The Syntax of Subordination [Reprint 2011 ed.] 9783110922134, 9783484303737

This study is concerned with the categorial status of subordinating conjunctions and the internal and external structure

186 65 39MB

English Pages 285 [288] Year 1997

Report DMCA / Copyright

DOWNLOAD FILE

Polecaj historie

The Syntax of Subordination [Reprint 2011 ed.]
 9783110922134, 9783484303737

Table of contents :
Acknowledgements
Introduction
Part I Early Generative Grammar, the Rise of Category C and the Categorization of Subordinating Conjunctions
1 Subordination in Early Generative Grammar – Recursiveness
1.1 Phrase Structure
1.2 Summary
2 Complementizers
2.1 Complementizers: Transformations and Phrase Structure Rules
2.2 Towards Uniform Sentence Structure
2.3 Summary
3 Subordinating Conjunctions – A Question of Categorization
3.1 Subordinating Conjunctions in Traditional Grammar
3.2 Subordinating Conjunctions as Complementizers
3.3 Subordinating Conjunctions as Prepositions
3.4 Summary
Part II Recent Conceptions of Phrase Structure
4 Phrase Structure
4.1 From Rules to Representations – Setting the Theoretical Background
4.2 The Inventory of Syntactic Categories
4.3 Summary
5 Clause Structure
5.1 Split-Infl
5.2 Arguments for Reordering T and Agr and for Two More Functional Heads between C and VP
5.3 Movement in Matrix Clauses – A Typology of Movement
5.4 Summary
Part III Lexical Properties of Complementizers, Prepositions and Subordinating Conjunctions
6 Lexical Properties of Complementizers and Properties of Category C
6.1 Complementizers
6.2 That
6.3 If
6.4 Whether
6.5 For
6.6 Summary
7 Lexical Properties of Prepositions
7.1 Lexical Prepositions and Their Projections
7.2 Summary
8 Subordinating Conjunctions Straddle the Dividing Line between the Lexical and the Functional Universe
8.1 Subordinating Conjunctions and Their Complement
8.2 The Specifier Position of SubconP
8.3 The Functional Architecture Dominating SubconP
8.4 Summary
9 Conclusion
10 Bibliography

Citation preview

Linguistische Arbeiten

373

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

Dagmar Haumann

The Syntax of Subordination

Max Niemeyer Verlag Tübingen 1997

Die Deutsche Bibliothek - CIP-Einheitsaufnahme Haumann, Dagmar: The syntax of subordination / Dagmar Haumann. - Tübingen : Niemeyer, 1997 (Linguistische Arbeiten ; 373) ISBN 3-484-30373-5

ISSN 0344-6727

D 468

© Max Niemeyer Verlag GmbH & Co. KG, Tübingen 1997 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. Gedruckt auf alterungsbeständigem Papier. Druck: Weihert-Druck GmbH, Darmstadt Buchbinden Industriebuchbinderei Hugo Nadele, Nehren

Contents Acknowledgements

Vffl

Introduction

1

Parti Early Generative Grammar, the Rise of Category C and the Categorization of Subordinating Conjunctions

5

1 1.1

Subordination in Early Generative Grammar - Recursiveness Phrase Structure 1.1.1 Noun Clauses 1.1.2 Adverbial) Clauses Summary

7 7 10 18 22

2.3

Complementizers Complementizers: Transformations and Phrase Structure Rules 2.1.1 Transformationalist Approaches to Complementizers 2.1.2 Complementizers as Deep Structure Nodes Towards Uniform Sentence Structure 2.2.1 Matrix Clauses are S's 2.2.1.1 Matrix wh-questions are S's 2.2.1.2 Matrix Declaratives are S's Summary

25 25 26 32 38 39 39 42 43

3 3.1 3.2 3.3 3.4

Subordinating Conjunctions - A Question of Categorization Subordinating Conjunctions in Traditional Grammar Subordinating Conjunctions as Complementizers Subordinating Conjunctions as Prepositions Summary

45 45 49 54 61

1.2 2 2.1

2.2

Part Π Recent Conceptions of Phrase Structure 4 4.1 4.2

Phrase Structure From Rules to Representations - Setting the Theoretical Background The Inventory of Syntactic Categories Adapting Minor Categories to the X-bar Format - An Aside 4.2.1 Lexical Categories ,

63 65 65 69 69 75

VI

4.3

4.2.2 Functional Categories Summary

82 89

5 Clause Structure 5.1 Split-Infl 5.2 Arguments for Reordering T and Agr and for Two More Functional Heads between C and VP 5.2.1 Arguments for Reordering T and Agr 5.2.2 Object Agreement (AgrO) 5.3 Movement in Matrix Clauses - A Typology of Movement 5.4 Summary

91 91 99 100 Ill 118 142

Part ΙΠ Lexical Properties of Complementizers, Prepositions and Subordinating Conjunctions

145

6 Lexical Properties of Complementizers and Properties of Category C 6.1 Complementizers 6.2 That 6.2.1That/w^/ca//ve) 6.2.2 Th&tsubfanctive) 6.3 If 6.4 Whether 6.5 For 6.6 Summary

147 147 150 150

7 Lexical Properties of Prepositions 7.1 Lexical Prepositions and Their Projections 7.1.1 Argument Structure, Case-properties and the Structure of PP 7.1.2 Functional Structure Dominating PP 7.1.2.1 The Internal Argument, Structural Case and Object Agreement 7.1.2.2 The Referential Argument, Degrees and Theta-Binding 7.1.2.3 The External Landing She for the Internal Subject 7.2 Summary

190 192 192 201 201 213 227 232

8 8.1

8.2 8.3 8.4

Subordinating Conjunctions Straddle the Dividing Line between the Lexical and the Functional Universe Subordinating Conjunctions and Their Complement 8.1.1 Subcon is a Functional Head within an Extended Projection 8.1.2 Subcon is not a Lexical Head Theta-marking its Complement The Specifier Position of SubconP The Functional Architecture Dominating SubconP Summary

163

171 174 178 187

236 237 237 247 251 254 260

νπ 9

Conclusion

262

10

Bibliography

271

Acknowledgements

I wish to take the opportunity of thanking a number of people who - in various ways contributed to this work. I wish to thank Gisa Rauh for encouraging me to take on this project, for her competent guidance and for the wisdom to let me have it my own way. Many thanks go to my colleagues within and outside the research project "Grammatische Eigenschaften englischer Präpositionen, Konjunktionen und Adverbien im Lexikon" for providing me with the opportunity to discuss parts of my work at various stages. I am grateful to Thilo Tappe and Friederike Pfingsten for many helpful discussions and critical comments. Thanks are also due Dieter Wolff for the helpful comments on the final draft. I am indebted to Colin Foskett for proofreading the final draft. Many thanks go to Jennifer Austin, Timothy Bell, Colin Foskett, John CHedhill and John McKeown for constantly being willing to check and discuss the data.

Introduction Recent developments in generative linguistic theory1 - such as the division of syntactic categories into lexical and functional categories, the rise of numerous functional projections dominating lexical ones and strict licensing conditions for head-complement and head-specifier relations within projections of lexical and functional heads - inevitably lead to a reconsideration of the categorial status of subordinating conjunctions and subordinate clauses. This discussion has been going on for roughly 20 years.2 The introduction of a special deep structure node for subordinating elements3, referred to as complementizer, paved the way for a uniform conception of clauses, which, in essence, predicts that clauses, be they matrix clauses or embedded clauses, are structurally uniform and thus of the same category, S', in early generative grammar, or CP, in recent models. The discussion of the categoryhood of subordinating conjunctions was initiated by Jackendoff (1973,1977) who - following Klima (1965) - argues that subordinating conjunctions such as after, before, since, etc., as opposed to that and //, should be categorized as prepositions because these elements are semantically identical to prepositions and because the subordinate clauses introduced by these elements pattern distributionalry like PPs, i.e. like phrases introduced by a preposition governing a nominal constituent. Hendrick (1976) and Emonds (1976,1985) take a more radical perspective than Jackendoff with respect to categorizing subordinating conjunctions. Whereas Hendrick (1976) treats all subordinators on a par as complementizers, Emonds (1976,1985) classifies them as prepositions. These three categorization options still prevail in recent linguistic theory: subordinating conjunctions are complementizers, subordinating conjunctions are prepositions and subordinating conjunctions are divided between these two categories. Whereas the functional categoryhood of complementizers is beyond controversy, the categoryhood of prepositions is not, as the elements constituting this class display anything but homogeneous syntactic behavior. Prepositions which in one of then* functions serve to introduce subordinate clauses, however, can unambiguously be characterized as figuring within the universe of lexical categories. Thus, subordinate clauses are headed either by a complementizer or by a preposition. While the structure of CP, which is considered an extended verbal projection4, is well studied, the structure of prepositional projections has been neglected, i.e. there are elaborate structures for extended verbal projections, e.g. CPs, and rather simple ones for prepositional projections.5 1 The generative models referred to center around the Principles and Parameters Theory as defined in Chomsky (1986a,b) and the Minimalist Program as developed in Chomsky (1992,1993,1994,1995). 2 Cf. Freidin (1971), Hendrick (1976), Jackendoff (1977), Emonds (1976,1985), Reuland (1979), Lasnik/ Saito (1992), Grimshaw (1991), van Riemsdijk (1994), Vikner (1994a,b,1995), Larson (1990), Dubinsky/ Williams (1995). 3 Cf. Bresnan (1970,1972,1973). 4 Cf. Grimshaw (1991). See Haider (1988) and van Riemsdijk (1990) for related concepts. 5 But see Zwarts (1992), Rooryck (1993) and Koopman (1993) for functional projections dominating PP.

Another disparity concerns the analyses of argument versus adjunct clauses in terms of licensing.6 Sentential elements that occur as arguments automatically fulfill the licensing requirement: they are licensed arguments by virtue of being theta-marked entities within the domain of the governing lexical head. It is far from clear how non-theta-marked sentential elements, Le. adjunct clauses, are licensed within a complex structure and what the principles are that govern their distribution. These two weak-spots, i.e. categorization and licensing, which at first glance appear to be disparate, make h worthwhile reconsidering the status of subordinate clauses and taking them as a touchstone for the syntax of subordination, which is what this study attempts to explore. This study falls into three parts: Part I: Early Generative Grammar, the Rise of Category C and the Categorization of Subordinating Conjunctions Chapter I summarizes the conception of phrase structure in early generative grammar taking into account the heritage of traditional grammar with respect to the classification of subordinate clauses. The central issue is the distribution of noun clauses and adverbial clauses which is accounted for by phrase structure rules introducing these elements as deep structure elements. Complementizers are central to chapter 2. Two traditional analyses of complementizers will be reviewed: the transformational approach and the phrase structure rule analysis which became the standard analysis for both subordinate and matrix clauses. In chapter 3,1 discuss the categorization options for subordinating conjunctions both from a traditional grammar point of view and from the perspective taken in generative grammar according to which subordinating conjunctions are either complementizers or prepositions. Part Π: Recent Conceptions of Phrase Structure Chapter 4 starts out with the conceptual shift from rule-driven to representation-based systems such as the Principles and Parameters Theory and the Minimalist Program. Central to this chapter is the inventory of syntactic categories, with the focus on the distinction between lexical and functional heads. In chapter 5, taking the "split-Infl"-Hypothesis and subsequent refinements as a starting point, I discuss clause structure on a larger scale. Central to this discussion is movement and the A- vs. A'-dichotomy for both XP- and head movement. Part ffl: Lexical Properties of Complementizers, Prepositions and Subordinating Conjunctions Chapter 6 is concerned with the lexical properties of complementizers, licensing of CP and licensing of the head-complement and head-specifier relation within CP. It will be argued that the licensing of both complement and specifier within CP (as well as movement within and out of CP) is based entirely on feature checking mechanisms. 6

The central issue in the discussion of these two types of clauses is the asymmetry these constructs display with respect to extraction.

In chapter 7, lexical properties of prepositions will be the central issue. It will be shown that all licensing, Le. licensing of the head-complement and the head-specifier relation within PP is based on theta-marking. Under the analysis developed in this chapter, matrix clauses are conceived as external arguments. It will be argued that PP is dominated by functional architecture and that the functional projections dominating PP are licensed on the basis of argument structure and Case-properties. Chapters 6 and 7 will be taken as the background against which the analysis of subordinating conjunctions that cannot be categorized as complementizers or prepositions is set. In chapter 8,1 will argue that there is a third catergory, Subcon, which conflates properties of complementizers and prepositions. It will be argued that the head-complement relation is licensed by feature checking mechanisms and that the head-specifier relation is licensed by theta-marking, Le. the matrix clause counts as the external argument. In the concluding chapter (chapter 9\ I will briefly discuss the derivation of the relative order of subordinate and matrix clauses.

Parti Early Generative Grammar, the Rise of Category C and the Categorization of Subordinating Conjunctions The first part of this study centers around three interrelated aspects of subordination in early generative grammar: the distribution of subordinate clauses, the structure of subordinate clauses and the category of subordinating conjunctions. As early generative grammar in its classification of subordinate clauses draws on notions provided by traditional grammar, each chapter contains a brief overview of the description of subordination from a traditional grammar point of view. Starting out with a sketch of the conception of phrase structure in early generative grammar, Chapter 1 focusses on phrase structure rules and recursiveness, i.e. on rules that allow for sentence embedding and that specify the distribution of subordinate clauses. Chapter 2 is concerned with the form of subordinate clauses. In this chapter two approaches to complementizers will be reviewed: the transformational approach and the phrase structure rule approach. The comparison of these two approaches focusses on the arguments which led to preferring the phrase structure rule approach according to which sentence structure is uniform, ie. there is only one phrase structure rule which generates all types of sentences. Chapter 3, finally, is concerned with the categorization of subordinating conjunctions, with the focus on adverbial subordinating conjunctions, ie. elements that do not hi any obvious way correspond to complementizers as characterized in chapter 2. Central to the discussion are elements such as after, although, before, because, since, until, in case, while, etc. which - in the relevant literature - are categorized as either complementizers or as prepositions.

l

Subordination in Early Generative Grammar An Instance of Recursiveness

In this chapter I present an overview of the syntactic analyses of two types of subordinate clauses in early generative grammar: noun clauses and adverbial clauses. Since early generative grammar, in its classification of subordinate clauses, draws on some of the notions of traditional grammar, some discussion of these will be necessary as well m the first part of this chapter a brief summary of the overall conception of phrase structure in early generative grammar is presented. The aspect of phrase structure which is highlighted is recursiveness. The distribution and functions of subordinate clauses are sketched from a traditional grammar point of view. The various classifications of subordinate clauses as conceived in traditional grammar serve as the background against which the analyses in early generative grammar are set. Noun clauses, Le. subordinate clauses which - according to traditional grammar - take on the syntactic functions of subject and object are discussed in 1.1.1, adverbial clauses are central to 1.1.2. 1.2 summarizes the the treatment of subordinate clauses in early generative grammar.

1.1 Phrase Structure Phrase structure in early generative grammar is determined by the categorial component and the lexical component (lexicon), which make up the base component, one of the two parts of the syntactic component. The categorial component consists of phrase structure rules which apply in a context-free fashion to generate phrase markers and whose function h is to "[...] define the system of grammatical relations and to determine the ordering of elements in deep structure." (Chomsky (1965/161990:123)). Besides the lexical hems (formatives) of a given language with all their phonological, semantic and syntactic properties specified, the lexicon contains rules which are associated with a lexical hem. These rules are referred to as subcategorization rules. The set of subcategorization rules is further subdivided into context-free and context-sensitive rules which introduce inherent features and contextual features respectively. The set of context-sensitive rules is again divided into selectional rules and strict subcategorization rules. From the set of context-sensitive rules only strict subcategorization rules are relevant for the present purposes since they specify the categorial contexts into which a lexical hem may be inserted. As already mentioned, phrase structure rules generate phrase markers in a context-free fashion. Lexical insertion rules apply to the terminal symbols of a phrase marker, replacing them with lexical formatives. Lexical insertion is restricted by subcategorization rules which, as already mentioned, specify the categorial contexts into which a given lexical hem may be inserted. The resulting structures, the so-called deep structures, serve as the input for the second part of the syntactic component, the transformational component.

Transformational rules, which are considered to be component-specific rules, map these input structures onto surface structures. Transformational rules are extrinsicalry ordered1 and hence obey the principle of cyclicity. Since each transformation is fully defined, transformational rules define the range of their own applicability; in this respect transformational rules can be considered filters for deep structure inputs (c£ Chomsky (1965/161990:138f£), Rosenbaum (1967:25)). Beside the syntactic component, which is considered to be autonomous, the model contains a semantic and a phonological component, both of which are merely interpretive.2 In this grammatical system, as well as k those to follow, it is held that main clauses and subordinate clauses do not differ structurally at the level of deep structure as there is only one set of rules that generates both mam clauses and subordinate clauses.3 Rosenbaum (1967:25), for example, states that "[...] there is no structure underlying any declarative sentence in English that cannot, in some other derivation, be the structure underlying a predicate complement sentence.", where predicate complements are complements, Le. strictly subcategorized elements, with sentential structure. Any superficial difference between these two structures is assumed to be derivational, Le. the result of transformational rules applying to deep structures. The deep structure of any sentence, main clause or subordinate clause, is derived by application of phrase structure rules such as those in (1): (1) a. S-> NP Λ Predicate Phrase b. Predicate Phrase-» Aux Λ W c. VP -»V... (adapted from Chomsky (1965/161990:102)) Since the set of structure building rules is necessarily finite, recursiveness of rule application is needed to guarantee that an infinite set of structures can be generated. Recursiveness must thus be encoded in the phrase structure rules of the base component.4 This recursive property of the phrase structure rules is what is important with respect to subordinate clauses, i.e. clauses which are embedded into higher clauses or into constituents of higher clauses. To see where recursiveness of this kind must apply, ie. how phrase structure rules must be designed to capture sentence embedding, a short survey of the distribution and functions of subordinate clauses is necessary. This survey is based on traditional grammar which, despite all its shortcomings from a theoretic perspective, attempts to fully reflect the distribution and functions of subordinate clauses and to give accurate classifications of them. This is not to say that traditional grammar constitutes anything like a theory of syntax (cf Andersson (1975: 4)). Cf. Chomsky (1965/161990:138ff.), Rosenbaum (1967:25), Lees (1960). As this chapter is concerned with phrase structure, I leave out the discussion of the various approaches to how the semantic and the phonological component interact with the syntactic components. So far, the uniformity hypothesis on sentence structure has been dominant. See chapter 5 for discussion. Phrase structure rules in Chomsky (1957) were not designed to be recursive. Instead, the recursive property of human language was captured by generalized transformations which, as opposed to singulary transformations, fuse two or more phrase markers either by embedding one into another or by joining them. In Chomsky (1965/161990:137), generalized transformations are rejected in favor of encoding recursiveness in the phrase structure rules.

In traditional grammar it has been observed that subordinate clauses pattern functionally with other, non-sentential constituents: "[...] dependent clauses stand to their principal clauses in relations similar to those in which single words stand." (Sweet (1891/1968:170)) As nouns and adverbs can have their syntactic function fulfilled by a subordinate clause, subordinate clauses, on the basis of their being functionally analogous to nouns and adverbs, are - both in traditional and early generative grammar - referred to as noun clauses (2) and adverb(ial) clauses (3).5 (2) a. b. c. d. e. £ g.

[T\hatyou should think so is quite natural We believed that the worst was past. We will make whoever asks for it a quotation for the whole job. V/etaJkeaofwhathewoulddo. [She is] conscious that something had happened. He called my attention to the fact that he had only just arrived in this country. [M]y opinion is that he is mistaken.6

(3) a. b. c. d. e. f g. h. L j.

He died before writing a will You are younger than I am. She has such confidence in herself that she can meet any situation. Whether we like it or not, Jane has been selected for the leading role. He has a house of his own, whereas his brother lives in aflat. The gun burst because it was not clean. Although there is much to be said in your favour, I think your action was unwise. The work was carried out as it had been planned. Unless the strike has been called offj there will be no trams tomorrow. We noticed many boys climbing the trees so that they might have a good view of the procession. k. I shall go to Stratford, where Shakespeare was born and died.7

Chief among the syntactic functions noun clauses can assume are those of subject (2a), direct and indirect object of verbs (2b) and (2c) respectively. Further, noun clauses may function as

Cf. Sweet (1891/1968), Krapp (1908/1970), Quirk et al. (1985), Jespersen (1924/1992), Pence/Emeiy (1947), Allen/Mason (1939/1965), Blake (1988:127), Bolinger (1972), Geis (1970).

The examples in (2) are taken from the following sources: Sweet (1891/1968:171), Krapp (1908/1970:30), Pence/Emery (1947:166), Jespersen (1924/1992:103), Jespersen (1924/1992:163), Allen/Mason (1939/1965: 283) and Sweet (1891/1968:171). These examples are taken from Quirk et al. (1985:1081), Pence/Emery (1947.182,182,183), Jespersen (1909-1949,Part V:360), Krapp (1908/1970:213), Allen/Mason (1939/1965:293,259), Quiik et al. (1985:

1089), Allen/Mason (1939/1965:290) and Jespersen (1909-1949,Part V:346).

10

objects of prepositions (2d) and adjectives (2e)*. In (2f), the noun clause has an appositive fimction and in (2g), finally, the noun clause has a predicative function. Adverbial clauses may function as modifiers of verbs in (3), of adjectives in (3b), of adverbs in (3c), or they may be used absolutely as in (3d). Subordinate clauses with adverbial function are generally subclassified into temporal (3a), comparative (3b), consecutive (3c), conditional (3d) and (3i), adversative (3e), causative (3f), concessive (3g), manner (3h), purposive (3j), spatial (3k) adverbial clauses, etcetera. As can already be seen from this brief survey, recursion of sentences occurs in the verbal, nominal, adjectival, prepositional and adverbial systems as well as in the context of the copula. In early generative grammar, i.e. in post-Aspects models,9 the distribution of subordinate clauses is captured in a systematic way by formulating phrase structure rules in such as way as to allow for recursiveness, ie. by the introduction of the relevant positions at the level of deep structure. In the following I present a survey of the syntactic analyses of noun clauses and adverbial clauses.

1.1.1 Noun Clauses Syntactic functions such as subject, object, etc. are inherently relational notions which, as mentioned above, are defined by phrase structure rules and which can be captured in terms of immediate dominance. The nominal constituent that is immediately dominated by S is the subject, the constituent immediately dominated by VP is the direct object and so forth (c£ Chomsky (1965/161990:71)). The syntactic function "subject of can be read off the phrase structure rule given in (la), repeated as (4). The subject is the nominal constituent that - together with the predicate phrase - is immediately dominated by S.10 (4) S ->NP A Predicate Phrase Given that noun phrases and noun clauses have the same distribution, Le. assume identical functions within a given sentence, the phrase structure rule in (4) should alternatively allow for the following expansion: (5) S -» SΛ Predicate Phrase

Quirk et al. (1985) describe these as complements of adjectives, Pence/Emery (1947:182) as adverbial clauses modifying an adjective. To be precise, the major shift which is highlighted in the following is initiated in Chomsky

(1965/161990:137), where generalized transformations are abandoned in favor of phrase structure rules

introducing recursiveness in the base component.

Subjects, since they are not realized internal to VP, are not considered to strictly subcategorize the verb.

11

This step, however, is not taken. Instead, subject clauses are generated by application of the phrase structure rule in (6)11 with N being realized as expletive it taking obligatorily a sentential complement as in (7)12: (6) NP - > N A S (7) h A S

The deep structure of sentences containing a sentential subject as in (8) is generated by application of the phrase structure rules in (4) and (7) and subsequent lexical insertion: (8) That you should think so is quite natural S

should think so

The surface structure is derived by a small set of cyclically applying transformational rules. The first transformation to apply is the so-called Complementizer Placement Transformation, which has four subparts (cf. Rosenbaum (1967:25)).13 By this transformation the complementizer that is inserted into the embedded S yielding: (9) [up [N it] [s that you should think so]] The expletive element it, in a subsequent transformational cycle, is deleted by the so-called Pronoun Deletion Transformation (Rosenbaum (1967:7)) yielding (10).

11

Cf. Chomsky (1965/161990), Rosenbaum (1967), Rutherford (1970), Andersson (1973), Lakoff (1968), Freidin (1971), etc. A similar phrase structure rule introducing sentential subjects as instances of NP is given in Andersson (1974:1). Here, NP expands into S. (i) NP -> S Emonds (1972:25) points out that subordinate clauses which are (optionally ) introduced by that or infinite clauses are not instances of NP. He (1976:135) rejects the analysis in (i) since it violates the base restrictions of endocentricity. If noun clauses are generated as NP, they must have a head. The head either contains the pleonastic element it or it is empty.

12

Expletive it must be strictly subcategorized for a sentential complement in order that it be neither preceded by a determiner or adjective nor followed by a PP. Cf. Emonds (1976:116) for discussion and criticism.

13

Although Rosenbaum (1967:25) considers the question of whether the Complementizer Placement Transformation is to be captured by context-free or context-sensitive rewriting rules an empirical matter, he adopts the transformational analysis. The set of complementizers, in his system (1967:24, 32), consists of thatjor-to, Poss-ing, uA-words, //and whether. See also Jacobs/Rosenbaum (1968:164). I will come back to complementizers and complementizer placement in chapter 2.

12

(10) [up 0 [g That you should think so]] Evidence for this analysis comes from (11), a construction in which both it and the subordinate clause co-occur. (11) is derived from (9) by a transformation extraposing S, leaving it in its base position (cf Rosenbaum (1967:6)): (11) //is quite natural that you should think so. Sentential subjects, under this analysis, are derivative entities, i.e. they are not base-generated as sentences but as NPs headed by expletive /'/. Although the system has undergone substantial changes over the past three decades, the analysis of subject sentences, by and large, has not.14 Jackendoff (1977:75f, 95ff), for example, assumes (14) to be the underlying representation of subject sentences such as the one in (8). Here, the sentential subject it is introduced by the phrase structure rule given in (12), a reformulation of (4) within X-bar syntax (though still a rule governed version at that time).15 The subordinate clause, Le. the V" category, is assumed to be base-generated as a V" complement16 by (13). (12) V"-*N"(M")V" (adapted from Jackendoff(1977:54)) (13) V"->VV™ (adapted from Jackendoff (1977:83)) (14) It is quite natural that you should think so γι»

be

quite natural

that

you should think so

(15) is derived from (14) by the so-called Intraposition Transformation which substitutes N1" forV".17 14 15

16

17

Cf. chapter 6 for discussion. The era of X-bar syntax, a generalized conception of phrase structure, began with the publication of "Remarks on Nominalization" (Chomsky (1970)) and has undergone substantial changes over the years. Among the V complements we find adverbial and prepositional phrases of manner, means, time, etc. Cf. Jackendoff (1977:72). Alongside the Intraposition Transformation, Jackendoff (1977: lOlf.) discusses the Extraposition Transformation. This Intraposition Transformation, however, which is due to Emonds (1970), has been rejected in favor of the Extraposition Transformation (cf. Emonds (1976:121ff.) for arguments).

13

(15) That you should think so is quite natural. As has already been mentioned and as will be seen in chapter 6, where I discuss subject sentences from the Principles and Parameters Theory point of view, the analysis of subject sentences has not substantially changed in that sentential elements are not considered to occupy the sentential subject position either in general, at the level of deep structure or at the level of surface structure. For the introduction of sentential direct objects to verbs into deep structure, however, two options are available: direct object sentences can either be introduced as NPs of the form given in (6) above, Le. in the same way subject sentences are introduced, or they can be introduced directly as Ss (4):18 (16) a. VP ->VNP b. VP ->VS In order to capture subordinate clauses that function as objects of prepositions and adjectives as well as those that function as attributes to nouns the following phrase structure rules are needed: (17) a. PP-^PNP b. PP->PS (18) a. AP ->ANP b. AP -»AS (19) a. NP ->NNP b. NP ->S In addition to these, a phrase structure rule is needed to introduce the indirect objects of verbs. (20) VP -+VNPPP Indirect objects are generally introduced as prepositional phrases headed by the preposition to or for. Surface deviations from this underlying sequence of the objects are due to an optional

18

Chomsky (1965/161990:128,132,134) uses S1 as a dummy symbol to indicate the position into which a sentential element is inserted by a generalized transformation. In subsequent generative work, especially after Bresnan (1970), S' is taken to indicate that the sentential constituent under consideration is an embedded one (but see chapter 2.1.2 for a detailed discussion). Since this chapter is not concerned with the internal structure of subordinate clauses, I will use the category labels the various authors use for referring to subordinate clauses. Provisionally, I will take S, S' and V" as interchangable labels.

14

transformational rule which is referred to as dative shift or indirect object movement and whose function it is to delete the prepositional element and to permute the two objects.19 The phrase structure rule for direct objects is given in (21): (21) PP->PNP Rosenbaum (1967), McCawley (1970) and Postal (1974) treat object clauses by analogy with subject clauses as in (7). They are generated by the phrase structure rule given in (6), yielding a string as in (22). (22) [NT [N it] [s the worst is past]] The same set of transformations, i.e. Complementizer Placement and Pronoun Deletion, is held to apply to derive the surface structure (23). (23) We believed that the worst was past. According to Rosenbaum (1967), there are no phrase structure rules as in (17b), (18b) and (19b) that introduce sentential objects directly, but only the phrase structure rules in (17a), (18a) and (19a) which, as in the case of subject sentences, introduce sentential objects as NPs, with NP expanding into itA S. Chomsky's (1965/161990) phrase structure rules, on the other hand, allow for the immediate introduction of sentential objects as S's by the phrase structure rules. S', the category assigned to sentence transforms, denotes the structural position in which embedded sentences may occur.

(24) VP ->VS' Again, additional phrase structure rules are needed to generate sentential objects of prepositions (25), adjectives (26), attributes to nouns (27) and the indirect object of a verb (28): (25) PP-»PS'20

(26) AP -»AS 1 19

See, for example, Fillmore (1962), Emonds (1972,1976:78f.) and Jackendoff (1977:68) for transformational analyses of the double object construction. 20 The phrase structure rule expanding PP into P and a sentential complement is introduced in Emonds (1976), Jackendoff (1977). In these works, however, the rule is designed to introduce sentential complements to prepositional elements with subordinating function such as before, after, etc. This rule, then, captures Jespersen's (1924/1992:89) intuition that subordinating conjunctions in fact are "sentence prepositions". Sentential complements to prepositions are discussed in Freidin (1971:62ff.). Chapter 7 will be devoted to the analysis of prepositions taking sentential complements.

15

(27) NP ->NS' (28) VP -»VNPS 1 Since phrase structure rules are taken to define "the system of grammatical relations" and to "determine the ordering of elements in deep structure" (cf Chomsky (1965/161990:123), the phrase structure rules in (24) - (28) guarantee that the subordinate clauses fulfill identical grammatical relations within a given structure. With respect to (24) - (28) this means that the subordinate clauses bear the grammatical relation "(direct)2^object-of" the lexical elements they strictly subcategorize.22 Before cross-categorial generalizations and layered structures were introduced in a systematic way23, the term complement was used to refer to any sequence of nodes following a head position without, to use Jackendoffs (1972,1977) terminology, distinguishing functional arguments and modifiers.24 A first step towards layered structures and cross-categorial generalizations is taken in Chomsky (1970). On the basis of a lexicalist account of derived nominale25, Chomsky proposes refining the categorial component in a twofold way. First, the fact that nouns, verbs and adjectives may take complements must be encoded in the phrase structure rules. Second, phrase structure rules must capture the similarites between the lexical heads V, N and A and their respective complements.26 Chomsky (1970) therefore introduces the variable X ranging over the lexical categories V, N and A. X is dominated by X' with X* containing X and its complements (Comp), X' in turn is dominated by X". X" expands into a specifier (Spec), which contains lexical material preceding the head X, and X1. (31) gives the refined generalized formulation of the phrase structure rules in (29), which are taken from Chomsky (1970:195):

21

22 23 24

The reason for putting "direct" in brackets is that in early generative grammar, as well as in traditional grammar, the term "direct object" is reserved for nominal expressions. See for example Jackendoff (1977:32). In early generative grammar, all and only strictly subcategorizing phrases are referred to as complements. Cf. Chomsky (1970) and especially Jackendoff (1977). The distinction between functional arguments and modifiers is made in almost every grammar. It roughly corresponds to the following dichotomies (taken from Handke 1984): (i) actant - circonstant (Tesniere (1959/41982)) (ii) nucleus - periphery (Longacre (1964)) (iii) complement - peripheral element (Matthews (1981)) (iv) complement - adjunct (Quirk et al. (1972,1985))

(v) argument - adjunct (PPT and beyond) 2^ The Lexicalist Hypothesis breaks radically with previous, transformational approaches to nominalization, which led to a proliferation of idiosyncratic transformational rules. Furthermore, the then prevailing analysis of possessive constructions and modifiers of nouns as derived from relatives is rejected in this paper. 26 The lexical categories V, N and A are analyzed as feature bundles that are made up of two binary features [±V] and [±N]. Verbs are classified as [+V.-N], nouns as a [-V.+N] and adjectives as [+V,+N]. Prepositions, in 1970, fall outside the realm of lexical categories. Only in the mid 70s are they recognized as a lexical category characterized by [-V.-N]. Cf. chapter 4.

16

(29) a. NP -»N-Comp b. VP -»V-Comp c. AP -»A-Comp (30) Comp-» NP, S, NP S, NP Prep-P, Prep-P Prep-P etc.27 (31) X"->[Spec,X']X' Χ1 -vXComp (adapted from Chomsky (1970:210)) Elaborating the X'-convention, Jackendoff (1977) introduces a third structural layer which allows for a more refined generalization of phrase structure rules, which he (1977:30) represents as in (32). (32) X»-»...X«*i... The maximal value for n, in his system, is 3, Le. all lexical categories V, N, A and Ρ have three supercategories. This generalization is called the "Uniform Three-Level Hypothesis". The generalized structure of phrases in English is given in (33): X1" non-restrictive modifiers2*

X"

restrictive modifiers complements)

The term complement is now restricted to those nodes following a head which are functional arguments, i.e. elements which strictly subcategorize the head. The traditional notions direct and indirect object are subsumed under the broader notion complement: if and only if the complement is categorialry realized as NP, it is referred to as object (cf. JackendofF(1977:32)). Whereas verbs and prepositions allow for sentential complements as well as for nominal complements (NPs), adjectives and nouns do not allow for an NP complement (34) and (35), except when it is governed by a preposition (36) and (37):29 (34) * She is anxious our safety. (35) *It is a matter trust.

27 2

By this rule, NP, together with S, is assigned recursive status. & Non-restrictive modifiers are referred to as appositives, i.e. X1" complements, and restrictive modifiers as X"

29

complements. Cf. Jespersen (1924/1992:163) for adjectives.

17

(36) She is anxious for our safety. (37) It is a matter of trust. The distribution of NPs is restricted to the complement positions of case-governing heads, i.e. NPs, as opposed to sentences, must be assigned case.30 Due to their not being entities that need case, the distribution of complements which are categoriauy realized as Ss is less restricted; hence they can occupy the complement position of heads which do not govern case, i.e. adjectives and nouns: (3 8) She is conscious that something had happened. (39) She is anxious to get the job. (40) He called my attention to the fact that he had only just arrived in this country. (41) The assumption that sentences do not need case is widespread. On the basis of data such as (34) - (37) h becomes clear that NPs and Ss cannot be freely substituted for one another.31 This is what we would expect if subordinate clauses were introduced by the phrase structure rule given (6) which I take to mirror the traditionalist intuition which underlies Rosenbaum's claim (1967:113) that "[...] both noun phrases and clauses are instances of noun phrases and whatever affects a noun phrase affects both nouns and clauses."32 That Rosenbaum's claim (1967:113), however, runs counter to distributional facts was, according to Freidin (1971:25), already formulated in Emonds (1969) for NPs and Ss: Given surface constituents X and Υ where X # Υ; Υ can be analyzed as an expansion of X only if it can appear in all surface structure positions that X does (i.e. it behaves like an X in surface structure).33 Before turning to adverbial clauses, I want to briefly present the phrase structure rule which introduces subordinate clauses which are used predicatrvery. Predicatively used elements occur typically - though not exceptionally - after the copula; in Chomsky (1965/161990:102) there is no special phrase structure rule introducing subordinate clauses as predicates: (42) VP -> be Predicate 3

^ On the basis of the requirement that NPs be case-marked, characterizations of noun clauses as "Kasuss tze" (Mann (1942:88)) turn out to be inappropriate 31 Van Riemsdijk (1984:4) points out that "[i]n most languages true noun phrases and S's do not have a completely identical distribution [...]". ·" As will be seen in chapter 6, NPs and sentences display similiar properties with respect to extraction. 33 Emonds (1969), rejecting a rule expanding NP into sentences, as is reported in Freidin (1971:28), claims that "there is no evidence that clause objects ever appear in the object position in the surface". -JA

18 As can be seen from the data in (43), predicates in the relevant sense can be realized by the categories AP, NP, PP and S: (43) a. b. c. d.

My car is orange. Bill is a liar. The copier is out of order. My opinion is that he is mistaken.

With "predicate" in (42) ranging over the categories AP, NP, PP and S, the phrase structure rules at work can be given as in (44): (44) a. b. c. d.

VP VP VP VP

->be A AP ->be A NP ->be A PP ->be A S

Having discussed the phrase structure rules by which noun clauses are introduced into the structure, I now turn to the second set of subordinate clauses which are relevant to this study: adverbial clauses.

1.1.2 Adverb(ial) Clauses Adverbial clauses constitute the by or largest set of subordinate clauses. They are, as mentioned above, subclassified with respect to the adverbial relation they specify: adversative, causative, comparative, concessive, conditional, consecutive, manner, purposive, spatial and temporal relations.34 In Chomsky (1965/161990:102, 106£), we find adverbial relations such as time, place and manner introduced by phrase structure rules: (45) a. Predicate-Phrase ->· Aux Λ VP (Place) (Time) f (NP) (Prep-Phrase) (Prep-Phrase) (Manner) ] b. VP -» V \ S1 \ [ Predicate J 34

It should be noted, however, that there is variation with respect to the classification of adverbial relations in traditional grammar, i.e. the number of adverbial relations identified varies because some authors present finer grained categories than others. Jespersen (1909-1949,Part V:345), in his discussion of clauses as tertiaries, conjectures that any subclassification of adverbial relations in terms of such relations cannot but fail in setting up a real system for the description of adverbial subordinate clauses. Therefore, he proposes a system in which the relation between the relata with respect to cohesion is focussed on, i.e. the relation between the two sentential entities is located on a scale from a looser to a closer relation (1909-1949:346); (i) mutually independent (ii) compared with one another (iii) contrasted (iv) dependent on one another.

19

c. Prep-Phrase -> Direction, Duration, Place, Frequency, etc. (Chomsky (1965/161990:106£)) What is striking about the phrase structure rules in (45) and what has subsequently led to much criticism35 is the Act that adverbial relations or functions such as place, time, manner, etc. in (45a) and (45b) are introduced as category labels, thereby obscuring strictly categorial information. (45c) is not less confusing as prepositional phrases, according to these rules, do not expand into category nodes but are rewritten as adverbial types and thus do not capture recursiveness in any obvious way. By introducing such rules, as Filhnore (1969:362) puts h, [...] either the strictly categorial information is lost, or else it is rescued by having nonbranching rules which rewrite each of these adverbial-type categories as Preposition Phrase. In any case the formal distinction between relations and categories is lost, and the constraints on the further expansion of these preposition phrases that depend on the types of adverbiale they manifest need to be provided [...] in ways that have not yet been made clear. Subordinate clauses which express the adverbial relations of place, time and manner (46) are not captured by these rules, nor do the rules capture the fact that, for example, NPs36 and AdvPs may express these relations (47). (46) a. She found the book where she had put it. (S) b. She bought the book -while she was in London. (S) c. You are younger than I am. (S) (47) a. She bought the book last summer. (HP) b. He did the dishes carefully. (AdvP) Andersson (1975:6), therefore, proposes having adverbial subordinate clauses introduced as follows:37 (48) AdvP->S Introducing adverbial subordinate clauses (and probably NPs and PPs) as instances of AdvP, i.e. the phrasal category of adverbs, mirrors the traditional grammar point of view sketched above.

35 36

37

Cf. Fillmore (1966,1969, passim), Steinitz (1969), etc. Within the framework of Fillmorian "Case-grammar", NPs would be introduced in deep structure as PPs, thus conforming to the rules given in (45). Later, Andersson (1975:7) gives (i) AdvP -> S1 (ii) S' -» Comp S as phrase structure rules relevant for the introduction of adverbial subordinate clauses. I will come back to S' and its expansion in chapter 2.

20 Again, as in the case of noun clauses, functional analogy is the criterion for classifying clauses.38 Emonds (1976:135), with respect to the analysis of noun clauses as NPs, argues that, m order not to violate the base restrictions of endocentricity, a clause which is generated as NP must have a head of the category N. But the phrase structure rule given in (48) is a violation of endocentricity.39 In order not to violate endocentricity and to keep functional and categorial concepts separate, phrase structure rule (48) has to be rejected in favor of (49), which allows the introduction of adverbial subordinate clauses as instances of category S: (49) S-»NPA Predicate Phrase Although the phrase structure rules in (45) had to be rejected because they inadequately introduced functional relations in terms of category labels, they are still relevant for the present purposes since they show in a quite appropriate - and yet to be refined - way the differences in cohesion between adverbial elements, here adverbial subordinate clauses, and the structures into which they are embedded. Sentential elements, as we have already seen in connection with noun clauses, occur recursively. They may function as complements, restrictive modifiers and as non-restrictive modifiers. The adverbial clauses in (50) function as complements, i.e. they strictly subcategorize the verb. The relation "complement of reflects the highest possible degree of cohesion. (50) a. He put the car where it belongs*® b. President Nkrumah [...] has resumed his holiday on the Black Sea and no one here professes to know when he will return. Traditional grammar, with its distinction between noun clauses and adverbial clauses on the basis of functional analogy, would not recognize that the subordinate clauses in (50) take on the function of nominal expressions, i.e. they pattern distributionally like noun clauses. Geis (1970), for example treats adverbial clauses of this type, ie. spatial and temporal adverbials, as relative clauses whose antecedent has been deleted.41 The deleted antecedent of where in (50a) would be the head of an NP with spatial reference, such as the place; the ante-

38

39 40 41

The syntactic category and the syntactic function "adverbial" are merged. As Sweet (1891/1968:170) pointed out, the subclassification of subordinate clauses has nothing to do with word classes. Endocentricity, up to date, is central to any theory of phrase structure in the generative tradition. (50a) is taken from Handke (1984:71), (50b) from LOB. A similar analysis is favored by Jaworska (1982), who takes adverbial clauses to be instances of complex nominals. Her analysis as well as Geis' (1970) focus on Ross' (1967) Complex Noun Phrase Constraint stating that adverbial subordinate clauses together with complex noun phrases are extraction islands, i.e. no element can be moved out of the constructions.

21

cedent of when in (50b) would be the head of an NP with temporal reference, such as the time V 43 Adverbial subordinate clauses that do not function in strict subcatgeorization can belong to one of a number of classes. From among the various analyses which have been proposed, I have chosen Williams (1975) as an illustration of the distribution of adverbial subordinate clauses. Williams identifies four classes of phrases that can follow the verb. The first class (I) he discusses comprises the phrases that strictly subcategorize the verb (cf ex. (50)). The second class (Π) contains manner adverbials, by- andjfawz-phrases and VP adverbs. (51) a. He left without saying a word. b. John got sick from eating so much. c. John offended Mary by telling jokes. (Williams (1 975 :253f)) The third class (HI) contains temporal adverbials introduced by during, before, after, while as well as subordinate clauses introduced by because**; (52) a. b. c. d. e.

He died before writing a will He left after dinner was over. The gun burst because ft was not clean. When I was 17, my father died. She got nervous while she was waiting for her publisher to show up.

Class four (IV) contains, for example, whether or «o/-phrases, certain w/Y/i-phrases and certain

(53) a. Whether we like it or not, Jane has been selected for the leading role. b. With John gone, havoc will break loose. c. He must be smart, because I was told that he solved the problem in no time. 42

43

44

45

These NPs can be either embedded into PPs: (i) I found it at the place where I had put it -> I found it where I had put it or they can be so-called bare-NP adverbs (cf. Larson 1985b): (ii) I don't know the time when he will return. -> I don't know when he will return. Subordinating conjunctions, in Geis' (1970:12ff.) system are assumed to be derived from different underlying structures. Whereas where, when and while are derived from relative adverbs (1970:37,60,passim), before and after derive from compared adjectives (1970:123,139,passim). Prepositions are held not to exist at the level of deep structure, instead they are considered degenerate VPs (1970:116ff.) Aspectual verbs are the basis for since and until (1970:139ff.,147,passim). As for adverbial subordinate clauses introduced by if, unless, although and because, he assumes that they function as noun complements (1970:15). If and unless, for example, are taken to be suppletive forms of the PPs in the event that and in any other event. (1970:14,164ff). The examples in (52) are taken from Quirk et al. (1985:1081), Krapp (1908/1970:213,213) and Jespersen (1909-1949,Part V.348). (53a) is taken from Pence/Emery (1947:183), (53b) from Williams (1975:253).

22

Evidence for these classes comes from certain movement facts, e.g. dative-movement, particle movement, passive, preposing, from which Williams (1974,1975) derives principles of cyclically applying rules.46 Williams (1975:269) presents the following schema giving the structural positions of adverbials.47

IV48 ΠΙ49

Aux

V

II I

The rules that are needed to allow for the introduction of adverbial clauses and hence capture their distribution within the verbal system are given in (55). (55) a. b. c. d.

V" V" V V

-> COMP V" V" -»NPV'V" ->VV" ->WM

1.2 Summary In this chapter I have reviewed the treatment of two types of subordinate clauses in early generative grammar. I have argued that the heritage of traditional grammar shows through, not only in the classification of subordinate clauses (as noun clauses and adverbial clauses), but also in the conception of the phrase structure rules introducing them in the earliest models of generative grammar. The introduction of noun clauses by phrase structure rule (6), repeated as (56), shows very clearly the impact traditional grammar had on early generative models. (56) NP 46

Further evidence for these different classes of adverbials comes from their behavior under negation and interrogation. Whereas elements of the classes I-III fall within the scope of negation and wh-movement, elements of class IV do not (cf. Williams (1974:202ff.), Haegeman (1985)). Further differences between these classes suface with respect to focus and clefting (cf. Jackendoff (1977:61ff), Haegeman (1985), Handke(1984:66ff.)). 47 I take COMP, i.e. the category designating subordinating elements, in this system to be adjoined to V", as proposed in Jackendoff (1977:47). 48 Elements of class IV correspond to Jackendoff s (1977:61ff.) non restrictive modifiers. 4 9 Class II and III correspond to Jackendoff s (1977:72) V complements, i.e. they function as restrictive modifiers.

23 Similarly, the introduction of adverbial clauses by phrase structure rule (48), repeated as (57), mirrors the traditional basis of early generative grammar. (57) AdvP-»S Although drawing on notions provided by traditional grammar, early generative grammar crucially diverges from the mere descriptive claim of traditional grammar in that it attempts to capture both sentence structure and distributional facts in terms of a finite set of rules and recursive application of these rules. Thus, clauses, no matter what their actual form or function is, do not differ at the level of deep structure. Different surface linear orders are assumed to be arrived at by transformations operating on deep structures (cf 1.1). In 1.1.1,1 have reviewed some of the refinements that led to capturing the distribution of noun clauses as instances of category S'. As a result noun clauses - except when occurring in subject position - are not any longer introduced by (56). Their distribution in the context of verbs (16), prepositions (17), adjectives (18) and nouns (19) is accounted for by introducing them as sentential. Encoding recursiveness in terms of phrase structure rules (cf Chomsky (1965/16:137)) has been argued to have paved the way for cross-categorial generalizations, known as the X'-convention, first formulated in Chomsky (1970)50 and then elaborated by Jackendoff(1977)51. I have approached adverbial clauses in 1.1.2 in the same spirit. Starting out with the design of phrase structure rules introducing adverbial clauses hi (45), repeated as (58), (58) a. Predicate-Phrase -> Aux Λ VP (Place) (Time) f (NP) (Prep-Phrase) (Prep-Phrase) (Manner) } b. VP -> V Ί S1 }· ( Predicate J c. Prep-Phrase -» Direction, Duration, Place, Frequency, etc. (Chomsky (1965/161990:106£ )) and with the subsequent criticism of those rules (cf Filhnore (1969), Andersson (1975)), I have reviewed the arguments that led to treating adverbial clauses as being deep structurally represented as instances of S1 and not as AdvPs as has been suggested by Andersson (cf (57)). Both sets of clauses have been argued to be generated by identical phrase structure rules, ie. the category of subordinate clauses is, depending on the notation chosen, V" or S. As this chapter has been primarily concerned with the distribution of subordinate clauses, virtually nothing has been said about their internal structure. In particular, nothing has been said about the elements introducing subordinate clauses. I have used the term complementizer to refer to these elements.

50

Cf. (29) above.

51

Cf. (32) and (33) above.

24

In the following chapter, I shall discuss two approaches to complementizers, one of which is transformationally based and one of which takes complementizers to be represented in deep structure.

"Complementizer", like so many terms in linguistics, is a rebarbative word. (Bresnan (1972:6))

2 Complementizers In the first part of this chapter I review two approaches to complementizers: the transformational analysis (Rosenbaum (1967) and Lakoff (1968)) and the phrase structure rule analysis (Bresnan (1970,1972,1973,1974)). Central to the discussion of complementizers introducing predicate complements, i.e. noun clauses, is their relation to the governing predicate. In the transformational approach (2.1.1), this relation is characterized as rule-feature governed (cf. Lakoff (1968)), whereas under the phrase structure rule analysis (2.1.2) it is captured by strict subcategorization (c£ Bresnan (1972), Grimshaw (1979)). The arguments which led to preferring the phrase structure rule analysis over the transformational analysis will be reviewed in some detail The second part of this chapter focusses on extentions of the phrase structure rule analysis to matrix clauses (2.2). As for the discussion of wA-interrogatives, wA-movement, as formulated in Chomsky (1977) and Chomsky/Lasnik (1977), will be central On the basis of wA-movement as discussed in connection with matrix interrogatives, embedded questions will be reconsidered (2.2.1.1). The generalized structure of embedded clauses and of matrix interrogatives serves as the background against which a brief discussion of matrix declaratives is set (2.2.1.2). A summary is presented in 2.3.

2.1 Complementizers: Transformations and Phrase Structure Rules Within early generative grammar, elements introducing predicate complements, Le. noun clauses, were classified us particles1, complementizing morphemes or complementizers? Among these we find elements like that, Poss-ing,for-to and wA-elements.3 Rosenbaum (1967:25) points out that Implicit hi the term "complementizer" is the idea that these morphemes are a function of predicate complementation and not the property of any particular sentence or set of sentences.4 Although complementation in those days, as has been pointed out in the previous chapter, was a rather fuzzy term since it subsumed not only complements in the strict sense, and the vast The terminology is based on traditional grammar where non-inflecting elements (prepositions, conjunctions, adverbs) were standardly subsumed under the lern particle. Cf. Rosenbaum (1967), Bresnan (1970,1972), Noonan (1985:42ff). Jacobs/Rosenbaum (1968:164) describe that as a clause complementizer, for-to as an infinitive complementizer and Poss-ing as a gerundive complementizer. This does not only imply that the sentences to which the Complementizer Placement Transformation applies have identical deep structures but also that there will be no complementizer in main clauses.

26

majority of analyses focussed on what today is referred to as a sentential argument, i.e. the noun clauses of traditional grammar.5 Only Bresnan's analysis goes beyond complementizers introducing predicate complements. As we have seen in 1.1.1, noun clauses and noun phrases are assumed to have the same distribution, i.e. they occur as subjects (1) and objects (2) of verbs: (1) a. That he was alone was obvious from the report. b. For you to leave right now would be inconvenient. c. Whether he hates cabbage or not simply doesnt matter. (Bresnan (1970:313)) (2) a. I know that he is wise. b. I prefer for you to speak English. c. I am adkmg whether you will accompany me. (Bresnan (1970:313)) The term complementizer, under this view, then, is a telling name: its function is to assign sentences the status of complements in the relevant sense:6 Complement types often have associated with them a word, particle, clitic, or affix whose function it is to identify the entity as a complement. (Noonan (1985:45))

2.1.1 Transformationalist Approaches to Complementizers In this section, two early generative grammar approaches to complementizers will be reviewed. These analyses crucially take for granted that complementizers are transformationally inserted. hi this respect, early generative grammar in an interesting way parallels structuralist theory a la Tesniere (1959/41982:543-617) where it is assumed that verbal nodes (sentences) cannot function as "actants" or "circonstants" unless they are transformed (translated) into these function-categories by means of an appropriate marker (translatif).7 As complementizers are held to be void of autonomous meaning8 they can - hi accordance with the postulate that "transformations [...] not introduce meaning-bearing elements" (cf Chomsky (1965/161990:132)) - be "[...] added to an embedded sentence" (cf Lakoff (1968:20)) without affecting the semantic interpretaion of the sentence. Complementizers, under Lakoff s (1968) tranfonnationalist approach, are introduced obligatorily by the so-called 5 6

7

Cf. Lees (1960), Rosenbaum (1967), Bresnan (1970,1972). Bresnan (1974:617) refers to complementizers as clause-making morphemes.

Subordinating conjunctions, i.e. elements which serve as "translatifs" for non-complements, will be dealt with in chapters 7 and 8. * This claim, in an interesting way, mirrors the structuralist claim that particles, i.e. elements of so-called closed classes, are void of meaning. Cf. Fries (1952), Hockett (1958). Jespersen (1909-1949,Part V:344) claimed that that"[...] in itself has no meaning, but serves to indicate that what follows is subordinate to and connected with the main statement."

27

Complementizer Placement Transformation, rather than by phrase structure rules.9 The rule which inserts complementizers, Le. the Complementizer Placement Transformation, must be sensitive to the element governing the embedded clause (cf Rosenbaum (1967:27), Lakoff (1968:19ff)). Rosenbaum (1967:28) argues that the fact that the choice of the complementizer is dependent on the matrix verb requires a "[..] descriptive apparatus which posits that verbs are marked with particular complementizer features in the lexicon that may appear in their verb phrase complements." Lakoff (1968) proposes introducing rule-features such as [+for-to] or [+poss-ing] into the lexical entries for verbs.10 As can be seen in (3), the choice of complementizer introducing the sentential complement, e.g. that, for-to or Pass-ing, is determined by the governing verb.11 (3) a. He says that the world is flat / *for the world to be flat / *the world's being flat. b. He conjectures that the world is flat / *for the world to be flat / *the world's being flat. c. He wants "that the gang leaves / for the gang to leave / *the gang's leaving d. He loathes *that the gang leaves / for the gang to leave / *the gang's leaving. e. She abhors *that he kills people / *for him to kill people / his killing people. £ She welcomes *that he gave up this job / *for him to give up this job / his giving up this job. But often verbs allow for variation with respect to the complementizers introducing their sentential complement: (4) We feared that he would lose the game / for him to lose the game / his losing the game. In the following I will briefly compare the Complementizer Placement Transformations proposed by Rosenbaum (1967) and Lakoff (1968). Rosenbaum (1967:5), as mentioned in 1.1.1, proposes a Complementizer Placement Transformation with four subparts12 which applies in the second cycle.13 Once the complementizer is inserted, either the Extraposition Transformation or the Pronoun Deletion Transformation may apply.14 Among the set of transformations specified in Rosenbaum, there are two Complementizer Deletion Transformations: one obligatory (5) the other - except in sentential subjects - optional (7).

9

10

The latter is an option Rosenbaum (1967:25) alludes to. However, he chooses the transformational analysis rather than the phrase structure analyis for reasons of familiarity. Cf. Kayne (1969), Bresnan (1970,1972) for criticism of rule-features.

11

The examples are construed on the basis of verb classifications given in Rosenbaum (1967:120ff).

12

These are assumed to apply in sentential complements to verbs and nouns. There are no transformations in the first cycle, the structure is the one generated by the phrase structure rules. Rosenbaum (1967:37, passim) refers to first-cycle structures as BASE. Cf. Rosenbaum (1967:6), see 1.1.1.

13

14

28

(5) Χ [4€]15 VP Υ 1 2 3 4 1 0 3 4

» (Rosenbaum (1967:6))

(5) is the transformational rule to apply last, Le. it follows all other transformations.16 (6) may serve as an example for one of the contexts in which (5) must apply: (6) a. 1st cycle: somebody trusts [John] [[ John] [do the work] BASE b. 2nd cycle: somebody trusts [John] [for [ John] [to do the work] Tcp somebody trusts [John] [for [0] to [do the work] Tffi somebody trusts [John] [ 0[0] [to [do the work] TCD (adapted from Rosenbaum (1967:96)) (7) X V ADJ 1 2 1 2

a. N [+D] NP Υ [+PRO] b. (NP) [-D] 3 4 5 6 3 0 5 6

->

One of the contexts in which (7) may apply is given in (8): (8) a. 1st cycle: They doubt [[it] [you will go]] b. 2nd cycle: They doubt [[ft] [that you will go]] They doubt [[ft]] [that you will go] They doubt [[ft]] [0 you will go] They doubt [[0]] [0 you will go]

15

BASE TCP TE T^ TPD (adapted from Rosenbaum (1967:43))

Rosenbaum (1967:26f.) proposes having complementizers represented by features: (i) that, for-to, Poss-ing [+C] (ii) that [-D] (iii) for-to, Poss-ing [+D] 16 Rosenbaum (1967:5ff.) lists the following extrinsically ordered transformations: Complementizer Placement Transformation (Tcp), Identity Erasure Transformation (T^, Subject-Object Inversion Transformation (TgQj), Passive Transformation (Tp), Extraposition Transformation (Tg), Optional Complementizer Deletion Transformation ((T^u), Auxiliary Transformation (TAUX), Pronoun Replacement Transformation (Tyg), Pronoun Deletion Transformation (TPD), Preposition Deletion Transformation (TPPD) and Obligatory Complementizer Deletion Transformation (TCD).

29

Complementizer Deletion rules, in feet, come as a logical consequence if one assumes, as Rosenbaum does, that complementizers are obligatorily inserted into deep structures. Lakoff (1968) follows Rosenbaum in the assumption that Complementizer Placement is an obligatory transformation. Her approach, however, differs crucially from Rosenbaum's in that she gives only one Complementizer Placement Transformation17 instead of four:

(9) X t - it - # - S - # - X2 1 2 3 4 5 6 1 2 thai + 4 0 6 (Lakoff (1968:25)) This rule applies in any context which is specified for sentence embedding. The complementizer which is introduced, irrespective of the governing element, is thai.™ In order to derive sentences as in (10), where the verb of the superordinate clause requires that a certain complementizer be realized in the sentential complement, (10) a. b. c. d. e. f

I hope for Tom to sing. I like to sing. I demand that he leave. I wonder how he did ft. I asked if I might stay, I wonder when he left.

she assumes a four part Complementizer-Change rule which replaces that. This rule, which is given in (11) is assumed to apply to any particle.19 (11) a. for-to Xj- that1 2 1 - 0

NP - VP X2 3 4 5 for+3 - t o + 4 - 5 (Lakoff (1968:26))

b. possessive-/«^ Xj- thatNP VP 1 2 3 4 1 - 0 poss+3 - ing+4 -

X2 5 5 (Lakoff (1968:27))

17

Included in the Complementizer Placement Transformation is the deletion of sentence boundaries (#). Cf. Lakoff (1968:25) 18 Jespersen (1909-1949,Part V:343f.) characterized that as the "maid of all work", which "in itself has no meaning but serves to indicate that what follows is subordinate to and connected with the main statement." 19 Particle, in this context, should be restricted to subordinating conjunctions.

30

c. //Mtf-non-finite20 X!- that- NP - VP X2 1 2 3 4 5 1 -2 -3 - C + 4 - 5

->

(Lakoff(1968:28)) d. how, f/finhe, when... Xj- /Λα/ S - X2 1 2 3 4 -> 1 - how 3 - 4 //-finite when (Lakoff( 1968:28)) Whether or not any of the rules just given applies is basically determined by the governing verb which " [...] is marked as having to undergo any of the complementizer-changing rules [...] (c£ Lakoff (1968:26)). Lakoffs (1968) Complementizer-changing rules, as opposed to Rosenbaum's (1976) Complementizer Placement, are rule governed, that is to say that verbs are lexically specified in terms of rules-features such as [+for-to] or [-tposs-ing] with respect to the kinds of complement types they allow. Beside the Complementizer Placement Transformation and the Complementizer Change Rules, the transformational component contains Equi-NP-Deletion, //-Substitution, Flip21, Passrvization, Extraposition, Λ-Deletion, Preposition-Deletion, TTwtf-Deletion and For/PossDeletion22. Needless to say, these are extrinsically ordered. Of these transformations, I want to briefly look at 77w3/-Deletion and For/Pass-Deletion. The TTiar-Deletion rule, as m Rosenbaum's system (1967:6), is optional. It is, however, specified with respect to the contextual environment of a governing verb.2·* (12) 77ia/-Deletion Xt- V - that - S - X2 1 2 3 4 5 1 - 2 - 0 - 4 - 5

->

Rule (12), however, must be blocked from applying in the context of verbs such as happen (13).

20

Lakoff (1968:28) conjectures that C in this rule is an abstract complementizer, adjoined to VP, whose function it is to remove all of the inflectional material on the verb. As will be shown in chapter 6.2.2, sentences containing verbs in the subjunctive are finite. If they were not, the subject would not be assigned nominative Case.

21

Flip accounts, for example, for raising structures.

22

A detailed discussion of these transformations is given in Lakoff (1968:29-55). ·* The contexts in which that may be absent are often referred to as bridge contexts.

2

31

(13) a. It happened that John left, b. *It happened John left. In order to rule out sequences such as in (13b), Lakoff( 1968:48) proposes having verbs that do not allow for their sentential complement's complementizer to be deleted marked in the lexicon. The application of For/Posj-Deletion rule is restricted to pre-VP environments: (14) For/Awj-Deletion XI- for/poss - VP - X2 1 2 3 4 1 0 3 - 4

->

(adapted from Lakoff(1968:48)) At a certain stage in the derivation of the sentences in (15), (15) a. b. c. d.

I want to go. I want Harry to go. I don't like doing that. I don't like John doing that. (Lakoff( 1968:48))

we encounter ungrammatical sequences as in (16): (16) a. b. c. d.

*I want for to go.24 *I want Harry for to go. *I dont like 's doing that. *I don't like John 's doing that. (Lakoff(1968:48))

These are the structures to which the deletion rule in (14) applies, Le. pre-VP for andposs ('s) are obligatorily deleted. In the two approaches sketched, a transformationalist perspective on complementizers was taken, Le. complementizers are not present at deep structure but they are obligatorily inserted on the second cycle. The insertion of complementizers - although on the second cycle - does not affect the semantic interpretation of the sentence since complementizers are considered void of semantic content. Whereas Rosenbaum (1967) introduces four rules which govern the insertion of complementizers, Lakoff (1967) introduces only one rule the output of which, depending on the governing verb, may be affected by one of the Complementizer-change rules given in (11).

24

According to Chomsky (1977:103), these structures are ruled out by the so-caüedfor-to-ßlter, which applies to surface structure.

32 Both authors assume that the choice of complementizers, either by insertion or by a changing rule, is dependent on the element governing the embedded clause. In order to account for surface structures that do not contain a complementizer, several deletion rules are necessary. Under transformationalist approaches such as Rosenbaum's (1967) and Lakoffs (1968), sentential complements are identical at the level of deep structure and the actual make-up of a given sentential complement is arrived at by transformations. Crucially, since they are sensitive to the rule-features of the governing head, complementizer-insertion transformations cannot take place on the first cycle but have to be deferred until the second cycle because only then can the transformational rules meet the structural description which obligatorily activates thenapplication.25 The assumption that complementizers do not have autonomous semantics and thus can be inserted transformationally leads to a proliferation of contextually specified transformational rules and thus systematically precludes generalizations. These two central assumptions of the transformationalist approaches are challenged by the phrase structure rule-driven analyses to complementizers which were initiated by Bresnan's (1970,1972) most influential work on complementation, which will be discussed in the following section. It will be shown that the phrase structure rule-driven analysis is not a mere alternative to the transformationalist approaches but superior to these for at least two reasons. First, it recognizes semantic differences between individual complementizers26 and thus allows for a principled account of the selection of sentential complements, and second, it paves the way for generalizations, thus minimizing the transformational burden.

2.1.2 Complementizers as Deep Structure Nodes Bresnan (1970:301ff., 310ff; 1972:33ff., 60ff), on the basis of distributional properties, semantic differences, subcategorization and from the interaction of complementizers with transformations, rejected the then prevailing view that complementizers are mere subordination markers lacking both semantic content and relevant syntactic function and hence are introduced transformationally. Concerning the alleged voidness of meaning, Bresnan (1972:60ff.)27 points out that complementizers such as that, for and wh- differ semanticaUy. In the examples in (17), the embedded clauses are governed by the verb say. (17) a. She never said that we should leave. b. She never said whether we should leave. 25

See, for example, the transformational rules given in (6), (8), (9) and (11) above. It should be noted that, since complementizer placement is sensitive to the rule-features of a governing head, there can be no root complementizers, i.e. the fact that there are no strings such as *That I hope to sing, is a consequence of the transformationalist approach. 26 Attributing semantic content to complementizers precludes them from being transformationally inserted, i.e. complementizers - contrary to what is assumed under the transformationalist approaches - affect the semantic interpretation of the sentence they figure in. 27 See also Bresnan (1970:302).

33 c. She never said for us to leave. (Bresnan( 1972:60)) If it were the case that complementizers did not contribute to the interpretation of the sentences they introduce, the sentences in (17) should be synonymous, which they are definitely not. Whereas the sentential complement in (17a) makes a statement, the sentential complement in (17b) states a decision and in (17c) it expresses an order. Further, Bresnan (1972:602) points out that, because of semantic differences between that and for, (17c) cannot be derived from (17a), which we would have to assume if we followed Lakoffs (1968) analysis. Bresnan (1970:300) argues that the peculiarities that adhere to the transformationalist analysis, e.g. Complementizer Placement on the second cycle or, in Lakoffs analysis, complementizer change, disappear in an analysis which introduces complementizers at deep structure, ie. by the phrase structure rule as k (18). (18) S' ->COMPS COMP, under this analysis, is a deep structure node that dominates complementizers (that, if, whether, for, than, as and wA-)28. Not only does a phrase structure rule based approach to complementizers and complementation allow for the reduction of transformations, it also leads to a more principled account of the relation between the governing head and the choice of the complementizer, a correlation which Lakoff (1968), for example, derives from rule-features on governing heads. Bresnan (1970) argues that complementizer choice, under the analysis which takes COMP as a category represented in deep structure, can be captured in a principled way if subcategorization, a concept which is available in the theory anyway, is drawn upon.29 As was argued with respect to the examples in (10) and as can be seen again in (19), certain verbs require that certain complementizers be realized in the sentential complement: (19) a. b. c. d. e. f g. h. L j.

28

29

We believed that he did the dishes. We assumed that he did the dishes. We suggested that he do the dishes. *We believed whether he did the dishes. *We assumed whether he did the dishes, *We suggested whether he do the dishes, We inquired whether he did the dishes, We asked whether he did the dishes. We wondered whether he did the dishes, *We inquired that he did the dishes.

Although Rosenbaum (1967:32) lists wfc-words, //and whether as complementizers, he does not bother to give Complementizer Placement Transformations to introduce these elements. Lakoff (1968:28) derives whcomplementizers by the complementizer-change rule given in (1 Id) above. Under this analysis, rule-features, such as assumed by Lakoff (1968), can be done away with.

34 k. *We asked that he did the dishes. L *We wondered that he did the dishes. Verbs like believe, assume and suggest allow for sentential complements introduced by that as in (19a) - (19c) but not for sentential complements introduced by whether or other whelements (19d) - (19f). Sentential complements introduced by that are illicit with verbs like inquire, ask and wonder (19j) - (191), these verbs are strictly subcategorized for sentential complements introduced by wA-elements such as whether in (19g) - (19i). Whereas the requirement that a certain complementizer be realized was met by complementizer-change rules under Lakoffs (1968) analysis, this requirement is met by strict subcategorization under Bresnan's (1970,1972) and also under GrimshaWs (1979) analyses, where complement-taking heads, e.g. verbs, are strictly subcategorized not only in terms of syntactic categories but also in terms of complementizers.30 Under this analysis, then, the verbs believe, assume and suggest in (19) are strictly subcategorized by that complements, the verbs inquire, ask and wonder are strictly subcategorized by Η'Λ-complements. Further, as Grimshaw (1979) argues, subcategorization not only makes reference to the syntactic type of the sentential complement, but also to its semantic type. The verbs wonder and think are both strictly subcategorized by S1, Le. they syntactically select a sentential complement. The subcategorization frame is given in (20). (20) wonder, think [

S1]

As we have already seen in the ungrammatical examples in (3) and (19), information about the syntactic category of the complement does not suffice to exclude ungrammatical strings such as those in (21b) and (2Id). (21) a. He wondered whether it was late. b. *He wondered that it was late. c. He thought that Bill was late. d. * John thought whether Bill was late. Semantically, the verbs wonder and think differ in that the former semantically selects a question whereas the latter semantically selects a proposition31. These semantic differences, as Grimshaw (1979:288) on the basis of Bresnan (1970,1972) convincingly argues, must be lexically specified. She therefore proposes having encoded in the subcategorization frames the semantic type of the sentential complement:

30

3

Lasnik/Saito (1992:2), for example, state that "[a] close look at the distribution of sentential complements in English immediately reveals that whether or not COMP contains a W7/-phrase depends on the matrix predicate."

M leave out exclamations, i.e. non-interrogative Ηώ-initial structures, which constitute the third semantic type (cf. Grimshaw (1979) for discussion).

35 (22) a. wonder b. think

[ [

S'], [ S'j, [

Q] (Q=question) P] (Proposition) (Grimshaw( 1979:288))

Since semantic selection, as well as categorial selection, can be multiple (23), the alternative realizations have to be encoded in the lexicon (24). (23) a. Bill found out that Tom left. b. Bill found out who left. For the examples in (23), the syntactic part of the subcategorization frame for the lexical hem find out contains S1, the semantic portion of it specifies that S' may be either a proposition or a question. (24) find out [

S1], [

P]

Q

(Grimshaw( 1979:288)) In addition to the strong correlations between the governing verb and the semantic type of its complement, there are strong correlations between the governing verb and the complementizer and between the complementizer and the semantic type of the complement it introduces. The latter type of correlation, Le. the correlation between the complementizer and the semantic type of the clause it introduces, is central in Bresnan's (1972:60ffi) discussion of the semantic functions of complementizers in noun clauses or predicate complements, where she focusses on w/t-elements, that and for, which she calls central complementizers. Crucially, under her analysis - as opposed to the transformationalist analyses discussed above - complementizers are assumed to have inherent semantic meaning which determines the semantic type of the complement. She argues that wA-complementizers such as those in (25) express the "undetermined" parts of the complements they introduce. (25) a. Π1 ask whether Linda is coming. b. She asked which one you liked. c. They are wondering what to do. (Bresnan (1972:62,63,68)) With respect to wA-complementizers such as which, hew and what, Bresnan (1972:621) draws on an idea put forward in Baker (1970), where the effect of wA-elements is considered to be "[...] a semantic function on determiners [...]", ie. either, some, so, that, and on pronouns, ie. then and there. Wk-, under this perspective, serves as an "undeterminer", Le. it converts the reference of the elements it is "prefixed" to, Le. wh- induces a change from determined to undetermined reference.

36

With respect to the bracketed elements in (26), this function may be phrased as in (27). (26) a. b. c. d. e. f

WH(some) WH(so) WH(that) WH(then) WH (there) WH (either)

= = = = = =

what how which when where whether (adapted fromBresnan (1972:63))

(27) For any X (X = either, some, so, that, there, where) WH-(X) undetennines the reference of X. (adapted fromBresnan (1972:62)) As a result of containing an undetermined element, the complement basically corresponds to what counts as an open sentence in logic, Le. a sentence which contains free variables and hence cannot be assigned a truth value.32 Bresnan (1972:65) argues that, based on the meaning of w/j-elements, the heads which allow or disallow wit-complements can be predicted: Interrogative and dubhative predicates leave aspects of their complements open to question; they are compatible with "undetermined", semanticalty open complements. On the other hand, predicates which assert or presuppose truth value, or which imply knowledge or certainty, should resist WH complements. (Bresnan (1972:65)) The examples given so far, e.g. (19), (21) and (25), demonstrate that this prediction is borne out. That, as opposed to ννΛ-complementizers, functions as a "defmitizer" with respect to the complement it introduces (cf Bresnan (1972:69)).33 Definiteness, in the relevant sense, precludes subordinate clauses introduced by that from functioning as complements to interrogative and dubhative verbs such as those in (25).34 That complements readily occur as complements to factive predicates and as complements to verba dicendi and verba sentiendL35 (28) a. He says that the world is flat. b. He conjectures that the world is flat. c. We feared that he would lose the game.

32

Cf. Bresnan (1972:65), Williams (1980:209£). She (1972:72ff) proposes analyzing that as a definiteness operator on propositions. Napoli (1993:387) proposes having definiteness introduced by a feature [+th]. 34 For exceptions cf. Bresnan (1972:70f.) 35 Ransom (1988:367) points ou that that semantically "[...] signals the modality of the complement, namely that it is inclined to be true." 33

37 d. I demand that he leave. e. It's rather odd that a man is chairing a women's meeting. (Bresnan( 1972:71)) Comparing that complements with for complements, h becomes obvious that that complements display a higher degree of definiteness.36 (29) a. It would be odd that a man is chairing a women's meeting. b. It would be odd for a man to be chairing a women's meeting. (Bresnan (1972:72)) c. It is always desirable that housework be done by a woman. d. It is always desirable for housework to be done by a woman. (Bresnan (1972:74)) Although ybr complements are less specific than that complements, Bresnan (1972:79) considers it inadequate to reduce the meaning of for to degrees of specificity. Instead, she argues that, with respect to for, factors like intentionality have to be taken into consideration. As can be seen in (30), intentional predicates are precluded as complements of factive verbs: (30) a. *That John eats donuts meansybr him to be in a bad mood. b. *Our retiring to bed was meant to imply for him to leave. c. *We didnt require for him to be honest. These observations strongly support the claim that the phrase structure rule analysis is to be preferred to the transformational analysis. Bresnan (1972:44£E) introduces the convention of marking complementizers as [±wh]. The observations reported so far essentially motivated the introduction of the binary feature [+wh] as distinguishing complementizers as to whether the clauses they introduce semanticaliy correspond to questions [+wh] or propositions [-wh], Le. that or for complements.37 That and for, however, must be further distinguished in terms of finheness. Whereas that invariably introduces finite complement clauses, for is restricted to introducing non-finite complement clauses. The relevant deep structure configurations of interrogative complements, Le. embedded questions, and that and for complements are given in (31):

36

37

This is true of indicative (29a) and (29b) and subjunctive complements (29c) and (29d) (cf. Bresnan (1972:74ff.)). Bresnan (1970:310) points out that whether and i/are related, but she does not identify i/with [+wh].

38

(31) a. embedded question VP S1

COMP [+WH]

(Bresnan (1972:44,1970:319)) b. that and for complements VP

The desideratum that rule (18) be a definition for subordinate clauses, and only for subordinate clauses, which was formulated by Andersson (1973:7,10), so far seems to be met by Bresnan1 s analysis.38 In the following I want to present an outline of the impact Bresnan's (1970,1972,1973,1974) analyses had on the overall conception of sentence structure.

2.2 Towards Uniform Sentence Structure Having convincingly argued that noun clauses are generated by the phrase structure rule in (18), repeated below as (32), Bresnan (1970,1972) extends the S1 analysis to matrix interrogatives and declaratives39, the structures of which are given in (33). (32) S' ->COMPS (33) a. interrogatives S1

COMP

[+WH]

The set of elements which introduce adverbial subordinate clauses, however, is not considered. Relative clauses and comparative clauses, which I will not discuss, are derived by the same set of rules.

39

b. declaratives S' COMP

(Bresnan (1972:44,45,1970:319)) These two types of sentences or clauses to which these phrase structure representations (generated by (32)) correspond will be discussed in turn.

2.2. 1 Matrix Clauses are S's In this section I want to give an outline of the analysis of matrix clauses under the S' analysis. The mahi argument for extending the S1 analysis to matrix clauses comes from matrix whquestions, with which I will start the discussion.

2.2. 1.1 Matrix wA-questions are S's The set of matrix interrogatives relevant to the present discussion is given in (34): (34) a. b. c. d. e. f. g. h.

Who promised to do the dishes? Who(m) did you promise to do the dishes? What did you promise? Which book did you buy? How do you like the book? Where did you buy the book? When did you buy the book? Why did you buy the book?

They resemble indirect questions in being introduced by an interrogative element, L e. by an element specified as [+wh]. The interrogative pronouns who, who(m) and which are identical in form to the relative pronouns who, who(m) and which, but differ from these in that they are specified as [-Hwhj, whereas relative pronouns are specified as [-wh]. The elements how, where and when, as has been pointed out by Bresnan (1972:63,passun), are also specified as [+wh]. The same is true of why. The rule which derives matrix w/i-questions as in (34) is the wA-movement rule given in (35): (35) move wA-phrase into COMP (Chomsky (1977:85))

40 #7t-movement has the effect of marking the base position of the moved constituent, a complement hi (36a) - (36d) and adjuncts in (36e) - (36h), with a trace, t: (36) a. b· c. de · £

[g, [Q^ Who [g t promised to do the dishes]]]? Is Icaav Whom(m) [s did you promise t to do the dishes]]]? k tccn* What ts did you promise t]]]? Utc^ Which book [g did you buy t]]]? Is [ccaf How [s do you like the book t]]]? ts- tcoBop Where [s did you buy the book t]]]?

g- [s· tcoo* When [s did you buy the book t]]]? h. [s [ο«. Why [s did you buy the book t]]]? The internal structure of S, Le. the sister of Comp, in these examples is different from the internal structure of the S in embedded wA-initial sentences in that the auxiliary precedes the subject. This surface linear order is basically due to two transformations. The first transformation to apply is an auxiliary placement transformation'*0, the second transformation, a subject-auxiliary inversion transformation, reverses the linear order of subject and auxiliary. With the exception of subject questions (36a), these transformations are obligatory, as can be seen from the ungrammatical examples in (37) a. b. c. d. e. £ g. h.

*Who(m) promised you to do the dishes? *What promised you? *How like you this book? *Why bought you this book? *Who(m) you did promise to do the dishes? *What you did promise? *How you do like the book? "Why you did buy the book?

PFfc-movement hi matrix interrogatives, besides moving [4wh]-elements into COMP, involves transformations internal to S (e.g. auxiliary placement and subject-auxiliary inversion). Before turning to matrix declaratives, I want to briefly come back to embedded questions. Both types of interrogatives, Le. matrix interrogatives and embedded questions, can be schematically represented as (38).42 4

" This transformation applies obligatorily unless there is an auxiliary at the level of deep structure. Up to the 16th century, matrix questions in English were formed according to the Germanic verb-second pattern, i.e. there was no auxiliary insertion and hence no subsequent subject-auxiliary inversion: (i) Seis thou noght hit that situs the besyde? (ii) Why come dogges so often to the churche? (iii) What tyme in the yere bereth a gose moost feders? (taken from Roberts (1993:246)) Insertion of dummy do increased with the loss of inflectional morphology. 42 They differ in two respects: First, embedded questions are selected by a governing predicate, matrix questions are not. Second, matrix questions involve subject-auxiliary inversion, embedded questions do not (cf. chapter 6.4). 41

41

(Bresnan (1972:45,1970:319)) Bresnan (1972:62f), as discussed above, refers to -which, haw, what, where and when alongside whether as [+wh] complementizers. As we have seen in the examples in (36) and (38), there is good reason to assume that which, haw, what, where and when are wA-moved elements rather than base-generated complementizers. Bresnan includes in her discussion of embedded questions the elements who, who(m) and why. She describes the function of interrogative w/t-elements in terms of undetermination of the reference of the element they replace, thereby converting the sentences which they introduce into open sentences.43 On the basis of this assumption, she correctly predicts that these constructions qualify as complements to interrogative and dubhative predicates, which "[...] leave aspects of their complements open to question [...]" (Bresnan (1972:65)). If this is a defining characteristic of embedded questions, then this analysis should carry over to who, who(m) and why, Le. to elements which, as we have seen in (36), are w/i-moved elements. The embedded questions in (39) are introduced by who, who(m) and why. They occur freely as complements to interrogative and dubitative predicates and are precluded from functioning as complements to fäctive verbs and verba dicendi or sentiendi (40), Le. they pattern distributionalry like complements introduced by which, how, what, where, when and whether. (39) a. b. c. d. e.

She wondered who had done the dishes. She wondered who(m) she might invite. She didn't know who(m) to invite. She pretended to know why he did it. She asked which one he would like to have.

(40) a. b. c. d.

*He says who did it. *He regretted why H should be this way. *We feared who(m) to invite. *I demand who will do the dishes.

Matrix w/j-questions and embedded questions do not differ at the highest level, Le. they are instances of S' with a Comp being filled by some [4-whJ-element.44 Evidence for not treating who, who(m), why, which, how, what, where and when on a par with whether comes from the observation that all [+wh]-elements except whether introduce

43 44

Cf. Bresnan (1972:65). Andersson (1973:7,10;1975:155) argues that only embedded clauses are instances of S'. Matrix ^-questions should be analyzed as instances of S. Subject-auxiliary inversion would then be triggered because the subject is preceded by a non-null element.

42

both matrix and embedded questions.45 Whether, as can be seen in (4 la) and (41b), is precluded from matrix questions. Embedded w>Ae/Aer-questions are alternative questions, this function is captured in matrix questions by moving the auxiliary into sentence initial positon, Le. by inverting it with the subject. (41) a. b. c. d.

*Whether will he come on time? 'Whether did he do the dishes? Will he come on tune? Did he do the dishes?

As we shall see in the following section, there is a general rule that deletes complementizers in matrix clauses, irrespective of whether the matrix clause is interrogative or declarative. Before turning to discuss the consequences of the S' analysis for matrix declarative sentences, I want to briefly come back to Geis' (1970) analysis of wA-initial spatial and temporal adverbial clauses as in (42), which he takes to be relative clauses whose antecedent has been deleted46: (42) a. I found it at the place where I had put it -> I found it 0 -where I had put it b. I dont know the time when he will return. ->· I dont know 0 when he will return. Independent of whether one adopts his analysis of wA-inhial adverbial clauses as relative clauses - which raises the question of how these adverbial clauses are distributed (cf 1.1.3) the formation of these adverbial clauses parallels the formation of relative clauses, i.e. when and where are wA-moved and left-adjoined to the complementizer47, that is to say that whinitial clauses, irrespective of whether they function as embedded questions or as adverbial complements, are captured by the structure given in (38) above.

2.2.1.2 Matrix Declaratives are S's If the S1 analysis, which has proved adequate with respect to the subordinate clauses discussed so far, carries over to mam declaratives, the schematic representation must be given as hi (43). (43)

S' COMP

[-WH]

(Bresnan (1972:45,1970:319)) 45 46 47

Cf. 6.4 for discussion: whether is a specC element See, for example, Jaworska (1982) and Jacobs/Rosenbaum (1969) for similar analyses. Similar to the pronouns who, who(m) and which, which can be either [+wh], i.e. interrogative pronouns, or [-wh], i.e. relative pronouns, where and when must be distinguished in terms of [+wh]; the [+wh] variant qualifies as an interrogative proform in matrix and embedded questions, the [-wh] variant qualifies as adverbial subordinator.

43

Matrix declarative sentences are introduced neither by a complementizer, nor by a wA-element: (44) a. *That he did the dishes.48 b. *That I left early. If they were introduced by a [-wh]-element, they would not be matrix clauses. If they were introduced by a [+wh]-element, contrary to the [-wh]-marking in (43), they would not be declaratives. Since there are no matrix declaratives with a lexically filled complementizer position, a reasonable question to ask is whether matrix declaratives are in fact instances of S' rather than instances of simply S. The main reason for describing both subordinate clauses and matrix clauses as instances of 1 S rests on the observation that the differences between the two kinds of sentences are slight. Within the generative framework, any superficial difference between these two is assumed to be derivational, i.e. the result of transformational rules applying to deep structures, so that there is no need to assume different rules to derive subordinate clauses and matrix clauses.49 As we shall see in later chapters, this perspective on sentence structure has been dominant up to now Since under the generalized S'-analysis subordinate clauses as well as matrix clauses are generated by (18), repeated as (45), (45) S1 -+COMPS the complementizer deletion transformation must apply obligatorily in matrix clauses.

2.3 Summary In this chapter I have provided a representation of the transformational approach to complementizers (2.1.1) and the phrase structure rule approach to complementizers (2.1.2). The comparison of these two approaches showed that the phrase structure rule approach is to be preferred over the transformational approach. The reasons for discarding the transformational approach in favor of the phrase structure rule approach are threefold. In the first place, the phrase structure rule approach recognizes the fact that complementizers are not void of semantic content in that they contribute to the interpretation of the sentences they introduce. Attributing to complementizers semantic content makes it impossible to insert them transformationally, as only elements which are not "meaning-bearing elements" can be transformationally inserted.50 Second, recognizing that complementizers have inherent semantic meaning makes it possible to capture both the relation between the embedded clause introduced by a comple4 49

50

(44a) is grammatical if interpreted as an exclamative. Cf. Grimshaw (1979) for discussion. This assumption is expressed in the following quote taken from Rosenbaum (1967:25):"[...] there is no structure underlying any declarative sentence in English that cannot, in some other derivation, be the structure underlying a predicate complement sentence." See, for example, Chomsky (1965/161990:132) and Lakoff (1968:20).

44

mentizer and the governing head and the relation between the complementizer and the semantic type of the clause it introduces. Third, idiosyncratic transformations can be discarded in favor of generalizations and thus the transformational burden can be reduced. Compelling evidence for the uniform structure of sentences comes from the analysis of embedded questions and matrix questions (2.2). These constructs are uniformly generated by the rule S1 -> COMP S. Surface structure differences between the various types of sentences are based on transformations and thus derivative. Powerful as this approach is, it does not - at least not in an obvious way - adequately describe adverbial subordinate clauses introduced by elements such as after, although, before, because, since, until, in case, while, etc. In the following chapter I turn to the syntactic structure of adverbial clauses. The major problem for their description, as will be seen, lies in the categorization of adverbial subordinating conjunctions. Elements like after, although, before, because, since, until, in case, while, etc. have been - and still are - categorized as complementizers or as prepositions. In the first part of the following chapter the characterizations of adverbial subordinating conjunctions will be reviewed from a traditional grammar point of view (3.1). Whereas in 3.2 the main arguments for categorizing adverbial subordinating conjunctions as complementizers will be considered, 3.3 focusses on the arguments in favor of categorizing adverbial subordinating conjunctions as prepositions.

3 Subordinating Conjunctions - A Question of Categorization According to the phrase structure rule approach to complementizers presented in chapter 2, sentence structure, at the level of deep structure, is uniform. Phrase structure rule (1) is taken to generate any sentence, Le. it is a generalization of sentence structure. (1) S'-»COMPS Arguments for the deep structure node COMP were presented in connection with complement clauses, both declarative and interrogative, matrix interrogatives and matrix declaratives. Central to the discussion of subordinate clauses in chapter 2 were the complementizers that and for and the wA-elements. Adverbial subordinate clauses, as was pointed out in 1.1.2, constitute the largest and most diverse set of subordinate clauses. This set has always been subclassified in terms of the relations the subordinating conjunctions specify. Traditional grammar recognizes adversative, causative, comparative, concessive, conditional, consecutive, manner, purposive, spatial and temporal adverbial clauses. From a generalized phrase structure perspective, the obvious question to be raised is where and how subordinating conjunctions are introduced in deep structure. A prerequisite to answering this question is knowing what the syntactic category of these elements is. The term subordinating conjunction, convenient as it is, is not precise enough as it is applicable to complementizers, such as that and ./or1 as well, and thus presupposes categorial identity of subordinating elements. Before I turn to the categorizations of subordinating elements which have been proposed in the relevant linguistic literature, I shall sketch the characterization of these elements in traditional grammar. Here already the categorization of subordinating elements proved problematic. As will be seen in 3.2. and 3.3, some of the characterizations and insights presented in the following section were adopted in generative grammar, leading to reclassifications. A summary is presented in 3.4.

3.1 Subordinating Conjunctions in Traditional Grammar Most traditional grammarians make a general distinction between the subordinating elements in (2) and those in (3).2

1 2

Cf. chapter 2. The examples are taken from Quirk et al. (1985:1049), Allen/Mason (1939/1965:204), Quirk et al. (1985: 1260,1047), Blake (1988:128), Allen/Mason (1939/1965:243), Pence/Emery (1947:102), Quirk et al. (1985: 1248,1247,1257,1081,1089,1089,1104,1104,1104,1101), Jespersen (1909-1949,Part V:346,348), Allen/ Mason (1939/1965:208).

46 (2) a. That the invading troops have been withdrawn has not affected our government's trade sanctions. b. I believed their intention to be that I should remain there only for the night. c. The belief that no one is infallible is well founded. d. You can tell whoever is waiting that Γ11 be back in ten minutes. e. He doubted whether they were at home. f. He asked me what I had done. g. I don't know how to tie a bow tie. h. They are delighted with the book that has just appeared. i. The woman who is approaching us seems to be somebody I know. j. This excellent book, which has just been reviewed, was published a year ago. (3) a. b. c. d. e. f. g. h. i. j.

He died before writing a will. Unless the strike has been called off, there will be no trains tomorrow. //"Colin is in London, he is undoubtedly staying at the Hilton. She watered the flowers because they were dry. Since the weather has improved, the game will be held as planned. Percy is k Washington, for he phoned me from there. Whether I shout at them or plead with them, I can't keep them quiet, I shall go to Stratford, where Shakespeare was bom and died. When I was 17, my father died. When a few weeks had elapsed, the whole affair was forgotten.

This general distinction is based on the formal morphological properties of the hems under consideration. In traditional grammar, the universe of lexical items is split into open class items and closed class hems.3 Elements of open classes, i.e. verbs, nouns, pronouns and adjectives, are inflecting elements. Open class elements take part in morphological processes such as inflection, derivation and compounding. Thus, open classes can be enlarged. The relative pronoun who and hs interrogative counterpart clearly display inflectional properties. The other elements, i.e. that, whoever, whether, how, what and which are subsumed under the category noun either because they serve to introduce noun clauses as in (2a) - (2g) or because they have a nominal antecedent as in the relative clauses hi (2h) - (2j). It is only by virtue of then: being related to the nominal system that these elements are attributed to the set of open classes, namely as pronouns. The set of non-inflecting elements which is standardly referred to as particles comprises adverbs, prepositions and conjunctions.4 These elements are characterized as constitumg socalled closed classes, that is to say that the elements do not have different morphological 3 In recent linguistic theory, we find the lexical-functional dichotomy dividing the universe of lexical items into two sets (cf. chapter 4.2). 4 Cf. Schibsbye (1965), Jespersen (1924/1992), Dudengrammatik (41984), Sweet (1891), Wackernagel (1926), Poutsma (1926), etc. I have left out coordinating conjunctions and interjections, since they will not be relevant at any point in this study.

47 shapes, i.e. they do not have inflectional paradigms. Since these elements are not subject to morphological processes such as derivation and compounding, the class cannot be enlarged.5 Whereas the subordinators in (3) are almost universally classified as subordinating conjunctions, i.e. as closed class items, the subordinators in (2) are classified as pronouns, i.e. they are open class items, although only the pronoun who displays inflectional properties. Although the italicized elements in (2) and (3) share the formal property of marking the clause they introduce as subordinate, they are not considered to belong to one syntactic category. Since this chapter is on subordinating conjunctions like those in (3a) - (3f), i.e. on elements that are neither affected by w/i-movement nor obviously belong to the category complementizer (cf chapter 2), I will not present traditional classifications of the subordinators in (2) and (3g)-(3j)-6 Subordinating conjunctions, as mentioned above, are considered to be members of a closed class, which is often referred to as "particles". Elements of this class are characterized as relation words (cf. Deutschbein (1959), Quirk et al. (1985)), or as Schibsbye (1965:289) puts it, as "bilaterally connected particles". As has been pointed out by most traditional grammarians, there is a vast amount of categorial overlap with subordinating conjunctions, prepositions and adverbs. Subordinating conjunctions and prepositions, although they share the relating function, differ insofar as the relata are of two kinds: sentences and noun phrases. As is noticed by most authors, there are particles which are polyfunctional, i.e. one and the same hem can, for example, function as a subordinating conjunction, a preposition and an adverb (4) or as a conjunction and a preposition (5). (4) a. I saw him before the party started, (conjunction) b. I saw him before the party, (preposition) c. I saw him before, (adverb) (5) a. b. c. d. e. f.

after his arrival after he had arrived before his breakfast before he had breakfast because of his absence7 because he was absent (Jespersen (1924/1992:89))

It should be noted, however, that particles, i.e. conjunctions, prepositions and adverbs both synchronically and diachronically are subject to word-formation processes. Within this class we find forms such as into, because (by cause), unless (on less), while, etc. The subordinating elements where and when are often treated as a separate class. Again, classifications vary from author to author. Allen/Mason (1939/1965:211), Sweet (1891/1968:133) and Pence/Emery (1947:101ff.), for example, classify them as relative adverbs, Jespersen (1909-1949,Part V:344) simply as relative words or as pronominal adverbs (Jespersen 1924/1992:85). Jespersen includes whether as in (3g). Jespersen ascribes the presence of of in this example to the nominal origin of this preposition.

48 These observations led Jespersen (1924/1992:89) to conclude that the fact that one and the same word, sometimes with slight differences, is used as a subordinating conjunction and as a preposition "[...] should not make us hesitate to affirm the essential identity of prepositions and conjunctions [...]". This statement, from a more recent linguistic perspective, can be taken to carry over to a subclass of adverbs as well.8 Sweet notes that "when an adverb introduces a sentence as a modifier [...] of the whole sentence, the adverb is indistinguishable from a conjunction" (1891:133). In a slightly different way, Deutschbein (1959:250) argues that, primarily from a diachronic point of view, the set of elements which is referred to as subordinating conjunctions is for from being homogeneous, i.e. the borderline between conjunctions and prepositions on the one hand, and the borderline between conjunctions and adverbs and prepositons and adverbs on the other hand, is fuzzy. Subordinating conjunctions and prepositions, in Deutschbein's grammar, are distinct subclasses of so-called relation words. Allen/Mason (1939/1965:259f), on the other hand, stick to the more conservative position hi claiming that the distinction between a subordinating conjunction introducing a clause and a preposition governing a noun is a necessary one. Categoryhood, here, is determined by the complement of the relation word. The lexical item till, for example, is classified as a subordinating conjunction if it is followed by a clause (6) and it is classified as a preposition if it governs a noun (7): (6) We decided to wait till dawn came. (7) We decided to wait //'// dawn. (Allen/Mason (1939/1965:260)) The lowest common denominator with respect to the characterization of subordinating conjunctions presented so far is the following: subordinating conjunctions are relation words that introduce sentence-like elements. In traditional grammar, subordinating conjunctions are considered to be elements which, because they specify specific relations between their relata, are not assumed to be void of semantic content9. Among these relations, as has been pointed out above, adversative, cause, comparative, concessive, conditional, consecutive, correlative, manner, purposive, spatial and temporal relations can be identified. In the following two sections, I will discuss the categorization of subordinating conjunctions in generative grammar. Chapter 3.2 deals with the arguments for categorizing subordinating conjunctions as complementizers. The categorization of subordinating conjunctions as prepositions will be central to 3.3.

8

9

In his (1909-1949,Part V:344) work, though, Jespersen lists subordinating elements as belonging to the following subcategories: conjunctions proper, relative words, adverbs, prepositions and preposition and regimen. This position, as pointed out in chapter 2, is not adopted in generative grammar until the 1970s. Structuralists would refute this claim outright. Cf. Fries (1952), Hockett (1958).

49

3.2 Subordinating Conjunctions as Complementizers Subordinating conjunctions like because, after, although, etc., in very much the same way as complementizers such as that, than, for, etc. indicate that the sentence they introduce is an embedded clause.10 As opposed to subordinate clauses introduced by a complementizer, e.g. that and for-to, subordinate clauses that are introduced by a subordinating conjunction are always adverbial clauses, ie. clauses which are precluded from the complement position of verbs but which figure prominently within VP or the whole sentence.1l The main reason for categorizing subordinating conjunctions as complementizers is a consequence of the system which takes sentences, whatever their surface form is, to be identical in deep structure. The complementizer, as we have seen in chapter 2, is taken to be obligatory in any sentence; hence, any sentence is an instance of S*. The initial position of sentences may contain either a w/i-moved constituent, which is left-adjoined to Comp, or a complementizer occupying Comp or it may be empty as a consequence of deletion. The assumption that sentences are uniform allows all elements that introduce subordinate clauses to be treated alike. This is implicit in Andersson's (1973:2) statement: The name of this category is of minor importance, [t]he important thing is that we have such a category. Under this analysis, which takes subordinating conjunctions to be identical with complementizers, the original characterization of complementizers as elements which "[...] are a function of predicate complementation and not the property of any particular sentence or set of sentences." (Rosenbaum (1967:25)) is lost. Complementizers are no longer elements whose function it is to assign sentences the status of complements in the relevant sense12, rather the notion complementizer is broadened to include elements that occupy a pre-S position and indicate that the clause they introduce is a subordinate one. According to this analysis, then, subordinate clauses such as those in (8) (8) a. Unless the strike has been called off, there will be no trams tomorrow. b. She watered the flowers because they were dry. c. Since the -weather has improved, the game will be held as planned. correspond to the by now familiar structure given in (9) which is generated by phrase structure rule(l).

Cf. Handke (1984), Andersson (1975) for pragmatic conditions on the usage of adverbial subordinate clause patterns as matrix declaratives. Cf. van Riemsdijk (1978:141), Hendrick (1976:116). The original characterization of complementizer, as will be seen in chapter 6, is revived in more recent linguistic work, where elements such as that are conceived as nominalizers. Cf. Webelhuth (1989,1990), Ouhalla (1991), Kayne (1982), Chierchia (1985), Williams (1987a,1987b,1994).

50 (9)

S' ^-" Comp i unless because since

the strike has been called off they were dry the weather has improved

This structure, as, for example, Hendrick (1976:117) and Lasnik/Saito (1992:113) argue, correctly predicts that sequences such as unless that, because that, since that, etc. are ungrammatical: ^ (10) a. * Unless that the strike has been called off, there will be no trains tomorrow. b. *She watered the flowers because that they were dry. c. "Since that the weather has improved, the game will be held as planned. Sequences like unless that, because that, since that, etc., they argue, could not be ruled out in a straightforward way if the subordinating conjunctions occured in a pre-S' position.14 The analysis of subordinate clauses as given in (9), however, does not yield the correct results with respect to adverbial subordinate clauses as they existed up to the 17th century. As the examples in (11) show, virtually every subordinating conjunction could co-occur with that: (11) a. for gifthat shrewdness maketh wretches b. whil I have tyme and space, er that I further in this tale pace (taken from Lightfoot (1979:322)) c. After that the king had burnt the roule d. Fra patt he wass full litell till patt he waxen was e. While that thou and I were coupled to geders, thou madest me to lede a fill vnthryfty lyf (taken from The Oxford English Dictionary) f. although that words chain'd with affection faile g. But neighbours I pray you be not angr Because that I am to bolde and homly h. where Desolation Was sole Inhabitant, until that one Poor Ermite chose his tamest habitation i. Although before that you did say, you forc'd me not a pin. j. Ο what a burden nature, do'st thou beare, Since that to sinne and live, seeme both but one! (taken from English Poetry (1994))

'·* Complementizers, as has been already pointed out by Bresnan (1970,1972), are mutually exclusive. 14 Cf. 3.3., where subordinating conjunctions are analyzed as prepositions taking S' complements. It should be noted that, in contemporary French, adverbial subordinate clauses are introduced by P-like elements and the complementizer que: apres qve, avant que, etc.

51

With respect to data as in (11), h has often been argued that subordinating elements - whatever their category - preceding that have been recategorized as complementizers, thus making superfluous the presence ofthat as a subordination marker.15 Hendrick (1976:116f£) refutes Jackendoffs (1973) and Emonds1 (1976) proposals to analyze subordinate clauses as instances of PP, with the subordinating conjunction being a preposition taking a sentential complement (cf. 3.3). His argumentation in favor of categorizing subordinating conjunctions as complementizers is based on the observation that adverbial subordinate clauses as opposed to PPs never function as complements, i.e. they are never daughters of V but sisters of V (12). PPs, on the other hand, may be sisters (13) and daughters of V ( 14). (12) a. John is tired because he went to bed late. b. We wont be satisfied until our demands are met. (Hendrick (1976:116)) c. *He assumed because he went to bed late. d. *He scheduled the ratification until our demands are met. (13) a. The hand-over, due in September, was delayed because of the Berlin crisis. (LOB) b. We never had to face this problem until last week. (14) a. Harpo ran down the street toward the garbage collector, b. The plane went from New York to London. (Hendrick (1976:108, 110)) Hendrick (1976:117) points out that the difference between the ungrammatical examples in (12c) and (12d) and the grammatical ones in (14) cannot be accounted for if subordinate clauses are analyzed as PP.16 Another argument which Hendrick (1976) produces for not treating subordinate clauses and PPs alike is based on wA-movement. As has been pointed out by Hendrick (1976), Emonds (1976) and others, movement out of constituents is possible only if the constituents are complements, i.e. sisters of V not sisters of V: (15) a. It is likely that California will elect Reagan. b. Who is it likely that California will elect f>. c. John is tired because he -went to bed late. d. * What is John tired because he went to t late? (adapted from Hendrick (1976:116ff))

13 Dubinsky/Williams (1995), for example, argue in favor of recategorizing prepositions as complementizers. 16 N.B. the lexical items introducing the subordinate clauses in (12c) and (12d) and the PPs in (13a) and (13b) are identical.

52

In (lSa), the subordinate clause introduced by that is a complement, hence extraction out of the subordinate clause yields a grammatical result (15b). The subordinate clause in (15c) is an adjunct, Le. a sister of V, hence extraction out of h is illicit as the ungrammatical string in (15d) shows. Consider the examples in (16) and (17), where both the PPs and the subordinate clauses are sisters to V.

(16) a. John wrote a paper on epithets for the New York conference. b. Which conference did John write a paper on epithetsybr f? c. John visited his aunt in San Francisco on his vacation. d. Which vacation did John visit his aunt on /? (adapted from Hendrick (1976:117)) (17) a. John is tired because he went to bed late. b. *What is John tired because he went to t late? c. We wont be satisfied until our demands are met by the officials. d. *Who wont be satisfied until our demands are met by t? (adapted from Hendrick (1976:116f)) On the basis of the prohibition that no element can be extracted out of a non-complement constituent, we would expect the examples in (16) to be as ungrammatical as those in (17). This expectation, however, is not borne out. In order to explain why extraction hi (16) yields grammatical results, Hendrick (1976:100f£) proposes a Complement Formation Transformation which may apply to PPs that are daughters of V" and sisters of V, lowering them into V, thereby assigning them sister status with respect to V. This rule applies to PPs but not to subordinate clauses.17 Hendrick attributes the ungrammaticatity of (17b) and (17d) to the feet that "[...] certain COMP-S sequences can be complements to V and others cannot." (Hendrick (1976:117)) If preposition lowering, Le. Complement Formation Transformation, were a valid option, the examples in (18) should be grammatical, which they are not. The prepositions - and according to Hendrick these are prepositions since they take an NP complement - do not permit extraction of their complement. (18) a. "What did they leave notwithstanding/? b. * Which break should we leave during f! (adapted from van Riemsdijk (1978:145)) c. *What was the hand-over, due in September, was delayed because of/? d. "When didnt we have to face this problem until fl

17

The preposition lowering rule, which is assumed to be structure preserving, goes back to Emonds (1976:176ff.).

53

A third argument for categorizing subordinating conjunctions as complementizers and adverbial subordinate clauses as instances of S' rather than PP is based on the observation that Sentence Adverb Fronting is possible only if a complementizer is present:18 (19) a. John said that, hopeßilly, he would arrive tomorrow. b. John will be in Paris next year, because, hopeßilly, hell get a Fulbright. c. * John said, hopeßilly, he would arrive tomorrow. (Hendrick(1976:118)) d. e.

Although never before in her life was her painting career interrupted, she helped her husband and continued to accumulate painting knowledge. So if you are made of atoms, you are just a big machine; and since should my theory prove right the universe is also made of atoms, it is just a supennachine.19 (Based on data taken from the LOB-corpus)

Sentence Adverb Fronting, in Hendrick (1976:118), is taken to be a diagnostic test for categoryhood. If all subordinating conjunctions, as Hendrick (1976) claims, were complementizers, adverb proposing should always be possible, which is not the case as can be seen in (20): (20) a. *She got nervous while at the conference she was waiting for her publisher to show up. b. *You will have to wait with your novel until when you are in London you find a publisher. c. *[...] this was easily the most heartening exhibition they had had since while quite sucessful Ben Hogan fell upon evil ways during his heyday [...]. d. *I seemed to know what they were going to say long before while we were driving home they said it. e. *After a few hours ago Mr Lloyd and his delegation landed at Accra this morning, hundreds of shop assistants demonstrated [...].20 On the basis of the data reconsidered in this section, the arguments produced in favor of analyzing subordinating conjunctions as complementizers prove to be too weak In the following section, I turn to analyses which take subordinating conjunctions to be prepositions.

18

This observation is due to Wexler/Culicover (1976). ^ It should be noted that this example is grammatical under the intended causal reading of since, but not under the temporal one (cf. (20c)). 20 Examples (20c)-(20e) are based on data taken from the LOB-corpus.

54

3.3 Subordinating Conjunctions as Prepositions As has already been observed in traditional grammar, there is a vast amount of categorial overlap in the so-called particle system. Elements such as before, after, since, until, because, etc. are often considered poryfuno tional, i.e. they may function as subordinating conjunction, preposition and adverb (21) or as preposition and conjunction (22). (21) a. I saw him before the party started, (conjunction) b. I saw him before the party, (preposition) c. I saw him before, (adverb) (22) a. b. c. d. e. f.

after his arrival before his breakfast because of his absence after he had arrived before he had breakfast because he was absent (Jespersen (1924/1992:89))

On the basis of data such as in (21) - (22) Jespersen (1924/1992:89) proposed "[...] to affirm the essential identity of prepositions and conjunctions just as we put all verbs in one class in spite of the fact that they cannot all take a complementary clause." Subordinating conjunctions, in his view, are nothing but sentence prepositions. The difference between subordinating conjunctions and prepositions lies in the complements they take, sentences and NP, respectively. Similarly, he argues with respect to on and in in (23) (23) a. Put your hat on (your head), b. He went in (the house). that it would be more natural"[...] to include them in one class and say that on and in are sometimes complete in themselves and sometimes followed by a complement [...]", instead of assigning these elements to two separate classes, adverb and preposition. These insights were first systematically incorporated into linguistic theory in Jackendoff (1973, 1977) and Emonds (1976).21 Instead of having bicategorial lexical items such as until, because, etc. or mufticategorial items such as before, since, etc., these elements are collapsed into one syntactic category, namely preposition. The fact that elements of this class take NP or 21

Freidin (1971:62f.) considers the option of introducing a phrase structure rule introducing sentences as complements of prepositions.

55

sentence complements or can be used intransitively, i.e. their poryfunctional nature, is accounted for by individual subcategorization frames.22 Before, as can be seen in (21) above, can be used transitively, Le. with an NP complement (21b) and with a sentential complement (2la), or it can be used intransitively (21c). After and because allow for an NP complement (22a) and (22e) or for a sentential complement (22b) and (22f). With respect to the range of complement types, since, as can be seen in (24), patterns like before, as given in (21) above. (24) a. I haven't seen him since Christmas. b. I haven't seen him since the party began. c. I haven't seen him since. After and until, in (25), unlike before and since, cannot be used intransitively. (25) a. She waited until midnight. b. She waited until dawn came. c. * She waited until d. She left after his arrival e. She left after he had arrived f *She left after. While, for example, allows only a sentential complement. NPs are precluded from the complement position. (26) a. She got nervous while she was waiting for her publisher to show up. b. * She got nervous while dinner. c. * She got nervous while. During, on the other hand, allows only an NP complement: (27) a. She got nervous during dinner. b. *She got nervous during she was waiting for her publisher to show up. c. *She got nervous during. Information about the type of complement these items take is encoded in the strict subcategorization frame. (28) covers before and since, (29) covers after, because and until, (30) while and (31) during.

(28) P-»[_NP]

Information about the categorial type of the complement is lexeme-specific and hence has to be encoded in the lexical entry.

56

(29) P->[_NP]23 (30)

(31) P->[_NP] Further evidence for collapsing prepositions, subordinating conjunctions and certain adverbs into one syntactic category comes from specification, modification, substitution by pro-forms, coordination, clefting and preposing.24 Emonds (1976:174), for example, points out that the particle right can only combine with temporal, spatial and directional prepositions (32a) (32d), but not with adjectives and manner or frequency adverbials. (32) a. b. c. d. e. f g. h.

He kept on drinking right until midnight. They kissed right after the ceremony. He came right before the party started. He kissed her right while her boyfriend was looking. *They came over right often. *John drove right carefully. *She is right pretty. *Fights happen right seldom around here. (Emonds (1976: 174))

Jackendoff (1977:79, 137) points out that PPs may contain modifiers. Modification applies irrespective of whether the complement of P is an NP or a sentence. (33) a. He h&ßve minutes before dinner. b. He le&ßve minutes after Bill had left. PPs, in very much the same way as NPs, may be replaced by pro-forms: (34) a. He left five minutes before dinner. b. He left five minutes after Bill had left.

c. He left then. d. When did he leave? Then may replace the sequence P-NP as well as the sequence P-S'; similarly when, as an interrogative pro-form, may question P-NP or P-S1.

23 24

In the case of because, the NP complement has to be introduced by of. Clefting and preposing, however, do not apply to all subcategories of subordinate clauses; comparative clauses, for example, resist preposing (cf. Emonds (1985:285)).

57

Coordination is considered a diagnostic test both for constituency and categorial identity as well as syntactic function.25 Elements which are coordinated must be identical with respect to phrase structure status, syntactic function and categoryhood and they must be semantically compatible: (35) a. He left before dinner and after Bill had left. b. We could try to meet while the conference is going on or after Christmas. c. *We met in room 32a of the British Museum and while it was raining. d. *We refused to meet at the exhibition and while it was raining. (35c) is ruled out because the spatial PP and the temporal PP are semantically incompatible, (35d) is ruled out because the non-finite clause and the wA/Ve-clause are functionally incompatible, the former is a complement, the latter an adjunct. These syntactic phenomena are taken as evidence for treating prepositions, subordinating conjunctions and certain adverbs alike. As has been pointed out in chapter 1, phrase structure, in X-bar Theory in the phrase structure rule-driven version which I use here as well as in most of the projective versions, which will be discussed in chapter 4 it is held that phrases are endocentric and parallel in structure. The syntactic phenomena sketched in (32) - (35) led automatically to the extension of the generalized phrase structure in (36) to prepositions.26 (36) a.

Xn->...X°-1... (Jackendoff(1977:30))

X1"

b. specifier

X" X1

X

non-restrictive modifiers restrictive modifiers

complement

Although this analysis allows for parallels between categories to be captured and consequently for further generalizations in the phrase structure component to be made, there remain at least two open questions. The first question is a very general one. Does the categorization of subordinating conjunctions as prepositions extend to all subordinating conjunctions or is it restricted to those elements which traditional grammar described as poryfunctional?

25 26

Cf. Radford(1988:69f.), Jackendoff(1977:50ff.). Cf. especially Jackendoff (1973,1977), Emonds (1970,1976,1985), van Riemsdijk (1978).

58

Emonds (1985:248ffi), for example, claims that all subordinating conjunctions with semantic content should be analyzed as prepositions.27 The arguments behind this claim correspond to the diagnostics given in (32) - (35), and in addition to these, he mentions the ability of PPs to undergo clefting and proposing. As was pointed out in the previous section, PPs of the form P-S' are precluded from complement positions. In his discussion of the examples in (37), Emonds (1985:249) states that"[...] before is a P which appears outside X1, regardless of whether P has an NP or an S complement."28 (37) a. John had several meetings before lunch. b. *John put his meeting before he left to eat. c. John had several meetings before he left to eat. (Emonds (1985:282)) He categorizes not only subordinating conjunctions as prepositions but also those elements which, since Bresnan (1970,1972), have been referred to as complementizers. Prepositions, in his system (1985:249,282, passim) come in three variants: (38) a. lexical prepositions: before, after, until, while, etc... b. grammaticalized prepositions: that, for, whether, i/29 c. complex or compound prepositions: because of, in case of, as for, now, that, in order that, etc. (the last element in each case is considered a grammaticalized P) A grammaticalized preposition, according to Emonds (1985:169), is a prepositional item"[...] that contains no purely semantic feature hi its lexical entry [...]". Introducing this subcategory of prepositions, which is taken to replace complementizers, has a side effect which leads immediately to the second open question. What is the status of the sentential complement? Is it an instance of S or an instance of S'? Emonds argues that there is no evidence for assigning the sentential complement the status of S1. Apart from complex or compound prepositions as in (38c), there are no constructions in which a preposition is followed by a complementizer, i.e. a grammaticalized P in his termi-

27

This argument, as Emonds (1985:248) points out, goes back to Klima. It should be noted, however, that prepositions in Emonds' (1985:159) system are not considered to be members of a lexical category but rather grammatical formatives. 28 Emonds (1985:291) points out that there are instances of P-S sequences occurring inside X*, e.g.: some PPs with semantic content (headed by 'where, as if), as well as PPs headed by grammaticalized P (formerly COMP). He further points out that prepositions"[...] of time, causality, and implication are excluded inside V equally well in both S and NP contexts." 29 For, that, whether and //are further distinguished in terms of [±wh], with //and whether being [+wh] and that ana for [-wh]. These elements also differ with respect to finiteness requirements on their sentential complement. That and // invariably take a finite sentential complement, ybr a non-finite sentential complement and -whether may take either a finite or non-finite sentential complement.

59

nology. Emonds (1985:251) claims that "[...] no COMP morpheme is ever permitted in P S."30 PPs headed by elements of the prepositional subcategory grammaticalized P are restricted to complement positions, i.e. they are strictly subcategorized for.31 Emonds (1985:304ff.), in my view anachronistically, argues that, since these elements are subcategorized for, they are not present in deep structure but inserted by a late lexical insertion rule.32 The context in which this rule applies is given hi (39).

(39)

S

Under Emonds1 (1985) analysis, sentences are reanalyzed as PP and COMP as P. FfTi-movement, under the analysis outlined in 2.2.1.1, creates a left-adjoined position to COMP; in the present system, however, w/i-movement is movement into P (c£ Emonds (1985:299ff.)). There is no adjunction since the target of w/j-movement, by definition, is an empty P: (40) A COMP is a P which dominates no lexical terminal element during the derivation of s-structure. (Emonds (1985:301)) By (40) the co-occurrence of a w/i-element and that, for example, is ruled out.33 With respect to the examples in (11), however, some of which are repeated in (41), Emonds1 (1985) analysis faces the same problems raised for the analysis presented in 3.2. (41) a. b. c. d.

for gifthat shrewdness maketh wretches whil I have tyme and space, er that I further hi this tale pace After that the king had burnt the roule While that thou and I were coupled to geders, thou madest me to lede a ful vnthryfty lyf e. although that words chain'd with affection faile f. But neighbours I pray you be not angr Because that I am to bolde and homry

In (4la) two grammaticalized Ps co-occur, in (41b) - (4If) a lexical preposition is followed by a grammaticalized one, without combining to form a compound preposition. 30

The COMP morpheme, as mentioned above, is assigned to the syntactic category P.

31

Cf. 1.1.1, 2.2 and 3.1. This, I think, comes close to the Complementizer Insertion Transformation discussed and rejected in chapter 2. A variant of this late lexical insertion rule is the so-called o/insertion rule (cf. Emonds (1985:304), Chomsky (1981:50,1986b: 192f.)). (40) is also considered a good explanation of why PPs can undergo u*-movement. Cf. Chomsky/Lasnik (1977:446) for the doubly-filled Compfilter or Lasnik/Saito (1992:3) for a more recent version of this filter.

32

33

60

These examples can be accounted for if the sentential complement of P is analyzed as S'. Jackendoff (1977) and van Riemsdijk (1978), for example, propose this analysis.

(42) P ->PS' With respect to Modem Standard English, however, the complementizer that in S' must be deleted obligatorily (cf Jackendoff (1977:79)). Van Riemsdijk (1978), referring to Dutch, states that the sentential complement of P must be S' (=V") since it may be introduced by a complementizer:34 (43) a. Nadat zij waren vertrokken was alles nistig. After that they were left was everything quiet. After they had left, everything was quiet. (adapted from van Riemsdijk (1978:60)) b. Piet heeft het gemaakt, doordat ie door Datting is gedwongen de school te doorloopen Piet has it made by that he by Datting is forced the school to pass through Piet has made it because he has been forced by Datting to finish his school (adapted from Reuland (1979:151)) Jackendoff (1977), Reuland (1979) and van Riemsdijk (1978), unlike Emonds (1985), do not reanalyze complementizers, at least the complementizer that, as prepositions, thereby allowing the phrase structure rules to account for Middle English data (41) and Dutch data (43). Having an S1 complement to P, on the other hand, structurally allows for clause-bound vvhmovement as well as for long wA-movement, both of which are ruled out in English: (44) a. "He left the party before what had he eaten /. b. * What did he leave the party before / he had eaten fl (44b) is ruled out because the subordinate clause is not a complement (cf. 3.2). (44a) must be ruled out although here wA-movement is clause bound. To account for the ungrammaticality of (44a), one would have to assume a filter of the form given in (45): (45) *P [ s - t c ^ ] S] 34

On the basis of data as in (43), Reuland (1979:154,161ff.) proposes analyzing P and COMP as one constituent which is internally structured (i) S1 -> Comp1 S (ii) Comp' -> P Comp In order to account for the complex subordinators omdat, opdat, zonderdat, nadat and doordat, the prepositions om, op, na and door must be lexically specified as obligatorily taking the complementizer dot (cf. Reuland (1979:151)). This specification is given in terms of the strict subcategorization feature +_Comp (cf. Reuland (1979:156)). The prepositions voor, sinds and tot take the complementizer dot optionally (cf. Reuland (1979:151)).

61

Under Emonds' (1985) analysis, structures like (44) are ruled out automatically since there is no S' intervening between P and S, i.e. there is no landing site for wA-moved constituents. On the basis of the data presented in this section, evidence for either analysis of the internal structure of sentence taking prepositions, i.e. S or S1, has been found. The immediate advantage that lies in the analysis of subordinating conjunctions as prepositions35 is that similarities between what are otherwise separate categories (prepositions, subordinating conjunctions and adverbs) can be captured hi a principled way.

3.4 Summary Among the arguments in favor of the two categorization options of subordinating conjunctions, i.e. as complementizers or as prepositions, there is only one argument that refers directly to the lexical properties of subordinating conjunctions, namely strict subcategorization (cf 3.3). Lexical elements are specified as to which kinds of complements they take. Encoding multiple selectional properties in the lexical entry, ie. in the strict subcategorization frame, leads to a principled account of syntactic and semantic similarities which cannot be captured if lexical elements are assigned a bicategorial or muhicategorial status. Of the set of elements which are categorized as prepositions only a subset may take sentential complements. If multiple selection, i.e. encoding categorial variants in the subcategorization frame, is considered a good answer to the question of categorization, the categorial status of subordinating elements like after, before, since, until and because as well as the category of the phrases they introduce can be given immediately. Categorization turns out to be problematic with subordinating conjunctions that do not allow for categorial variation of their complement. Strict subcategorization thus proves a heuristic means for those elements which have traditionally been assigned a bicategorial or muhicategorial status but fails to include subordinating conjunctions proper, i.e. those elements which take exclusively sentential complements. The other arguments for categorizing subordinating conjunctions as either complementizers or prepositions are derived from the syntactic behavior of these elements or the constructions they introduce. Movement facts such as the non-extractabilhy of elements from adverbial subordinate clauses are - in the approaches sketched - not related to lexical properties of the subordinating conjunctions but to the fact that adverbial subordinate clauses are barred from complement positions (cf. 3.2, 3.3). Clefting and proposing, i.e. the movability of subordinate clauses, are taken to be tied to the category of the moved constituent36. Under Emonds1 (1985:289) analysis, only PPs headed by lexical prepositions and NPs can be focussed, subordinate clauses which are introduced by grammaticalized prepositions cannot undergo these movements. These movement properties, if they really refer to lexical properties of subordinating conjunctions, could be used as a diagnostic test for categoryhood.

35

I exclude complementizers, i.e. Emonds' grammaticalized Ps, from the discussion.

36

Cf. Emonds (1985:249,289,passim).

62 Hendrick (1976) argues that the possibility of adverb proposing in subordinate clauses as in (19) is closely tied to the presence of a complementizer. On the basis of adverb preposing as a diagnostic test for categoryhood, however, subordinating conjunctions as in (20), cannot be classified as complementizers. Categorizing subordinating conjunctions as complementizer or as prepositions to some extent requires that the complement of this element be of a certain category. Since it is uncontroversial that sentences are at least instances of S and that any sentence must have a pre-S position, both phrase structure rules given in (46) and (47) seem to be appropriate: (46) S'

->COMPS

(47) PP -»PS Taking into account the requirement that phrases be endocentric, (47) predicts that the sentence PP has a head, whereas the sentence, ie. S1, in (46) does not. Both rules, which can be taken to represent Hendrick's (1976) and Emonds* (1985) analyses, respectively, predict that there are no sequences like unless that, because that, since that in Modem Standard English.37 Under Jackendoffs (1977) analysis, where the sentential complement of a preposition is taken to be an instance of S', these data can be accounted for.

(48) a. PP -»PS 1 b. S' -»COMPS

(V

->COMPV")

This analysis requires that, in Modern Standard English, there be an obligatory deletion rule applying to preclude that and other complementizers from following prepositions introducing subordinate clauses. In conclusion, then, we can state that the criteria for categorizing subordinating conjunctions as either complementizers or as prepositions are not reliable. Whereas there is good evidence for categorizing that and if, for example, as complementizers and before, after, since and until, for example, as prepositions, both categorizations are problematic with respect to subordinating conjunctions such as although, in case, unless, etc.

These phrase structure rules, on the other hand, oil to account for the Middle English and Early Modern English data

Part II Recent Conceptions of Phrase Structure Part Π centers around conceptions of phrase structure as conceived in recent linguistic theory. The central aim of the following two chapters is to set the theoretical background against which the analysis of subordinate clauses is developed in this study. The focus will be on phrase structure and clause structure in general and not so much on specific types of clauses since - as has been pointed out in part I - sentence structure is held to be uniform, i.e. matrix clauses and subordinate clauses are assumed to be identically structured. Chapter 4 is an introduction to the theory of phrase structure as it figures in recent generative models such as the Principles and Parameters Theory (PPT) and the Minimalist Program (MP). Syntactic categories as well as the their division into lexical and functional categories will be discussed. Chapter 5 is concerned with the functional architecture of the clause. As recent syntactic theory abounds with functional heads, the arguments for positing functional heads, thenarrangement, their status with respect to the A vs. A'-distmction and the licensing of functional heads and projections will be discussed.

4 Phrase Structure This chapter is concerned with the conception of phrase structure within the Theory of Principles and Parameters (PPT), which developed from early generative grammar via Government and Binding Theory (GB) and subsequent work as put forward in Chomsky (1981). In the first section I will present a brief overview of the major shift within generative grammar, namely the shift from rules to representations. The "new generation" of generative grammar is essentially based on the evaluation of representations rather than on rules which create and change structures. In 4.1, I will sketch the Principles and Parameters Theory (PPT) and the Minimalist Program (MP). The inventory of syntactic categories which is assumed in PPT and beyond will be central in 4.2. The integration of so-called minor categories into the X-bar format will be discussed in some detail for auxiliaries and complementizers and will be sketched for determiners, all of which are now referred to as functional categories. The rise of functional categories is accompanied by the division of syntactic categories into lexical and functional categories which - although internally structured in a parallel fashion - are fundamentally different in nature (c£ 4.2.1 and 4.2.2). A summary is presented in 4.3.

4.1 From Rules to Representations The new generation, i.e. GB and PPT, differs crucially from previous models of generative grammar in that phrase structure rules and transformational rules are supplanted by generalizations such as X-bar Theory and move-α. * The shift from rules to representations, which figures at any level of syntactic representation2, led to substantial changes in the overall system of grammar. X-bar Theory, in this framework, is conceived as a system that governs the well-formedness of phrases in terms of endocentricfty and hierarchical structure and not - like the phrase structure rules of early generative grammar - a device that combines phrases and determines their linear order. The representations, i.e. the structures which are created by X-bar Theory and which are affected by move-α, are subject to and restricted by a set of interacting modules the principles of which must be met at a given level of representation: (1) (i) (ii) (iii) (iv)

Θ-Theory Case Theory Binding Theory Control Theory

Move-α, a generalization of movement, captures NP-movement, Wi-movement and head movement. D-structure, s-structure and Logical Form (LF) are conceived as the levels of syntactic representation. Move-α relates d-structure to s-structure and s-structure to LF. There is no interaction between LF and PF (Phonological Form).

66 (v) Government Theory (vi) Bounding Theory (adapted from Chomsky (1981:5)) The representational level of d-structure is taken to reflect the thematic properties of predicates, i.e. the number and types of arguments a given predicate takes. Θ-Theory is the module of grammar that regulates the assignment of Θ-roles under government and sisterhood. The core of Θ-Theory, i.e. its basic principle, is the Θ-Criterion which requires that each argument be assigned a Θ-role uniquely and that each Θ-role be uniquely assigned to an argument (cf. Chomsky (1981:36,1982:6)). The Θ-Criterion is not a level-specific principle but is active at all levels of syntactic representation. Arguments as well as argument chains - and their members are subject to the Θ-Criterion.3 Case Theory is the module that governs the distribution of NPs at the level of s-structure. In order for a sentence that contains NPs to be well formed, NPs must be assigned case, either morphologically overt or abstract case. The basic principle of Case-Theory is the Case-filter which states that any lexical NP must bear Case. Just like the Θ-Criterion, the Case-filter extends to chains, i.e. an NP-chain must bear Case.4 Another module that is closely tied to the category NP is Binding Theory which consists of three principles which s-structurally govern the interpretation of anaphoric NPs (Principle A), pronominal NPs (Principle B) and referential NPs (Principle C). Non-lexicalry overt occurrences of NPs, i.e. big PRO, are subject to Control Theory which applies at s-structure. Government Theory is a module which interacts with any of the modules sketched so far. Government, at any level of representation, is defined in terms of syntactic configurations. The core principle of this module is the Empty Category Principle (ECP) which states that nonpronominal empty categories, i.e. traces, must be property governed. The ECP applies at LF. The last of the modules given in (1), Bounding Theory, is concerned with movement. Bounding Theory, which can be regarded as a generalized successor of Ross' (1967) Island Constraints, defines - in interaction with other modules, especially Θ-Theory - the boundaries or barriers for movement, i.e. it determines "how far" a given element can move within a given structure. There are two more Principles which are not listed in (1) above, the Projection Principle and Principle of Full Interpretation (FI).5 The Projection Principle expresses the requirement that lexical properties - e.g. selectional properties or thematic information - be syntactically represented, i.e. they are projected onto syntactic structure (cf Chomsky (1986a:84)). FI is an LF-active principle to which the results of non-overt movement, i.e. movement from s-structure to LF, such as covert w/i-raising and quantifier raising are subjected. Besides The Θ-Criterion is extended to apply to chains, i.e. to constituents that have been moved out of a base position that is assigned a Θ-role, to the base position and to intermediate traces. Θ-Theory and Case-Theory are held to be closely tied to each other. See for example Burzio's generalization (1986) and the visibility condition derived from it (Chomsky 1986a). FI bears close relations to Θ-Theory and Government Theory in that it is fed by the ©-Criterion and the ECP.

67

applying to the results of non-overt movement, FI applies to thematic structure. (c£ Chomsky (1986a:95ffi)). I now turn to the structure building device of PPT, X-bar Theory. As has been pointed out in the previous chapters, X-bar Theory is a generalization of phrase structures which predicts that phrases - for any category X6 - are uniform and obey endocentrichy.7 Heads (X) and their projections (X1 and X") are licensed by X-bar Theory: (2) a. X1 =X X" b. X" =X" X'

(adapted from Chomsky (1989a: 89)) The combination of phrases, all of which internally are subject to X-bar Theory, is governed by independent modules of grammar the most important of which, in subsequent interaction with other modules, is ©-Theory. This modular system of grammar with its interacting subsystems allows well-formedness conditions to be expressed in independent terms rather than formulating them in terms of phrase structure rules and transformational rules. With X-bar Theory, the Projection Principle and independently derived well-formedness conditions, i.e. licensing principles, there is no need to assume that structure-creating and structure-changing devices are rule-driven. The occurrence of maximal projections, such as complements and specifiers, within other maximal projections must be licensed on independent grounds, Le. they must meet licensing conditions imposed by various modules of the grammar at some level of syntactic representation. These licensing conditions will be discussed in the following two sections, in which I shall discuss the inventory of syntactic categories in PPT: lexical and functional categories. As will be seen in 4.2, lexical and functional heads differ crucially with respect to the licensing of maximal projections within their own projections. In the remainder of this section I shall present an outline of the Minimalist Program8, which, as far as linguistic representations and their evaluations are concerned, is an even more radical approach within the generative tradition than PPT is. In MP linguistic representation is reduced to the pair PF (phonological form) and LF (logical form). PF and LF are non-related interface levels of the articulatory-perceptual system (A-P) and of the conceptual-intentional system (C-I), respectively.9 Any linguistic expression in human language is conceived as made up of a phonological representation paired with a logical representation.10 Since PF and LF are the only levels of representation, with linguistic structure mediating between them, and since they are interface levels, any linguistic expression, 6 7

8 9

10

X ranges over the categories N, V, A and P. The version of X-bar Theory which has been used so far was rule-governed, i.e. phrase structure was generated by successive application of phrase structure rules (cf. chapter 1 and 2). Chomsky (1992,1993,1994,1995). A-P and C-I are assumed to be the only two systems that interact with the computational system for human languages (C^), which is assumed to be both unique and optimal (cf. Chomsky (1992,1993,1994,1995)). Cf. Chomsky (1994:3f, 1993:5).

68 in order to be well formed, must converge at both levels. The two levels are considered bare output conditions because either the derivation converges, or h crashes.11 All syntactic well-formedness principles are subsumed under the overarching well-formedness principle of Full Interpretation (FI), which applies at both LF and PF. The make-up of MP differs from that of PPT in that the representational levels of dstructure and s-structure are done away with, and with them the modules of PPT. The sole remnants are the lexicon, the computational system, spell-out and the above-mentioned interface levels PF and LF. Spell-out is characterized as an operation (rather than a stage) in the derivation that feeds the representational levels PF and LF. At spell-out, structures are divided in such as way as to allow for unambiguous shifting of LF-relevant information to LF, and PF-relevant information to PF, i.e. there may be no LF information at PF and vice versa because otherwise the derivation would crash. It is held that lexical items consist of bundles of features, with features having one of two qualities: they are either strong or weak. A condition on convergence at PF is that there may be no strong features left, i.e. they must be eliminated before spell-out (c£ below). Individual linguistic expressions are created by the structure building device numeration, which takes fully inflected lexcial items out of the lexicon. These elements are subjected to the computational system, where they are combined to form partial trees by the operation merge. Merge applies successively to any two elements, be they simple or complex, to form larger units. 12 Move, the second operation of the system, applies to structures or rather representations composed by merge. Movement in MP, in an even stricter sense than in PPT, is always for a reason, Le. it is not unconstrained. Basically there are two constraints on movement: procrastinate and greed. Procrastinate is an economy constraint requiring that movement be deferred until LF unless there is good reason for movement to apply before LF, Le. before spell-out. A case in point for pre-LF movement is greed, which is an operation "[...] driven by morphological necessity: certain features must be checked in the checking domain of the head, or the derivation will crash" (Chomsky (1993:32)). Greed applies where there are strong features which must be checked, Le. eliminated before spell-out, because at PF these features are illegitimate and induce "crash". In MP, some of the PPT modules can be observed to have survived at LF. Case Theory, now subsuming Control Theory13, is conceived as a checking device14, Binding Theory is conceived as an LF-reconstruction mechanism15, Theta-theory is subsumed under FI, Le. there may be neither superfluous elements nor may there be elements missing. 11 12 13

14

15

Cf. Chomsky (1994:4, 1995:390), Cf. Chomsky (1995:396). PRO is assumed to be assigned Null Case which must be checked in the specifier position of a non-finite TP (cf. Chomsky/Lasnik (1993:561)). Cf. Chomsky/Lasnik (1993:539); early versions of Case checking can be found in Chomsky (1981) and Safir (1985). See chapter 9.

69 Government Theory is discarded entirely as it is ultimately based on only two relations: headcomplement and head-specifier. The core principle of Government Theory, the ECP, together with Bounding Theory, is replaced by general economy conditions such as "shortest move" and "minimal link".16 A further radical change relates to X-bar Theory, which in PPT is considered the structurebuilding device and which in MP has been replaced by merge. These two structure-building devices differ in that X-bar Theory is an "all at once operation" (Chomsky (1993:21)) in the sense that structural positions, such as complement and specifier, are generated automatically, whereas under merge, which is a kind of self-assembling structure building device, these structural positions are taken as a consequence of merge. The theoretical model I confine myself to in this study could be characterized as "late PPT" or "near minimalism", as I will make use of notions and concepts of both models. The perspective I take on licensing complements and specifiers is feature-driven, as in MP. Central to my analysis will be the fleshing out of the credo of both PPT and MP that syntactic structure is determined by the lexical properties of lexical heads. In the following section, I shall review the inventory of syntactic categories, Le. the building blocks of structure.

4.2 The Inventory of Syntactic Categories Syntactic categories, in recent linguistic theory, are assumed to be of two types: lexical and functional categories. Whereas the concept of lexical categories is familiar from both traditional grammar and the major categories of early generative grammar (cf chapter 1), the concept of functional categories is a fairly recent one (cf Fukui/Speas (1986:130)) that originally made its way into syntactic theory with the systematic integration of so-called closed-classes or minor categories, e.g. Aux, Comp and Det. Before I turn to these two sets of categories, I shall present a short overview of how minor categories became adapted to the X-bar format.

Adapting Minor Categories to the X-bar Format - An Aside: In early versions of X-bar Theory, only the lexical categories V(erb), N(oun), A(djective) and P(reposhion) are assumed to be projecting heads, Le. only these have corresponding phrases.

(3) a. b. c. d.

16

V N A P

VP NP AP PP

See for example Chomsky (1993.14i).

70 So-called minor categories17, for example determiners, auxiliaries and complementizers, are considered as non-projecting, thus falling outside the realm of X-bar Theory. S and S' do not fit into X-bar Theory either, they are exocentric categories: (4) a. NP -» DetN b. S1 -» Comp S c. S -» NPAuxVP Det, in (4a), and Comp, in (4b), are heads which have no projection. S1, as (4b) suggests, is the projection of S, S is the head of S1. In (4c), however, S is not a head since it has three immediate constituents, NP, Aux and VP. Whereas NP and VP are maximal projections of N and V, Aux is a head that does not have a projection. S is neither a head, nor is h the projection of a head. A closer look at the relation between the elements that occur in Aux, verbs and finiteness features, reveals that Aux must be the head of S. In the first place, Aux and tensed verbs are mutually exclusive (5). (5) a. *He did left early, b. *He might left early. This fact strongly suggests that inflectional tense is realized outside VP, namely on Aux, i.e. in the Aux-position. Supportive evidence for this analysis comes from constructions as in (6), where the VP is pseudo-clefted, i.e. the VP is moved and thus separated from its inflectional tense. (6) a. They said she may attempt to leave, and attempt to leave she -will. (Jackendoff( 1977:47» b. Fix the car, I wonder whether he will (Chomsky (1986b:20)) Whether the Aux position is lexically filled or not, it is present in any sentence. If there is a lexical realization of Aux, tense is marked on the auxiliary and if the Aux-position is not lexically filled, inflectional tense combines with the lexical verb.1S In subsequent work, the node Aux has been relabeled I(nflection) with I being a head that projects to IP, thus yielding an endocentric sentence structure (7).

17

In Jackendoff (1977), some of the minor categories were taken to have projections, e.g. Deg, Q, etc. S, in his system, was reanalyzed as V", a "projection" of the verb. 1* Cf. 5.1 for how tense becomes realized on the verb.

71

(=S)

ff

(7)

spec

Γ I

VP

The complement position of I is invariably filled by VP, the specifier position by the subject. I contains all verbal inflection, Le. h is a syntactic instantiation of the traditional categories of verb morphology such as tense, number and person agreement. In finite clauses, the subject and the finite verbal form agree in person and number. Models lack overt person and number morphology. English verbal agreement, in the case of lexical verbs, Le. non-auxiliaries and non-modals, is reduced to two forms in the present tense, Le. to the opposition between the 3rd person singular and all other singular and plural forms, and to only one form hi the past tense. The correlation between tense and agreement, in pre-"split-INFLll-analyses, is captured by assuming that I hosts two binary features, namely: [±Tense], [±Agr(eement)].

ttTense] t±Agr]

Overt agreement is triggered by the [+Tense]-variant of I. In non-finite forms, where Tense is [-Tense], there is no agreement between the subject and the verb. The IP analysis allows for further generalization of phrase structure in terms of X-bar Theory. Sentences, Le. former Ss, are no longer exocentric constructions; they conform in terms of general phrase structure in that they are both endocentric and hierarchically structured. Further, the projection of I, like the projections of lexical heads, contains a specifier position and a complement position. If S is replaced by IP, how can the relation between IP and S1 be described? S1 has two immediate constituents, only one of which is a head, namely Comp. S, or rather IP, is phrasal There are several arguments that led to the assumption that sentences (here S's) are headed by the complementizer. The first argument for analyzing Comp as the head of S1 comes from selectional properties. In chapter 2.1, I discussed Bresnan's (1970,1972) arguments for introducing Comp as a deep structure node. Complementizers, under her and Grimshaw's (1979) analyses, must be subclassified according to whether they introduce interrogative or declarative sentences (cf 2.1). The set of complementizers - and here I refer to complementizers in the strictest possible sense, namely those elements that introduce complement clauses (cf 3.2) consists of the lexical elements that, (interrogative) if, whether19 and for, which are often 19

The X-bar status of whether will be reconsidered in 6.4.

72

subclassified in terms of a [+wh]-feature that indicates whether the complementizer introduces an interrogative [+wh] or declarative [-wh] subordinate clause. (9)

that -

WH

for -

if whether + +

Not only the phrasal category S1, but also complementizers were identified as figuring in the strict subcategorization of verbs (cf chapter 2). Comp as a head, in recent linguistic theory, is ejected to have a projection conforming to and restricted by X-bar Theory thus corresponding to (10).

(10)

CP spec

C C

complement

The complement position in (10) is invariably filled by IP. As already mentioned in 2.1, there is a correlation between individual complementizers and their IP complement in terms of finiteness. That and if invariably take a finite IP complement, for allows only a non-finite IP complement, and whether allows both types of IP complements. (11) a. b. c. d. e. £ g. h.

It would be odd that a man is chairing a women's meeting. *It would be odd that a man to be chairing a women's meeting. It would be odd for a man to be chairing a women's meeting. *It would be odd for a man is chairing a women's meeting. If Colin is in London, he is undoubtedly staying at the Hilton. *If Colin be in London, he is undoubtedly staying at the Hilton, She wondered whether she should behave as expected. She wondered whether to behave as expected.

These correlations can be added to the [±wh]-distinction given above, yielding (12). (12)

that WH finite +

for -

if whether + + + + or -

Another argument supporting the structure given in (10) has to do with w/i-movement. In chapter 2 and 3,1 reviewed early generative analyses of wA-movement in relative clauses and embedded questions where the w/wnoved constituent was assumed to end up Chomsky-adjoined to the left of the complementizer. The specifier position, which according to the schema given in (2) above is reserved for a maximal projection, is taken as the landing site for wA-moved constituents, ie. maximal projections that are marked as [+wh]:

73

(13)

CP XPi [+wh]

IP ...tj...

Supportive evidence for this wA-movement analysis comes from wA-interrogatives which - as pointed out in chapter 2 - involve subject-auxiliary-inversion, a process which is barred in embedded interrogatives: (14) a. b. c. d. e. £

What did he offer to do for her right away? "They wondered what did he offer to do for her right away? When will you leave for Granada? *She asked when will you leave for Granada? Why did she call him? *She has been asked why did she call him?

The structure given in (13) provides a position for the auxiliary in non-embedded clauses. If the specifier of CP contains a wA-constituent as in (14e), the auxiliary is taken to leave the IP, i.e. the head position I, and to move into the head C:

did she t

call him tj..

Sentences (S1), under the CP analysis, as well as Ss under the IP analysis, conform in that they are both endocentric and hierarchically structured. Sentence structure is thus no longer exceptional with respect to X-bar Theory, both the sentential projections IP and CP are parallel in structure, i.e. they both contain a specifier position in which some XP may occur and a complement position which is filled by either VP, in the case of I, or by IP, in the case of C. Moreover, IP and CP are parallel in structure to the projections of V, N, A and P. With the emergence of a systematized account of elements such as I and C, now classified as functional categories, X-bar Theory carries over to determiners as wefl. Since the syntax of nominal expressions, i.e. the internal structure of DPs, will not be crucial in this study, I will restrict myself to a mere enumeration of the arguments which led to the phrase structure in (16).

74

Abney (1987:25, passim) proposes assigning noun phrases a sentence-like structure. Some of the reasons he gives for the analysis in (16) are well known from earlier versions of generative grammar20, but were not captured in a principled way. IBs argument develops from an analysis of "poss-ing" constructions, which are notoriously ambivalent: internally they behave like verbs, externally like NPs. The subject of "poss-ing" constructions is formally identical with the subject of nouns, i.e. it is assigned genitive case.21 (17) John's sister is a pest. (18) John's singing the Marseillaise bothers me. (Abney (1987:223)) As we have already seen in connection with IP, the subject, i.e. the nominal constituent that occupies the specifier position, is assigned nominative case only if the sentence is finite, i.e. if I contains the relevant agreement features. Abney (1987) and Fukui/Speas (1986) assume that genitive case, parallel to nominative case in the sentence, is assigned under agreement. Agreement in the nominal system is located in D, the head of DP. Genitive, in these systems, is assigned to the XP which occurs in the specifier of DP if D contains 's, Le. nominal agreement features. D, as well as I and C, may contain lexical material, namely those elements after which the category is named: determiners. Another parallel between D and the heads of IP and CP lies in their not allowing for multiple categorial realization of their complement. These heads invariably take NP, VP and IP as their complement, respectively. The integration of so-called minor categories into the X-bar Theory leads to homogeneous phrase structure which is advantageous in so far as it makes constituent specific rules superfluous. Evidence for parallel structure across category projections come from language acquisition.22 X-bar theory, as it stands, predicts that category projection does not differ across lexical categories (V, N, A, P) and functional categories (I, C, D). The projection, in each case, conforms to the X-bar Theory - disregarding the differences between lexical and functional categories:

20

It has already been pointed out in Chomsky (1970) that with respect to their internal structure NPs pattern like VPs. In subsequent work, NPs and IPs are treated on a par NPs and IPs (=S) are bounding nodes, NP and IP may have a subject, NP and IP may function as subject or object. 21 The same is true of so-called ing-of constructions: (i) John's singing of the Marseillaise bothers me. 22 Cf. Radford (1990), Deprez/Pierce (1993), Clahsen/Parodi/Penke (1992).

75

(19) a. X1 = XX" b. Χ" = Χ" Χ' (adapted from Chomsky (1989a: 89» In the following two sections I want to briefly review the defining characteristics of lexical and functional categories. A closer examination of individual categories of either group, i.e. of elements of category C for functional and elements of category Ρ for lexical categories, is postponed until chapter 6 and chapter 7. The focus in chapter 5 will be on the overall structure of clauses which is relevant for the analyses of clauses introduced by either C or P (cf. chapters 6 and 7) and clauses which are introduced by subordinating elements that cannot in any obvious way be classified as either complementizers or prepositions (c£ chapter 8).

4.2.1 Lexical Categories In recent linguistic theory, the categories V(erb), N(oun), A(djective) and Preposition) are considered lexical categories. Since Chomsky (1970) these four categories have been assumed to be represented by the two basic binary features [±V] and [±N]:23

(20) V Ν A Ρ

V N + - + + + - -

These features, on the one hand, are taken to be categorial features which define distinct classes, i.e. verbs, nouns, etc. On the other hand, they allow for cross-classification among these classes. So, for example, the feature [-N] defines the set of lexical categories that qualify as structural Case-assigners.24 Thus, with respect to structural Case-assignment, verbs and prepositions form a natural class. In connection with Case-assignment, another lexically specified property, namely directionality of Case-assignment, is identified.25 In English, verbs and prepositions assign structural Case to the right. The structural configurations to be met are government and adjacency (cf Chomsky 1986a:188), with government defined as in (21): 23

24 25

Cf. chapter 1. In Chomsky (1970), however, prepositions, i.e. the [-N-V]-caiegory in (20), are not mentioned. In Chomsky (1981:44) prepositions are classified as a [-N-V]-category. Their status as a lexical category, however, is not acknowledged. In Chomsky (1989a:89), the [-N-V>specification applies to "particles" which - at least implicitly - are subsumed under the set of lexical heads. The status of Ρ as a lexical category, as we shall see in chapter 7, is not uncontroversial. Cf. Chomsky (1981), Stowell (1981), Rauh (1993a,b), etc. Directionality of Case-assignment is a parametrized property; in English, Case-assignment by lexical categories is uniformly to the right (cf. Chomsky (1986a:193), Speas (1990:5)). But see chapter 5 for structural Case generalized to Case by agreement, i.e. a uniform treatment of structural case across categories.

76

(21) α governs iff α m-commands and every barrier for dominates α (Chomsky (1986b:8)) The second defining aspect for lexical categoryhood is lexical argument structure, or, to use the more traditional term, Θ-structure or the Θ-grid. The approach to lexical argument structure that I will defend in this study is based on the assumption that the arguments contained hi this structure are of two kinds: thematic arguments and referential arguments.26 Thematic arguments, in this system, correspond to what have traditionally been referred to as Θ-roles. Since Williams (1981) they have been divided into external and internal arguments; i.e. thematic arguments roughly specify the number of syntactic A(rgument)-positions a given lexical head (L) is associated with. A lexical head is thus related to a lexical argument structure that contains often [kissed Mary]. (Wilder/Cavar (1994:29)) As strong features must be eliminated before "spell-out", the highest verbal element in the structure, namely do, checks off the strong V-feature in T.34 They assume that have and be as well as do undergo movement into T and that lexical verbs carry a weak auxiliary feature (AUXoo) which is eliminated at LF. Assuming that this feature is part of the feature make-up of lexical verbs guarantees that these occur only in structures where there is an abstract doelement.35 Under this analysis do-insertion is no longer conceived as a last-resort mechanism but as "early altruism", i.e. the highest verbal element moves overtly into T to eliminate strong features in T. "Early altruism" allows lexical verbs to obey the economy constraint Procrastinate. The idea that do-insertion serves to eliminate the strong V-feature in T is central in Watanabe (1994). His analysis of do-insertion rests on two assumptions:

32

33 34

Cf. Chomsky (1992:66,note 22;1993:46,note 22); Chomsky/Lasnik (1993:558). Ofia strings as in (i) is considered a pleonastic element whose insertion is last resort-driven. See chapter 7

for a discussion of Case-prepositions (i) The elimination of strong features must take place prior to LF. In the Minimalist Program, FI is assumed to apply at LF and PF. Unless have or be occur, do counts as the highest verbal element.

·" Although not stated explicitly, have and be do not cany this feature as otherwise strings as in (i) would not be ruled out:

(i) *John does have kissed Mary. *John does being sleeping.

Ill

(47) In the context of Neg, T has a strong V-feature.36 (adapted from Watanabe (1994:161» (48) Economy of Representation Lexical elements can be inserted into structure if and only if insertion leads directly to satisfaction of some feature checking or contributes to meaning.37 (Watanabe (1994:160)) What is unclear, essentially, is how negation induces a strong V-feature in T. Ignoring this weak spot for the time being, the feet that rfoinsertion does not take place in non-negative clauses and in clauses containing modals and aspectual auxiliaries38 comes automatically under this analysis. So far I have considered some cases of head movement involving either overt movement to T or to T and Agr. Overt movement has visible results, i.e. overt movement is pre-LF movement. Parametric variation as well as variation within a given language depends on how far a given verb may move in a pre-LF structure. The "how far" in turn is dependent on several feetors such as the transparency or opacity of a functional head, which can be stated in terms of "strong" versus "weak" phi-features, SHA, the HMC, etc. In the following two sections, I concentrate on two more functional heads to be added to the ones discussed so fer.

5.2.2 Object Agreement (Agro) As we have seen in 5.1 and 5.2, the head of AgrP is the place where phi-features (e.g. number and case) are located. AgrP in a finite clause assigns nominative case to hs specifier under SHA, the mechanism Chomsky (1986b:24) referred to as feature sharing: subject and predicate agree in features. In this section I shall review the main arguments that led to postulating a second agreement phrase, a phrase for object agreement. The data discussed in this section are taken from French, Dutch and English. French, for example, can be observed to display overt agreement (number and gender) between the lexical verb and its direct object.39 Overt agreement, however, surfaces only if the direct object leaves hs base position, i.e. the position in which h is assigned a theta-role: 3

*> With respect to c/o-insertion in matrix interrogatives, Watanabe (1994:161) proposes the following condition: (i) [T] in the matrix clause has to have a strong V-feature which can only be satisfied by a modal [an aspectual auxiliary or by do], when wh-movement takes place in the matrix clause. [D.H.]. See also chapter 6 for a discussion of the syntactic properties of modals. 37 According to Watanabe (1994:160), (47) is needed to restrict the insertion of expletive elements. What is not clear, though, is what exactly is meant by the expeletive's contribution to meaning as expletives are generally assumed no to contribute to meaning, i.e. they are ignored at LF. 38 Wheras modals are base-generated T-heads, aspectual heads move into T to check off strong features. 39 The term direct object, as in traditional grammar, refers only to nominal constituents.

The crucial data are given in (50) and (51): (50) a. Combien de hvresj U a ecrits t,! b. Quelle voiturej a-t-il achetee tj? (51) a. Qui a ecrit ces livres? Paul leSj a ecrits tj.

b. Qui a achete ces voitures? n lesj a achetees tj. The cases m (50) illustrate wA-movement of the direct object. In (50a) wA-movement ofcombien de livres is to a topic position outside CP. The vfA-moved direct object agrees with the participle ecrits in number (pL) and gender (masc.). In (50b) the target of wA-movement is the specCP position, again there is number (sg.) and gender (fern.) agreement between the wAmoved direct object and the participle.40 In (51) the direct object is realized as an object pronoun which surfaces as a proclitic on the finite verb form, the auxiliary a, in Agr§. In both cases, the clitic pronoun and the participle agree in number (pi) and gender (masc. in (5 la) and fern, in (5 lb)). If the direct object, however, is neither wA-moved nor realized as an object clitic, there is no overt agreement between the lexical verb and the direct object: (52) a. Qui a ecrit ces hvres? b. Qui a achete ces voitures? Based on these observation, which are due to Kayne (1987), Chomsky (1989b:57fE) proposes having a separate agreement projection for object agreement. To tell the two Agr heads and projections apart, Chomsky (1989b) uses subscripts - Agr that relates to subject-verb agreement is Agrg, and Agr that relates to verb-object agreement is AgTQ. The structure of CP is given in (53) with Agrg being close to the verb, i.e. VP is the complement of AgTQ. AgrgP, in turn, is the complement of Asp (cf 5.2. 1 above):

40

Kayne (1987), according to Chomsky (1992:58), assumes that in French the participle may or may not agree in number with its wA-extracted object: (i) Combien de tablesi Paul a repeintes \ (ii) Combien de tables, Paul a repeint ^ These two strings are argued to be synonymous.

113

V

DP

(modified version of Chomsky (1989b:58)) Overt verb-object agreement as in (50) and (51) is brought about by the direct object moving from its base position to specAgrg and by the lexical verb head moving into AgfQ. Agreement between the wA-moved object and the participle in (50) as well as agreement between the object clitic and the participle is due to the wA-element and the object clitic having passed through the specAgT0P position as is indicated in (54) and (55): (54) a. CombiendehVreSi[ua[Agrop[ti][Agr0.p[Agr0ecriteSj][VPtjti]. b. Quel voiturei [a-t-il [^p [t,] [Agro'P iAgro acheteej ] [VP tj tj? (55) a. Paul lesj a b. Π lesj a

[tj]

ecrites j] [VP tj ti ]. [^ acheteesj] [VP tj tj.

Overt object agreement holds if and only if in the course of the derivation the direct object enters a specifier head relation with AgtQ, Le. when the nominal element or its trace occurs in specAgro? as in (54) and (55). SHA, as has been pointed out in 4.2.2, is conceived as "sharing of phi-features" with phifeatures comprising person, number, gender, case, etc.41 The cases considered so far are cases 41

Cf. Chomsky (1986b: 24).

114 in which object verb agreement is visible, ie. both kinds of movement, verb movement to A§TQ and movement of the direct object to specAgTQ take place in overt syntax, i.e. movement is pre-LF. SHA, in the cases considered, is visible as agreement is overtly marked on the verbal head and on the direct object; these elements overtly share the features number and gender. There are two more movement cases to be considered in which AgTQ and its projection are assumed to play a crucial role: overt movement into Agr0 and specAgTQ and covert movement, i.e. LF-movement, into AgrQ and specAgrQ. The case of overt movement into AgTQ and specAgrg will be illustrated for Dutch. Kayne (1992,1993) maintains that typological differences among languages, i.e. parametric variation,42 should be accounted for on the basis of a restrictive theory of word order in conjunction with a restrictive theory of syntactic movement. The core of his restrictive theory of word order can be stated as in (56) - (58): (56) a. Specifiers generally precede heads. b. Heads generally precede complements. (Kayne (1992)) The universal skeleton of phrase structure is thus as in (57): (57)

XP

X1

specifier X

complement

On the basis of (56), head final structures are excluded, Le. all languages, despite appearances, are head-initial. Different surface linear orders are derived by leftward movement of heads and maximal projections: (58) All syntactic movement is leftward movement. On the basis of Kayne's (1992,1993) theory, Zwart (1994) develops an analysis of Dutch, a language which is traditionally conceived as having an underlying SOV structure:

(59)

IP subject;

Γ VP

spec

tj

I

V

object

V

SOV is the unmarked word order in embedded clauses: 42

Under this analysis, the various attested word order patterns such as SOV, VSO, etc. (cf. Greenberg (1963)) are derivative , i.e. they reduce to movement cases.

115

(60) ... dat Jan zijn rijbewijs haalt ... that John his drivers license gets (Zwart (1993:4)) The unmarked word order in matrix clauses, SVO, is taken to be derivative, Le. the verb is assumed to leave its base position and to end up in C, while the subject, leaving its base position, moves to specCP. (61) Jan haalt zijn rijbewijs John gets his drivers license (Zwart (1993:4)) Whereas nominal internal arguments, in embedded clauses, precede the verb, sentential internal arguments follow h: (62) a.

... dat Jan wil [dat hij zijn rijbewijs haalt] ... that John wants that he his drivers license gets b. *... dat Jan [dat hij zijn rijbewijs haalt] wil ... that John that he his drivers license gets wants (Zwart (1993:5))

This asymmetry is related to the fact that nominal constituents, unlike sentential constituents, have to bear case in order to be licensed. The case assigned by haalt is oblique case, ie. structural case. Structural case assignment, ie. nominative, genitive and oblique case, as is stated in Sportiche's (1990) Strong Correlation Hypothesis (63), is generalized to Case by (specifier head) agreement, whereas inherent Case is generalized to Case by government:4^ (63) Strong Correlation Hypothesis Structural Case is identical to Agreement Case. Inherent Case is identical to Governed Case. (Sportiche (1990:74)) What I will assume in the following is that structural Case, in fact, is agreement Case. Case checking under SHA requires that the nominal constituent and the verb be in specifier head relation. This SHA relation is instantiated within AgrgP, with the object being in specand with the verb being in Agrg.

It should be noted that Sportiche explicitly states that the second part of the Strong Correlation Hypothesis does not necessarily imply that governed Case is inherent Case. Accusative, though governed Case, is not inherent Case. Further, nominative Case, as has been pointed out by Roberts (1993:18ff.,passim), can be agreement Case or governed Case (cf.5.2): In English, nominative Case is assigned under government in matrix interrogatives, i.e. when the finite verb form, as a consequence of head movement, is in C.

116

The situation we face with respect to the Dutch data is similar to the situation we faced with the French data above. The only difference is that agreement hi Dutch is not morphologically overtly marked on both the verb and the object. If we assume that objective Case is Case by agreement, ie. it is checked under agreement, the underlying structure of the Dutch data in (60) - (62) conforms to what has been postulated by Kayne (1992,1993) (see (56) and (58) above):

Jaiifc

oat haaltj

Jan^ tt

zijn rijbewijSi haalt· zijn rijbewijSi tj

D-stmcturaUy, L e. in the underived structure, the direct object follows the verb. As it is categorialry realized as a nominal constituent, it must move to specAgrQP to have its Case checked, Le. it has to enter a specifier head relation with the verb. Movement is triggered by Case requirements. As sentential constituents are not subject to the requirement imposed by Case theory, movement of a sentential internal argument is not triggered (62a). Were it to take place, as in (62b), the resulting structure would be ungrammatical since the verb and the sentential elements have no features to share. Having pointed out the relevance of Agr0 and Agr0P on the basis of French and Dutch data, where there are visible reflexes of movement to both Agr0 and specAgr0P, I now turn to English, where there are no visible reflexes of these movements, Le. there is neither overt agreement between the verb and its direct object as in French, nor is there variation in the linear order of the verb and its direct object. It is generally assumed that (in)visibility of SHA effects is due to movement having taken place in overt syntax or at LF. As there are overt reflexes of SHA in French and Dutch, movement to AgTQ and specAgro? is characterized as pre-LF movement. In English, as there are no

117 visible effects of SHA, movement to AgrQ and specAgrQP is assumed to be LF movement. The argument for assuming this kind of movement at all is essentially based on the correlation between structural Case and Case by agreement. The verb in (65) assigns structural Case to its nominal internal argument:

(65) ...

VP

spec

V

V

|

hit

DP

Δ

him

As structural Case cannot be checked in a head-complement configuration but only under SHA, we are forced to assume that - at some point in the derivation - the direct object moves to specAgroP and that the verb moves to AgrQ. As these movements, were they to take place in overt syntax, yield ungrammatical strings as in (66a) and (66b), we have to assume that these movements are deferred until LF in (66c) and (66d), respectively, where SHA is instantiated between the direct object hi specAgrQP and the trace of the head moved verb in AgrQ. (66) a. They him hit. b. ... that they him hit. c UgrsP [toeyiJ tAgrs' Ugrs tv hit+T+AgrJJ hp [r [T tj] d · tAgroP P1^] [Agro'tAgro *i] tvP foci tv tv »J tyl The obvious question to ask, then, is why does movement take place overtly in French and Dutch whereas it is deferred until LF in English. As we have seen above in connection with Agrg, features which are located in functional heads may be either "strong" or "weak". Strong Agr§, for example, is assumed to trigger overt verb movement and overt movement of the subject, as in the case of subject verb agreement in French. Weak Agrg, as we have seen in the analysis of the English data, does not trigger overt movement. Only at LF do the subject and the inflected verb enter an SHA relation. A similar situation obtains with respect to Agr0: The phi-features in AgrQ, Le. number, gender and Case, are strong in French and in Dutch thus triggering overt movement. In English, the phi-features are weak, thus deferring movement until LF. This perspective on structural Case assignment requires that Case Theory be changed in such a way as to incorporate the option of having structural Case checked either at "sstructure" or at LF:44 (67) Structural Case is checked either at S-structure or at LF. (Sportiche( 1990:85))

44

Inherent Case is assumed to be closely tied to theta-marking.

118 If the phi-features in Agrg are strong, overt movement is obligatory and Case checking will take place prior to LF. If the phi-features in A§TQ are weak, there will be no overt movement and Case checking will take place at LF. In this section I have presented an overview of (a part of) the inventory of functional categories within the extended projection of the verb, namely Agrg, T, Neg, Asp and AgTQ. These functional heads have not only been demonstrated to interact morphosyntactically with lexical verbs, auxiliaries and modals but were also shown to figure in the derivation of different word order patterns both within a given language and across various languages. Head movement, as has been shown, can be either overt movement, i.e. taking place before LF, or LF-movement. In either case, movement has to meet the requirements imposed by the HMC. I have considered some cases of head movement involving either overt movement of the verb to T or to T and Agr. Overt movement has visible results, i.e. overt movement is pre-LF movement. Parametric variation as well as variation within a given language depend on how far a given verb may move hi a pre-LF structure. The distance of movement in turn is dependent on several factors such as the transparency or opacity of a functional head, which can be stated in terms of "strong" versus "weak" phi-features, SHA, the HMC, etc. In connection with SHA in terms of number, gender and case, overt movement of maximal projections has been considered for French and Dutch. Covert movement, i.e. LF-movement has been considered for heads and maximal projections in English (and partly in French, i.e. the case of Agr-to-T-movement of non-finite verbs.) The second kind of covert movement concerned the movement of a nominal maximal projection to specAgrQP. In the following section I shall consider clause structure and movement on a larger scale so as to include matrix interrogatives and topicalization. The analysis of clause structure will include some refinements concerning the available landing sites for heads and maximal projections.

5.3 Movement in Matrix Clauses - A Typology of Movement In the previous sections, I have reviewed the mam arguments that led to the "split-Infl"-hypothesis, which gave rise to a lively debate on how the new functional heads Agr and T should be ordered relative to one another in the hierarchical structure of the clause. The mam arguments in favor of ordering Agr higher than T, which is now the standard analysis, come from movement facts in conjunction with the internal structure of inflected verbal forms, supporting Baker's (1988:4) Mirror Principle. It has been argued that head movement is strictly local in that h obeys the HMC. A problem with respect to head movement and the requirement that head movement be strictly local centers around negation. The question is whether Neg is present universally or only if Neg or specNeg contains lexical material. If we pursue the idea of Neg being always present, i.e. a universal ingredient of any clause, we face two options: either we can assume that there is head movement through Neg

119 and XP movement through specNeg, or we have to assume that under certain circumstances the HMC can be circumnavigated. In addition to the arguments that led to postulating the functional heads Agr, T and Neg, I have reviewed the arguments that led to introducing Asp and a second Agr, Le. AgTQ. In connection with AgrQ and its specifier, the crucial distinction between overt and covert movement, which, in connection with Agr§, T and Neg, applies to heads, was shown to apply to maximal projections as well, e.g. overt movement of subjects in French and English and overt movement of direct objects in Dutch vs. covert movement of direct objects in English, Not much has been said so far about the functional head C and its projection. For the functional heads introduced and discussed (except Neg), head movement into any functional head could be observed to interact with morphological properties and requirements, namely the the checking of tense and agreement features and, in conjunction with XP movement to the respective specifier positions, I.e. the requirement of instantiating SHA relations. In this section I will present a survey of head and XP movement within CP. The range of structures which will be discussed in this section is restricted to matrix clauses. Subordinate clauses are excluded from the discussion in this section for two reasons, m the first place, it is far from being clear whether subordinating elements are base generated in C or whether they take CP as their complement (c£ chapter 3). Secondly, abstracting away from the question whether subordinating conjunctions are elements of category C or P, there is no movement to this position in subordinate clauses introduced by a subordinating conjunction. The discussion of movement is based on Rizzi's (1990) "Relativized Minimality" approach to XP movement and on Roberts' (1993) adaptation of Relativized Minimality to head movement, both of which make the distinction between A- and A'-movement. A- versus A'-movement crucially depends on the nature of the elements being moved and the nature of their respective landing sites, which are also distinguished in terms of the A- vs. A'-dichotomy. This typology of movement in terms of A- vs. A'-movement (for both heads and phrases) is relevant with respect to the ECP, which requires of both heads and phrases that at LF empty categories be properly governed. The arguments in favor of splitting I into Agrg and T as well as the arguments in favor of introducing the functional heads ArgQ and Asp, which were reviewed in some detail in 5.1, derive from word order variation. The two kinds of movement, ie. head movement and XP movement, were argued to take place either overtly, Le. pre-LF, or covertly, Le. at LF. Independent of where movement takes place, Le. pre-LF or at LF, movement creates head chains such as those in (68) and XP chains such as those in (69): (68) [CP [spec] [c [c [v V+Agr+Asp+T+Agrlj] [^p [spec] [^ [^ tj] [χρ [spec] [r Mil [As?1* [Asp1 [Asp tj] [Agjop [spec] f^. [^ tj] [yp [spec] [v [v tj] [DP ]]]]]]]]]]]]] (69) a. subject trace [CP [spec] [c [c ] [AgrsP [DPil Ugrs' Ugrs 1 hp [spec ] [r [T ] [AspP ft] [Asp1 [Asp Ugrop [spec] [^ U^] [w ft] [v [v ] [DP ]]]]]]]]]]]]]

120

b. direct object trace [CP [spec] [c [c ] Ugrep [spec] [A^· [^ ] [χρ [spec] [Γ [τ ] [AspP [Asp1 [Asp] [AgroP P>Pil Ugro' Ugro 1 Ivp [·Ρ~1 [v [V1 fop till]]]]]]]]]] Chains, as can be seen from the data discussed in the previous sections and from the structures given in (68) and (69) are uniform with respect to the X-bar status of the element which is moved. Heads move into head positions, phrases move into phrase positions, to be precise into specifier positions of functional heads. Chain composition thus obeys the uniformity condition: (70) A chain is uniform with regard to phrase structure status. (Chomsky (1994:18))45 In order for the chain to be well-formed, any member of a chain must meet the ECP: (71) Traces must be properly head-governed. ^ Head government is defined as in (72): (72) Head Government: α head-governs iff α = X° governs

in X1 (the immediate projection of X0)47 (Roberts (1993:76))

The formulation of the ECP given in (71) is conceived as a formal licensing condition on traces. In addition to the requirement that traces be formally licensed, they are required to be identificationally licensed. Identification of traces is necessary for the recovery of "some contentive property of the null element on the basis of its immediate structural environment." (c£ Rizzi (1990:32)). What this means, basically, is that, at LF, traces have to satisfy the requirement imposed by the Princnile of Full Interpretation (FI). Traces are identified if they are bound or antecedent governed, i.e. binding and antecedent government are means of identifying traces: (73) Antecedent Government X antecedent-governs Υ iff (i) X and Υ are nondistinct48 4

^ Object clitics as in (i), at first glance, seem to violate the uniformity condition on chains in (70) since they are generated in XP positions and end up procliticized, i.e. left adjoined to the moved verb in ArgS: (0 Je Uers Ugrs bpVl [avaisjlll tj vu t. The fact that clitic movement into a head position does not violate (70) is due to their being lexically specified as XP/X°-elements. See, for example, Muysken (1982), Roberts (1993), Chomsky (1994), etc. for the X-bar status of clitics. 46 In Rizzi's (1990:30) original formulation (71) reads as follows: A non-pronominal empty category must be properly head-governed. 47 X° ranges over all lexical and functional heads except C, which is inert for government; i.e. within a given extended projection of the verb, the verb may not move higher than C. 4 * Nondistinctness is nondistinctness in terms of categorial feature specifications.

121 (ii) X c-commands Υ (iii) no barrier intervenes (tv) Relativized Minimality is respected (Rizzi (1990:92)) (74) X binds Υ iff (i) X c-commands Υ (ii) X and Υ have the same referential index49 (Rizzi (1990:87)) (75) α c-commands

iff α does not dominate and every γ that dominates α dominates (Chomsky (1986b:8))

(76) Relativized Minimality X α-governs50 Υ only if there is no Z such that (i) Z is a typical α-governor for Υ (ii) Ζ c-commands Υ and does not c-command X. (Rizzi (1990:7)) Identification of traces is arrived at by binding or by antecedent government. Binding is the mode of identification that applies to the traces left behind by theta-marked XPs as in (77): (77) a. Whoj do you [yp meet tj? b. WhOi do you think [CP t, they win [yp invite tj]? c. Whatj do you think [cp tj they will say [Q> tj he has [yp read tJ]]? All three cases in (77) are instances of u'/t-movement. In (77a) the direct object of meet is whmoved into specCP. In (77b) the direct object of invite is moved through the specifier of the CP which is the object of think into the specifier position of the matrix clause. FfTt-movement of the direct object of read m (77c) crosses two sentence boundaries, Le. it passes through the specifier of the CP object of say and through the specifier of the CP object of think. In all three cases the w>A-moved element is theta-marked, hence it bears a referential index The traces lea behind are identified by being bound by the antecedents, who in (77a), who in (77b) and what in (77c). Antecedent binding has to apply to traces left behind by non-theta marked XPs as in (78): (78) a. Why| did you [yp [yp meet him] tj? b. Whyj do you think [cp tj they will [yp [yp invite them] tj]]? c. Why} do vou think tCP *i they wu^ ΜΥ tCP *i *^at ^e ^ [\T tVP dating ner

49

Referential indices, in Rizzi's (1990) approach, are associated with theta-marked elements that are referential, e.g. direct object. 50 "α-government" is a shorthand notation for two kinds of government given in (72) - (73).

122 Again, as in (77), the wA-moved element ends up in the specifier position of the matrix CP. The members of the chain of the wA-moved element why do not bear referential indices, as h is not theta-marked in its base position (which I provisionally take to be a VP-adjoined position). The traces must hence be identificationalry licensed by being antecedent governed by the head of the chain, i.e. by why. In (78) the structural conditions under which antecedent government holds are met (cf (73)). In (79), however, antecedent government is impossible as Relativized Minimality is not observed: (79) a. *Whyj do you wonder [Q> whoj they will [yp [yp invite tj tj]]]? b. *Whyj do you think [QPI tj they will wonder [CK whoj he is [yp [yp dating t Jtj]]]? Why in (79a) cannot antecedent govern the trace tj within the VP as who in specCP is a potential antecedent governor for this trace. In (79b), -why antecedent governs the trace in specCPl, but the trace in specCPl cannot antecedent govern the trace within VP since -who in specCP2 counts as a potential antecedent governor. Although extraction out of w/z-islands is always marginal, there is a difference in grammaticality between ^-extraction of theta-marked elements and of non-theta-marked elements, as the contrast between (80) and (81) shows: (80) a. "^WhOj do you wonder [CP whyj they will [yp [yp invite tj] tj]]? b. ?Whoi do you think [^Ρ1 1, they will wonder [CK whyj he is [yp [yp dating tj] tj]]]? (81) a. *\Vhyj do you wonder [Q> whoj they will [yp [yp invite tj tj]]? b. *\Vhyj do you thmk [^Ρ1 tj they will wonder [CP2 whoj he is [yp [yp dating tj tj]]]? This contrast shows that binding of traces by (74) is not blocked by intervening wA-elements, whereas antecedent government is. If we take into consideration examples such as those in (82), we are led to reconsider binding as a sufficient condition on licensing by identification for theta-marked elements: (82) a. *Whoj do you wonder [Q> whatj they will [yp [yp give tj to tj]]] b. *Whatj do you wonder [Q> whoj they will [yp [yp give tj to What we are led to expect is that the traces tj and tj are bound by their respective antecedents what and who. As the extractions in (82) are worse than those in (80) - both extractions out of wA-islands - we have to assume that the offending wA-elements, i.e. what in (82a) and who in (82b) block identification. The situation we face in (82) is similar to the situation in (79), i.e. there are minimality effects at work. Although bearing different indices, what in (82a) can be

123

structurally constructed as a potential antecedent governor for tj, and who in (82b) as a potential antecedent governor for tj. The basic approach to licensing traces which has been outlined above has been refined in such a way as to take into account not only the properties of the moved elements but also the properties of their landing sites. Recent linguistic theory has it that not only the base positions of XP but also XP movement is of two kinds: Α-movement and A'-movement, with Α-movement being conceived as movement of DPs, and A'-movement as movement of operator-like elements, e.g. wA-moved phrases51: (83) Α-positions are restricted [DP] positions, A'-positions are unrestricted [i.e. they may contain XPs other than DP]. [D.H.] (Sportiche(1990:8)) The respective landing sites of Α-movement and A'-movement are Α-positions52 and A'-positions53. In the current view, Α-positions are licensing positions for DP arguments. Within lexical projections, these positions are positions that are assigned a theta-role. Within functional projections, Α-positions are specifier positions of functional heads where the moved DPs enter an SHA-relation with a functional head. What happens under SHA in this case, basically, is sharing or checking of phi-features such as number, gender and case.54 A'-positions have traditionally been defined ex negative, i.e. as positions that are not thetamarked and into which lexical material is moved. SpecCP, in classical GB analyses, is a prototypical A'-position in non-V2 languages as it is the target of wA-movement. With a more systematized account of functional projections, the definition of A'-positions has become more elaborate. Within lexical projections, A'-positions, in fact, are positions that are not theta-marked. A'-positions in lexical projections are conceived as adjoined position, e.g. the VP-adjoined positions in (80) and (82) above. Within functional projections, A'-positions as well as Α-positions are specifier positions. The difference between A- and A'-positions, however, lies in the kind of SHA they enter. Whereas Α-specifiers agree with their heads in phi-features, A'-specifiers agree with then: heads in operator features such as [±wh], [+neg], etc. The distinction of specifier positions in terms of A- versus A'-makes it possible to generalize the Relativized Minimality part of Antecedent Government defined as in (76) to A-movement (84) and A'-movement (85): 51

52

53

54

Cf. May (1985), Chomsky (1981,1986b), Sportiche (1990), Rizzi (1990), Manzini (1992), Cinque (1990), Roberts (1993,1994), etc. These Α-positions are not to be confused with Α-positions in lexical projections, i.e. positions that are thetamarked. Α-position in the sense used here, i.e. contrasting with A'-positions, are DP landing sites. Roberts (1993:39ff.) proposes having the A-A'-dichotomy extended to heads. This proposal is based on the following assumptions: "Α-heads have properties that are relevant for the determination of argument structure; while A'-heads are relevent for the A'-system. More precisely, Α-heads license Α-chains (by assigning Case or a Θ-role), while A'-heads license A'-chains (e.g. by licensing an operator)." Cf. Chomsky (1992), Roberts (1993), Sportiche (1990).

124 (84) Z is a typical potential antecedent governor for Υ, Υ in an Α-chain = Ζ is an A specifier c-commanding Y. (85) Ζ is a typical potential antecedent governor for Υ, Υ in an A'-chain = Ζ is an A' specifier c-commanding Y. (Rizzi (1990:7)) In the previous section, we have considered two types of DP movement: DP movement to specAgr0P as in (86) and DP movement to specAgr§P as in (87) (only the relevant traces are given): (86) a. You have not met Bill t> [CP [spec] [c [c ] [Agrsp [you] [Agrs'tAgrs havel trP [spec] Μτ 1 [NegP t8?60! [Neg'[Negnot] t^P [Asp'[Asp] [^p [Billj] [4^. [Apo met] [yp [spec] [v [ V 1 fop ti ]]]]]]]]]]]]]]] (87) a. You have not met Bill t> [CP [Φ60! Ic1 Ic 1 UgrsP [yowil tAgrs'tAgrs h»ve] [χρ [spec] [r[T ] [NegP [spec] [Neg1 tNeg not] [Asp? Μ [Asp'[Asp] [^p [Bui] [A^. [^ met] [yp ft] [v tvl fop ]]]]]]]]]]]]]]] In (86) and (87) both the direct object DP and the subject DP have been moved to specifier positions. Movement of DP is required for Case reasons. The case assigned to these DPs is structural Case which, as we have seen above, is Case by agreement. Objective Case is assigned or checked in specAgrQ? and nominative Case in specAgr§P, the respective landing sites for the direct object DP and the subject DP. The trace of the direct object in (86) is formally licensed by being properly head-governed and it is identified as it is antecedent-governed by the head Bill in specAgr0P.55 There is no specifier of a functional projection intervening between the head and the tail of the chain such that the Minimalized Relativity effect cannot arise. DP movement to specAgTQP is thus unproblematic. With respect to DP movement to specAgrgP, the situation is different. Between the head of the chain in specAgrgP and the tail in specVP in (87) there are four intervening specifier positions: specAgrQP, specAspP, specNegP and specTP, none of which offends minimality, ie. the identification requirement on traces is met.56 The fact that DP in specAgroP does not prevent tj in specVP from being antecedent governed by the head of the chain, or, as we shall see, by the trace in specAspP, can be accounted for on the basis of the features which are shared under SHA. These features are not identical, as specAgr0P and Agr0 share features for objective Case and specAgrgP and Agr§ share features for nominative Case.

" Needless to say, it is bound. Cf. Rizzi (1990:32).

56

125 What seems to be necessary is to integrate the inherent/intrinsic properties of the specifiers into the definition of what counts as a typical potential antecedent governor.57 What we have to exclude, basically, is that a specifier position in which objective case is checked may count as a typical potential antecedent governor of a trace which - by inheritance - is marked as bearing nominative Case. With respect to antecedent government in Α-chains, the definition of Relativized Minimality must encapsulate the idea that, other things being equal, the inherent properties of what might be construed as an antecedent governor are relevant. Under this perspective, then, AgTQ? cannot be construed as an intervening potential antecedent governor for the trace in specVP. Similarly, Agr§P cannot be construed as an intervening potential antecedent governor for the direct object trace in specAgr0P in (88): (88) *John seems that Bill has met. [CP Ε5?60] tc tc ] tAgrsP [JohnJ [Agrs'tAgrs see™8 1 ···· tvP [sPec] tv tv 1 tcP [spec] [c- [c that ] [^p [Bill] [Agrs'tAgrs hasl ΙτΡ [specl ίτίτ 1 [As?1* [Asp'fAsp] UgroP P»a tAgro'Ugro 1 tvP h>~] [v [v meet] [DP ή ]]]]]]]]]]]]]]]...]]]] In (88), the DP John, which originates in the direct object position of the W of the embedded clause, moves to specAgro? in the embedded clause for Case reasons, i.e. this DP bears objective Case. From specAgrQP, DP moves on and ultimately ends up in specAgrsP of the matrix clause, from where it fails to antecedent govern its trace in specAgrQ?. Examples like this have been used as evidence for the adequacy of (84), i.e. (88) is ruled out because the DP that occupies the specAgrsP of the embedded clause counts as a typical potential antecedent governor for the trace left behind by the DP John. If we assume that what counts as a typical potential antecedent governor is sensitive to inherent properties of a specifier, then (88) is ruled out because the movement of the DP John, which enters an SHA relation with AgTQ in which objective Case is assigned or checked, into specAgrsP would violate the requirement that head and specifier agree. As DP bears objective Case and as Agr§ assigns/checks nominative Case, DP and Agr§ do not agree, hence there is no licensing of DP. Moreover, the phi-features of Agr§ (person, number and Case) cannot be checked. In chapter 5.2.1 I argued that DP, on its way to specAgr§P, passes through specAspP. In the examples given in (30) and (31) this position was observed to be occupied by a floated quantifier, i.e. by a segment of the extended projection of a noun. On the basis of these observations, specAspP is classified as an A-posftion. The chain created by DP movement from specVP to specAgrgP thus has three links. As the specifiers of NegP and TP can neither be observed to function as intermediate landing sites for DP movement nor to block antecedent government of the DP trace in specAspP, they cannot be Α-specifiers. The logic of the system predicts that, if they are not Aspecifiers, specTP and specNegP count as A'-specifiers. What we expect, then, is that they do

57

Clause (i) of (73) Rizzi's (1990:92) definition of Antecedent Government should be maintained as categorial non-distinctness is still relevant.

126

not block antecedent government in Α-chains. This expectation, as the grammatical examples in (89) show, is borne out:

They;

may

not allj

have

read it

The subject DP, on its way to specAgrgP, can arguably not have passed through specTP, as T bears no features to share with DP, Le. there is not SHA between DP and TP. Similarly, DP cannot have passed through specNegP, as specNegP hosts a negative operator (negOp).58 In French, as we have seen in 5.1, specNegP contains an overt negative operator (pas, point, plus, guere, etc.). With respect to English it seems reasonable to assume that specNegP hosts an empty negative operator which enters an SHA relation with the the head. Specifier positions hosting operators are A'-specifiers, and as such, by definition, they do not interfere with antecedent government relations in A-chains. Before I turn to discuss the role of specNegP in antecedent government in A'-chains, I want to come back to specTP, which has been argued not to enter at any point in the derivation an SHA relation with the DP which is now in specAgrgP.59 Not entering an SHA relation with 58 59

Cf. (Ill)below. But see Chotnsky/Lasnik (1993:561), who assume PRO moves into the specifier position of a non-finite TP to have ist Null Case checked. Chomsky (1993:45.note 11) alludes to the possibility of DP - on its way to specAgrsP - to pass through specTP. Under such an analysis, Case checking and phi-feature checking would take place in two different specifiers, i.e. in specTP and specAgrsP, respectively.

127

DP or a trace of it can be taken as evidence ofthat position either being an A'-posftion or an Α-position whose features are such that they make the position an impossible potential antecedent governor. Independently of whether we classify specTP as an A- or as an A-poshion, it has to agree with the head of T. In either case we would expect specTP to host some element with which T has features to share. As we have seen above, T hosts features that specify (non-)finheness: if T is marked as [+fin], it is further specified in terms of [±past], if h is marked as [-fin], no further distinctions are made. The obvious question to ask is which elements interact with the features hosted in T on an SHA basis? And what is the status of these elements: are they operators or DP-like elements, i.e. is specTP an A'- or an A-position? With respect to the first question, i.e. which elements interact with features in T, consider the contrasts in (90): (90) a. They left yesterday. b. *They left tomorrow. c. They will leave tomorrow. d. *They will leave yesterday. e. They left then. f They will leave then. As temporal adverbs such as tomorrow and yesterday are inherently specified hi terms of [±past], it is likely that they interact with the [ipastj-specification of T, e.g. yesterday as a [-Hpast] adverb can occur only if T is [+past].60 Adverbs such as then - and its interrogative counterpart when - are not inherently specified for [+past], hence they co-occur with both [+past] and [-past] T. Elaborating on work by Reichenbach (1947), En9 (1987), Zagona (1988, 1992, passim) and Stowell (1993), Zeller (1994) assumes that reference time can be lexicalized by temporal adverbs which are located in the specifier position of the functional head.61 Zeller (1994:228ff.) proposes having Z(eit) (time) represented as a functional head different from T(ense). T, in his 60 Larson (1985b) assumes that bare-NP adverbs, i.e. elements like yesterday, tomorrow, today, that day, Sunday, now, then, etc. are lexically marked as [+temporal]. 61 Drawing on seminal work by Reichenbach (1947), Zeller develops a syntactic analysis in which the temporal location of events is related to the time of speech via a reference time. (i) Tom will have done the dishes tomorrow. In (i), the event time, i.e. "Tom's having done the dishes", is to be located in the future relative to the time of speech. The relative order of event time (E) and time of speech (S) is given in (ii): (ii) S E The event time, i.e. 'Tom's having done the dishes" is not only related to the time of speech but to a reference time "tomorrow": (iii) S E R R, in the example just given, is expressed by a temporal adverb, but R is not always explicitly expressed, i.e. its effects are not always visible as can be seen in (iv) and (v), where E and R collapse: (iv) Paul laughed. (v)E,R S.

128 analysis, takes ZP as its complement; T hosts the [±fin] specification and Z hosts the [ipast] specification: The temporal adverb yesterday in specZP is taken to instantiate the relation between event time and reference time hi terms of sharing F-features (cf. Zeller (1994:228f£)).

(91) spec time of speech [+tns]

spec reference time yesterday

[+pst] -ed

laugh-

Abstracting away from the fact that Zeller (1994) splits temporal information between two functional heads, this analysis paves the way for a systematized account of temporal adverbials that can easily be adopted in the present framework. Sharing of features, basically, amounts to SHA. Although not stated explicitly, there is an SHA relation between specZP and Z in (91). The crucial feature is [+past]. Adopting the idea that temporal adverbs are located in the specifier position of a functional head, I will assume that the contrasts in (90) can be explained on the basis of SHA between specTP and T. Only a [+past] T can host a [+past] temporal adverb in its specifier; only a [-past] T can host a [-past] temporal adverb in its specifier. This analysis gives us the structure in (92), where specTP contains the temporal adverb yesterday:

129

yesterday [+past]

[+past]

laugh

As can easily be seen on the basis of vert) movement facts outlined in 5.1 and 5.2, specTP cannot be the position where temporal adverbs such as yesterday, tomorrow, today, that day, Sunday, then, etc. are base generated, Le. they must have moved into specTP.62 This assumption is supported by the ungrammatical sequences in (93): (93) a. b. c. d. e. £ g.

"They yesterday left. *They have yesterday left. *They might yesterday have left *They will tomorrow leave. *They will tomorrow have left. "They then left. *They will then leave.

These observations strongly suggest that temporal adverbs occur only covertly in specTP, i.e. as a result of LF-movement. This leads us to the second question: what is the status of these elements? What is their base position, are they operators or DP-like elements, i.e. do they head A'- or rather Α-chains? Is specTP an A'- or an A-poshion? In the relevant literature on temporal expressions, e.g. Εης (1987), Partee (1973), Larson (1985b), we find tense and temporal expressions, Le. temporal adverbs, treated on a par with

Alexiadou (1994) presents the same analysis.

130

referential nominal expressions.63 Basically, temporal adverbs are conceived as fixing the reference time with respect to which the sentence is interpreted.64 The referential property of temporal adverbs as opposed to adverbs of manner, measure and reason, for example, is highlighted in Rizzi (1990:86fE).65 He assumes that temporal adverbs are theta-marked by the verb, i.e they count as temporal arguments in the spirit of Zagona (1988). Evidence for this assumption comes from observations by Cinque (1984) and Aoun (1986), namely that argument extraction, i.e. extraction of theta-marked XPs out of wA-island is marginally acceptable, whereas extraction of non-arguments, e.g. manner and reason adverbs is not: (94) a.

??

[A ehe ora]j non ti ricordi [ehe cosa abbiamo detto tj? At what time don't you remember what we said? b. *[In ehe modo]; non ti ricordi [ehe cosa abbiamo detto tj]? In what way don't you remember what we said? c. *[Pre ehe ragione]j non ti ricordi [ehe cosa abbiamo detto tj? For what reason dont you remember what we said? (Rizzi (1990:91)) ?

d. [When]j do you wonder [whether he will arrive tj. e. *[How]; do you wonder [whether he will react tj. f *[Why]j do you wonder [whether he has done it tj These examples pattern like the examples given in (80) and (81) above, where w -extraction of arguments out of w/i-island yields marginally acceptable results, whereas wA-extraction of adjuncts results in ungrammaticality. A further argument suggesting that these temporal adverbs are arguments, i.e. DPs, rather than adjuncts comes from distributional facts :

63 The crucial idea is that in sentences as in (i) (i) He will do the dishes tomorrow. the tense is conceived as an anaphoric element which needs a temporal antecedent, i.e. tenses and their temporal antecendents are subject to Binding Theory. This requirement can be summarized as "tenses must be anchored" (cf. Εης (1987), Partee (1973), Zeller (1994)). Larson (1985b) assumes that adverbs of the type under consideration are NP-like elements (bare NP-adverbs) which need to be Case-marked. 64 Partee (1984) points out that "Intuitively, noun phrases 'refer1 and tenses don't; nominal anaphora can be viewed as involving a generalization and explication of the notion inaccurately but intuitively labeled 'coreference'. Temporal anaphor is more subtle because of the categorial variety of expression involved tenses, adverbs, adverbial clauses, and main clauses [...]. It is still fair to say that tenses, like pronouns, are anaphoric, and like pronouns can be construed with either linguistic or non-linguistic antecedents; but it doesn't follow that they 'refer1 to times to the degree that pronouns 'refer* to individuals." (cf. Partee (1984:275)). 65 Spatial adverbs are assumed to pattern like temporal adverbs with respect to referential properties.

131

(95) a. I wont be free before then/that time. (Larson (1985b:612)) b. That day passed very quickly. (Larson (1985b:609)) (96) a. b. c. d.

Tomorrow is another day. I like Thursdays. Yesterdays bomb attack was terrible. There was an article by Bill in this -weeks edition of TIME.

Rizzi (1990:91), drawing on tfigginbotham (1985), assumes, that, as events take place in time66, temporal elements, such as those under consideration, "[...] are selected by the head containing the event specification in the clausal structure". Their being selected or thetamarked, according to Rizzi (1990:91), allows for long-distance construal of temporal adverbs such as in (94) above. Alexiadou (1994:135) points out that wA-movement of temporal adverbs in (98) gives rise to weak crossover (WCO) effects identical to the effects yielded by wA-extraction of DPs over a bound pronoun in (97). This parallel can be accounted for if temporal adverbs have DPargument status: (97) *Whok does herk husband beat? (98) *Which dayk did you read a poem about its^ sunset? (Alexiadou (1994:135)) A further case in point, as is reported by Alexiadou (1994:136), is the symmetry of DParguments (99) and temporal adverbs (100) in antecedent-contained deletion structures: (99) a. Max saw the same person that Fritz did. b. Eunice would believe anyone that Hank would. (Larson (1987:241)) (100) a. When did Mary read everything that Bill did? b. On what day did Mary see everybody that Bill did? (Alexiadou (1994:136)) On the basis of the syntactic behavior of temporal adverbs, which parallels the syntactic behavior of DPs with respect to long-distance movement, e.g. movement out of wA-islands (94), occurring as arguments (96), WCO-effects (98) and antecedent-contained deletion (100), we are led to conclude that temporal adverbs are theta-marked, i.e. they occupy A-poshions within lexical projections.67 Temporal adverbs, as we have seen above, interact with the tense specification in T. Therefore, I assume that, at LF, in order to enter an SHA-relation, these 66

Rizzi's (1990:91) discussion covers both temporal and spatial adverbs.

67

See also Alexiadou (1994). Cf. Zeller (1994:234f.) for a discussion of the Α-status of temporal adverbs.

132

elements are specTP and that specIP is an A-posftion.68 Movement of temporal adverbs into specTP is thus A-movement. If we come back for a moment to the semantic function of temporal adverbs, i.e. the anchoring function of tenses or their indication of the reference time69, we are led to assume that sentences cannot be interpreted if there is no temporal adverb. Sentences like those in (101) would be non-mterpretable: (101) a. John left. b. John has left. c. John will leave. A solution to this problem as well as to the problem that would arise for SHA, i.e. for the discharging of the tense features in T, is presented in Zeller (1994:241f), where he proposes having non-lexicalized reference time, Le. null temporal adverbs, located in specZP.70 Under our analysis, temporal adverbs as well as their null-realizations are, at LF, located in specTP, where they enter an SHA relation with T. Alternative analyses of non-lexicalized temporal adverbs are those given by Larson (1990) and Dubinsky/Williams (1995). The basic idea behind these analyses is to capture the long distance construal of subordinates, m (102) the event / saw Jill in New York can be construed as having taken place either "before the time of her claiming that she would meet Tom" or "before the time of her actually meeting Tom": (102) I saw Jill in New York before she claimed that she would meet Tom Larson (1990:177£), and in subsequent work Dubinsky/Williams (1995), propose tracing back the scopal differences between the two interpretations to the actual depth of the temporal adverb which they take to be an operator tmpOp What Larson (1990) and Dubinsky/Williams (1995) refer to as a temporal operator is identical to what I describe here as temporal adverbs in specTP and to Zeller's (1994) lexicalized reference time. The two interpretations of (102) are given in (103) and (104) together with their respective underlying structures: (103) a. I saw Jill in New York before the time of her claiming that she would meet Tom. b. I saw Jill in New York [χρι before [χρ2 she claimed [χρ3 that she would meet Tom]

68 Cf. Alexiadou (1994:130), who states that "[tjemporal adverbs are specified as being marked for temporal features similar to the ones located in T°. A plausible proposal would be to assume that they must be checking this agreement relations [sic!] in a Spec-head configuration in TP, if T° is primarily marked for temporal features and is primarily responsible for temporal interpretation." 69 70

See for example Εηφ (1987) and Zeller (1994). Zeller (1994:242) has these null-adverbs represented as PROs, which, in order for the sentence to receive a temporal interpretation, must be contextually bound N.B. this analysis differs crucially from analyses that are based on the assumption of temporal operators, e.g. Larson (1985b, 1990), Dubinsky/Williams (1995).

133

(104) a. I saw Jill in New York before the time of her actually meeting Tom. b. I saw Jill in New York [χρ! before [χρ2 she claimed [χρ3 that she would meet Τοπικό?]]] In (103) the temporal operator originates within the VP in XP1, in (104) within the VP of the lower XP.71 The two different interpretations, according to Larson (1990), are arrived at by tmpOp, roughly corresponding to when12, moving into spec XP2 . As Larson (1985b, 1990) assumes that adverbs of the type under consideration are NP-like elements, they must move into Case-positions. SpecXPl, under his analysis, is such a position. Dubinsky/Wifliams (199S), taking Larson's (1990) analysis as a starting point, object that his analysis, though on the right track, falls short of explaining why neither the head nor the specifier of XP2 can be lexically filled, Le. why both (105) and (106) are ruled out. (105) 'John left [χρ! bn [x after] [χ^ [spec] [χ2 [x that] ρ told him to]]]]]. (106) 'John left ^ bd [x after] [χρ2 [when] [χ·2 [χ e] [I told him to]]]]]. (adapted from Dubinsky/Williams (1995: 130)) In order to rule out these examples, they propose reanalyzing temporal prepositions taking sentential complements as complementizers with an IP complement, or, in more recent terminology, an AgrgP complement. The temporal operator, as under Larson's (1990) analysis, is located in specCP. Instead of assuming that tmpOp moves for Case reasons, Dubinsky/Williams assume that the relation between C and specCP is an SHA relation satisfying the requirements imposed by a "relative" of Rizzi's (1991:2) 0%-Criterion: (107) Temporal Operator Criterion a. A temporal operator must be in a spec-head configuration with a [-Kemporal] X° b. A [-Kemporal] X° must be in a spec-head configuration with a temporal operator. (Dubinsky/Williams (1995: 134)) On the basis of the proposal made above, namely that temporal adverbs move into specTP, as well as in view of the analysis of temporal subordinate» which win be developed in chapter 7, I shall argue that temporal SHA as defined (107) does not take place in CP but in TP, thereby maintaining the Α-chain status of the moved elements. Now that specTP is classified as an Α-specifier, we still have to explain why Α-movement of the subject DP across specTP, an intervening A-poshion, does not give rise to an ECP violation, Le. why specTP does not count as an offending Α-specifier with respect to antecedent government of the subject trace in specAspP by the head of the Α-chain in specAgrsP in (108)73: 71

72 73

I use XP here because Larson (1990) classifies before as a preposition and Dubinsky/Williams (1995) call it a complementizer. Cf. Larson (1990:178). Only the relevant parts of the structure are given.

134 (108)

AgrsP spec

Theyj

Agrs'

yesterday^ [+past]

[+past]

laugh tj

I assume that specTP, although an Α-specifier, does not qualify as a potential antecedent governor, as its feature specification, namely [+past], precludes its being construed as such. It is clearly not the case that Rizzi (1990) could be accused of having made the wrong predictions about what counts as a typical antecedent governor, as the functional system dominating VP at that time was relatively parsimonious, i.e. his analysis is essentially based on a twofunctional-categories model, ie. on a CP-IP-system. As a refined system requires refined definitions, I propose to maintain Rizzi1 s (1990:7) original definitions hi a slightly refined way: (109) Z is a typical potential antecedent governor for Υ, Υ in an Α-chain = Ζ is an A specifier c-commanding Υ and Ζ bears features compatible with Y. Relativized Minimality effects can thus be observed to be at work only if a potential antecedent governor, ie. an Α-specifier in Α-chains and an A'-specifier in A'-chams, is realty a typical potential antecedent governor, Le. if ft bears a compatible feature specification. Taking this perspective, the situations in (110) can be ruled out, which they could not if the notion "typical potential antecedent governor" were merely dependent on the Α-status of the specifier. (110) A-movement *specAgroP binds t in specVP => due to non-compatible Case specifications, Le. objective vs. nominative *specAgrgP or specAspP binds t in specAgTQP => due to non-compatible Case specifications, i.e. nominative vs. objective (cf raising constructions as in (88))74 *specTP bkds t in specAspP => due to non-compatible feature specifications

in the Minimalist Program specAgrgP and specAgr0P are assumed to be identical in nature, i.e. both Arg°s contain phi-features. Spec AgrQP not binding the subject trace in VP is accounted for by "shortest move" and "equidistance". See for example Chomsky (1993:17ffi).

135

Above, specNegP was classified as an A'-specifier: specNegP does not interfere with A-chains and h contains a negation operator. According to the Neg-Criterion proposed by Haeberli/Haegeman (1995:91), specNegP and Neg enter an SHA relation: ( 1 1 1) The Neg Criterion75 a. An X°[+NEG] must be in a Spec-head configuration with an Operator [+NEG]. b. An Operator [+NEG] must be in a Spec-head configuration with an X° [+NEG]. Taking specNegP as an A'-specifier, we expect adjunct extraction across NegP, ie. A'-movement, to yield ungrammatical results as, according to (85), repeated as (1 12), specNegP counts as a typical potential antecedent governor for adjunct traces. (112) Z is a typical potential antecedent governor for Υ, Υ in an A'-chain = Ζ is an A specifier c-commanding Y. (Rizzi (1990:7)) The relevant cases to look at are cases of long movement76 as in (1 13) and (1 14): (1 13) a. *Whyj didn't they say [CP tj he had [yp [yp missed the train] tj]]? b. *HoWj didn't you say [Q> tj he had [γρ [γρ done the dishes] tj]]? ( 1 14) a. *Why do you think [CP tj they will not [yp [yp invite them] tj]]? b. *Why do you think [Q> tj they will not say [Q> tj he is not [yp [yp dating her] tj]]]? In (113) and (114), the wA-moved adjuncts are taken to originate in a VP-adjoined position of the embedded verb77 and to have undergone successive cyclic movement to specCP of the matrix clause. We take the fact that the wA-moved adjunct in specCP cannot be reconstmed as having originated in the VP-adjoined position of the embedded VPs as evidence for the failure of antecedent government. The relevant specifiers to look at with respect to the examples in (114) are specCP of the matrix clause, specNegP and specCP of the embedded clause, Le. the sentential argument of the matrix verb. The chain fink [CP t, ] - [yp [yp tj] is unproblematic as short movement is possible, i.e. the trace in specCP antecedent governs the trace in the VPadjoined position. Crucially, why in the matrix specCP fails to antecedent govern tj in the embedded CP:

75

The Neg Criterion in (111) can be considered a "relative" of Rizzi's (1991:2) formulation of the »7i-criterion. See also Rizzi (1991:1 If). 76 Short movement, i.e. clause bound movement, always yields grammatical results: (i) Why have you not [yp [yp met him] tj? 77 Rizzi (1990:50) assumes that manner adverbs are adjoined to VP, whereas reason adverbs are adjoined to TP. He proposes analyzing why as a base-generated specCP element (1990:46).

136

he had missed the train ^

The failure of why to antecedent govern its trace in specCP is attributed to specNegP, an intervening A'-specifier, qualifying as a typical potential antecedent governor of t, in specCP (cf Rizzi(1990:17f)). With respect to the example given in (H4a), we have to concentrate on specCP of the sentential argument of the matrix verb, specNegP and the trace in the VP-adjoined position, with respect to (114b) the highest specCP position, specNegP and the lowest specCP position are relevant. Again, antecedent government fails because specNegP qualifies as a typical potential antecedent governor for tj in the relevant specCPs. Adjunct extraction out of negative islands is parallel to adjunct extraction out of wA-islands (cf (81), (94b), (94c), (94e) and (94f) above). With respect to A'-movement and island violations, (112) makes the correct predictions and thus allows us to reduce island effects to the ECP(cf Rizzi(1990:17)).78 There is another specifier position we have to look at: specCP. On the basis of the \vhmovement data discussed so far, the A'-status of specCP is beyond controversy.79 The XPs 78

It should be noted that specNegP displays what might be called "unselective" A'-binding properties, i.e. it counts as an interving A'-specifier independent of the quality of the trace it antecedent-governs. On the basis of these observations, however, the requirement of non-distinctness of the antecedent governor and the governee in terms of categorial features (cf. (73)) has to be reconsidered 79 In V2-languages, specCP and C count as Α-positions if specCP hosts a nominal element. See also Roberts (1993:61).

137 that move to specCP are wA-operators which, as we have seen in 4.2.2, may have an overt (115a) or a null realization, e.g. the case of matrix alternative questions in (115b).80

(115) a

CP spec

C C

whyj [+whj

AgrgP

hasj [+whj

he tj refused to remove the garbage t;?

CP

r-ΗΛ]

he tj refused to remove the garbage tj?

In both cases, movement of (Asp-to-JT-to-Agrg-to-C is triggered by C being specified as [+wh], whh [+wh] designating that CP is interrogative. According to standard assumptions, head and specifier in a functional projection must agree in features. The requirement of SHA within CP, Le. between C and specCP, is captured by Rizzi's (1991) WH-Critenon: ( 1 17) The Wh-Criterion81 a. A Wh-Operator must be in a Spec-head configuration with an X°r+wj,i. b. An X°[+wb] must be in a Spec-head configuration with a Wh-Operator (Rizzi (1991:2)) The elements that end up in specCP can be either arguments ( 1 18) or adjuncts (119): ( 1 1 8) a. [CP WhOj do you [yp meet t4]]? b. [Q> Whatj do you think [Q> tj they will [yp buy tj]]7 ( 1 19) a. [CP Whyj did you [yp [yp meet him] tj]? b. [CP How did he [yp [yp fix the sink] tj]? The fact that specCP may host A- elements seems to spoil the neat picture of A-elements moving to Α-specifiers and A'-elements movmg to A'-specifiers.

80 81

In both cases, according to Roberts (1993:18ff, passim), the head moved verbal form in C assigns nominative Case to SpecAgrs under government (cf. note 43). Rizzi's (1991) W -criterion is based on May (1985:17).

138

SpecCP does not interact with the Α-system as it is an operator position.82 The SHA relation the wA-moved moved elements in (118) enter into is not based on nominal features such as Case and number, but on the [+wh]-feature. The [+wh]-feature is an operator feature which is an integral part of the overall feature representation of who, what, why and how, ie. an argumental XP which, in addition to Case and number or tense features, carries operator features suchas[+wh]. The chains we get with w/z-movement of an argument as in (118) have the structure in (120): (120)A'-A(-A-... -A) Chains of this type are well formed, whereas chains of the type in (121), which are generally referred to as "improper movement" cases, are not:83 (121)*A-A'(-A-...-A) So far, we have assumed that AgrgP is only dominated by CP. What we are led to assume on the basis of (122) is that movement of the negative element is movement to specCP. (122) a. b. c. d.

[Under no circumstances] will I remove it. [Never before] have I refused their invitation. [Never again] will she call him an idiot. [On no account] will I do the dishes.

There is good evidence, however, that the A'-specifier occupied by the preposed negative element is not specCP but a specifier position between specCP and AgrgP which I will refer to as specTop(ic)P.84 Crucially, I take topicalized elements to be located in the specifier position of an autonomous functional projection TopP and not to be adjoined to AgrgP.85 Evidence for positioning TopP between C and AgrsP comes from embedded contexts as in (123), where the complementizer that precedes the preposed negative element:

82

In general, if movement precedes from an A'-specifier, it can be only to another A'-specifier. As has been pointed out above, specCP counts as an Α-position in V2-languages if it hosts nominal elements. It seems reasonable to assume that there are two Cs, one which is an Α-head and one which is an A'head. See Roberts (1993:61f.) for a refined typology of Cs. 83 See for example Chomsky (1986b), Cinque (1990), M ller/Sternefeld (1993), Roberts (1993), May (1985), etc. 84 See, for example, Hoekstra (1992:194), M ller/Sternefeld (1993:486), Alexiadou (1994:84,98) for Topic Phrases as projections of a functional head Top. 85 Lasnik/Saito (1992), Rochemont (1989) and Chomsky (1977), for example, assume that topicalization is adjunction to AgrgP-equivalents, i.e. EP or S; Kiparsky (1995) assumes that topicalization is adjunction to CP. Others, e.g. Diesing (1990), Santorini (1994,1995), assume that topicalization in symmetrical V2 languages (e.g. Icelandic, Yiddish) is movement into specIP.

139 (l23)a. She said [CP [c1 tc matl [τορΡ [under no circumstances^ [Top. would she tj marry him]]]]. b. She hoped [Q> [c1 tc ^at] tropP [never again]; [Top· would he tj pop the question]]]]. Assuming an extra projection86 for topicalization structures gives us an additional specifier position, specTopP, and an additional head position, Top. As can be seen in (123) and (124), the finite verb precedes the subject in specAgrgP, i.e. h has head moved into a higher functional head projection, namely Top. Overt movement is induced by the negative element occurring in specTopP:

that under no circumstances,

would

she

t

In addition to assuming that Top and specTopP share the operator feature, [+top], we can assume that they share [+neg]. This latter assumption is crucial as only negative elements in specTopP induce overt Argg-to- Top-movement. Non-negative topics, as in (125), do not trigger overt Agrs-to-Top-movement87, i.e. head movement to Top - due to weak features is deferred until LF: (125) a. Tom says [CP [c [c that] [TopP [cheesej [Top-. b. Tom promised [cpfc· [Q that] [Topp[tomorrowj] dishes ti]]]]].

Jane wouldn't eat tj]]]]]. he will do the

With respect to matrix clauses, the relative position of TopP is hard to test as there are no main clause complementizers in English. I assume that the mam clauses in (126) are identical in 86

87

No matter whether we call them CP-recursion structures (cf. Vikner (1994a,b,1995), Watanabe (1992), Authier (1992), Culicover (1993)), AgrsP-recursion structures (cf. Roberts (1991)), or simply TopP (cf. Hoekstra (1992), M ller/Sternefeld (1993), Alexiadou (1994)) Their not triggering overt Argg-to-Top-movement is presumably due to weakness of features.

140 structure to the embedded clauses in (124), the only difference being that the projection of C is not present.88 (126) a. b. c which I take to be adjoined to CP:

(130)

In this

t^tj sleep ti

Assuming a structure such as in (130), in principle allows for lexical material to intervene between the two TopPs. The data in (131), where an overt ννΛ-element is specCP is flanked by TopPs, are marginally acceptable:90 (131) a. [CP [TopPexi h this room] [CP [ who] [c [c ] [TopPmt t [Topint did]... [Agrsp sleep]]]]]]

never once

] [τορ'»ι

Above, I conjectured that in main clauses, as long as neither head nor specifier of CP are filled, CP may be missing.

142 b [CP hopPed (A8 for) BU1] ICP l w11086 mother] [c [c ] [jopPb, [ never again] [Top'tohopmtwi11]··· [AgrsPhe mvfte]]]]]] c · [CP [TopPot Under pressure] [CP [ which book] [c [c ] [Toppiit t on no account] [Top'mttTopmtWill]... [^p he read]]]]]] d [TopPbt [τορΡβί During all the time John was in Greece] [jopPsm t not once] [Top'int [Topmt didJ-UgrsP he drink ouzo]]]] e John - tTopPfat [TopP«t T™8 room] tTopPmt t never once] [Top'mt tTopint sleep in it]]]] f John - tTopPint [TopPext m *is room] [TopP5llt [ never once] [τορ-irt [Topinl did]... deep]]]] and specTopP^ figure within the A'-system; Top^ adjoins to either CP (13 la) - (13 Ic) ortoTopP^ (13 Id) -(13 If). On the basis of the data discussed k this section and in 5.2, the typology of movement in terms of the A- vs. A'-distinction can be summarized as in (132):

(132) A- system specifier head specAgroP Agr0 specAspP Asp specTP T specAgrgP Agrs specCPrvii91 C[V21

A'-ssy stem specifier head specNegP Neg specTopPbt Toptt specCP C

5.4 Summary At the outset of this chapter I discussed the split-Infl hypothesis and the impact ft had on recent linguistic analyses, especially with respect to parametric variation. The focus in 5.2 was on the arguments that led to reordering the functional heads T and AgrgP. Chief among the arguments were linearization facts, which have been argued to be accounted for by the interaction of the HMC (133) and the mirror principle (134). (133) An X° may only move into the Y° which properly governs ft. (Travis (1984:131) quoted in Baker (1988:53)). (134) Morphological derivations must directly reflect syntactic derivations. (Baker (1985:4)) In the course of the discussion, the functional heads Neg and Asp were discussed.

91

Cf. Roberts (1993:61) for a typologgy of Cs. See also chapter 6.

143

Evidence for both Neg and Asp as functional heads ranked below T and above W has been argued to derive from word order facts in finite and ηοη-finhe clauses. SpecNegP, following Haeberli/Haegeman (1995), has been argued to host a negative operator; Asp, following Koopman/Sportiche (1991), has been argued to temporarily host the subject DP that originates in specVP. Evidence for DP having passed through specAspP comes from floated quantifiers. In 5.2.2 the central issue was object agreement. Based on proposals by Chomsky (1989b), Kayne (1992) and Zwart (1994), I discussed data involving either covert or overt movement into specAgr0P and AgTQ. In connection with this discussion, I reviewed Sportiche's (1990) Strong Correlation Hypothesis, according to which structural Case is generalized to agreement Case. 5.3 focussed on the A- vs. A'-distinction of movement in terms of Relativized Minimality. I proposed a refinement of Rizzi's definition of "typical antecedent governor" to the extent that "typical" refers to features such as "nominative Case" or "objective Case" instead of simply referring to the A- vs. A'-status of the element having moved, its base position and its landing site. A finer grained characterization makes h possible, for example, to prevent DP in AgroP from binding the subject trace in specVP. With respect to the question of what counts as an A- or A'-specifier or as an A- or A'-head in a derived structure, I have discussed movement into the specifier and the head of each of the projections introduced. Both agreement phrases, i.e. specAgrgP and specAgrQ?, figure in the Α-system. The specifier positions are the checking sites for structural Case, i.e. they are licensing positions for DPs. Since their heads bear Case features, they count as Α-heads and as such they license Α-chains. AspP, on the basis of data containing floated quantifiers (135) has been argued to be an A-poshion as the subject, which originates in specVP, can be observed to pass through this position on its way to specAgr§P: (135) [The kidsj ]j may not [all [ tj]]j have tj watched Sesame Street. Licensing operators, SpecNegP, specCP[non_v2] and the specifier of ΤορΡώι have been argued to be A'-specifiers, the respective heads A1- heads. External topics have been argued to be adjoined positions. With respect to specTP and T, I have argued (contrary to Roberts (1993)) that TP figures in the Α-system. I have assumed that specTP is the landing site for temporal adverbs that originate within VP. SpecTP and Τ have been argued to enter an SHA relation based on tense.92

92

This analysis precludes PRO from occurring in specTP in the course of the derivation. See Chomsky/Lasnik (1993:561) for the assumption that specTP is the checking site for the Null Case of PRO.

Part III Lexical Properties of Complementizers, Prepositions and Subordinating Conjunctions and the Structure of Subordinate Clauses Phrase structure, as has been argued in Part Π of this study, is conceived as a projection of lexically defined properties. From among the set of lexically defined properties I will consider only those which are relevant in the computational system, i.e. those which are syntactically relevant in that they determine the architecture of phrases. In order to approach subordination, I first investigate the lexical properties of the categories subordinate conjunctions have traditionally been associated with: complementizers and prepositions. I shall take these two categorization options as the extremes on the spectrum ranging from lexical to functional categories, showing that the majority of subordinating conjunctions are - due to their displaying mixed properties, that is to say lexical and functional properties - in the grey area between these two extremes. Chapter 6 is concerned with the lexical properties of complementizers and the properties of category C. Developing my analysis, I will assume that complementizers make up a functionally defined category. Following a short overview of the treatment and feature specifications of complementizers in PPT, I turn to discuss the invidual hems. In Chapter 71 shall investigate the lexical properties of prepositions, the structure of PP and the functional architecture dominating PP which will be argued to have at most three layers, each of which has to be motivated and licensed by the lexical properties of the prepositional nucleus. The analyses of "pure complementizers" and of "pure prepositions" will serve as the background to the analysis of subordinating conjunctions such as although, because, in case, unless, conditional -whether, etc. in Chapter 8. These elements will be shown to resist clear cut classification in terms of categorial membership, i.e. they cannot be unambiguously attributed to either category C or category P. I will argue that these elements make up a distinct syntactic category (Subcon) which systematically conflates properties of both the functional head C and the lexical head P. In the concluding chapter, Chapter 9,1 briefly discuss the derivation of the relative order of subordinate and matrix clauses within complex sentences.

6 Lexical Properties of Complementizers and Properties of Category C Since Bresnan's seminal work on complementation (1970, 1972, 1973), complementizers have been recognized as a syntactic category which, in recent developments in the concept of X-bar Theory and phrase structure, plays a crucial role in the universe of functional categories. With the rise of category C and its projection in the system of phrase structure, we have not only a category label for "complementizing morphemes" but also a syntactic head position. This head position being a functional one, we also have a potential landing she for verbal head movement, one of the ingredients of parametric variation (cf. chapter 5). As category C can host either head-moved lexical material or lexical material tagged C, both the properties of genuine C-elements and the properties of the syntactic position C have to be investigated. As we have seen in 5.3, the properties of the syntactic position C are not only relevant with respect to head-movement, but also with respect to the licensing of the specifier under SHA. This section starts out with a brief review of the characterization of complementizers in PPT, where I see complementizers as they were originally conceived in the strictest possible sense, namely as elements that serve to introduce argument clauses (cf chapter 2). Taking this perspective, complementizers, i.e. that, for, //and whether, are a subset of subordinating conjunctions which is defined by distributional criteria. These elements, although conceived as fulfilling identical functions, differ from each other not only with respect to selection by governing heads but also with respect to the internal structure of the clauses they introduce. Discussing these elements and the structure of the clauses they are assumed to head, I will isolate the lexical properties of lexical elements that are - so to speak - "base generated" in the C position. In 6.2 I shall discuss the element that and show that there are three different complementizers that, two distinct indicative ones and a subjunctive one. Evidence for this classification comes from selectional facts, movement within the CPs they head and extraction out of CP. The interrogative complementizer (/'will be considered in 6.3. With respect to the elements whether ana for, it will be argued - contra PPT - that these elements are not complementizers. Whether, following Kayne (1990), will be reanalyzed as an interrogative specifier in 6.4. For, in 6.5, will be reanalyzed as a structural Case-assigner whose presence is triggered in certain syntactic environments. 6.6 summarizes the main results.

6.1 Complementizers The most conservative characterization of complementizer is the one put forward in Rosenbaum (1967), Noonan (1985) and Ouhalla (1991), where "complementizer" is taken to be a function fulfilled by a small set of lexical items:1 1

Cf. Stowell (1981:147), Kayne (1982), Webelhuth (1989:184f), Sternefeld (1991:171ff.) and Williams (1994:35f.) for arguments concerning the nominal status ofthat.

148

[...] C elements are basically nominalisers, that is nominal elements whose function h is to nominalise otherwise verbal argument clauses. [...] In view of the standard idea that argumenthood is a property of nominal categories only, it follows that for a sentential clause to be able to function as an argument it has to be nominal(ised). The process of nominaHsation is performed by the C elements [..]. [D.H.] (Ouhalla( 1991:195)) Under this view, complementizers are elements that turn sentential elements, which standardly do not have the status of complements hi the relevant sense, into complements (c£ chapter 2).2 The function of complementizers, then, can be described as converting prototypical propositions into arguments. Thus, following van Riemsdijk (1984:4), we can provisionally state that, "when a clause has a lexical complementizer, h is quite uncontroversial that h is of a specific category [...]", namely CP. The distribution of clauses introduced by "complementizing morphemes" under this perspective straightforward: CPs occur in argument positions. From among the set of subordinating conjunctions only that, for, //and whether can be observed to introduce argument clauses. That these elements are complementizers, Le. elements which occur in C, seems to be beyond controversy. Within lexical projections, the phrases these elements head are restricted to argument positions, ie. they are theta-marked. D-structuralry, they are barred from non-theta-marked positions, Le. they do not serve to introduce adjunct clauses: (1) a. b. c. d.

Joe asked if Tom knew about the gossip. Bill said that Tom had been dating Jane. She wondered whether to invite them or not. It would be inconvenient for you to leave right now.

Current linguistic theory has h that the set of complementizers is subclassified by two binary features [±wh] and [±fmhe], the introduction of which, essentially, is based on insights from subcategorization and w/i-movement.3 Interrogative and dubhative verbs, such as ask and wonder, semantically select interrogative complements, ie. complements which are marked as [+wh]. Only CPs that are [+wh] qualify as embedded questions (la) and (Ic). In order for CP to be [+wh], either the head or the specifier must contain lexical material tagged [+wh]4, in either case, head and specifier enter an SHA relation. The set of lexical complementizers can thus be subclassified by the feature [±wh]: (2) wh

whether +

if that + -

for -

Cf. Tesniere (1959/41982:543-617), who assumes that sentences, i.e. vertal nodes, cannot function as "actants" unless they are "prefixed" by a so-called "translatif'. See chapter 2 for disussion. Cf. Bresnan (1970,1973), Chomsky (1977) and Grimshaw (1977,1979). The latter case involves movement of a [+wh]-phrase into the specifier of CP.

149

The subclassification of the set of complementizers in terms of the [±wh]-specification is relevant not only with respect to semantic selection but also with respect to extraction out of argument clauses. As we have seen in 5.3, extraction out of argument CPs that are [+wh] gives rise to whisland violations. Depending on whether the element which leaves CP is an argument or an adjunct, wA-island violations are either strong or weak: (3) a. b. c. d.

?

Who1 did you ask if Bill had been dating tj? AVhOj do you wonder whether we will meet tj at the opera? *Whyj do you ask if Bill is in a bad mood tj? *Whyj do you wonder whether Bill will quit his job t,?

Whereas specCP hosts a [+wh]-operator in (3a) and (3c), it hosts a wA-moved phrase in (3b) and (3d). In both cases, according to the wA-criterion, C and specCP enter an SHA relation, with [+wh] being the feature that is shared: (4) The Wh-Criterion a. A Wh-Operator must be in a Spec-head configuration with an X0r+wh]· b. An X°r+wfa] must be in a Spec-head configuration with a Wh-Operator (Rizzi(1991:2)) If extraction takes place out of CPs whose head and specifier agree in [-wh], no ungrammatical strings result: (5) a. Whoj do you think that Bill had been dating tj? b. Whoj do you hope for him to meet tj in Prague? A further subclassification of the set of complementizers is derived from the finheness properties of the complement of C, i.e. C is specified as taking either finite or non-finite sentential complements. Since certain verbs not only select sentential arguments of a certain semantic type, but are also selective with respect to whether the sentential argument is finite or nonfinite, finiteness is expected to be visible on CP.5 Whereas if and that can only take finite sentential complements (6a) - (6d) and for only non-finite sentential complements (6e) and (6f), whether can take either (6g) and (6h). (6) a. b. c. d. e. £ 5

The King asked if Columbus had reached India. *The King asked if Columbus to reach India. Everybody thought that he reached India. 'Everybody thought that he to reach India. They hoped for Columbus to reach India. *They hoped for Columbus reached India.

Cf. Wilder/Cavar (1994:24f). This feet is also reflected in Reulaiuf s (1986) [OA]-classification of C, where the phi-features of IP, namely tense, are picked up by C.

150

g. Columbus wondered whether to take some Indians on the ship. h. Columbus wondered whether he should take some Indians on the ship. On the basis of the contrast given in (6), the (non)finiteness distinction is assumed to be a defining one with respect to elements of category C (c£ Wilder/Cavar (1994:24f)). Extending the matrix given in (2), we get the following cross-classification of complementizers: (7)

whether if that wh + + finite ± + +

for -

Taking this rough subclassificaton of the set of complementizers as a first working hypothesis, I now turn to isolating the lexical properties of the individual members of category C and their respective relations to their complement.6 In particular, I will reject the assumption that the relation between C and AgrgP is best characterized by a "mechanism" called f(unctional)selection (cf Abney (1987) and Fukui/Speas (1986)). Grimshaw (1991) assumes that functional heads, as opposed to lexical heads, are non-selecting categories.

6.2 That In this section I return to an issue that was already touched on in chapter 2. Argument clauses introduced by that are of two kinds: indicative and subjunctive. For the purposes of this section it is relevant to keep these two types of //urf-complements apart since - as will be shown - there are two complementizers that which have fundamentally different properties that are relevant not only for the morphological shape of the verbal nucleus they contain and the architecture of the sentential complements of the two lexical items that, but also for selectional properties of the governing predicate. The analysis starts out with the indicative complementizer that, which, for ease of reference, I will indicate as thatind(icative) as opposed to the subjunctive complementizer that, represented as

6.2.1 Thatind Argument clauses introduced by thatin(j, as mentioned above, are always finite and then* interpretation is a definite one; they are distributionally restricted to the internal argument position of "predicates which assert or presuppose truth value, or which imply knowledge or certainty" (Bresnan (1972:65)). Ransom (1988:367), for example, points out that thatind semanticaOy "[...] signals the modality of the complement, namely that it is inclined to be true." Bresnan (1972:69) characterizes thatind as a "definitizer" with respect to the complement it introduces. A similar characterization is found hi Napoli (1993:387), who proposes having 6

See 4.2.

151

definiteness - which is induced by thatind - represented by the feature [+th] on the complementizer. Similarly, Ransom, in her (1986) study on complementation, argues that complementizers are semi-lexical and that that is compatible with definiteness. This basic approach to thatind in terms of inducing definheness has to be refined as argument CPs headed by thatind are of two kinds: they are either factive or non-factive, with definiteness being associated only with factive CPs.7 Regret, realize, point out, admit, ignore and resent in (8e) - (8j) are factive predicates; the events their sentential arguments refer to are definite. In order for some person X to regret Y, Υ must be "true". Truth, in these cases is presupposed. The arguments of non-factive predicates such as believe, conclude, figure and claim in (8a) - (8d), on the other hand, do not have this presupposition, their very nature is assertive. (8) a. b. c. d. e. f g. h. ι j.

He believes that the world is flat. Having heard her arguments, he concluded that she was right. He figured that there was another problem. She claimed that "complementizers" is a rebarbative word. She regretted that she had not accepted the offer, She realized that she had run out of coffee. He pointed out that movement out of adjuncts is impossible, She admitted that she had broken the mugs, He ignored the fact that these elements resist movement, She resented the fact that he had messed it up again.

Factive and non-factive complements are argued to differ with respect to extraction and embedded topicalization.8 Whereas both argument and adjunct extraction out of non-factive arguments is possible (9), argument extraction out of factive arguments yields a weak violation and adjunct extraction yields a strong violation (10): (9) a. WhOj did John believe that Susan invited tj? b. How, did Bill believe that Anne solved the problem tj? (10) a. ?Whoi did Fred confess that he fired tj? b. *Howj did Bill reveal that Anne solved the problem tj? (adapted from Melvold (1991:102))

7

8

See for example Kiparsky/Kiparsky (1971), Bresnan (1970,1972), Hooper (1976), Melvold (1991), Watanabe (1992). It will be argued, following Melvold (1991:104), that there are two /Aoiimrcompleiiieiitizers. Cf. Kiparsky/Kiparsky (1971), Hooper (1976), Melvold (1991) and Watanabe (1992).

152

Extraction out effective arguments (10), which are often referred to as factive islands, patterns like extraction out of w/t-islands (11), Le. argument extraction gives rise to weak island violation, adjunct extraction to strong violation:9 (11) a. ?Whoi do you wonder [CP whyj they will [yp [yp iavhe tj tj]]? b. *Whyj do you wonder [Q> whoj they will [γρ [γρ invite tj tj]]? A further difference between non-factive and factive CPs lies in the fact that only the former may lack the complementizer:10 (12) a. He believes the world is flat. b. Having heard her arguments, he concluded she was right. c. He figured there was another problem. (13) a. *She regrets she had not accepted the offer. b. * She realized she had run out of coffee. c. *He pointed out movement out of adjuncts is impossible. To account for the differences between factive and non-factive complements with respect to Wt-extraction, Melvold (1991) proposes analyzing factive CPs as operator constructions, with the operator in specCP rendering this position unavailable as a landing she (cf ex. (10)).11 Non-factive CPs, on the other hand, do not contain an operator, hence their spec-position is available for w/t-movement (cf ex. (9)). On the basis of these observations concerning definiteness of CP and movement out of CP, Metvold (1991) proposes assuming two tfiaf-complementizers: That in factive CPs has semantic content, Le. it bears the feature [+def(initeness)] which under spec-head-agreement - licenses the iota operator. As"[...] factive verbs select a particular type of functional element as head of their complement [...], namely one which bears the feature [-Kief].", this that is present at d-structure (cf (1991:104)).12 The other that, i.e. the one in non-factive CPs, is, according to Melvold (1991:104), semantically inert and inserted at PF. With respect to extraction out of finite sentential complements and the optionally of that, it should be noted that, traditionally, governing predicates which allow for arguments with these properties are referred to as bridge verbs (or bridge predicates). Bridge verbs constitute a subset of transitive verbs which have the lexically defined property of rendering their finite sentential argument transparent, a means by which they can properly 9

See for example Cinque (1990:29f), Melvold (1991:102), Rizzi (1990:105f), etc. Cf. Melvold (1991), Hooper/Thompson (1973), Authier (1992). It should be noted, however, that native speakers' judgements vary as to whether (13a) and (13b) are grammatical. H Under her analysis, specCP of factive complements hosts an iota operator that binds the referential argument of the verbal nucleus; this operator is not present in non-factive CPs. The referential argument of the verbal nucleus of non-factive CP is assumed to be existentially bound. Cf. Rizzi (1990:106) for a similar proposal. 12 Cf. the contrast between (12) and (13). 10

153 govern the w/t-trace in the embedded specCP as in (9) above.13 This property is crucial with respect to long movement. Since non-bridge verbs lack the property of rendering their argument CP transparent, they cannot property govern a wA-trace in the specifier position of h, hence the ungrammaticality of (10). The frequently stated correlation between long movement and the optionality of /Aa/ind deserves some comments.14 That-m^ is obligatory in bridge contexts, or non-factive contexts, if either extraction (14), or embedded topicalization (15), or negative inversion take place (16), otherwise it is optional (17). (14) a. *WhOj did Bill say that John believed Susan invited tj? b. *Howj did Bill say that he believed Anne solved the problem tj? (15) a. * John says Suei; Bill doesn't like ti (adapted from Authier (1992:329£)) b. *John thinks pastaj, Mary detests tj. (16) a. *Mary kept saying never in her lifej had she tj seen such a thing. (adapted from Authier (1992:329f)) b. *I thought never agakj would the solution tj be so tantaHzingry near. (17) a. She said (that) Bill would like this movie. b. Having solved this problem, she figured (that) there was another one ahead. Muller/Stemefeld (1993) propose an interesting solution to this problem. They "[...] assume that Germanic phrase structure is homogeneous in the sense that h always contains a CP that embeds a T[op]P." [D.H.] (1993:486).15 Crucial to their analysis is the assumption that sentences are always doubly headed, ie. by C, to which inherently nominal properties are attributed, and by Top, to which inherently verbal properties are attributed. As a clause cannot be nominal and verbal at the same time, C and Top are considered competing heads, one of which must be "activated" to perform the function of the designated head, i.e. the one which is being selected.16 They assume that by default C is the designated head. Top, in English, can be the designated head if and only if Top is visible, i.e. if Top contains lexical material17

13

See, for example, van Riemsdijk/Williams (1986:94ff), Stemefeld (1991:146), Müller/Stemefeld (1993) and Haider (1986;1993:72f). 14 See, for example, Kiparsky/Kiparsky (1971), Melvold (1991), Müller/Sternefeld (1993). 15 In their analysis they refer to TopP as TP. In order to avoid confusion with TP (Tense Phrase), I will replace their TP by TopP. !6 In the matrix, where obviously no selection takes place, the designation of C is captured via movement, i.e. I-to-Top-to-C- and wA-to-specCP-movement. 17 Müller/Sternefeld (1993:489) propose the following "Visibility Condition for Clausal Functional Heads": A clausal functional head (C or Top) is designated if and only if it is visible at S-structure.

154

A clear case of Top containing lexial material is (18), the structure of which is given in (19), where preposing of the negative adverbials induces embedded inversion (c£ 5.3 for movement into the head of Top).18 (18) I thought that never again; wouldj the solution tj tj be so tantalizingry near.

(19)

CP C

TopP

AgrsP

s that

never again;

the solution t be tj so tantalizingly near

Here, the finite verb form, would, occurs in Top via head-movement, thus making the verbal head of the clause visible and rendering it the designated head. As in (19) both clausal heads, C and Top, are lexically filled, both count as designated clearly an undesirable situation. To solve this problem, Müller/Sternefeld (1993:429) - on the basis of the "Visibiltiy Condition for Clausal Functional Heads" (c£ note 17) - claim that that in (18) and (19) is not present in the derivation until after S-structure. They propose a PF-rule, given in (20), which is operative hi "[...] the presence of an overt topic in SpecTop." (1993:492). (20) C-tthat/

topic (Müller/Sternefeld (1993:492))

Their proposal is compatible with the one produced by Mervold (1991:104), namely that thatina in non-active CPs, or bridge contexts, is semanticaDy inert and inserted at PF. Embedded topicalization and negative inversion, as Watanabe (1992) points out, based on an observation reported in Kiparsky/Kiparsky (1971), is only allowed hi non-factive complements (21).

18

N.B. Müller/Stemefeld (1993:481) subsume embedded inversion induced by negative-preposing under embedded topicalization.

155

Both processes, which he takes to be instances of CP-recursion, are precluded from occurring in factive complements (22).19 (21) a. John says that Suej, Bill doesnt like tj. b. Mary kept saying that never in her lifej had she tj seen such a thing. (adapted from Autbier (1992:329f)) (22) a. *John regretted that 'Gone with the Wind'j, we went to see tj. b. "John regretted that ηβνεη had he tj seen 'Gone with the Wind*. (adapted from Authier (1992:334)) A further difference between factive and non-factive arguments lies in the fact that only the former may be wA-initial, i.e. clause bound wA-movement is permitted only in factive CPs. (23) a. Fred figured out who j Peter eloped with t j. b. The scientists realized immediately what; they had discovered tj. (adapted from Mervold (1991:101)) c. Fred figured out whyj Peter eloped with Jane tj. d. The scientists realized immediately how, they had discovered the gene tj. (24) a. *Fred insisted who; Peter eloped with tj. b. *The scientists claim what, they had discovered tj. (adapted from Mervold (1991:101)) At a first glance it might seem as if wA-introduced arguments, which, for example, Bresnan (1972) and Baker (1970), take to be indefinite arguments, override the definiteness requirement imposed by factive predicates20; although the argument CPs are [+wh], they are definite. This problem can easily be solved if we assume that [4-whJ-elements can be either interrogative or non-interrogative. I propose adopting Rizzi' s (1991:22, note 3) idea of

19 The data in (i), however, suggest that this generalization is too strong as both embedded topicalization and negative inversion are possible: (i) a. She regrets that under no circumstances would he accept the offer. b. She realized that never again would she get such an offer. c. He pointed out that under no circumstances would movement out of adjuncts yield grammatical results. d. He pointed out that Mary, he wouldn't accept as chairperson although Joan would be acceptable to him. In the case of sentential arguments to nouns, Authier (1992:332) points out that, in general, embedded topicalization and negative inversion are ruled out. Again, the data in (ii) suggest that this is not the case: (ii) a. She wondered whether to take seriously the claim that under no circumstances can CPs iterate. b. The conclusion that under no circumstances can CPs iterate is far-fetched. c. He ignored the fact that under no circumstances would these elements resist movement. d. She resented the fact that under no circumstances could she be crowned Queen. 20 Cf.2.1.

156

representing (non-) interrogativity in terms of the binary feature [±Q].21 Under this analysis all wA-initial22 CPs are marked [+wh], but only those which, in addition, are marked as [+Q] are interrogative and thus are indefinite. Non-interrogative CPs, such as relative clauses and adjunct clauses introduced by wA-elements, are marked as [-Q]. This move allows the wA-criterion to be interpreted as a purely syntactic requirement on the distribution of wA-phrases, i.e. they are required to move overtly irrespective of whether they are [-KH or [-Q]. (25) The Wh-Criterion a. A Wh-Operator must be in a Spec-head configuration with an X°[+whj. b. An X°[+wh] must be in a Spec-head configuration with a Wh-Operator (Rizzi (1991:2)) As Rizzi (1991) himself conjectures, the original formulation is misleading since it addresses only the form of the moved element, namely wA- and not its interpretation, i.e. the wA-criterion as it stands would carry over to relative pronouns and other non-interrogative variants of wAphrases, which clearly cannot be intended. In order to circumvent this undesirable situation, h seems reasonable to shift the orginal (non-)interrogative aspect of vt>A-elements to a lower ranked criterion which addresses interrogativity (and thus mdefinheness) directly in terms of

(26) The Q-Criterion a. A [-i-Q]-Operator must be in a Spec-head configuration with an X°[+Q]· b. An Χ°[-κ}] must be in a Spec-head configuration with a [Q]-Operator The vcA-arguments of the factive verbs figure out and realize in (23) are clearly not interrogative, Le. they are [-Q], and as such they count as definite in the relevant sense. Specifying C as [+wh] and [-Q] accounts for the CPs not being interpreted as indefinite. The wA-elements thus do not have the status of interrogative operators but simply the status of pronouns. In the following I will modify Mervold's (1991) analysis, showing that there is a close relation between definite CPs and non-interrogatrvity on the one hand, and between indefinite CPs and interrogativity on the other hand.23

21

22 23

Cf. Baker (1970) for the introduction of the question morpheme Q and the Q-universal. In addition, see Rizzi (1990:68) for subclassifiying C and CP. CPs that contain an empty w/j-operator in their specifier are w/i-initial. I will not, however, adopt her analysis of binding the event role of the embedded predicate.

157

Meb/old (1991:104), following Abney (1987), assumes that factive predicates select CPs whose head is specified as [-Kief].24 The complementizer that^ which is specified as [+def] licenses an operator hi the specCP. Given that thatm& which introduces factive CPs, bears the feature [+defj, we expect thatm^ to enter an SHA relation with specCP. Agreement between a [-klefj-head and its specifier can be formulated as in (27), analogous to the w -Criterion and the Neg-Criterion (cf 5.3): (27) a. A [-KlefJ-Operator must be in a Spec-head configuration with an X°[4finiteness requirement on CPs whose specifier is lexically filled will be discussed in 6.4.

173

In 5.3 we saw that a CP whose specifier contains an overt [+Q]-operator, i.e. a wA-moved element, is an island. W7i-extraction out of such a CP cannot be through specCP as this position is already filled by a [+Q]-element, thus w/z-extraction, if it takes place across a [+Q]specifier, gives rise to an island violation. Depending on whether the extracted element is an argument or an adjunct, the violation is either weak (68) or strong (69): (68) a. ?Whoj do you wonder [^p why; they will [γρ [yp invite t J t;]]? b. ?WhOj do you think [^pi tj they will wonder [ςρ2 why; he is [yp [yp dating t J

(69) a. *Whyj do you wonder [^p who, they will [yp [γρ invite tj tj]]? b. *Whyj do you think fcpi tj they will wonder [^p2 whoj he is [yp [yp dating t J tj]]]? The asymmetry of argument vs. adjunct extraction out of wA-islands was argued to be derivable from the ECP. Whereas -why in (69) cannot govern its trace within VP because of the presence of \vho in the embedded specCP intervening as a potential antecedent governor, -who in (68) is not prevented from antecedent governing its trace within the embedded VP, as no potential antecedent governor intervenes. As CPs headed by //host an empty interrogative operator, [4Q]Op, we expect extraction out ofif-CPs to pattern like extraction out of CPs whose specifier contains an overt interrogativeelement. This expectation is borne out, as the examples in (70) and (71) show: (70) a. "^Vhoj do ask [CP [j+QjOp] [ς· [Q if] Tom has [yp seen tj]]] b. "^Whatj did you ask [CP [[+Q]Op] [c- [c if] Tom [yp knew t,]]] (71) a. *Whyjdoyouask[CP[[+Q]Op] [c. [ c ifJTomhas[yp[ypseen Jane]tj]]]? b. *Howj did you wonder [^p [r+oiOp] [Q· [^ if] Tom [yp [yp made h home] tj]]]? Again, antecedent government of an argument trace is possible (70), whereas antecedent government of an adjunct trace is blocked by an intervening A'-specifier, i.e. by [+QjOp in specCP (71). CPs headed by //resemble fkctive CP arguments in that they constitute extraction-islands, and they resemble non-factive CPs in that both are instances of indefinite expressions with interrogative arguments displaying a higher degree of mdefinheness than non-factive arguments. Bresnan (1972) and Baker (1970), for example, take wA-introduced arguments, Le. interrogative arguments, to be indefinite arguments.50 In 6.2, it was argued that [+wh]-elements should be specified for (non-)interrogatrvity, that is to say in terms of the binary feature [±Q] (cf Rizzi (1991:22, note3)). Under this analysis any CP whose specifier hosts a wA-element which is marked as [+Q] is interrogative and thus counts as indefinite. Indefiniteness in con-

50

Bresnan (1970:310) does not identify if as [+wh].

174 junction with mterrogativity precludes CPs headed by //from functioning as arguments of noninterrogative heads (cf. (65e) - (65h)).

6.4 Whether The distribution of argument CPs introduced by whether is identical to the distribution of argument CPs headed by if, both elements are specified as interrogative (cf (7) above). CPs introduced by whether occur as the internal argument of an interrogative head (72a) - (72c), and they are barred from non-interrogative contexts (72d) - (72f). (72) a. Columbus wondered whether he should take some Indians on the ship. b. I am asking whether you will accompany me. c. He doubted whether they were at home. d. *Cohunbus said whether he should take some Indians on the ship. e. *I think whether you will accompany me. £ *He pointed out whether they were at home. Argument CPs introduced by whether differ from argument CPs headed by ;/ or that^^^ in that they can be either finite or non-finite: (73) a. b. c. d. e. f

Columbus wondered whether he should take some Indians on the ship. I am asking whether you will accompany me. He doubted whether they were at home. She wondered whether to invite them or not. Columbus wondered whether to take some Indians on the ship. I wonder whether to go home right now.

Under the perspective we have taken so far, complementizers, and hence CPs, are specified as either [+fin] or [-fin]. CPs introduced by whether fall out of the system as they can be either [+fin] or [-fin]. If the CP introduced by whether contains a finite T, the subject is lexical and it is assigned nominative Case under SHA hi AgrsP.51 If T is [-fin], the subject is PRO. One of the features of PRO is that this element is barred from governed positions.52 The relevant case to look at is (74), as the complementizer for, which exclusively heads non-finite arguments, does not embed complements containing an empty pronominal subject, i.e. PRO: (74) a. *We hoped/or PRO to win. b. *For PRO to leave right now would be inconvenient. ^ * I.e. nominative Case assignment properties are parasitic on a finite T. 32 PRO, specified as [+anaphoric,+pronominal], is taken to be subject to the Principles A and B of Binding Theory. The so-called PRO-Theorem is derived from the fact that PRO is subject to two principles. The PRO-Theorem essentially states that PRO must be ungoverned (cf. Chomsky (1981,1986a,b)). PRO is governed if C contains lexical material.

175 Given that ./or and whether are complementizers (75), why is it that only for and not whether illicitly governs PRO? Or to put the same question in a more archaic form, why is there a "forto-filter"53 and not a "whether-to-filter"? (75)

CP C

AgrSP *for whether

PRO to win PRO to win

Among the set of complementizers only whether allows for finite and non-finite complements, with the latter containing PRO as in (75). There is no correlation between whether and its complement in terms of finfteness. Argument CPs introduced by whether share these two properties with sentential arguments introduced by w/j-moved interrogative elements such as when, where and who in (76): (76) a. He doesn't know whether to go to the movies. b. He doesnt know when to go to the movies. (Kayne(1990:16)) c. d. e. f. g. h.

He doesn't know where to go to the movies. He doesn't know who to go to the movies with. I wonder whether we will go the the movies. I wonder when we will go to the movies. I wonder where we will go to the movies. I wonder who we will go to the movies with.

Kayne (1990:16f), drawing on work by Katz/Postal (1964), Bresnan (1972) and Larson (1985a), who point out that whether behaves like the wh-form of either5*, argues that analyzing whether as a wA-phrase in specC accounts for the fact that CPs introduced by whether internally pattern like CPs introduced by other w/i-elements, i.e. the CPs can be either finite or ηοη-finhe, PRO in non-finite CPs is not governed. Whether as a specCP-element thus is not specifed for [±fin] as this feature is a head feature. Supportive evidence for Kayne's (1990) claim that whether is a specifier and not a head comes from earlier stages of English where whether and that, a genuine C-head, could cooccur

53

54

Cf. Chomsky (1977:103).

See 2.2.3.1 for discussion. Cf. Chomsky/Lasnik (1993:56) for an analysis of whether as a specCP element.

176 in the same way wA-moved elements and that cooccured in embedded interrogative CPs.55 In (77), both specC and the head position contain lexical material: (77) a. ye desire to knowe whedyr that I shulde abide here stille or mowe b. men shal wel knowe who that I am (taken from Lightfoot (1979:321f.)) In 2.2.3.1 I reviewed Bresnan's (1972) arguments for assigning a unique structure to interrogative clauses, irrespective of whether they are matrix interrogatives or embedded ones. In our system, interrogative clauses are instances of CP whose specifier position contains a [+QJelement. Of the set of interrogative wA-elements only whether is precluded from matrix interrogatives: (78) a. 'Whether will he come on time? b. *Whether did he do the dishes? CPs introduced by -whether are alternative questions. In matrix alternative questions there is no overt interrogative operator in specCP. The fact that AgrgP-to-C movement takes place is accounted for by SHA between C and the empty operator r+oiOp in specCP. (79) a. Will he come on time? b. Did he do the dishes? Unlike overtly introduced matrix interrogatives, there is no movement into specCP in matrix alternative questions. The same situation obtains in embedded interrogatives. What I will assume is that -whether is the lexical counterpart of the empty interrogative operator. In 6.2 it was argued that [+Q], being a strong feature, must be lexicalized. Thanks to SHA, [+Q] can be lexicalized either in specCP (80), in C (81) or in both specCP and C (82). (80) a. He doesn't know who to go to the movies with, b. I wonder whether we will go the movies. (81) a. Will he come on time? b. Did he do the dishes?

55

A similar phenomenon is claimed in Rizzi/Roberts (1989:4) for matrix questions. They give examples of Quebec French matrix questions which involve a root complementizer: (i) Qui que tu as vu? Who that you have seen? (ii) Qui as-tu vu? Who have-you seen? (iii) *Qui qu'as-tu vu? Who that have-you seen?

177

(82) a. What is this good for? b. Why has he done the dishes? It is generally assumed that Agr§-to-C movement is barred from embedded contexts56, i.e. legalization of [+Q] in argument CPs is predicted to be legalization of specCP. The examples m (83), which are often referred to as semi-indirect speech, show very clearly that the ban on embedded Agrg-to-C movement is too strong, as there is embedded V2.57 (83) a. You asked them would they marry him. (Rizzi( 1991:5)) b. I wondered was he a spy. CPs hosting either an overt wA-element or an empty interrogative operator in their specifier position constitute extraction islands (cf. 6.3). Argument CPs introduced by whether as well as the V2-mterrogatives in (83) are wA-islands. For by now familiar reasons, argument extraction out of these CPs yields a weak island violation (84), adjunct extraction a strong one (85): (84) a. ?Whoj do you wonder whether we will meet tj at the opera? b. AVhoj did you ask her would she marry tj? (85) a. *Whyj do you wonder whether Bill will quit his job t,? b. *Whyj did you ask her would she marry Tom tj? Analyzing whether as a genuine specCP element reduces the set of overt complementizers. What we have in the case of CPs introduced by -whether, in fact, is an empty [+Q]-complementizer, which - by SHA - agrees with specCP thus satisfying the Q-criterion (see (26), 6.2). [+Q] being a strong feature, it requires that either the head or the specifier be lexicalized.58 Apart from the fact that whether is not a complementizer but a base-generated specifier, these CPs underty the same restrictions and they display the same properties as CPs headed by // Furthermore w/ie/Ae/--initial CPs, since they are specified as [4Q], are interrogative and thus, by definition, indefinite. There is one property whether and if do not share, and this is the [+fin]-specification which is irrelevant for embedded interrogatives.

56

See for example Rizzi/Robeits (1989) and Gnmshaw (1994), who assume that Agrs-to-C movement in CP arguments would give rise to a violation of the projection principle. But see note 41 above. ^ Similarly, in overtly introduced interrogative CP arguments C may be lexically filled: (i) I wondered what kind of party had he in mind. (Radford (1990:117)) (ii) I wondered where had he been. Further evidence for the availability of embedded Agrs-to-C movement comes from counterfactual conditionals (cf. 6.2): (iii) Had I repaired the car, the accident would not have taken place. '* In non-alternative interrogatives, both head and specifier are filled.

178 6.5 For

Unlike //w/m(j, subject.

and //, for takes non-finite declarative complements with a lexical

(86) a. We hope for him to succeed. b. *We hope for PRO to succeed. c. It is inconvenient for him to iron the shuts. d. *It is inconvenient for PRO to iron the shuts. The lexically realized subject in these constructions is assumed to occur in specAgr§P where it is assigned structural Case, i.e. objective Case, under government by for in C59:

(87)

tj succeed tj iron his shirts

As for in (87) is characterized as a (de-prepositional complementizer that qualifies as a structural Case-assigner, the examples in (86b) and (86d) are ruled out because PRO, which,

Structural Case, as was pointed out in chapter 5, can be governed Case. The cases we looked at in that context were cases in which specAgrgP is assigned Nominative Case. In the constructions under discussion, for in C is taken to assign objective Case to specAgrsP under government. Structural Case-assignment by for takes place in very much the same way as in Exceptional) C(ase) M(arking) constructions, i.e. across an AgrsP-boundary.

179 by the PROTheorem is precluded from governed positions, would be governed by an overt complementizer. 60 In Middle English, as the data in (88) show, for and to could occur adjacent to each other. (88) a. b. c. d.

For to go is necessary I for to go is necessary it is good for to go it is necessary a man for to go (taken from Lightfoot (1979:187))

Roberts (1993:260) points out that there are two analyses available: either for-to is a complex complementizer, or to is in C, with CP governed by the preposition for. As there is good evidence for both analyses, Roberts (1993) claims that, whichever analysis is adopted, the relevant outcome is that to is in C. Evidence for this assumption derives from data as in (89), where the non-finite verb which occurs in T bears infinitival inflection (-en). (89) Ne herte hath noon to stonden at deffense (taken from Roberts (1993:260)) In Middle English, movement of V-to-T was triggered because T was strong as it hosted the suffixal infinitive marker -en. Infinitival T became a host for to, the substitute for infinitival inflection, only after the decline of verbal inflection. To, originally a complementizer, was reanalyzed as part of the infinitval inflection.61 The well-formedness of (88a) and (88c) is argued to be due to infinitives in Middle English being conceived as nominal. With the loss of verbal inflection, i.e. the infinitive suffix -en,

60 Rizzi (1990) proposes a slightly different structure. He assumes that the preposition for is inserted into C, i.e. into a non-genuine head position. For, under his analysis, would not be a complementizer but a preposition. CP

for

61

him

to

With respect to (88b) and (80d), where a nominative DP precedes the non-finite clause, Roberts (1993:260) points out that constructions of this type are not very frequent and he decides "[...] to treat them as a sporadic reanalysis of the compound_/orto complementizer as an infinitival marker."

180 infinitives were reanalyzed as being sentential62 Being sentential, Modem English to-infinitives are barred from the complement position of prepositions. In Middle F.nglijji> sequences like vw/A-to, through-to, etc. were perfectly well-formed. Roberts (1993:262) argues ih&tfor, as opposed to other Ps, "[...] was semanticalty empty and phonologicalty light enough to be reanalyzed as a complementizer." The only property for, originally a preposition, has kept in these constructions is the property of assigning objective Case. Distributionally, CPs headed by for can be described as a subset of non-finite sentential arguments with a PRO subject: Whenever a sentential argument introduced by for is possible, a sentential argument of the type PRO-to-...-VP is possible as well (90). Since the reverse is not the case, as can be seen in (91), we have to conclude that these two types of non-finite arguments are not distributionally equivalent. (90) a. b. c. d.

We hope for him to succeed. We hope PRO to succeed. I prefer for him to do the dishes. I prefer PRO to do the dishes.

(91) a. b. c. d. e. f g. h.

He tried PRO to remove the garbage. *He tried for him to remove the garbage. He promised PRO to remove the garbage. *He promised for him to remove the garbage. I forced him PRO to remove the garbage, *I forced for him to remove the garbage. I convinced him PRO to remove the garbage, *I convinced him for him to remove the garbage.

PRO in argument clauses must be obligatorily controlled by a lexically overt DP. Arbitrary control of PRO is possible in extraposed clauses (92) but not in declarative sentential arguments (93):63 (92) a. PRO to drive home drunk is dangerous. b. PRO to behave oneself in public is necessary. c. PRO not to be invited to this party is a relief (93) a. b. c. d.

*He tried PRO to behave oneself in public. *He promised PRO to behave oneself in public. *I forced him PRO to behave oneself in public. *I convinced him PRO to behave oneself in public.

62

Cf. Lightfoot (1979), Roberts (1993), Stockwell (1973). 63 Manzini (1983) points out that arbitrary control in sentential arguments is possible if these are interrogative:

(i) John asked how PRO to behave oneself. (ii) John was asked how PRO to behave oneself. (Manzini (1983:425))

181 Obligatory control of PRO can be control by the matrix subject (9 la) and (91c), or the matrix object (9 le) and (91g).64 As control properties are lexically defined properties, e.g. try is lexically specified for subject control, force for object control, these verbs do not take sentential arguments other than those of the format "PRO-to-...-VP".65 Among the set of sentential arguments verbs like hope, prefer and want allow for, we also find the "PRO-to-...-VP"-type. PRO, in these examples, is controlled by the matrix subject. These constructions are often characterized as involving optional control, Le. the matrix verb has control capacities which it need not exercise unless the requirement has to be met that PRO be controlled if in an argument clause (94b), (94d) and (94t).66 (94) a. b. c. d. e. £

We hope for him to succeed. We hope PRO to succeed. I prefer for him to do the dishes. I prefer PRO to do the dishes. We wanted for him to buy these books. We wanted PRO to buy these books.

There is another construction type which alternatively allows for "PRO-to-...-VP" and Uieforinhial to-infinitive construction: (95) a. b. c. d. e. f

It is necessary for him to leave on time. It is necessary PRO to leave on time. It is inconvenient for him to iron the shirts. It is inconvenient PRO to iron the shirts. It is not sufficient for Tom to do his job. It is not sufficient PRO to do one's job.

In these examples, as well as in (96), PRO is subject to arbhray control (96) a. PRO to leave on time is necessary. b. PRO to iron the shuts is inconvenient. c. PRO to do one's job is not sufficient.

64

Cf. Manzini (1983:423).

65

Cf. Napoü (1993:321ff).

66

The sentential argument may alternatively be realized as a non-factive CP (i) or as a subjunctive CP (ii): (i) We hope (that) he will succeed (ii) I prefer that he do the dishes.

182 In (95) the subject of the matrix predicates is an expletive, i.e. an element which is not thematic.67 Expletive it is assumed to be coindexed with the "PRO-to-...-VP" argument. In chapter 1, constructions of this type were dealt with under the topics "extraposition structures"68 and "intraposition structures"69. The analysis which seems to be strongly favored in recent linguistic theory is one which is similar to that presented in Emonds (1970,1976) and Jackendoff (1977). The similarity, however, does not go beyond the assumption that movement of "PRO-to-...-VP" is leftward movement and that pleonastic it and the "PRO-to-...-VP" argument are coindexed and thus form a chain. As specAgrgP is an Α-specifier to which nominative Case is assigned under SHA and as clauses are not entities that have to satisfy the Case fiber, there is no reason to assume that movement of "PRO-to-...-VP" is movement into specAgrsP. Koster (1978) and Emonds (1976) observe that so-called subject sentences do not take part in the regular subject-auxiliary-inversion processes, Le. there is no Agrg-to-C movement. They propose having "subject sentences" introduced as topics (cf Chomsky s (1977:91) phrase structure rule for topics).70 If we follow Koster (1978) and Emonds (1976) in assuming that sentential subjects do not occur in the subject position, we have to take (97) to be the structural representation of the data in (96). "PRO-to-...-VP" movement is into specTopP.71

(97)

TopP spec

to leave on time

is necessary tj

There is, however, one problem with this analysis. Expletive // is deleted when "PRO-to-...VP" is moved into specTopP. As expletives are not interpretable at LF since they are semantically vacuous, they must be removed. Expletives, according to Chomsky (1993, 1994), can be 67

68

69 70 71

Arbitrariness of control is most obvious in (95f) where the indefinite possessive pronoun takes PRO as its antecedent. See also note 63. Cf. Chomsky (1965/161990), Rosenbaum (1967), Rutherford (1970), Andersson (1973), Lakoff (1968), Freidin (1971), etc.

Cf. Emonds (1970,1976) and Jackendoff (1977) for discussion. Cf. Stowell (1981: 153f.) and Webelhuth (1989:181f.) for supportive evidence. But see Webelhuth (1989:185), who assumes that CPs headed by that can function as the verb's external argument.

183 removed by adjoining the coindexed phrase. For the purpose of the argument we can assume that the expletive is replaced at LF by the coindexed phrase. As the expletive is in specAgrgP for Case reasons, we expect replacement of the expletive by the coindexed phrase to take place in specAgrgP. This strategy would yield a structure in which "PRO-to-...-VP", contrary to what Koster (1978) and Emonds (1976) assume, ends up in the sentential subject position. Alternatively, to rule out this situation, we could assume that "PRO-to-...-VP" passes through specAgrgP deleting expletive it. I have no solution to this problem. With respect to the data in (95a), (95c) and (95e), we have to race the question whether expletive it is coindexed with "for-DP-to-...-VP" or only with "PRO-to-...-VP". I will argue that both options are available. In (98), "for-DP-to-...-VP" precedes the finite verb, and in (99) only "to-...-VP" occupies specTopP. The phrase headed by for remains in situ. (98) a. For him to leave on time is necessary. b. For him to iron the shirts is inconvenient. c. For Tom to do his job is not sufficient. (99) a. To leave on time is necessary for him. b. To iron the shirts is inconvenient for him. c. To do his job is not sufficient for Tom. Consider next the data in (100), where for-DP is extracted and the matrix clause remains intact, Le. // is not deleted and thus it forms a chain with the infinitival argument: (100) a. For him, h is necessary to leave on tune. b. For him, h is inconvenient to iron the shirts. c. For Tom, it is not sufficient to do his job. Neither the data hi (99) nor those in (100) can be explained if we assume (87) to be the representation of the internal structure of the string introduced by for. To derive (99) TP would have to be extracted from within the argument CP. To derive (100) the complementizer and the subject DP would have to be extracted. In the light of (87) both extractions are ruled out, and the only way to account for the wellformedness of (99) and (100) is to assume that predicates like be necessary, be inconvenient and be sufficient in the cases considered select a non-finite argument, and that, in addition to selecting "PRO-to-...-VP", these predicates optionally allow for selecting a PP headed by /or.72 Assuming that there are two arguments, Le. a PP and a CP, not only gives an explanation of why (99) and (100) are well-formed, but it also paves the way for explaining why expletive it does, in fact must, occur in (100). As has been argued above, only DPs can function as

72

The structure I assume here corresponds to the structure Lightfoot (1979:196) and Roberts (1993:259) assume to be the adequate structural analysis for Middle English. With respect to Modem English they assume that for-DP is not a PP, i.e. for is analyzed as C and DP as specAgrgP.

184

subjects, i.e. what is being construed as a "subject sentence" is in fact in specTopP. The CP remains in its base position and is coindexed with expletive it. Taking these observations as a background, the data given in (98) are well-formed under so-called comma-intonation, i.e. the two moved constituents occupy two different specTopPs; (101) a. For him, to leave on time is necessary. b. For him, to iron the shirts is inconvenient. c. For Tom, to do his job is not sufficient. The second option, namely that "fbr-DP-to-...-VP" is coindexed with expletive it in (98), requires us to consider the data in (94), repeated as (102). (102) a. b. c. d. e. f

We hope for him to succeed. We hope PRO to succeed. I prefer for him to do the dishes. I prefer PRO to do the dishes. We wanted for him to buy these books, We wanted PRO to buy these books.

Here it is implausible to assume that the governing verbal heads optionally take two internal arguments, as neither the string "for-DP" nor the string "to-...-VP" can be extracted: (103) a. *For himj we hope tj to succeed. b. *For himj I prefer tj to do the dishes. c. *For himj we wanted tj to buy these books. (104) a. *To succeedj we hope for him tj. b. *To do the disheSj I prefer for him tj. c. *To buy these books we wanted for him tj. But what is the structure of the non-finite argument? As we have seen in connection with subjunctive CPs and with interrogative CPs, selection is a strictly local process. A governing head cannot select across its complement, Le. verbs like hope, prefer and -want can only select for features visible either on C or specCP. The feature they select for is clearly [-fin], and [-fin], under PPT and minimalist analyses is a feature for which./br is lexically specified. Having this specification of for makes it possible to instantiate a second local relation, namely the relation between for in C and to in T. In order to guarantee selection in the case of non-finite CPs whose head is not lexically filled, one has to assume that for has a non-lexical counterpart, which, except for the Case-marking properties of for, is characterized by identical features.73 The situation that obtains here is similar to the one we faced in connection with the subjunctive in section 6.2.2, where I assumed - following Roberts (1993) - that the modal should has a non-lexical conterpart characterized by identical features.

185

Infinitival arguments, though not distributionally equivalent (cf above), are functionally equivalent in that they are associated with intentionality.74 Intentionality, in the relevant sense, is closely tied to notions such as future-oriented. It has often been argued that non-finite sentential arguments containing to, although they are non-finite, are associated with future tense. Stowell (1982), for example, argues that the events referred to by to-infinitives (105), as opposed to those referred to by gerunds (106), are never interpreted as contemporaneous with the event in the matrix, but rather as potential or future events:75 (105) a. Jenny remembered PRO to bring the wine, b. Jim tried PRO to lock the door. (106) a. Jenny remembered PRO bringing the wine, b. Jim tried PRO locking the door. (Stowell (1982:563)) Van Gelderen (1993:114) proposes having governing verbs such as try, convince, hope, want, etc. specified as selecting complementizers which are specified as [+fut]76, with [+fut] being conceived as a grammatical feature in the sense of Ouhalla (1991:16). In Van Gelderen's system, [+fut], in turn, selects a non-finite T, Le. a T which is lexically filled by to. She (1993:112f£, passim) assumes that there is no AgrgP in non-finite CPs and that specTP is an A-poshion that hosts the subject. The [+fut] feature in C may be lexicalized by_/br, or it may be realized as a 0-complementizer.

(107)

(adapted from van Gelderen (1993:114))

74

75 76

Whereas Bresnan (1970,1972) argues that ./or is associated with intentionality, Grimshaw (1977:90) argues that intentionality is not so much a property of the lexeme for but a property of non-finite clauses, specifically of non-finite clauses containing the ίο-infinitive. See, for example, Bresnan (1970) and van Gelderen (1993) for related arguments. I will not adopt her analysis of subjunctive CPs as non-finite clauses.

186 Since, as has been argued above, "PRO-to-...-VP" is possible in all contexts in which "for-DPto-...-VP" is possible (but not vice versa), there is good evidence that for is indeed never selected. Instead, for surfaces only if the occurrence of a lexical subject is not precluded by control requirements imposed by the governing verb, i.e. in cases where the governing verb is an optional and not an obligatory control verb. For has similarities to both dummy-do (108) and the Case-prepositions77 of, on and at in (109). (108) a. What did you buy? b. *Whatyoubuy? c. I don't know who did it. d. *I not know who did it. e. * I do cannot pick you up at the station. (109) a. b. c. d. e. f g. h. i j.

The refusal of my offer upset me. "The refusal my offer upset me. She was proud of him. * She was proud him. She wasn't keen on housework, *She wasn't keen housework. She wasnt good at h. * She wasn't good it. *Of the documents have been stolen. *On the documents have been stolen.

Diunmy-db surfaces if and only if serves to check off strong features. Insertion of do in contexts other than these leads to ungrammaticality. Similarly, Case-prepositions surface if and only if a nominal constituent, otherwise licensed hi its position i.e. by being theta-marked, would violate the Case-filter.78 Case-prepositions are not inserted blindly to rescue the structure; the governing head determines which Case-preposition has to surface. Surfacing of on or of m contexts in which DP can be assigned Case, e.g. under SHA by Agr§, gives rise to ungrammaticality as hi (109i) and (109j). As a lexical subject cannot be assigned Case under SHA by AgrgP in non-finite clauses, there must be some other means of Case assignment in order for DP to be licensed. In the cases under consideration Case is assigned under government by for surfacing in C:

77 78

Of. Rauh (1993a,b,1995a,b) for a detailed discussion of Case-prepositions. Of, as Chomsky/Lasnik (1993:558ff) argue, can be considered a realization of inherent Case. Rauh (1992: lOff.) points out that inherent Case in the context of nouns and adjectives may be realized by prepositional forms other than of and she maintains that verbs and prepositions alongside nouns and adjectives qualify as inherent Case-assigners (1992:27).

187

hope

for

For, as opposed to the Case-preposition of m (109), is not selected by the governing verb and h is not a realization of inherent Case. Unlike dummy-do, for contributes to the structure in that it assigns objective Case to DP in specAgrsP. The only trigger for the occurrence of for in C is the presence of a lexical DP in specAgrgP. This situation arises if and only if the governing verb imposes no control restrictions on the ηοη-finhe clause it selects. I assume that for - as opposed to that^d/^j - is not a complementizer but a Case-assigner whose presence is required for reasons that have nothing to do with selection. Neither [-fin] nor [+fut] can be observed to be strong features. If they were strong features, we would expect them to be lexicalized. For, arguably, cannot be a legalization of these features, as the same feature specification accounts for the CPs of the form "PRO-to-...-VP", where there is no overt lexical material in C. CPs which are headed by an empty C which is specified as [-fin], [+fut]79 share with nonfactive CPs, subjunctive CPs and with embedded interrogatives the property of being [-def].

6.6 Summary In this chapter I have analyzed the elements which introduce argument clauses and which PPT classifies as complementizers, i.e. as elements that take sentential complements. From among the set of elements which PPT treats as complementizers, I have argued that only / a/jnd, that^^ and if are complementizers. Whether, I have argued, following Kayne (1990) and others, is not a complementizer but a specifier. For was analyzed as a Caseassigner which, given that certain conditions are met, is inserted into C. On the basis of feature selection, internal processes and extraction out of argument CPs, I have tried to isolate lexical properties of both complementizers and the functional head position C. The features which were considered relevant in this chapter are listed in (111): 79

Specifying C as [-fin], [+fut] is not a redundant specification as gerunds, which are also [-fin], are not [+fiit] but [-fut] (cf. note 82).

188 (111) [+wh]: [+Q]: [-Q]: [+defj: [-def]: [+fin]: [-fin]: [+subj]: [+fut]: [-tut]:

governs the distribution of wA-elements (ie. their occurrence in specCP) marks [+wh]-elements as interrogative marks [-Hvh]-elements as non-interrogative marks definiteness marks indefinheness marks finheness marks non-finiteness marks subjunctive mood marks future tense (ίο-infinitives) marks co-temporaneous tense (gerunds)

Of these features, [±fin], [+subj] and [±fut] were observed to interact both with selectional requirements imposed by the governing verb and with the morphological properties of the embedded verb. In all the contexts considered, [±fin] and [±fiit] are weak features and as such need not be lexicalized. Any movement of verbal material into a C head which is specified by any of the possible combinations of these features is covert movement, i.e. it takes place at LF. [+subj] was shown to be a strong feature which must be lexicalized, with legalization in English being restricted to the occurrence of that^. [±wh], [±Q] and [±defj are operator features and as such they do not interact with the morphological properties of the embedded verb but only with the selectional requirements of the governing verbs. The positively specified variants of these features, i.e. [+wh], [+Q] and [+defj, are strong features and as such must be lexicalized, with lexicalization, due to SHA, taking place either in the specifier or in the head position. Lexicalization of [+wh] and [+Q], with the exception of if, takes place hi specCP. [+defj is invariably lexicalized in C. The various combinations of these features were argued to account for selectional facts, internal movement and extraction out of CP. (112)-(117) summarize the feature specifications which, by SHA, are visible on both C and specCP and then- respective lexicalization options in English: (112) [-Kief], [+fin]: f ctive thatind CPs specified with this feature specification are extraction islands.80 ^-extraction out of these CPs, leads to interpreting wA-moved elements as interrogative, i.e. [+Q]. The traces of the wA-moved element are [+QJ. SpecCPj+^fj,[+fin] hosting an intermediate trace that is specified as [+Q] gives rise to the by now familiar island effects. An overt [+QJ wA-element in specCPr+defj^+fin] is incompatible with definiteness and is thus ruled out. fFA-elements which 316 [~Q] can occupy this position. (113) [-def], [+fin]: non-factive thatind

80

Similar effects can be observed with extraction out of DPs (cf. Abney (1987), Melvold (1991)). Wh-txtraction out of indefinite DP yields grammatical results, whereas wA-extraction out of definite DP results in ungrammaticality (cf. note 24).

189

H7i-extraction out of CPs specified as [-defj, [+fin] yields grammatical results because of the basic compatibility of indefiniteness and interrogativity. The feet that overt [-KJ] w/z-elements in specCP[_(jef] r+finj are ruled out derives from the selectional properties of the governing head. (114)[-def], [+fin], [+subj]: Subjunctive CPs resemble non-fective CPs in that they are both finite and indefinite. Long movement out of these CPs is possible because indefiniteness and interrogativity are compatible. As the governing verbs selects [+subj]81 but not [+Q], the feet that overt [+Q] wAelements are precluded from occurring in specCP^q^r+fin],[+subj] can ^e accounted for. (115) [-defj, [+fin],[-K}]:/y CPs that are specified as [-defj, [+fin], [+Q], as opposed to those specified as in (113), are islands for wA-extraction. This feet is accounted for by the [+Q] feature which, besides being a strong feature and thus requiring to be lexicalized, is selected by the governing verb. (116) [-defj, [-*$: whether With respect to long wA-movement and selection, CPr^f] Γ+QI patterns like CPs specified as in (115), i.e. both types of CP constitute wA-islands and embedded interrogatives. They differ from CPs headed by //in that they can be either finite or non-finite and in that the strong [-fQ] feature is lexicalized in specCP in situ. The characterization given in (116) carries over to CPs into whose specifier interrogative wA-elements are moved. (117)[-def],[-fin],[+rut]:0 All this feature specification gives us is an empty C which, as [+fut] is not a strong feature, hosts a T containing the infinitive marker to, but does so only at LF.82 Neither selectional requirements of the governing verb nor morphological requirements of the embedded verb force [-defj, [-fin], [+fut] to be lexicalized. The insertion of/or in this position is required by Case-requirements of the DP in specAgrgP. For, under this analysis, is not a complementizer and it does not serve to lexicalize any of the features since legalization is not required. All it contributes to the structure is objective Case which ft assigns to DP hi specArggP under government.

[+sub] is a strong feature which requires to be lexicalized. Gerundive arguments could be assumed to be introduced by an empty C specified as {[-def], [-fin], [-fut]}, with {[-fin], [-fut]] arguably being hosted in T. But see van Gelderen (1993:123), who suggests analyzing gerundive arguments as bare VPs whose head is specified as [-tense].

7 Lexical Properties of Prepositions Prepositional elements, as the various characterizations of these elements from traditional grammar to generative grammar show, cannot be taken to represent a homogeneous class (c£ chapter 3). This is the reason for the longstanding question of whether prepositions are lexical or functional heads. There are basically three positions in recent linguistic work: prepositions are lexical heads1, prepositions are functional heads2 and prepositions straddle the dividing line between the lexical and the functional universe. The analysis in this section is based on the view that prepositions cannot in toto be assigned to either universe, i.e. they straddle the line between lexical and functional categories (c£ Abney(1987)). Most of the arguments in favor of dividing the set of prepositional elements into two groups, Le. attributing them to one of the two universes, derive from thematic properties individual lexical items display or do not display.3 Prepositional elements that display thematic properties are often termed lexical prepositions* or semantic prepositions*. Rauh (1993a,b,1995a,b) characterizes this subclass on the basis of its individual members having a lexical argument structure with one external, one referential and < 1 internal argument, allowing multiple selection of the internal argument, having descriptive content6 and Case-assigning properties, syntactically and semantically heading their own projections and allowing for modification.7 (1) a. b. c. d. e. £ g. h. 1

2

3

4 5 6

7

I didn't see him before the conference. I didn't see bun before the conference started. I didnt see him before. He came right before the conference started. He arrived two hours before me. The dizzy turns started long before the conference started. He stood before me. He stood right before me.

Cf. Jackendoff (1977), van Riemsdijk (1978, 1990), Zwarts (1992), Higginbotham (1985), Dubinsky/ Williams (1995), Koopman (1993), Sportiche (1990). See, for example, Lasnik/Saito (1992), Vikner (1994), Grimshaw (1991), but see 4.2.2. Emonds (1976,1985) considers them grammatical formatives. A similar situation obtains with verbal elements which are argued to figure in both universes. Besides lexical verbs, there are verbal elements, such as aspectual auxiliaries and modals, which are assigned a functional status within the extended verbal projection. With respect to prepositions the situation is different as functional prepositions, such as Case-prepositions, do not form an extended projection with a lexical preposition but participate in extended projections that figure as arguments in the projection of lexical heads. See, for example, Emonds (1985), Rauh (1993a,b,1995a,b), Zwarts (1995a,b), Haumann (1993). Cf. Williams (1989). Descriptive content here refers to spatial or temporal semantics and thus is closely tied to the referential argument (cf. 4.2.1). Zwarts (1995a) points out that, in addition to these properties, lexical prepositions can be the "input to morphology", i.e. they participate in word-formation processes.

191

Lexical prepositions contrast with prepositional elements that go by names such as grammatical prepositions*, grammaticalized prepositions9 or Case-prepositions™. These elements, according to Rauh (1993a,b,1995a,b), are not associated with a lexical argument structure, they have one obligatory structural nominal complement, i.e. there is no categorial variation, they do not have descriptive content and hence do not semantically head their projection and they do not allow for modification.11 All these elements contribute to the structure they occur in, as can be seen from the term "Case-preposition", is Case.12 (2) a. He is good at tennis. b. He is an expert on computers. (Rauh(1993a:20)) c. The dean appealed to everyone to attend the meeting. d. Unfortunately, water came out of the throttle. These two classes of prepositions can be considered as marking the extremes on a scale ranging from lexical to functional categories, implying that there are elements in between, i.e. elements that display mixed properties to different extents. One such class is constituted by prepositional elements that lack a referential argument, i.e. do not display spatial or temporal semantics. Rauh (1993a: 17,21) argues that, lacking a referential argument, projections of these elements are precluded from argument positions; their function is that of a modifier. In her (1993a: 17ff, 1995b:3 Iff) terminology, these prepositional forms are classified as grammatical prepositions; but this class must not be confused with the elements Williams (1989) refers to as grammatical prepositions: 13 (3) a. She came home with Bill. b. He payed $3Qfor the curtains. c. As usual, talks were on a high level

8

Cf. Williams (1989). Cf. Emonds (1985). 10 Cf. Rauh (1993a,b,1995a,b), Rooryck (1993). These prepositional elements - and only these, according to Rauh (1995b) - correspond to P in GrimshaVs (1991) extended nominal projection. 11 This set of prepositional elements, although not corresponding to Zwarts' (1995a) funtional prepositions, shares the property of not being subject to word-formation processes with functional prepositions. They can, however, be argued to play a role in inflectional morphology since they are Case-like and thus can be attributed to nominal inflectional paradigms. 12 Following Rauh (1993a,b,1995a,b), I assume that prepositions together with nouns and adjectives qualify as inherent Case-assigners (pace Chomsky (1981,1986a)) and that, contrary to what Chomsky and Lasnik (1993:558ff.) assume, inherent Case may be realized by prepositional forms other than o/(cf. 6.5). 13 in her (1993a) paper, Rauh takes the prepositional elements in (4) to constitute a second class of grammatical prepositional items that display mixed behavior. Rauh (1993a:18) classifies these element as fixed or lexicalized grammatical prepositions. These elements differ from grammatical prepositions in that, according to Rauh (1993a:17f), the constructions they occur in are lexically fixed and thus preclude internal variation.

9

192 (4) a. b. c. d.

He did it on the quiet and then left in a huny. In the light of the overwhelming evidence, he had no choice but to confess. He came in disguise. Look back m anger.

As this chapter will be concerned with those lexical prepositions that in one of their various functions serve to introduce subordinate clauses, the reader is referred to Rauh (1993a,b,1995a,b) for a detailed discussion of non-lexical prepositions.

7.1 Lexical Prepositions and their Projections In this section I will discuss the internal structure of lexical PPs and the functional structure dominating them.14 The focus will be on lexical prepositions that may function as subordinators or subordinating conjunctions. This subset is constituted by the temporal prepositions after, before, since and until. The temporal elements Wie«15 and while have to be added to this class. When and while, although differing from the other elements in that their structural complement is not categorially variable, have to be categorized as prepositions. The first part of this section concentrates on argument structure and Case-properties of lexical prepositions. Adopting the internal subject hypothesis, I will argue that the theta-identification mode of discharge can be, in fact must be, dispensed with as it turns out to be incompatible with assumptions about licensing. I will argue that theta-marking and theta-binding (in a revised version) are necessary and sufficient means for licensing maximal projections within the lexical projection of a head. The central issue in the second part of this section is the functional structure dominating PP. It will be argued that the number of functional projections above PP is determined by the lexical properties of the prepositional element heading the entire structure. It will be argued that lexical prepositions are dominated by at least two functional projections, each of which is motivated and licensed on the basis of the lexical properties of the prepositional nucleus.

7.1.1 Argument structure, Case-properties and the Structure of PP To set the background for my arguments, I want to briefly come back to the conception of argument structure which was outlined in 4.2.1.

14

15

Whereas there are numerous proposals concerning the functional structure dominating NP - DP, NumP, KP, QP, etc. (cf. Abney (1987), Olsen (1989a,b,1991a,b), Label (1992), Valois (1991), Ritter (1991), Felix (1990), Hale/Keyser (1991), Lamontagne/Travis (1986), Huppertz (1992)), there are fewer conceptions of functional projections dominating AP and PP. But see Corver (1990) and Abney (1987) for DegP as the functional projection topping AP, and Zwarts (1992), Rooryck (1993) and Koopman (1993) for functional projections dominating PP. When as a temporal preposition is not to be confused with the temporal adverb when (cf. 5.3).

193

The argument structure of lexical heads, in addition to having thematic arguments, i.e. one external and