and Ivan A. Sag (). Head-driven phrase structure grammar. Stanford, CA: Center for the Study of Language and Information. Pollmann, Thijs (). Een regel die subject en copula deleert? Spektator : –. Pollock, Jean-Yves (). Verb movement, UG and the structure of IP, Linguistic Inquiry (): –. Poplack, Shana (). Sometimes I’ll start a sentence in Spanish and termino en Español, in Jon Amastae and Lucia Elfas-Olivares (eds), Spanish in the United States: Sociolinguistic aspects. Cambridge: Cambridge University Press, –. Poplack, Shana (). Code-switching (linguistic), in Neil J. Smelser and Paul B. Baltes (eds), International encyclopedia of the social and behavioral sciences. Amsterdam: Elsevier, –. Poplack, Shana, Lauren Zentz, and Nathalie Dion (). Phrase final prepositions in Quebec French: An empirical study of contact, codeswitching, and resistance to convergence, Bilingualism: Language and Cognition (): –. Postal, Paul M. (). Anaphoric islands, in Robert I. Binnick, Alice Davison, Georgia M. Green, and Jerry L. Morgan (eds), Papers from the th regional meeting of the Chicago Linguistic Society. Chicago: Chicago Linguistic Society, –. Postal, Paul M. (). On raising. Cambridge, MA: MIT Press. Postal, Paul M. (). Parasitic and pseudoparasitic gaps, Linguistic Inquiry (): –. Postal, Paul M. (). Three investigations of extraction. Cambridge, MA: MIT Press. Potsdam, Eric (a). English verbal morphology and VP ellipsis, in Kiyomi Kusumoto (ed.), NELS : Proceedings of the th annual meeting of the North East Linguistic Society. Amherst, MA: Graduate Student Linguistic Association, –.





Potsdam, Eric (b). NegP and subjunctive complements in English, Linguistic Inquiry (): –. Potsdam, Eric (). Malagasy sluicing and its consequences for the identity requirement on ellipsis, Natural Language and Linguistic Theory (): –. Potsdam, Eric (). A direct analysis of Malagasy phrasal comparatives, in Lauren Eby Clemens, Gregory Scontras, and Maria Polinsky (eds), Proceedings of the th annual meeting of the Austronesian Formal Linguistics Society (AFLA ). Cambridge, MA: Austronesian Formal Linguistics Association, –. Potter, David (). A heterogeneous approach to gapping, in Robert E. Santana-LaBarge (ed.), Proceedings of the st West Coast Conference on Formal Linguistics. Somerville, MA: Cascadilla Press, –. Potts, Christopher (). The logic of conventional implicatures. Oxford: Oxford University Press. Potts, Christopher, Ash Asudeh, Seth Cable, Yurie Hara, Eric McCready, Luis Alonso-Ovalle, Rajesh Bhatt, Christopher Davis, Angelika Kratzer, Tom Roeper, and Martin Walkow (). Expressives and identity conditions, Linguistic Inquiry : –. Prinzhorn, Martin and Viola Schmitt (). Discontinuous DP-coordination in German, Linguistic Variation Yearbook : –. Pritchett, Bradley (). Subjacency in a principle-based parser, in Robert C. Berwick, Steven P. Abney, and Carol Tenny (eds), Principle-based parsing: Computation and psycholinguistics. Dordrecht: Kluwer Academic Publishers, –. Progovac, Ljiljana (). Detereminer phrase in a language without determiners, Journal of Linguistics : –. Progovac, Ljiljana (). Structure for coordination, in Lisa L.-S. Cheng and Rint Sybesma (eds), The second Glot International state-of-the-article book. Berlin: Mouton de Gruyter, –. Progovac, Ljiljana (). The syntax of nonsententials: Small clauses and phrases at the root, in Ljiljana Progovac, Kate Paesani, Eugenia Casielles, and Ellen Barton (eds), The syntax of nonsententials: Multidisciplinary perspectives. Amsterdam and Philadelphia: John Benjamins, –. Progovac, Ljiljana (). Non-sentential vs. ellipsis approaches: Review and extensions, Language and Linguistics Compass : –. Progovac, Ljiljana, Kate Paesani, Eugenia Casielles, and Ellen Barton (a). Epilogue: Wherefrom and whereto?, in Ljiljana Progovac, Kate Paesani, Eugenia Casielles, and Ellen Barton (eds), The syntax of nonsententials: Multidisciplinary perspectives. Amsterdam and Philadelphia: John Benjamins, –. Progovac, Ljiljana, Kate Paesani, Eugenia Casielles, and Ellen Barton (b). Introduction, in Ljiljana Progovac, Kate Paesani, Eugenia Casielles, and Ellen Barton (eds), The syntax of nonsententials: Multidisciplinary perspectives. Amsterdam and Philadelphia: John Benjamins, –. Progovac, Ljiljana, Kate Paesani, Eugenia Casielles, and Ellen Barton (eds) (c). The syntax of nonsententials: Multidisciplinary perspectives. Amsterdam and Philadelphia: John Benjamins. Prüst, Hub (). On discourse structuring, VP anaphora, and gapping. PhD thesis, University of Amsterdam. Prüst, Hub (). Gapping and VP anaphora. PhD thesis, University of Amsterdam. Prüst, Hub, Remko Scha, and Martin van den Berg (). A discourse perspective on verb phrase anaphora, Linguistics and Philosophy (): –. Pullum, Geoffrey K. (). People deletion in English, Working Papers in Linguistics : –. Pullum, Geoffrey K. (). The morpholexical nature of English to-contraction, Language (): –. Pullum, Geoffrey K. (). Hankamer was!, in Sandra Chung, James McCloskey, and Nathan Sanders (eds), Jorge Hankamer Webfest. Santa Cruz: UCSC. Pullum, Geoffrey K. and Arnold Zwicky (). Phonological resolution of syntactic feature conflict, Language : –.





Pulman, Stephen G. (). Higher order unification and the interpretation of focus, Linguistics and Philosophy (): –. Purver, Matthew (). Clarie: Handling clarification requests in a dialogue system, Research on Language and Computation (): –. Purver, Matthew, Ronnie Cann, and Ruth Kempson (). Grammars as parsers: Meeting the dialogue challenge, Research on Language and Computation (–): –. Purver, Matthew, Arash Eshghi, and Julian Hough (). Incremental semantic construction in a dialogue system, in Johan Bos and Stephen Pulman (eds), Proceedings of the th International Conference on Computational Semantics. Oxford: Association for Computational Linguistics, –. Purver, Matthew, Jonathan Ginzburg, and Patrick Healey (). On the means for clarification in dialogue, in Jan van Kuppevelt and Ronnie Smith (eds), Current and new directions in discourse and dialogue. Dordrecht: Kluwer Academic Publishers, –. Purver, Matthew, Eleni Gregoromichelaki, Wilfried Meyer-Viol, and Ronnie Cann (). Splitting the ‘I’s and crossing the ‘You’s: Context, speech acts and grammar, in Pavel Lupkowski and Matthew Purver (eds), Aspects of Semantics and Pragmatics of Dialogue: SemDial , th Workshop on the Semantics and Pragmatics of Dialogue. Poznan: Polish Society for Cognitive Science, –. Purver, Matthew and Ruth Kempson (). Incremental context-based generation for dialogue, in Anja Belz, Roger Evans, and Paul Piwek (eds), Proceedings of the rd International Conference on Natural Language Generation (INLG). Brockenhurst: Springer, –. Quer, Josep and Joanna Rossello (). On sloppy readings, ellipsis and pronouns: Missing arguments in Catalan Sign Language (LSC) and other argument-drop languages, in Victoria Camacho-Taboada, Ángel L. Jiménez-Fernández, Javier Martin-Gonzalez, and Mariano Reyes-Tejedor (eds), Information structure and agreement. Amsterdam and Philadelphia: John Benjamins, –. Quirk, Randolph, Sidney Greenbaum, Geoffrey Leech, and Jan Svartvik (). A comprehensive grammar of the English language. London: Longman. Radford, Andrew (). Syntactic theory and the structure of English. Cambridge: Cambridge University Press. Radford, Andrew and Eiichi Iwasaki (). On swiping in English, Natural Language and Linguistic Theory (): –. Ramos, Sergio. (). A look at fragment answers under code-switching. Talk given at the UIC Talks in Linguistics. University of Illinois at Chicago. Rankin, Tom (). Variational learning in L: The transfer of L syntax and parsing strategies in the interpretation of wh-questions by L German learners of L English, Linguistic Approaches to Bilingualism (): –. Rankin, Tom (). Quotative inversion and particles. Ms. University of Vienna. Ranta, Aarne (). Type theoretical grammar. Oxford: Oxford University Press. Raposo, Eduardo (). On the null object in European Portuguese, in Osvaldo Jaeggli and Carmen Silva- Corvalán (eds), Studies in Romance linguistics. Dordrecht: Foris Publications, –. Rapp, Irene and Arnim von Stechow (). Fast ‘almost’ and the visibility parameter for functional adverbs, Journal of Semantics (): –. Rasekhi, Vahideh (). Missing object in Persian. Ms. Stony Brook University. Reeve, Matthew (). If there’s anything cleft ellipsis resembles, it’s (pseudo)gapping, in Robert E. Santana-LaBarge (ed.), Proceedings of the st West Coast Conference on Formal Linguistics. Somerville, MA: Cascadilla Press, –. Reich, Ingo (). Toward a uniform analysis of short answers and gapping, in Kerstin Schwabe and Susanne Winkler (eds), On information structure, meaning and form. Amsterdam and Philadelphia: John Benjamins, –. Reich, Ingo (). Ellipsis, in Claudia Maienborn, Klaus von Heusinger, and Paul Portner (eds), Semantics: An international handbook of natural language meaning. New York: Mouton de Gruyter, –.





Reich, Ingo and Marga Reis (). Koordination und Subordination, in Jörg Meibauer, Markus Steinbach, and Hans Altmann (eds), Satztypen des Deutschen. Berlin and New York: Mouton de Gruyter, –. Reinhart, Tanya (a). Anaphora and semantic interpretation. Chicago: University of Chicago Press. Reinhart, Tanya (b). Coreference and bound anaphora: A restatement of the anaphora questions, Linguistics and Philosophy : –. Reinhart, Tanya (). Elliptic conjunctions—non-quantificational LF, in Asa Kasher (ed.), The Chomskyan Turn. Cambridge, MA: Blackwell Publishers, –. Remmele, Bettina, Sophia Schopper, Robin Hörnig, and Susanne Winkler (forthcoming). (Non) strategic production planning and ambiguity: Experimental evidence, in Matthias Bauer and Angelika Zirker (eds), Strategies of ambiguity. London: Routledge. Renato, Lacerda (). Contrastive topicalisation in child English. Ms. University of Connecticut. Repp, Sophie (a). Interpreting ellipsis: The changeable presence of the negation in gapping. PhD thesis, Humboldt-Universität zu Berlin. Repp, Sophie (b). ¬(a&b) Gapping, negation and speech act operators, Research on Language and Computation : –. Repp, Sophie (). Negation in gapping. Oxford Studies in Theoretical Linguistics. Oxford: Oxford University Press. Repp, Sophie (). Contrast, in Caroline Féry and Shinichiro Ishihara (eds), The Oxford handbook of information structure. Oxford: Oxford University Press, –. Richards, Norvin (). What moves where when in which language? Doctoral dissertation, Massachusetts Institute of Technology, Cambridge, MA. Richards, Norvin (). Movement in language: Interactions and architectures. Oxford: Oxford University Press. van Riemsdijk, Henk (). A case study in syntactic markedness. Dordrecht: Foris Publications. Rieser, Hannes and David Schlangen (). Special issue on incremental processing in dialogue, Dialogue and Discourse . Rijkhoek, Paulien (). On degree phrases and result clauses. PhD thesis, University of Groningen (GroDiL ). Rissanen, Terhi (). Viittomakielen perusrakenne [The basic structure of sign language]. Helsinki: University of Helsinki. Rissanen, Terhi (). The categories of nominals and verbals and their morpholoy in Finnish Sign Language. Master’s thesis, University of Turku. Ritter, Elizabeth (). Two functional categories in noun phrases: Evidence from Modern Hebrew, in Susan Rothstein (ed.), Syntax and semantics : Perspectives on phrase structure: Heads and licensing. San Diego and London: Academic Press, –. Rivero, María-Luisa (). The location of nonactive voice in Albanian and Modern Greek, Linguistic Inquiry : –. Rizzi, Luigi (). Relativized minimality. Cambridge, MA: MIT Press. Rizzi, Luigi (). The fine structure of the left periphery, in Liliane Haegeman (ed.), Elements of grammar. Dordrecht: Kluwer Academic Publishers, –. Rizzi, Luigi (). Locality and left periphery, in Adriana Belletti (ed.), Structures and beyond: The cartography of syntactic structures, vol. . Oxford: Oxford University Press, –. Roberts, Craige (a). Information structure in discourse: Towards an integrated formal theory of pragmatics, in James H. Yoon and Andreas Kathol (eds), OSU working papers in linguistics : Papers in semantics (). Ohio: Ohio State University, –. Roberts, Craige (). Information structure in discourse: Towards an integrated formal theory of pragmatics, Semantics and Pragmatics (): –, revision of  version of paper originally circulated in . Roberts, Ian (b). Have/be raising, move F, and procrastinate, Linguistic Inquiry (): –.





Roberts, Leah, Ayumi Matsuo, and Nigel Duffield (). Processing VP-ellipsis and VP-anaphora with structurally parallel and non-parallel antecedents: An eye-tracking study, Language and Cognitive Processes (–): –. Rochemont, Michael S. (). Question/answer congruence and focus phrase. Ms. University of Tuebingen and University of British Columbia. Rochemont, Michael S. (). Givenness, in Caroline Féry and Shinichiro Ishihara (eds), The Oxford handbook of information structure. Oxford: Oxford University Press, –. Rodgers, Richard and Lorenz Hart (). Where or when. Van Nuys, CA: Alfred Music Publishing. Rodrigues, Cilene, Andrew Nevins, and Luis Vicente (). Cleaving the interactions between sluicing and preposition stranding, in Danièle Torck and Leo Wetzels (eds), Romance languages and linguistic theory . Amsterdam: John Benjamins, –. Roelofsen, Floris (). Free variable economy, Linguistic Inquiry (): –. Roelofsen, Floris (). Algebraic foundations for the semantic treatment of inquisitive content, Synthese (): –. Roeper, Tom (). The role of universals in the acquisition of gerunds, in Eric Wanner and Lila R. Gleitman (eds), Language acquisition: The state of the art. Cambridge: Cambridge University Press, –. Roeper, Tom (). Universal bilingualism, Bilingualism: Language and Cognition (): –. Roeper, Tom (). Multiple grammars, feature attraction, pied-piping, and the question: Is AGR inside TP?, in Natascha Müller (ed.), Vulnerable domains in multilingualism. Amsterdam and Philadelphia: John Benjamins, –. Roeper, Tom (). Strict interface principles and the acquisition engine: From unlabeled to labeled and minimal modular contact, Language Sciences : –. Roeper, Tom (). Multiple grammars and the logic of learnability in second language acquisition, Frontiers in Psychology (). . Roeper, Tom, Barbara Zurer Pearson, and Margaret Grace (). Quantifier spreading is not distributive, in Nick Danis, Kate Mesh, and Hyunsuk Sung (eds), Proceedings of the th annual Boston University Conference on Language Development (BUCLD ). Somerville, MA: Cascadilla Press, –. Roeper, Tom and Harry N. Seymour (). The place of linguistic theory in the theory of language acquisition and language impairment, in Yonata Levi (ed.), Other children, other languages: Issues in the theory of language acquisition. Hillsdale, NJ: Lawrence Erlbaum Associates, –. Roeper, Tom and Muffy E. A. Siegel (). A lexical transformation for verbal compounds, Linguistic Inquiry (): –. Roeper, Tom, Uri Strauss, and Barbara Zurer Pearson (). The acquisition path of the determiner quantifier ‘every’: Two kinds of spreading, in Tanya Heizmann (ed.), Current issues in first language acquisition. Charleston: Booksurge Publishing, –. Romaine, Suzanne (). Bilingualism. Malden, MA: Blackwell. Rombi, Marie-Françoise and Pierre Alexandre (). Les parlers comoriens, caractéristiques différentielles, position par rapport au swahili, in Marie-Françoise Rombi (ed.), Études sur le bantu oriental. Paris: SELAF, –. Romero, Maribel (). Focus and reconstruction effects in wh-phrases. PhD thesis, University of Massachusetts, Amherst. Romero, Maribel (). Antecedentless sluiced wh-phrases and islands, in Kerstin Schwabe and Ning Zhang (eds), Ellipsis in conjunction. Tübingen: Niemeyer, –. Romero, Maribel (). Correlate restriction and definiteness effect in ellipsis, in Kerstin Schwabe and Susanne Winkler (eds), The interfaces: Deriving and interpreting omitted structures. Amsterdam and Philadelphia: John Benjamins, –. Ronat, Mitsou (). Une contrainte sur l’effacement du nom, in Mitsou Ronat (ed.), Langue: Théorie generative étendue. Paris: Collection Savoir, –. Rooryck, Johan (). Evidentiality, part I, GLOT International : –.





Rooth, Mats (). A comparison of three theories of verb phrase ellipsis, in Wynn Chao and Dierdre Wheeler (eds), University of Massachusetts Occasional Papers in Linguistics. Amherst, MA: GLSA, –. Rooth, Mats (a). Ellipsis redundancy and reduction redundancy, in Steve Berman and Arild Hestvik (eds), Proceedings of the Stuttgart Ellipsis Workshop, Stuttgart. Heidelberg: IBM. Rooth, Mats (b). A theory of focus interpretation, Natural Language Semantics (): –. Rooth, Mats (). Focus, in Shalom Lappin (ed.), The handbook of contemporary semantic theory. Oxford: Blackwell Publishers, –. Rooth, Mats (). Alternative semantics, in Caroline Féry and Shinichiro Ishihara (eds), The Oxford handbook of information structure. Oxford: Oxford University Press, –. Rosen, Carol (). Guess what about?, in Alan Ford, John Reighard and Rajendra Singh (eds), NELS : Proceedings of the sixth annual meeting of the North East Linguistic Society. Amherst, MA: GLSA, –. Ross, John Robert (). Constraints on variables in syntax. PhD thesis, Massachusetts Institute of Technology. Ross, John Robert (a). Auxiliaries as main verbs, in W. Todd (ed.), Studies in philosophical linguistics, Series One. Evanston, IL: Great Expectations Press, –. Ross, John Robert (b). Guess who?, in Robert I. Binnick, Alice Davison, Georgia M. Green, and Jerry L. Morgan (eds), Proceedings of the th annual meeting of the Chicago Linguistic Society. Chicago: Chicago Linguistic Society, –. Ross, John Robert (). Gapping and the order of constituents, in Manfred Bierwisch and Karl E. Heidolph (eds), Progress in linguistics. The Hague: Mouton, –. Ross, John Robert (). Infinite syntax! Norwood, NJ: Ablex. Rottman, Isaac and Masaya Yoshida (). Sluicing, idioms, and island repair, Linguistic Inquiry : –. Roussarie, Laurent and Pascal Amsili (). Interpréter les pronoms phrastiques, Modèles Linguistiques : –. Rouveret, Alain (). VP ellipsis, phases and the syntax of morphology, Natural Language and Linguistic Theory : –. Rowlett, Paul (). The syntax of French. Cambridge: Cambridge University Press. Rowling, Joanne K. (). Harry Potter and the prisoner of Azkaban. London: Bloomsbury Publishing. Rowling, Joanne K. (). Harry Potter and the goblet of fire. London: Bloomsbury Publishing. Rudin, Catherine (). On multiple questions and multiple wh-fronting, Natural Language and Linguistic Theory : –. Rullmann, Hotze (). Maximality in the semantics of wh-constructions. PhD thesis, University of Massachusetts at Amherst. Runic, Jelena (). A new look at clitics, clitic doubling, and argument ellipsis: Evidence from Slavic. PhD thesis, University of Connecticut. Ruwet, N. (). Théorie syntaxique et syntaxe du français. Paris: Editions du Seuil. Ruys, E. G. (). The scope of indefinites. PhD thesis, Utrecht Univeristy. Saab, Andrés (). El dominio de la elipsis nominal en español. Master’s thesis, University of Comahue. Saab, Andrés (). Hacia una teoría de la identidad parcial en la elipsis. PhD thesis, University of Buenos Aires. Saab, Andrés (a). (Im)possible deletion in the Spanish DP, IBERIA: An International Joumal of Theoretical Linguistics (): –. Saab, Andrés (b). Silent interactions. Spanish TP-ellipsis and the theory of island repair, Probus : –. Saab, Andrés (). Ineffable narratives in Spanish: Another case of overgeneration by e-GIVENness, Probus (): –. Saab, Andrés (). A note on someone (else): An island repair solution and its competitors, Linguistic Inquiry (): –.





Saab, Andrés and Anikó Lipták (). Movement and deletion after syntax: Ellipsis by inflection reconsidered, Studia Linguistica (): –. Saab, Andrés and Pablo Zdrojewski (). Anti-repair effects under ellipsis, in Irene Franco, Sara Lusini, and Andrés Saab (eds), Romance languages and linguistic theory : Selected papers from ‘Going Romance’ Leiden . Amsterdam and Philadelphia: John Benjamins, –. Sabbagh, Joseph (). Ordering and linearizing rightward movement, Natural Language and Linguistic Theory (): –. Sabbagh, Joseph (). Right node raising and extraction in Tagalog, Linguistic Inquiry (): –. Sabbagh, Joseph (). Right Node Raising. Ms. University of Texas. Sacks, Harvey (). Lectures on conversation. Malden, MA: Blackwell. Saeboe, Kjell Johan (). Anaphoric presuppositions and zero anaphora. Ms. University of Oslo. Saeboe, Kjell Johan (). Anaphoric presuppositions and zero anaphora, Linguistics and Philosophy (): –. Sáez, Luis (). Peninsular Spanish prenominal possessives in ellipsis contexts: A phase-based account, in Julia Herschensohn (ed.), Romance linguistics : Selected papers from the th Linguistic Symposium on Romance Languages (LSRL). Amsterdam and Philadelphia: John Benjamins, –. Safir, Ken (). Vehicle change and reconstruction in a’-chains, Linguistic Inquiry (): –. Sag, Ivan A. (a). Deletion and logical form. PhD thesis, MIT. Sag, Ivan A. (b). A logical theory of verb phrase deletion, in Salikoko S. Mufwene, Carol A. Walker, and Sanford B. Steever (eds), Papers from the th regional meeting of the Chicago Linguistic Society. Chicago: Chicago Linguistic Society, –. Sag, Ivan A. (). Deletion and logical form. Bloomington, IN: Indiana University Linguistic Club Publications. Sag, Ivan A. (). Deletion and logical form. New York: Garland Publishing. Sag, Ivan A. (). Sign-based Construction Grammar: An informal synopsis, in Hans C. Boas and Ivan A. Sag (eds), Sign-based Construction Grammar. Stanford: CSLI Publications, –. Sag, Ivan A., Hans C. Boas, and Paul Kay (). Introducing sign-based Construction Grammar, in Hans C. Boas and Ivan A. Sag (eds), Sign-based Construction Grammar. Stanford: CSLI Publications, –. Sag, Ivan A. and Janet Dean Fodor (). Extraction without traces, in Paul Aranovich, William Byrne, Susanne Preuss, and Martha Senturia (eds), Proceedings of the th West Coast Conference on Formal Linguistics. Somerville, MA: Cascadilla Press, –. Sag, Ivan A., Gerald Gazdar, Thomas Wasow, and Steven Weisler (). Coordination and how to distinguish categories, Natural Language and Linguistic Theory (): –. Sag, Ivan A. and Jorge Hankamer (). Toward a theory of anaphoric processing, Linguistics and Philosophy : –. Sag, Ivan A., Philip Hofmeister, and Neal Snider (). Processing complexity in subjacency violations: The Complex Noun Phrase Constraint. Ms. Stanford. Sag, Ivan A. and Joanna Nykiel (). Remarks on sluicing, in Stefan Müller (ed.), Proceedings of the th International Conference on Head-Driven Phrase Structure Grammar. Stanford: CSLI, –. Sag, Ivan A., Thomas Wasow, and Emily Bender (). Syntactic theory: A formal introduction. Stanford: CSLI Publications. Sailor, Craig (). Tagged for deletion: A typological approach to VP ellipsis in tag questions. Master’s thesis, UCLA. Sailor, Craig (a). Inflection at the interfaces. Ms. UCLA. Sailor, Craig (b). Tag questions and the typology of VP ellipsis. Ms. UCLA. Sailor, Craig (). The variables of VP-ellipsis. PhD thesis, UCLA. Sailor, Craig and Grace Kuo (). Taiwanese VP ellipsis and the progressive prohibition. Paper presented at the nd North American Conference on Chinese Linguistics.





Sailor, Craig and Gary Thoms (). On the non-existence of non-constituent coordination and nonconstituent ellipsis, in Robert E. Sandana-LaBarge (ed.), Proceedings of the st West Coast Conference on Formal Linguistics. Somerville, MA: Cascadilla Proceedings Project, –. Saito, Mamoru (). Some asymmetries in Japanese and their theoretical implications. PhD thesis, Massachusetts Institute of Technology. Saito, Mamoru (). Three notes on syntactic movement in Japanese, in Issues in Japanese linguistics. Dordrecht: Foris Publications, –. Saito, Mamoru (). Scrambling as semantically vacuous A0 -movement, in Mark R. Baltin and Anthony S. Kroch (eds), Alternative conceptions of phrase structure. Chicago: University of Chicago, –. Saito, Mamoru (). Long-distance scrambling in Japanese, Journal of East Asian Linguistics (): –. Saito, Mamoru (). Ellipsis and pronominal reference in Japanese clefts, Nanzan Linguistics : –. Saito, Mamoru (). Notes on East Asian argument ellipsis, Language Research : –. Saito, Mamoru (). Japanese syntax in a comparative perspective. New York: Oxford University Press. Saito, Mamoru and Keiko Murasugi (a). N0 -deletion in Japanese, in Javier Ormazabal and Carol Tenny (eds), University of Connecticut working papers in linguistics . Connecticut: University of Connecticut, –. Saito, Mamoru and Keiko Murasugi (b). N0 -deletion in Japanese: A preliminary study, in Hajime Hoji (ed.), Japanese and Korean linguistics . Stanford: CSLI Publications, –. Saito, Mamoru, Jonah T.-H. Lin, and Keiko Murasugi (). N0 -ellipsis and the structure of noun phrases in Chinese and Japanese, Journal of East Asian Linguistics (): –. Sakamoto, Yuta (). Absence of case-matching effects in Mongolian sluicing. Ms. University of Connecticut. Sakamoto, Yuta (forthcoming). Japanese null arguments as mixed anaphora: Evidence for the LF-copy analysis, in Proceedings of the th Workshop on Altaic Formal Linguistics (WAFL). MIT Working Papers in Linguistics. Cambridge, MA: MITWPL. Samiian, Vida (). Origins of phrasal categories in Persian: An X-bar analysis. PhD thesis, UCLA. San Pietro, Steve, Ming Xiang, and Jason Merchant (). Accounting for voice mismatch in ellipsis, in Nathan Arnett and Ryan Bennett (eds), Proceedings of the th West Coast Conference on Formal Linguistics. Somerville, MA: Cascadilla Press, –. Sandler, Wendy and Diane Lillo-Martin (). Sign language and linguistic universals. Cambridge: Cambridge University Press. Santos, Ana Lúcia (). Minimal answers: Ellipsis syntax and discourse in the acquisition of European Portuguese. PhD thesis, University of Lisbon. Santos, Ana Lúcia (a). Early VP ellipsis: Production and comprehension evidence, in Acrísio Pires and Jason Rothman (eds), Minimalist inquiries into child and adult language acquisition: Case studies across Portuguese. Berlin and New York: Mouton de Gruyter, –. Santos, Ana Lúcia (b). Minimal answers: Ellipsis syntax and discourse in the acquisition of European Portuguese. Amsterdam and Philadelphia: John Benjamins. Sato, Yosuke (). P-stranding generalization and Bahasa Indonesian: A myth?, Snippets : –. Sato, Yosuke (). Sluicing in Bahasa Indonesia, P-stranding, and interface repair, in Sarah Berson, Alex Bratkievich, Daniel Bruhn, Amy Campbell, Ramon Escamilla, Allegra Giovine, Lindsey Newbold, Marilola Perez, Marta Piqueras-Brunet, and Russell Rhomieux (eds), Proceedings of the annual meeting of the Berkeley Linguistics Society, vol. . Berkeley: Berkeley Linguistics Society, –. Sato, Yosuke (). Minimalist interfaces: Evidence from Indonesian and Javanese. Amsterdam and Philadelphia: John Benjamins.





Sato, Yo (a). Local ambiguity, search strategies and parsing in Dynamic Syntax, in Ruth Kempson, Eleni Gregoromichelaki, and Christine Howes (eds), The dynamics of lexical interfaces. Stanford: CSLI Publications, –. Sato, Yosuke (b). P-stranding under sluicing and repair by ellipsis: Why is Indonesian (not) special? Journal of East Asian Linguistics : –. Sato, Yosuke (). Fragments, ellipsis and PF repair: New evidence from Indonesian, Canadian Journal of Linguistics : –. Sato, Yosuke (). Cyclic Spell-Out and modal complement ellipsis in Javanese, Linguistic Analysis (–): –. Sato, Yosuke (). Argument ellipsis in Javanese and voice agreement, Studia Linguistica : –. Sauerland, Uli (). Sluicing and islands. Ms. MIT. Sauerland, Uli (). Guess how? Proceedings of the th conference of the Student Organization of Linguistics in Europe (ConSoLE ). Leiden: SOLE, –. Sauerland, Uli (). The meaning of chains. PhD thesis, Massachusetts Institute of Technology, Cambridge, MA. Sauerland, Uli (). The silent content of bound variable pronouns, in Kyle Johnson (ed.), Topics in ellipsis. Cambridge: Cambridge University Press, –. Savolainen, Leena (). Interrogatives and negatives in Finnish Sign Language, in Ulrike Zeshan (ed.), Interrogative and negative constructions in sign languages. Nijmegen: Ishara Press, –. Schachter, Paul (a). Constraints on coordination, Language : –. Schachter, Paul (b). Does she or doesn’t she? Linguistic Inquiry (): –. Schachter, Paul (). English propredicates, Linguistic Analysis : –. Schank, Roger and Robert Abelson (). Scripts, plans, goals, and understanding: An inquiry into human knowledge structures. Hillsdale, NJ: Lawrence Erlbaum Associates. Schegloff, E. A. (). Sequence organization in interaction: A primer in conversation analysis I. Cambridge: Cambridge University Press. Schiebe, Traugott (). Zum Problem der grammatisch relevanten Identität, in Ferenc Kiefer and Nicholas Ruwet (eds), Generative grammar in Europe. Dordrecht: D. Reidel, –. Schlangen, David (). A coherence-based approach to the interpretation of non-sentential utterances in dialogue. PhD thesis, University of Edinburgh. Schlenker, Philippe (). Non-redundancy: Towards a semantic reinterpretation of Binding Theory, Natural Language Semantics (): –. Schmeh, Katharina, Peter W. Culicover, Jutta Hartmann, and Susanne Winkler (). Discourse function ambiguity of fragments: A linguistic puzzle, in Susanne Winkler (ed.), Ambiguity. Berlin: Mouton de Gruyter, –. Schoorlemmer, Erik and Tanja Temmerman (). Head movement as a PF phenomenon: Evidence from identity under ellipsis, in Jaehoon Choi, E. Alan Hogue, Jeffrey Punske, Deniz Tat, Jessamyn Schertz, and Alex Trueman (eds), Proceedings of the th West Coast Conference on Formal Linguistics. Somerville, MA: Cascadilla Press, –. Schütze, Carson T. (a). On the nature of default case, Syntax (): –. Schütze, Carson T. (b). Semantically empty lexical heads as last resorts, in Norbert Corver and Henk van Riemsdijk (eds), Semi-lexical categories: The function of content words and the content of function words. Berlin: Mouton de Gruyter, –. Schuyler, Tamara (). Wh-movement out of the site of VP ellipsis. Master’s thesis, UCSC. Schwabe, Kerstin and Klaus von Heusinger (). On shared indefinite expression in coordinative structures, Journal of Semantics : –. Schwabe, Kerstin and Susanne Winkler (eds) (). The interfaces: Deriving and interpreting omitted structures. Amsterdam and Philadelphia: John Benjamins. Schwarz, Bernhard (). On the syntax of either . . . or, Natural Language and Linguistic Theory (): –. Schwarz, Bernhard (). Topics in ellipsis. PhD thesis, University of Massachusetts, Amherst, MA.





Schwarzschild, Roger (). GIVENness, Avoid F and other constraints on the placement of focus, Natural Language Semantics (): –. Schwarzschild, Roger (). Singleton indefinites, Journal of Semantics (): –. Schwarzschild, Roger (). Measure phrases as modifiers of adjectives, Recherches linguistiques de Vincennes : –. di Sciullio, Anna Maria, Pieter Muysken, and Rajendra Singh (). Government and codeswitching, Journal of Linguistics : –. Scott, Tatiana V. (). Whoever doesn’t HOP must be superior: The Russian left-periphery and the emergence of superiority. PhD thesis, Stony Brook University. Sczcegielniak, Adam (). VP ellipsis and topicalization, in L. Bateman and C. Ussery (eds), NELS : Proceedings of the th annual meeting of the North East Linguistic Society. Amherst, MA: GLSA, –. Selkirk, Elisabeth O. (). Sentence prosody: Intonation, stress and phrasing, in J. A. Goldsmith (ed.), The handbook of phonological theory. Cambridge, MA: Blackwell Publishers, –. Selkirk, Elisabeth O. (). On the determiner systems of noun phrases and adjective phrases. Ms. MIT. Selkirk, Elisabeth O. (). Phonology and syntax: The relation between sound and structure. Cambridge, MA: MIT Press. Selkirk, Elisabeth O. (). Focus types and tones in English. Paper presented at the North American Phonology Conference . Selkirk, Elisabeth O. (). Contrastive FOCUS vs. presentational focus: Prosodic evidence from Right Node Raising in English, in Bernard Bel and Isabelle Marlien (eds), Speech Prosody : Proceedings of the First International Speech Prosody Conference. Aix-en-Provence: Université de Provence, Laboratoire Parole et Langage, –. Sener, Serkan and Daiko Takahashi (). Ellipsis of argument in Japanese and Turkish, Nanzan Linguistics : –. Seymour, Harry N., Tom Roeper, and Jill de Villiers (). DELV: Diagnostic evaluation of language variation-norm referenced. London: Pearson. Shabani, Mansour (a). Erteqâye gereye râst dar zabâne fârsi [Right Node Raising in the Persian language], Adabe pazhohi : –. Shabani, Mansour (b). Pâreye jomlehâ dar zabâne fârsi [Fragments in the Persian language], Pazhoheshhâye zabâni : –. Shabani, Mansour (). Erteqâye gereye râst: Sâxtâriye nâhamgen dar zabâne fârsi [Right Node Raising: A nonhomogeneous construction in Persian], Jostârhâye zabâni : –. Shafiei, Nazila (). Ellipsis in Persian complex predicates: VVPE or something else. Paper presented at the annual conference of the Canadian Linguistic Association. Shannon, Claude E. (). Claude E. Shannon: Collected papers. Hoboken, NJ: John Wiley & Sons. Shapiro, Lewis P., Betty Gordon, Nancy Hack, and Jennifer Killackey (). Verb argument structure processing in complex sentences in Broca’s and Wernicke’s aphasia, Brain and Language : –. Shapiro, Lewis P. and Arild Hestvik (). On-line comprehension of VP-ellipsis: Syntactic reconstruction and semantic influence, Journal of Psycholinguistic Research : –. Shapiro, Lewis P., Arild Hestvik, Lesli Lesan, and A. Rachel Garcia (). Charting the time-course of sentence processing: Evidence for an initial and independent structural analysis, Journal of Memory and Language : –. Shapiro, Lewis P., David Swinney, and Susan Borsky (). Online examination of language performance in normal and neurologically impaired adults, American Journal of Speech-Language Pathology : –. Sherman (Ussishkin), Adam (). VP ellipsis and subject positions in Modern Hebrew, in Adam Zachary Wyne (ed.), Proceedings of the th annual meeting of the Israel Association of Theoretical Linguistics. Jerusalem: Akademon, –.





Shimoyama, Junko (). On ‘sluicing’ in Japanese. Ms. University of Massachusetts at Amherst. Shinohara, Michie (). Nihongo-no sakuzyo gensyoo-nituite [On the deletion phenomena in Japanese]. BA thesis, Nanzan University. Shiobara, Kayono (). Derivational linearization at the syntax–prosody interface. Tokyo: Hituzi Syobo. Shopen, Timothy (). A generative theory of ellipsis: a consideration of the linguistic use of silence. PhD thesis, UCLA. Shopen, Timothy (). Ellipsis as grammatical indeterminacy, Foundations of Language : –. Shriberg, Elizabeth (). Preliminaries to a theory of speech disfluencies. PhD thesis, University of California at Berkeley. Siegel, Muffy E. A. (). Compositionality, case, and the scope of auxiliaries, Linguistics and Philosophy (): –. Šimík, Radek (). Modal existential wh-constructions. PhD thesis, University of Groningen. Simons, Mandy (). Semantics and pragmatics in the interpretation of ‘or’, in Effi Georgala and Jonathan Howell (eds), SALT : Proceedings of the th Semantics and Linguistic Theory Conference. Ithaca, NY: CLC Publications, –. Simons, Mandy, Judith Tonhauser, David Beaver, and Craige Roberts (). What projects and why, in SALT : Proceedings of the th Semantics and Linguistic Theory Conference. Ithaca, NY: CLC Publications, –. Sleeman, Petra (). Noun ellipsis in French, Probus : –. Sleeman, Petra (). Licensing empty nouns in French. Amsterdam: Holland Institute of Generative Linguistics. Sleeman, Petra and Aafke Hulk (). L acquisition of noun ellipsis in French and in Dutch: Consequences for linguistic theory, in Sergio Baauw, Frank Drijkoningen, Luisa Meroni, and Manuela Pinto (eds), Romance languages and linguistic theory. Amsterdam and Philadelphia: John Benjamins, –. Smith, Edward E. and John Jonides (). Storage and executive processes in the frontal lobes, Science : –. Smith, Jennifer (). ‘You Ø na hear o’ that kind o’ things’: Negative do in Buckie Scots, English World-Wide (): –. Sneddon, James Neil, Alexander Adelaar, Djenar Dwi N., and Michael C. Ewing (). Indonesian: A comprehensive grammar. London: Routledge. Snider, Neal and Jeffrey Runner (). Structural parallelism aids ellipsis and anaphor resolution: Evidence from eye movements to semantic and phonological neighbors. Talk presented at the Linguistic Society of America Annual Meeting. Snider, Neal and Jeffrey Runner (). Semantic and phonological neighbors are retrieved in verb phrase ellipsis: Evidence for syntactic structure. Ms. University of Rochester. Soh, Hooi Ling and Hiroki Nomoto (). Degree achievements, telicity, and the verbal prefix meN- in Malay, Journal of Linguistics : –. Speer, Shari and Allison Blodgett (). Prosody, in Matthew J. Traxler and Morton Ann Gernsbacher (eds), Handbook of psycholinguistics. Amsterdam: Elsevier, –. Sportiche, Dominique (). Inward bound: Splitting the wh-paradigm and French relative qui. Ms. UCLA. Sprouse, Jon (). The Accent Projection Principle: Why the hell not? University of Pennsylvania Working Papers in Linguistics : –. Sprouse, Jon and Norbert Hornstein (eds) (). Experimental syntax and island effects. Cambridge: Cambridge University Press. Stabler, Edward P. (). Derivational minimalism, in Christian Retoré (ed.), Logical aspects of computational linguistics. Dordrecht: Springer, –. Stabler, Edward P. (). Two models of minimalist, incremental syntactic analysis, Topics in Cognitive Science (): –.





Stainton, Robert J. (a). Neither fragments nor ellipsis, in Ljiljana Progovac, Kate Paesani, Eugenia Casielles, and Ellen Barton (eds), The syntax of nonsententials: Multidisciplinary perspectives. Amsterdam and Philadelphia: John Benjamins, –. Stainton, Robert J. (b). Words and thoughts: Subsentences, ellipsis, and the philosophy of language. Oxford: Oxford University Press. Stanley, Jason (). Context and logical form, Linguistics and Philosophy : –. Starke, Michal (). Move dissolves into Merge: A theory of locality. PhD thesis, University of Geneva. von Stechow, Arnim (). Comparing semantic theories of comparison, Journal of Semantics (): –. von Stechow, Arnim (). Syntax und Semantik, in Arnim von Stechow and Dieter Wunderlich (eds), Semantik: Ein Handbuch der zeitgenoessischen Forschung/Semantics: An international handbook of contemporary research. Berlin and New York: de Gruyter, –. von Stechow, Arnim (). The different readings of wieder ‘again’: A structural account, Journal of Semantics : –. Steedman, Mark (). Combinatory grammars and parasitic gaps, Natural Language and Linguistic Theory (): –. Steedman, Mark (). Gapping as constituent coordination, Linguistics and Philosophy : –. Steedman, Mark (). The syntactic process. Cambridge, MA: MIT Press. Steedman, Mark (). On ‘the computation’, in Gillian Ramchand and Charles Reiss (eds), The Oxford handbook of linguistic interfaces. Oxford: Oxford University Press, –. Steedman, Mark and Jason Baldridge (). Combinatory Categorial Grammar, in Robert D. Borsley and Kersti Börjars (eds), Non-transformational syntax: Formal and explicit models of grammar. Oxford: Blackwell, –. Stepanov, Arthur (). On wh-fronting in Russian, in Pius N. Tamanji and Kiyomi Kusumoto (eds), NELS : Proceedings of the th annual meeting of the North East Linguistic Society. Amherst, MA: Graduate Linguistic Student Association, –. Stepanov, Arthur (). The end of CED? Minimalism and extraction domains, Syntax (): –. Stepanov, Arthur and Penka Stateva (). When QR disobeys superiority, Linguistic Inquiry (): –. Sternefeld, Wolfgang (). Reciprocity and cumulative predication, Natural Language Semantics (): –. Stirling, Colin (). Decidability of higher-order matching, Logical Methods in Computer Science (): –. Stjepanović, Sandra (). VP ellipsis in a verb raising language and implications for the condition on formal identity of verbs, in Jeong-Seok Kim, Satoshi Oku, and Sandra Stjepanović (eds), Is the logic clear? Papers in honor of Howard Lasnik. Connecticut: University of Connecticut, –. Stjepanović, Sandra (). Short-distance movement of wh-phrases in Serbo-Croatian matrix clauses. Paper presented at the Comparative Slavic Morphosyntax Workshop. Stjepanović, Sandra (). Multiple sluicing and superiority in Serbo-Croatian, in Pius N. Tamanji, Masako Hirotani, and Nancy Hall (eds), NELS : Proceedings of the th annual meeting of the North East Linguistic Society. Amherst, MA: GLSA, –. Stjepanović, Sandra (). Multiple wh-fronting in Serbo-Croation matrix question and the matrix sluicing construction, in Cedric Boeckx and Kleanthes K. Grohmann (eds), Multiple wh-fronting. Amsterdam and Philadelphia: John Benjamins, –. Stjepanović, Sandra (). P-stranding under sluicing in a non-P-stranding language? Linguistic Inquiry (): –. Stjepanović, Sandra (). Two cases of violation repair under sluicing, in Jason Merchant and Andrew Simpson (eds), Sluicing: Cross-linguistic perspectives. Oxford: Oxford University Press, –.





Stolterfoht, Britta (). Processing word order variations and ellipses: The interplay of syntax and information structure during sentence comprehension. Leipzig: Max Planck Institute for Human, Cognitive, and Brain Sciences. Stolterfoht, Britta, Angela D. Friederici, Kai Alter, and Anita Steube (). Processing focus structure and implicit prosody during reading: Differential ERP effects, Cognition (): –. Stowe, Laurie A. (). Parsing wh-constructions: Evidence for on-line gap location, Language and Cognitive Processes : –. Stromswold, Karin, David Caplan, Nathaniel Alpert, and Scott Rauch (). Localization of syntactic comprehension by positron emission tomography, Brain and Language : –. Stump, Gregory T. (). Pseudogapping. Ms. Ohio State University. Sturt, Patrick (). The time-course of the application of binding constraints in reference resolution, Journal of Memory and Language : –. Sudo, Yasutada (). Hidden nominal structure in Japanese clausal comparatives, Journal of East Asian Linguistics (): –. Sugisaki, Koji (). ‘One’ issue in acquisition, in Yukio Otsu (ed.), Proceedings of the sixth Tokyo Conference on Psycholinguistics. Tokyo: Hituzi Syobo, –. Sugisaki, Koji (a). NP ellipsis and parameters: An acquisitional perspective, in Alyona Belikova, Luisa Meroni, and Mari Umeda (eds), Proceedings of the nd Conference on Generative Approaches to Language Acquisition North America (GALANA). Somerville, MA: Cascadilla Press, –. Sugisaki, Koji (b). The syntax of swiping: The view from child language. Ms. University of Connecticut. Sugisaki, Koji (forthcoming). Argument ellipsis in child Japanese revisited, in Proceedings of the th Workshop on Altaic Formal Linguistics (WAFL ). Cambridge, MA: MITWPL. Suñer, Margarita and María Yépez (). Null definite objects in Quiteño, Linguistic Inquiry (): –. Suomalaisen viittomakielen perussanakirja [The basic dictionary of FinSL] (). Helsinki: Kuurojen Liitto ry ja Kotimaisten kielten keskus. Surányi, Balázs (). Multiple operator movements in Hungarian. PhD thesis, Utrecht University. Surányi, Balázs (a). ‘Incorporated’ locative adverbials in Hungarian, in Katalin É. Kiss (ed.), Adverbs and adverbial adjuncts at the interfaces. Berlin and New York: Mouton de Gruyter, –. Surányi, Balázs (b). Verbal particles inside and outside vP, Acta Linguistica Hungarica (–): –. de Swart, Henriette (). Négation et coordination: La conjonction ‘ni’, in Reineke Bok-Benneman, Bob de Jonge, Brigitte Kampers-Manhe, and Arie Molendijk (eds), Adverbial Modification. Amsterdam: Rodopi, –. Syrett, Kristen (). Acquisition of comparatives and degree constructions, in Jeffrey Lidz, William Snyder, and Joe Pater (eds), The Oxford handbook of developmental linguistics. Oxford: Oxford University Press, –. Syrett, Kristen and Jeffrey Lidz (). QR in child grammar: Evidence from antecedent-contained deletion, Language Acquisition (): –. Szabolcsi, Anna (). Combinatory grammar and projection from the lexicon, in Ivan A. Sag and Anna Szabolcsi (eds), Lexical matters. Stanford, CA: Center for the Study of Language and Information, –. Szabolcsi, Anna (). The noun phrase, in F. Kiefer and K. É. Kiss (eds), Syntax and semantics : The syntactic structure of Hungarian. San Diego: Academic Press, –. Szabolcsi, Anna (). Ways of scope taking. Dordrecht: Kluwer Academic Publishers. Szabolcsi, Anna (). Strong vs. weak islands, in Martin Everaert and Henk van Riemsdijk (eds), The Blackwell companion to syntax. Malden, MA and Oxford: Blackwell, –. Szabolcsi, Anna (). Quantification. Cambridge: Cambridge University Press. Szabolcsi, Anna and Frans Zwarts (). Weak islands and an algebraic semantics for scope taking, Natural Language Semantics (): –. Szczegielniak, Adam (). Relativization and ellipsis. PhD thesis, Harvard University. Szczegielniak, Adam (). All sluiced up, but no alleviation in sight. Ms. Boston College.





Szczegielniak, Adam (). Relativization that you did. Cambridge, MA: MIT Press. Szczegielniak, Adam (). Islands in sluicing in Polish, in Natasha Abner and Jane Bishop (eds), Proceedings from the th West Coast Conference on Formal Linguistics. Somerville, CA: Cascadilla Proceedings Project, –. Takahashi, Daiko (). Movement of wh-phrases in Japanese, Natural Language and Linguistic Theory (): –. Takahashi, Daiko (). Sluicing in Japanese, Journal of East Asian Linguistics (): –. Takahashi, Daiko (). Apparent parasitic gaps and null arguments in Japanese, Journal of East Asian Linguistics (): –. Takahashi, Daiko (a). Noun phrase ellipsis, in Shigeru Miyagawa and Mamoru Saito (eds), The Oxford handbook of Japanese linguistics. Oxford: Oxford University Press, –. Takahashi, Daiko (b). Quantificational null objects and argument ellipsis, Linguistic Inquiry (): –. Takahashi, Daiko (). Argument ellipsis, anti-agreement, and scrambling, in Mamoru Saito (ed.), Japanese syntax in comparative perspective. Oxford: Oxford University Press, –. Takahashi, Daiko (). Koo syooryaku [Argument ellipsis], in Keiko Murasugi, Mamoru Saito, Yoichi Miyamoto, and Kensuke Takita (eds), Nihongo bunpoo Handobukku [A handbook of Japanese grammar]. Tokyo: Kaitakusha, –. Takahashi, Daiko and Sichao Lin (). Two notes on multiple sluicing in Chinese and Japanese, Nanzan Linguistics : –. Takahashi, Shoichi (). Pseudogapping: The view from Scandinavian languages. Paper presented at the Comparative Germanic Syntax Workshop . Takahashi, Shoichi (). Pseudogapping and cyclic linearization, in Keir Moulton and Matthew Wolf (eds), NELS : Proceedings of the th meeting of the North East Linguistic Society. Amherst, MA: Graduate Linguistic Student Association, –. Takahashi, Shoichi and Danny Fox (). MaxElide and the re-binding problem, in Effi Georgala and Jonathan Howell (eds), Proceedings of Semantics and Linguistic Theory (SALT XV). Ithaca, NY: CLC Publications, –. Takita, Kensuke (). NP-internal honorification and N0 -deletion in Japanese, Nanzan Linguistics (): –. Takita, Kensuke (). ‘Genuine’ sluicing in Japanese, in Ryan Bochnak, Nassira Nicola, Peet Klecha, Jasmine Urban, Alice Lemieux, and Christina Weaver (eds), Proceedings of the th annual meeting of the Chicago Linguistic Society. Chicago: Chicago Linguistic Society, –. Takita, Kensuke (). An argument for argument ellipsis from -sika NPIs, in Suzi Lima, Kevin Mullin, and Brian Smith (eds), NELS : Proceedings of the th annual meeting of the North East Linguistic Society. Amherst: GLSA, –. Takita, Kensuke and Nobu Goto (). On (im)possible N0 -deletion within PPs, Nanzan Linguistics : –. Tanaka, Hidekazu (). Voice mismatch and syntactic identity, Linguistic Inquiry : –. Tancredi, Christopher (). Deletion, deaccenting and presupposition. PhD thesis, Massachusetts Institute of Technology, Cambridge, MA. Tanenhaus, Michael K. and Greg N. Carlson (). Comprehension of deep and surface verb phrase anaphors, Language and Cognitive Processes (): –. Tang, C.-C. Jane (). Chinese phrase structure and the extended X0 -theory. PhD thesis, Cornell University. Tao, Hongyin and Charles F. Meyer (). Gapped coordinations in English: Form, usage, and implications for linguistic theory, Corpus Linguistics and Linguistic Theory (): –. Temmerman, Tanja (). Multidominance, ellipsis, and quantifier scope. PhD thesis, Leiden University. Temmerman, Tanja (). The syntax of Dutch embedded fragment answers: On the PF-theory of islands and the wh/sluicing-correlation, Natural Language and Linguistic Theory : –.





Tesnière, Lucien (). Éléments de syntaxe structurale. Paris: Klincksieck. Tesnière, Lucien (). Elements of structural syntax. Amsterdam and Philadelphia: John Benjamins. Testelets, Yakov (). Ellipsis v russkom jazyke: Teoretičeskij i opisatelnyj podxody. Paper presented at the conference Tipologija morfosintaksičeskix parametrov. Thompson, Andrea (). Beyond deep and surface: Explorations in the typology of anaphora. PhD thesis, UCSC. Thoms, Gary (a). From economy to locality: Do-support as head movement. Ms. Strathclyde University. Thoms, Gary (b). ‘Verb floating’ and VP-ellipsis: Towards a movement account of ellipsis licensing, in Jeroen van Craenenbroeck (ed.), Linguistic variation yearbook . Amsterdam: John Benjamins, –. Thoms, Gary (). Towards a movement theory of ellipsis licensing. Paper presented at UiL-OTS. Thoms, Gary (). MaxElide and clause structure in Scottish Gaelic, Linguistic Inquiry : –. Thoms, Gary (). Syntactic identity, parallelism and accommodated antecedents, Lingua : –. Thoms, Gary (). Short answers in Scottish Gaelic and their theoretical implications, Natural Language and Linguistic Theory : –. Thoms, Gary, David Adger, Caroline Heycock, and Jennifer Smith (). ‘Remarks on negation in varieties of Scots’. Paper presented at the ReCoS Workshop on English Dialects. Thoms, Gary, David Adger, Caroline Heycock, and Jennifer Smith (). On some surprising contractions in Scots dialects. Paper presented at LAGB (Linguistic Association of Great Britain). Thoms, Gary and Craig Sailor (). When silence gets in the way: Extration from do-ellipsis in British dialects. Paper presented at the LSA. Thoms, Gary and George Walkden (). vP-fronting with and without remnant movement. Ms. University of Glasgow and University of Manchester. Thornton, Rosalind (). Minimal answers: Ellipsis, syntax and discourse in the acquisition of European Portuguese, Language (): –. Thornton, Rosalind and Kenneth Wexler (). Principle B, VP ellipsis, and interpretation in child grammar. Cambridge, MA: MIT Press. Tian, Ye and Jonathan Ginzburg (). No, I am: What are you saying ‘no’ to. Ms. Université ParisDiderot. Ticio, Emma (). On the structure of DPs. PhD thesis, University of Connecticut at Storrs. Timm, Lenora A. (). Spanish-English code-switching: El porqué and how-not-to, Romance Philology : –. Titov, Elena (). Encoding focus and contrast in Russian, in Ad Neeleman and Reiko Vermeulen (eds), The syntax of topic, focus and contrast: An interface-based approach. Berlin: Mouton de Gruyter, –. Todorovic, Neda (). VP-ellipsis and all its phases: The role of aspect in VP-ellipsis in Serbian, in Cassandra Chapman, Olena Kit, and Ivona Kučerová (eds), Proceedings of Formal Approaches to Slavic Linguistics . Ann Arbor: Michigan Slavic Publications. Tomioka, Satoshi (). A sloppy identity puzzle, Natural Language Semantics (): –. Tomioka, Satoshi (). The semantics of Japanese null pronouns and its crosslinguistic implications, in Kerstin Schwabe and Susanne Winkler (eds), The interfaces: Deriving and interpreting omitted structures. Amsterdam and Philadelphia: John Benjamins, –. Tomioka, Satoshi (). A step-by-step guide to ellipsis resolution, in Kyle Johnson (ed.), Topics in ellipsis. Cambridge: Cambridge University Press, –. Tomioka, Satoshi (). Sluicing in Mandarin Chinese, in Jason Merchant and Andrew Simpson (eds), Sluicing: Cross-linguistic perspectives. Oxford: Oxford University Press, –. Toosarvandani, Maziar (). v-stranding VPE: Ellipsis in Farsi complex predicates, in Christopher Davis, Amy Rose Deal, and Youri Zabbal (eds), NELS : Proceedings of the th annual meeting of the North East Linguistic Society. Amherst, MA: GLSA, –.





Toosarvandani, Maziar (). Wh-movement and the syntax of sluicing, Journal of Linguistics : –. Toosarvandani, Maziar (). Ellipsis in Farsi complex predicates, Syntax (): –. Toosarvandani, Maziar (a). Corrective but coordinates clauses not always but sometimes, Natural Language and Linguistic Theory : –. Toosarvandani, Maziar (b). Gapping is VP-ellipsis: A reply to Johnson. Ms. MIT. Toosarvandani, Maziar (). Embedding the antecedent in gapping: low coordination and the role of parallelism, Linguistic Inquiry : –. Toribio, Almeida Jacqueline and Kay González-Vilbazo (). Operator movement in EnglishSpanish and German-Spanish codeswitching, in Jeff MacSwan (ed.), Grammatical theory and bilingual codeswitching. Cambridge, MA: MIT Press, –. Traugott, Elizabeth (). Towards a constructional framework for research on language change, Cognitive Linguistic Studies (): –. Traxler, Matthew J. and Lyn Frazier (). The role of pragmatic principles in resolving attachment ambiguities: Evidence from eye movements, Memory and Cognition (): –. Traxler, Matthew J. and Martin J. Pickering (). Plausibility and the processing of unbounded dependencies: An eye-tracking study, Journal of Memory and Language (): –. Ueno, Yoshio (). An automodular view of ellipsis. Tokyo: Waseda University Press. Ueyama, Aymui (). Two types of dependency. PhD thesis, University of Southern California. Ueyama, Aymui (). Two types of scrambling constructions in Japanese, in Andrew Barss (ed.), Anaphora: A reference guide. Malden, MA: Oxford University Press, –. Uriagereka, Juan (). Multiple spell-out, in Samuel Epstein and Norbert Hornstein (eds), Working minimalism. Cambridge, MA: MIT Press, –. Vallduví, Enric (). Information structure, in Maria Aloni and Paul Dekker (eds), The Cambridge handbook of formal semantics. Cambridge: Cambridge University Press, –. Valmala, Vidal (). On Right Node Raising in Catalan and Spanish, Catalan Journal of Linguistics : –. Valois, Daniel, Phaedra Royle, Ann Sutton, and Evelyne Bourdua-Roy (). L’ellipse du nom en français: Le rôle des données de l’acquisition pour la théorie linguistique, Canadian Journal of Linguistics (): –. Van Valin Jr, Robert D. (). An empty category as the subject of a tensed S in English, Linguistic Inquiry : –. Van Valin Jr, Robert D. and Randy J. LaPolla (). Syntax: Structure, meaning and function. Cambridge: Cambridge University Press. Vanden Wyngaerd, Guido (). Object shift as an A-movement rule, MIT Working Papers in Linguistics : –. Vanden Wyngaerd, Guido (). Gapping constituents. Ms. HUB. Vasić, Nada, Sergey Avrutin, and Esther Ruigendijk (a). The economy hierarchy of referential dependencies in child language: Evidence from Dutch VP-ellipsis, in Kamil Ud Deen, Jun Nomura, Barbara Schulz, and Bonnie D. Schwartz (eds), University of Connecticut Occasional Papers in Linguistics : Proceedings of the Inaugural Conference on Generative Approaches to Language Acquisition—North America (GALANA). Storrs: University of Connecticut Linguistics Club, –. Vasić, Nada, Sergey Avrutin, and Esther Ruigendijk (b). Interpretation of pronouns in VPellipsis constructions in Dutch Broca’s and Wernicke’s aphasia, Brain and Language : –. Vergnaud, Jean-Roger (). French relative clauses. PhD dissertation, MIT. Vicente, Luis (). Negative short replies in Spanish, in Jeroen van de Weijer and Bettelou Los (eds), Linguistics in the Netherlands . Amsterdam and Philadelphia: John Benjamins, –. Vicente, Luis (). Syntactic isomorphism and non-isomorphism under ellipsis. Ms. UCSC. Vicente, Luis (a). A note on the movement analysis of gapping, Linguistic Inquiry (): –.





Vicente, Luis (b). On the syntax of adversative coordination, Natural Language and Linguistic Theory (): –. Vicente, Luis (a). Morphological case mismatches under sluicing. Ms. University of Potsdam. Vicente, Luis (b). Sluicing and its subtypes. Ms. University of Potsdam. Vicente, Luis (). Morphological case mismatches under sluicing. Snippets : –. de Villiers, Jill, Peter A. de Villiers, and Tom Roeper (). Wh-questions: Moving beyond the first phase, Lingua (): –. Vlachos, Christos (). The mechanics of sluicing, Linguistic Analysis : –. de Vries, Gertrud (). On coordination and ellipsis. PhD thesis, Tilburg University. de Vries, Mark (). Ellipsis in nevenschikking: Voorwaarts deleren maar achterwaarts delen, Tabu : –. de Vries, Mark (). On multidominance and linearization, Biolinguistics : –. de Vries, Mark (). Multidominance and locality, Lingua : –. de Vries, Mark (). Across-the-board phenomena, in Martin Everaert and Henk van Riemsdijk (eds), The Wiley Blackwell companion to syntax. nd edition. Hoboken: Wiley-Blackwell, –. Wagner, Michael and Duane Watson (). Experimental and theoretical advanes in prosody: A review, Language and Cognitive Processes (–): –. Walenski, Matthew, Tracy Love, Michelle Ferrill, Isabelle Deschamps, Peter Pieperhoff, Katrin Amunts, Yosef Grodzinsky, and Lewis P. Shapiro (). Structure–function correspondences in Broca’s aphasia: Evidence from MRI and comprehension of verb phrase ellipsis constructions. Paper presented at the Clinical Aphasiology Conference. Walker, Edward, Phillip Gough, and R. Wall (). Grammatical relations and the search of sentences in immediate memory. Paper presented at the Midwestern Psychological Association. Walters, Minette (). The dark room. London: Pan Books. Wang, Chyan-An (). Sluicing and resumption, in Emily Elfner and Martin Walkow (eds), NELS : Proceedings of the th annual meeting of the North East Linguistic Society. Amherst: GLSA, –. Ward, Gregory and Julia Hirschberg (). Implicating uncertainty: The pragmatics of fall-rise intonation, Language : –. Ward, Gregory, Richard Sproat, and Gail McKoon (). A pragmatic account of so-called anaphoric islands, Language : –. Warner, Anthony (). The structure of English auxiliaries: A phrase structure grammar. Bloomington, IN: Indiana University Linguistics Club. Warner, Anthony (). Ellipsis conditions and the status of the English copula, York Papers in Linguistics : –. Wasow, Thomas (). Anaphoric relations in English. Doctoral dissertation, MIT, Cambridge, MA. Wasow, Thomas (). Anaphora in generative grammar. Ghent: E. Story-Scientia. Wasow, Thomas and Tom Roeper (). On the subject of gerunds, Foundations of Language (): –. Webber, Bonnie (). A formal approach to discourse anaphora (Outstanding Dissertations in Linguistics). New York: Garland. Weber, David (). Relativization and nominalized clauses in Huallaga Quechua. Berkeley: University of California Press. Weber, Heinz J. (). Dependenzgrammatik: Ein Arbeitsbuch. Tübingen: Gunter Narr Verlag. van Weelden, Lisanne (). Discourse integration in children: An eye-tracking study to children’s interpretation of bare cardinals. Master’s thesis, Utrecht Univeristy. Weerman, Fred (). The V conspiracy: A synchronic and diachronic analysis of verbal positions in Germanic language. Dordrecht: Foris Publications. Weir, Andrew (). Scots na: Clitic or affix? Ms. University of Massachusetts at Amherst.





Weir, Andrew (). Fragments and clausal ellipsis. PhD thesis, University of Massachusetts at Amherst. Weir, Andrew (). Fragment answers and exceptional movement under ellipsis: A PF-movement account, in Thuy Bui and Deniz Özyildiz (eds), NELS : Proceedings of the th annual meeting of the North East Linguistic Society. Amherst, MA: GLSA, –. Wentz, J. and E. McClure (). Ellipsis in bilingual discourse. In Papers from the th regional meeting of the Chicago Linguistic Society. Chicago: Chicago Linguistic Society, –. Werner, Edwin (). The ellipsis of ‘ellipsis’: A reanalysis of ‘elided’ noun phrase structures in German. Master’s thesis, Utrecht Univeristy. Wesche, B. (). Symmetric coordination. Tübingen: Niemeyer. Wescoat, Michael (). Sloppy readings with embedded antecedents. MS. Stanford University. Wexler, Kenneth and Peter Culicover (). Formal principles of language acquisition. Cambridge, MA: MIT Press. Whitman, Neal (). Semantics and pragmatics of English verbal dependent coordination, Language (): –. Whitman, Neal (). Right-node wrapping: Multimodal categorial grammar and the ‘Friends in Low Places’ coordination, in Erhard Hinrichs and John Nerbonne (eds), Theory and evidence in semantics. Stanford: CSLI, –. Wiedmann, Natalie and Susanne Winkler (). The role of prosody in children’s processing of ambiguous sentences, in Susanne Winkler (ed.), Ambiguity. Berlin: Mouton de Gruyter, –. Wijnen, Frank and Edith Kaan (). Dynamics of semantic processing: The interpretation of bare quantifiers, Language and Cognitive Processes (): –. Wijnen, Frank, Tom Roeper, and Hiske van der Meulen (). Discourse binding: Does it begin with nominal ellipsis?, in Jacqueline van Kampen and Sergio Baauw (eds), Proceedings of GALA  on language acquisition. Utrecht: Utrecht University, –. Wilder, Chris (). Coordination, ATB, and ellipsis, Groninger Arbeiten zur Germanistischen Linguistik : –. Wilder, Chris (). Some properties of ellipsis in coordination, in Artemis Alexiadou and T. Hall (eds), Studies on Universal Grammar and typological variation. Amsterdam: John Benjamins, –. Wilder, Chris (). Right Node Raising and the LCA, in Sonya Bird, Andrew Carnie, Jason D. Haugen, and Peter Norquest (eds), Proceedings of the th West Coast Conference on Formal Linguistics. Somerville, MA: Cascadilla Press, –. Wilder, Chris (). Shared constituents and linearization, in Kyle Johnson (ed.), Topics in ellipsis. Cambridge: Cambridge University Press, –. Williams, Edwin (). Rule ordering in syntax. PhD thesis, Massachusetts Institute of Technology. Williams, Edwin (a). Across-the-board application of rules, Linguistic Inquiry (): –. Williams, Edwin (b). Discourse and logical form, Linguistic Inquiry : –. Williams, Edwin (c). On deep and surface anaphora, Linguistic Inquiry : –. Williams, Edwin (). Transformationless grammar, Linguistic Inquiry (): –. Williams, Edwin (). The ATB-theory of parasitic gaps, The Linguistic Review : –. Williams, Edwin (). Blocking and anaphora, Linguistic Inquiry (): –. Winkler, Susanne (). Ellipsis at the interfaces: An information-structural investigation of sentence-bound and discourse-bound ellipsis in English. PhD thesis, University of Tübingen. Winkler, Susanne (). Ellipsis and focus in generative grammar. Berlin: Mouton de Gruyter. Winkler, Susanne (). Ellipsis and the division of labor at the interfaces, in Valeria Molnár and Susanne Winkler (eds), The architecture of focus. Berlin and New York: Mouton de Gruyter, –. Winkler, Susanne (). Syntactic diagnostics for extraction of focus from ellipsis site, in Lisa L.-S. Cheng and Norbert Corver (eds), Diagnosing syntax. Oxford: Oxford University Press, –. Winkler, Susanne (). Exploring ambiguitiy and the ambiguity model from a transdisciplinary perspective, in Susanne Winkler (ed.), Ambiguity. Berlin: Mouton de Gruyter, –.





Winkler, Susanne (). Ellipsis and information structure, in Caroline Féry and Shinichiro Ishihara (eds), The Oxford handbook of information structure. Oxford: Oxford University Press, –. Winkler, Susanne and Kerstin Schwabe (). Exploring the interfaces from the perspective of omitted structures, in Kerstin Schwabe and Susanne Winkler (eds), The interfaces: Deriving and interpreting omitted structures. Amsterdam and Philadelphia: John Benjamins, –. Winter, Yoad (). Flexibility principles in Boolean semantics. Cambridge, MA: MIT Press. Winterson, Jeanette (). The passion. Westminster: Penguin Books. Wittgenstein, Ludwig (). Philosophical investigations. Oxford: Basil Blackwell. Wold, Dag (). Antecedent-contained deletion in comparative constructions. Ms. MIT. Wong, Kwong-Cheong and Jonathan Ginzburg (). Investigating non-sentential utterances in a spoken Chinese corpus, in Proceedings of PACLING , Tokyo. Woolford, Ellen (). Bilingual code-switching and syntactic theory, Linguistic Inquiry (): –. Woolford, Ellen (). On the application of wh-movement and inversion in code-switching sentences, Revue Québécoise de Linguistique (): –. Wulf, Alyssa, Paul Dudis, Robert Bailey, and Ceil Lucas (). Variable subject presence in ASL narratives, Sign Language Studies : –. Wurmbrand, Susi (). Stripping and topless complements, Linguistic Inquiry : –. Xiang, Ming (). A phrasal analysis of Chinese comparatives, in Amy Franklin, Jon Cihlar, David Kaiser, and Irene Kimbara (eds), Proceedings of the th annual meeting of the Chicago Linguistic Society. Chicago: Chicago Linguistic Society, –. Xiang, Ming, Jason Grove, and Jason Merchant (). Ellipsis sites induce structural priming effects. Ms. University of Chicago. Yamauchi, Nobuyuki (). Some properties of vice versa: A corpus-based approach, Journal of Culture and Information Science : –. Yang, Charles D. (). Knowledge and learning in natural language. Oxford: Oxford University Press. Yatabe, Shuichi (). The syntax and semantics of left-node raising in Japanese, in Daniel P. Flickinger and Andreas Kathol (eds), Proceedings of the th International Conference on Head-Driven Phrase Structure Grammar. Stanford: CSLI Publications, –. Yatabe, Shuichi (). A linearization-based theory of summative agreement in peripheral-node raising constructions, in Jong-Bok Kim and Stephen Wechsler (eds), Proceedings of the th International Conference on Head-Driven Phrase Structure Grammar. Stanford: CSLI Publications, –. Yatabe, Shuichi (). Comparison of the ellipsis-based theory of non-constituent coordination with its alternatives, in Stefan Müller (ed.), Proceedings of the th International Conference on HeadDriven Phrase Structure Grammar. Stanford: CSLI Publications, –. Yoshida, Masaya (). The rightward movement analysis of gapping in NP and its structural implications, in John Aldrete, Chung-Hye Han, and Alexei Kochetov (eds), Proceedings of the West Coast Conference on Formal Linguistics. Somerville, MA: Cascadilla Press, –. Yoshida, Masaya (). ‘Antecedent-contained’ sluicing, Linguistic Inquiry (): –. Yoshida, Masaya, Lauren Ackerman, Rebekah Ward, and Morgan Purrier (a). The processing of backward sluicing. Paper presented at the th Annual CUNY Conference on Human Sentence Processing. Yoshida, Masaya, Katy Carlson, and Michael Walsh Dickey (). Incremental parsing, gapping, and connectives. Paper presented at the th Annual CUNY Conference on Human Sentence Processing. Columbus, SC. Yoshida, Masaya, Michael Walsh Dickey, and Patrick Sturt (a). Predictive processing of syntactic structure: Sluicing and ellipsis in real-time sentence processing, Language and Cognitive Processes (): –. Yoshida, Masaya, Jiyeon Lee, and Michael Walsh Dickey (b). The island (in)sensitivity of sluicing and sprouting, in Jon Sprouse and Norbert Hornstein (eds), Experimental syntax and island effects. Cambridge: Cambridge University Press, –.





Yoshida, Masaya, Tim Hunter, and Michael Frazier (). Parasitic gaps licensed by elided syntactic structure, Natural Language and Linguistic Theory : –. Yoshida, Masaya, Nina Kazanina, Leticia Pablos, and Patrick Sturt (a). On the origin of islands, Language, Cognition and Neuroscience : –. Yoshida, Masaya, Chizuru Nakao, and Iván Ortega-Santos (b). The syntax of why-stripping, Natural Language and Linguistic Theory : –. Yoshida, Masaya, Honglei Wang, and David Potter (b). Remarks on ‘gapping’ in DP, Linguistic Inquiry (): –. Zagona, Karen (). Government and proper government of verbal projections. PhD thesis, University of Washington. Zagona, Karen (a). Proper government of antecedentless VPs in English and Spanish, Natural Language and Linguistic Theory : –. Zagona, Karen (b). Verb phrase syntax: A parametric study of English and Spanish. Dordrecht: Kluwer Academic Publishers. Zamparelli, Roberto (). Coordination, in Klaus von Heusinger, Claudia Maienborn, and Paul Portner (eds), Semantics: An international handbook of natural language meaning. Berlin: Mouton de Gruyter, –. Zipf, George K. (). Human behavior and the principle of least effort. Cambridge: Addison Wesley. Zocca, Cynthia (). O que não está lá: Um estudo sobre morfologia flexional em elipses. Master’s thesis, Universidad Estatal de Campinas. Zoerner, Cyril Edward, III (). Coordination: The syntax of &P. PhD thesis, University of California at Irvine. van Zonneveld, Ron (). Kleine syntaxis van het Nederlands. Dordrecht: ICG Publications. Zribi-Hertz, Anne (). Relations anaphoriques en français: Esquisse d’une grammaire générative raisonnée de la féflexivité et de l’ellipse structurale. PhD thesis, Université Paris . Zubizarreta, María-Luisa (). On the relationship of the lexion to syntax. PhD thesis, MIT. Zurif, Edgar B., David Swinney, Penny Prather, Julie Solomon and Camille Bushell (). An on-line analysis of syntactic processing in Broca’s and Wernicke’s aphasia, Brain and Language : –. Zwaan, Rolf A. and Gabriel A. Radvansky (). Situation models in language comprehension and memory, Psychological Bulletin (): –. Zwart, C. Jan-Wouter (). Dutch syntax: A Minimalist approach. PhD thesis, University of Groningen. Zwart, C. Jan-Wouter (). Dutch is head-initial, The Linguistic Review (–): –. Zwart, C. Jan-Wouter (). The syntax of Dutch. Cambridge: Cambridge University Press. Zwarts, Frans (). Extractie uit prepositionele woordgroepen in het Nederlands, in Ans van Berkel, Willem Blok, Gerard Brummel, and Theo A. J. M. Janssen (eds), Proeven van Neerlandistiek. Groningen: Nederlands Instituut, –. Zwicky, Arnold (). Stranded to and phonological phrasing, Linguistics (/): –.

INDEX ................... acceptable ungrammaticality  acquisition , , , –, , , –, , , –, –, –, , , , , ,  acquisition path –, –, , , , ,  across-the-board movement , , , , , , –, , –; see also ATB agentive ,  agreement , , , , , –, , , , , , , , , , , , , , , , –, –, , , , , –, –, , , , –, , , –, , , , , , , , , , , , , , , ,  align , , , , –, , ,  alternative semantics  ambiguity , , , , –, , , , , , , –, , –, , –, –, , , , –, , , , , , , , , – anaphor deep anaphor , , , , , –, –, –, –, , –, , –, –, –, , , , ,  discourse anaphora ,  do it anaphor , , –,  do so anaphor ,  one anaphor ,  sentential it anaphor  surface anaphor , , , , –, –, , , , –, –, , –, , , , , , ,  VP anaphor , , , , , , , , – antecedent ACD –, , , –, –, , , , , , , , ; see also antecedent-contained deletion antecedent-contained deletion , , , , , , , , –, , , , , , , , ; see also ACD antecedentless –, –, , ; see also non-linguistic antecedent; implicit correlate; pragmatic antecedent; pragmatic control linguistic antecedent , –, –, –, , , –, –, , , , , , , , , , , , , –, , –, , , , , 

missing antecedent , , –, ,  nominal antecedent , , –,  non-linguistic antecedent , , , , , ; see also antecedentless; implicit correlate; pragmatic antecedent; pragmatic control pragmatic antecedent –, , –; see also antecedentless; non-linguistic antecedent; implicit correlate; pragmatic control split antecedent –, ,  anti-connectivity , , – anti-government constraint – aphasia , , , –, –, – appositives , ,  argument cluster conjunction – argument deletion  argument drop –, , , –, , –,  argument ellipsis , , , , , , , –, , , , , , , , , , –, , , , , , , –, –, , ,  Argument Realization Principle  argument-adjunct asymmetry  assertion , , , , , –, , , ,  ATB , , –, –, –, –, , –, –, , , , –, –, ; see also across-the-board movement augment , –, ,  auxiliary , , , , –, , , –, , , , , –, –, , , , , , , , –, , , , –, –, , , –, , , , –, –, , , , , , , , –, –, , , –, , , , , –, –, , –, , –, , , , , –, , –, , , , –,  Avoid Pronoun Principle  BAC , ; see also Backward Anaphora Constraint Backward Anaphora Constraint , , , ; see also BAC backward conjunction reduction –, , ,  backward ellipsis , , , ,  bare nominal , – bilingualism , , –, –,  binary branching , 





binding , , , –, , , , , –, –, , , , , , , , –, , , , , , –, –, , , , , , , , –, , , , , , , , , , , , , , , , , ,  borrowing  bound pronouns ,  bound variable , , –, –, –, , , , , , –, –, , – British English do , , , , –, , , – Broca , , –, , –,  c-command –, , , , , –, , , , –, , , , , , , , , , , , –, ,  cascaded ellipsis  Case Condition ,  case marking , –, –, , , , , , , , , , , , –, , , , ,  case matching , , , , , , , , , , , , , ,  case-marked cleft –, –, ,  cataphora , ,  Categorial Grammar , –, , , ; see also CG catena , –, –,  CG , –, , –, , –; see also Categorial Grammar clarification –, , , –, , , –, , , , , –, ,  clausal ellipsis , , , , , , , , , , , , , , , –, –, , , –, , –, , ,  cleft , , , , –, , , –, –, –, –, –, , , , , , , –, –, , , –, –, –, , , , , ,  codeswitching , , –, –, –, – coherence , , –, –, –, –, , ,  coindexing , , , , ,  communicative motivation  comparative attributive comparative ,  case-marked comparative –, , , –, ,  clausal comparative , , , , , –, – comparative deletion , , , , , , , –, , , , , , , , , , ,  comparative ellipsis , , , , , , , , ,  comparative subdeletion –, ,  direct analysis –, , , –, , –,  matching analysis –, 

non-case-marked comparative , ,  non-elliptical comparative ,  phrasal comparative , , , –, –, , , , –, – complete subtree , – complex predicate , , , ,  concealed cleft  Condition C , , , , , , ; see also Principle C conjunction , , , , , , –, , , , , , , , –, , , , , , , , –, , , –, –, –, , , –, –, , , –, , , –, –, , , , , , , , , ,  conjunction reduction , , , , , , , , , –, , –, , –, –, , , –, , , , , ; see also backward conjunction reduction; CR; forward conjunction reduction connectivity , , , , –, , –, –, , , , , –, , , , , , , , , –, , , , –, –, – Constituent Raising  constraint , –, , , , , –, –, , –, –, , –, –, , –, , , , , , , , , , , , , , –, –, , , , –, , –, –, , , –, , , , , –, –, –, , , , , , –, –, , , , , , , , , –, –, , –, –, , –, , –, , , –, , , , –, –, , , , , , , , , , –, , , , , , , , , –, , , , , , , –, , , , , , , , , –, –, –, , , , , , , , , –, – constraint-based ,  constraint-free ,  constructed action –, –, – Construction Grammar , , , ,  context-dependence , , ,  contrast contrast sluicing , , –, , –,  contrastive ellipsis , –, , ,  contrastive focus , –, , , , , , , , –, , , , , –, , , –, – contrastive remnant condition , –, , ,  contrastive stress , , , , , , ,  contrastive remnant , –, , , , , , , – contrastive topic , , –, , –, , , , –, –, , – conversational dialogue ,  coordinate ellipses , , , , –, 

 coordination , , –, , , –, –, –, , , , , –, , , , –, , , , –, –, , –, , , , , –, –, –, , , , , –, , , , –, , , , –, –, , , , , , , –, , , , –, , , , , , –,  coordination ellipsis ,  copula drop –, , ,  copular clause –, , , , ,  copy , , , , , , , –, –, , , , , , , , , , , –, , , , , , , –, , , , , , , , , , –,  copy theory of movement ,  coreference , –, , , , , , –, , , ,  Coreference Rule –,  correlate –, , , , , , , , –, , , , –, , –, , , , , , , , , , , , –, , , , , , –, –, , , , , –, , , , –, , –, –, –, –, , , ,  CR , , –, , –, –, –, –, , , –, , , –; see also backward conjunction reduction; conjunction reduction; forward conjunction reduction cross-modal priming ,  cumulative agreement ,  D-linked –, , , , ,  deaccentuation , , , –, , –, , , , , , –, –, , , ,  default case , , , , ,  definite article , – definite constituent ellipsis – definite correlate ,  definite ellipsis , ,  definite null complement , –,  definiteness , , , , , – degree , , , , , , , , , –, , , , , , , , –, , –, –, –, , , , –, , , , ,  Degree Phrase , , ,  degree quantifier , , , ,  deontic ,  Dependency Grammar , ; see also DG DG , –, , –, –; see also Dependency Grammar DGB , , –, –, ; see also Dialogue Gameboard Dialogue Gameboard , ; see also DGB direct compositionality , , –, 



direct interpretation approach , –, ,  discourse –, , , , , , , , , , , –, , , , –, –, –, , , , , , , , , , , , –, , , , , –, , , , , , , , , , –, –, –, , , –, , –, –, –, –, , , –, , , , , , –, , –, , –, , , , –, , , –, , , –, , –, , , , , , , –, , , , , ,  discourse deletion ,  discourse drop  discourse orientation ,  discourse relations , –, –, ,  disjunction , , –, , , , , , –, ,  Distributed Morphology  do-support , , , , –,  Downward Bounding Constraint  Dynamic Semantics , , ,  E-feature , , , , , , , , , , , ,  e-GIVENness –, –,  Economy , , , , , , ,  ellipsis resolution , , , , , , , , , , , ,  embedded language  embedded question ,  emphatic do  empty noun –, , –, , , ,  entailment , –, –, –, ,  epistemic , , , , , , – EPP ; see also Extended Projection Principle equation , , – Evaluation Metric  exophoric ellipsis  Extended Projection Principle ; see also EPP extraction , –, , , , , , –, , , , –, –, , –, , , –, , , , , , , , , –, , –, , –, , , , –, –, , , , , , –, , –, –, , , , , –, , , – extraposition , –, –, , ,  feature checking  focus FEC –, –; see also Focus Establishing Constituent Focus Establishing Constituent ; see also FEC focus fronting , , – focus intonation – focus movement –, –, –, , , –, , , –, , , 





focus (cont.) focus projection , , , , , ,  focus-verb adjacency ,  forward conjunction reduction , , –, –, –, –, , ; see also backward conjunction reduction; conjunction reduction; CR fragments case-marked fragment , ,  contrastive fragment answers ,  declarative fragments , ,  non-case-marked fragment , ,  full reconstruction , , ,  function composition –, – g rule –, – gapping , , , , , –, , , –, –, , , , , –, –, –, –, , , –, –, –, –, –, , –, –, , , –, –, , , , , , –, , , , –, , –, –, , –, , –, , –, , , , –, , , , , –, , –, –, –, , , –, –, , –, , , –, , –, , , –, , , , –, –, , –, , , , – gender , , , , , , –, , , , –, , –, –, , , –, , , , ,  gender mismatch effect ,  general inference , –, ,  gesture , , –, – givenness , –, , –, –, , , ,  government , , , , , , –, –, , , , , , , ,  grounding , – Head-Complement Rule  Head-Driven Phrase Structure Grammar , –, ; see also HPSG Heavy NP shift –, –, –, –,  Higher Operator Phrase ; see also HOP higher-order matching  higher-order unification , , , , , ,  HOP –; see also Higher Operator Phrase HPSG –, –, , , , , –, ; see also Head-Driven Phrase Structure Grammar identity condition , , , , –, , , , , –, , , , , , , –, –, , , , , , , ,  implicit arguments , ,  implicit correlate ; see also antecedentless; non-linguistic antecedent; pragmatic antecedent; pragmatic control implicit questions , – implicit structure 

in-situ analysis – incrementality , , , , , –, –, , –, –, , , – indefinite ellipsis ,  indefinite null complement – Indirect Licensing , –,  inference , , , , , , , , , , , –, , , , –, , , ,  infinitival to – infinitive , , , –, , –, , , , –, –, , , –, , , , , , –, , ,  inflectional mismatches  innateness , , –,  inquisitive semantics , –, –, , – interaction –, –, –, , –, –, , , –, , , , , –, , , , , , , , , , , , , , , , , , , , , , , , , –, ,  interactive coordination  interrogative , , , , , , , –, , , –, –, , , , , , –, , –, , , , , , , , , , –, , , , , –, –, –, ,  intonation , , , , –, , , , , , , ,  intonational contour ,  intonational phrase  inverse scope , , , –, , ,  invisible syntactic structure –, ; see also unpronounced syntactic structure islands Adjunct Island , –, , , , ,  Complex NP Constraint , , , , , , –, –,  Coordinate Structure Constraint –, , , , , , ; see also CSC CSC , , –, –, , , ; see also Coordinate Structure Constraint island constraint , , , , , , , , , , , , , , , , , , , , , –, , ,  island effect , , , –, , , , –, –, , –, , , , , , , , , , –, , –, –, ,  island evasion –, –,  island repair –, , , , –, , –, , , , , , , ,  island violation , –, –, , , , , , , , , , –, –, , , –, , , , , , , , , , , , , , , , ,  LBE –, ; see also Left-Branch Condition

 Left-Branch Condition –, , –, , ; see also LBE PF islands ,  Subject Condition , , –,  weak islands , ,  isomorphic reconstruction  Kase –,  KoS , , , , , , ,  lambda abstraction , –, , , – Late Merge , –, , –,  Law of Coordination of Likes ; see also LCL LCL –, , ; see also Law of Coordination of Likes learnability , , , , ,  LEE , , , , , , –, ; see also left-edge ellipsis left periphery , , , , , , , , , –, , , , , , , –,  left-edge ellipsis , , ; see also LEE lexical verb complement ellipsis , – LF Pied-Piping  LF-copy , , , –, , , ,  LF-extraction  LF-projection ,  licensing –, –, , –, , –, , –, –, , –, , , , , , , –, –, –, –, –, , , , , , , , –, –, , –, , , –, , –, , , , –, , , , , , –, , , , , , –, , –, , , , , , –, , , , , –, , –, –, , , , , , , , , , –, –, –, ,  lift –, , ,  linearization –, , , ,  Logical Form , , , , , , , , ,  low coordination , –, , –, , – main assertion principle  matrix language  Matrix Language Framework  MCE –, –, , , –, , –; see also Modal Complement Ellipsis MD , –, ; see also multidominance Meaning-Text Theory  Merge , –, –, , –, , –, –, –, , , , , –, , –, , , ,  mime , –, , – Minimalist framework , ,  mismatch –, , , –, , , –, , –, –, –, –, , , –, ,



, –, , , , , , , , , –, ,  mixed reading ,  modal , , , , , , , –, , , , –, –, , , –, , , , , –, , , –, , , , , –, , , –, , , , – Modal Complement Ellipsis –, , , , , , , ; see also MCE modal ellipsis , , ,  move-and-elide , –,  movement-then-deletion , ,  multidominance , , , , –, ; see also MD multiple cleft , – multiple wh-movement , ,  N’-deletion –, – N-deletion , ,  NCA , , , –, –, , , , –, –, –, –, , , –, , ; see also Null Complement Anaphora negation , , , –, , –, –, –, , , , –, , , , , , , , , , , , , , , , , , , , , , , –, , –, – negative polarity item , ,  nominal ellipsis , –, , , , , , , –, –, , , –, , –, , , , –, , ,  nominalization , , – non-actuality implicature  non-constituent coordination , , , ,  non-constituent deletion , ,  non-movement account  non-sentential utterances –, , , ; see also NSUs NP-ellipsis , , , , , , , , –, , –, , , , , –; see also NPE NPE , , , , , –, –, –, , –, –, –, –, –, –, , , , –, –, –, , , –, , –; see also NP-ellipsis NSUs , –, , , , , ; see also non-sentential utterances null argument , , –, , , , , , –, –, , , –,  Null Complement Anaphora , , , , , , , , –, , , –, , , , , , , , ; see also NCA Null Complement Rule – null elements , , , , , , , , ,  null object construction , – null subject , , , 





number , , –, , , , , , , , –, , , , , , –, , , –, , , , –, , , , , , , , , , , , , , , , , , , , , , , , –, , , , , , , , , , –, –, , , , , , , , –, , , , , , , , , , –, , , , , , , , , –, , , , , , , , , ,  object drop , , ,  object shift , –, – Occam’s Razor , , , , , ,  orphan , ,  overt correlate , , , , , –,  Parallel Architecture ,  parallelism , , , –, , –, , , , , –, , , , , , –, –, , , , , , , , –, –, , –, , , –, , , , –, –, –, , –, –, , , , –, , , , , , , , , –, , , , –, , , –,  parameter identification , , – parasitic gap –, –, , ,  parasitic scope – parataxis  parsing , , , –, , –, –, –, , , , , , , ,  partitivity , , –, , , , – passive –, –, , , , , , –, , –, –, –, , –, , , , , , , , , , , , , , , , –, ,  PCC , , –, ; see also Phrase Cluster Coordination perfect –, , –, , –, , , , , –, , , , , , , – PF-deletion , , , , , , , , –, –, –, ,  phrasal ellipsis  Phrase Cluster Coordination , ; see also PCC Pied-Piping , , , , ,  pitch accent –, , –, –, –, –, –, –, , ,  pitch extraction contour , , , ,  pointer , , –, , , – polarity , –, , , , , , , , , , , , , , –, –, –, , , , –, ,  polarity ellipsis , –, , , –, ,  possessive , –, –, –, , , , , –, , –, , , , –, –, , ,  possessor , , , , , –, , ,  possessor doubling , 

poverty of the stimulus  pragmatic control ; see also antecedentless; implicit correlate; non-linguistic antecedent; pragmatic antecedent pragmatics , , , , , , –, , , , ,  predicate ellipsis , –, , –, , , , , –, , , , , , , , –, , , , , , , , , , ,  preposition omission , , –, , , – preposition stranding , , , –, –, –, , , , , –, , , , , , , , , –, , , –, –, , , , , , , , –, , , , –, –, –, –,  Preposition Stranding Generalization , , , , , , , , , , , ; see also PSG presentational focus – presupposition inheritance , ,  preverb , , –,  preverb stranding , , –,  preverbal particle , , ,  Primary Linguistic Data  priming , , , , –, –, – Principle C –, , , , ; see also Condition C Principles and Parameters  pro , , , –, , –, , , , , –, , , , –, –, , –, , –, , , , –, , , , –, , –, , , –, , , –, –, , , , , , , –, –, –, , , , ,  pro-form , , –, , , , , , , , –, –, , , , –, , , –, , –, –, –, , , –, –, , , , , , ,  processing , , , , , , , , , , , , , –, , –, –, , –, –, –, –, , , , , , –, , , , , , –, , –, –, , , , –, –,  progressive , , , –, , , , , , –,  Projection Principle  pronominal possessor  pronoun , , , –, , , –, , , –, , , –, –, , , , , –, , , , , , , –, , , –, , , , , –, –, , , –, –, –, –, –, –, , , , –, –, , , –, –, , , , , , –, , , , , , , , , , , , , , –, –, , –, , , , , , , , –, –, , , , , , –, , , , , , , , –, –, –, , ,

 –, –, , , , , –, , , , , –, , ,  pronoun resolution ,  Pronoun Rule – proposition , , , –, –, , , , , , , –, , , , , , , , –, , , , , , , , –, , , –, –, , –, , , , , –, , ,  prosodic boundaries –, ,  prosodic contour  prosodic licensing , , – prosodic phrasing –, , , – prosody , , –, –, –, , –, –, –, , , , , , , , , ,  pseudocleft , , , , , ,  pseudogapping –, , , , , , , , , –, –, , –, , , –, , –, , , , –, , –, –, , , , , –, , , , –, , –, , , , , , , , , , , , –, –, , –, , –, , , , –, , ,  PSG , –, , , , –, , –, , ; see also Preposition Stranding Generalization psycholinguistics , , , –, , –, , , , , , –, , –, , , ,  punctuated paths , ,  QR , , –, , , –, –, , , , –; see also Quantifier Raising Qu-Ans –, –, ,  quantifier , , , , , , , –, , , , , , , , –, , , , , , , , , , , –, , –, , , , , ,  Quantifier Raising , , , –, , , , , , ; see also QR QUD , , , , –, , , –, –, , , , –, , –, –, –, –, –, ; see also Question Under Discussion Question Under Discussion , , , , , , , –, , , –, , ; see also QUD reaction time ,  record , , –, , , –, –, , ,  record types , – recoverability –, –, , –, –, , , –, , , –, , , –, , –, , , , , , , , , , , , , , , –, –, , , , , –, , , , , – Recycling Hypothesis  Reduction Analysis ,  relational adjective 



relative clause , , , , , , –, , , , –, , , , , , , , , , , , , , –, , ,  repair , , –, , , , , –, , , –, , , , , –, , , –, –, , , , –, , –, , , –, –, , , , , , –, , , , , –, , , ,  replacives –,  reprise –, , –, , , –, ,  restructuring ,  resumptive pronoun  Right-Node Raising non-coordinate RNR , , – RER –, , , ; see also Right-Edge Condition; Right-Edge Restriction Right-Edge Condition ; see also RER; Right-Edge Restriction Right-Edge Restriction ; see also RER; Right-Edge Condition Right-Node Raising , , , , , , , , –, , , , , , , , , –, , –, , , –, –, , , , , –; see also non-coordinate RNR; RNR Right-Node Wrap – RNR , –, , , , –, , –, –, –, –, –, –, –, –, , –, , , –, –, ; see also Right-Node Raising root , , , –, , , , , , , , , , , , –, , , –, , –, –, , , –, –, , –, , , , , , – Saxon genitive –,  scope , , , , , , , –, –, , , , –, , , , –, , –, , , , , , , , –, –, , , –, , –, , –, –, –, –, , , –, , , , , –, , , , –, , , , –, –, – scrambling , , , –, , , , , , ,  script , ,  self-paced reading , , , ,  semantic identity –, , , , –, , , , –, –, , , , , , , , , , , ,  sentence processing , , , –, ,  short answer , , –, , , –, , –, , , –, –,  short do replies – sign language –, , , , –, – Simpler Syntax Hypothesis ; see also SSH





sloppy identity –, , , , , , , , –, , –, , –, , , , –, , , , –, , , –, , , , , , –, –, –, , –, –, , , ; see also sloppy-like reading; strict identity; strict-sloppy identity sloppy-like reading , –, –; see also sloppy identity; strict identity; strict-sloppy identity Slot Grammar  sluicing backward sluicing –,  bare sluice –, , – case-marked sluicing –,  contrast sluicing , , –, , –,  embedded sluicing ,  isomorphic sluicing , , –,  matrix sluicing ,  multiple coordinated sluicing ,  multiple sluicing , –, , , –, , –, –, , –, , , , , –, , , ,  non-isomorphic sluicing , , –,  pronominal sluicing  pseudosluicing , , ,  spading , , , , –,  sprouting , , –, , , , –, –, –, , , , –, , , –, , , , , , , , –,  swiping , , , –, –, , ,  split questions ,  spontaneous production  SSH –, , ; see also Simpler Syntax Hypothesis stress , , , , , , , –, , , , , , , , , , , , , ,  strict identity , –, , , –, –, –, , , , , , –, , , –, , , , –, , ; see also sloppy identity; sloppy-like reading; strict-sloppy identity Strict Identity Hypothesis , ,  strict interfaces –, , –, –, ,  strict-sloppy identity , –, , , , ; see also sloppy identity; sloppy-like reading; strict identity string deletion –,  string-vacuous movement  stripping BAE , , , –, , –, ; see also Bare Argument Ellipsis Bare Argument Ellipsis , , , , , , , , , , , , ; see also BAE case-marked stripping –, –, ,  multiple stripping  non-case-marked stripping –, ,  subjacency , –, –, –, –, – subject drop , 

subordination –, , , , ,  subsentential , , , , ,  subtract –, , ,  successive cyclicity  summative agreement ,  Superiority , , , – symmetric entailment –, – syntactic complexity – syntactic identity , –, , –, –, , , , , , , , , , , –, , , –, –, , , ,  syntactic matching , , –,  syntactic mismatch –, –, –, , , ,  tag questions , ,  TDH –, , –; see also Trace-Deletion Hypothesis Theta Criterion  topic –, , , , , , , , , , , , , , , , , –, , , , , , , , –, , , –, , , , , , , , , , , , , , , –, , –, , –, , , , , , , – topicalization –, , –, , , , , , –, , –, , , , , , –, – TP ellipsis , , , –, –, , –, , –,  Trace-Deletion Hypothesis ; see also TDH transfer ,  trigger , , , –, , , , , , , , , , , , , , –,  TTR , –, , , ; see also Type Theory with Records Type Theory with Records , , , ; see also TTR UCC –, ; see also Unlike Category Coordination unification , , –, , , , –, –, ,  Unlike Category Coordination ; see also Law of Coordination of Likes; UCC unpronounced syntactic structure , , , , , –, , , , –, , , ; see also invisible syntactic structure Upward Bounding Constraint  variable , , , , , , –, –, , –, , , , , , –, , , –, –, –, –, –, , –, , –, , , , –, , –, –, , , –, , , , –, –, , , –,  variable-free semantics , – vehicle change , , , , , , , , , –, , , –, , , 

 verbal phrase ellipsis – voice –, –, , , –, , , –, , , , , , , , , , , , –, –, , , –, , , , , , , ,  voice alternations  voice mismatch –, –, , , –, , , , , , , –, –, –, , ,  VP-ellipsis auxiliary-stranding ellipsis – Derived Verb Phrase Rule  verb-stranding VP-ellipsis , , –, , , , , –, ; see also VVPE Verbal Identity Requirement ; see also VIR VIR –; see also Verbal Identity Requirement VP Deletion , , – VVPE –, , –, –; see also verbstranding VP-ellipsis wh-fronting –, –, –, –, , –, – wh-in-situ , , , 



wh-movement , , , –, –, , , , –, –, , , –, –, , –, , , , , , –, –, , , , , –, ,  wh-phrase , , , , , , , , –, , –, –, –, –, , , , –, , , , –, –, , , , , –, –, , –, , , , , , –, , , , , , , –, , –, –, , , , , , –, , , –, –, , , – wh-question , , , , , , , , –, , , , –, –, –, , –, , , , , , , , , , , , , , , , – wh-remnant –, , , , , , , , , , –, , –,  WM –, –; see also working memory Word Grammar  working memory , –; see also WM

O X F O R D H A N D B O O K S IN LI N G U I S T I C S THE OXFORD HANDBOOK OF AFRICAN AMERICAN LANGUAGE Edited by Sonja Lanehart

THE OXFORD HANDBOOK OF APPLIED LINGUISTICS Second edition Edited by Robert B. Kaplan

THE OXFORD HANDBOOK OF ARABIC LINGUISTICS Edited by Jonathan Owens

THE OXFORD HANDBOOK OF CASE Edited by Andrej Malchukov and Andrew Spencer

THE OXFORD HANDBOOK OF CHINESE LINGUISTICS Edited by William S.-Y. Wang and Chaofen Sun

THE OXFORD HANDBOOK OF COGNITIVE LINGUISTICS Edited by Dirk Geeraerts and Hubert Cuyckens

THE OXFORD HANDBOOK OF COMPARATIVE SYNTAX Edited by Gugliemo Cinque and Richard S. Kayne

THE OXFORD HANDBOOK OF COMPOSITIONALITY Edited by Markus Werning, Wolfram Hinzen, and Edouard Machery

THE OXFORD HANDBOOK OF COMPOUNDING Edited by Rochelle Lieber and Pavol Štekauer

THE OXFORD HANDBOOK OF COMPUTATIONAL LINGUISTICS Edited by Ruslan Mitkov

THE OXFORD HANDBOOK OF CONSTRUCTION GRAMMAR Edited by Thomas Hoffman and Graeme Trousdale

THE OXFORD HANDBOOK OF CORPUS PHONOLOGY Edited by Jacques Durand, Ulrike Gut, and Gjert Kristoffersen

THE OXFORD HANDBOOK OF DERIVATIONAL MORPHOLOGY Rochelle Lieber and Pavol Štekauer

THE OXFORD HANDBOOK OF DEVELOPMENTAL LINGUISTICS Edited by Jeffrey Lidz, William Snyder, and Joe Pater

THE OXFORD HANDBOOK OF ELLIPSIS Edited by Jeroen van Craenenbroeck and Tanja Temmerman

THE OXFORD HANDBOOK OF ERGATIVITY Edited by Jessica Coon, Diane Massam, and Lisa deMena Travis

THE OXFORD HANDBOOK OF EVIDENTIALITY Edited by Alexandra Y. Aikhenvald

THE OXFORD HANDBOOK OF GRAMMATICALIZATION Edited by Heiko Narrog and Bernd Heine

THE OXFORD HANDBOOK OF HISTORICAL PHONOLOGY Edited by Patrick Honeybone and Joseph Salmons

THE OXFORD HANDBOOK OF THE HISTORY OF ENGLISH Edited by Terttu Nevalainen and Elizabeth Closs Traugott

THE OXFORD HANDBOOK OF THE HISTORY OF LINGUISTICS Edited by Keith Allan

THE OXFORD HANDBOOK OF INFLECTION Edited by Matthew Baerman

THE OXFORD HANDBOOK OF INFORMATION STRUCTURE Edited by Caroline Féry and Shinichiro Ishihara

THE OXFORD HANDBOOK OF JAPANESE LINGUISTICS Edited by Shigeru Miyagawa and Mamoru Saito

THE OXFORD HANDBOOK OF LABORATORY PHONOLOGY Edited by Abigail C. Cohn, Cécile Fougeron, and Marie Hoffman

THE OXFORD HANDBOOK OF LANGUAGE AND LAW Edited by Peter Tiersma and Lawrence M. Solan

THE OXFORD HANDBOOK OF LANGUAGE EVOLUTION Edited by Maggie Tallerman and Kathleen Gibson

THE OXFORD HANDBOOK OF LEXICOGRAPHY Edited by Philip Durkin

THE OXFORD HANDBOOK OF LINGUISTIC ANALYSIS Second edition Edited by Bernd Heine and Heiko Narrog

THE OXFORD HANDBOOK OF LINGUISTIC FIELDWORK Edited by Nicholas Thieberger

THE OXFORD HANDBOOK OF LINGUISTIC INTERFACES Edited by Gillian Ramchand and Charles Reiss

THE OXFORD HANDBOOK OF LINGUISTIC MINIMALISM Edited by Cedric Boeckx

THE OXFORD HANDBOOK OF LINGUISTIC TYPOLOGY Edited by Jae Jung Song

THE OXFORD HANDBOOK OF LYING Edited by Jörg Meibauer

THE OXFORD HANDBOOK OF MODALITY AND MOOD Edited by Jan Nuyts and Johan van der Auwera

THE OXFORD HANDBOOK OF MORPHOLOGICAL THEORY Edited by Jenny Audring and Francesca Masini

THE OXFORD HANDBOOK OF NAMES AND NAMING Edited by Carole Hough

THE OXFORD HANDBOOK OF PERSIAN LINGUISTICS Edited by Anousha Sedighi and Pouneh Shabani-Jadidi

THE OXFORD HANDBOOK OF POLYSYNTHESIS Edited by Michael Fortescue, Marianne Mithun, and Nicholas Evans

THE OXFORD HANDBOOK OF PRAGMATICS Edited by Yan Huang

THE OXFORD HANDBOOK OF SOCIOLINGUISTICS Second edition Edited by Robert Bayley, Richard Cameron, and Ceil Lucas

THE OXFORD HANDBOOK OF TABOO WORDS AND LANGUAGE Edited by Keith Allan

THE OXFORD HANDBOOK OF TENSE AND ASPECT Edited by Robert I. Binnick

THE OXFORD HANDBOOK OF THE WORD Edited by John R. Taylor

THE OXFORD HANDBOOK OF TRANSLATION STUDIES Edited by Kirsten Malmkjaer and Kevin Windle

THE OXFORD HANDBOOK OF UNIVERSAL GRAMMAR Edited by Ian Roberts