Symmetric Coordination: An Alternative Theory of Phrase Structure 9783110934687, 9783484303324

Table of contents :
Acknowledgements
Introduction
I Transformational Approaches to Coordination
1 The Grammatical Framework
1.1 X̄-Theory
1.2 Relation to Non-Transformational Approaches
2 Coordination Types
2.1 Asymmetric Coordination
2.2 Symmetric Coordination
2.3 Relationship between the Types
2.4 Beyond Symmetric Coordination
3 Past and Present Analyses
3.1 Basic proposals
3.2 Further Development
3.3 CSC & ATB
3.4 Redundancy and Recoverability
3.5 Coordination within GB
3.6 Summary
4 Neither Deletion Nor Reconstruction
4.1 PhC and Deletion
4.2 RNR and Deletion
II Shared Constituent Coordination
5 German Conjunctions
5.1 Classification of Conjunctions
5.2 Phrasal Coordination
5.3 Right Node Raising
5.4 Gapping
6 The Domain of Coordination
6.1 Right SCC
6.2 Left SCC
6.3 The role of basic phrase structure
6.4 Medial SCC
6.5 Contrast Sets and Conjunct Structure
6.6 Summary
III An Alternative Theory of Phrase Structure
7 The Generalized X¯-scheme
7.1 The Abstract Coordination Lexeme
7.2 From D-structure to S-structure
7.3 Obviating CSC and ATB
7.4 Parallelism of Syntactic Structures
8 Linearization
8.1 Linearization without Verb Factors
8.2 Adjacency
8.3 Linearization with Verb Factors
8.4 Interaction of Linearization Rules
8.5 Coordinate Structures as Factors
9 Modes of Interpretation
9.1 Non-SCC
9.2 SCC
9.3 Relation to Plurals
10 Linearization with Adjuncts
10.1 Factor Adjunction
10.2 Adjunction to Shared Finite Verb
10.3 Coordination as Constituency-Test
10.4 Initial versus Non-Initial Conjunctions
11 Closed versus Open Type Categories
11.1 Strictly Isomorphic Structures
11.2 Apparently Non-Isomorphic Structures
11.3 Closed Type Categories in Other Frameworks
12 Related Phenomena
12.1 Comparatives
12.2 Coordination and Word-Syntax
12.3 Genuine or Pseudo-Asymmetries?
13 Summary
Bibliography

Citation preview

Linguistische Arbeiten

332

Herausgegeben von Hans Altmann, Peter Blumenthal, Herbert E. Brekle, Gerhard Heibig, Hans Jürgen Heringer, Heinz Vater und Richard Wiese

Birgit

Wesche

Symmetric Coordination An Alternative Theory of Phrase Structure

Max Niemeyer Verlag Tübingen 1995

Die Deutsche Bibliothek - CIP-Einheitsaufnahme Wesche, Birgit: Symmetric Coordination : an alternative theory of phrase structure / Birgit Wesche. - Tübingen : Niemeyer, 1995 (Linguistische Arbeiten ; 332) NE: GT ISBN 3-484-30332-8

ISSN 0344-6727

© Max Niemeyer Verlag GmbH & Co. KG, Tübingen 1995 Das Werk einschließlich aller seiner Teile ist urheberrechtlich geschützt. Jede Verwertung außerhalb der engen Grenzen des Urheberrechtsgesetzes ist ohne Zustimmung des Verlages unzulässig und strafbar. Das gilt insbesondere für Vervielfältigungen, Übersetzungen, Mikroverfilmungen und die Einspeicherung und Verarbeitung in elektronischen Systemen. Printed in Germany. Druck: Weihert-Druck GmbH, Darmstadt Einband: Hugo Nädele, Nehren

Contents

Acknowledgements

ix

Introduction

1

I

3

1

Transformational Approaches to Coordination The Grammatical Framework

5

1.1

6

X-Theory 1.1.1

1.2 2

3

German Phrase Structure

Relation to Non-Transformational Approaches

Coordination Types

8 11 13

2.1

Asymmetric Coordination

13

2.2

Symmetric Coordination

14

2.3

Relationship between the Types

17

2.4

Beyond Symmetric Coordination

19

2.4.1

VP-Deletion

19

2.4.2

Sluicing

20

2.4.3

N-Gapping

20

Past and Present Analyses

23

3.1

3.2

Basic proposals

23

3.1.1

Chomsky

23

3.1.2

Ross

25

Further Development

29

3.2.1

RNR

29

3.2.2

Gapping

32

3.2.3

PhC

33

3.2.4

Gapping versus R N R

36

3.3

CSC & ATB

38

3.4

Redundancy and Recoverability

40

Coordination within G B

41

3.5

3.6

3.5.1

Gapping in G B

43

3.5.2

R N R in G B ?

46

Summary

48

vi 4

Neither Deletion Nor Reconstruction 4.1 PhC and Deletion 4.2 RNR and Deletion

II

Shared Constituent Coordination

49 50 52

57

5

German Conjunctions 5.1 Classification of Conjunctions 5.2 Phrasal Coordination 5.2.1 Non-initial Conjunctions 5.2.2 Initial Conjunctions 5.2.3 Ungrammatical initial PhC of CP 5.3 Right Node Raising 5.3.1 Non-Initial Conjunctions 5.3.2 Initial Conjunctions 5.4 Gapping 5.4.1 Non-Initial Conjunctions 5.4.2 Initial Conjunctions

63 63 64 64 66 68 69 70 70 71 71 72

6

The Domain of Coordination 6.1 Right SCC 6.1.1 Against lexical PhC 6.2 Left SCC 6.3 The role of basic phrase structure 6.3.1 Left and Right SCC: universal rules 6.4 Medial SCC 6.5 Contrast Sets and Conjunct Structure 6.5.1 Admissable interruptions of contrast sets with Medial SCC 6.6 Summary

75 76 77 81 82 83 86 88 90 96

III

An Alternative Theory of Phrase Structure

7

The 7.1 7.2 7.3 7.4

Generalized X-scheme The Abstract Coordination Lexeme From D-structure to S-structure Obviating CSC and ATB Parallelism of Syntactic Structures

8

Linearization 8.1 Linearization without Verb Factors 8.2 Adjacency

99 103 105 106 115 116 119 119 122

vii

8.3 8.4 8.5 9

Linearization with Verb Factors Interaction of Linearization Rules Coordinate Structures as Factors

129 137 143

M o d e s of Interpretation 9.1 Non-SCC 9.2 SCC 9.2.1 Medial SCC 9.2.2 Peripheral SCC 9.2.3 Binding and Control with SCC 9.3 Relation to Plurals

145 147 148 148 149 151 153

10 Linearization w i t h A d j u n c t s 10.1 Factor Adjunction 10.1.1 Left Adjunction 10.1.2 Left versus Right or Medial Adjunction 10.1.3 Adjunction in Functional versus Lexical Projections 10.2 Adjunction to Shared Finite Verb 10.3 Coordination as Constituency-Test 10.4 Initial versus Non-Initial Conjunctions

155 155 160 162 165 168 169 171

11 Closed versus O p e n T y p e Categories 11.1 Strictly Isomorphic Structures 11.2 Apparently Non-Isomorphic Structures 11.2.1 Active and Passive 11.2.2 Raising and Active 11.3 Closed Type Categories in Other Frameworks

173 173 182 182 185 186

12 R e l a t e d P h e n o m e n a 12.1 Comparatives 12.2 Coordination and Word-Syntax 12.2.1 Head-drivenness and XP-domain? 12.3 Genuine or Pseudo-Asymmetries? 12.3.1 Default versus non-default linearization 12.3.2 Parenthetical Constructions

191 191 193 195 198 198 203

12.3.3 Movement of Single List Items?

204

13 Summary

209

Bibliography

213

Acknowledgements

I do not have to assume, because I know for certain that there exists in my own possible world a set of people who could be coordinated as follows: NP[COOr ( Jürgen Allgayer; Peter Bosch; Heike Doebele; Bart Geurts; Christa Hauenschild; Otthein Herzog; Doris Janhsen; Jim Kilbury; Tibor Kiss; Esther König; Gerd Kortüm; Gabi Kreutzner; Duri Mayer; Martin Mezger; Claudia Mrotzek; Petra Naerger; Ingo Raasch; Ingrid Renz; Claus Rainer Rollinger; Marc-Thomas Schmidt; Gabi Warkus; Andrea Weß; Astrid, Holger, Vera Wesche; Dirk Wolken; Dieter Wunderlich )

As can easily be seen the order of this list is determined by the alphabet. The necessary relation of contrast between the list members is given by the different type of support provided: • judging over and over again the most natural coordinations • being enthusiastic about discussing every single epoch-making idea • always jumping to handle most urgent literature requests • letting me share reliable, and thus ever so valuable, Tex macros • happily adjusting Tex style files to any level of idiosyncracy • sharing the fun of formatting • synchronizing my native speaker mixture of British and American English • translating passages from English to German and vice versa, when I could not tell right from wrong • being the world's best managing editor of IKBS reports • without 'ifs' and 'buts' taking the trouble of reviewing parts or even all of the following pages • and, foremost, constantly providing all the encouragement so vitally needed Thank you very much!

Introduction

Symmetric coordination is assumed to comprise the three universal coordination types of PHRASAL COORDINATION, RIGHT NODE RAISING, a n d GAPPING. T h i s p h e n o m e n o l o g -

ical view is agreed upon by transformational as well as non-transformational grammarians. Previous analyses of the phenomenon of symmetric coordination have generally been determined by the question of constituency. Within the transformational framework a coordinate structure is assumed to be based on two (or more) full-fledged constituents of the same t y p e from which certain items may be deleted. Coordination of overt full-fledged constituents is referred to as PHRASAL COORDINATION ( P H C ) . If (at least) t h e finite verb is deleted in non-first conjuncts we speak of GAPPING structures. If rightmost items are deleted in non-final conjuncts we speak of RIGHT NODE RAISING ( R N R ) structures. P H C comprises t h e various types of CONSTITUENT COORDINATION, while R N R and GAPPING are attributed to the class of NON-CONSTITUENT COORDINATION, since after deletion has applied their conjuncts normally do not have genuine constituent status anymore. The transformational approach to symmetric coordination which we will present here does not take recourse to any type of deletion. Especially concerning R N R constructions numerous counterexamples have been raised against a deletion t r e a t m e n t , in t h e course of which a preceding raising operation is put into question, too. Consequently, we also deny procedural interdependencies between the three coordination types, according to which t h e non-constituent coordination types R N R and GAPPING are derived by deletion from the base generated constituent coordination type P H C . Instead we propose a direct phrase structure account which is a natural extension to the classical X-scheme, and which uniformly covers not only the syntactic structures of constituent and non-constituent coordination but also conventional simplex structures. Thus, our proposal contributes to the grammar of phrasal structures in general. This dissertation is structured as follows. Part one contains a detailed characterization of the three symmetric coordination types PHC, RNR, and GAPPING. We present an overview of previous analyses which illustrates how the three types evolved. Finally we summarize t h e counterarguments against deletion type approaches to non-constituent coordination. In part two we motivate our own view on the relevant phenomena. We shift t h e focus of attention from deleted items to contrasting items in a coordinate structure. Constituents hitherto assumed to be subject to deletion are viewed as being structurally shared by these items of contrast. This leads to a classification of symmetric coordination into NONSHARED CONSTITUENT COORDINATION ( N O N - S C C ) on t h e o n e h a n d v e r s u s SHARED

CONSTITUENT COORDINATION ( S C C ) on the other, the latter being subdivided into LEFT S C C , RIGHT S C C , a n d MEDIAL S C C .

2 In part three we present an alternative theory of phrase structure which provides a uniform account of coordinate as well as simplex structures. Its effects on the various levels of syntactic representation are discussed for each level separately. After a general presentation we refine this theory, addressing among other things adjunction type structures, before we conclude with a discussion of related phenomena such as word-coordination, comparatives, and asymmetric coordination. However, except for specific coordinations hitherto assumed to be asymmetric in nature, neither of the three lends itself to a treatment in terms of contrast and sharing of the kind we propose for symmetric phrasal coordination.

Part I

Transformational Approaches to Coordination

Chapter 1 The Grammatical Framework

The approach to symmetric coordination we will present is couched in the transformational framework of Government & Binding (GB) as it is originally presented in [Chomsky,1981], and further elaborated in [Chomsky,1986a] and [Chomsky,1986b], Within this model of grammar we distinguish between four distinct levels of representation for a given sentence. • D-structure • S-structure • Phonetic Form (PF) • Logical Form (LF) Mapping functions specify the relation between these levels of representation. Phrase structure (PS) rules and lexical items determine the D-structure of a sentence. The component of transformations—within the GB framework restricted to the single transformation move a—relates D-structure to S-structure. Interpretative rules (rules of construal) link S-structure and the level of LF, which forms the basis for the sentence's semantic interpretation. Finally, phonological rules map S-structure onto PF, which yields the actual phonetic surface structure of a sentence. Thus the model of grammar is organized as follows:

PS-rules lexicon D-structure move a S-structure phonological rules PF

LF

6 The rule systems are constrained by several subtheories: 1. X-theory 2. θ-theory 3. Case theory 4. Government theory 5. Bounding theory 6. Binding theory 7. Control theory The effects of X-theory will be central to our investigation. Arguments resting on the remaining theories will be discussed where appropriate. The main tenet of our approach to symmetric coordinate structures will lie in an extension of the data structures admitted as terminal nodes—a step which eventually leads to an alternative theory of phrase structure. We claim that coordinate structures are strictly endocentric. This generalization is accounted for within a "three-dimensional" tree structure representation, generated by our phrase structure rules. Moreover, the extended X-scheme not only applies to coordinate structures, but is equally valid for conventional simplex structures. With this generalized representation our alternative theory of phrase structure directly contributes to the general theory of sentence grammar. Since we will mainly examine coordination phenomena from the perspective of German constructions we will briefly sketch the subtheory most relevant to our purposes, viz. X-theory, giving an overview of German sentence structure as assumed within the GBframe work.

1.1

X-Theory

The central idea of X-theory is that phrases are projections of lexical categories, which comprise nouns (N), verbs (V), prepositions (P), and adjectives (A), and that theses phrases inherit the lexical category's type, the latter being referred to as the X-theory's principle of endocentricity. Hence we get the corresponding phrases NP, VP, PP, and AP. A lexical category X is the head of an XP phrase, and XP in turn is the maximal projection of X. The head category X is also termed the minimal projection of X. The single levels of projection are marked by a respective number of "bars", hence X-bar theory. In general not more than two bar levels are assumed to be an appropriate characterization of the maximal projection XP. The X-scheme determines the recursive structuring of an arbitrary phrase:

7 X-Scheme Χ —> X Complement Χ —• Specifier Χ

Both Complement and Specifier are maximal projections, with zero or more occurrences in a given phrase. The ordering of the head with respect to the Complement(s) or Specifier varies according to the chosen language and the type of phrase. The figure in (1) sketches a basic phrasal structure: (1)

YP (Specifier) X (head)

ZP (complement)

In [Chomsky, 1986b] the X-scheme is extended to the non-lexical categories mentizer (C) and INFL (I). Thus the PS rule (2) S

Comp

Comple-

S

is rendered obsolete. The clausal category conventionally labeled S is replaced by the category IP, and accordingly S by CP. By the extended X-scheme as presented in [Chomsky, 1986b] the basic structure of an English sentence involving a transitive verb looks as follows: 1 (3)

IP Spec NP,ub]

I' ^

^

\ VP

I

I

V'

/\

NP

V

The present investigation will be based on this extended X-scheme, into which our account of symmetric coordinate structures will be naturally embeddable. 1 The subject NP is base-generated here at the Spec-IP node. This move calls for an extension of the definition of "direct 0-marking" in order for the subject to receive its appropriate 0-role from the main verb. Case is assigned to the subject by I(NFL). Furthermore, since the choice of X is only forced when there is a specifier, V may be omitted.

8 1.1.1

German Phrase Structure

In German we distinguish between three different sentence types, according to the position of the finite verb: 2 • verb-final (V-F) e.g.: subordinate clauses ..., daß Bart Spaghetti gekocht hat. (... that Bart spaghetti cooked has.) • verb-initial (V-l) e.g.: yes/no questions Hat Bart Spaghetti gekocht? (Has Bart spaghetti cooked?) • verb-second (V-2) e.g.: declarative sentences, wh-questions Bart hat Spaghetti gekocht, um zu . . . . (Bart hits spaghetti cooked, in order to . . . . ) A theoretical framework which offers a suitable model for describing the word order phenomena found in German, as in most Germanic languages, is the positional field framework (PFF). It divides the sentence into three topological regions, called the 'initial', 'middle' and 'final' fields (IF, MF, and FF, respectively). The MF is framed by the verbal bracket, the latter being so called because it predominantly contains elements of the verb cluster. Within the P F F , then, the above examples are assigned the following structure: (4) IF

Bart

VBieft ..., daß Hat hat

MF Bart Spaghetti Bart Spaghetti Spaghetti

VBright gekocht hat. gekocht? gekocht,

FF

um zu

The final field is mainly reserved for extraposed phrases, such as infinitival complements, or subordinate clauses. Transformational approaches to the description of German verb order have been attempted since the early sixties (see, e.g., [Bierwisch,1963]). Most of the more recent approaches to the varieties of German verb order go back to Thiersch and den Besten

2 The examples should not imply that there is a one-to-one correlation between V-l, V-2, and V-F structures and respective sentence modes such as declaratives, questions, or subordinate clauses. Thus, declaratives also display V-l structure, and subordinate clauses can have V-l or V-2 structure.

9 ([Thiersch,1978]; [den Besten,1983]). As for most Germanic languages it is generally assumed for German that the basic position of the finite verb is clause-final. In these transformational proposals, V-l and V-2 variants are derived by rule from V-F sentences. As can be seen in the matrix above, complementizers and finite verbal elements are distributed complementarily over the left verbal bracket. Thiersch introduces a pair of fronting rules, the first of which moves the finite verb out of the final into the initial (left VB) position, provided the latter does not contain a complementizer. Given the constructions that result from applying this rule, the second rule makes possible the topicalization of an arbitrary phrasal constituent, i.e. moving it out of the M F into the IF before the fronted finite verb. Fillers for this preverbal position may be any obligatory or optional argument to the main verb, including complex non-finite verbal clusters. With these two rules the interdependence of the above three sentences becomes apparent: (5)

a.

(..., daß) Bart Spaghetti gekocht hat —•

b.

Hatj Bart Spaghetti gekocht t¡ —•

c.

Bartj hat; tj Spaghetti gekocht t¡

The regularities as stated within the P F F , together with Thiersch's rules, can be naturally incorporated into the GB framework. Here the initial field is assumed to correspond to the Spec-CP node, and the left verbal bracket to the head of CP (i.e. the C node). In accordance with the constraint of head-to-head movement the verb is moved (via the head node of IP, 3 where the verb receives its TENSE and AGReement features) to the head node of CP. While this rule is an adaptation of Thiersch's first rule, his second rule is restated in GB terms as movement of an arbitrary phrasal constituent into the Spec-CP node, which by definition is reserved precisely for phrasal nodes. The effect of these rules is illustrated in (6) through (8). 4

3 In contrast to English, the I-head is right-branching in German, reflecting the V-F base structure. 4 Instead of base-generating the NP subject at the Spec-IP node, here, and in the following, we alternatively assume, in accordance with, e.g., Fukui & Speas [Fukui and Speas,1986], that the NP subject is base-generated adjoined to the VP-node so that it may be assigned its 0-role by the main verb. Irrespective of whether we deal with a V-l, V-2, or V-F structure it is obligatorily moved to the Spec-IP node where it may receive (nominative) Case from I(NFL). If given, it may in turn be moved to the Spec-CP node from here.

10

(?)

CP

NP

VP NP Spaghetti

V gekocht t,

11 (8)

1.2

Relation to Non-Transformational Approaches

The proposed treatment of symmetric coordination is not necessarily bound to the GB framework. The main tenets of the approach to be presented here could also be captured in any of the non-transformational grammatical approaches such as Generalized Phrase Structure Grammar (GPSG) (cf. [Gazdar et al., 1985]), Lexical Functional Grammar (LFG) (cf. [Kaplan and Bresnan,1982]), or Head-Driven Phrase Structure Grammar (HPSG) (cf. [Pollard and Sag,1987]). The reason for this lies in the fact that our approach in essence effects the Phrase Structure component, extending the permitted range of terminal nodes to include—besides the hitherto exclusively atomic values—also higher order concepts. Apart from the transformation move a no further transformation, be it Deletion or any other transformation, is applied. Every grammatical framework relies on a PS component, each after its own fashion. And, the phenomena captured within the GB framework by employing move a are equally well accounted for within the nontransformational approaches, again each after its own fashion. So, the necessary provision is made for an adaptation of the proposed approach to non-transformational frameworks. The reason for embedding our analysis of symmetric coordination into the GB framework is due to the modularity this multi-stratal system offers. The concepts which GB employs permit to make the process of generating a coordinate structure transparent in a most elegant way. The distinction between the various levels of representation enables us to clearly elaborate the laws controlling symmetric coordination and to explain in a satisfying way the modularity of coordination processes in relation to the single levels of syntactic representation. Most significantly, the distinction between the two levels of D-structure and S-structure on the one hand, and the level of Phonetic Form on the other, will provide for an elegant illumination of the specific requirements to be obeyed within

12 a coordinate structure. As we shall see we would be hard put if we wanted to explain certain coordinate structures, which to a large extent underly phonetic constraints, and no explicit division into syntactic structure on the one hand, and phonetic surface structure on the other were available.

Chapter 2 Coordination Types

The construction of coordination is divided into the two global domains of symmetric coordination on the one hand, and asymmetric coordination on the other. The former is distinguished by conjuncts of the same syntactic type, while the latter displays conjuncts with different syntactic status.

2.1

A s y m m e t r i c Coordination

Sentences (9) and (10) contain asymmetric coordinations ([Wunderlich, 1988b] and [Höhle,1983], respectively): (9)

CP & C'(V-l): [,cp In den Wald ging der Jäger] und [ c jagte einen Helsen], ([Into the forest went the hunter] and [hunted a hare].)

(10)

C'(V-F) & CP: [c'(v-F) Wenn jeder sich zusammenreißt] und [cp alle reichen ihr Abstract rechtzeitig ein ...] ('[If everybody themselves together-pulled], and [all handed their abstracts on time in...]'; If everybody tried hard, and all (of us) handed in their abstracts on time...)

In sentence (9) we observe a coordination of a verb-second CP-structure with a verb-first C'-structure. The first conjunct's subject der Jäger is also the understood subject of the second conjunct. The asymmetric coordinate structure in sentence (9) stems from the fact that the adverbial phrase in den Wald has been preposed to the Spec-CP position instead of the subject der Jäger. If the latter had been preposed to the Spec-CP position within the first conjunct a symmetric coordination of C'-nodes would have resulted: (11)

C & C': Der Jäger [c> ging in den Wald] und [c< jagte einen Hasen], (The hunter [went into the forest] and [hunted a hare].)

In sentence (10), on the other hand, stylistic operations seem to be at work. Here, instead of the expected analogous verb-final structure, a verb-second main clause can serve as second conjunct.

14

However, in the course of this presentation questions about asymmetric coordination structures will only be answered in case there is an overlap with the proposed approach to symmetric coordination.

2.2

Symmetric Coordination

Most of the earlier investigations into symmetric coordination (cf., e.g., [Ross,1967a], [Hankamer,1973], [Sag,1976], and [Neijt,1980]) assume a taxonomy of several levels. On the first level we distinguish between Constituent Coordination (CC) versus Non-Constituent Coordination (NCC). The former is represented by the rule of PHRASAL COORDINATION (PHC), which comprises lexical coordination as well as fully established phrasal coordination, i.e. a coordination of two complete XP's. Sentence (12) contains a lexical coordination of verbs, while sentences (13) and (14) show coordinations of complete phrases, here NP and CP, respectively. (12)

Ich [c kaufe] und [c esse] jeden Tag eine Tüte Gummibärchen. ( Ί [buy] and [eat] each day a pack of gumbears.')

(13)

Ingrid mag [jv/> laute Musik] oder [ ν ρ Greta-Garbo-Filme]. (Ingrid likes [loud music] or [Greta Garbo films].)

(14)

[cp In Hamburg regnet es] und [cp in Stuttgart scheint die Sonne]. ('[In Hamburg rains it], and [in Stuttgart shines the sun].')

We speak of Non-Constituent Coordination when one or more conjuncts appear in a so-called reduced form, which does not correspond to a constituent structure in the conventional sense. A major intention of the present study is to refute the idea that in case of NCC-constructions we are faced with reduced conjuncts—a hypothesis which stems from, as traditionally assumed within transformational approaches to grammar, a deletion transformation which operates on a fully established phrasal coordination. For expository purposes, however, we will—for the time being—still refer to the "deletion" metaphor in the following sections. The domain of NCC-constructions in turn is divided into two major groups, depending on the type and on the position of the "missing material". The first group is characterized by absence of the finite verb (and maybe further constituents) within the non-first conjuncts. The first conjunct displays a complete sentential (and thus constituent) structure, while the remaining conjuncts, since the finite verb is missing, obviously do not. Phenomena of this type are subsumed under the term GAPPING. The classical GAPPING construction shows two remnants within the non-first conjuncts, with only the main finite verb missing. Gapped conjuncts with more than two remnants are possible, as well as gaps comprising more constituents than the finite verb only. Gapped conjuncts with one remnant only have been termed STRIPPING constructions (cf., e.g., [Neijt,1980]).

15 (15)

Classical GAPPING:

[Johann liest die Zeitung] und [Maria φ ein Buch], ([Johann reads the newspaper] and [Maria φ a book].) φ = liest (reads) (16)

GAPPING with three remnants: [Johann schenkt Maria ein Buch] und [Peter φ Susanne eine Schallplatte], ([Johann gives Maria a book] and [Peter φ Susanne a record].) φ — schenkt (gives)

(17)

GAPPING over verbal projections:

[Meine Schwester liest die Zeitung am Morgen] und [mein Bruder φ am Abend]. ([My sister reads the newspaper in the morning] and [my brother φ in the evening].) φ = liest die Zeitung (reads the newspaper) (18)

STRIPPING:

[Ist Claus für das Projekt verantwortlich], oder [φ Peter φ\1 ([Is Claus responsible for the project], or [φ Peter φ\7) φ...φ'' = ist ... für das Projekt verantwortlich (is ... responsible for the project) Since GAPPING affects the finite verb, its domain, consequently, is confined to the phrasal levels of IP and CP. VP-GAPPING (which in GB terms amounts to GAPPING in C' for German), as assumed, e.g., by Neijt [Neijt,1980], for a construction such as (19)

Johann [c gibt Maria Blumen] und \c> φ Susanna eine Schallplatte], ([Johann gives Maria flowers] and [φ Susanna a record].)

will be shown to be otiose. More general principles will allow for attributing these instances to either IP or CP coordination. The second major group within NCC is distinguished by reduced first conjuncts. An arbitrary number of constituents at the right end of a phrasal structure is omitted, which may result in conjuncts with non-constituent structure, if only the remaining overt constituents are considered. (20)

[cp Johann liebt φ] und [cp Maria haßt Austern]. ([Johann loves φ] and [Maria hates oysters].) φ = NP-object: Austern (oysters)

(21)

\cp Johann wollte φ), aber [cp Maria durfte die Tagung besuchen]. ([Johann wanted to φ], but [Maria was allowed to attend the conference].) φ — VP-inf: die Tagung besuchen (attend the conference)

16 (22)

Ich glaube, daß {¡P Peter heute φ] und [/p Maria morgen kommt]. ( Ί think that [Peter today φ] and [Maria tomorrow comes].') φ = V-fin: kommt (comes)

(23)

[NP die kleinen φ] und [jvp die großen Kinder], ([the small φ] and [the big children]) φ = Ν: Kinder (children)

(24)

[pp für die kleinen φ] aber nicht [ρρ für die großen Kinder], ([for the small φ] but not [for the big children]) φ — Ν: Kinder (children)

Unlike GAPPING these constructions are not constrained to a specific domain. While GAPPING ranges over IP and CP structures only, this second type of NCC coordination occurs in any type of phrasal structure, as the above constructions exemplify. Depending on the chosen approach, this coordination type has been termed differently. The most prominent terms are BACKWARD CONJUNCTION REDUCTION or BACKWARD GAPPING (cf. [Ross,1967b], [Ross,1967a]), RIGHT NODE RAISING (cf. [Postal,1974]), SHARED CONSTITUENT COORDINATION (cf. [Radford,1988]), and RIGHT PERIPHERY ELLIPSIS (cf.[Höhle,199l]). For the time being we will use the term RIGHT NODE RAISING ( R N R ) when referring to this rule of coordination. Summarizing, the outlined classification of coordination structures is illustrated in (25). (25)

Classification

of Coordination

/

Structures

Coordination

symmetric

asymmetric

NCC

CC

/

Gapping (class. Gapping,

RNR

/ lexical

PhC

\

phrasal

Stripping, etc.) Symmetric coordination is sufficiently defined by the three types PHC, R N R , and GAPPING.1 This view on symmetric coordination is shared by, e.g., [Sag,1976], [Neijt,1980], [Wunderlich,1988b], and [Höhle,1991]. First, there are no further symmetric coordinate structures which cannot be covered by these three typical rules. Thus, no further rules are necessary. Second, a further reduction of these three rules to two, or even 1 See chapter 3 for a detailed exposition of the evolution of symmetric coordination into these three distinct types.

17 only one, is inadequate. T h e three coordination types are clearly distinguished by their typical characteristics as three disjoint types. Thus, symmetric coordination comprises PHRASAL COORDINATION, RIGHT N O D E RAISING, a n d GAPPING, a n d t h e s e only.

2.3

Relationship between the Types

Within transformational approaches to grammar it is traditionally assumed t h a t R N R and GAPPING constructions are derived via a deletion transformation from a maximal P H C construction. This relationship between the coordination types will be denied here. We will argue that P H C , R N R , and GAPPING are independently motivated phenomena, without any procedural dependencies between them. One of the reasons for this view lies in the fact that all three rules may apply to a given sentence: (26)

[Johann liebt grüne und rote φ\] und [Maria φ ι gelbe und weiße Gummibärchen]. ([Johann loves green and red φι] and [Maria φι yellow and white gumbears].) PHC = ... R N R = φ\ (Gummibärchen / gumbears) GAPPING = φ2 (liebt / loves)

It is certainly not desirable to assume a two- or threefold step derivation for sentences like the above one. Therefore, these sentences are in need of a more elegant and appropriate derivational explanation. But although we want to give up the view that the relationship between P H C , R N R , and GAPPING should be captured in terms of an interdependent sequential ordering of the three rules with respect to each other, we cannot deny the very fact of a relationship per se. Just as the overlapping application of the coordinate rules may serve as an argument for viewing these rules as independently motivated phenomena, at the same time it is precisely these overlapping occurrences which call for an integrative t r e a t m e n t . Characteristically, we encounter the following constructions, displaying an interaction of rule application: (27)

P H C and

a.

RNR:

[Johann repariert φ] und [Bill verkauft Autos und Fahrräder], ([Johann repairs φ], and [Bill sells cars and bicycles].) φ ( R N R ) = Autos und Fahrräder / cars and bicycles

b.

[Johann und Maria heiraten φ] und [Hans und Anna verloben sich morgen]. ([Johann and Maria get married φ], and [Hans and Anna get engaged tomorrow] . ) φ ( R N R ) = morgen / tomorrow

18

(28)

P H C and GAPPING:

a.

[Johann interessiert sich für Tanzen und Schwimmen] und [Maria φ für Fußball und Rugbyl. ([Johann is interested in swimming and dancing], and [Maria φ in football and rugby].) φ (GAPPING) = interessiert sich / is interested

b.

(29)

[Johann kocht und ißt oft Spaghetti] und [Maria φ Gulasch]. ('[Johann cooks and eats often spaghetti], and [Maria φ goulash].') φ (GAPPING) = kocht und ißt oft / cooks and eats often R N R and GAPPING:

a.

[Johann sagte Maria φι] und [Hans φ2 Anna, daß die Vorlesung ausfiele]. ([Johann told Maria], and [Hans φ Anna that the lecture was cancelled].) φι ( R N R ) = daß die Vorlesung ausfiele / that the lecture was cancelled φ2 (GAPPING) = sagte / told

[Hans kennt einen Mann, der drei φι], und [Maria Φ2 eine Frau, die vier Häuser besitzt]. ('[Hans knows a man who three φι], and [Maria Φ2 a woman who four houses possesses].') φι ( R N R ) = Häuser besitzt / houses possesses Φ2 (GAPPING) = kennt / knows

And finally, recall the above example displaying all three symmetric coordination types: (30)

P H C . R N R and GAPPING:

[Johann liebt grüne und rote φι] und [Maria φ2 gelbe und weiße Gummibärchen]. ([Johann loves green and red φι], and [Maria Φ2 yellow and white gumbears].) φι ( R N R ) = Gummibärchen / gumbears φ2 (GAPPING) = liebt / loves

An adequate analysis of symmetric coordination must strive for providing a framework for systematically correlating the three coordination types. If we deny a sequential interdependence between PHC, R N R , and GAPPING, two alternative approaches suggest themselves for nevertheless establishing a common framework. First, we could try to develop three distinct modes of analysis with a clear elaboration of their mutual interplay. Second, we could elicit the essential commonalities of the three coordination types and develop an analysis which is directed by these general characteristics of symmetric coordination. The aim of the present study is to develop such a common framework, which equally suffices all three of the coordination types P H C , R N R , and GAPPING, building on their common aspects, but still respecting their distinct status.

19

2.4

Beyond Symmetric Coordination

The three rules PHC, R N R , and GAPPING form a coherent group in that each of them consistently complies with the characteristic criteria for symmetric coordination, which essentially comprise the following: 1. Deletion of constituents under identity with respective linguistic antecedents 2. No occurrence of anaphoric references in reduced conjuncts 3. Restriction to coordinate structures only According to these criteria, elliptical rules such as VP-DELETION (cf., e.g., [Sag,1976]), SLUICING (cf., e.g., [Ross,1969] or [Sag a n d H a n k a m e r , 1 9 7 6 ] ) , a n d N - G A P P I N G (cf., e.g.,

[JackendofF,197l]), do not fall within the scope of symmetric coordination. 2.4.1

VP-Deletion

VP-DELETION violates all three items of the above list. (31) a.

I will leave when you do φ. φ ( V P - D E L E T I O N ) = leave

b.

Gwendolyn hit a single after Sandy did φ. φ ( V P - D E L E T I O N ) = h i t a single

(cf. [Sag,1976]) First, in example (31a) the deletion operation obviously cannot have been deletion under identity, since the fully expanded sentence is ungrammatical. Second, in example (31a) and (31b) do and did, respectively, refer anaphorically to the before mentioned verb phrase of the first conjunct. And, third, the examples in (31) show that VP-DELETION is not necessarily restricted to coordinate contexts. In German the scope of VP-DELETION is restricted to a very limited range of structures. In English the full range of variable VP-DELETION is available for a sentence like the following:2 (32)

I want to try to begin to write a novel, but you i.

don't φ.

[ii. don't want to φ. iii. don't want to try to φ. iv. don't want to try to begin to φ. A German quasi-analogue can only be constructed for the first variant: 2

The English example is taken from [Neijt,1980, p.20].

20 (33)

Ich möchte versuchen zu beginnen, eine Novelle zu schreiben, aber du i'.

φ nicht φ. φ (VP-DELETION) = möchtest . . . versuchen zu beginnen, eine Novelle zu schreiben.

For the remaining English constructions in (32) no direct German correspondence is available. In each case the anaphoric pronoun es (it) obligatorily refers to the respective antecedent VP. (34)

Ich möchte versuchen zu beginnen, eine Novelle zu schreiben, aber du ii'. möchtest es nicht. iii'. möchtest es nicht versuchen. iv'. möchtest es nicht versuchen zu beginnen.

2.4.2

Sluicing

The rule of SLUICING violates items (1) and (3) of the list of distinguishing characteristics of symmetric coordination. (35) a.

Johann will ein Auto kaufen, aber er weiß noch nicht, welches φ. φ (SLUICING) = Auto er kaufen sollte

a'.

John wants to buy a car, but he doesn't know yet which φ. φ (SLUICING) = car he should buy

b.

John is determined to buy a car on Saturday, although he doesn't know yet which φ. φ (SLUICING) = car he should buy

T h e German-English opposition in (35a) and (35a') demonstrates that SLUICING is not a rule of deletion under identity, and (35b) provides evidence that SLUICING also applies outside t h e realm of coordinate structures. 2.4.3

N-Gapping

T h e rule of N-GAPPING violates items (2) and (3) of the above list, and by implication also item (1). (36)

After t h e dog with a cow's liver died, he wanted to buy one (? φ) with five legs. φ (N-GAPPING) = dog

[Neijt,1980, p. 28] In sentence (36), a non-coordinated structure, the pronoun one takes up the aforementioned N P the dog, referring to a different dog, however. A deletion operation would presuppose the existence of an originally fully established N P one dog, obviously an inadequate assumption. Things look slightly different for the German equivalent:

21 (37)

Nachdem der Hund mit der Kuhleber gestorben war, wollte er einen φ mit fünf Pfoten kaufen. φ (N-GAPPING) = Hund

The structure of this sentence is ambiguous: einen (one) can equally function as referential pronoun and as indefinite determiner. The latter implies deletion (under identity), while the former does not. In German the nominal gap can occur in either conjunct in case of a coordinate structure: (38) a.

b.

[Claus mag schwachen Kaffee], aber [Bart bevorzugt starken φ]. ('[Claus likes weak coffee], but [Bart prefers strong

b.

(48)

Ich aß Fisch, und Bill Reis. (I ate fish, and Bill rice.) German—SOV:

a.

..., daß ich Fisch aß, und Bill Reis aß. ('...that I fish ate, and Bill rice ate'.)

b.

..., daß ich Fisch, und Bill Reis aß. ('...that I fish, and Bill rice ate'.)

Ross does not provide a formal definition of the two rules of FORWARD GAPPING and BACKWARD GAPPING, but illustrates them via phrase structure trees. The structures in b) are derived from the corresponding underlying structures in a): (49)

FORWARD GAPPING:

a.

NP NP

b.

NP

27 (50)

BACKWARD G A P P I N G :

NP

b.

NP

In [Ross, 1967b] Ross proposes a treatment of coordination structures which affect other constituents than the (finite) verb. Again he develops a mirror-image rule, termed CONJUNCTION REDUCTION, which operates leftwards and rightwards. Other than Chomsky, he understands his rule as an explicit deletion rule, augmented by a raising transformation: 4 A copy of some constituent which occurs in all conjuncts is Chomskyadjoined to the right or left of the coordinated node, on a right or left branch, respectively, with subsequent deletion of the original nodes (cf. [Ross,1967b, p.97]). Both, LEFTWARD a n d R I G H T W A R D C O N J U N C T I O N REDUCTION o p e r a t e across-the-board—only

elements

occurring in all conjuncts can be adjoined and deleted. (51)

L E F T W A R D CONJUNCTION

REDUCTION:

a.

The University's students are intelligent and the University's students are committed to freedom.

b.

The University's students are intelligent and (are) committed to freedom.

These sentences are assigned the corresponding structural representations:

4

In order not to confuse his and Chomsky's C O N J U N C T I O N R E D U C T I O N rule, Ross later spoke of a transformation here (cf. [Dougherty, 1971]).

N O D E RAISING

28 (52)

a.

S /\ NP VP b.

NP

\s

/ \s

S VP (53)

S / \ NP VP

VP

RIGHTWARD CONJUNCTION

REDUCTION:

a.

John picked these grapes, and Sally will prepare these grapes.

b.

John picked, and Sally will prepare these grapes.

Accordingly, the structural representation: (54)

a.

V NP b.

NP

29 Like Ross's GAPPING rule, the rule of CONJUNCTION REDUCTION, too, is meant to be a u n i v e r s a l r u l e . W h e t h e r LEFTWARD or RLGHTWARD CONJUNCTION REDUCTION a p p l i e s ,

is simply a matter of basic phrase structure of the affected node. W i t h t h e s e t w o b i - d i r e c t i o n a l r u l e s , GAPPING a n d CONJUNCTION REDUCTION, t h e

ground was laid for the division into the three distinct coordination types PHC, GAPPING, a n d R N R . FORWARD GAPPING r e m a i n e d t o b e t h e e s s e n t i a l G A P P I N G

phenomenon.

BACKWARD G A P P I N G a n d RIGHTWARD CONJUNCTION REDUCTION m e r g e d i n t o t h e c o o r d i n a t i o n t y p e of R N R . A n d LEFTWARD CONJUNCTION REDUCTION d e v e l o p e d i n t o PHC.

3.2 3.2.1

Further Development RNR

In [Ross, 1967a] Ross himself discusses an alternative analysis of backward gapped sentences. Since many speakers perceive a considerable intonational break before the rightmost finite verb 5 he suggests, for a sentence as in (55), the possibly more adequate structural representation in (56a) as opposed to the original representation of a backward gapped sentence, which is repeated in (56b): (55)

BACKWARD GAPPING in G e r m a n :

..., daß Johann Maria und Peter Susanne liebt. ('... that John Mary and Peter Susan loves.')

5 Te Velde [teVelde, 1988] presents quite interesting data which render convincing evidence for this conjectured intonational break. Thus we do not get good contraction between a so-called raised constituent and an adjacent constituent to its left: i. Jim believes that Sue, and George that Mary, will (* Mary'll) be the fastest. ii. The referee claims that we, and the fans that they, are (* they're) up to bat first.

30 (56)

a. alternative representation: 5

^ ^ S S

/

\

NP

VP

/

V

s

NP

NP b. original

NP

representation:

5

NP

I

VP

VP

I

NP

/

NP

5

\VP

NP

/\V

In (56a) the rightmost finite verb has been raised to the coordinate node, which renders a structure identical to a structure which results from applying the subsequently proposed rule of RIGHTWARD CONJUNCTION REDUCTION, as observed in (54b). Hankamer [Hankamer,197l] addresses precisely the question of conflating the two rules into one, arguing for an obliteration of a separate rule of BACKWARD GAPPING. First, he presents arguments that LEFTWARD CONJUNCTION REDUCTION and RLGHTWARD CONJUNCTION REDUCTION cannot be viewed as a mirror-image rule, since "they do not produce the same structural change" ([Hankamer,1971], cited from [Hankamer, 1979, p.95]). W h i l e Hankamer agrees with Ross concerning the operation of RlGHTWARD CONJUNCTION REDUCTION, he rejects a raising-plus-deletion approach to the data which the rule of LEFTWARD CONJUNCTION REDUCTION is supposed to cover. For the latter he simply assumes a deletion transformation to be at work, which may imply an obligatory restructuring process ("reduction" in Hankamer's terms) to render the correct structural representation (cf. [Hankamer, 1979, p.66]). The structure in (57a)

31 (57) a.

V

NP

V

cooked

the eggplant

ate

\

NP

the eggplant

is converted by deletion to:

and is automatically restructured ("reduced") to: c.

NP the eggplant

V

V

cooked

ate

Hankamer then proceeds to show that GAPPING is not a mirror-image rule either, but that, in fact, we are faced with two disjoint operations here. While he accepts simple deletion of the main verb for FORWARD GAPPING, he argues that for BACKWARD GAPPING a structural analysis consisting of raising plus deletion (as shown in (56a)) is to be preferred. From adopting the latter on the one hand, and the rule of RlGHTWARD CONJUNCTION REDUCTION on the other, it is only a small step to collapsing the two rules into one, generally viewing common rightmost constituents (whether finite verb or another type of constituent) as raised, with subsequent deletion of their identical antecedents.

32 A further consideration by Hankamer, namely that this rule is constrained to exactly one rightmost (major) constituent ([Hankamer, 1979, p.96]), was taken up by Postal [Postal,1974], Postal stipulated this constraint to be a necessary as well as a sufficient condition on t h e a p p l i c a t i o n of this, as he t e r m e d it, RIGHT NODE RAISING

(RNR)

transformation. The definition of R N R is formulated like the RLGHTWARD CONJUNCTION REDUCTION rule: a constituent occurring in all conjuncts on a right branch is raised and Chomsky-adjoined to the coordinate node, with subsequent deletion of the original nodes: (58) a.

XP XP '

\

YP WP

b.

ZP

Ζ

/\

WP

XP WP

Though it is disputable to use R N R as a constituency test (cf. [Bresnan,1974]), or to approach the relevant data by a raising-plus-deletion process (cf. [Jackendoff,1977], [Levine, 1985], [Abbott, 1976] ), it has been widely accepted since that the coordination structures which are intended to be covered by the rule of R N R constitute a distinct class (cf., e.g., [Hankamer,1971], [Sag,1976], [Neijt, 1980], [Höhle,1983], or [Wunderlich, 1988b]). 3.2.2

Gapping

H a v i n g c o l l a p s e d t h e rules of RIGHT CONJUNCTION REDUCTION a n d BACKWARD GAP-

PING, Hankamer proceeds in attempting to equally develop a uniform account for the rule of FORWARD GAPPING a n d DELETION & REDUCTION, t h e l a t t e r b e i n g H a n k a m e r ' s a l t e r -

native to Ross's LEFTWARD CONJUNCTION REDUCTION. A very general universal rule of COORDINATE DELETION should delete constituents from succeeding conjuncts under conditions of identity and parallel structure with the initial conjunct. This proposal, however, is refuted in [Sag,1976], where Sag presents convincing arguments for viewing FORWARD

33 GAPPING and DELETION & REDUCTION6 as two disjoint rules, explaining different sets of data. Following up, Sag elaborates a precise description of the rule of GAPPING, formerly FORWARD GAPPING, which remains the sole GAPPING rule. He shows his rule to be superior to the GAPPING proposals made by [Langendoen,1975], [Stillings, 1975], and [Kuno,1976], and—in its aspect of a comprehensive description of the relevant GAPPING phenomena—Sag's rule, represented in (59), has in fact been valid up to date. Sag sums up the observations hitherto made and postulates the basic characteristics of GAPPING to comprise: 7 • application in sentential categories • major categories (X 2 's) as remnants • possibility of discontinuous deletion

(59) X2 1 1

Wj 2 2

[X 2 ]· 3 3

W2] 4 4

CONJ 5 5

[5

X2 6 6

W, 7

Φ

[X2]* 8 8

w2; 9

φ

X, W = variables CONJ = and, or [.X2]* = any number of X 2 , s [Sag,1976, p.212]

3.2.3

PhC

W h i l e G A P P I N G a n d R I G H T N O D E RAISING a r e t h e r e s u l t of a r e f i n i n g p r o c e s s of g i v e n

deletion transformations, the rule of PHC is due to an essential change of perspective on instances hitherto handled by the various versions of "LEFTWARD CONJUNCTION REDUCTION". Instead of assuming a deletion operation to be responsible for the coordinate output in, e.g.: (60) a.

b.

John is erudite and Bill is erudite,

John and Bill are erudite.

6 Sag himself here coins a third term for these phenomena, namely LEFT PERIPHERAL DELETION. 7 Relevant examples are listed below (in (75), (76), (77)), where the essentials of GAPPING are contrasted with the ones of RNR.

34 (61) a.

Curvai hit Duelos and Curvai frightened Julie.

b.

Curvai hit Duelos and frightened Julie.

it is suggested (cf., e.g., [Dougherty,1970], [Dik,1968]) to directly base-generate the coordinate phrase by a phrase structure rule roughly stated as: (62)

X

X

(Conj X ) n

(n > 0)

So, instead of base-generating a sentential coordination, where the identical elements of the non-first conjuncts will subsequently be deleted (together with appropriate restructuring if necessary—recall Hankamer's analysis in (57)), the sentences in (60) and (61) immediately contain a coordinate structure of NP's and VP's, respectively, as represented in (63) and (64): (63)

5

NP

/\

NP

VP NP

(64)

NP

VP

/ \VP

VP

Thus, for Constituent Coordination phenomena the "Conjunction Reduction Hypothesis" is replaced by a "Phrase Structure Rule Hypothesis" [Dougherty,1970, p.851]. This move is motivated by constructions for which the position of unique sentential bases simply is untenable: (65) a.

John and Mary are a curious couple.

*

< -

b.

John

(66) a.

John

*

« -

b.

John

(67) a.

John

*

« -

b.

John

35 Via the "Phrase Structure Rule Hypothesis" the problem of derivation is circumvented: instances of Constituent Coordination, where all conjuncts are of the same syntactic category, are handled directly now. Dougherty's detailed phrase structure rule covers precisely the prototypical constructions in the above examples: (68) X where:

(Q)

X"

(ADV)

Q = distributive quantifier: each, all, both, either, A D V = distributive adverb: at once, etc. X = major category: S, NP, VP [Dougherty,1970, p.864]

neither

While for the above examples, involving symmetric predicates, the inadequacy of a sentential base is quite obvious, Höhle [Höhle,1991] extends this view to less obvious cases, affecting, e.g., indefinite NP's. His convincing refutation of any LEFTWARD CONJUNCTION REDUCTION in general is predominantly fostered by semantic considerations on the non-equivalence between allegedly reduced and unreduced structures. Thus, sentence (69a) is not synonymous to its alleged sentential basis (69b). The preferred reading of the latter sentence involves two different teachers being referred to, while preferably only one teacher is being spoken of in the former construction. (69) a.

daß ein Lehrer [den Kindern ein Buch zeigt] und [den Eltern einen Film zeigt] ('that a teacher [the children a book shows] and [the parents a film shows]')

b.

daß [ein Lehrer den Kindern ein Buch zeigt] und [ein Lehrer den Eltern einen Film zeigt] ('that [a teacher the children a book shows] and [a teacher the parents a film shows]') [Höhle,1991, p.150]

In view of a generalized approach to coordination, the "Phrase Structure Hypothesis" becomes extended to coordination constructions in general—be it Constituent or Non-Constituent Coordination—in that the basis for an arbitrary coordinate structure is generated by a general phrase structure rule, such as Dougherty's, under which sentential coordination is simply subsumed as a specific instantiation. In case of GAPPING or R N R constructions, then, deletion transformations operate on the respective base generated phrasal coordination. Hence, the assumed relationship between the three coordination types: (70)

P H R A S A L COORDINATION

GAPPING

R I G H T N O D E RAISING

36 (71)

GAPPING o p e r a t i n g on P H C :

John reads the newspaper and his parents read a book.

John reads the newspaper and his parents φ a book. (72)

R N R operating on PHC: John loves oysters and Mary hates oysters.

John loves φ and Mary hates oysters. Any coordination structure is sufficiently described by one of the three coordination types. No further operation is needed. Nor is evidence given for a further conflation of any of the three rules presented in the foregoing. 8 The examples of the next section corroborate this position. 3.2.4

Gapping versus R N R

The distinction between Constituent Coordination (PHC) on the one hand, and NonConstituent Coordination (GAPPING, R N R ) on the other, seems well motivated. But what about the two rules which are responsible for Non-Constituent Coordination? The following list of contrasting characteristics, illustrated in the examples below, should provide enough evidence for classifying GAPPING and R N R as two distinct rules. 9 (73)

A. B. C. D. E.

GAPPING

RNR

non-initial deletion discontinuous deletion possible only maximal projections as remnants deletion in sentential categories no phonological identity, but sense identity

non-final deletion only continuous deletion non-maximal projections as remnants possible deletion in arbitrary phrasal categories phonological identity, and sense identity

8 A recent attempt at nevertheless subsuming symmetric coordination rules—including even VPDELETION—under a unitary deletion rule, operating on sentential PHC constructions only, is presented in [Oirsouw,1987]. See [Neijt,1989] for a convincing refutation of this proposal. 9 See also [Neijt,1980, pp.39ff] for a discussion of opposed phenomena. Furthermore, [Wunderlich, 1988b] presents a list of criteria of admitted conjunct structures, taking into account the possibility of simultaneous application of G A P P I N G and R N R .

37 (74) ad A): GAPPING:

Claus drives a BMW, and Peter φ an Opel. *

Claus φ a BMW, and Peter drives an Opel.

RNR: They kissed φ, and they hit him. *

They kissed him, and they hit φ.

(75) ad B): GAPPING:

John gave his mother roses, and Bill φ his girlfriend φ. RNR: John gave φ φ, and Bill sold his mother roses. *

John φ his mother φ, and Bill gave his girlfriend roses.

(76) ad C): GAPPING:

Ingrid is interested in adverbials, and Tibor φ in passives. *

Ingrid is interested in adverbials, and Tibor φ φ infinitives.

RNR: Dirk cannot decide between the blue φ or the white Volvo. (77) ad D): GAPPING:

CP: Who wants to see "One, two, three", and who φ "Dead men don't wear plaid"? IP: Claus likes "One, two, three", and I φ "Dead men don't wear plaid"! RNR: CP: Who is working φ, and who is sleeping in the garden? P P : both in front of the blue φ and behind the white house NP: the youngest φ and the eldest member of parliament

38

(78) ad E): GAPPING:

sense-identity: *

John loves Mary, and Peter φ to swim,

no phonological identity: John goes to the cinema, and his parents φ to the theatre. RNR: sense-identity: Mary repairs φ, and John cleans planes. *

Mary saw φ, and John likes flying planes,

phonological identity: *

John loves φ, and Mary hates himself/herself, but in German: Johann liebt φ und Maria haßt sich.

In case of R N R constructions, "sense" identity is not to be confused with referential identity. No referential identity is required in Mary repairs, and John cleans, planes. Mary and John may be repairing and cleaning two totally different planes. It is obligatory, however, that the common constituent be of a uniform syntactic type with respect to each of the conjuncts. Hence, in Mary saw, and John likes, flying planes the rightmost constituent flying planes cannot at the same time function as the direct object for the first conjunct, and as a small clause complement in the sense of to fly planes for the second (the given judgement pertains to this ungrammatical usage). The same is true for GAPPING constructions—the chosen verb must uniquely conform to all of the conjuncts. These examples should suffice for illustrating that a separation of Non-Constituent Coordination structures into the two distinct rules of GAPPING on the one hand, and R N R on the other, is justified. 3.3

C S C ic

ATB

Two major constraints still need to be mentioned. The first, the Coordinate Constraint (CSC) [Ross,1967b, p.89], is formulated as follows:

Structure

The 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. By this constraint the ungrammaticality of the following sentences (taken from [Ross,1967b, p.88f]) is accounted for:

39 (79) a. * [What sofa], will be put the chair between [some table] and [t¿]? b. * [What table], will be put the chair between [t,·] and [some sofa]? c. * [The lute] which Henry [plays t¿] and [sings madrigals] is warped. d. * [Which trombone], did [the nurse polish t,] and [the plumber computed my tax]? Williams [Williams,1978], then, designs an approach to handling the class of systematic exceptions to the CSC. T h e CSC prohibits movement of, or out of, only one conjunct. If, however, movement uniquely applies to all conjuncts—viz. applies "across-the-board (ATB)"—we may get acceptable results: (80) a. b.

[Which film], did [the critics hate t,], and [the audience love t,·]? [On which topic]; did [Mary give a talk t¿], and [Peter write an article ti]?

Williams suggests a four-fold approach: First, the conjuncts in a coordinate structure are ordered in a particular way, in that their respective basic structures are written on top of each other. This representation Williams refers to as "ATB format": (81)

COMP

[The critics hated which film] [The audience loved which film]

and

Second, for WH-movement Williams "factorizes" this layered structure as follows: (82)

COMP 1

[The critics hated [The audience loved 2

which film] which film] 3

and J 4

Third, factorization is followed by the "ATB rule application". Provided the factorization process yielded a factor consisting of superficially identical constituents ("simultaneous factor" in Williams's terms), we may prepose a copy hereof, moving it to the C O M P node, while deleting the simultaneous factor. 1 0 (83) COMP

[the critics hate

which film

[the audience love

and

10 Within the examples Williams discusses he neglects the question of "do-insertion", focussing predominantly on indirect questions. We simply assume that it will occur at a later stage during the derivation.

40 Fourth, a rule applies which "linearizes" the coordinate structures given in ATB format into coordinate surface structures: (84) Xi and

[Xi (and) . . . and X n ]

[Williams,1978, p.42]

3.4

Redundancy and Recoverability

The development outlined before is embedded in the models of transformational grammar, according to which, roughly speaking, the syntax is divided into a base component (supplied by PS-rules and lexical insertion rules), determining the D-structure, and a transformational component, yielding the S-structure. (85) Base: PS-rules lexical insertion rules D-Structure Transformations S-Structure

The format of a transformational rule is determined by the two input and output levels of "Structural Description" and "Structural Change", respectively. If given, further constraints on the nature of the involved constituents can be stated. Each deletion transformation is subject to the recoverability constraint, which demands that no nonredundant material be deleted. (86) "We are proposing the following convention to guarantee recoverability of deletion: a deletion operation can eliminate only a dummy element, or a formative explicitly mentioned in the structure index (for example, you in imperatives), or the designated representative of a category (for example, the ΐ/Λ-question transformations that delete Noun Phrases are in fact limited to indefinite Pronouns—cf. Chomsky, 1964, §2.2), or an element that is otherwise represented in the sentence in a fixed position. [Chomsky, 1965, pp.l44f]

41 Much attention has been devoted to the question on which level of representation this constraint is to hold (cf. [Neijt,1980] for a discussion of [Lees,I960], [LakofF,1968], [Sag,1976]), but this should not concern us here. As to the question of redundancy, we will take a different perspective here. Our view— which correlates directly with denying any deletion operation to be at work—is that in a coordinate structure there simply are no redundant constituents involved to begin with. Each coordinate structure ("reduced" or "non-reduced") is a base generated phenonemon. In each case we observe (apart from movement and phonetic rules) a one-to-one mapping between base generated constituents and their occurrence in the actual surface structure form. Hence, mechanisms for recovering redundant constituents become irrelevant.

3.5

Coordination within GB

Within the GB-framework the transformational component is reduced to the single rule of move a. The twofold format of Structural Description and Structural Change is rendered obsolete; general principles, such as "Subjacency", govern the application of move a . Deletion processes are since assumed to take place on the level of PF: 1 1 D-Structure I

movement

S-Structure

PF (deletion, etc.)

LF (rules of control, etc.)

phonetic representation

semantic representation

Within this model of grammar coordination constructions have received little attention. In particular, comprehensive investigations to symmetric coordination in its full range have been very rare. 12 If, at all, we find studies of single coordination rules. We will outline the most significant ones here. Also within the GB framework, the prevailing idea as to the relationship between the single coordination types is that a fully established PHC construction serves as base for the application of either R N R or GAPPING. Höhle [Höhle,1991, p.l42f] reformulates 11 See [Riemsdijk and Williams,1986, p.l72f] for a more detailed illustration of this as well as the preceding models of transformational grammar. 12 An exception here is the work of [teVelde, 1988], who proposes to augment the grammatical system of GB by features in order to capture the three significant coordination types. However, te Velde does not question hitherto presented analyses, nor does he try to derive his approach from other grammatical principles independent of coordination.

42 these two deletion rules in terms of ellipsis licensing principles, assuming both to apply optionally. (87) Right Periphery Ellipsis (RPE): In some 'B a string at the right periphery is ellipsed at surface structure under phonological and sense identity with a string at the right periphery of the final conjunct "B. (88) Gapping: In some 'B an independent verb is ellipsed (possibly together with certain other elements of the same conjunct) at surface structure under sense identity with corresponding elements in the initial conjunct 'B. The simultaneous application of both, RPE and GAPPING, may licence surface structures with reduced conjuncts only: (89)

[Karl hört dir i], [Heinz φι mir φ\] oder [Walter φ2 ihr zu]. ('[Karl listens you ^i], [Heinz φι me φ\] or [Walter φι her to'].) φι = zu / to Φ2 — hört / listens

Obviously, with deletion transformations not being available any longer, R N R and GAPPING phenomena are in need of alternative explanations. In contrast, for PHC constructions one continues, somewhat uncontroversially, to assume a direct base generation along the Phrase Structure Rule Hypothesis. Pesetsky [Pesetsky,1982] is an exception in this respect. He incorporates PHC coordination into his overall grammatical system, according to which any syntactic structure is licensed by the conditioned establishment of valid paths between the various nodes in a tree. Not questioning the general hypothesis as to PHC constructions, Pesetsky then suggests relating the grammaticality of a given phrasal coordination to the establishment of licensed paths. In particular, he can show the regularities hitherto handled by the Coordinate Structure Constraint (see section 3.3) to follow directly from his approach. Munn [Munn,1987] takes up this line of argumentation and proposes relating the licensing of ellipsis in coordinate structures to the licensing of parasitic gaps. 13 According to him both can be shown to obey the same principles as to valid path establishment. Hereby the gaps occurring with an ATB-type extraction are interpreted such that the gaps in non-first conjuncts are parasitic on the initial gap occurring in the first conjunct. Since the existence of parasitic gaps in, e.g., German still remains a matter of debate, the universality of such a proposal, however, must be doubted. 13 The occurrence of a parasitic gap may be observed in: i. Which papersi did you file ti without reading e¿? The second gap e so-called parasitically bears on the existence of, and must be licensed by, the presence of the real gap i.

43

3.5.1

Gapping in GB

Following the assumption that deletion transformations apply on the level of P F , Sjoblom [Sjoblom,1980] develops an approach in which reduced coordinate structures are controlled by the position of contrastively accented categories. He symbolizes these contrastively accented categories by X and X , without specifying their exact intonational contour. According to Sjoblom a GAPPING construction shows the following pattern: (90)

Peter loves Betsy, and Betsy φ Peter.

T h e filter he designs ('C-filter') constrains the position of grave accented nodes. T h e presence of a conjunction C is a necessary condition for licensing a string of grave accented nodes. This string is called s. Hence, a string s is ruled out, unless preceded by a conjunction C. T h e filter nicely accounts for GAPPING as well as P H C instances, since s is furthermore confined to either (possibly multiple) maximal projections only, or single constituents (minimal or maximal) with no further constituents except for t h e conjunction intervening between the contrastively accented categories. T h e former accounts for GAPPING including STRIPPING constructions, the latter for any instance of P H C (U here stands for any maximal unaccented string): (91) C-Filter: * θ unless

/

u C

. [ X, and u = 0 where s — < - » 1 (Xa) [Sjoblom,1980, p.94] R N R is not handled by Sjoblom's system. This is not surprising, because in (92) the string Bill hates obviously is neither a constituent, nor a string of maximal projections: (92)

John likes, and Bill hates, rice.

Sjoblom classifies R N R constructions as instances of "Secondary Conjunction", and simply excludes these structures from his investigation of "Primary Conjunction", which— according to him—are those coordinate structures, i.e. P H C and GAPPING, which comply with the proposed pattern. Neijt [Neijt,1980] relates the conditions governing the application of GAPPING to the conditions governing the rule of move a. She assumes GAPPING to operate on t r u e variables. In analogy to the very general rule of move a , which says: "Move" (any constituent anywhere), Neijt formulates the rule of GAPPING simply as: (93) Gapping: "Delete" [Neijt,1980, p.95]

44 She then designs an extended version of the Subjacency condition, which is meant to sufficiently constrain the range of application of both of the very general rules, move a ("Move") and GAPPING ("Delete"). (94) Strict Subjacency No rule may involve X, Y in . . . . X . . . [„ ... Y . . . ]

...X...

where a is NP, or α is S iff its specifier is lexically filled, unless X binds a trace t^ in a , and there is no cyclic boundary between t r and Y. [Neijt,1980, p.185] Thus Neijt succeeds in explaining a significant number of GAPPING constructions in Dutch and English. Though very promising, some GAPPING constructions, however, remain inaccessible to this approach. It is at a loss, for example, when it comes to explaining the grammaticality of sentence (95). 14 (95)

This doctor suggested [5 I should buy tunafish], and that doctor φ salmon.

In relating that doctor to the second remnant salmon we clearly cross a cyclic sentential boundary (a = S) with a lexically filled specifier (the subject / ) , which is prohibited by Strict Subjacency, unless the matrix NP that doctor would bind a trace in a , which obviously is not the case. It must be noted, though, that this GAPPING construction seems to be specific for English. Neither Dutch nor German permit equivalent structures. Goodall [Goodall,1984] in his approach to parallel structures, under which he subsumes coordination, focusses on the common finite verb in GAPPING constructions. He develops a sophisticated system which admits the generation of "Reduced Phrase Markers". Via an operation of "union of phrase markers" he furthermore provides for the generation of phrase markers which consist of two or more independently well-formed phrase markers, i.e. structures which may contain two or more distinct terminal strings (cf. [Goodall,1984, p.24]). Goodall thus adds a third dimension to the structural representation in terms of tree diagrams, traditionally thought of as being two-dimensional, representing precedence and dominance relations only. Any coordination structure is represented as a union of phrase markers in Goodall's system. It amounts to a type of superimposing two (or more) trees upon each other, with identical nodes merging into one unique representation. The result of a union of phrase markers is reflected in the following diagram, which displays precedence relations only: John (96)

Mary loves

Peter

Susan

14 For a discussion of this construction see also section 6.5.1.

45 A Linearization Principle, which applies on the level of P F , guarantees the corresponding GAPPING surface structure output: (97)

John loves Mary, and Peter Susan.

T h e Linearization Principle establishes a precedence relation between constituents previously not ordered with respect to each other—as in the above example between John and Mary, and Peter and Susan. However, Goodall can only explain structures with phonetically identical verbs. Grammatical GAPPING sentences such as: (98)

John reads the newspaper, and his parents a book.

elude his system. No representation, analogous to the one in (96), can be provided. In fact, across languages judgements as to the grammaticality of GAPPING constructions with phonetically non-identical finite verb forms vary. For some speakers only strict identity licenses GAPPING. However, we agree with Höhle [Höhle,1991], among others, in assuming that GAPPING is merely bound to sense identity, but not to phonetic identity. W i t h phonetically identical material being the focus of his investigation the incorporation of R N R lies at hand. Though in the course of his presentation Goodall discusses a number of R N R constructions, such as the following ([p.49] and [p.59], respectively): (99) a. b.

John eats, and Mary cooks, pies. I know t h a t the boys, and I believe that the girls, like ice cream.

he does not a t t e m p t to extend his system to this coordination type. In fact, he even denies the applicability of his account for R N R constructions (cf. [Goodall,1984, p. 122f]). Chao [Chao, 1988] looks at ellipsis phenomena which according to her include—among VP-DELETION, SLUICING, STRIPPING and Null arguments—also GAPPING constructions. Though we do not agree with this specific classification of linguistic phenomena, it is worthwhile to consider Chao's interesting proposal. Instead of defining deletion operations to be at work in case of GAPPING, Chao allows for generating t h e reduced surface structure directly via phrase structure rules. 1 5 Her "defective" X-scheme, being responsible for GAPPING constructions, which characteristically lack the verbal head (therefore classified as an "H-" type), is designed as follows: (100) Defective X-schema: H- series X-

Y*

X-

(SPEC)

(X-)

Y*

15 A similar proposal is contained in [Wunderlich, 1987] and [Wunderlich, 1988b] (the latter developed within the framework of GPSG), where it is suggested viewing reduced coordination as a base generated phenomenon, reconstructing given missing constituents in the course of semantic interpretation.

46 This overgenerating rule is further constrained via a "licensing condition" ("All Xnodes in a structure must be in the H- projection whose highest node is t h e highest node of the e[lliptical]-clause" [p.40]) to admit exactly the grammatical GAPPING structures. On the level of Logical Form the missing constituents are reconstructed ("E-Reconstruction" in Chao's terms), in accordance with the complete first conjunct ([Chao, 1988, p.48]): 16 (101) E-Reconstruction E-reconstruct m a j o r constituents of an H + projection under the corresponding nodes in the H- projection of a corresponding e-clause. T h e result of Ε-reconstruction is a complete logical representation of the gapped sentence. But, if the information contained in the INFL node is reconstructed by identity, again we are unable to account for GAPPING sentences with non-identical INFL content: (102)

John admires Joe Jackson, and his friends φ Elvis Costello. φ = admire

A [third person, singular] agreement specification in the first conjunct is opposed by a [third person, plural] specification in the second conjunct. 3.5.2

R N R in G B ?

For the coordination type R N R no alternative account has been provided which is directly embedded into the GB-framework. Two proposals, related to earlier versions of transformational grammar, should be mentioned here, however. The first comes from Erteschik-Shir [Erteschik-Shir, 1987a], where she relates R N R constructions to the ATB-format presented in Williams [Williams,1978] (see section 3.3). Erteschik-Shir proposes a treatment of factorizing R N R constructions into t h e conjuncts on the one hand, and the allegedly raised constituent on the other. For a sentence such as ([Erteschik-Shir,1987a, p.5]): (103)

I thank Mary and respect Susan for their help,

this would amount to the following factorized structure: (104)

I

[thank [respect

Mary] Susan]

and ]

J for their help

The sentence in (103) presents a convincing counterexample against an analysis via raising (moving) plus deletion. First, the prepositional phrase for their help cannot possibly have originated in the assumed positions. 16 "H+ projections" here mean projections containing a verbal head, "H- projections" the opposite; "e-clause" refers to the elliptic clause; the term "major constituent" applies to all of the immediate daughters of the nodes in an Η projection sequence, hence, e.g., the INFL node, the verbal head, and, if given, further complements and adjuncts.

47 (105) *

I thank Mary for their help and respect Susan for their help.

Second, if—nevertheless—raised to the right of the coordinate VP-node, the pronoun contained in the raised P P would c-command its antecedents Mary and Susan, certainly an undesirable consequence.

thank Mary^i t,

respect Susan j 2 t,

With the factorized representation in (104) neither deletion nor movement to the raised position need be assumed. Instead, following the above representation the factors can be interpreted in situ, and, furthermore, Mary and Susan, being contained in one factor, can serve as appropriate antecedents for the pronoun their. Finally, like in Williams' system, "Linearization" will render the actual surface structure output, shown in (103). 17 However, in [Williams, 1978] no definite account is given of how this factorized representation is to come about. Erteschik-Shir concedes that Williams simply stipulates that ATB representations exist without stating how they come into existence. She suggests overcoming this deficiency by adopting a representation such as Goodall's, superimposing factorization onto his parallel structures. Unfortunately, however, she leaves the matter at that, and does not work out a concrete proposal. The second proposal comes from McCawley (cf. [McCawley, 1982], [McCawley,1988]). Agreeing with the counterarguments against traditional R N R approaches, McCawley presents a structural representation, which builds on the assumption of "Fusion of Identical Constituents", and which reflects this "in situ" character of the allegedly raised constituent. His structural proposal accounts for the discontinuity of the rightmost constituent, but technically this structure is no longer a tree, because it contains a constituent with more than one mother:

17 See the next chapter for further counterarguments against a raising-plus-deletion approach to R N R type constructions.

48

(107)

S NP John

However, we are not provided with a formal definition of the FUSION operation, which is responsible for generating this discontinuous structure.

3.6

Summary

We outlined the development of the three significant symmetric coordination rules, and sketched approaches to single rules within the framework of GB. While grammarians agree on the classification of the linguistic phenomena, a comprehensive approach to symmetric coordination has not been presented yet. In particular, also within GB the idea prevails that each so-called reduced coordination must be derived from a fully established PHC coordination, which implies the application of deletion at some point in the derivation (here on PF). In the following chapter, then, we will summarize the counterarguments against any type of deletion approach, laying the ground for a different perspective on the constructions under study here.

Chapter 4 Neither Deletion Nor Reconstruction

As argued in the previous chapters, symmetric coordination comprises the three distinct coordination types P H R A S A L C O O R D I N A T I O N , R I G H T N O D E R A I S I N G , and G A P P I N G . The approach we will propose will respect their distinguishing characteristics. But in abstracting away from their specific features, and clearly focussing on what the three coordination types have in common, we will elaborate a comprehensive and very general account of symmetric coordination, equally valid for all three coordination types. In this respect, any type of approach which is inadequate for one of the three coordination types must consequently also fail as a general approach to symmetric coordination. It has been mentioned repeatedly that an approach to R N R constructions via deletion (following raising) is inappropriate. Therefore, our approach will not rely on any type of deletion, nor reconstruction in the sense of Chao, after "defective" generation of a phrase. Accordingly, we deny any procedural dependency between PHC and R N R , and—by implication—also between P H C and G A P P I N G . The examples to be discussed below will, once again, refute the assumption of deletion under redundancy, and make way for a different view: (108) In coordinate structures there do not exist any redundant constituents. Apart from movement (and phonetic/stylistic rules applying on PF), each constituent is spelled out on the surface as generated in D-Structure. In our view, coordination, in its different varieties, is a base generated phenomenon. In part III we will, therefore, present a phrase structure account which allows for directly generating any type of symmetric coordination. The counterarguments against deletion (or reconstruction) will be viewed against the model of grammar assumed within the GB-framework. We will start with examining approaches to P H C , and then proceed to R N R constructions. For G A P P I N G constructions, counterexamples against a deletion approach are not so easy to find.1 But since we want to propose a unified account of symmetric coordination, we will abandon a deletion approach here, too, embedding the treatment of G A P P I N G into the overall treatment.

1

But see the examples in section 5.4.1 for possible counterevidence.

50

4.1

P h C and Deletion

With respect to the treatment of coordinate structures a major drawback of the GB-model of grammar as outlined in (85) lies in the lack of correspondence between the two levels of representation Phonetic Form and Logical Form. Sjoblom [Sjoblom,1980], despite the numerous counterarguments, still assumes a general sentential base (except for N P coordination in the context of symmetric predicates). He postulates t h a t every m a j o r coordinate structure (for him only P H C and GAPPING) displays the same stress pattern. Thus, on the level of PF—as intended in this model 2 — deletion rules apply, provided the output shows the desired intonation. Let us examine this approach with respect to the relation between deriving the actual surface structure output on the one hand, and deriving the sentence's corresponding logical form, which subsequently will be the input for its semantic interpretation. A sentence such as (109)

Yesterday a man attacked three old ladies and a man robbed a bank,

would be reduced on P F to (110)

Yesterday a man attacked three old ladies and robbed a bank.

Sentence (109) is not only input to P F , but also to LF. However, the meaning we obtain for sentence (109) is not identical to the meaning of sentence (110) 3 . In sentence (109) we talk of two different men, in sentence (110) we refer to one and the same man. For the above two sentences this problem might be circumvented by a coindexing mechanism: (111)

Yesterday a man¿ attacked three old ladies and a man, robbed a bank.

On LF, then, both subjects would be identified as one and the same logical entity, and Sjoblom's deletion rule could be extended so as to delete not only under (phonetic) identity, but also under coindexation, conveying referential identity. Things look different, however, if instead of a man we refer to the same man in a given context. Jackendoff notes t h a t

2 To be precise, Sjoblom elaborates his approach within the model of "Filters k Control", as introduced by Chomsky & Lasnik [Chomsky and Lasnik,1977]. However, since the critical distinction into P F on the one hand and LF on the other was introduced in this paper, arguments bearing on the lack of correspondence between the two levels carry over to the GB framework. 3 For a discussion of this—as they call it—"notoriously non-meaning-preserving" character of CONJUNCTION REDUCTION see, e.g., [Partee and Rooth,1983, pp.376ff].

51 . . . e v e n Dougherty's (1970-1971) highly restricted version of of Conjunction Reduction comes to grief in examples like these:

T h e same m a n

got drunk and was arrested by the cops is rarely easy to please and eager to please praised you and seemed to h a t e you

Dougherty invokes a Conjunction Reduction transformation 4 specifically in order to deal with conjoined derived verb phrases. But the underlying source he needs is ungrammatical here: ii. * T h e same man got drunk and the cops arrested the same man. * It is rarely easy to please the same man and the same man is rarely eager to please. * T h e same man praised you and it seemed [(for) the same m a n to h a t e you]. [Jackendoff,1977, p,193f] It should be noted t h a t the fully established sentential coordinations listed in ii. are by themselves acceptable, if the two instances of the same man are non-coreferential, e.g.: (112)

T h e same man, got drunk and the cops arrested the same m a n j .

But coreferentiality, the necessary condition for deriving the sentences in i., is ruled out. T h e same line of reasoning applies to any reconstruction approach. Chao suggested a "defective" phrase structure generation for G A P P I N G constructions (as described in section 3.5.1). Provided we adopted this approach, adequately modified, also for t h e domain of PHC constructions. Accordingly we could generate: (113)

Yesterday the same man attacked three old ladies and robbed a bank.

On LF we are able to derive the adequate logical form for this sentence as it is. W h a t we certainly want to avoid, though, is to reconstruct the "missing" instance of the same man as subject of the second conjunct. It seems obvious, then, that also for this model of g r a m m a r a "Phrase Structure Rule Hypothesis" is to be preferred over a "Conjunction Reduction Hypothesis". Adopting Dougherty's proposals—originally conditioned to t h e transformational model of Standard Theory—-also for the model of GB, we would directly generate any type of constituent coordination via a general phrase structure rule such as: (114)

X

-»

(and, or)

Xn

4 The term Conjunction Reduction transformation is misleading here. In fact, Dougherty sets out to show "that a Coordinate Transformation . . . should be a substitution transformation, not a reduction transformation" [Dougherty, 1970, p.850].

52

4.2

R N R and Deletion

For R N R constructions, counterexamples along the same line as for PHC constructions can be construed. Concerning coindexation, however, we are presented a slightly different picture. (115)

Peter writes a letter; and Mary reads a letter —>

Peter writes and Mary reads a letter, In (115) two different letters are involved. This obtains for the reduced as well as for the unreduced form. Thus, while phonetic identity is a crucial condition for deletion, referential identity is not necessarily required. This behaviour of rightward deletion is contrary to assumed leftward deletion, as observed in (109) and (110). Here, we do obtain referential identity with the so-called reduced sentence, as opposed to arbitrary reference in the corresponding unreduced sentence. Similar to PHC structures it is equally difficult to provide a complete sentential basis, when wide scope inducing predicates are involved. The following well-known examples originate from [JackendofF,1977, pp. 192f] :5

5 It is interesting to note, though, t h a t the chosen verb form in (116b) seems to play an essential role, too. Exchanging simple past for, e.g., past progressive leads to questionable, if not ungrammatical, constructions:

T h e simple past form is ambiguous as to singular or plural agreement, whereas the auxiliary with the progressive form makes agreement explicit. Since we expect a plural subject for doing s.th. together the ambiguous simple past form qualifies, whereas the non-plural progressive form does not. If, instead, we take plural subjects here, we again obtain well-formed sentences, allowing even for both, the reduced and the unreduced reading.

Similarly, for certain German R N R instances we find t h a t with singular subjects we get an ungrammatical construction, while with plural subjects b o t h readings (reduced find unreduced) seem to be available: i. * Karl pfiff [3.p.sg.] und Maria s u m m t e [3.p.sg.] gemeinsam. (Karl whistled [3.p.sg.] and Mary hummed [3.p.sg.] together.) ii. Die Männer pfiffen [3.p.pl.] (gemeinsam) und die Frauen summten [3.p.pl.] gemeinsam. (The men whistled (together) and the women hummed together.)

53 similar issues (116) a.

John avoided and Bill ignored

the same man men with the same birthday the same tune

b.

John whistled and Mary hummed

at equal volumes together

As Jackendoff [p. 192] notes "none of these sentences can arise from Conjunction Reduction or Right Node Raising, since the presumed sources would either have the wrong reading or be ungrammatical" : similar issues (117) a.

John

avoided

and

the same man

Bill

ignored

men with the same birthday similar issues the same man men with the same birthday the same tune b.

John

whistled

* at equal volumes

and

Mary

hummed

* together the same tune * at equal volumes * together Here, exactly the same arguments hold as presented for PHC constructions with wide scope inducing predicates: due to the lack of correspondence between PF and L F we will obtain non-matching phonetic and semantic representations of a given sentence, if deletion were to take place on the level of PF. And again, a reconstruction approach would be just as inappropriate here. Further convincing data against the deletion account of R N R are presented by Abbott, who points to the problematic nature of a deletion derivation for the example in (118) in view of the non-equivalence (at least for one of the readings) to the corresponding sentences in (119) [Abbott,1976, p.642]: (118) a. b.

(119) a.

I borrowed, and my sister stole, a total of $3000 from the bank. John gave Mary, and Joan presented to Fred, books which looked remarkably similar. I borrowed a total of $3000 from the bank and my sister stole a total of $3000 from the bank.

b.

John gave Mary books which looked remarkably similar and Joan presented to Fred books which looked remarkably similar.

54

For these RNR sentences a phrase structure solution, as adopted for similar PHC constructions, is not so easily available, since we predominantly deal with non-constituent structures here. Before proceeding let us sum up the remaining counterarguments against a raising and deletion approach to RNR constructions. Further counterevidence stems from the fact that the allegedly raised constituent is felt to be still present in the sites from which it has been displaced and subsequently deleted (recall example (103) from the previous chapter). Among others, Levine [Levine,1985] argues that "it seems that pronouns in the conjuncts are only coreferential with NP's in the raised constituent when the raised constituent corresponds to a gap in the conjunct such that the pronoun would not ccommand the NP if the raised constituent filled the gap" [p.496]. Thus the sentences in (120) are ungrammatical, while the ones in (121) are well-formed. (120) a. * She; said—and I happen to agree—that Mary, needs a new car. b. * She, disliked, and I hated, that picture of Mary;. (121) a. b.

Though he, was polite, Mary knew—and though he, seemed generous, Tom suspected—that John, was operating from dubious motives, I told the man he; had insulted, and the woman he,· had screamed at, that John; had an unbalanced mind.

Second, the RNR transformation may violate island constraints. Though in general movement out of, e.g., a relative clause is prohibited, raising seems to be allowed here. Nevertheless, when displaced from such a (here wh-) island, the constituents seem to retain the information about their origin. Hence, as Wexler L· Culicover [Wexler and Culicover,1980] point out, questioning of (or out of) a raised constituent is possible with RNR constructions in the course of which no island constraints have been violated (as in (122)), but impossible otherwise (as in (123)). (122) a. b.

Mary buys, and Bill sells, pictures of Fred. Who; does Mary buy, and Bill sell, pictures of t,?

(123) a.

Mary knows a man who buys, and Bill knows a woman who sells, pictures of Fred. b. * Who, does Mary know a man who buys, and Bill know a woman who sells, pictures of t,?

A third argument attacks the assumption that RNR affects one, and only one, rightmost constituent, which led Postal [Postal,1974] to the hypothesis of employing RNR as a test for constituenthood. In a number of articles this view has been questioned, however (see, e.g., [Bresnan,1974], [Abbott,1976], [G rosu,1976], [Zwarts, 1986]). In English, constructions such as the following (taken from Abbott [p.640]): (124)

John offered, and Mary actually gave, a gold Cadillac to Billy Schwartz.

55 seem perfectly a c c e p t a b l e for a large community of speakers. A possible solution, suggested by Grosu [Grosu,1976, p.643], might lie in allowing R N R to apply iteratively t o t h e constituents in question. T h i s approach is at a loss, however, when it comes t o

RNR

constructions in, e.g., G e r m a n , where the operation m a y extend into t h e internal s t r u c t u r e of constituents: (125)

Karl hat einen M a n n , der zwei

, und A n n a hat, [γρ [ΛΓΡ eine Frau, die

[νρ [ΛΓΡ drei Hunde] besitzt]], gekannt t,·] . (Karl

has

a

man

who

two

, and

Anna

has

a

woman

who

three

dogs possessed known.) T h e s e are t h e constructions which Hoehle (cf. [Höhle,1983] and [Höhle,1991]) takes as evidence to abandon a raising approach altogether, and to assume simple deletion to b e at work, which applies (under phonetic identity) to a d j a c e n t rightmost i t e m s in t h e left c o n j u n c t only, thus not affecting the right conjunct at all. His Right Periphery Ellipsis ( R P E ) (as cited in ( 8 7 ) ) , in essence a re-implementation of R o s s ' s BACKWARD CONJUNCTION REDUCTION, not only adequately covers the t y p e of sentences illustrated in ( 1 2 5 ) . W i t h R P E we are also able to explain t h e above mentioned puzzle as t o t h e coreferentiality between pronoun and allegedly raised antecedent NP, since t h e affected items are left in place. B u t still, also R P E fails as soon as wide scope inducing predicates are involved: ( 1 2 6 ) a.

Die Grünen bekämpfen und die R e c h t e n unterstützen dieselben Leute. ( T h e greens fight and the rightists support the same people.)

b.

T h e R e d Sox b e a t , and the Giants were b e a t e n by, different t e a m s . [Gazdar,1981, p.180]

T h e discussed examples should suffice as counterevidence against any t y p e of approach to R N R constructions which involves deletion at any point.

Part II

Shared Constituent Coordination

59 This part is devoted to a different view of what should be regarded as the adequate classification of the given phenomena of symmetric coordination. The discussion will lead to the general classification of coordinate constructions into two distinct types: first Shared Constituent Coordination (SCC),1 which comprises RNR, GAPPING, and PHC of non-maximal projections, and second Non-Shared Constituent Coordination (Non-SCC), which is represented by PHC of fully established XP-conjuncts. Since our approach to symmetric coordination will concentrate on German coordination phenomena, we will start by presenting the range of German coordinate conjunctions, followed by a set of German examples employing these conjunctions. The examples cover phrasal and lexical coordination of (projections of) lexical and non-lexical categories, each in turn viewed, if appropriate, within the three coordination types PHC, RNR, and GAPPING.

Subsequently, we will review the presented examples and hereby motivate our alternative classification of symmetric coordination. The discussion will lead to a classification viewing the following coordinate phenomena as constituting distinct classes:

1 Referring to R N R constructions Radford [Radford, 1988] also preferably uses the term Shared Constituent Coordination. As described here, we use the term in a more extensive way.

60 (127) Re-Classification of Coordinate Structures I. NON-SCC formerly: - PHC of fully established XP's II.

SCC

i. peripheral SCC * Right SCC formerly: • RNR

• PHC of lexical heads governing rightwards * Left SCC formerly: • PHC of lexical heads (and non-maximal projections hereof) governing leftwards ii. "medial" SCC formerly: *

GAPPING

Furthermore, we consider any coordinate construction, be it Constituent or NonConstituent Coordination, to range over maximal projections. The assumed XP-domain of coordination is easy to see for PHC of complete XP's, as well as for R N R and GAPPING constructions, which presumably are derived from a fully established phrasal PHC. The generalization to assuming an XP-domain also for so-called lexical PHC constructions is motivated by the fact that the involved complements required by the respective lexical heads (whether governing rightwards or leftwards) underly exactly the same constraints as the "raised" constituent in an R N R construction. A uniform view of R N R and non-maximal PHC coordination phenomena as belonging to the same class opens the way to a common treatment in which we keep the advantages of previous analyses of the one phenomenon, and do away with the problematic aspects of the other. Hitherto, R N R constructions have been assumed to be due to (raising and) deletion, an approach which leaves a number of questions unanswered. PHC constructions, on the other hand, have been approached directly via phrase structure rules, sufficiently covering the constructions in question. For a uniform treatment of both it lies at hand to generalize the latter account to cover also the instances of R N R constructions. Hence, our new classification will lay the ground for pursuing a general treatment in terms of a direct phrase structure rule account, without further recourse to any type of deletion operations.

61 Our alternative theory of phrase structure, which we will present in detail in part III, will evolve as a natural consequence of our S C C view of symmetric coordination. This SCC view bears crucially on two concepts: first—as the name already suggests—on the concept of shared constituents, and second on the concept of Contrast Sets, both concepts being vital for any type of symmetric coordination. The former will be focussed during the review of R N R and lexical PHC constructions. The latter will be the focus within the discussion of GAPPING constructions. We will outline that GAPPING cannot be captured in terms of a universal structural description, the implicit hypothesis of any type of deletion approach, but that the range of admitted GAPPING constructions is subject to language-specific constraints. Viewing GAPPING as a type of S C C , too, in which the focus is on the contrasted constituents—the members of the contrast sets—instead of on the deleted constituents, will create a further argument in favour of a direct phrase structure approach to symmetric coordination. Consistently viewing every symmetric coordination as a type of S C C implies a different view of conjunct structures. We understand a conjunct to be a direct projection of the contrast sets which contain the contrasting constituents. The question of constituency of a conjunct becomes irrelevant. We no longer deal with fully established XP-conjuncts which—if given—bear deletion traces. Only movement traces may occur. Thus, within our approach a conjunct may, but need not necessarily, form a constituent in the conventional sense. It follows directly from this assumption that we deny the adequacy of coordination as a means for testing constituenthood in the hitherto assumed way. Finally, we will view a set of R N R structures of various languages, which will, first, support our assumption of a distinction between S C C and NON-SCC, and, second, render convincing evidence for the appropriateness of a direct phrase structure account of symmetric coordination. We will see that the grammaticality of coordinate structures is simply due to basic syntactic structure. This explains why certain R N R constructions are possible in one language, but impossible in another. Hence, a phrase structure account will prove to be the most adequate for providing the desired universal approach to symmetric coordination.

Chapter 5 German Conjunctions

The acceptability of a given coordinate structure is to a large extent controlled by the chosen conjunction. First of all, the conjunction determines the extent of contrast between the coordinated constituents in question. But the conjunction may also impose structural requirements on the conjuncts. As the conjunct structures vary with the three different coordination types, so vary the respective requirements. This is what we will be concerned with in this chapter.

5.1

Classification of Conjunctions

We distinguish between binarily versus iteratively used conjunctions on the one hand, and initial versus non-initial conjunctions on the other. Within this paradigm German conjunctions are classified as shown in (128). German conjunctions comprise und (and); oder (or); sowie, wie, wie auch (as well as); aber (but); NEG ... sondern (NEG . . .but); weder ... noch (neither . . . nor); sowohl ... als auch (both . . . and); NEG ... noch (NEG . . . nor); entweder ... oder (either . . . or); and nicht nur ... sondern auch (not only . . . but also). (128)

iterative

non-initial und oder sowie, wie, wie auch

binary

(129)

aber

Non-initial

/ iterative

initial weder...noch sowohl . . . als auch NEG . . . n o c h e n t w e d e r . . . oder nicht nur . . . sondern auch NEG . . . sondern

Coordination:

a.

Hans (und) Peter und Maria

b.

Johann (oder) Claus oder Susanne

(130)

Initial / iterative

Coordination:

a.

sowohl Hans (als auch) Peter als auch Maria

b.

entweder Johann (oder) Claus oder Susanne

64 (131)

Non-initial / binary Coordination: sonnig, aber kalt (sunny, but cold)

(132)

Initial / binary Coordination: nicht nur schön, sondern auch intelligent (not only beautiful, but also intelligent)

During the following discussion we will only list binary coordinate constructions—a permissible step, since constraints holding for a binary usage of an iterative conjunction will also be valid for a respective iterative usage. Since we ignore binary versus iterative usage we will hence only be concerned with the distinction between initial versus noninitial conjunctions. The discussion will be organised as follows. First, we will look at PHC in the context of non-initial and initial conjunctions. Second, we will likewise examine R N R constructions, and finally, GAPPING sentences. The usage of non-initial versus initial conjunctions each will be examined first for lexical (minimal and maximal) projections of Ν, V, P, and A, and then for non-lexical, i.e. functional, projections of I and C. Throughout we will investigate phrasal coordination only. Word coordination such as auf- und abladen ('onand download') lies beyond the scope of our investigation. 1

5.2

Phrasal Coordination

5.2.1

Non-initial Conjunctions

Lexical Projections PHC of maximal lexical projections 2 when coordinated by non-initial conjunctions is illustrated by the following examples: 3 (133)

Non-initial PHC of major lexical phrases: a. ΝΡ: [die jungen Jecken] und [die alten Närrinnen] ([the young jesters] and [the old foolish women]) b. ΡΡ: [in der Stadt] und [auf dem Land] ([in the town] and [in the country])

1 In chapter 12, however, we will address the question in which respect the alternative approach we will propose for treating phrasal coordination might be applicable to word-coordination, too. 2 For expository purposes we will refer throughout to an NP structure which traditionally has been assumed to be base generated by a PS rule such as: NP —(Det) (A) Ν This restricted version of NP structure suffices for the time being, since here we merely want to motivate our SCC view on coordination. The discussion of our alternative approach in part III will reveal, however, that a DP-analysis is in fact more favourable. 3 Within the examples the conjuncts in question will be enclosed in brackets: [...].

65 c. AP: [äußerst fröhlich] und [sehr ausgelassen] ([extremely gay] and [very frolicsome]) d. VP: ..., daß sie [Karneval feierten] und [Spaß hatten] ('...that they [carnival celebrated] and [fun had]') Likewise we can coordinate minimal lexical projections: (134)

Non-initial

PHC of lexical heads:

a. N: die einheimischen [Jecken] und [Närrinnen] (the local [jesters] and [(foolish) women]) b. P: [in] und [um] Düsseldorf ([in] and [around] Düsseldorf) c. A: von dem Trubel [begeistert] und [überwältigt] ('about/by the hurly-burly [enthusiastic] and [overwhelmed]') d. V: ..., weil sie den Karnevalszug [beobachteten] oder [begleiteten] ('..., because they the carnival procession [watched] or [accompanied]')

Non-lexical projections With non-lexical (maximal and minimal) projections we get the following non-initial coordination structures: (135)

Non-initial a. IP:

PHC of major functional

phrases:

..., daß [Peter nach Holland fährt] und [Claus nach Berlin fliegt]. ('...that [Peter to Holland drives] and [Claus to Berlin flies].')

b. CP: [Peter fährt nach Holland], aber [Claus fliegt nach Berlin]. ([Peter drives to Holland], but [Claus flies to Berlin].) (136)

Non-initial C:

PHC of functional

heads:

[Obwohl] oder [(gerade) weil] er kommen wird, ist sie nervös. ('[Although] or [(precisely) because] he come will, is she nervous.')

Since we deny a base generation of auxiliaries at the functional I-head node, 4 a discussion of the coordination behaviour of INFL becomes superfluous. Under the assumption of a bearer of TENSE and AGReement features only, INFL simply is not lexically filled. Therefore, only the functional category C is of relevance here.

4 One of the counterarguments is tied precisely to coordination behaviour, revealing that the auxiliary is better assumed to be basegenerated within the verbal projection—cf. chapter 10 for a discussion.

66

5.2.2

Initial Conjunctions

Again, we will look at lexical projections first, and then at non-lexical ones. Lexical Projections (137)

Initial P H C of major lexical phrases:

a. NP: sowohl [die Narren] als auch [die Zuschauer] (both [the fools] and [the spectators]) b. PP: nicht nur [im verschneiten Februar] sondern auch [im sonnigen Mai] (not only [in snowy February] but also [in sunny May]) c. AP: entweder [fantasievoll gekleidet] oder [bunt bemalt] (either [fancifully dressed] or [colourfully painted]) d. VP: ..., daß sie weder [Trübsal bliesen] noch [ihr Kommen bereuten]. ('...that they neither [moped] nor [their coming regretted]'.) (138)

Initial P H C of lexical heads:

a. N: sowohl [Ärger] als auch [Freude] über den Karnevalszug (both [annoyance] and [joy] about/at the carnival procession) b. P:

weder [in] noch [um] Düsseldorf (neither [in] nor [around] Düsseldorf)

c. A:

nicht nur [begeisterte] sondern auch [ablehnende] Mienen (not only [enthusiastic] but also [critical] faces)

d. V: ..., weil sie den Karnevalszug entweder [beobachten] oder [begleiten] wollten. ('...because they the carnival procession either [watch] or [accompany] wanted to'.) Non-Lexical Projections Again, we will only consider the coordination behaviour of C, but not of INFL, for initial PHC of functional heads: (139)

Initial PHC of functional

a. C:

heads:

Entweder [obwohl] oder [(gerade) weil] er kommen wird, ist sie nervös. ('Either [although] or [(precisely) because] he come will, is she nervous.')

(140)

Initial PHC of major functional

a. IP:

phrases:

..., daß entweder [Peter nach Holland fährt] oder [Claus nach Berlin fliegt], ('...that either [Peter to Holland drives] or [Claus to Berlin flies]'.)

b. CP: * Sowohl [Peter fährt nach Holland] als auch [Claus fliegt nach Berlin]. (Both [Peter drives to Holland] and [Claus flies to Berlin].)

67

The examples in (140) reveal an interesting fact: While PHC with initial conjunctions invariably renders acceptable results in case of IP-conjuncts, it yields throughout ungrammatical structures if CP-conjuncts are involved. (141)

Initial P H C of IP:

a.

..., daß sowohl [Hans heute kocht] als auch [Maria den Pizzaservice ruft], ('...that both [Hans today the cooking does] and [Maria the pizza-service calls].')

b.

..., daß weder [Hans heute kocht] noch [Maria den Pizzaservice ruft], ('...that neither [Hans today the cooking does] nor [Maria the pizza-service calls].')

c.

..., daß [Hans heute nicht kocht], noch [Maria den Pizzaservice ruft], ('...that [Hans today the cooking not does] nor [Maria the pizza-service calls].')

d.

..., daß nicht nur [Hans heute kocht], sondern auch [Maria den Pizzaservice ruft]. ('...that not only [Hans today the cooking does], but also [Maria the pizzaservice calls].')

e.

..., daß entweder [Hans heute kocht] oder [Maria den Pizzaservice ruft], ('...that either [Hans today the cooking does] or [Maria the pizza-service calls].')

(142)

Ungrammatical

initial P H C of CP:

a. *

Sowohl [Hans kocht heute], als auch [Maria ruft den Pizzaservice]. (Both [Hans does the cooking today], and [Maria calls the pizza-service].)

b. *

Weder [Hans kocht heute], noch [Maria ruft den Pizzaservice]. (Neither [Hans does the cooking today], nor [Maria calls the pizza-service].)

c. *

[Hans kocht heute nicht], noch [Maria ruft den Pizzaservice]. ([Hans does not the cooking today], nor [Maria calls the pizza-service].)

d. *

Nicht nur [Hans kocht heute], sondern auch [Maria ruft den Pizzaservice]. (Not only [Hans does the cooking today], but also [Maria calls the pizzaservice].)

e. ? * Entweder [Hans kocht heute], oder [Maria ruft den Pizzaservice]. (Either [Hans does the cooking today], or [Maria calls the pizza-service].) The ungrammatical CP coordinations with initial conjunctions will be the subject of the following section.

68

5.2.3

Ungrammatical initial PhC of CP

While in G e r m a n with initial IP coordination t h e conjunction is adjoined to t h e coordin a t e d node, t h e usage of initial conjunctions with C P conjuncts seems to underly a strict V-2 constraint. T h e conjunction particles of an initial conjunction occupy t h e initial field, i.e. t h e Spec-CP position. This blocks t h e filling of this position with a sentential topic, as it is normally t h e case for G e r m a n V-2 sentences. Hence, as d e m o n s t r a t e d in (142), explicit CP-coordination with initial conjunctions is prohibited. If we are faced with a purely structural condition for t h e occupation of t h e SpecC P position, we conjecture t h a t sentential coordination would still be possible. Only t h e conjuncts would not exhibit overt V-2 structure, but V - l structure, with t h e conjunction particles filling t h e Spec-CP position. This is indeed t h e case for weder ... noch (neither . . . nor): (143)

Weder [v_i kocht Hans heute], noch [v_i ruft Maria den Pizzaservice]. (Neither [does Hans t h e cooking today], nor [calls Maria t h e pizza-service].)

T h a t this actually is a coordination of declarative sentences (by default displaying V-2 s t r u c t u r e ) , is supported by t h e fact t h a t t h e analogous question (with overt V - l s t r u c t u r e ) is u n g r a m m a t i c a l : (144) *

Weder [v-i kocht Hans heute], noch [v_i ruft Maria den Pizzaservice]? (Neither [does Hans t h e cooking today], nor [calls Maria t h e pizza-service]?)

In a question t h e Spec-CP position is phonetically empty, but most likely occupied by an abstract W H - o p e r a t o r , which is incompatible with t h e strict V-2 constraint of t h e initial conjunct particles. Non-initial conjunctions do not d e m a n d a specific position, they are simply adjoined to t h e coordinated nodes in t h e final o u t p u t structure. Initial conjunctions are p a r t l y composed of conjunction particles, in which one of t h e m is identical t o a non-initial conjunction, or at least behaves alike. If adjunction applies here, too, we expect an overt V-2 conjunct structure. This indeed can be observed with entweder and nicht

nur ...sondern

conjunction, sondern

... oder (either . . . or)

auch (not only . . . b u t also). Oder (or) is also a non-initial

(auch) (but (also)) behaves like one. Accordingly, we get a V - l / V - 2

combination for t h e conjuncts: (145) a.

Entweder [v_i kocht Hans heute], oder [v-2 Maria r u f t den Pizzaservice]. (Either [does Hans t h e cooking today], or [Maria calls t h e pizza-service].)

b.

Nicht nur [v-i kocht Hans heute], sondern [v-2 Maria ruft (auch [noch]) den Pizzaservice]. (Not only [Hans does t h e cooking today], b u t also [Maria calls t h e pizzaservice] . )

69 For the conjunction NEG ... noch (NEG . . . nor) it is not astonishing that we get a V-2 structure for the first conjunct, since the negation of a declarative sentence will be expressed with the finite verb, but not in isolation in the Spec-CP position. For the following conjuncts, however, we again get V-l structure, since the conjunction particle noch (nor) does not behave like an adjoining non-initial conjunction, but demands the Spec-CP position. (146)

[v-2 Hans kocht heute nicht], noch [v-i ruft Maria den Pizzaservice]. ([Hans does not the cooking today], nor [calls Maria the pizza-service].)

Sowohl .. .als auch (both . . . and) imposes the strictest restrictions. It neither allows for a V-l structure of the conjuncts, nor for any V-l/V-2 combination. (147) a. * Sowohl [v-i kocht Hans heute], als auch [v-i ruft Maria den Pizzaservice]. (Both [does Hans the cooking today], and [calls Maria the pizza-service].) b. * Sowohl [v-2 Hans kocht heute], als auch [v-i ruft Maria den Pizzaservice]. (Both [Hans does the cooking today], and [calls Maria the pizza-service].) c. * Sowohl [y_i kocht Hans heute], als auch [v-2 Maria ruft den Pizzaservice]. (Both [does Hans the cooking today], and [Maria calls the pizza-service].)

The weakest restrictions, on the other hand, are imposed by entweder ... oder (either . . . or). The default realization is a V-l/V-2 combination of conjuncts, as shown in (145a), but a V-2/V-2 combination is on the borderline of acceptability (therefore the questionable judgement in (142e)). Many speakers of German already accept a sentence such as: (148)

Entweder [v-2 Hans kocht heute], oder [v-2 Maria ruft den Pizzaservice]. (Either [Hans does the cooking today], or [Maria calls the pizza-service].)

Entweder (either) here tends to behave like a non-initial conjunction, being adjoined to the coordinated CP node, instead of occupying the Spec-CP position. This concludes our discussion of the behaviour of PHC with respect to the different conjunctions and the different domains of application. Except for the specific conditions with initial CP coordination, PHC does not seem to underly any specific constraints as to the choice of conjunction and the domain of application.

5.3

Right Node Raising

R N R belongs to the class of Non-Constituent Coordination. We speak of non-constituents when a string of constituents cannot be derived to one single constituent. Since the coordination of lexemes automatically means Constituent Coordination, we are, consequently,

70 concerned here with t h e coordination of phrasal levels only. Again we will look at t h e usage of non-initial versus initial conjunctions, each section divided into R N R constructions in maximal lexical and non-lexical projections. 5 5.3.1

Non-Initial Conjunctions

Lexical Projections (149)

Non-initial

R N R in major lexical

phrases:

a. NP: [die jungen] und [die alten] Narren ([the young] and [the old] fools) b. P P : [auf dem großen] und [hinter dem kleinen] Wagen ([on the big] and [behind the small] carriage) c. AP: [so schön wie] aber [intelligenter als] eine Karnevalsfigur ([as beautiful as] but [more intelligent than] a carnival character) d. V P : ..., daß der Karnevalsprinz [den Erwachsenen gute Unterhaltung] und [den Kindern viele b u n t e Bonbons] versprochen hat. ('...that the carnival prince [the adults good entertainment] and [the children many colourful bonbons] promised has.') Non-Lexical Projections (150)

Non-initial

R N R in major functional

phrases:

a. IP: ..., daß [der Karnevalsprinz den Erwachsenen] und [die Karnevalsprinzessin den Kindern] einen herrlichen Umzug versprochen hat. ('...that [the carnival prince the adults] and [the carnival princess the children] a marvellous procession promised has.') b. CP: [Der Karnevalsprinz versprach den Erwachsenen] und [die Karnevalsprinzessin prophezeite den Kindern] einen herrlichen Umzug. ([The carnival prince promised the adults] and [the carnival princess foretold t h e children] a marvellous procession.) 5.3.2

Initial Conjunctions

Lexical Projections (151)

Initial R N R in major lexical

phrases:

a. NP: sowohl [die jungen] als auch [die alten] Narren (both [the young] and [the old] fools)

5 Within the examples the conjuncts are again shown in brackets, with the allegedly raised constituent^) to their right. We hereby follow an RNR analysis, according to which both conjuncts are subject to deletion. Deletion traces, however, will be omitted.

71

b. P P : entweder [auf dem großen] oder [hinter dem kleinen] Wagen (either [on the big] or [behind the small] carriage) c. AP: nicht nur [so schön wie] sondern auch [intelligenter als] eine Karnevalsfigur (not only [as beautiful as] but also [more intelligent than] a carnival character) d. V P : ..., daß der Karnevalsprinz weder [den Erwachsenen gute Unterhaltung] noch [den Kindern viele bunte Bonbons] versprochen hat ('...that the carnival prince neither [the adults good entertainment] nor [the children many colourful bonbons] promised has.') Non-Lexical Projections (152)

Initial R N R in major functional a. IP:

phrases:

..., daß sowohl [der Karnevalsprinz den Erwachsenen] als auch [die Karnevalsprinzessin den Kindern] einen herrlichen Umzug versprochen hat. ('...that both [the carnival prince the adults] and [the carnival princess the children] a marvellous procession promised has.')

b. CP: Entweder [der Karnevalsprinz versprach den Erwachsenen] oder [die Karnevalsprinzessin prophezeite den Kindern] einen herrlichen Umzug. (Either [the carnival prince promised the adults] or [the carnival princess foretold the children] a marvellous procession.)

5.4

Gapping

A binary G A P P I N G coordination consists of a complete sentential first conjunct and a reduced sentential second conjunct, in which the finite verb (and m a y b e f u r t h e r constituents) are missing. Therefore, the domain of rule application is automatically restricted to the non-lexical maximal projections of IP and CP. 5.4.1

Non-Initial Conjunctions

T h e following constructions. 6 (153)

examples

Non-initial

will

contain

both

classical

GAPPING

GAPPING in IP:

a.

..., daß [Hans Spaghetti kocht], und [Maria Lasagne], ('...that [Hans spaghetti makes], and [Maria lasagne].')

b.

..., daß [Hans Spaghetti kocht], oder [Maria Lasagne], ('...that [Hans spaghetti makes], or [Maria lasagne].')

6

Again, brackets indicate the reduced and unreduced, here sentential, conjuncts.

and

STRIPPING

72 Non-initial

(154)

5.4.2

GAPPING

in

CP:

a.

[Hans kocht Spaghetti], und [Maria Lasagne]. ([Hans makes spaghetti], and [Maria lasagne].)

b.

[Hans kocht Spaghetti], oder [Maria Lasagne], ([Hans makes spaghetti], or [Maria lasagne].)

Initial Conjunctions

(155)

Initial

GAPPING

in

IP:

a.

..., daß sowohl [Hans Spaghetti kocht], als auch [Maria], ('...that both [Hans spaghetti makes], and [Maria lasagne].')

b.

..., daß weder [Hans Spaghetti kocht], noch [Maria Lasagne], ('...that neither [Hans spaghetti makes], nor [Maria lasagne].')

c.

..., daß [Hans nicht (keine) Spaghetti kocht], noch [Maria Lasagne], ('...that [Hans not (no) spaghetti makes], nor [Maria lasagne].')

d.

..., daß nicht nur [Hans Spaghetti kocht], sondern auch [Maria], ('...that not only [Hans spaghetti makes], but also [Maria].')

e.

..., daß entweder [Hans Spaghetti kocht], oder [Maria Lasagne], ('...that either [Hans spaghetti makes], or [Maria lasagne].')

(156)

Initial

a.

GAPPING

in

CP:

Sowohl [Hans kocht Spaghetti], als auch [Maria]. (Both [Hans makes spaghetti], and [Maria].)

b. ? Weder [Hans kocht Spaghetti], noch [Maria Lasagne]. (Neither [Hans makes spaghetti], nor [Maria lasagne].) c.

[Hans kocht nicht (keine) Spaghetti], noch [Maria Lasagne]. ('[Hans makes not (no) spaghetti], nor [Maria lasagne].)

d.

Nicht nur [Hans kocht Spaghetti], sondern auch [Maria], (Not only [Hans makes spaghetti], but also [Maria].)

e. ? Entweder [Hans kocht Spaghetti], oder [Maria Lasagne], (Either [Hans makes spaghetti], or [Maria lasagne].) Both with non-initial conjunctions (in IP and CP constructions alike) and with initial conjunctions in IP constructions the G A P P I N G sentences could have been derived from an underlying fully established PHC construction, which here means sentential coordination. But what about initial G A P P I N G in CP? Interestingly, also the constructions as presented in (156) suggest a fully established PHC basis from which they have been derived. However, such an assumption is incompatible with the observation that initial PHC in explicit CP, i.e. V-2, structures renders ungrammatical results throughout (recall the sentences in example (142)).

73 The constructions in (156b) and (156e) are questionable. They become better, however, when we adhere to the structural conditions imposed by the involved initial conjunctions. As shown, the default conjunct structure for weder ... noch (neither . . . nor) and entweder ...oder (either . . . o r ) is V - l / V - 1 and V - l / V - 2 , respectively, when used with C P coordinations. Indeed, the two sentences become perfectly acceptable when construed accordingly: (157) a.

Weder [kocht Hans Spaghetti], noch [Maria Lasagne]. ('Neither [makes Hans spaghetti], nor [Maria lasagne]'.)

b.

Entweder [kocht Hans Spaghetti], oder [Maria Lasagne]. ('Either [makes Hans spaghetti], or [Maria lasagne]'.)

Since the verb within the second conjunct assumedly is deleted, differing positions of the finite verb have no overt effects. The difference in acceptability is only conveyed by the alternative structure of the respective first conjuncts, here V-l instead of the former V-2 structure. 7 Nevertheless, though V-l is preferable, a V-2 structure for the first conjuncts is not completely unacceptable. But within the deletion paradigm it is problematic to explain the derivation of these overt GAPPING sentences, since no corresponding grammatical source in terms of a fully established P H C construction is available for them. Another difficulty is contained in the GAPPING sentences in (156a) and (156d). They represent overt STRIPPING constructions, which means that within the second conjunct the entire verb phrase instead of merely the finite verb is deleted, leaving behind one remnant only. Hence, we assume the following underlying structural bases: (158) a.

Sowohl [Hans kocht Spaghetti], als auch [Maria kocht Spaghetti], (Both [Hans makes spaghetti], and [Maria makes spaghetti].)

b.

Nicht nur [Hans kocht Spaghetti], sondern auch [Maria kocht Spaghetti]. ('Not only [Hans makes spaghetti], but also [Maria makes spaghetti].)

First, it is surprising t h a t here we do have regular V-2 conjuncts with initial conjunctions which either do not allow for overt C P coordination at all (sowohl .. .als auch—both . . . a n d ) , or call for a different conjunct structure—in the case of nicht nur ...sondern auch (not only . . . b u t also) for a V - l / V - 2 combination of conjuncts. But given this fact, it is, second, surprising that no corresponding classical GAPPING construction seems to be available here: 8 7 Also for the coordination in (156c) it is most likely that, instead of having a symmetric CP (i.e. V-2) source, the underlying structure rather conforms to the requirements of the initial conjunction ΝEG ... noch (NEG .. .nor), which calls for a V-2/V-1 combination. Only, that with the finite verb being deleted the overt structure is ambiguous with respect to either base. 8 However, when we apply the default conjunct structures required by nxchi nur ... sondern auch (not only ...but also) in the context of CP coordination (i.e. V-l/V-2), we may get a GAPPING surface structure:

74 (159) a. * Sowohl [Hans kocht Spaghetti] als auch [Maria Lasagne]. (Both [Hans makes spaghetti], and [Maria lasagne].) b. * Nicht nur [Hans kocht Spaghetti], sondern [Maria (auch [noch]) Lasagne]. ('Not only [Hans makes spaghetti], but [Maria (also [additionally]) lasagne].)

Both of these facts seem to suggest that, in fact, the coordinations in (156a) and (156d) are not due to deletion, but rather to an operation which has become known under the term CONJUNCT MOVEMENT (cf., e.g., Höhle [Höhle, 1983]). With this rule it is assumed that the second conjunct of a fully established PHC of NP's is moved to the end of the sentence. Hence, no recourse is taken to deletion. The effects of this rule are shown in (160):

(160) a.

[jvf> Sowohl [Hans] als auch [Maria]] kochen Spaghetti. ([¿vf> Both [Hans] and [Maria]] make spaghetti.)

Sowohl [Hans] kocht Spaghetti, als auch [Maria], (Both [Hans] makes spaghetti, and [Maria].) b.

[jvp Nicht nur [Hans] sondern auch [Maria]] kocht Spaghetti. ([ΛΓΡ Not only [Hans] but also [Maria]] makes spaghetti.)

Nicht nur [Hans] kocht Spaghetti, sondern auch [Maria]. (Not only [Hans] makes spaghetti, but also [Maria].) One small argument against CONJUNCT MOVEMENT and in favour of a deletion approach may lie in the fact that the former is necessarily coupled with an alternation of AGReement features at least in the case of sowohl ... als auch (both . . . and). While the source contains plural agreement, the target structure needs to have singular agreement— a change which is not necessary when we assume the basic structure to be as in (158a), and subsequently apply deletion (under identity). But if we abandon CONJUNCT MOVEMENT and return to a deletion approach, we again have to ask the question about the general availability of corresponding fully established CP coordinations with initial conjunctions. Thus, the examples of this section may provide further evidence against the deletion paradigm and in favour of an alternative treatment of so-called Non-Constituent Coordination.

? Nicht nur [kocht Hans Spaghetti], sondern [Maria (auch [noch]) Lasagne]. ('Not only [makes Hans spaghetti], but [Maria (also [additionally]) lasagne]. But this raises the question why we should have different CP conjunct structures with this initial conjunction.

Chapter 6 The Domain of Coordination

In this chapter we will lay out our motivation for viewing non-maximal PHC, R N R , and GAPPING as b e l o n g i n g t o t h e d i s t i n c t class of SHARED CONSTITUENT COORDINATION.

One of the main issues we will address is the question of what should be regarded as the domain of a coordination construction. For instance, is it actually justified to assume lexical P H C of adjectives for a construction such as (161)? (161)

Er ist seiner Frau sowohl [4 treu] als auch [A überdrüssig]. ('He is his wife (dat/gen) both [loyal to] as well as [tired of].')

Or is it more appropriate to think of this construction in terms of a coordination of maximal adjectival projections, which includes the (shared) complement seiner Frau? We will argue in favour of the latter alternative, which assumes that a coordination structure always ranges over maximal phrases. We will see that any coordination structure is licensed by the existence of multiple contrasting heads, but that its range is never limited to these heads. The relevant complements and specifiers within their maximal projection cannot be chosen arbitrarily, because they have to comply uniformly with the requirements of each of the lexical heads. Therefore, we will conclude that the domain of a coordination structure always is the maximal projection of its "coordinated" heads. The assumption of a uniform phrasal domain obviates the coordination type of lexical PHC. What we are left with is, on the one hand, PHC of fully established XP's (NONS C C ) , and, on the other hand, coordination structures which range over XP's and involve a sharing of one or more constituents somewhere within the phrasal structure ( S C C ) . In the following sections we will leave aside uncontroversial NON-SCC constructions like, e.g., the P P coordination (162)

(ein Papier) [pp auf einer interessanten Konferenz] oder [pp in einer renommierten Zeitschrift] ((a paper) [at an interesting conference] or [in a renowned journal])

and concentrate on peripheral S C C , which we take to comprise RIGHT S C C and LEFT S C C . Subsequently, we will turn to GAPPING, or MEDIAL SCC, structures.

76

6.1

Right SCC

Recall the initial R N R constructions in (151), which are repeated here as (163): (163)

Initial R N R in major lexical phrases: a. NP: sowohl [die jungen] als auch [die alten] Narren (both [the young] and [the old] fools) b. PP: entweder [auf dem großen] oder [hinter dem kleinen] Wagen (either [on the big] or [behind the small] carriage) c. AP: nicht nur [so schön wie] sondern auch [intelligenter als] eine Karnevalsfigur (not only [as beautiful as] but also [more intelligent than] a carnival character) d. VP: ..., daß der Karnevalskönig weder [den Erwachsenen gute Unterhaltung] noch [den Kindern viele bunte Bonbons] versprochen hat ('...that the carnival king neither [the adults good entertainment] nor [the children many colourful bonbons] promised has.')

The phrases in (164) present slightly modified versions: (164) a.

sowohl [junge] als auch [alte] Narren (both [young] and [old] fools)

b.

entweder [auf] oder [hinter] dem kleinen Wagen (either [on] or [behind] the small carriage)

c.

nicht nur [schöner] sondern auch [intelligenter] als eine Karnevalsfigur (not only [as beautiful as] but also [more intelligent than] a carnival character)

d.

..., daß der Karnevalskönig bunte Bonbons weder [gegessen] noch [geworfen] hat ('...that the carnival king colourful bonbons neither [eaten] nor [thrown] has.')

Superficially, it looks as if we were faced here with simple constituent coordination of lexical heads of the PHC type. The example in (164a) contains a coordination of adjectives, the one in (164b) coordinated prepositions, (164c) again contains coordinated (comparative) adjectives, and (164d) coordinated non-finite verbs. So, do the sentences in (164) represent genuine instances of lexical PHC? Or are they somehow instances of the same type of coordinate structures as their obviously similar counterparts in (163)? The distinguishing characteristic of R N R is that we have rightmost constituents which bear the same relationship to both of the conjuncts. Let us say that these constituents are shared by the respective conjuncts. We already mentioned the crucial constraint of phonetic identity for the raised, i.e. shared, constituent(s). This means that the grammatical

77 requirements imposed by each of the conjuncts have to result in one unique surface structure form for the shared element, otherwise no valid R N R construction will be obtained. 1 In this respect, the examples in (164a-d) clearly represent instances of the R N R type, which, following the considerations outlined above, is more adequately termed SHARED CONSTITUENT COORDINATION (SCC). Thus, we regard the noun Narren in (164a) as the shared constituent, in (164b) the dative NP dem kleinen Wagen, in (164c) the comparative phrase als eine Karnevalsfigur, and in (164d) the finite auxiliary hat. Consequently, instead of being lexical coordinations these constructions amount to coordinations over the phrasal levels of NP, PP, AP, and VP, respectively—in accordance with their analogues in (163). 6.1.1

Against lexical P h C

Lexical PHC is subsumed as a special case under phrasal PHC, which is taken to be generated by the familiar phrase structure rule: X —>

X

Conj

X

The variable X may be instantiated by any type of constituent, be it phrasal or lexical. Furthermore, we assume that the resulting mother constituent inherits the syntactic type of its daughters. Accordingly, it is predicted that the coordinate node distributes like any of its simplex daughter nodes. When restricted to the syntactic type of the affected cons t i t u e n t s ) this seems to be a valid generalization. 2 But what about, for example, case assigning criteria? It is not prohibited to coordinate two lexemes which assign different cases, but of what type should the coordinate node be? Take, for example, the two adjectives mentioned already: treu (loyal to) and überdrüssig (tired of). The first assigns dative, the second genitive case. (165)

A-[???] —

A-[dat]

Conj

A-[gen]

It is desirable that also with respect to case assignment the resulting node should be of a unique type, since we want to be able to say that it distributes like a simplex node. In case of dative and genitive this problem could be circumvented by generalizing the two cases to a common oblique case. But if we deal with dative versus accusative case assignment such a solution is not so easily available. Consider the two prepositions mit (dat) and ohne (acc) in the following constructions: (166) a.

mit(da 2: (Xj, X2 ..., X m ) ii. Factors are sequences of length one: (Y)

The exact relationship between Factors and Ordered Lists will be made more precise as we proceed. With the introduction of these higher order structures, the term terminal node becomes somewhat awkward, since it is normally used to refer to atomic values, not lists. But since we will continue to use tree structures representations for any type of phrase, we will stick to the term terminal node, referring to a specific position in the tree. We propose the following extended X-scheme: (219) Coordinate X-scheme: X[coord] [coord] [coord]

( X l v i

Xm)

(Compl l v .., Cornpl m ) —" (Spec l v .., Spec m ) X'[coori] *

^ [ c o o t j¡

1

This extended X-scheme generates conflated trees. A conflated tree Γ consists of m simplex structure trees T'. Therefore it is essential that sequences are ordered: it is necessary for determining the simplex trees that Τ contains. Given that m is equal for each base generated sequence the simplex structure trees T ' that Τ contains are defined as follows: (220)

i. Each sequence 5 i n T i s replaced with the i t h element, with 1 < i < m. ii. A sequence of length one is part of each of the simplex structures 7". iii. Condition: Each simplex structure tree T ' must be well-formed with respect to the grammatical requirements of each level of syntactic representation.

104 A coordinate structure is licensed by an obligatory Ordered List (i.e. sequence of length > 2) of heads. With this condition we account for the fact that any coordination is head-driven, being licensed by multiple heads. Complements and specifiers, on the other hand, may be either generated as Ordered Lists, as Factors, or recursively as both Ordered Lists and Factors, the latter allowing for a sharing of constituents across embedded structures. As it stands, the extended X-scheme in (219) is designed for coordinate structures only. With this X-scheme we require an obligatory Ordered List of heads. If, however, we weaken this strict condition, we open the way to a general X-scheme, which is valid for simplex and coordinate structures alike: (221) Generalized X-scheme: X

—•

X' X"

—• X (Compii,..., Compl m ) -» (Speci,..., Spec m ) X'

(Xlv-5

Xm)

In this generalized X-scheme we uniformly assume an extension into sequences at every projection level. The sequences are of arbitrary length, although they all must have the same length. We then postulate that within this X-scheme a head may be instantiated as either (222a) or (222b): (222) a.

X :=

X[ 3lmp ( el ]

b.

X :=

X[coorii]

For an instantiation via (222a) we require that all of the sequences provided for in the X-scheme in (221) be set to length one. In our terms this means that we generate a unique Factor at each level of projection. (223) Simplex X-scheme: X[stmp/ex] X'[«mplex]

X" [simplex]

' —»

(X)

X [simp/ex] (Spec)

(Compi) X'[s,mp/ex]

This X-scheme is simply the conventional X-scheme for simplex structures. We will take this instantiation to be the default one, and will, therefore, omit the feature specification [simplex] in the following. On the other hand, for an instantiation via (222b) we require that the head position be filled by a sequence of length > 2, i.e., an Ordered List. Complements and specifiers, as we mentioned above, may either be Ordered Lists or Factors, or even consist of both Ordered Lists and Factors. Such an instantiation will amount to the coordinate X-scheme as represented in (219).

105 The full range of possible phrases which we provide for by the generalized X-scheme can summarized as follows: At one end of the scale we get simplex structures, in which only Factors are involved; one step up we have Shared Constituent Coordination, which requires Ordered Lists as well as Factors; and finally at the opposite end of the scale there is Non-Shared Constituent Coordination, which involves Ordered Lists only. For the remainder of this study we will leave aside familiar simplex structures and be concerned with coordinate structures only, viz. Shared and Non-Shared Constituent Coordination. The coordinate X-scheme we propose suffices to account for the complexity of coordinate structures. In including Ordered Lists and Factors as terminal nodes we augment the hitherto two-dimensional tree structure with a third dimension. This "threedimensionality" is much in the spirit of Goodall's three-dimensional approach (cf. section 3.5.1). The crucial difference is that we directly embed this third dimension into the X-scheme, whereas Goodall proposes a sophisticated phrase structure component which is completely independent from the phrase structure theory for simplex sentences. Furthermore, with our coordinate X-scheme we provide for a direct generation of factorized syntactic structures in the sense of Williams (cf. section 3.3). Our Ordered Lists and Factors may result in, and thus be equivalent to, Williams' factorized sections. The same holds for the representation as suggested by Erteschik-Shir (cf. section 3.5.2), which follows Williams' ATB-format. However, factors as understood by Williams and Erteschik-Shir are not to be confused with our notion of factors. By a Factor occurring in a coordinate structure we mean a unique base generated constituent within a given lexical or functional projection, which by definition has a multiple character. At best, there is a superficial similarity to what Williams calls a "simultaneous factor", in which two or more identical occurrences of a given constituent may be merged into one instance, if this factor is to undergo movement. In our approach, however, Factors always are unique, regardless whether they are moved or not.

7.1

The Abstract Coordination Lexeme

Since, as we showed, any type of conjunction ranges over an entire phrase, we will generate the conjunction with the maximal projection. However, like Ross [Ross,1967b], we will not generate the actual conjunction, but an abstract COORD lexeme. This abstract COORD lexeme will be adjoined to the maximal coordinate projection. So we get the following phrase structure rule: (224)

X"

-

COORD ( c o n j ]

X" [ c o o r i ]

In a given coordination, the abstract COORD lexeme will be specified for the chosen conjunction by a specific feature, e.g., COORD[onli]. Further features, encoding structural constraints holding for particular conjunctions (as, e.g., outlined for initial conjunctions in German in section 5.2.3), could be added. Thus, these features parametrize the general phrase structure rule in (224). The advantage of this set up is that we do not have to

106 modify the basic X-scheme responsible for coordinate structures and may continue to regard them as the general phrase structure rules, irrespective of the conjunction chosen. We assume that the specifying conjunction feature is always filled and thus exclude asyndetic coordination. This position is based on the observation that scope assignment differs between coordination with conjunctions and coordination without conjunctions— viz. asyndetic coordination—in which the conjuncts are simply juxtaposed. Note the difference in the following two constructions (cf. Lang [Lang,1991, p.615]): (225)

Die Sonne scheint, die Vögel singen nicht. ('The sun shines, the birds sing not.')

versus (226)

Die Sonne scheint, und die Vögel singen nicht. ('The sun shines, and the birds sing not.')

The negation particle nicht (not) may have scope over the entire preceding coordinate complex in (226), while this is excluded in (225).

7.2

From D-structure to S-structure

The effects of the generalized X-scheme on D-structure and S-structure shall be discussed, first, for a number of constructions in which no movement is involved and, secondly, for constructions where we do observe movement. First, consider the NON-SCC structure in (227): (227)

[auf dem Tisch] und [unter dem Teppich] ([on the table] and [under the carpet])

Here, no constituent is shared, and therefore we have to deal with Ordered Lists only. According to the head-drivenness condition this prepositional coordination is licensed by multiple prepositional heads, in this case auf, unter (on, under). Via the coordinate X-scheme and the conjunction adjunction rule we generate the following phrase structure tree: 1

1

Ordered Lists are given in curly braces, Factors in future examples will be delimited by vertical lines.

107 (228)

PP

\

COORD [und]

Ρ P\coord\ p> [coord\

f

P\coord] auf auf 1

^ unter J

NP[coord\ Det[COOTIq

N[coorq

R D. Ι

R Tisch

[ d_ J

I Teppich J

Ι

I on l i the(.) 1 Í table Ì I under J \ the(_) J | carpet J Before discussing this phrase structure representation, a remark on a non-structural aspect need to be made. As can be seen from this NON-SCC type coordination, not all of the simple contrast sets need to be directly contrastive. The simple contrast set of determiners, which are not yet specified for their appropriate case, surely does not convey any contrast. The essential condition of contrast, a necessary condition for every coordinate structure, is not claimed to hold for each Ordered List, but rather for the complex contrast set of the superordinate coordinate projection (here PP). Its elements must be contrasting with respect to the given context. However, with the question of contrast not being a matter of syntax, verifying whether or not the condition of contrast is satisfied is a matter that lies outside the realm of a purely syntactic description of coordinate structures, with which we are concerned here. It is important that a given coordinate structure has to comply with the condition that each simplex structure tree contained in the conflated tree structure be well-formed according to the grammatical requirements on the various syntactic levels of representation. In (228) we have the D-structure representation of a PP-coordination. Given this structure, it must be guaranteed that linewise the prepositional heads can assign the required (lexical) case to their prepositional objects. In (228) this condition is met: auf (on) may assign dative case to its complement