Parameters and Functional Heads
OXFORD STUDIES IN COMPARATIVE SYNTAX Richard Kayne, General Editor
Principles and Pa...
110 downloads
1169 Views
17MB Size
Report
This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form
Parameters and Functional Heads
OXFORD STUDIES IN COMPARATIVE SYNTAX Richard Kayne, General Editor
Principles and Parameters of Syntactic Saturation Gert Webelhuth Verb Movement and Expletive Subject in the Germanic Languages Sten Vikner Parameters and Functional Heads: Essays in Comparative Syntax Edited by Adriana Belletti and Luigi Rizzi Discourse Configurational Languages Edited by Katalin E. Kiss Clause Structure and Language Change Edited by Adrian Battye and Ian Roberts Dialect Variation and Parameter Setting: A Study of Belfast English and Standard English Alison Henry Parameters of Slavic Morphosyntax Steven Franks The Polysynthesis Parameter Mark C, Baker
PARAMETERS AND FUNCTIONAL HEADS Essays in Comparative Syntax
Edited by
Adriana Belletti Luigi Rizzi
New York Oxford OXFORD UNIVERSITY PRESS 1996
Oxford University Press Oxford New York Athens Auckland Bangkok Bombay Calcutta Cape Town Dar es Salaam Delhi Florence Hong Kong Istanbul Karachi Kuala Lumpur Madras Madrid Melbourne Mexico City Nairobi Paris Singapore Taipei Tokyo Toronto and associated companies in Berlin Ibadan
Copyright © 1996 by Oxford University Press, Inc. Published by Oxford University Press. Inc., 198 Madison Avenue, New York, New York 10016 Oxford is a registered trademark of Oxford University Press All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording, or otherwise, without (he prior permission of Oxford University Press. Library of Congress Cataloging-in-Publication Data Parameters and functional heads : essays in comparative syntax / edited by Adriana Bellelti, Luigi Rizzi. p. cm. — (Oxford studies in comparative syntax) Papers prepared at a program Certificat de specialisation en theorie syntaxique et syntaxe comparative which was given during 1988-1990 at the University of Geneva. Includes bibliographical references. Contents: The verb always leaves IP in V2 clauses / Bonnie D. Schwartz, Sten Vikner—Residual verb second and the Wh-criterion / Luigi Rizzi—Complex inversion in French / Luigi Rizzi, Ian Roberts—Negative concord in West Flemish / Liliane Haegeman, Raffaella Zanutlini—On the relevance of tense for sentential negation / Raffaella Zanuttini—A cross-linguistic study of Romance and Arberesh causatives / Maria Teresa Guasti—Hebrew noun phrases: generalized noun raising / Tal Siloni—Three kinds of subject clitics in Basso Polesano and the theory of pro / Cecilia Poletto. ISBN 0-19-508793-3 ISBN 0-19-508794-1 (pbk.) 1. Grammar, Comparative and general—Syntax. 2. Principles and parameters (Linguistics). 3. Lexical-functional grammar. I. Belletti, Adriana. II. Rizzi, Luigi, 1952- . III. Scries. P291.P36 1996 415—dc20 93-37260
2 4 6 8 9 7 5 3 I Printed in the United States of America on acid free paper
Contents Contributors, vii Introduction, 3 Adriana Belletti and Luigi Rizzi 1. The Verb Always Leaves IP in V2 Clauses, 11 Bonnie D. Schwartz and Sten Vikner 2. Residual Verb Second and the Wh-Criterion, 63 Luigi Rizzi 3. Complex Inversion in French, 91 Luigi Rizzi and Ian Roberts 4. Negative Concord in West Flemish, 117 Liliane Haegeman and Raffaella Zanuttini 5. On the Relevance of Tense for Sentential Negation, 181 Raffaella Zanuttini 6. A Cross-Linguistic Study of Romance and Arberesh Causatives, 209 Maria Teresa Guasti 7. Hebrew Noun Phrases: Generalized Noun Raising, 239 Tal Siloni 8. Three Kinds of Subject Clitics in Basso Polesano and the Theory of pro, 269 Cecilia Poletto
This page intentionally left blank
Contributors Adriana Belletti Universita per stranieri, Perugia Maria Teresa Guasti DIPSCO, Fondazione San Raffaele, Milano Liliane Haegeman Universite de Geneve Cecilia Poletto Universita di Padova Luigi Rizzi Universite de Geneve Ian Roberts University of Wales, Bangor Bonnie D. Schwartz University of Durham Tal Siloni Tel-Aviv University Sten Vikner Universitat Stuttgart Raffaella Zanuttini Georgetown University
This page intentionally left blank
Parameters and Functional Heads
This page intentionally left blank
Introduction Adriana Belletti and Luigi Rizzi
The essays collected in this volume originate, directly or indirectly, from the Certificat de specialisation en theorie syntaxique et syntaxe comparative which was held at the University of Geneva in 1989-90. This program gave rise to a substantive body of research in comparative syntax which seemed to us original and coherent enough to justify a collective publication. Two leading ideas lie in the background of much current work in comparative syntax: the parametric approach to the cross-linguistic comparison and the focus on the formal properties of functional heads. The "Principles and Parameters" model claims that fundamental aspects of the cross-linguistic variation are amenable to the different fixation of some simple parameters of Universal Grammar (Chomsky 1979, 1981). Each parameter represents a choice point, the primitive site of differenciation between grammatical systems. The primitive bifurcation induced by a parameter interacts with the deductive structure of Universal Grammar, thus multiplying its effects and giving rise to various observable differences. Clusters of properties which co-vary across languages, often following complex implicational relations, are sometimes amenable to the operation of a single abstract parameter, entering into complex interactions with other principles and parameters of Universal Grammar. This research strategy has proven successful in different domains of synchronic and diachronic studies, particularly in the area of comparative Romance and Germanic syntax (see Kayne 1984, ch. 5; Rizzi 1982, ch. 4; Platzack 1987; Roberts 1993, just to mention a small sample of an impressive body of literature). The second leading idea concerns the nature of syntactic configurations. Structural representations are now seen as arising from the integration of lexical projections (NP, VP, AP, PP) into a configurational skeleton provided by functional heads (inflections, complementizers, determiners ... ) and their projections (IP, CP, DP ... ). Lexical heads provide the descriptive content and the basic argumental (thematic) structure; functional heads determine the configurational geometry and provide such grammatical specifications as tense, mood, definiteness, etc., thus contributing to the determination of the interpretation as well as of the form of linguistic expressions. Much attention is devoted in the recent literature to the study of the interplay between functional and lexical elements across languages. The issue is closely interconnected 3
4
PARAMETERS AND FUNCTIONAL HEADS
with the parametric approach in that it has been claimed that the central parameters of Universal Grammar involve properties of heads (Borer 1984; Chomsky 1991; Rizzi 1989), and possibly only of functional heads (Ouhalla 1991). The specific research trend which has more directly inspired our work is the study of head movement, as instantiated in Baker's (1988) analysis of incorporation, and in Pollock's (1989) comparative study of V-to-I movement, based on Emonds (1978). The major insight which characterizes this trend is the idea that certain cases of formation of morphologically complex words are amenable to a syntactic process of head-to-head movement; some of the morphosyntactic properties of complex words are therefore explainable in terms of familiar syntactic principles, such as the Empty Category Principle; moreover, the enriched syntactic representations that this approach requires provide enough structure to permit an illuminating analysis of traditionally neglected empirical domains, such as the adverbial (Belletti 1990) and adjectival (Abney 1987) positions in sentences and NPs, respectively. One of the most significant results of the head movement idea is the fact that it permits a principled analysis of word order properties at the clausal level, in particular of Verb Second (V2) phenomena. As different aspects of V2 are the main empirical focus of three of the following essays, some introductory remarks on the topic are in order here. The fundamental word order property of main clauses (and some embedded clauses) in the Germanic languages — with the major exception of Modern English — involves an initial constituent followed by the inflected verb, with the rest of the clause in third position: (1) [XP V+Infl [ . . . ]] This tripartite structure is provided by the projection of the complementizer, under X-bar Theory, as an ordinary instance of Specifier-Head-Complement configuration (see Chomsky 1986:1-2 for the original proposal of this approach). The peculiar word order is obtained via a double structure-preserving movement into the CP level: starting from the basic order instantiated by embedded clauses (as in 2), the inflected verb moves to C°, and any constituent (the subject, the object or the adverbial in 3) can move to the specifier of C, the third position being realized as the complement of C, the IP: (2) ... [
dass [Hans gestern ein Buch gekauft hat] ]
'... that Hans yesterday a book bought has' (3) a. [Hans hat [t gestern ein Buch gekauft t'] ] 'Hans has yesterday a book bought' b. [Ein Buch hat [Hans gestern t gekauft t'] ] 'A book has Hans yesterday bought' c. [Gestern hat [Hans t ein Buch gekauft t'] ] 'Yesterday has Hans a book bought' This approach preserves the idea of the classical analysis of V2 according to which this peculiar order is derived via a double movement into the initial position of the
INTRODUCTION
5
clause (Den Besten 1977/83; Thiersh 1978; and, for different refinements, the papers collected in Haider and Prinzhorn 1987). It improves over the classical analysis in that it explains, under X-bar Theory and structure preservation, the following properties: 1. Exactly two positions are involved (rather than one, or three, or four ... ) because the head C takes exactly one specifier, under X-bar Theory and Binary Branching (Kayne 1984); so, only two landing sites are available for structurepreserving movement (occasional instances of VI in Germanic may involve either a Spec filled by a null operator, as in yes-no questions, or a case of radical absence of the specifier, among other possibilities). 2. These two positions are an XP and a head, rather than two XPs, or two heads; this follows from the general X-bar schema, and in particular from the assumption that specifiers are maximal projections; 3. The fact that the two positions are in the order XP-X0, rather than the opposite order, or free order; this follows from the fact that specifiers precede heads in the languages in questions, and perhaps more generally. Schwartz and Vikner's article directly addresses the question of the categorial status of V2 clauses. The authors take issue with two recent proposals: the asymmetric analysis of V2 (Travis 1984, 1991; Zwart 1991; among others) according to which V2 is not a unified phenomenon, in that V2 clauses introduced by the (local) subject would not involve the CP layer; and the IP internal analysis (Diesing 1990; Santorini 1989, etc.), according to which V2 clauses typically involve movement to an inflectional head lower than C. Schwartz and Vikner provide detailed evidence for a unified analysis of V2, based on the interaction with cliticization, adverbial positions and other reordering processes, considered comparatively across Germanic. Their approach then supports the classical claim that all V2 clauses involve a category higher than IP, hence, under ordinary assumptions on the clausal structure, involve movement of the inflected verb into the complementizer layer. The current trend of work on head movement in V2 and other structural configurations raises a host of theoretical issues. Two particularly prominent (and interconnected) questions are the following: 1. What causes head movement? 2. What is the derived structure of head movement? The second question is addressed in Rizzi and Roberts's article in the context of an analysis of French Complex Inversion. Pursuing the parallelism with movement of a maximal projection, one is lead to expect that head movement may also involve adjunction or substitution. The host will always be another head position, under the generalized structure preservation constraint proposed by Chomsky (1986). Head movement qua adjunction appears to be involved in the processes of cliticization, as in the classical analysis of Kayne (1975): e.g., object clitics in Romance adjoin to a functional head of the inflectional system, presumably the highest agreement head in finite clauses:
6
PARAMETERS AND FUNCTIONAL HEADS
As for head movement qua substitution, two distinct cases should be distinguished: a. selected substitution, i.e., substitution into a slot subcategorized for by the host head, as in the case of verbal affixation, e.g., movement of the lexical verb to T to pick up the tense morpheme, or movement of V+T to Agr to pick up subject agreement in the formation of mang-er-a (will-eat) in Italian (cf. also Pollock's suggestion that -er- may be better analyzed as a mood affix, 1992:837):
b. straight substitution into a radically empty head. This case is assumed by Rizzi and Roberts to be operative in the root instances of I-to-C movement (as in structure 7, below), with the root character of the process derivable from the Projection Principle. Various refinements of this typology of derived structures for head movement are discussed in Guasti's article in this volume and references cited there (see in particular Roberts 1993; Guasti 1993). The question of the causal force triggering head movement is related in part to the question of the derived structure. In cases like 5, in which the moved head picks up an affix at each step, the obvious functional motivation of head movement is the satisfaction of morphological well-formedness, the combination of roots and affixes to form complete words. The causal force triggering V2 type phenomena is much less straightforward, as the inflected verb does not pick up any (overt) affix in C. Why does it have to move then? One aspect of this question is addressed in Rizzi's article. The empirical problem discussed there is Residual Verb Second, i.e., the construction-specific manifestations of V2 in languages, such as English and the modern Romance languages (with the
INTRODUCTION
7
exception of Raetho-Romansh), which do not manifest V2 as the general word order property of main declaratives. Instances of residual V2 are Subject-Aux Inversion in English, Subject Clitic Inversion in French, and other inversion processes typically involving the subject and the inflected verb in interrogatives. The functional motivation of residual V2 in main interrogatives is identified in the necessity of satisfying the Wh-Criterion, the principle determining the SS distribution and LF interpretation of wh-operators. The following formulation is adapted from May (1985): (6) a. A wh-operator must be in a Spec-head configuration with an X0 [+wh]. b. An X0 [+wh] must be in a Spec-head configuration with a wh-operator. This principle, in essence, requires wh-operators and heads of interrogative clauses (marked with the feature [+wh]) to be in an agreement (Spec-head) configuration on the appropriate level of representation. On the assumption that [+wh] can be licensed on some inflectional head, I-to-C movement is instrumental to create such an agreement configuration:
A generalization of this approach is now strongly invited by the checking theory of Chomsky (1993). Essentially along lines already foreseen by Kuroda (1986), the Wh-Criterion can now be viewed as a special case of feature checking in an agreement configuration, in the terms of Chomsky (1993). So, the first clause of the Criterion can be seen as the A' analogue of the extended clause of the Extended Projection Principle: clauses must have subjects because the Agr head must have its Phi features supported by a nominal in the relevant checking (Spec-head) configuration; similarly, h[+wh] clausal heads must be matched by appropriate wh-specifiers. The second clause of the Wh-Criterion can be seen as the A' analogue of the fact that a nominal must move to a position in which its case feature can be checked by an appropriate head in the A system; similarly, a wh-operator must have its defining [+wh] feature checked by an appropriate head. A second group of articles focusses on different cross-linguistic properties of negation, a topic linked in many ways to WH and V2. Negation is now viewed as a component of the functional structure of the clause, a functional head projecting
8
PARAMETERS AND FUNCTIONAL HEADS
its own X-bar structure. Moreover, questions and negative operators share important structural and interpretive properties. In particular, 1. They both determine Weak Island Effects, selectively blocking adjunct movement (irrelevantly, the following examples are well-formed with main clause construal of why): (8) a. *Why do you wonder whether Bill was fired t ? b. *Why didn't they say that Bill was fired t ? 2. They both trigger residual V2 (Subject-Aux Inversion) in English: (9) a. In what case would you do that? b. In no case would I do that Moreover, they both license polarity items: (10) a, b.
Did you see anyone? I didn't see any one
and they both involve a special operator morphology in the languages analyzed in Haik (1990) and references cited there. In all these respects, question and negative operators pattern differently from other kinds of operators, e.g., universal quantifiers. A partially parallel analysis seems to be in order. Haegeman and Zanuttini propose that negative operators are constrained by a Negative Criterion, the close analogue of the Wh-Criterion, requiring them to end up (at the latest at LF) in a Spec-head configuration with a negative head. Then, the obligatory application of residual V2 can be treated on a par with the WH case; as for the Weak Island Effects induced by negation, if a negative operator must be in an A' spec position at the latest at LF to fulfill the Negative Criterion, this case falls under the relevant case of Relativized Minimality (Rizzi 1990), again on a par with WH. The articles by Zanuttini and by Haegeman and Zanuttini deal with the structural position, selectional properties, cross-linguistic parametrization of the syntactic projection of the negative marker, the Negative Phrase. The possibility of Negative Concord (the co-occurrence of several negative markers which correspond to a unique negation in the interpretation, as in Italian: Nessuno ha mai detto niente 'Nobody has never said nothing' = 'Nobody ever said anything') is shown to be cross-linguistically related to the structural realization of negation (hence it is generally possible in Romance, generally absent in Germanic, but possible in West Flemish, a Germanic language which possesses a negation of the Romance type), and to be constrained by the Negative Criterion. Zanuttini provides ample cross-linguistic evidence for a systematic dependency between the tense specification and the overt realization of the clausal negation as a head. If a negative head bears a selectional relation to tense, then one expects that a tenseless construction such as the imperative will be unable to be negated in a language involving an overt negative head, like Italian (mangia il
INTRODUCTION
9
dolce! 'eat the cake' but not *non mangia il dolce! 'don't eat the cake'); in this case, a subsidiary infinitive form is used (Non mangiare il dolce!). Other aspects of the interaction between lexical and functional structures in clausal and nominal constructions are investigated in the other articles. Immediately related to Baker's approach to causative formation qua incorporation is Guasti's analysis of causatives in Romance and in some Albanian dialects spoken in Southern Italy. Evidence is provided in support of the assumption that syntactic incorporation of the head of the embedded clause into the causative verb is always involved in Romance causatives. Hence, the distinction between morphological (Chichewa) and analytical (Romance) causatives cannot be attributed to the level of application of incorporation (syntax vs. LF). Rather, the nature of the host is subjected to a simple parametrization which determines the various attested cases. Siloni's article develops Abney's approach to the structure of NP/DP. The paper stresses the analogy of nominal and clausal structures, both arising from the integration of a lexical layer into one or more functional layers, with word order determined in part by raising of the lexical head into the functional structure. A simple analysis is provided of construct and free state in Semitic. The key syntactic process, responsible for word order and the case properties of the two constructions, is movement of the noun into the determiner position, an instance of head-to-head movement in the nominal system. Poletto investigates the status of subject clitics in different Romance varieties, with particular reference to the Basso Polesano variety of the Veneto dialects; the standard hypothesis that subject clitics in the northern dialects are manifestations of Agr is considerably refined; a new, articulated typology of subject clitics is introduced, based on the structural position and argumental/non-argumental status; its consequences are investigated for the theory of clausal structure and the theory of pro; in particular, new evidence is provided in support of the dissociation between formal licensing and identification of null pronominals. In concluding these brief introductory remarks, we wish to express our gratitude to all the participants in the Certificat for contributing, in different functions, to the creation of a lively and enthusiastic research group: Maria Cristina FigueiredoSilva, Marc-Ariel Friedemann, Corinne Grange, Liliane Haegeman, Richard Kayne, Christopher Laenzlinger, Umesh Pattnaik, Cecilia Poletto, Genoveva Puskas, Ian Roberts, Manuela Schoenenberger, Tali Siloni, Alessandra Tomaselli, Sten Vikner, Eric Wehrli, and Raffaella Zanuttini.
References Abney, S. 1987. "The English Noun Phrase in its Sentential Aspect," PhD dissertation, MIT. Baker, M. 1988. Incorporation: A Theory of Grammatical Function Changing. Chicago: University of Chicago Press. Belletti, A. 1990. Generalized Verb Movement. Aspects of Verb Syntax. Turin: Rosenberg and Sellier. den Besten, H. 1977/83. "On the Interaction of Root Transformations and Lexical Deletive Rules," ms, University of Amsterdam. Published (1983) in W. Abraham (ed.), On the Formal Syntax of the Westgermania. Amsterdam: John Benjamins. 47-131.
10
PARAMETERS AND FUNCTIONAL HEADS
Borer, H. 1984. Parametric Syntax. Dordrecht: Foris. Chomsky, N. 1980. "Principi e parametri nella teoria sintattica," Rivista di grammatica generativa 4:3-75. . 1981. Lectures on Government and Binding. Dordrecht: Foris. . 1986. Barriers. Cambridge, Mass.: MIT Press. . 1991. "Some Notes on the Economy of Derivation and Representation," in R. Freidin (ed.), Principles and Parameters in Comparative Grammar. Cambridge, Mass.: MIT Press. 417-454. . 1993. "A Minimalist Program for Linguistic Theory," in K. Hale and S.J. Keyser (eds.), The View from Building 20. Cambridge, Mass.: MIT Press. 1-52. Diesing, M. 1990. "Verb Movement and the Subject Position in Yiddish," Natural Language and Linguistic Theory 8:41-79. Emonds, J. 1978. "The Verbal Complex V' V in French" Linguistic Inquiry 9:151-175. Freidin, R. (ed.) 1991. Principles and Parameters in Comparative Grammar. Cambridge, Mass.: MIT Press. . 1993. Causative and Perception Verbs: A Comparative Study. Turin: Rosenberg and Sellier. Haider, H. and M. Prinzhorn. 1986. Verb Second Phenomena in Germanic Languages. Dordrecht: Foris. Haik, I.1990. "Anaphoric, Pronominal, and Referential Infl," Natural language and Linguistic Theory 8:347-374. Kayne, R. 1975. French Syntax. Cambridge, Mass.: MIT Press. . 1984. Connectedness and Binary Branching. Dordrecht: Foris. Kuroda, Y. 1986. "Whether We Agree or Not," ms, University of California/San Diego. May, R. 1985. Logical Form. Cambridge, Mass.: MIT Press. Ouhalla, J. 1991. "Functional Categories and the Head Parameter," GLOW Newsletter 26:3839. [Abstract of paper presented at the 1991 GLOW Conference.] Platzack, C. 1987. "The Scandinavian Languages and the Null Subject Parameter," Natural Language and Linguistic Theory 5:377-401. Pollock, J.-Y. 1989. "Verb Movement, Universal Grammar, and the Structure of IP," Linguistic Inquiry 20:365-424. . 1992. Review of Belletti (1990), Language 68:836-840. Rizzi, L. 1982. Issues in Italian Syntax. Dordrecht: Foris. . 1989. "On the Format for Parameters," Behavioral and Brain Sciences 12:355-356.. . 1990. Relativized Minimality. Cambridge, Mass.: MIT Press. Roberts, I. 1993. Verbs and Diachronic Syntax. Dordrecht: Kluwer. Santorini, B. 1989. "The Generalization of the Verb Second Constraint in the History of Yiddish," PhD dissertation, University of Pennsylvania. Thiersh, C. 1978. "Topics in German Syntax," PhD dissertation, MIT. Travis, L. 1984. "Parameters and Effects of Word Order Variation," PhD dissertation, MIT. . 1991. "Parameters of Phrase Structure and Verb Second Phenomena," in R. Freidin (ed.), Principles and Parameters in Comparative Grammar. Cambridge, Mass.: MIT Press. 339-364. Zwart, J. W. 1991. "Clitics in Dutch: Evidence for the Position of Infl," Gronginen Arbeiten im Generative Linguistik 33:71-92.
1
The Verb Always Leaves IP in V2 Clauses Bonnie D. Schwartz and Sten Vikner
1
Introduction
The verb-second (V2) phenomenon, as it is found in the Germanic languages, has been the focus of much attention within recent syntactic research. In most of the literature on V2 (e.g., den Besten 1977, 1989; Thiersch 1978; Koopman 1984; Holmberg 1986; Platzack 1986a, 1986b, 1987; Taraldsen 1986a; Schwartz and Vikner 1989; Tomaselli 1990a, 1990b; Roberts 1993; and Vikner 1994c), it is assumed that the verb in all V2 clauses has moved to a head position outside IP, e.g., C°. In Schwartz and Vikner (1989) we claimed that all V2 clauses were CPs, and we referred to this analysis as the "traditional" analysis. In this paper1 we shall call it the "V2-outsideIP" analysis, and by using this term we want to convey that although in what follows we will adhere to the view that the verb moves to C°, any analysis in which the verb moves into an X0 which is the sister of IP may be compatible with what we say here.2 Various alternatives to this analysis have been explored in the literature, and here we will address two in particular: One alternative is that there is an asymmetry between subject-initial and non-subject-initial V2 clauses, the former being only IPs and the latter CPs, as suggested by Travis (1984, 1986, 1991) and Zwart (1990, 1991). Below we will refer to this analysis as the "V2 asymmetry" analysis. The other alternative analysis is that V2 takes place inside IP, as suggested by Diesing (1988, 1990), and also in slightly different forms by Rognvaldsson and Thrainsson (1990), by Reinholtz (1989) and by Santorini (1989), and accordingly we shall group these under the term the "V2-inside-IP" analysis.3 Below we will first discuss the V2 asymmetry account in section 2, then the V2inside-IP account in section 3, and finally some facts concerning V0-to-I0 movement in German and Dutch in section 4. 11
12
PARAMETERS AND FUNCTIONAL HEADS
2
The Asymmetry Analysis
This analysis, as found in Travis (1984, 1986, 1991) and in Zwart (1990, 1991), has it that subject-initial V2 clauses are smaller than non-subject-initial V2 clauses: the former are only IPs, whereas the latter are CPs. We will argue that the position of the finite verb is the same in all types of V2 clauses, irrespective of whether the preverbal XP is a subject.4 The asymmetry analysis is forced to assume that I0 in Dutch (Du.) and German (Ge.) is to the left of VP in order to account for the position of the verb in 1 as opposed to its clause-final position in embedded clauses, as in 2: (1) Ge. a. *Die Kinder den Film gesehen haben b. Die Kinder haben den Film gesehen The children (have) the film seen (have) (2) Ge. a. Ich weiB, daB die Kinder den Film gesehen haben b. *Ich weB, daB die Kinder haben den Film gesehen I know that the children (have) the film seen (have) Indeed, 2 points to another consequence of the asymmetry analysis, namely that V0to-I0 movement is not obligatory in German (and Dutch), at least not at S-structure; in fact, in this analysis, V0-to-I0 movement is impossible at S-structure in non-V2 clauses. We will come back to this issue of V0-to-I0 movement in section 4. Below we will discuss some issues which are relevant yet problematic for either the asymmetry account and/or the V2-outside-IP analysis.
2.1 Adjunction to V2 Clauses Positing adjunction to IP would seem to be the only way to account for the position of adverbials like German letzte Woche 'last week', or Swedish (Sw.) trots allt 'after all/nevertheless/despite everything', in the examples below. The analysis has the following steps: a. The subject is taken to be in IP-spec, as it occurs to the left of another adverbial (in 3: German tatsdchlich 'actually'; in 4: Swedish inte 'not'), which we take to be adjoined to VP. b. The adverbials left of the subject are therefore left of IP-spec and hence must be adjoined to IP. This is demonstrated in three different types of clause: in 3a and 4a in an embedded clause; in 3b and 4b in a main clause (yes/no) question; and in 3c and 4c in a main clause topicalization:5 (3) Ge. a. Ich weiB, [cp daB letzte Woche [lp Peter tatsachlich ein Buch gelesen hat]] I know that last week Peter actually a book read has b. [cp Hat letzte Woche [Ip Peter tatsachlich ein Buch gelesen] ] ? Has last week Peter actually a book read? c. [cp Dieses Buch hat letzte Woche [IP Peter tatsachlich gelesen] ] This book has last week Peter actually read
THE VERB ALWAYS LEAVES IP IN V2 CLAUSES
13
(4) Sw. a. Jag beklagar [CP att trots allt [IP Johan inte vill lasa de har bokerna] ] I regret that despite all Johan not will read these here books b. [CP Vill trots allt [IP Johan inte lasa de har bokerna] ] ? Will despite all Johan not read these here books? c. [CP De har bokerna vill trots allt [IP Johan inte lasa] ] These here books will despite all Johan not read If a subject-initial main clause is an IP (as it is according to the asymmetry analysis), then 5 and 6 ought to be grammatical, as they should be completely parallel to 3 and 4: The adverbial should be able to adjoin to IP. However, these examples are not grammatical: (5) Ge. *Letzte Woche [? Peter hat tatsachlich ein Buch gelesen] Last week Peter has actually a book read (6) Sw. Trots allt [? Johan vill inte lasa de har bokerna] Despite all Johan will not read these here books If a subject-initial main clause is a larger constituent than an IP (e.g., a CP), as it is according to the approach we want to defend here, 5 and 6 are not predicted to be grammatical; instead they should be completely parallel to 7 and 8: The adverbial cannot adjoin to the V2 clause (i.e., to the CP), giving the correct prediction. (7) Ge. *Letzte Woche [CP ein Buch hat [IP Peter tatsachlich gelesen] ] Last week a book has Peter actually read (8) Sw. *Trots allt [CP de har bokerna vill [IP Johan inte lasa] ] Despite all these here books will Johan not read Summing up: There is independent evidence that adjunction to IP is allowed and that adjunction to CP is not. The fact that adjunction to a subject-initial V2 clause is impossible is therefore a natural consequence of the V2-outside-IP approach but left unexplained within the asymmetry approach.
2.2
Sentence-Initial Weak Pronouns
2.2.1
Weak Object Pronouns Impossible Sentence-Initially in Dutch and German
In section 2.1, the asymmetry analysis was seen to make the wrong predictions, because there was no difference between the behavior of subject-initial main clauses and non-subject-initial main clauses with respect to adjunction. In this section we will discuss some facts where such an asymmetry does exist. The more well-known of these facts fall out naturally from the asymmetry analysis, as we shall see; other problems nevertheless are raised that only the V2-outside-IP analysis can handle in a unified manner. As Travis (1986:20, 1991:359) shows, the German unstressed personal pronoun (third person neuter singular) es 'it', may only occur sentence-initially if it corresponds to a subject (cf. 9), but not if it corresponds to an object (cf. 10):
14
PARAMETERS AND FUNCTIONAL HEADS
(9) Ge. a. Das Kind hat das Brot gegessen b. Es hat das Brot gegessen The child/it has the bread eaten (10) Ge. a. Das Brot hat das Kind gegessen b. *Es hat das Kind gegessen The bread/it has the child eaten The facts are parallel in Dutch, as shown by Zwart (1991:80, ex. 28, 29):
(11) Du. a. Ik zag hem b. 'k zag hem I saw him (12) Du. a. Hem zag ik b. *'_m zag ik Him saw I In the asymmetry account, this difference (9b and 1 1b vs. l0b and 12b) is linked to the hypothesis that sentence-initial subjects are in IP-spec, whereas sentence-initial objects are in CP-spec. This difference is analysed in two distinct ways. Travis (1986:20, 1991:359) suggests that only XPs carrying focal stress may move to CP-spec, and that es (and by extension all reduced pronouns) cannot bear focal stress. In this way, 9b and 11b are permitted, since subject es and 'k are in IP-spec, and l0b and 12b are ruled out. According to Zwart (1990:4, 1991:80, n. 13), the difference between unstressed sentence-initial subject pronouns and unstressed sentence-initial object pronouns may be accounted for in a similar but distinct way: Adapting the suggestion by Kayne (1991:647) that all Romance pronominal clitics left-adjoin to a functional head, Zwart suggests that all weak pronouns in Dutch (which he maintains are clitics) right-adjoin to a functional head. This means that a weak pronoun in CP-spec will have nothing to cliticize to (see 13a), as there is no functional head to its left, and such structures will therefore be ruled out (see l0b and 12b). If in contrast the sentence-initial weak pronoun is a subject, it will first be in IP-spec (as is the case with all sentence-initial elements if and only if they are subjects (see 13b); and then it may cliticize to the right of the empty C0, satisfying the above cliticization requirement (see also the discussion in section 4.2).6 The V2-outside-IP account, on the other hand, would assume all the sentence-initial elements above to move to CP-spec. This account thus does not have recourse to a structural difference to which the difference in grammaticality can be linked. There are, nonetheless, at least two attempts in the literature to reconcile the V2-outside-IP account with these data, on the basis of which we will then propose a third. Tomaselli (1990a:438, 1990b: 124-126) follows Travis' suggestion that only pronouns carrying stress may occur in CP-spec. Her solution to the difference in behavior between unstressed subject and unstressed object pronouns is that the subject ones may cliticize (at the level of phonetic form) to C0 but the object ones may not. However, since this cliticization takes place from CP-spec onto C°, and since both
THE VERB ALWAYS ALWAYS LEAVES IP IP IN IN V2 V2CLAUSES CLAUSES
1515 1515
types of sentence-initial unstressed pronouns move to CP-spec, the difference with respect to cliticization must fall out from another difference between subject and object unstressed pronouns. According to Tomaselli, this other difference is that only the subject agrees with C°, as she assumes that C° agrees with IP-spec, as shown by evidence from dialects of German and Dutch (e.g., Bavarian and West Flemish). An objection to this might be that cliticization in, for example, the Romance languages does not seem to require agreement between a clitic and the head to which it cliticizes. Holmberg (1986:123-127) suggests an analysis of a rather different kind, making an appeal to Binding Theory. He proposes that sentence-initial unstressed pronouns cannot be operators, and therefore their traces are not variables but rather anaphors, following a suggestion in Taraldsen (1986b). Anaphors must be bound in their governing category and the governing category for subjects is CP, whereas for nonsubjects the governing category is only IP (the latter is essentially the Specified Subject Constraint). Therefore a trace of a pronoun which has moved to CP-spec is bound in its governing category only if the trace itself is in subject position; and since anaphors must be bound, a pronoun in CP-spec must therefore have its trace in the subject position. This solution thus requires accepting the claim that a trace may have its antecedent in CP-spec but still not be a variable. An account that retains the insights of the two proposals above can be found if we adopt some of Luigi Rizzi's recent ideas. According to Rizzi (1991a, 1992:11), a position is an A-position if it is either assigned a thematic role or construed with agreement. Consequently IP-spec is always an A-position, but CP-spec can be an A-position only if it is coindexed with C°, i.e., if the subject has moved into CP-spec (cf. Tomaselli's condition above on cliticization in CP-spec and cf. Rizzi 1990a:5 Iff on C° agreement). If we now assume with Holmberg (1986:123-127) that the unstressed pronouns in German and Dutch cannot be operators, i.e., they cannot be moved into CP-spec by A-bar-movement, then the only way for them to reach CP-spec is through A-movement. However, in accordance with Rizzi's proposals about A-positions, if the moved element is not the subject, then CP-spec cannot be an A-position and hence A-movement is impossible. (Notice that the same result is guaranteed by the Relativized Minimality restrictions on movement; cf. Rizzi 1990a: if A-movement out of IP does not go via IP-spec, it will violate Relativized Minimality, as IP-spec will be an intervening A-position.) Thus the same effect is
16
PARAMETERS AND FUNCTIONAL HEADS
reached as in Holmberg (1986), on slightly different assumptions, provided we accept his idea that the unstressed pronouns under discussion cannot be operators. 2.2.2
Weak Object Pronouns Impossible Sentence-Initially also in Danish and Norwegian
Holmberg claims that this restriction, namely, unstressed object pronouns such as German es 'it' (and also e.g., Du. unstressed 'm 'him') not being able to occur in CP-spec, is rather limited in its application and does not apply in Scandinavian. In support of this he gives the following examples from Swedish (1986:123, ex. 130): (14) Sw. a. Det har Johan atit It has Johan eaten b. Henne kannerjag faktiskt inte Her know I actually not c. Den tar jag hand om It take I care of These examples, however, do not necessarily show that unstressed (object) pronouns may occur sentence-initially in main clauses, even if these sentences are acceptable with contrastive stress on e.g., the subject: It is possible that the pronouns here are not really unstressed forms. The situation displayed above may in fact be identical to the one concerning German er/ihn 'he'/'him', or sie 'she'/'her': There is no difference in form between the unstressed version and the normal version of the pronoun, and therefore the difference between 9b and l0b above is not reproduced: (15) Ge. a. Die Mutter hat den Sohn in die Schule gebracht b. Sie hat den Sohn in die Schule gebracht The mother/she has the son(acc) to school brought c. Die Tochter hat der Vater in die Schule gebracht d. Sie hat der Vater in die Schule gebracht The daughter/her has the father(nom) to school brought To the unstressed sie corresponds a stressed form sie, whereas the unstressed es is different from its stressed variant das. So the possibility remains that the restriction against unstressed pronouns occurring sentence-initially in main clauses is also valid in Scandinavian, but that this is simply not discernible in the Swedish examples above, as the stressed and unstressed forms are indistinguishable. In fact, there is evidence from Scandinavian dialects that the restrictions on unstressed pronouns discussed in the previous subsection are not limited to German and Dutch. One case comes from an Oslo dialect of Norwegian (No.) (as discussed by Christensen 1984) and another from a Copenhagen dialect of Danish (Da.): One Norwegian unstressed pronoun is a 'she', and a Danish one is 'd (phonetically [9]) 'it', i.e., the unstressed form of the neuter pronoun. 7 Both Norwegian a and Danish 'd may occur in the post-verbal subject position in a main clause (16a, c), but not sentence-initially (16b, d):8
THE VERB ALWAYS LEAVES IP IN V2 CLAUSES
17
(16) No. a. Bar a ikke bodd her? Has she not lived here? b. * A har ikke bodd her (Christensen 1984:1, ex. 1 a) She has not lived here Da. c. Maske vil '_d ikke koste mere end tusind kroner Maybe will it not cost more than thousand kroner d. *^d vil maske ikke koste mere end tusind kroner It will maybe not cost more than thousand kroner The question is whether 16b, d are evidence that al'd are generally impossible as the initial element in a V2 clause. Following Christensen (1984:6), we will argue that there is no such general constraint, and that the reason al'd are impossible in 16b, d is that there is then nothing to the left of subject al'd to which they may cliticize. This is supported by the fact that if we take an embedded V2 clause, then al'd are both acceptable as the clause-initial element. This is shown by the embedded V2 clauses in 17a and 18a, which are just as grammatical as the embedded non-V2 clause in 17b and 18b:9 (17) No. a. Vi vet at a har ikke bodd her We know that she has not lived here b. Vi vet at a ikke har bodd her We know that she not has lived here (18) Da.
(Christensen 1984:28, ex. iv) (Christensen 1984:1, ex. 3a)
Marie sagde ogsa . . . Marie said also ... a. ... at ^d ville sikkert ikke koste mere end tusind kroner . . . that it would probably not cost more than thousand kroner b . . . . at \i sikkert ikke ville koste mere end tusind kroner ... that it probably not would cost more than thousand kroner
However, now it might look as if the restriction on al'd is only a phonetic one, i.e., that al'd must occur to the immediate right of something phonetically overt to which it can cliticize. That this constraint is both too strong and too weak is shown by the following: (19) No. a. *Vi vet at ikke a har bodd her (Christensen 1984:1, ex. 3b) We know that not she has lived here b. Vi vet hva a har gjort (Christensen 1984:27, ex. ii) We know what she has done Da. c. Marie vidste ikke hvorfor \J var sa billigt Marie knew not why it was so cheap 19a shows that it does not suffice to require that a simply appear to the right of another overt element, as a is not allowed here (there is no Danish version of 19a because sentential negation generally cannot occur left of the subject in Danish). 19b, c show that al'd are possible even when immediately right-adjacent to an empty element (we
18
PARAMETERS AND FUNCTIONAL HEADS
are assuming that in embedded questions like 19b, c, C° is empty). We therefore follow the analysis of Christensen (1984): Subject al'd must cliticize to the right ofaC 0 . 1 0 Summing up the discussion of subject al'd in Norwegian and Danish, we have seen that it is possible for a subject weak pronoun to occur initially in a V2 clause, provided that it occur to the immediate right of a C°. We now turn to al'd as objects:11 (20) No. a. Jon hadde ikke sett a f0r Jon had not seen her before b. *A hadde ikke Jon sett f0r Her had not Jon seen before Da. c. Marie ville ikke give tusind kroner for ^d Marie would not give thousand kroner for it d. *^d ville Marie ikke give tusind kroner for It would Marie not give thousand kroner for 20a, c show that object al'd are possible in their base position inside the main clause, whereas 20b, d show that they are not possible initially in a main clause. As with the weak subject pronouns, the question is whether object al'd are generally impossible as the initial element in a V2 clause or whether they are only impossible here because there is no C° to their left. Consider now the following examples: (21) No. a. *Jon sa dessuten at a hadde han ikke sett f0r Jon said moreover that her had he not seen before Da. b. *Marie sagde ogsa at ^d ville hun ikke give tusind kroner for Marie said also that it would she not give thousand kroner for As shown in the ungrammatical 21, object al'd (parallel to object es in German) are also impossible initially in an embedded V2 clause, even though here there is a C° to their left (filled by at). This is different from the subject al'd, which were allowed initially in an embedded V2 clause (cf. the grammatical 17b and 18b). Thus, modulo the restriction that a weak pronoun occur to the immediate right of a C° when it is the initial element of a V2 clause, the same subject-object asymmetry exists in Norwegian and Danish as exists in German and Dutch (contrary to the claims of Holmberg 1986:123, 127): Weak subject pronouns can but weak object pronouns cannot occur as the initial element in a V2 clause. Let us now consider the relevance of these Scandinavian data to the opposing analyses (discussed in the previous subsection) of the asymmetry between subject and object pronouns originally noticed for only German and Dutch. While the data from Norwegian and Danish fall out straightforwardly under the various versions of the V2-outside-IP account, they in fact undermine Zwart's account. What is crucial to our argument is the fact that Norwegian and Danish embedded V2 clauses must always follow a complementizer (as opposed to embedded V2 in German, for example, where the complementizer is impossible), as can be seen in all the Norwegian and Danish examples of embedded V2 above.
THE VERB ALWAYS LEAVES IP IN V2 CLAUSES
19
Recall that according to Zwart, the difference between unstressed subject and object pronouns in sentence-initial position is due to the idea that clitics must always right-adjoin to a functional head: Unstressed subject pronouns, which are taken to move to IP-spec, may appear initially because they may cliticize from IP-spec to C° (cf. 13b); by contrast, unstressed object pronouns, which would have to move to (or through) CP-spec when sentence-initial, are not possible initially because there is no functional head to the left of CP-spec to which they could cliticize (cf. 13a). However, this line of argumentation cannot be valid in view of the facts concerning unstressed subject and object pronouns as the initial element in embedded clauses in Norwegian and Danish discussed above. Since embedded V2 in Norwegian and Danish must always take place under an overt complementizer, then Zwart would necessarily predict there to be no asymmetry between unstressed subject and object pronouns in embedded clauses: There will always be a functional head to the left of the unstressed object pronoun (the C° containing at) to which it should be able to cliticize.12 As for the V2-outside-IP accounts (Tomaselli 1990a, 1990b; Holmberg 1986; and our combination of these two based on Rizzi 1991 a, 1992), they will apply to not only the Dutch and German facts but also the data from Danish and Norwegian: In these accounts, the unstressed object pronoun (as opposed to the unstressed subject one) is impossible in CP-spec because it does not agree with C° and because it would have to move across the subject in IP-spec on its way to CP-spec. Thus a single analysis covers all the data, in Norwegian and Danish as well as Dutch and German, and in embedded as well as main clauses. This section has thus shown not only that the impossibility of unstressed object pronouns sentence-initially is more widespread than previously thought, but also, more importantly, that the Norwegian and Danish data can crucially decide between the asymmetry account and the V2-outside-IP account. Whereas the asymmetry account makes the wrong prediction concerning the occurrence of unstressed object pronouns sentence-initially in Norwegian and Danish, the V2-outside-IP account treats the data in a unified manner in all four languages and thus makes the correct predictions. 2.2.3
Weak Expletive Pronouns Sentence-Initially in German, Yiddish, and Icelandic
So far we have discussed unstressed pronouns which were arguments. Let us now turn to unstressed expletive pronouns (in German, Yiddish (Yi.), and Icelandic (Ic.)), as they show a different kind of distribution which will again be seen to pose more problems to the asymmetry account (as noted by Tomaselli 1990b:140) than to the V2-outside-IP account. Compared to the unstressed object pronouns, the (unstressed) expletive pronouns have an almost mirror-image distribution. Whereas the former cannot occur sentenceinitially, the unstressed expletives of German, Yiddish and Icelandic seem to occur only in sentence-initial position (in CP-spec) (these facts have been discussed in the literature as early as Breckenridge 1975 and Thrainsson 1979):13
20
PARAMETERS AND FUNCTIONAL HEADS
(22) Ge. a. Es ist ein Junge gekommen b. *pro ist ein Junge gekommen There is a boy come c. *Gestern ist es ein Junge gekommen d. Gestern ist pro ein Junge gekommen Yesterday is there a boy come (23) Yi. a. Es iz gekumen a yingl b. *pro iz gekumen a yingl There is come a boy c. *Nekhtn iz es gekumen a yingl d. Nekhtn iz pro gekumen a yingl Yesterday is there come a boy (24) Ic, a. pa8 hefur komiS strakur b. *pro hefur komiS strakur There has come (a) boy c. *I g<er hefur baS korm'5 strakur d. I ga;r hefur pro komiS strakur Yesterday has there come (a) boy The asymmetry account has no difference in positions to appeal to in order to explain the grammaticality difference above between 22a, 23a, and 24a, on one hand, and 22c, 23c, and 24c, on the other: In both cases the overt expletive is in IP-spec. Whereas Travis (1986,1991) and Zwart (1990,1991) do not address this problem, Travis (1984:169) suggests that only if 1° is phonetically empty (i.e., in non-subjectinitial clauses) may VP properly govern IP-spec, and thus allow it to be empty. We find this proposal difficult to accept for three reasons: First, although this might allow IP-spec to be empty, it is difficult to see how it would force it to be empty. Second, VP as a proper governor seems rather controversial (as noted by Tomaselli 1990b: 141), given that proper governors are normally heads. Third, why should this proper government depend on phonetic adjacency (i.e., be limited to cases where 1° is phonetically empty)? As for the V2-outside-IP account, here there is a difference in position: The expletive in the grammatical 22a, 23a, and 24a is in CP-spec, whereas in the ungrammatical 22c, 23c, and 24c the expletive is in IP-spec. The next question is why expletives are not allowed in IP-spec. Within the V2-outside-IP account, two different approaches may be taken: The expletive may be generated in CP-spec, as argued by Tomaselli (1990b: 140) for German and by Sigur5sson (1989:11,165,284) and references therein for Icelandic. It then follows that it could never appear in IP-spec. An alternative is that the expletive is generated in IP-spec and then obligatorily moved to CP-spec, as suggested in Cardinaletti (1990a, 1990b). We agree with the latter approach, as it is the only one compatible with our view that expletives need to be assigned case (and that nominative case is assigned only under government in V2 languages). The requirement that expletives be case assigned receives further support from the following difference in languages like Danish and Swedish (as well as, e.g., Norwegian and Dutch): 14
THE VERB ALWAYS LEAVES IP IN V2 CLAUSES
21
(25) Da. a. ... at deq faktisk ser ud til tj ikke at blive danset til festen . . . that there actually seem out to not to become danced at party-the (= . . . that there actually seems not to be any dancing at the party) b. *... at det faktisk ser ud til der ikke at blive danset til festen . . . that it actually seem out to there not to become danced at party-the (26) Sw. a. ... att detj faktiskt verkar tj inte dansas pa festen b. *... att detj faktiskt verkar det inte dansas pa festen ... that it actually seems (it) not danced-be at party-the These examples can be accounted for only if the expletive must be assigned case: In 25a and 26a the expletive is raised out of its own clause and receives case as subject of the higher embedded clause. In 25b and 26b, on the other hand, no raising takes place, and the expletive in the most deeply embedded clause does not receive case. If the expletive did not require case, 25b and 26b should be well-formed. 2.2.4
Conclusion
Summing up: In section 2.2.1, we saw that the distribution of data concerning the unstressed argument pronoun es in German (and unstressed argument pronouns in Du.) may be captured in the asymmetry analysis by linking it to the difference in position of the first element in a subject-initial main clause as opposed to the position of the first element in a non-subject-initial main clause. The V2-outside-IP analysis has no such difference to appeal to, but there may nevertheless be ways of accounting for the difference, if we assume either that these object pronouns cannot bear stress or that they cannot be operators. Section 2.2.2, on the other hand, showed that this difference in position could not be appealed to in order to account for the parallel data in Norwegian and Danish, given that in embedded sentences, both sentence-initial subjects and sentence-initial objects (topicalized objects) occur to the immediate right of a C°. Only the V2-outside-IP analysis was able to provide an account for this set of data. Finally, in section 2.2.3, we argued that the asymmetry analysis makes the wrong predictions concerning the expletive pronoun es in German (and the expletive pronouns Yi. es and Ic. pad). In contrast, these facts are more easily accounted for in the V2-outside-IP analysis, even if not all problems are solved, as shown in the discussion of whether this expletive is generated in CP-spec or not.
2.3
Extractions from Embedded V2 Structures
Below we consider an argument that attempts to support the V2-outside-IP approach; our focus will be on extractions from embedded V2 structures in German, and we show that Travis' version of the asymmetry approach cannot capture the distribution of the data. This argument is based on one made by Holmberg (1986:110) using Swedish, but (as we discussed in detail in Schwartz and Vikner 1989:35-38) we think that Travis' version of the asymmetry approach is in fact not susceptible to
22
PARAMETERS AND FUNCTIONAL HEADS
those criticisms. We do think, however, that when the argumentation is carried over to German data, Travis' version of the asymmetry approach finds itself in an insoluble dilemma. Following this discussion we will then address the difference between Travis' and Zwart's versions of the asymmetry approach and the reason Zwart's version is not subject to the objections made here. After that, however, we will consider some additional data that are not amenable to his analysis. 2.3.1
Travis'Version of Asymmetry
German has embedded clauses with V2 under matrix verbs like say and believe, but they are only possible without daft 'that'. 27 shows that with daft, the finite verb must remain at the end of the embedded clause, whereas 28 and 29 show that when there is no complementizer, the finite verb has to move, resulting in a V2 structure: (27) Ge. a. *Sie glaubte daB das Kind hatte dieses Brot gegessen b. Sie glaubte daB das Kind dieses Brot gegessen hatte She thought that the child (had) this bread eaten (had) (28) Ge. a. Sie glaubte das Kind hatte dieses Brot gegessen b. *Sie glaubte das Kind dieses Brot gegessen hatte She thought the child (had) this bread eaten (had) (29) Ge. a. Sie glaubte dieses Brot hatte das Kind gegessen b. *Sie glaubte dieses Brot das Kind gegessen hatte She thought this bread (had) the child eaten (had) (= She thought that the child had eaten this bread) Now consider what happens when extraction takes place out of the complementizerless embedded clause. The results are only grammatical if the finite verb precedes all of the rest of the clause:15 (30) Ge.
Womit glaubte sie,... What-with thought she . . . a. . . . hatte das Kind dieses Brot gegessen b. *... das Kind hatte dieses Brot gegessen . . . (the child) had (the child) this bread eaten
In the V2-outside-IP approach, there is a straightforward account for these facts, parallel to the analysis of main clauses, i.e., all V2 structures receive the same analysis: The finite verb is in C°. This entails that das Kind in 30a is in IP-spec but in CP-spec in 30b:
THE VERB ALWAYS LEAVES IP IN V2 CLAUSES
23
In 3la the extraction does not violate any constraints; the empty embedded CPspec contains an intermediate trace of the extracted adjunct. In 31b, on the other hand, since the embedded CP-spec is filled, there is no room for an intermediate trace there, and the extraction is ruled out.16 This analysis of 30b would, on first view, seem not to be open to Travis' version of the asymmetry approach, as there it is claimed that the subject of subject-initial V2 structures is in IP-spec. This leaves the Travis version of asymmetry with two possible analyses of 30b: either a. the subject is in IP-spec; and there is no C°-projection at all, or b. the subject is in IP-spec; and CP-spec and C° are present but empty. Let us start with the former: In a structure like 28a, one might be tempted to propose that glauben 'believe/think', takes only an IP as a complement, on parity with the asymmetry analysis of subject-initial main clauses. However, in 29a glauben must be followed by a CP, since the object dieses Brot 'this bread', precedes the finite verb and the subject. Thus, this analysis runs into the conceptual problem of stating that glauben subcategorises for an IP (only when the clause is subject-initial) and for a CP (in all other cases). More significant are the empirical problems encountered in such a proposal. If the subject of 30b is in IP-spec and there is no C°-projection, the sentence, which is ungrammatical, would be predicted to be good:
Movement of the adjunct does not violate any constraints on extraction here.17 Hence, the first alternative to the V2-outside-IP approach must be rejected, since the sentence must be ruled out. Let us now turn to the second alternative, where das Kind in 30b is in IP-spec, but CP-spec and C° are present but empty:
The question is whether or not the embedded CP is the complement of the matrix verb in 33, glauben. Let us first discuss an analysis which assumes that the CP is the complement of glauben. This would again incorrectly predict 30b to be grammatical, because the empty C° would be properly governed (by glauben, hence no ECP violation
24
PARAMETERS AND FUNCTIONAL HEADS
would result within Travis' conceptualisation of the ECP)18 and CP-spec would also be available for an intermediate trace of womit 'what-with'. Thus, within Travis' version of the asymmetry approach, it cannot be assumed that the embedded CP is the complement of the matrix verb. In fact, the grammaticality of 30a clearly shows that glauben does not properly govern the embedded C°: The finite verb unquestionably is in the embedded C°, since it precedes the subject, and this would not be possible if the C° was properly governed (cf. n. 18, above). If the embedded CP is not the complement of the matrix verb in 33, then this verb cannot identify the empty C° and the empty C° would thus violate (Travis' version of) the ECP if it were to remain empty. However, although C° is empty at D-structure, there is nothing that would force the embedded finite verb (hatte) to not move into C°, and the subject (das Kind) to not move into CP-spec, given that the subject precedes the finite verb. It should be remembered that it is this kind of head movement that is relied on to prevent violations of (Travis' version of) the ECP. In this instance, however, these steps of movement would amount to the analysis of the V2-outside-IP approach, which was given as 3 Ib. In sum, we find that working within Travis' version of the asymmetry approach, there is no way to rule out 30b, except if it is analysed as 31b: The finite verb (hatte) is in C°; the subject (das Kind) is in CP-spec, and since CP-spec is filled, there is no room for an intermediate trace of womit. Thus it seems that there is no analysis which can simultaneously maintain the subject in IP-spec and rule out 30b under this version of asymmetry. The general conclusion of the discussion above must be that embedded V2 clauses are larger than IPs, irrespective of whether they are subject-initial or not. This would seem to imply that Travis' version of an asymmetry approach has to either be given up completely or be maintained in a much weaker form: While conceding that all embedded V2 clauses are CPs, proponents of Travis' version of the asymmetry approach could still maintain that V2 subject-initial main clauses are IPs, necessitating two different explanations for what seems to be only one phenomenon, viz. V2. Though this is theoretically possible, it is less desirable given the existence of an analysis which provides a unified explanation of this phenomenon. In the remainder of this section, we will discuss another example of a phenomenon which cannot receive a unified explanation under Travis' version of the asymmetry approach. This will thus be a further argument for why even this much weaker variant of Travis' version of asymmetry should be rejected. The relevant data concern the behavior of es, as discussed in section 2.2.1. There it was shown that (Travis' version of) the asymmetry approach could provide an elegant analysis of the distribution of sentence-initial es, by assuming that the specifier position to which sentenceinitial subject es moves is IP-spec (and therefore grammatical), whereas the specifier position to which sentence-initial object es moves is CP-spec (and therefore ruled out, as CP-spec only can be occupied by elements that may bear stress). We start by considering the following contrast:19 (34) Ge. a. Womit glaubst du hat es dieses Brot gegessen b. *Womit glaubst du es hat dieses Brot gegessen What-with think you (it) has (it) this bread eaten
THE VERB ALWAYS LEAVES IP IN V2 CLAUSES
25
At first glance, it might appear that Travis' version of asymmetry could account for this difference in grammaticality in a parallel fashion to its account of 9b and lOb, repeated below as 35. In 35a es is in IP-spec (allowed), but in 35b es is in CP-spec (disallowed):
Carrying this over to 34, in 34a hat would be in C° and the intermediate trace of womit would be in CP-spec; the ungrammaticality of 34b could then be due not only to there not being any intermediate trace (there is no room for it in CP-spec) but also to the unstressed es occurring in CP-spec, which is explicitly excluded under Travis' version of the asymmetry approach (cf. 35b and section 2.2.1). As also argued in the discussion of 30-33 above, however, this presupposes that the embedded clause is a CP, whether it is subject-initial, as in 36a (= 28a), or objectinitial, as in 36b (= 29a) (note that we have put 1° to the left of VP here, as the asymmetry approach would): (36) Ge. a. Sie glaubte [cp das Kind hatte [Ip 11 dieses Brot gegessen] ] She thought the child had this bread eaten b. Sie glaubte [cp dieses Brot hatte [Ip das Kind 11 gegessen] ] She thought this bread had the child eaten (= she thought that the child had eaten this bread) This in turn leaves Travis' version of asymmetry without an account for the difference in grammaticality between the es versions of 36a and 36b, viz. 37a and 37b, as in both cases es must be in CP-spec: (37) Ge. a. Sie glaubte es hatte dieses Brot gegessen She thought it (the child) had this bread eaten b. *Sie glaubte es hatte das Kind gegessen She thought it (this bread) had the child eaten In fact, Travis' version of asymmetry would incorrectly predict 37a to be ungrammatical, precisely because es must be in CP-spec. Let us briefly run through, once more, why this must be so: a. glauben must be followed by a CP (cf. the discussion of 32) b. CP cannot be the complement of glauben (cf. the discussion of 33) c. C° is not properly governed, so the verb must be in C° (cf. the discussion of 33)
26
PARAMETERS AND FUNCTIONAL HEADS
d. es cannot be in IP-spec, as it precedes the verb in C° e. es is in CP-spec, where it must not occur (cf. section 2.2.1) f. the sentence 37a is hence ruled out incorrectly In other words, Travis' account of 37 which refers to a difference in position between sentence-initial subject es (in IP-spec) and sentence-initial object es (in CP-spec) is not tenable for embedded clauses (it would incorrectly predict 37a to be ungrammatical). So it is precisely the claim that unstressed es cannot be in CP-spec (which was the prime motivation for the idea that subject-initial (main) clauses are IPs) that turns out not to be able to account for the completely similar facts in embedded clauses. Thus, this is another example of a phenomenon for which Travis' version of asymmetry now has to have two different explanations, one for main clauses and another for embedded clauses (whatever the latter might be). Summing up: We have shown that two important assumptions of Travis' version of the asymmetry approach, (a) that subject-initial V2 clauses are IPs, and (b) that unstressed es cannot occur in CP-spec, cannot possibly hold for embedded clauses, as embedded subject-initial V2 clauses are CPs (cf. 33) and unstressed es may occur in an embedded CP-spec (cf. 37a). This leaves three possibilities: a. the assumptions of Travis' version of asymmetry are maintained, though only for main clauses. The costs for this are that facts which are completely parallel in main and embedded clauses thus do not receive unified explanations. b. the assumptions of Travis' version of asymmetry are rejected and the relevant phenomena receive parallel analyses: Both main and embedded V2 clauses are CPs and the restrictions for unstressed es in CP-spec are the same in main and embedded clauses. c. a rather different version of asymmetry is adopted Option (c) is what will be explored in the following section. 2.3.2
Zwart's Version of Asymmetry
As for Zwart's (1990,1991) version of the asymmetry approach, it differs sufficiently from Travis' version so as to be able to rule out the problematic 30b, repeated here as 38b: (38) Ge.
Womit glaubte sie,... What-with thought she ... a. . . . hatte das Kind dieses Brot gegessen b. * . . . das Kind hatte dieses Brot gegessen . . . (the child) had (the child) this bread eaten
38b is not possible in Zwart's account because of the following filter (Zwart 1991:74, ex. 14):20
THE VERB ALWAYS LEAVES IP IN V2 CLAUSES
27
(39) *[YP X°], where YP is an operator, and X° is empty. With respect to 38b, the consequences of 39 are: a. that the embedded clause is a CP, with the subject in IP-spec, and therefore b. that the example is ruled out because although CP-spec is filled (by a trace of womit), C° is not filled. 39, however, does not apply to (unstressed) subject-initial clauses: The subject has no reason to move to CP-spec, since unstressed subjects cannot be considered operators, according to Zwart (1991:75). The analysis of subject-initial sentences follows from 40 instead (Zwart 1991:85; recall that for Zwart, 1° precedes VP): (40) The finite features, which are located in 1°, must be licensed by either (a) or (b): a. V°-to-I° movement b. lexicalized C° Thus, in a subject-initial main clause, the verb moves to 1° in order to license the finite features there (cf. 40a). 40b can be used to explain the contrast in grammaticality in 27 and 28, repeated here as 41 and 42: (41) Ge. a. *Sie glaubte daB das Kind hatte dieses Brot gegessen b. Sie glaubte daB das Kind dieses Brot gegessen hatte She thought that the child (had) this bread eaten (had) (42) Ge. a. Sie glaubte das Kind hatte dieses Brot gegessen b. *Sie glaubte das Kind dieses Brot gegessen hatte She thought the child (had) this bread eaten (had) As C° is lexicalized in 41, the finite features in 1° are licensed; moreover, in the spirit of Chomsky (1991), Zwart rules out V°-to-I° movement by principles of economy, and hence 41 a is ungrammatical. In contrast, C° is not lexicalized in 42 and therefore V°-to-I° movement is required in order to license the finite features of 1°. One potential counterexample to 39 may be furnished by exclamatives (adapted from Naf 1987:143): (43) Ge. a. Wie riesig sind die Pflanzen geworden! b. *Wie riesig die Pflanzen sind geworden! c. Wie riesig die Pflanzen geworden sind! How enormous (are) the plants (are) become (are)! The grammaticality of 43a would seem to indicate that C° is not lexically filled by an empty element in C°.21 If we apply Zwart's analysis, 43a should follow from 39, and so then the ungrammaticality of 43b would be due to violating 39. Nevertheless, this leaves 43c unexplained, for it too should violate 39: As CP-spec is filled, C° should be
28
PARAMETERS AND FUNCTIONAL HEADS
filled. Unlike the solution given for 4Ib where a lexicalized C° was said to license the features of 1°, here no such analysis can be appealed to — given the grammaticality of 43a with verb movement to C°. In sum, it is not at all clear how both 43a and 43c are grammatical under an analysis like Zwart's: Either the features of 1° are not licensed in 43c, or 43a violates principles of economy — yet both sentences are equally acceptable. Note, finally, that within a V2-outside-IP approach, the contrast in 43 is unproblematic: The verb in 43a is in C°, the verb in 43c is in 1°, whereas 43b is ungrammatical because there is no X°-position at all between the subject and the VP.22
2.4
Some Adjunction Consequences from Zwart (1990)
Zwart, in his attempt to account for the primary differences between V2 languages like German and Dutch, on the one hand, as opposed to non-V2 languages like English, on the other, ends up positing the following (Zwart 1990:11, ex. 52) as his fundamental proposal:23 (44) a. A strong functional head licenses the features of the functional head it governs (as well as its own features) if lexically filled b. A strong functional head attracts adjunction to its projection Apart from these two statements, Zwart is claiming that virtually the only other thing that needs to be specified to capture the differences between V2 and nonV2 languages is which functional head counts as "strong" (following Koster 1986): In German and Dutch it is C°, but in English 1° is the "strong" functional head. The somewhat non-intuitive claim being made here notwithstanding (i.e., that 1° is "stronger" in English than, say, in German), there are a few empirical consequences one can explore that follow in particular from 44b, above. So, first we will consider a non-V2 language, English, and next we will briefly consider a V2 language, Dutch 2.4.1
Topicalization in English: Adjunction to IP?
Given that in English it is 1° that is strong, then Zwart (1990:7, ex. 37), following Lasnik and Saito (1992:85), gives the following as an example supporting the idea that IP attracts adjunction: (45) En.
A man to whom freedom [1P we would never g r a n t . . . ] . . .
In this case, it appears that since to whom is in CP'-spec, freedom must be adjoining to IP (as originally suggested in Baltin 1982:18, ex. 71). However, the contrasts within the pairs of sentences like the following seem to show that adjunction to IP is not the only possible analysis for 45: (46) En.
This is the man . . . a. . . . to whom only in America could liberty be granted b. * . . . to whom only in America liberty could be granted
THE VERB ALWAYS LEAVES IP IN V2 CLAUSES
(47) En.
29
He is the kind of guy . . . a. . . . who under no circumstances should you trust b. *... who under no circumstances you should trust
(48) En. a. He's the jerk who never in my life will I want to see again b. *He's the jerk who never in my life I will want to see again c. *He's the jerk who never in my life I want to see again In 46a, 47a, and 48a, although the embedded w/z-phrase is in CP-spec, the negative constituent (only in America/under no circumstances/never in my life) that follows does not appear to be adjoined to IP, for otherwise one would not expect the auxiliary verb to precede the subject (libertylyoull); moreover, that 46b, 47b, and 48b, c are ungrammatical indicates that the negative constituent cannot simply be adjoined to IP. Returning now to 45, one can in parallel fashion propose that freedom is not adjoined to IP but rather is in some Spec position between CP and IP, but because V2 effects are rather limited in English,24 the auxiliary does not move. Miiller and Sternefeld (1993:481-482) give another argument that embedded topicalizations are different from adjunctions to IP, contrary to Lasnik and Saito (1992). Their point is that extraction is impossible out of embedded topicalizations but possible out of sentences in which scrambling to IP has taken place. The limited effects of V2 in English and in other residual V2 languages seem to suggest, therefore, that the difference between V2 and non-V2 languages cannot simply be linked to the "strength" of functional heads, for one would not want to say that in constructions like 46a, 47a, and 48a above — not to mention normal w/z-questions — the nature of the functional categories changes, i.e., that in these constructions C° becomes strong. Moreover, we have seen that the general claim in 44b, namely, that the specification of a functional head as 'strong' leads to its maximal projection being the landing site for adjunction, is empirically questionable. In the next section, we will briefly show that 44b is necessarily misguided. 2.4.2
Adjunction and the Strong Functional Head in V2 Languages
The flipside of the coin for 44b as it applies to V2 languages also is empirically inaccurate. In principle, the idea of having a single property (of functional heads) be able to account for the bifurcation between V2 and non-V2 languages is certainly desirable; however, it appears that what is in fact captured is only (not even) half of the story: Namely, one half of the verb movement story (in V2 languages) and one half of the adjunction story (in non-V2 languages). Above we saw that adjunction to IP is not the only possibility for fronting in English, where 1° is said to be strong. However, it should also be noted that if Zwart (1990) is indeed on the right track, then one straightforward prediction from 44b is, since C° is 'strong' in Dutch (and German), that adjunction to CP should be possible. That this is an incorrect prediction is shown by the following examples:25 (49) Du. a. *In de biblioteekj [cp de hockey heeft het kind 11; gelezen] In the library the books has the child read
30
PARAMETERS AND FUNCTIONAL HEADS
b. *Ik geloof niet in de biblioteekj [cp dat het kind de boeken tgelezen heeft] I think not in the library that the child the books read has Summing up, then, the facts of adjunction to the maximal projection of the respective 'strong' functional heads do not follow from the proposal in 44b: Although 1° is categorized as the strong functional head in English, in certain circumstances adjunction to IP is excluded; in Dutch (and other V2 languages), where C° is strong, adjunction to CP is not permitted.
3
The V2-inside-IP Analysis
Here we will mainly address the analyses of Yiddish in Diesing (1988, 1990) and of Icelandic in Rognvaldsson and Thrainsson (1990) and not the analyses of Yiddish in Santorini (1989) and of Danish in Reinholtz (1989). All four analyses agree that in a non-subject-initial V2 clause, the first element is in IP-spec and that IP-spec in this instance is an A-bar-position. The above analyses also agree that in a subject-initial V2 clause, the subject is in IP-spec. Where the analyses differ, however, is in regard to the status of IP-spec when filled by a subject. According to Diesing and to Rognvaldsson and Thrainsson, when occupied by a subject, IP-spec is an A-position (in short, IP-spec varies with respect to Aand A-bar status). On the other hand, both Santorini and Reinholtz suggest that a sentence-initial subject is completely parallel to a sentence-initial non-subject; not only are both in IP-spec, but IP-spec is an A-bar-position in both cases as well (in short, IP-spec is always an A-bar-position).26 Let us recapitulate the differences between the three major lines of analysis discussed in this paper:
(50)
Position of the first element in:
The Asymmetry analysis (discussed in section 2) The V2-outside-IP analysis (advocated by this paper) The V2-inside-IP analysis (discussed in this section)
3.1
a subject-initial V2 clause
a non-subject-initial V2 clause
IP-spec
CP-spec
CP-spec
CP-spec
IP-spec
IP-spec
Extraction out of Embedded V2 Clauses in Yiddish
Diesing (1990:62, ex. 30; 74, ex. 52) uses the following evidence from Yiddish to argue that V2 in embedded clauses takes place inside IP: (51) Yi.
Vemenj hot er nit gevolt az lot di bikher zoln mir gebn t;l ? Who(dat) has he not wanted that PRT the books should we give?
THE VERB ALWAYS LEAVES IP IN V2 CLAUSES
31
The V2 clause is bracketed, and the element in the specifier-position of the V2 clause is underlined. Diesing argues that as extraction must take place via CP-spec, ot di bikher in 51 cannot be in CP-spec but must instead be in IP-spec. Therefore embedded V2 clauses like 51 must involve topicalization to IP-spec, and the subject must be inside the VP. Following Vikner (1994c, section 4.8), it seems that it is only argument extraction from an embedded V2 clause that is perfectly grammatical. If adjunct extractions are considered, then we see that extraction is not necessarily as free as the data in Diesing might lead us to expect (1991:118, ex. 170):27 (52) Yi.
Viazoy hot zi gezogt... How has she said . . . a. ??... az lin shul hobn di kinder gelernt geshikhte t] ? ... that in school have the children learned history? b. ??... az [di kinder hobn gelernt geshikhte t] ? . . . that the children have learned history?
Furthermore, the more crucial of the two kinds of extraction is adjunct extraction, for the following reason: Whereas every trace in the chain of an adjunct extraction has to be antecedent governed, the intermediate traces in an argument extraction chain have to observe only subjacency. That the intermediate traces in argument extractions do not have to have antecedent government inside the chain has (at least) two different motivations in the literature: According to Chomsky (1986:17-18), following Lasnik and Saito (1984), a chain ending in an argument position must be licensed with respect to the ECP (through gamma-marking) at S-structure, whereas chains ending in non-argument positions must be licensed at LF. At LF, all empty categories must be or have been licensed. This means that the intermediate trace t' properly governing the trace t in an argument position may do so at S-structure and then disappear at LF (Lasnik and Saito 1984:258). In contrast, the intermediate trace t' properly governing the trace t in the base-generated position of the adjunct must do so at LF; hence t' must exist at LF and will itself have to be properly governed. Thus in an argument chain, it is only the trace in the base position that must observe the ECP by being properly governed; in non-argument chains, on the other hand, all traces, including the intermediate ones, must observe the ECP by being properly governed. An alternative motivation for claiming that intermediate traces in argument chains do not have to be antecedent governed comes from Rizzi (1990a: 85-95). Within the Relativized Minimality framework, argument extraction and adjunct extraction are alike in that the trace in the base position of both (non-subject) arguments and adjuncts may satisfy the head-government requirement ("formal licensing") of the ECP (Rizzi 1990a:87, 82). The two kinds of extraction differ in the way their lowest trace is linked to the moved element ("identification"): When an argument is extracted, the extraction is subject only to subjacency. This is because the extracted element may be linked to its trace through binding, as an argument has a referential index. Extraction of an adjunct, on the other hand, is subject to antecedent government (as well as to subjacency). The extracted element may not be linked to its trace through binding, since an adjunct has no referential index (Rizzi 1990a:86, 76-80).
32
PARAMETERS AND FUNCTIONAL HEADS
As examples of this difference, Rizzi (1990a:88, ex. 32b, c) gives the following contrast:
In either case the embedded CP-spec cannot be part of the chain between the whelement at the head of the main clause and its trace inside the embedded clause. That 53a is not as unacceptable as 53b is ascribed to the fact that the link between which problem and the trace in its base-generated position is not subject to the antecedent government requirement. Whichever account is preferred, argument extraction is subject only to subjacency requirements (though see the discussion of 54, below), whereas non-argument extraction is subject both to subjacency and to an antecedent government requirement. In what follows, we shall phrase our analysis in the terms of Rizzi (1990a), but as far as we can tell, it could also have been done in terms of Chomsky (1986) and Lasnik and Saito( 1984). When seen in the light of the above discussion, the Yiddish data of 51 and 52 are open to another interpretation: It is not the case that Yiddish always allows extraction out of embedded V2 clauses. That argument extractions are permitted and adjunct ones are not seems to suggest instead that Yiddish allows violations of subjacency.28 Let us now go through the crucial examples of Diesing (1990:71-75, ex. 49-54), both discussing the problems they pose for her analysis and showing how they may be accounted for under a V2-outside-IP analysis. Such an analysis will make the following two basic assumptions: a. complementizer-less embedded clauses are V2 and therefore CPs. That this is also the case in German can be seen from the ungrammaticality of 28b in section 2.3.1. b. Yi. az 'that', (like Ic. ad 'that') obligatorily selects CP (viz. [CP [c; az/a5 [CP . . . ] ] ] ). We start with object extractions (the base position is thus to the right of the embedded verb leyenen 'read'). As above, the V2 clause is bracketed and the element in the specifier-position of the V2 clause is underlined (the underlined element is thus in IP-spec in Diesing's analysis and in CP-spec in ours) (Diesing 1990:71-72, ex. 49): (54) Yi. a. Vos hot er nit gevolt az [mir zoln leyenen -] ? What has he not wanted that we should read ? b. ?Vos hot er nit gevolt az [es zoln mir leyenen -] ? What has he not wanted that it should we read? c. *Vos hot er nit gevolt az [_ zoln mir leyenen -] ? What has he not wanted that should we read?
THE VERB ALWAYS LEAVES IP IN V2 CLAUSES
33
d. Vos hot er nit gevolt [— zoln mir leyenen —] ? What has he not wanted should we read? e. *Vos hot er nit gevolt [mir zoln leyenen —] ? What has he not wanted we should read? Under our account, 54a violates subjacency since the extraction is crossing IP and CP (and CP-spec is filled by mir), but as we have just seen this does not seem to matter in Yiddish (cf. 51 and 52, above). 54b violates only subjacency, not the ECP, just like 54a, and it should thus be acceptable both under our analysis and under Diesing's (cf. the following quote from Diesing 1990:73, n. 25: "[example 54b] is marginal for some speakers. I have no explanation of why this should be so." 54c is a violation of the ECP: The intermediate trace (in IP-spec from Diesing's point of view but in CP-spec according to ours) is not properly head governed, as az 'that', cannot be a proper governor (following Diesing 1990:74). Notice that in 54c, the ungrammaticality is caused by an intermediate trace of an argument extraction, seemingly in contradiction to both Chomsky's and Rizzi's ideas that intermediate links in an argument extraction are subject to only subjacency. So while it is obvious that if the intermediate trace exists, it violates the ECP, the question is: Why does it exist at all? In other words, why can it not delete as other intermediate argument traces are allowed to do? The answer seems to stem from the V2 phenomenon. That V2 overrides the possibility of deleting intermediate traces of an extracted argument is motivated by the idea that only traces not contributing to the interpretation of a sentence can be deleted (Chomsky 1990, class lectures). In fact, when viewed from this perspective, two accounts of the ungrammaticality of 54c fall out. The first is as suggested above: The trace, which must exist in order to satisfy the V2 constraint, is unable to be properly governed by az. If, in contrast, the specifier of zoln is empty, the sentence violates whatever it is that forces movement of an XP into a specifier of the pre-subject finite verb in all V2 languages. In either derivation, 54c is ruled out. The fact that 54d is well-formed supports the above analysis of 54c and in particular that an intermediate argument trace filling the initial position in adherence to V2 cannot be deleted. Only in a derivation in which there is an intermediate trace in the specifier position of the embedded verb zoln can this sentence be acceptable; this trace, in turn, is properly governed by the matrix verb gevolt. (The alternative derivation, with the specifier of the verb zoln left empty will be ruled out, but this is irrelevant.)29 54e seems to us to be somewhat problematic for both Diesing's and our approaches. Considering the available positions in the structure, three different derivations are possible in principle. Below we schematise the possible positions of mir and
(55) a. b. c.
... ... ...
CP-spec
C°
mir mir
zoln
IP-spec
1° zoln
mir
zoln
34
PARAMETERS AND FUNCTIONAL HEADS
Notice that Diesing explicitly rules out the possibility that embedded clauses like the one in 54e have no CP level at all: The complementizer az is optional in many embedded clauses. I will assume that in these case there simply is no CP node. In the case of w/j-movement from an embedded clause, the CP node is obligatorily expanded by the w/i-word moving to [Spec, CP], requiring government of C. (Diesing 1990:75-76, n. 27) This means that to Diesing, 55a, b are excluded, as there is no room in CP-spec for an intermediate trace of vos. She furthermore excludes 55c (and 55a again) by saying that if there is no verb in C°, then C° is not properly governed and it should be. The claim that C° needs to be properly governed (under the derivation in either 55a or 55c) is not self-evident. Diesing seems to be following the proposals of Travis (1984, 1986, 1991), resorting to the idea that a base-generated empty X° category is subject to the ECP (whereas, according to the standard view, the ECP applies only to categories which are empty as a result of movement). For a detailed discussion of the problems with such an approach, see Schwartz and Vikner (1989) as well as section 2, above. To us, 55a, c are excluded as violations of whatever motivates V2. The question for the V2-outside-IP analysis, however, is what excludes the third derivation, 55b. It should be a subjacency violation, on a par with 54a (in both examples an argument extraction has to cross both an IP and a CP without having a trace in CP-spec), and yet 54e is much worse than 54a. The problem for us is that it seems that Yiddish sometimes allows violations of subjacency and sometimes not. It would appear that these violations are allowed when the relevant subjacency barrier is the CP selected by az, 54a, but not when it is the CP selected by the matrix verb, 54e. Consider now the next set of examples which show subject extractions, also from Diesing (1990:75, ex. 53): (56) Yi. a. ?Ver hot er moyre az [es vet kumen ]? Who has he fear that it will come? b. *Ver hot er moyre az [ vet kumen ]? Who has he fear that will come? c. Ver hot er moyre [ vet kumen ]? Who has he fear will come? 56a is just like 54b: subjacency is violated, which itself does not result in ungrammaticality in Yiddish 56b is parallel to 54c: The (undeletable) trace in the specifier of the V2 clause violates the ECP: Since az is not a proper governor, the trace is not properly governed. 56c is like 54d above: The trace needed to observe V2 is properly governed by the matrix head that selects CP, in this case the N° moyre. Note, moreover, that the fact that this is a V2 clause is clear from the German version of it, where the finite verb precedes the infinitive:
THE VERB ALWAYS LEAVES IP IN V2 CLAUSES
35
The last of Diesing's examples is 58 (1990:75, ex. 54):
58, which also should be a subjacency violation, is parallel to 54a and 51. Let us finally reconsider the adjunct extractions in 52, above (repeated here as 59):
According to our analysis, 59a, b are ruled out. Following Rizzi (1990a), the trace in the base position of the extracted element, t, cannot be linked to the intermediate trace in the specifier of az, t', by referential indices, as adjuncts do not have referential indices. Binding cannot provide the linking necessary to identify the trace, t, so we have to look to antecedent government to do this. The problem is that t' does not antecedent govern t, as there is an intervening governor of the relevant kind (A-bar), namely the underlined element in the specifier of the embedded finite verb, hobn. Contrast this with the following example where there is no intervening A-bargovernor, since, in fact, the intermediate trace itself is in the specifier of hobn (Vikner 1994c, section 4.8.1):
It would appear that Diesing's analysis has no way of accounting for the difference in acceptability between argument and adjunct extractions (but cf. n. 27, above). Under her approach, the underlined elements in 59a, b would be in IP-spec, and would therefore not induce any kind of violation, as extractions out of IP are generally allowed. In this section we have shown that neither the V2-inside-IP approach of Diesing nor the V2-outside-IP approach presents us with a completely satisfactory set of predictions concerning extraction from embedded V2 clauses. We will therefore have to look to other phenomena to indicate which analysis is preferable; this is what we will do below, considering adverbial positions in section 3.2, subject-verb agreement in section 3.3, topicalization in relative clauses and in embedded questions in section 3.4, and inversion with a topicalized element in section 3.5.
36
PARAMETERS AND FUNCTIONAL HEADS
3.2 Adverbial Positions In this section we will show how data concerning adverbials which occur to the right of the subject in embedded V2 clauses in Icelandic and Yiddish provide an argument against the claim of the V2-inside-IP analysis that in non-subject-initial V2 clauses, the subject is in VP-spec. The data argue only against the subject being in VP-spec in such clauses and argue only indirectly in favor of the initial element being in IP-spec, as this presupposes that there is only one functional head between C° and V°, namely 1°. We will start out with Icelandic since the data are clearer here. In Icelandic embedded non-subject-initial V2 clauses, the subject always precedes the sentential adverbial: (61) Ic. a. Hann veit aS k-innski las Jon aldrei bokina b. *Hann veit a5 kannski las aldrei Jon bokina He knows that maybe read (Jon) never (Jon) book-the c. Hann veit a9 kannski hefur Jon ekki lesiS bokina d. *Hann veit ad kannski hefur ekki Jon lesiQ bokina He knows that maybe has (Jon) never (Jon) read book-the If the finite verb is in 1° (and there is no IP-recursion), then the subject must be either in VP-spec or in the specifier of some intermediate projection, e.g., TP-spec. As both Diesing (1988, 1990) and Rognvaldsson and Thrainsson (1990) explicitly take the subject to be in VP-spec, we will mainly argue against this. The first argument applies to the possibility of the subject occurring in either IPspec or in TP-spec. The data concern the position of adverbials in relation to the subject and a participle: The negative sentence adverb aldrei 'never' in 6la and the negation ekki 'not' in 6 Ic, like other sentential adverbials, should only occur adjoined to, or in the specifier position of, an XP relatively high in the tree, for reasons of scope. Furthermore, within the Relativized Minimality framework (Rizzi 1990a), the fact that ekki induces a negative island (as would also negative adverbs such as aldrei) points towards the negation being in TP-spec (cf. the following data): (62) Ic. a. Hversu margar baekur hefur Jon Iesi61? How many books has Jon read? b. Hversu margar baekur hefur Jon ekki lesiS t? How many books has Jon not read? (63) Ic. a. HyaS hefur Jon lesiS [t margar bajkur] ? What has Jon read many books? (= How many books has Jon read?) b. *Hya8 hefur Jon ekki Iesi3 [t margar btekur] ? What has Jon not read many books? Asking for the number of books that Jon has read can take two forms, either How many books ... , 62a, or What ... many books ... , 63a.31 If, however, you want to know the number of books that Jon has not read, only the former strategy works,
THE VERB ALWAYS LEAVES IP IN V2 CLAUSES
37
62b; the latter does not, 63b. Following Rizzi (1990a: 15-22) this may be accounted for by assuming the negation to be in TP-spec, blocking the A-bar-movement of non-arguments across it (i.e., of elements which do not have a referential index). The whole object can be moved across negation into CP-spec, 62b, but just a part of the object cannot, 63b, as it does not have a referential index and the negation is blocking antecedent government of the trace, thereby leaving the trace without any link to its antecedent. Carrying this conclusion over to 61, if the negation in 61 a, c is in TP-spec, then the subject, which is to the left of the negation, can neither be in VP-spec nor in TP-spec. The second argument only goes against the subject occurring in VP-spec. The adverbial drugglega 'surely', like its English counterpart, has two different interpretations which depend on its position in the sentence: (64) Ic. a. Vilhjalmur mun orugglega hitta epliS b. Vilhjalmur mun hitta epliS orugglega Vilhjalmur will (surely) hit apple-the (surely) In 64a drugglega is a sentence-adverbial, with the meaning 'definitely/certainly/absolutely'. In 64b drugglega is a VP-adverbial, with the meaning 'in a sure manner'. Let us now consider what happens in embedded clauses: (65) Ic.
Eg held a31 gaer hafi . . . I believe that yesterday has . . . a. ??... orugglega Vilhjalmur hitt epli5 b. . . . Vilhjalmur orugglega hitt epliS . . . Vilhjalmur hitt epliS orugglega ... (surely) Vilhjalmur (surely) hit apple-the (surely)
The subject cannot occur to the right of the adverbial, 65a, parallel to 61b and 6Id (insofar as 65a is interpretable, the adverbial would not be a sentence adverbial but would modify only Vilhjalmur 'at least Vilhjalmur' or 'certainly Vilhjalmur', as opposed to anybody else). If the subject is in VP-spec, as claimed by Rognvaldsson and Thrainsson (1990), the adverbial in 65b, which only has the sentence-adverbial interpretation, is adjoined to V-bar. The adverbial in 65c, which has only the VPadverbial interpretation, can either be adjoined to V-bar or to some larger constituent, e.g., VP or TP. If the adverbial in 65c is adjoined to V-bar, then the adverbial in 65b and the one in 65c would be adjoined to the same constituent, and we would expect them to have identical interpretations or scopal properties, which is not the case. If the adverbial in 65c is adjoined to VP or higher, we would expect it to have wider scope than the one in 65b, exactly contrary to fact. Let us now compare the analysis just given which posits the subject in VP-spec with an analysis that assumes the subject to be in IP-spec. Under such an analysis, the adverbial in 65b can be adjoined to TP and the one in 65c to VP. If, following the argumentation concerning 62-63, we assume that the negation is in TP-spec, then the following data support the analysis that when drugglega occurs right of the subject and left of the participle it is adjoined to TP:
38
(66) Ic.
PARAMETERS AND FUNCTIONAL HEADS
Eg held a8 a morgun muni Vilhjalmur... I believe that tomorrow will Vilhjalmur . . . a. . . . orugglega ekki hitta epliS b. *... ekki orugglega hitta epliS ... (surely) not (surely) hit apple-the
Another argument of a closely related nature against the subject being in VP-spec concerns the scope interactions between adverbials and quantified objects:32 (67) Ic.
Helgi sagQi... Helgi said . . . a. . . . aS bess vegna hafSi Jon oft Iesi5 margar baekur b. . . . a3 bess vegna haf5i Jon lesiS margar baekur oft . . . that therefore has Jon (often) read many books (often)
The interpretations of 67a, b differ in exactly the same way as those of their English counterparts: 67a means that Jon often reads many books (for some particular reason), whereas 67b means that there are many books which (for some particular reason) Jon often reads. This again clearly shows that when the adverbial occurs between the subject and the participle it has higher scope than when it occurs sentence-finally. As the sentence-final adverbial in 67b cannot possibly have a position in the tree lower than adjoined to V-bar (as it is preceded by the object), the adverbial in 67a must occur in a higher position, which means that it in turn cannot be lower than in VP-spec or adjoined to VP. Both of these in turn exclude the subject being in VP-spec in 67a. The fourth argument against the subject being in VP-spec is conceptual in nature and is based on X-bar-Theory. If the subject is in VP-spec in 6la, c, then the fact that the adverbial would have to occur between VP-spec and the complement of V° implies a particular D-structure representation. This structure (before the verb leaves VP) would have to be the following, as is in fact explicitly assumed by Rognvaldsson andThrainsson(1990:10,ex. 10, 11):
We take a structure like 68 to be explicitly ruled out in the X'-system of Chomsky (1986), as adjunction to an X-bar is impossible. (This point is also made for Danish inRcinholtzl989:107.) 33 In sum, since the subject in 61 a, c cannot be in VP-spec (and if we disregard the TP-spec option), then there is no possible analysis of these well-formed examples with the finite verb in 1°. A different analysis therefore needs to be found for 61a, c.
THE VERB ALWAYS LEAVES IP IN V2 CLAUSES
39
Let us now turn to the Yiddish data. Here, in contrast to the Icelandic data, both the subject-adverbial and the adverbial-subject orders seem to be possible: (69) Yi. a. . . . ... b. . . . ...
az morgn vet dos yingl in emesn zen a kats that tomorrow will the boy in truth see a cat az morgn vet in emesn dos yingl zen a kats that tomorrow will in truth the boy see a cat
What is important for us is that 69a is possible at all. Following the argumentation above, the adverbial in emesn 'in truth', in 69a must be adjoined to VP, for reasons to do with both scope and X-bar-Theory, and then the subject dos yingl 'the boy', must be outside the VP. As with Icelandic, this means that (at least in 69a) the subject must be in IP-spec (again provided we disregard the TP-spec option), which in turn implies that the initial element, morgn 'tomorrow', in this non-subject-initial V2 clause must be in a specifier outside IP, e.g., CP-spec. A very similar argument may be made on the basis of data concerning object shift in Icelandic (cf. among others Holmberg 1986:218 and Vikner 1989, 1994b) and movement of weak pronouns in Yiddish (cf. den Besten and Moed-van Walraven 1986:123-125). Consider the following Icelandic examples: (70) Ic.
Hann veit... He knows ... a. ... a6 bess vegna lasy Jon bokinaj ekki ty t; b. *... a5 bess vegna lasv b6kina; Jon ekki tv t; . . . that therefore read (book-the) Jon (book-the) not
We know from the fact that bokina precedes the negation that it has left its base position (which is to the right of the verb trace which again is to the right of the negation). The question is now what position the object has moved to in 70a. If the subject, Jon, were in VP-spec, then bokina must have adjoined to V-bar, something we also take to be excluded by X-bar-Theory. If on the other hand Jon is in IP-spec, then it is possible that bokina has adjoined to VP, which is perfectly compatible with X-bar-Theory. As was the case concerning the adverbials above, the argument concerning object movement can likewise be repeated for Yiddish but in a weaker form. Consider the following: (71) Yi. a. Miriam hot gezpgt az dos Bukh hot Mendele ir gegebn b. Miriam hot gezogt az dos Bukh hot ir Mendele gegebn Miriam has said that the book has (her) Mendele (her) given On the assumption that the object ir 'her', is generated right of the main verb, then it must have moved in order to get to the position it has in 7 la; if the subject Mendele is in VP-spec, then the object ir must have adjoined to V-bar, which is problematic with respect to X-bar-Theory. On the other hand, if Mendele is in IP-spec, then ir may have adjoined to VP in 71a, and then 71b would have to be analyzed as adjunction of ir to IP.34 So far we have seen five reasons why the subject in 61 a, c (and in 69a) cannot be in
40
PARAMETERS AND FUNCTIONAL HEADS
VP-spec, only one of which also argues against it being in TP-spec. There is another argument to be made against the subject being in TP-spec, and that is that TP-spec is an A-bar-position. This assumption is made, for example, by Roberts (1993), and it is supported by the Relativized Minimality analysis of negative islands (cf. 62-63, above) and of the so-called "pseudo-opacity" phenomena (cf. Rizzi 1990a: 12-15, which is based on Obenauer 1976, 1984). Summing up, if the subject in 61 a, c and 69a can occur neither in TP-spec nor in VP-spec, as has been argued above, then it is not possible to analyse these well-formed examples as having the finite verb in 1°. In contrast, however, were we to posit that the finite verb is in C°, then a third possibility for the position of the subject in 61a, c and 69a may be taken into consideration: The subject could be in IP-spec. This analysis suffers from none of the defaults discussed above, and as the subject being in IP-spec excludes IP-spec as the landing site of topicalization, we shall take this to be an argument against the V2-inside-IP analysis.
3.3
Subject-Verb Agreement Facts
In Yiddish and Icelandic, the finite verb agrees in number and person with the subject, irrespective of whether the initial element in the V2 clause is the subject, the object, or an adverbial. Consider the following examples, where the initial element in the embedded V2 clause is the object. At the top, we have pointed out which position the elements are claimed to occupy, both according to the V2-outside-IP analysis and according to the V2-inside-IP analysis:35
(72)
V2-outside-IP
CP-spec
V2-inside-IP
IP-spec
C° 1°
VP-spec
IP-spec
a. b.
Yi. Ic.
dos bukh . . . az bokina ... ad the book . . . that
hobn hafa have
di kinder bornin the children
geleyent nekhtn lesiS i gsr read yesterday
c. d.
Yi. Ic.
dos bukh *. . . az bokina *. . . ao" . . . that the book
hot hefur has
di kinder bornin the children
geleyent nekhtn lesiQ i gaer read yesterday
In the V2-outside-IP analysis, the subject-verb agreement is a realisation of SpecX° agreement inside IP: The subject in IP-spec agrees with the verbal inflection generated in 1°. That the verbal inflection then has to move to C° along with the verb stem does not alter this. In the V2-inside-IP analysis, the subject-verb agreement seems to be more difficult to account for (to our knowledge, this issue has not been addressed by proponents of this analysis). If the subject is in VP-spec, we would expect that the only head which could show agreement with the subject would be the verb stem, which is generated in V°. This may not be a problem, as the verb stem does not show any agreement at all, and so the idea of agreement between subject and verb stem cannot be empirically rejected. However, if the object is in IP-spec, we would expect the verbal inflection,
THE VERB ALWAYS LEAVES IP IN V2 CLAUSES
41
which (we assume) is generated in 1°, to show agreement with the object and not the subject. This is clearly not the case, as it would predict 72a, b to be ungrammatical and 72c, d to be grammatical—exactly the wrong prediction. The only way for the V2-inside-IP analysis to avoid this problem would seem to be to give up the idea that morphological agreement is a manifestation of a head-specifier relationship, something which can be retained in the V2-outside-IP analysis.
3.4
Topicalization in Relative Clauses and Embedded Questions in Yiddish and Icelandic
Within the V2-inside-IP analysis, the IP-spec position can serve both as an A-position (when the subject moves there) and an A-bar-position (when any other XP — including wft-elements—moves there). Consequently, no CP is assumed to exist in main clauses in Diesing (1988, 1990) or in Rognvaldsson and Thrainsson (1990). In embedded clauses, in contrast, CP is needed to accommodate either an overt complementizer in C° or a w/z-word/trace in CP-spec. It is thus expected that in embedded clauses, it should be possible to have both topicalisation (of the subject or a nonsubject) and w/z-movernent. In other words, given that in embedded contexts (only), topicalisation and w/z-movement are claimed to be movement to distinct postions (IP-spec and CP-spec, respectively), it should be unexceptional to find them cooccurring. Specifically, this predicts that topicalization in embedded questions and relative clauses (where something other than the subject occurs after the w/z-word but before the verb) should be unexceptional—but they are not. In this section, we will first discuss the data and show that topicalization in embedded questions and relative clauses is a rather restricted phenomenon, and then we will discuss the proposals that Diesing (1990) offers to account for these restrictions. We first show that topicalizations in embedded questions and relative clauses are not as easily accepted as the particulars of the proposal by Diesing (1990:62-67) for Yiddish or by Rognvaldsson and Thrainsson (1990) for Icelandic would seem to predict. In relative clauses, topicalization is perfectly acceptable when the subject is relativized: (73) Ic. Flokkur sem [urn fjogurra ara skei8 hefur veriS i stjorn] tapa5i kosningunum A party that in four years' course have been in government lost election-trie (= A party which had been in government for four years lost the election) (Rognvaldsson 1984:6, ex. 12) (74) Yi. ... nokh epes, [vos oyfn hitl iz geven]... ... still something that on-the little-hat is given . . . (= something else that was on the little hat) (Santorini 1989:56, ex. 36a) However, if the object is relativized, not only are these sentences not nearly as good as expected, they are in fact unacceptable:
42
PARAMETERS AND FUNCTIONAL HEADS
(75) Ic. *Helgi hefur keypt bok, [sem trulega hefur Jon ekki lesiS] Helgi has bought a book that probably has Jon not read (Thrainsson, personal communication) (76) Yi. *Der yid [vos in Boston hobn mir gezen] iz a groyser lamdn The man that in Boston have we seen is a great scholar (Lowenstamm 1977: ex. 34d) It would seem that in order to improve the acceptability of sentences like 76, the topicalized element (which is in IP-spec according to the V2-inside-IP analysis) needs rather heavy contrastive stress, as demonstrated by Santorini (1989:56-57, ex. 38): (77) Yi. Der yid vos mir hobn gezen in Niu-York iz an amorets, ober der yid [vos in Boston hobn mir gezen] iz a groyser lamdn the man whom we have seen in New York is an ignorant but the man whom in Boston have we seen is a great scholar In embedded questions as well, topicalization is not always well-formed. We must point out, however, that with respect to embedded questions, it is less easy to classify even superficially the different groups. The following are grammatical examples: (78) Ic.
Eg spurSi [hvar henni hefSu flestir aSdaendur gefi5 blom] I asked where her(dat) had most fans given flowers (= I asked where most fans had given her flowers) (Thrainsson 1986:186, ex. 28b)
(79) Ic. a. ?Ikh veys nit [tsi dos bukh hot er geleyent] b. Ikh veys nit [tsi ot dos bukh hot er geleyent] I know not whether (PRT) the book has he read (Diesing 1990:66, ex. 40) Note that the sole difference between 79a and 79b is the addition of the particle ot, which, according to Diesing (1990:66), gives the NP "contrastive emphasis". Yet, ungrammatical examples of non-subject (NP) topicalization in embedded questions are also found: (80) Ic.
Eg veil ekki [af hverju . . . I know not why ... a. *... Mariu hefur Olavur eiginlega lofaS bessum hring] . . . Maria(dat) has Olavur actually promised this ring(acc) b. * . . . bessum hring hefur Olavur eiginlega Iofa3 Mariu] . . . this ring(acc) has Olavur actually promised Maria(dat) (Thrainsson, personal communication)
(81) Yi. *Ikh veys nit [vemen zuntik hot zi gezen] I know not whom Sunday has she seen (Diesing 1990:63, ex. 32a, citing Travis 1984)
THE VERB ALWAYS LEAVES IP IN V2 CLAUSES
43
Furthermore, Diesing suggests that topicalizations of non-NPs may be more acceptable than topicalization of NPs, and she gives an example of the following type as evidence (Diesing 1990:61, ex. 41): (82) Yi. a. Ikh veys nit [far vos in tsimer iz di ku geshtanen] b. *Eg veit ekki [af hverju i herberginu hefur kyrin staSiS] I know not why in the room has the cow stood (Notice that the parallel example is not possible in Icelandic). As suggested in Vikner (1994c, section 4.1.4), the grammaticality of 82a may be linked to the fact that the indirect question is introduced by why. If we try topicalization embedded under other w/i-elements, the result is completely unacceptable: (83) Yi. a. *Ikh veys nit [ven in tsimer iz di ku geshtanen] b. *Eg veit ekki [hvenaer i herberginu hefur kyrin stadiS] I know not when in the room has the cow stood (84) Yi. a. *Ikh veys nit [vu nekhtn iz di ku geshtanen] b. Eg veit ekki [hvar i gaer hefur kyrin sta5i9] I know not where yesterday has the cow stood (ex. 82-84; Prince, Thrainsson, personal communications) (Notice that these are ill-formed both in Yiddish and in Icelandic.) So, on the face of it, this area is problematic to all of the analyses proposed, as none is capable of accounting for the complete distribution of the grammatical and the ungrammatical sentences. It therefore follows that both analyses will have to say something extra to account for these data. Diesing (1990:66-67) finds the good examples 'a sufficiently robust phenomenon to warrant generation by the grammar', and goes on to suggest an analysis based on "independent discourse conditions on topicalization and Whmovement." In short, she argues that since in the embedded context there are two landing sites for movement, only one of which can be emphasized, a contradiction arises when movement of a w/i-element ("inherently emphasized") and topicalisation of a non-subject co-occur: . . . [T]he difference between embedded topicalization of the subject vs. . . . nonsubjects with a [+Wh] CP can be formulated in terms of the A vs. A-bar distinction . . . As an A-position . . . [Spec, IP] is an unemphasized position. A-bar movement to [Spec, IP] results in an operator interpretation which requires an added emphasis. This interpretation of non-subject topicalization is odd in the context of an embedded question or relative clause unless there is some additional emphasis . . . The added emphasis resolves the clash between Wh-extraction and the topicalization. (Diesing 1990:67; emphasis ours)
So, as she states, the embedded sentences/questions with a non-subject in initial position improve if stressed. As we have seen, such a strategy does not capture all of the data — not even considering only the Yiddish facts. In addition, it does not seem to us to be intuitively
44
PARAMETERS AND FUNCTIONAL HEADS
appealing to claim, on the one hand, that in Yiddish, A-bar-movement to IP-spec parallels traditional A-bar-movement to CP-spec in main clauses but that in embedded contexts this generalisation is excluded. Moreover, it seems to be the case that IPspec, in so far as it is an A-bar-position, simultaneously is an inherently emphasized position but also requires additional emphasis (cf. 77, above). Notice that when IPspec is an A-bar-position in main clauses no such additional emphasis is needed. Nor is it the case that material in the CP-spec position needs emphasis — even in double w/i-questions, despite the fact that they, too, are both operator-variable chains — and which would, pursuing this proposal, need emphasis to "resolve . . . the clash" (Diesing 1990:67). And finally, even if there were a way to make the idea of discourse constraints sound more plausible, it still seems to require a syntactic reason for IPspec as an A-bar-position to be treated differently in main as opposed to embedded clauses. Although we have no alternative to offer, given the array of data considered, we find Diesing's proposal suggestive but unfortunately insufficient. In this section, we have tried to show, on the one hand, that the data concerning topicalization in relative clauses and embedded questions is quite difficult to account for, and on the other hand, that although Diesing's attempt to give such an account is very laudable, it is also rather counterintuitive. In conclusion, it seems to us that this is another area where it is not yet possible to find a reason to prefer either the V2-inside-IP or the V2-outside-IP approach. It therefore seems that the data and arguments such as discussed in sections 3.2 and 3.3, which favored the V2-outside-IP approach, are perhaps more revealing.
3.5
Inversion with a Topicalized Element in Yiddish
In the previous subsection we saw that Diesing (1990:54, ex. 20) allows topicalization to take place inside an embedded question, with wft-movement into CP-spec and topicalization into IP-spec (e.g., 82a). This is not the case in main clauses, however, since w/z-elements do not occur in CP-spec in main clauses, but only in IP-spec, like topicalized elements. As pointed out by Heycock and Santorini (1992, section 2.2), such an analysis makes the following prediction: In an extraction from an embedded clause, it should not matter whether IP-spec contains a subject or a non-subject, i.e., the verb should be able to precede either a subject or a topicalized element, since both should be able to occur in IP-spec. 85 shows that this prediction is not borne out (Heycock and Santorini 1992:4, ex. 8). (85) Yi.
Yemen; hot er nit gevolt... Who(dat) has he not wanted . . . IP-spec
CP-spec
C°
a.
. . . tj
zoln should
min we
t: gebn tj ot di bikher? give PRT the books?
b.
*. . . tj
zoln should
ot di bikherj PRT the books
mir tj t: gebn? we give?
THE VERB ALWAYS LEAVES IP IN V2 CLAUSES
45
The same point was raised in Vikner (1990, section 2.3.3.1) as a problem for Santorini's (1989:98) account of VI declaratives, which are analyzed with the verb in C°: Why can the verb in VI declaratives not be followed by a topicalized element as well as by the subject, if it is the case that IP-spec may contain either? In part in response to data like 85, Heycock and Santorini (1992) propose an analysis where main clause questions are CPs but main clause topicalizations IPs. In the following we summarize what is essential in the theoretical apparatus they propose, and then we will attempt to argue that the analysis that results is itself less than completely satisfactory. Heycock and Santorini (1992) are attempting to explain why, when the verb is in C°, the sole element that may immediately follow the verb is the subject. Therefore much of their analysis depends on the licensing of XP-positions. They hypothesize three ways in which XP-positions can be licensed: theta-assignment, predication, and case assignment. Licensing via theta-assignment occurs only at D-structure, whereas predication and case assignment can license new positions only at S-structure. For example, the direct object position is licensed under theta-assignment at D-structure, whereas IP-spec — not being a theta-position — can be licensed only at S-structure in one of two ways, depending upon the final landing site of the verb (and note that IP-spec must be licensed (1992, section 3.1): As the landing site for topicalization (as well as for clause-initial subjects), IP-spec is licensed via predication; however, in w/z-questions, for example, where the verb moves to C°, a different way of licensing IP-spec is needed, namely nominative case assignment. It is worth pointing out that Heycock and Santorini explicitly rule out the possibility of IP-spec being licensed via predication in this configuration (i.e., when the verb precedes the subject) by proposing that "the relationship between licensing mechanisms and the positions licensed by them is a one-to-one relationship" (1992, section 3.1). The net result of this system is that while thematic positions are licensed at D-structure, the licensing of non-thematic positions depends on the final landing site of the finite verb (and hence that verb traces are insufficient for this kind of licensing). Within this approach, 85b is ruled out because the finite verb in C° can only license IP-spec through nominative case assignment, and therefore the object ot di bikher 'just these books', cannot occur there. Nevertheless, it seems to us that this approach leads to some unwanted predictions. Consider the following two cases of exceptional case marking (either as yes/noquestions or as VI declaratives): (86) Yi. a. Zet der yid plutsling den shokhn kumen t (?) Sees the man(nom) suddenly the neighbor(acc) come (?) b. Lozt der yid plutsling den tsigar fain t (?) Lets the man(nom) suddenly the cigar(acc) fall (?) In both 86a, b, the NP with accusative case is no longer in its base-generated position, which follows the embedded (infinitival) verb kumenlfaln 'come'/'fall'. As it is not in its base-generated position anymore, the question arises as to how the position it occupies is licensed at S-structure. It cannot be via predication, as it is not in a Spec-X° agreement relationship with the verb; the only other possibility is via case assignment. But this in turn cannot be possible either: The finite verb is licensing
46
PARAMETERS AND FUNCTIONAL HEADS
(through nominative case assignment) the subject in IP-spec and cannot also be licensing the surface position of the accusative NP, both because of the one-to-one relationship and because there is no position for the accusative NP to occupy such that it is head-governed by C° and not by 1°, in accordance with minimality (section 3.2). Thus, it seems that under this approach there is no way to license den shokhn in 86a and den tsigar in 86b. Under the V2-outside-IP analysis, the accusative NP in 86a, b is licensed (i.e., assigned accusative case) in the specifier of the embedded VP, but there is no prohibition against this licensing being carried out by a trace. As for the examples in 85, which were problematic for Diesing (1990), they would fall out from our general assumptions: IP-spec is only an A-position, and therefore only the subject may occur there.
4
The Position of 1° in Dutch and German
Having discussed the asymmetry and the V2-inside-IP approaches, we will now turn to a related but still somewhat independent topic: the position of 1° in Dutch and German. As outlined in section 2, it is crucial to the asymmetry analysis that 1° in Dutch and German be to the left of the VP, as it has to provide the landing site for the verb in subject-initial V2 clauses. Furthermore, if one were to suggest that the V2-inside-IP analysis also covered German and Dutch (a suggestion which to our knowledge has not been explicitly made so far), this would also presuppose the sequence I°-VP. If the order is I°-VP, there can be no obligatory V°-to-I° movement (at S-structure) in German and Dutch (cf. the data in 2, above, repeated here: (87) Ge. a. Ich weifi, dafi die Kinder den Film gesehen haben b. *Ich weB, da6 die Kinder haben den Film gesehen I know that the children (have) the film seen (have) The problem for this account, then, is how to explain just what prevents V°-to1° movement in finite embedded clauses. In essence, under either version of the asymmetry approach the idea is that since 1° is "filled" with features, the finite verb is unable to move into it. As noted in Schwartz and Tomaselli (1990), the prohibition against V°-to-I° movement in embedded clauses under Travis' version of the asymmetry analysis is grounded in a rather unorthodox usage of the ECP, namely that proper government by a lexical complementizer of the (phonetically) empty 1° blocks movement of the verb there. In Zwart's (1991:85) version, as summarized above, economy is said to prevent V°-to-I° movement in embedded clauses (with a complementizer), as the complementizer licenses the finite features in 1°. Nevertheless one might still wonder why the complementizer is able to identify the finite features.36 As for the V2 outside IP analysis, it is compatible with the order of either I°-VP or VP-I°. Notice that if the correct order is I°-VP, then it must be concluded that V°-to-I° movement cannot be obligatory; if, on the other hand, the correct sequence is VP-l", then V°-to-I° movement may be obligatory and hence uniform for all tensed
THE VERB ALWAYS LEAVES IP IN V2 CLAUSES
47
sentences. It is of course also perfectly possible within a VP-I° analysis that V°-to-I° movement is not obligatory (or maybe even impossible) in embedded clauses and that the verb thus moves to 1° only when it is on its way to C° (similar to what is commonly claimed about Danish, Norwegian and Swedish). Below we will discuss data from two areas which have been used to support arguments both in favor of and against the obligatoriness of V°-to-I° movement in German and Dutch.
4.1
Richness of Inflection
Several analyses have linked obligatory V°-to-I° movement to the richness of inflection (e.g., Holmberg and Platzack 1988). The basis of comparison comes from examining S VO languages where V°-to-I° movement can be directly observed to take place (or not to take place), such as Italian, Icelandic, French, English, Swedish, etc. (cf. Emonds 1978; Pollock 1989; Holmberg and Platzack 1988; Belletti 1990; and others). The generalization seems to be that the languages which have a relatively rich verbal inflection also have V°-to-I° movement (Italian, Icelandic, and French), whereas the languages that have a poor verbal inflection lack V°-to-I° movement (English, Swedish, and Danish). We should perhaps emphasise that when we refer to V°-to-I° movement, we are primarily focusing on whether it is possible for 1° to be the final landing site for the finite verb at S-structure. Consider this generalization in light of the inflectional system in the Germanic V2 languages: Icelandic, which has V°-to-I° movement, distinguishes between four or five of the six possible forms in the verbal paradigm, whereas Danish, which does not have V°-to-I° movement, makes no distinctions in the verbal paradigm: (88) Ic.
eg tek, bu tekur, hann tekur, vi5 tokum, bi3 takiS, beir taka
(89) Da. jeg tager, du tager, han tager, vi tager, I tager, de tager I take, you take, he takes, we take, you take, they take In view of this difference, let us now consider German: Here we notice that German too has rich inflection, in fact just as rich as that found in Icelandic (four or five different forms out of a possible six): (90) Ge.
ich nehme, du nimmst, er nimmt, wir nehmen, ihr nehmt, sie nehmen I take, you take, he takes, we take, you take, they take
Intuitively, then, German seems to have the same kind of motivation as Icelandic has to force movement of the verb from V° to 1°, even in embedded clauses.37 The verbal inflection in Dutch is much poorer than that in German, even if it is not as poor as that of English or Danish: (91) Du.
ik neem, jij neemt, hij neemt, wij nemen, jullie nemen, ze nemen I take, you take, he takes, we take, you take, they take
However, even though there is a sizeable literature on the differences between German and Dutch, it has to our knowledge never been suggested that one of them has V°-to-I° movement and the other one does not.
48
4.2
PARAMETERS AND FUNCTIONAL HEADS
Dutch Unstressed Pronouns
Jaspers (1989) and Zwart (1991:82-83) both argue that Dutch unstressed pronouns are clitics and that the fact that they obligatorily occur immediately right of the subject may be explained by assuming that they have to cliticize to a functional head which occurs between the subject and the VP. Zwart (1991:82-83) furthermore takes this head to be 1°, and the position of the unstressed pronouns to be evidence that in Dutch, 1° is to the left of VP: (92) Du. a. *... dat Jan het boek gisteren Marie gegeven heeft b. . . . dat Jan It gisteren Marie gegeven heeft . . . that Jan the book/it yesterday Marie given has (93) Du. a. *... dat Marie het boek Jan heeft zien lezen b. . . . dat Marie 'J. Jan heeft zien lezen . . . that Marie the book/it Jan has see read The interesting point about this analysis is that it is compatible with the fact that the full NP cannot undergo this movement, as the ungrammaticality of 92a and 93a shows. It is not obvious that a VP-I° analysis has anything to say about these cases, except that they cannot be cliticization (as there would be no head to which the pronoun could cliticize), and so they must be instances of scrambling instead. That this might be the case is suggested by the fact that both 92a and 92b are grammatical in German:38 (94) Ge. a. ... daB ich das Buch gestern Maria gab b. . . . daB ich es gestern Maria gab . . . that I the book/it yesterday Maria gave It seems to us that the claim that the pronouns in 92b and 93b cliticize to 1° makes a wrong prediction (as noted in Vikner and Sprouse 1988:12), namely that they should move along with the verb when it moves from 1° to C°. In order to see this in the asymmetry approach, a non-subject-initial main clause is required, e.g., a question like the following: (95) Du. a. *Waarom 't heeft [,p Jan gekocht] ? b. *Waarom heeft't [IP Jan gekocht] ? c. Waarom heeft [IP Jan '_t gekocht] ? Why (it) has (it) Jan (it) bought? Underlined are the elements that have been moved from 1° to C° (still from an asymmetry point of view), and it is clear that both of the two logically possible verb + clitic combinations are impossible. Only the verb itself may occur in C° in these cases. It is commonly assumed that once something adjoins to an X°, it can no longer be separated from this X° (see, for example, Kayne 1991:649, who says that a trace cannot be "a proper subpart of a X° constituent" and cites a rule in Baker 1988:73 to the same effect). If this were a generalized constraint, then 95a, b
THE VERB ALWAYS LEAVES IP IN V2 CLAUSES
49
would be ruled out. However, in Roberts (1991), an attempt is made to refine and distinguish different types of head movement. Elaborating on a proposal in Rizzi and Roberts (1989), in which a distinction is made between head movement as adjunction vs. head movement as substitution, Roberts argues that extending the ideas of the Relativized Minimality framework (Rizzi 1990a) to these two types of head movement can explain both the possibility and the impossibility of certain cases of "excorporation": Excorporation is defined as successive cyclic movement of a head which first incorporates into another governing head but then moves out of the head, leaving something behind. The conditions under which excorporation is disallowed are when a host-head morphologically subcategorizes for another head; this is the case of tense and agreement marking in 1° with respect to the verb-incorporee, and it is this that forces verb movement to 1° in this system (see Rizzi and Roberts 1989). In head movement as adjunction, in contrast, moving an adjoined head up to another governing head and stranding either the host or the incorporee is possible, since proper government of the trace of the moved head will still be allowed (i.e., the non-moved head will not count as a closer intervening governor, see Roberts 1991:214-216 for details). If we now apply this system to the facts presented above on Dutch clitics,39 we see that moving the verb to C° does not entail that the clitic move as well. Movement of the verb into 1° by substitution guarantees that the verb cannot be separated from its verbal affix; however, adjoining the clitic to 1° allows the clitic to be stranded. Thus, positing 1° to the left of VP can accommodate the facts of weak pronouns in Dutch not moving along with the verb to C°, as in 95c. Nevertheless, it should be pointed out that other instances of verb movement to C° do not allow the clitic to be stranded, in direct opposition to the Dutch facts. If we follow Baker's proposal, then the pair of sentences in 96a, b is explained: Clitics in French (Fr.), which normally adjoin to 1°, must move along to C° in questions: (96) Fr. a. Ou 1'avait-ilj [IP t; achete] ? b. *Ou avait-ili [Ip tj T achete] ? Where (it) had he (it) bought? The elements that have been moved from 1° to C° have been underlined, and clearly the clitic must move along with the verb when it moves to C0.40 Roberts' system, on the other hand, has no way of ruling out stranding the clitic in 1° in French, as in 96b. Summing up so far, an extension of Roberts (1991) can explain why it is possible for a clitic to stay behind in 1° when the verb moves to C°, as in the Dutch 95c, but not why this is not possible for the French clitic, 96b, nor why it is impossible for the Dutch clitic to stay attached to the verb when it moves into C°, 95a, b. Baker's (1988:73) rule (which disallows traces internal to a word), on the other hand, gives exactly the right predictions for French and exactly the wrong ones for Dutch, provided the Dutch pronoun in 95c is in fact cliticized to 1° (if there is no cliticization, then Baker's rule correctly predicts 95a, b to be ungrammatical). Notice furthermore that it is commonly assumed that cliticization is taking place in French, whereas this is only one of many theoretically possible analyses as far as Dutch is concerned. It therefore seems to us that if an account for the Dutch pronouns as clitics means losing (part of) the explanation for French, then this should be taken
50
PARAMETERS AND FUNCTIONAL HEADS
as a sign that the analysis of Dutch may not be on the right track, a point of view which receives further support from the German data in 94.41 Let us conclude this section by giving a different argument against the analysis of the Dutch unstressed pronouns in 92, 93, and 95 as clitics. This last argument has to do with the interaction between object cliticization to 1° and the realisation of verbal inflection. Although Zwart (1990, 1991) does not address this issue, it seems that in his analysis, affix hopping from 1° to V° would have to be the mechanism by which the finite verb in embedded clauses becomes morphologically encoded (as suggested, for example, in Travis 1984). The problem with this is that the result has the object clitic being cliticized to an 1° which contains nothing but the trace of the verbal inflection (which itself is realised on the verb in V°). Apart from the fact that there is no way of deriving this without violating strict cyclicity, it is also ruled out by the rule from Baker (1988:73) and Kayne (1991:649) discussed above: There would be an X° constituent, 1°, which would contain both the clitic and the trace of the verbal inflection. In short, we find that the problems with assuming that unstressed Dutch pronouns are clitics are bigger than the advantages of this assumption.42
5 Conclusion In this paper we have tried to compare and evaluate three different approaches to verb second: V2-outside-IP, asymmetry (two versions), and V2-inside-IP. In an attempt to sort out the strengths and weaknesses of each approach, particularly in comparing the V2-outside-IP approach to the others, data from across the Germanic V2 languages related to a variety of linguistic phenomena were examined. In some cases, this led us to point out areas where the V2-outside-IP approach fares better than the other approaches. In other instances, we sought to consider data that either seem or have explicitly been claimed to illustrate the superiority of one of the other approaches over the V2-outside-IP approach; in most cases we found that on closer inspection, the data are either indeterminate with respect to the competing approaches or in fact more adequately handled by the V2-outside-IP approach. The following are some of the particular findings regarding both versions of the asymmetry account (Travis 1984,1986, 1991; Zwart 1990,1991): a. Neither is able to account for, e.g., the facts concerning adjunction to a V2 clause (section 2.1) or the fact that expletive es in German occurs in CP-spec (section 2.2.3). b. In contrast, both of these accounts were shown to have ready though distinct explanations for the impossibility of German and Dutch weak object pronouns occurring sentence-initially (section 2.2.1), though other analyses compatible with the V2-outside-IP approach were also briefly discussed; in contrast, the additional data from dialects of Norwegian and Danish, which were shown to be parallel to the Dutch and German facts, can be accommodated only under the V2-outside-IP analysis (section 2.2.2).
THE VERB ALWAYS LEAVES IP IN V2 CLAUSES
51
c. With respect to Travis' version of asymmetry, we considered the facts concerning extraction from embedded clauses (section 2.3.1) and showed that no account for the data can be found within her system. d. And finally we argued that Zwart's version of asymmetry gives rise to problems in the areas of exclamatives (section 2.3.2) and V°-to-I°-to-C0 movement, in so far as the "clitic" must be stranded in 1° (section 4.2), in addition to it making rather unfortunate predictions concerning adjunction to IP (in non-V2 languages) and to CP (in V2 languages) (section 2.4). Turning now to the V2-inside-IP approach (Diesing 1988, 1990; Rognvaldsson and Thrainsson 1990; Heycock and Santorini 1992), we have similarly tried to show that it is indeed not so apparent that it is superior to the V2-outside-IP approach, as argued by its proponents: a. Data concerning extraction out of embedded V2 clauses in Yiddish (section 3.1) as well as the facts about topicalization in relative clauses and in embedded questions in both Yiddish and Icelandic (section 3.4) are not completely captured under either approach. b. We have argued, nevertheless, that the Yiddish and Icelandic facts concerning the positioning of adverbials, especially in relation to the position of the subject in non-subject-initial V2 clauses (section 3.2), subject-verb agreement (section 3.3), and inversion with a topicalized element (section 3.5) can be satisfactorily accounted for only under the V2-outside-IP approach. In section 4, arguments concerning the position of 1° in German and Dutch were presented: a. One favors the order I°-VP, namely, the data concerning unstressed pronouns in Du. b. In spite of these facts remaining essentially unexplained in the V2-outside-IP approach, we pointed out that the account offered by Zwart (1990, 1991) is not without its problems, viz. that the clitic cannot move along with its hosting verb to C°. c. Another argument, concerning the relation between the richness of verbal inflection and verb movement (section 4.1), favors the order VP-I°. Although we have tried to maintain the superiority of what we originally termed the "traditional account" in Schwartz and Vikner (1989), we would like to emphasize that the V2-outside-IP approach also has a number of problems of its own: No really satisfactory solution has been suggested concerning, for example, weak object pronouns (sections 2.2.1 and 2.2.2), or extraction and topicalization in embedded V2 clauses (sections 3.1 and 3.4). Nevertheless, we feel that this analysis is still the one that comes closest to giving a straightforward account of much of the data. In light of these specific and non-trivial shortcomings, it remains an open question whether V2 actually always takes place at the CP-level or whether embedded V2 takes place at the level of a category which contains IP but is itself contained by CP (namely, the "Bigger-Than-IP-But-Smaller-Than-CP" approach . . . ).
52
PARAMETERS AND FUNCTIONAL HEADS
Notes 1. This paper is a completely revised, much altered, and considerably extended version oi? Schwartz and Vikner (1989). We would like to thank Markus Bader, Maria Beck, Anna Cardinaletti, Noam Chomsky, Kathrin Cooper, Molly Diesing, Lynn Eubank, Giuliana Giusti, Christine Haag-Merz, Liliane Haegeman, Hubert Haider, Arild Hestvik, Teun Hoekstra, Hans Kamp, Jim McCloskey, Ad Neeleman, Christer Platzack, Luigi Rizzi, Ian Roberts, Ramona Romisch-Vikner, Beatrice Santorini, Manuela Schonenberger, Halldor SigurSsson, Rex Sprouse, ThiloTappe, Hoskuldur Thrainsson, AlessandraTomaselli, Lisa Travis, Heike Zinsmeister, and Jan-Wouter Zwart. Of course are all errors still our own fault. 2. Alternatives to C° as the head that selects IP include, for example, the F° of Tsimpli (1990:246), the Agrl" of Roberts (1993, section 1.4) and Cardinaletti and Roberts (1991), or the Topic" of Muller and Sternefeld (1993:485). 3. It should be noted that the topics covered in this paper are not intended nor claimed to be a complete discussion of the merits of the V2-outside-IP account. For additional reasons to prefer the traditional account of V2, see, for example, Holmberg (1986), Giusti (1991), Tomaselli (1990b), and Johnson and Vikner (1994). 4. An exception to this general claim is the position of the verb in questions in residual V2 languages like English (En.) and French (which we do not address here—but see Rizzi 1990b:377). There is a real asymmetry with respect to the verb positions in English questions: Only in subject questions like who saw an accident? is a form of do not required, as opposed to non-subject questions like what did you see? or where did you see_ an accident?. 5. If adjunction to IP is possible (as shown here for German and Swedish), we have a reason to prefer the conditions on proper government of Rizzi's (1990a) Relativized Minimality framework over those of Chomsky's (1986) Barriers framework (see also the discussion in section 3.1). As we will show in section 2.3.1, extraction of an adjunct out of an embedded clause in German is impossible unless there is an intermediate trace in the embedded CP-spec. If adjunction to IP is possible, then the Barriers framework cannot prevent adjunct-extractions from adjoining to the embedded IP. Then, however, even extractions across a filled CP-spec are predicted to be grammatical (CP would not inherit barrierhood from IP, since IP would not be a blocking category), though this is clearly not a desirable prediction (cf. 30b and 34b in section 2.3.1). In the Relativized Minimality framework, the possibility for adjoining to IP makes no difference; the extraction still has to go across CP-spec, which still is a typical potential antecedent governor of the relevant type (A-bar). Thus in the chain there will be a trace that is not properly governed (either the one adjoined to the embedded IP, or the one adjoined to the embedded VP) and the relevant examples are predicted to be ungrammatical, which is the correct prediction (see section 2.3). 6. Haegeman (1991:52-55) points out that the analysis of Zwart (1990, 1991) and, by extension, also the one of Travis (1986, 1991), is not compatible with the data concerning cliticization in West Flemish. She shows that regardless of whether the initial element in a V2 clause is an unstressed subject pronoun or unstressed object pronoun, the same number of functional heads need to be posited in order to accommodate all of the cliticization possibilities. If there are the same number of functional heads available in the two types of clause, Haegeman argues, the finite verb must occupy the same position in each, which is incompatible with the "asymmetry" hypotheses of Zwart and Travis (but compatible with the other analyses discussed below).
THE VERB ALWAYS LEAVES IP IN V2 CLAUSES
53
7. This argumentation can be reproduced with respect to other weak pronouns in the two languages: e.g., Danish d'n, the weak form of den 'it' (common gender), and Norwegian 'n, the weak form of ham 'him'. 8. All the examples in 16-21 below (including those that are ungrammatical) will be fully grammatical if the weak pronoun is replaced either by its corresponding strong form or by a full NP. Thus a can be replaced by hun 'she', in 16-19 and by henne 'her', in 20-21 or by tante Sofie 'Aunt Sofie', in all of 16-21. The same applies to 'd, which can be replaced by del 'it', or by del herfjernsyn 'this TV set', in all of 16-21. 9. The word order in 17b and 18b, where negation or sentence adverbials precede the finite verb, is always an option and sometimes the only option in embedded clauses in Danish (and Norwegian and Swedish). The word order in 17a and 18a, where the finite verb precedes negation or sentence adverbials, could in theory be analyzed either as embedded V2 or as an instance of V(l-to-I° movement. Two facts support the former interpretation: The word order in 17a and 18a is not always possible (whereas the other one is, leading to the conclusion that Danish, Norwegian, and Swedish are languages which lack V°-to-I° movement), and whenever the word order in 17a and 18a is possible, then embedded topicalization, which must be taken to be cases of embedded V2, is possible as well (cf. Platzack 1986a, 1986b; Vikner 1994a, 1994c; and references therein). 10. In view of the examples to be discussed below, this rule could be generalized to cover all occurrences of al'd (and other weak pronouns): al'd must cliticize to the right of an X° which c-commands it. In addition to this syntactic rule, there is also a phonological constraint on the occurrence of Danish 'd: The word to which 'd cliticizes phonologically must end in a vowel (as observed by Jensen 1986:92). In spite of the Danish orthography, this constraint is not violated in any of our examples. 11. Thanks to Arild Hestvik for providing those Norwegian examples which have not been taken from Christensen (1984). 12. Furthermore, Zwart's proposal would also not be able to account for the fact that initially in a main clause neither unstressed subject nor unstressed object pronouns are possible in Norwegian and Danish (cf. 16), as opposed to Dutch (cf. 11). The unstressed subject pronouns should have moved to IP-spec (and not to CP-spec) and therefore be able to cliticize to the empty C° (cf. also 19b, c, which show that cliticization to an empty C" is possible). 13. It should be noted that 22b, 23b, and 24b all have grammatical readings: 22b is fine as a question; 23b and 24b are fine either as questions or as so-called VI declaratives. VI declaratives are restricted to narratives and similar kinds of contexts (cf. e.g., SigurSsson 1990 for Icelandic and Santorini 1989:68 for Yiddish). Also, please note that we are consciously avoiding the rather thorny issue of the analysis of embedded clauses in Icelandic and Yiddish, where the expletive is allowed to the right of the complementizer, as opposed to embedded clauses in German. For a discussion of embedded clauses in Icelandic and Yiddish in general, see section 3; and for a discussion of the interaction between expletives and embedded clauses in Icelandic and Yiddish which is compatible with the proposals of this paper, see Vikner (1994c, ch. 4). 14. Notice that the ungrammatical 25b and 26b cannot be ruled out simply by saying that the expletive cannot occur in IP-spec, as this is not the case (in contrast to German, Yiddish, and Icelandic): Da. a. Sad der en fugl pa taget? Sw. b. Satt det en ragel pa taket? Ge. c. *SaB es ein Vogel auf dem Dach? Sat there/it/it a bird on the roof?
(Platzack 1983:85, ex. 9b)
54
PARAMETERS AND FUNCTIONAL HEADS
Notice also that whether the language makes the distinction between it and there, as Danish and Dutch do, or whether the form it is used for both, as in Norwegian and Swedish, plays no role. 15. The examples in 30 are adjunct-extractions, with the base-generated position of the adjunct being left-adjoined to the embedded VP. Note that the same results are obtained when we extract the subject (i), or the object (ii): (i) Ge. a. Welches Kind glaubte sie hatte dieses Brot gegessen b. *Welches Kind glaubte sie dieses Brot hatte gegessen Which child thought she (had) this bread (had) eaten (ii) Ge. a. Welches Brot glaubte sie hatte das Kind gegessen b. ^Welches Brot glaubte sie das Kind hatte gegessen Which bread thought she (had) the child (had) eaten As for the distinction between argument and non-argument extraction and for why argument extraction is subject only to subjacency requirements whereas non-argument extraction is subject both to subjacency and to the ECP, see the discussion of 51 and 52 in section 3.1. 16. An intermediate trace in CP-spec is necessary whether one adopts the conditions on proper government in Chomsky (1986) or those in Rizzi (1990a). In Chomsky's Barriers framework, there cannot be proper government across both an IP and a CP, as the CP would then be a barrier, inheriting its barrierhood from IP. In Rizzi's Relativized Minimality framework, the filled CP-spec is a 'typical potential antecedent governor' of the relevant type (i.e., A-bar), and thus in order for the trace adjoined to the embedded VP (or to the embedded IP, cf. n. 5) to be properly governed, this CP-spec position must contain an antecedent for the trace. As these are adjunct-extractions, subject to the ECP (cf. section 3.1), the conditions are that each link of the extraction chain properly govern the next one. (Note also that we have omitted the intermediate trace adjoined to the matrix VP in all of these examples.) 17. In other words, each link in the chain (including the trace adjoined to the matrix VP, which we have omitted) properly governs the next one, as no barriers intervene (only IPs) with respect to Chomsky (1986) or as no typical potential antecedent governors intervene (because of the absence of CP-spec) with respect to Rizzi (1990a). 18. In order to see how this works, let us briefly review Travis' version of proper government: A properly governed head must remain empty, i.e., nothing can move into it, because it is filled in some sense by features. To put this in Travis' (1986:12, 18, 1991:351, 357) terms, the head is identified by proper government, and in this way it receives features which must remain recoverable, thus preventing anything moving into this position (cf. also n. 11 in Schwartz and Tomaselli 1990:270). This is how complementizers like dap prevent V°-to-I° movement from taking place. Another kind of proper government concerns complements. By definition (Travis 1991:351, ex. 22a), the (head of a) complement of a is properly governed by a. Therefore, if CP is the complement of glauben, the embedded verb necessarily cannot move into the properly governed C". In fact, this same line of argumentation applies to the first proposal discussed above concerning IP being the complement of glauben, as pointed out to us by Molly Diesing (personal communication). As IP is a complement to the matrix verb, then 1° should be properly governed — and hence should not be able to be filled (on a par with complementizers properly governing the empty 1°). This incorrectly predicts 28b to be grammatical. One would therefore have to conclude that if glauben should be able to take an IP as complement, i" for some reason would not count as being properly governed by glauben (cf. 28a). 19. Teun Hoekstra has pointed out to us that 30 and 34 might not be interpreted as extractions out of embedded clauses; instead they might have an alternative interpretation under which
THE VERB ALWAYS LEAVES IP IN V2 CLAUSES
55
glaubte sie 'believed she', is a so-called parenthetical, inserted between the matrix CP-spec and the matrix C". If so, then the examples would not support our argument, as 30b and 34b would now be straight forward violations of the V2 constraint: The finite matrix verb hatte would not have moved to C°, though it should have, as nothing else occurs in C". There are, however, at least two reasons to reject a parenthetical analysis of the examples in 30 and 34 (cf. also Tappe 1981). One is that the judgments (of both 30 and 34) are the same with more complicated matrix clauses: (i) Ge.
Womit hast du mir gestern gesagt,... What-with have you me yesterday told . . . a. . . . hatte das Kind dieses Brot gegessen b. *... das Kind hatte dieses Brot gegessen . . . (the child) had (the child) this bread eaten
Here the parenthetical analysis is unlikely, as the parenthetical would consist of hast du mir gestern gesagt 'have you me yesterday told'. Another argument can be made on the basis of examples containing bound variables: (ii) Ge. An welcher Universitat wiirde jeder Linguist sagen wiirde er am liebsten arbeiten? At which university would every linguist say would he preferably work? Bound variables have to be c-commanded by their antecedent, as opposed to the E-type pronouns of Evans (1980:339-340). It is clear that the pronoun in (ii) is a bound variable rather than an E-type pronoun as the quantifier antecedent may be negative. (iii) Ge. An welcher Universitat wiirde kein Linguist sagen wiirde er am liebsten arbeiten? At which university would no linguist say would he preferably work? Er in (ii) and (iii) clearly has jeder Linguistlkein Linguist as an antecedent, and therefore jeder Linguistlkein Linguist must c-command er. This in turn means that jeder Linguistlkein Linguist are part of the matrix clause and cannot be inside a parenthetical. 20. As far as we can tell, "an operator" simply means an element in an A-bar-position. 21. Zwart (1991:75, n. 3) assumes an empty complementizer is present in embedded questions (i), and in relative clauses (ii), since CP-spec is filled but verb movement to C° is impossible: (i) Ge. a. *Ich frage mich was hat Peter gekauft b. Ich frage mich was Peter gekauft hat I ask myself what (has) Peter bought (has) (ii) Ge. a. *Die Zeitung, die hat Peter gekauft, war teuer b. Die Zeitung, die Peter gekauft hat, war teuer The newspaper, that (has) Peter bought (has), was expensive 22. It should be added here that the fact that 43c is grammatical is also a problem under Travis' version of the asymmetry approach, as 1° in 43c must be empty, but there is nothing that could possibly properly govern 1° ('fill it with features') and therefore 43c should violate the ECP (twice, in fact, as C" is also empty). For more discussion of the position of 1° in German and Dutch, see section 4. 23. Discussion of what differentiates V2 from non-V2 languages is greatly reduced in Zwart (1991) as compared to Zwart (1990). In the more recent version (1991:76), he merely states that English is an IP-oriented language whereas Dutch (and German) is a CP-oriented language. For this reason, we rely on Zwart's earlier paper. 24. They may be analyzed as results of the Negative Criterion, cf. Rizzi (1991b, section 5) and Haegeman and Zanuttini (1991:244), which is a biconditional principle to the effect that a
56
PARAMETERS AND FUNCTIONAL HEADS
negative XP-element must be in a Spec-X0 agreement relation with a head which has the feature [+neg], i.e., the verb in 46,47, and 48. 25. Zwart (1991:76) follows Roster's (1978) analysis of non-subject-initial V2 clauses: Topicalization is adjunction of an XP to CP, accompanied by a co-indexed pronoun in CP-spec, a so-called "d-word", which may be empty. 39 then applies, forcing the verb to move to C°. One might expect that this would rule out 49a, as de boeken 'the books', would then be adjoined to CP rather than in CP-spec. However, as nothing should prevent multiple adjunction to CP (notice that multiple adjunction to IP is possible in both German and Swedish), this still does not explain why 49a is impossible. 26. In some respects, the position of Reinholtz (1989) and of Santorini (1989) may be considered a notational variant of the V2-outside-IP analysis (their IP-spec always corresponds to our CP-spec; their VP-spec always corresponds to our IP-spec), and therefore not all of the arguments below will actually be problematic for their position. However, Heycock and Santorini (1992), discussed in section 3.5, cannot be considered a notational variant, as they suggest that while topicalized elements occur in IP-spec, w/i-elements occur in CP-spec; the verb therefore is in C° in questions but in 1° in topicalizations and other declarative clauses. 27. Molly Diesing and Beatrice Santorini have each pointed out that many speakers find 52 either acceptable or not nearly as unacceptable as a that-tta.ee violation like the one in 54c. This is problematic in so far as we would predict 52 and 54c to violate the same condition, the ECP (though two different parts of the ECP: 52 violates the restriction on antecedent government, 54c the one on head government). If 52 should turn out to be good in Yiddish (and/or in Icelandic, for example), one possible analysis could be that the language in question has a difference between topicalization and w/z-movement, so that the two kinds of movement cannot interfere with one another (cf. the suggestions of Miiller and Sternefeld 1993:484), much like A-movement and A-bar-movement cannot interfere with one another in the Relativized Minimality framework of Rizzi (1990a). 28. Other languages have also been claimed to allow violations of subjacency. The best-known examples are perhaps the Mainland Scandinavian languages, as discussed in the papers in Engdahl and Ejerhed (1982). A characteristic example is the following (Engdahl 1982:152, ex. 4): (i) Sw. Nobelpriset i medicin: ska vi snart fa reda pa vem; som tj har fatt tj Nobel-prize-the in medicine shall we soon get clarity as-to who that has got (= the Nobel prize for medicine, we will soon know who has received) 29. The analysis suggested for 54c and 54d will also extend to Danish and German, if we make the two following (independently empirically motivated) assumptions: (i) a. Embedded V2 is only possible in Danish if the complementizer 'at' is present, b. Embedded V2 is only possible in German if the complementizer 'daB' is not present. (iia) and (iiia) correspond to 54c; (iib) and (iiib) to 54d: (ii) Da. a. *Hvad sagde han at skulle vi k0be b. *Hvad sagde han skulle vi k0be What said he (that) should we buy?
? ?
(iii) Ge. a. *Was hat er gesagt daB _ sollten wir kaufen b. Was hat er gesagt sollten wir kaufen ? What has he said (that) should we buy?
7
THE VERB ALWAYS LEAVES IP IN V2 CLAUSES
57
(lib) and (iiia) are ruled out by (ia, b) above, (iia) is ruled out because the trace before skulle is not properly governed (by at), just like 54c. (iiib) is possible, the trace before sollten being properly governed (by gesagt), just like 54d (see also the discussion in section 2.3.1). Neither (ia) nor (ib) holds for Yiddish (cf. the quote below from Diesing 1990:75). 30. Above we bracketed 54e corresponding to 55c, even though Diesing's (1990:72, ex. 49e) bracketing corresponds to 55a. 31. This is therefore a parallel to the so-called was ... fur split in other Germanic languages (cf. e.g., den Besten 1984:34-39; Corver 1991; Vikner 1994c, section 2.4). 32. Thanks to Halldor Sigur5sson for pointing out the different interpretations of oft in 67. 33. Versions of X-bar-Theory exist in which a VP-internal subject may precede a VP-adjoined adverbial. In Koopman and Sportiche (1991:212) and Sportiche (1988:425), a VP structure is suggested which would allow an adverbial to occur between the subject and V° without necessitating adjunction to V-bar:
According to Koopman and Sportiche (1991:212), Vmax, which is "a small clause whose predicate is V P . . . "the maximal projection" of V", whereas VP is "the phrasal projection" of V". This gives two possibilities for the position of the adverbial, neither of which presupposes adjunction to V-bar: Either the adverbial is in VP-spec (as opposed to Vma*-spec which is occupied by the subject) or it is adjoined to VP. Notice that though VP-adjunction is adopted by Sportiche (1988:432), it is only suggested for a manner adverbial like French soigneusement 'carefully'. In fact, Sportiche assumes that a sentential adverbial would have to be "adjacent (adjoined) to I" and thus to the left of Vmax. The option of positing a structure like (i) in order to save the idea that the adverbial is generated between the subject and the verb in sentences like 61a, c is thus unavailable, since the position of the adverbial in (i) is feasible only for non-sentential adverbials. 34. The potential weakness of this argument is that it presupposes that objects in Yiddish are generated to the right of the verb, an assumption that is generally made but also questioned (cf. den Besten and Moed-van Walraven 1986 and GeilfuG 1991). 35. Potentially problematic in this approach is that in Icelandic sentences with a nominative object and a non-nominative subject, the finite verb agrees with the object rather than the subject. However, as argued by Jonsson (1991:24-26) and by Vikner (1994c, section 4.6), for example, this kind of agreement is different from the standard subject-verb agreement, as it only takes place in the third person and as it also seems to be optional. 36. In other words, one might wonder about the naturalness of the disjunctive ways in which finiteness features in 1° are supposedly licensed (cf. 40a, b). 37. For other arguments that at least one functional head should follow VP, see Giusti (1991) on infinitivals. 38. There is no direct German parallel to the construction in 93. 39. It should be pointed out that for Zwart, V°-to-l" movement could not possibly be forced by 1° morphologically subcategorizing for V", as V"-to-I° movement is only motivated by the need to license the finiteness features in 1°, and there is an alternative way for these features to be licensed: by a lexicalized C° (cf. Zwart 1991:85 and ex. 40 in section 2.3.2).
58
PARAMETERS AND FUNCTIONAL HEADS
40. A further complication is that in French questions (e.g., 96), the subject pronoun does not stay in IP-spec but cliticizes to C°, as shown by the indexing. For more discussion of this and other details of this construction, see Rizzi and Roberts 1989. 41. It should also perhaps be pointed out that it is Kayne's (1991) analysis of the French clitics (which explictly relies on Baker's (1988) rule) that serves as the fundamental basis for Zwart's proposals about the Dutch reduced pronouns. Hence, it becomes somewhat questionable (not to mention ironic) to have to conclude that accepting Zwart's extension of Kayne's idea in order to account for the Dutch pronouns seems to necessitate abandoning the analysis of the French pronouns as cliticization to 1°. 42. Ad Neeleman (personal communication) has pointed out that the data concerning the possible positions of reduced pronouns in Dutch are more complicated than depicted in Zwart (1990, 1991). Recall that it is on the basis of data like 92b and 93b where the reduced form of the pronoun occurs immediately to the right of the subject (in embedded clauses) — coupled with the assumption that clitics must cliticize to a functional head — that Zwart concludes that there must be a functional head to the right of IP-spec and the left of VP. However, Neeleman notes the grammaticality of the following as well: (i) Du, a. ... dat Jan gisteren '_t eindelijk gekocht heeft . . . that Jan yesterday it finally bought has b. . . . dat Jan 't gisteren 'm eindelijk gegeven heeft . . . that Jan it yesterday him finally given has As a supplement to the data presented in Zwart (1991), the data in (i) show that there are minimally two positions for reduced pronouns in Dutch and moreover that 't in (ia) and 'm in (ib) could not possibly be in the head (i.e., 1°) of the projection whose specifier is occupied by Jan (i.e., IP-spec), since both 't in (ia) and 'm in (ib) follow the adverb gisteren (for related discussion, see also Haegeman 1991:520-555).
References Baker, M. 1988. Incorporation: A Theory of Grammatical Function Changing, Chicago: University of Chicago Press. Baltin, M. 1982. "A Landing Site Theory of Movement Rules," Linguistic Inquiry 13:1-38. Belletti, A. 1988. "The Case of Unaccusatives," Linguistic Inquiry 19:1-34. . 1990. Generalized Verb Movement. Aspects of Verb Syntax. Turin: Rosenberg and Sellier. den Besten, H. 1977/83. "On the Interaction of Root Transformations and Lexical Deletive Rules," ms, University of Amsterdam. Published (1983) in W. Abraham (ed.), On the Formal Syntax of the Westgermania. Amsterdam: John Benjamins. 47-131. [Also part of den Besten 1989.] . 1984. "The Ergative Hypothesis and Free Word Order in Dutch and German," in J. Toman (ed.), Studies in German Grammar. Dordrecht: Foris. 23-64. [Also in den Besten 1989.] . 1989. Studies in West Germanic Syntax. Amsterdam: Rodopi. den Besten, H. and C. Moed-van Walraven. 1986. "The Syntax of Verbs in Yiddish," in H, Haider and M. Prinzhorn (eds.), Verb Second Phenomena in Germanic Languages. Dordrecht: Foris. 111-135. Brcckenridge, J. 1975. "The Post-Cyclicity of e.v-Insertion in German," in R.E. Grossman, L.J. San and T.J. Vance (eds.), Papers from the llth Regional Meeting of the Chicago Linguistics Society. 81-91.
THE VERB ALWAYS LEAVES IP IN V2 CLAUSES
59
Cardinaletti, A. 1990a. Impersonal Constructions and Sentential Arguments in German. Padua: Unipress. . 1990b. "Es, pro and Sentential Arguments in German," Linguistische Berichte 126: 135-164. Cardinaletti, A. and I. Roberts. 1991. "Clause Structure and X-Second," ms, University of Venice and University of Geneva. [To appear in W. Chao and G. Horrocks (eds.), Levels of Representation. ] Chomsky, N. 1986. Barriers. Cambridge, Mass.: MIT Press. . 1990. Class lecture notes, MIT. . 1991. "Some Notes on the Economy of Derivation and Representation," in R. Freidin (ed.), Principles and Parameters in Comparative Grammar. Cambridge, Mass.: MIT Press. 417-454. Christensen, K. Koch. 1984. "Subject Clitics and A'-Bound Traces," Working Papers in Scandinavian Syntax 15:1-31. Corver, N. 1991. "The Internal Syntax and Movement Behavior of the Dutch 'wat voor'Construction," Linguistische Berichte 133:190-228. Diesing, M. 1988. "Word Order and the Subject Position in Yiddish," Proceedings ofNELS 18. 124-140. . 1990. "Verb Movement and the Subject Position in Yiddish," Natural Language and Linguistic Theory 8:41-79. Emonds, J. 1978. "The Verbal Complex of V'-V in French," Linguistic Inquiry 9:151-175. Engdahl, E. 1982. "Restrictions on Unbounded Dependencies in Swedish," in E. Engdahl and E. Ejerhed (eds.), Readings on Unbounded Dependencies in Scandinavian Languages. Stockholm: Almquist and Wiksell. 151-174. Engdahl, E. and E. Ejerhed (eds.) 1982. Readings on Unbounded Dependencies in Scandinavian Languages. Stockholm: Almquist and Wiksell. Evans, G. 1980. "Pronouns," Linguistic Inquiry 11:337-362. GeilfuB, J. 1991. "Jiddisch als SOV-Sprache," Working Papers of Sonderforschungsbereich 340 11:3-17. [Universities of Stuttgart and Tubingen.] Giusti, G. 1991. "ZK-Infinitivals and Sentential Structure in German," Rivista di Linguistica 3:211-234. Haegeman, L. 1991. "On the Relevance of Clitical Placement for the Analysis of SubjectInitial Verb Second in West Flemish," Groninger Arbeiten zur Gemanistischen Linguistik 24:29-66. Haegeman, L. and R. Zanuttini. 1991. "Negative Heads and the Neg Criterion," The Linguistic Review 8:233-251. Heycock, C. and B. Santorini. 1992. "Head Movement and the Licensing of Non-Thematic Positions," ms, Oakland University and Northwestern University. [Paper presented at WCCFL11.] Holmberg, A. 1986. Word Order and Syntactic Features in the Scandinavian Languages and English. Stockholm: Department of General Linguistics, University of Stockholm. Holmberg, A. and C. Platzack. 1988. "On the Role of Inflection in Scandinavian Syntax," Working Papers in Scandinavian Syntax 42:25-42. Jaspers, D. 1989. "A Head Position for Dutch Clitics," in D. Jaspers, W. Klooster, Y. Putseys, and P. Seuren (eds.), Sentential Complementation and the Lexicon. Dordrecht: Foris. 241-252. Jensen, P. Anker. 1986. "Syntaksogufonologi," CEBAL 8:90-115. Copenhagen: NytNordisk Forlag Arnold Busck. Johnson, K. and S. Vikner. 1994. "The Position of the Verb in Scandinavian Infinitives," Working Papers in Scandinavian Syntax 53:61-84.
60
PARAMETERS AND FUNCTIONAL HEADS
Jonsson, J. Gi'sii. 1991. "Stylistic Fronting in Icelandic," Working Papers in Scandinavian Syntax 48:1-44. Kayne, R.S. 1991. "Romance Clitics, Verb Movement, and PRO," Linguistic Inquiry 22:647686. Koopman, H. 1984. The Syntax of Verbs. Dordrecht: Foris. Koopman, H. and D. Sportiche. 1991. "The Position of Subjects," Lingua 85:211-258. Koster, J. 1978. "Why Subject Sentences Don't Exist," in S. Jay Keyser (ed.), Recent Transformational Studies in European Languages. Cambridge, Mass.: MIT Press. 53-64. Lasnik, H. and M. Saito. 1984. "On the Nature of Proper Government," Linguistic Inquiry 15:235-255. . 1992. Move a. Cambridge, Mass.: MIT Press. Lowenstamm, J. 1977. "Relative Clauses in Yiddish: A Case for Movement," Linguistic Analysis 3:197-216. Miiller, G. and W. Sternefeld. 1993. "Improper Movement and Unambiguous Binding," Linguistic Inquiry 24:461-507. Na'f, A. 1987. "Gibt es Exklamativsatze?" in J. Meibauer (ed.), Satsmodus zwischen Grammatik und Pragmatik. Tubingen: Niemeyer. 140-160. Obenauer, H.-G. 1976. Etudes de syntaxe interrogative dufrancais. Tubingen: Niemeyer. . 1984. "On the Identification of Empty Categories," The Linguistic Review 4:153-202. Platzack, C. 1983. "Existential Sentences in English, Swedish, German and Icelandic," in F. Karlsson (ed.), Papers from the Seventh Scandinavian Conference of Linguistics. 80-100. . 1986a. "COMP, INFL, and Germanic Word Order," in L. Hellan and K. Koch Christensen (eds.), Topics in Scandinavian Syntax. Dordrecht: Reidel. 185-234. . 1986b. "The Position of the Finite Verb in Swedish," in H. Haider and M. Prinzhorn (eds.), Verb Second Phenomena in Germanic Languages. Dordrecht: Foris. 27-47. . 1987. "The Scandinavian Languages and the Null-Subject Parameter," Natural Language and Linguistic Theory 5:377-401. Pollock, J.-Y. 1989. "Verb Movement, Universal Grammar, and the Structure of IP," Linguistic Inquiry 20:365-424. Reinholtz, C. 1989. "V-2 in Mainland Scandinavian: Finite Verb Movement to Agr," Working Papers in Scandinavian Syntax 44:101-117. Reuland, E. and A. ter Meulen (eds.). 1987. The Representation of (In)defmiteness. Cambridge, Mass.: MIT Press. Rizzi, L. 1990a. Relativized Minimality. Cambridge, Mass.: MIT Press. . 1990b. "Speculations on Verb Second," in J. Mascara and M. Nespor (eds.), Grammar in Progress. GLOW Essays for Henk van Riemsdijk. Dordrecht: Foris. 375-386. . 199la. "Proper Head Government and the Definition of A-Positions." Talk at GLOW 14, Leiden. [Abstract in GLOW Newsletter 26:46-47.] . 1991b. "Residual Verb Second and the Wh-Criterion." Technical Reports in Formal and Computational Linguistics, no. 2. University of Geneva. [Also published in this volume, 63-90.] . 1992. "Early Null Subjects and Root Null Subjects," ms, University of Geneva. Rizzi, L. and I. Roberts. 1989. "Complex Inversion in French," Probus 1:1-30. [Also published in this volume, 91-116.] Roberts, I. 1991. "Excorporation and Minimality," Linguistic Inquiry 22:209-218. . 1993. Verbs and Diachronic Syntax. Dordrecht: Kluwer. Rognvaldsson, E. 1984. "Icelandic Word Order and /wi5-Insertion," Working Papers in Scandinavian Syntax 8:1-21.
THE VERB ALWAYS LEAVES IP IN V2 CLAUSES
61
Rognvaldsson, E. and H. Thrainsson. 1990. "On Icelandic Word Order Once More," in J. Maling and A. Zaenen (eds.), Modern Icelandic Syntax (Syntax and Semantics 24). San Diego: Academic Press. 3-40. Santorini, B. 1989. "The Generalization of the Verb-Second Constraint in the History of Yiddish," Ph.D., University of Pennsylvania. Schwartz, B.D. and A. Tomaselli. 1990. "Some Implications from an Analysis of German Word Order," in W. Abraham, W. Kosmeijer and E. Reuland (eds.), Issues in Germanic Syntax (Trends in Linguistics, Studies and Monographs 44). Berlin: Mouton de Gruyter. 251-276. Schwartz, B.D. and S. Vikner. 1989. "All Verb Second Clauses are CPs," Working Papers in Scandinavian Syntax 43:27-49. Sigurdsson, H. Armann. 1989. "Verbal Syntax and Case in Icelandic," Ph.D., University of Lund. . 1990. "VI Declaratives and Verb Raising in Icelandic," in J. Maling and A. Zaenen (eds.), Modern Icelandic Syntax (Syntax and Semantics 24). San Diego: Academic Press. 41-69. Sportiche, D. 1988. "A Theory of Floating Quantifiers and Its Corollaries for Constituent Structure," Linguistic Inquiry 19:425-449. Tappe, T. 1981. "Wer glaubst du hat recht — Einige Bemerkungen zur COMP-COMPBewegung im Deutschen," in M. Kohrt and J. Lenerz (eds.), Sprache: Formen und Strukturen, Aktendes 15. Linguistischen Kolloquiums. Tiibingen: Niemeyer. 203-212. Taraldsen, T. 1986a. "On Verb Second and the Functional Content of Syntactic Categories," in H. Haider and M. Prinzhorn (eds.), Verb Second Phenomena in Germanic Languages. Dordrecht: Foris. 7-25. . 1986b. "Som and the Binding Theory," in L. Hellan and K. Koch Christensen (eds.), Topics in Scandinavian Syntax. Dordrecht: Reidel. 149-184. Thiersch, C. 1978. "Topics in German Syntax," Ph.D., MIT. Thrainsson, H. 1979. On Complementation in Icelandic. New York: Garland. . 1986. "VI, V2, V3 in Icelandic," in H. Haider and M. Prinzhorn (eds.), Verb Second Phenomena in Germanic Languages. Dordrecht: Foris. 169-194. Tomaselli, A. 1990a. "COMP° as a Licensing Head: An Argument Base on Cliticization," in J. Mascaro and M. Nespor (eds.), Grammar in Progress. GLOW Essays for Henk van . Riemsdijk. Dordrecht: Foris. 433-445. ——. 1990b. La sintassi del verbofinito nelle lingue germaniche. Padua: Unipress. Travis, L. 1984. "Parameters and Effects of Word Order Variation," Ph.D., MIT. . 1986. "Parameters of Phrase Structure and V2 Phenomena," ms, McGill University. [Presented at the Princeton Workshop on Comparative Syntax, March 1986.] . 1991. "Parameters of Phrase Structure and Verb Second Phenomena," in R. Freidin (ed.), Principles and Parameters in Comparative Grammar. Cambridge, Mass.: MIT Press. 339-364. Tsimpli, I. 1990. "The Clause Structure and Word Order of Modern Greek," in J. Harris (ed.), UCL Working Papers in Linguistics, vol. 2. London: University College London. 228-255. Vikner,S. 1989. "Object Shift and Double Objects in Danish," Working Papers in Scandinavian Syntax 44:141-155. . 1990. "Verb Movement and the Licensing of NP-Positions in the Germanic Languages," Ph.D., University of Geneva. . 1994a. "Finite Verb Movement in Scandinavian Embedded Clauses," in N. Hornstein and D. Lightfoot (eds.), Verb Movement. Cambridge: Cambridge University Press. 117-147.
62
PARAMETERS AND FUNCTIONAL HEADS
. 1994b. "Scandinavian Object Shift and West Germanic Scrambling," in N. Corver and H. van Riemsdijk (eds.), Studies on Scrambling. Berlin: de Gruyter. 487-517. . 1994c. "Verb Movement and Expletive Subjects in the Germanic Languages," ms, University of Stuttgart. [To appear. New York: Oxford University Press.] Vikner, S. and R.A. Sprouse. 1988. "Have/Be Selection as an A-Chain Membership Requirement," Working Papers in Scandinavian Syntax 38:1-48. Zwart, J.-W. 1990. "Clitics in Dutch: Evidence for the Position of Infl," ms, University of Groningen. . 1991. "Clitics in Dutch: Evidence for the Position of Infl," Groninger Arbeiten zur Germanistischen Linguistik 33:71-92.
2
Residual Verb Second and the Wfo-Criterion Luigi Rizzi
Some natural languages do not allow the subject to intervene between the w/z-element and the inflected verb in main questions. This constraint is illustrated by English and Italian: (1) a. * What Mary has said? b. *Che cosa Maria ha detto? The two languages apparently use different strategies to avoid the forbidden sequence: English preposes the inflected Aux, Italian (like Spanish, Catalan, Romanian, etc.) uses zero realization or postposing of the subject: (2) What has Mary said? (3) a. Che cosa ha detto? 'What has said?' b. Che cosa ha detto Maria? 'What has said Maria?' Two questions arise: 1. What excludes the forbidden sequence? 2. Are the English and Romance salvaging strategies as different as they look? Focusing initially on the first question and on the English case, we assume that the required adjacency between wh- and I is to be expressed in terms of Chomsky's (1986) approach to the structure of clauses: in main questions, I-to-C movement must apply, and create a Spec-head configuration involving the w/z-element and the inflected verb. Subject-auxiliary inversion can thus be reduced to a special case of Verb Second (as in den Besten 1983), in turn a particular instance of head-to-head 63
64
PARAMETERS AND FUNCTIONAL HEADS
movement. I will call "residual V2" such construction-specific manifestations of I-to-C movement in a language (like English and the modern Romance languages except Raetho-Romansch) which does not generalize the V2 order to main declarative clauses. Question 1 can then be restated as: What triggers residual V2? I would like to propose that the application of I-to-C movement in this and other similar cases is enforced in order to satisfy the W/z-Criterion, a general well-formedness condition on vv/z-structures, which is also ultimately responsible for the SS distribution and LF interpretation of w/z-operators. Sections 1 and 4 introduce and refine the WhCriterion; sections 2 and 3 show how the application of Subject-Aux inversion in English is enforced by the Wz-Criterion; section 5 extends the analysis to inversion with negative operators; section 6 deals with Subject Clitic Inversion in French along similar lines; in sections 7 and 8 we go back to question 2, and provide an analysis of inversion in Romance interrogatives partially in terms of the Wz-Criterion.
1
The Wh -Criterion
In English, the Spec of Comp of an interrogative clause must be filled by a w/z-element at S-structure, hence the in situ strategy is excluded: (4) *I wonder [ [you saw who] ] Wh-in situ is possible in multiple questions, but the wft-element must stay in an argument position; if it is moved to an A-bar position which is not the appropriate scope position, as the embedded Spec of C in 5b, the structure is excluded: (5) a.
Who believes [ [Mary went where] ]
b. *Who believes [where [Mary went t] ] Following standard practice, we will assume that the complementizer of a question is marked by the feature [+wh]. We can then state the following principle: (6) The W/z-Criterion: X"
A. A w/z-operator must be in a Spec-head configuration with r +wn ] • 11
x i must be in a Spec-head configuration with a wft-operator. B. An f +wn 6 is the updated version of the principle first proposed in May (1985), made compatible with the theory of Comp of Chomsky (1986). Here I develop the analysis sketched outinRizzi(1990b). As the feature [+wh] on a clausal head (most typically a C°) designates the fact that the projection of that head (CP) is a question, the Wi-Criterion simply expresses the fact that at the appropriate level of representation interrogative operators must be in the spec of CPs which are interpreted as questions and, reciprocally, CPs interpreted as questions must have interrogative operators as specifiers. The WTz-Criterion thus requires configurations of the following shape:
RESIDUAL VERB SECOND AND THE WH-CRITERION
65
As a general well-formedness principle on the scope of w/z-operators, 6 can be taken as a criterial condition applying universally at LF. So, in languages lacking syntactic w/z-movement, such as Chinese and Japanese, question operators must be moved in the syntax of LF to satisfy the W7z-Criterion at this level, thus giving rise to ECP and other locality effects that have been much discussed since Huang (1982). On the other hand, it can be argued that the W/z-Criterion applies earlier in other languages. For instance, the impossibility of 4 and 5b can be naturally accounted for through the assumption that in English the Wz-Criterion must be fulfilled at Sstructure. 4 violates clause B of the criterion: The verb wonder selects an embedded question, hence a CP whose C° is marked [+wh]; this C° is not in a Spec-head relation with a w/z-operator at S-structure, in violation of B. As for 5b, it contains a w/z-operator that is not in a Spec-head configuration with a w/z-head (believe selects a declarative, hence a CP whose head is [—wh]), and therefore it violates clause A of the Criterion (we will come back in section 4 to the fact that if the w/z-element remains in an argument position, as in 5a, it does not determine a violation of the Wz-Criterion at S-structure).
2
Subject-Aux Inversion in English
We can now show that the Wz-Criterion provides a simple account of the fundamental paradigm of Subject-Aux inversion in English. We assume that this process involves structure-preserving movement of I to C.1 To simplify matters, let us concentrate on cases of w/z-movement of the direct object, and see how it interacts with I-to-C movement. We have to deal with the following eight representations, depending on whether w/z-movement has taken place or not, whether I-to-C movement has taken place or not, and whether the interrogative is independent or embedded: (8) a. *[ [Mary has seen who] ] b. *[Who [Mary has seen t] ] c. * [has [Mary t seen who]] d.
[Who has [Mary t seen t] ]
(9) a. *I wonder [ [Mary has seen who] ] b. I wonder [who [Mary has seen t] ] c. *I wonder [has [Mary t seen who] ] d. *I wonder [who has [Mary t seen t] ]
66
PARAMETERS AND FUNCTIONAL HEADS
Only two of the possible combinations are well-formed. Let us consider the embedded paradigm first. The verb wonder selects an indirect question, hence an embedded C marked [+wh]. The D-structure representation then is
(10) I wonder r+FC"l w jU [Mary has seen who] ] If nothing happens, the corresponding S-structure is ruled out by clause B of the WhCriterion, as a w/z-clausal head is not in the required configuration with a w/z-operator at S-structure; this accounts for 9a. In 9b w/z-movement has applied, the required Spec-head configuration has been created and the Wz-Criterion is satisfied. 9d is excluded by whatever principle accounts for the root character of I-to-C movement: Rizzi and Roberts (1989) argue that a restrictive enough formulation of the Projection Principle rules out such cases ; alternatively, one may think that the specification [+wh] fills the embedded C, and makes it unavailable as a landing site for I-to-C movement (as any other filled C) .2 9c is ruled out at the same time by clause B of the Mi-Criterion and by whatever principle rules out embedded I-to-C movement. Let us now consider the main clause paradigm, shown in 8. The first question to ask is how the w/z-specification can occur in main clauses. I will assume that this, as well any other substantive feature specification cannot occur "for free" in a structure, and must be licensed somehow. The occurrence of [+wh] in an embedded Comp is determined by a standard licensing device, lexical selection. What about main questions? Of course, the theory of licensing cannot be too demanding: There must be at least a position in a structure whose properties and specifications are independently licensed, i.e., a point which the chain of licensings can be anchored to, and start from. It is natural to assume that such a position can be the main inflection (or one of the main inflectional heads, if some version of the Split Infl hypothesis is adopted, as in Pollock 1989), the head that also contains the independent tense specification of the whole sentence. I would like to propose that among the other autonomously licensed specifications, the main inflection can also be specified as [+wh]. That a verbal inflection can carry such a specification is strongly suggested by the fact that in some natural languages the verb manifests a special morphology in interrogatives (see Clements 1984 for Kikuyu; Chung 1982 for Chamorro; Georgopoulos 1985, 1992 for Palauan; Haik, Koopman and Sportiche 1985 for Moore; Tuller 1985 for Hausa; Haik 1990 for a comparative analysis of these cases; see also Kayne 1984 and Roberts 1993 on interrogative -ti in colloquial French). If we make the assumption that I can carry [+wh], the functional role of Subject- Aux inversion becomes clear: this instance of residual I-to-C movement moves the w/z-specification high enough to allow satisfaction of the Wi-Criterion. Let us see how this system works. If the main I is specified [+wh] , the common D-structure of the different cases of 8 is the following: (11) [C [Mary
seen who]]
If nothing happens, the representation 8a is ruled out by clause B of the Wz-Criterion at S-structure (no w/z-operator in the Spec of inflection); (8)b and c are ruled out for the same reason. 8d is well-formed: I carrying [wh] is moved to C, the w/z-operator is moved to its spec, and the configuration required by the Wz-Criterion is met:
RESIDUAL VERB SECOND AND THE WH-CRITERION
67
(12) [who[+h£sn] [Mary t seen t]] We thus obtain the result that obligatory Subject Aux inversion in interrogatives is enforced by the same principle which is responsible for the distributional and interpretive properties of w/z-operators.3
3
Wh -Movement of the Subject
If the application of do support is a reliable cue that I-to-C movement has applied, we must conclude from the following examples that I-to-C movement cannot apply when a subject is moved (irrelevantly, 13a is possible with emphatic do in I): (13) a. * [who does [t t love Mary] ] b.
[who C [t loves Mary] ]
Actually, two distinct problems arise here: 1. Why is I-to-C movement incompatible with subject movement? 2. Why is I-to-C movement allowed not to apply in 13b without violating the Wz-Criterion? Starting from the first question, three additional cases suggest that I-to-C movement creates a configuration in English which does not licence a subject trace: a. Heavy NP Shift of the subject is impossible in interrogatives (as well as in declaratives; see Rizzi 1990a for discussion): (14) a. Have they left? b. *Have t left all the people who you invited? b. Embedded V2 may be triggered by a proposed negative element (see below for discussion); in that case, object extraction has the flavor of a Wi-Island violation, while subject extraction across the preposed I is distinctly worse: (15) a. I think that never did he help her b. ??The woman who I think that never did he help t c. *The man who I think that never did t help her C. McCloskey (1991) points out that in Hiberno-English I-to-C movement can take place in embedded questions, and familiar subject-object asymmetries arise in cases of extraction (examples adapted from McCloskey 1991:294296; we leave open here the question of why this instance of I-to-C movement is not root): (16) a.
You asked them would they marry him
68
PARAMETERS AND FUNCTIONAL HEADS
b. ??Which one did you ask them would they marry t c. *Which one did you ask them would t marry him In Rizzi (1990a) it is argued that 13a and 14b are ruled out as violations of the proper head government requirement of ECP. Suppose this is stated as follows: (17) t must be head-governed by X° within X-bar (the immediate projection of X) If the derived structure of I-to-C movement in English (and more generally in residual cases of V2) is the following, as argued in Rizzi and Roberts (1989):
then a trace in subject position is not allowed to occur: It is not governed by C° (inert for government in non-V2 languages), and it is governed by 1°, but not within its immediate projection I-bar; hence, the proper head government requirement of 17 is violated. This account of 13a-14b can be immediately extended to cover 16c, 17c (the latter extension is explicitly envisaged by McCloskey 1991). It remains to be determined why the structure not involving I-to-C movement is well-formed, and does not violate the Wz-Criterion. A possible approach would be to assume that the subject does not move at all, hence the representation simply is the following, rather than 13b: (19) [[who gloves Mary]] The subject simply remains in the spec of Infl endowed with the feature [wh], and the W/z-Criterion is satisfied within IP in this case. Some technical problems are raised by this minimal solution: a. If Infl is associated to the lexical verb through affix hopping in English (Chomsky 1957, 1989), and this process involves the whole content of the node, then the feature [wh] too would be lowered into the VP, hence it could not be in a Spec-head configuration with the subject at S-structure. b. The Spec of IP should be allowed to count as an A-bar position in this case. c. There is no obvious position for the variable in 19 (even assuming the "subjectwithin-VP" hypothesis, an empty category in VP internal (or adjoined) position would not receive Case). These technical problems (and, more forcefully, the empirical argument concerning similar French cases discussed in Friedemann 1990) suggest that representation 13b
RESIDUAL VERB SECOND AND THE WH-CRITERION
69
is to be favored (problems b. and c. do not arise, and problem a. is not worse). But why is the Wz-Criterion not violated? I would like to propose that the Wz-Criterion must be interpreted as requiring that the chain of the relevant X° position has the feature [+wh], not necessarily the position itself. Can we build the proper chain in (13)a? 1° and the inflection containing [+wh], lowered to V, are co-indexed and already form a chain. If the subject locally moved to the Spec of C triggers agreement in C (Rizzi 1990a), we obtain the following indexing pattern: (20) [WhOi q [t; Ij love- t4h] Mary] ] Within the standard assumption that agreement is minimally expressed by co-indexation, the subject trace is co-indexed with Infl as well as with who; the latter is co-indexed with C°, hence, by transitivity, C° is co-indexed with 1°. Assuming that two co-indexed positions in a local binding relation can always be put together into a single chain (provided that no independent well-formedness condition is violated: see Rizzi 1986), C° forms a chain with 1° and with the lower inflection containing [+wh]. Hence, the Wz-Criterion is met at S-structure. We must now show that this extension does not overgenerate. Why is the chain option restricted to the local movement of the subject? Consider the pattern of indexation that arises when any other element is moved to Comp, for example, the direct object: (21) Who C° [Mary 1° loves t] In this case, the w/z-operator must be contra-indexed with the subject because of strong cross-over (if they were co-indexed, the variable would also be co-indexed with the subject, and principle C would be violated), hence, by transitivity, C° is contra-indexed with 1° and no chain can be formed. As C° cannot be endowed with the [+wh] feature through chain formation in this case, the only available device to fulfill the Wz-Criterion is I-to-C movement. We then derive the conclusion that Subject-Aux inversion applies obligatorily in all the cases in which the moved element is not the local subject.
4 Functional Definition of Wife-Operators Wh-in situ is impossible in English in single questions; in multiple questions it is possible, provided that one w/z-operator has been moved to Comp: (22) a. *You gave what to whom? b.
What did you give t to whom?
This may suggest, at first sight, that the two clauses of the Wz-Criterion apply asymmetrically in English: B must be fulfilled at S-structure, but A can be delayed until LF. According to this approach, 22a is excluded because clause B is violated at S-structure, as the [+wh] feature in I is not supported by a w/z-operator in its
70
PARAMETERS AND FUNCTIONAL HEADS
Spec. On the other hand, a tvft-operator can be left in situ in 22b if clause A can be delayed until LF, as clause B is fulfilled in the now familiar way (I-to-C movement and w/z-movement of one of the two operators). This corresponds, in essence, to the interpretation of May (1985). I would like to consider a different possibility here. Some empirical reasons have already been mentioned to assume that both clauses of the Wz-Criterion apply at S-structure in English (we will come back to them in a moment). But, if this is so, why is wh-in situ at all possible? I believe the solution is provided by a refinement of the notion "w/z-operator". The needed refinement is independently justified to solve a paradox that w/z-constructions raise for Theta Theory. Consider the following two fairly uncontroversial statements: (23)
i. The Theta-Criterion applies at DS, SS, LF ii. Variables are arguments
Consider also DS (24a) and its SS (24b), as well as the multiple question (25) at SS: (24) a. Mary saw whom b. Who did Mary see t (25) Who t saw whom By 23i, the verb see assigns a theta-role to its object at DS, hence who, the assignee, must be an argument in 24a. By 23i the verb assigns a theta-role at SS, and by 23ii this role is received by the variable in 24b; hence who must be a non-argument here, otherwise there would be one argument too many. Therefore, the same element functions as argument and as non-argument at different levels. The same paradox arises in 25 with two distinct occurrences of the same element at the same level, SS: here the subject theta-role is assigned to the variable, therefore who must be a non-argument; on the other hand, whom is the only possible recipient of the object theta-role, hence it must be an argument. This paradox was noticed in Chomsky (1981:115), and is also discussed in Cinque (1986); I will assume an adapted version of the approach proposed in these references, cutting some corners (see also Brody 1990 for relevant discussion). Suppose that the notion w/z-operator is defined in part in functional terms, in the following manner:4 (26) w/z-operator = a w/z-phrase in an A-bar-position A w/z-phrase as such is an argument, unless it is an operator according to 26. So, in 24a whom is in an A-position, hence it does not qualify as an operator, it is an argument and receives the object theta-role; in 24b it is in an A-bar-position, it qualifies as an operator, it is not an argument, hence the object theta-role can be assigned to the variable. Similarly, in 25 who qualifies as an operator, and the subject role can be assigned to the variable, while whom is an argument, and receives the object role. The paradox is thus resolved.5 We can now go back to the basic paradigm of wh-in situ. The well-formedness of 22b is now compatible with the assumption that the Mi-Criterion applies entirely
RESIDUAL VERB SECOND AND THE WH-CRITERION
71
at SS in English: the w/z-element in situ is in an A-position, therefore it does not qualify as an operator under the functional definition of 26, hence clause A of the Mi-Criterion does not apply to it and no violation is produced at SS. Clause A applies to the w/z-operator in Spec of C, and it is satisfied in the familiar manner. In addition to the conceptual advantage of allowing a uniform application of both clauses of the Mi-Criterion, a system involving the Mi-Criterion at SS and the functional definition of 26 has the interesting empirical consequence of deriving a generalization observed by Aoun, Hornstein and Sportiche (1981). These authors pointed out that w/z-movement in English seems to behave as follows: It can move an element from the position of the variable to the appropriate scope position in the syntax or (in multiple questions) in LF, but this movement must take place entirely in one component; in other words, in multiple questions a w/z-element can be left in situ in an argument position at S-structure, as in 27a, and undergo LF movement from there to its scope position at LF, but it cannot be moved to an intermediate A-bar-position in the syntax, as in 27b, to continue its movement to its scope position in LF (see also Georgopoulos 1991 for a recent discussion): (27) a.
Who thinks [C [Mary saw whom ] ]
b. *Who thinks [whom C [Mary saw t] ] This generalization is explained by the Wz-Criterion applying at SS and the independently needed functional definition of w/i-operators from 26: in 27a at S-structure whom does not qualify as an operator because it is not in an A-bar-position, hence clause A of the Mz-Criterion is not violated. It must then be moved in the syntax of LF when 26 is superseded by the general requirement mentioned in n. 5. In 27b, on the other hand, whom is in an A-bar position at SS, it qualifies as an operator according to 26, but then the structure is ruled out at SS by clause A of the Mz-Criterion: a w/z-operator is not in the required configuration with a [+wh] head (the corresponding C is [-wh], given the selectional properties of think). There are other instances of the same descriptive generalization which are now subsumed by the Mz-Criterion applying at SS. 1. Lasnik and Saito (1984, 1992) point out that an embedded topicalized constituent cannot be a w/z-element involved in a multiple question: (28) a.
Who believes that John, Mary likes t
b. *Who believes that whom, Mary likes t In 28b, whom is in an A-bar-position, hence it qualifies as a w/z-operator, and the structure is ruled out as a violation of clause A of the Mz-Criterion at SS.6 2. In French certain quantificational specifiers of the direct object can be extracted and moved to a VP-initial position (an A-bar-specifier position, according to the analysis of Rizzi 1990a): (29) a. II a lu beaucoup de livres 'He has read a lot of books'
72
PARAMETERS AND FUNCTIONAL HEADS
b. II a beaucoup lu [t de livres] 'He has a lot read of books' Obenauer (1976) noticed that when the NP specifier is interrogative (combieri), it can be extracted (30) or left in situ within the NP (3la), but not left in situ in the VP-initial position (3 Ib): (30) Combien a-t-il lu [t de livres]? 'How many did he read of books' (31) a.
II a lu combien de livres? 'He read how many books?'
b. *I1 a combien lu de livres? 'He has how many read of books?' We will come back in section 6 to the general possibility of leaving a w/z-element in situ in main interrogatives in French. The impossibility of 31 b can be attributed to the Wz-Criterion applying at SS, plus the functional definition of 26: combien in 31b qualifies as a w/z-operator, as it is a w/z-element in an A-bar-position; it is not in a Spec-head relation with a [+wh] head, hence it violates clause A of the W/z-Criterion at SS.7 3. It has been noticed that a direct object can be scrambled in pre-subject position in German, but a w/z-element in situ does not allow this process (examples from Grewendorf and Sternefeld 1990): (32) a. Warum hat Peter dieses / welches Buch gekauft? 'Why has Peter this / which book bought?' b. Warum hat dieses /*?welches Buch Peter gekauft? 'Why has this / which book Peter bought?' If the scrambled position is an A-bar-position, this restriction follows from the WhCriterion: the w/z-element qualifies as a w/z-operator, hence clause A of the criterion is violated at SS (thanks to Guglielmo Cinque and Sten Vikner for pointing out this consequence of our analysis). Notice that this restriction follows from the fact that the Wz-Criterion applies at SS in German; we predict that a scrambling language in which the WTz-Criterion applies at LF only should not disallow scrambling of a w/z-element. This prediction appears to be borne out in Japanese (Saito 1985). Before leaving this topic, we must introduce a refinement of our functional definition of w/z-operator. Different considerations suggest that the A/A-bar distinction is too rough: 1. A w/z-element in situ is possible in heavy NP-shifted position, presumably an A-bar-position (e.g., a shifted NP licenses a parasitic gap, see Chomsky 1982 for discussion; the following example is due to Kayne): (33) Which of the students borrowed t from you which of the theses?
RESIDUAL VERB SECOND AND THE WH-CRITER1ON
73
2. In French certain adverbials (for instance, manner adverbials) can be left in situ: (34) II a parle comment? 'He spoke how?' Presumably the positions involved in 33 and 34 are A-bar, and still the in situ strategy is possible. The natural refinement that immediately comes to mind is that the functional definition of operator refers to a more articulated notion of scope position: (26)' w/z-operator = a wh-phrase in a scope position where, by scope position, we mean a left-peripheral A-bar-position (either a Spec or an adjoined position). This excludes right-peripheral positions and the base-generated position of VP adverbials.8'9
5
Negative Inversion and the Negative Criterion
When a negative constituent is preposed in English, I-to-C movement applies obligatorily: (35) a. I would do that in no case b. *In no case I would do that c. In no case would I do that It seems quite natural to try to relate this case to the obligatory application of I-to-C movement in questions. The relation between questions and negatives in this context is strengthened by the observation that negation patterns with the w/z-operators in the selection of a special inflection in the languages discussed in Hai'k (1990; see section 1). Moreover, question and negative operators pattern alike in blocking adjunct extraction (see Rizzi 1990a, chapter 1, for relevant discussion). In the system of the latter reference, such a blocking effect is due to the fact that these operators differ from other operators in that they fill an A-bar specifier position at LF. I would now like to state this scope requirement as resulting from the fact that such affective operators must fulfill at the appropriate level of representation an appropriate generalization of the Wi-Criterion: Informally, affective operators must be in a Spec-head configuration with a head marked with the relevant affective feature. The negative counterpart of 7, the Negative Criterion, would then involve the following configuration:10
74
PARAMETERS AND FUNCTIONAL HEADS
How is a clausal head endowed with the feature [+neg]? Following Pollock (1989), I will assume that negative sentences involve an independent clausal projection, the Negative Phrase; following Belletti (1990) I will assume that it is an intermediate projection between the Agr Phrase and the Tense Phrase (37). I will also assume, as is natural, that the feature [+neg] is licensed in the head position of the Neg P, and that an inflected verbal element can be associated with this feature when it passes through Neg under head-to-head movement, as proposed in Moritz (1989).
We can now understand the pattern in 35 if we assume that the negative counterpart of (clause A of) the Mi-Criterion applies at SS in English. In 35a the negative element is not in a scope position, hence the (negative counterpart of) functional definition of 26' does not apply, and at SS there is no negative operator to worry about. In 35b the negative element is in a scope position, hence it qualifies as a negative operator; but then clause A of the Criterion is violated at SS, as C is not endowed with the [+neg] feature. I-to-C movement can salvage the structure in 35c by moving the [+neg] feature to C. The triggering force of I-to-C movement then is, in essence, the same as in the interrogative case.1' Putting together some previous observations, we can notice that an important asymmetry arises between questions and negatives in embedded contexts: The former disallow I-to-C movement in standard English, while negative proposing in embedded clauses requires it: (38) a. I wonder when (*did) he helped her b. I think that never *(did) he help her We have already discussed possible accounts of 38a. As for 38b, Rizzi and Roberts (1989) have proposed that this is a case of CP recursion, with the higher C filled by that and the lower C hosting I-to-C movement. The contrast ultimately reduces to the fact that, in English, verbs select for interrogative clauses, but not for negative clauses. Embedded interrogatives then sharply differ from main interrogatives, in which no selection is involved, while embedded negatives (in cases of negative proposing) must resort to the same device as main negatives, a marked option made possible by CP recursion.
RESIDUAL VERB SECOND AND THE WH-CRITERION
75
6 Subject Clitic Inversion in French A pronominal subject and the inflected verb invert in French interrogatives, a process which has also been analyzed as a special instance of residual V2 (e.g., Rizzi and Roberts 1989 and references cited there). If we compare the English and French paradigm in main clauses, the salient emerging fact is that French allows more options: Restricting again our attention to the movement of the object and of the inflected verb, we find that three of the four possible combinations are well-formed in French, while only one is possible in English (cf. 8): (39) a. b.
[[Elle a rencontre qui]]? 'She has met who?' [Qui [elle a rencontre t]]? 'Who she has met?'
c. *[a-t [elle t rencontre qui]]? 'Has she met who?' d.
[Qui a-t [elle t rencontre t] ] ? 'Who has she left?'
That is, in addition to the simultaneous movement of the object and the inflected verb (36d), the simple movement of the object (36b) and the full in situ strategy (36a) are also possible, the combination of I-to-C movement and wh-in situ (36c) being the only excluded option. The first approach that comes to mind to express such a less restrictive system would be to assume that the W/z-Criterion does not apply at SS in French, hence the creation of the required spec head configuration could be delayed until LF. But this approach is clearly inadequate in view of the fact that the embedded paradigm is as restrictive as the English equivalent: Only the structure resulting from the simple movement of the object is well formed: (40) a. *Je ne sais pas [ [elle a rencontre qui] ] 'I don't know she has met who' b. Je ne sais pas [qui [elle a rencontre t] ] 'I don't know who she has met' c. *Je ne sais pas [a-t [elle t rencontre qui] ] 'I don't know has she met who' d. *Je ne sais pas [qui a-t [elle t rencontre t] ] 'I don't know who has she left' In order to deal with this paradigm, we are lead to assume that the W/z-Criterion must be satisfied at SS in French: the embedded C is specified [+wh] because of lexical selection, hence 40a and 40c are ruled out as violations of clause B of the W/z-Criterion at SS; whatever principle excludes embedded applications of I-to-C movement (in some languages) will rule out 40d (and redundantly 40c).
76
PARAMETERS AND FUNCTIONAL HEADS
This leaves open the question of the extra options that are allowed in the main paradigm (39), in particular the option ofwh-in situ. Notice that the latter instantiates a kind of LF root phenomenon: In French wh-in situ in single questions (i.e., LF tv/z-movement) is restricted to unembedded clauses (in the sense that a wh-in situ is always construed with the main C; of course the w/z-element can be located in an embedded clause). If the root/non-root asymmetry is ultimately to be understood as a consequence of the Projection Principle, it would be desirable to link our LF root phenomenon to the same principle. I would like to propose that the two additional well-formed structures of the main paradigm in French are made possible by an extra option concerning the licensing of the [+wh] feature on the head of a clausal constituent. We have assumed that lexical selection and free licensing in the main Infl are the only available devices in English. Suppose that French disposes, as an extra option, of the following agreement process: (41) w/z-op X° =>• w/z-op That is to say, we assume that a w/z-operator can endow a clausal head of the [wh] feature under agreement. Of course, the very configuration required by the MiCriterion is an agreement configuration with respect to the feature [wh] (as pointed out in a different context by Kuroda 1986); but we are now distinguishing agreement as a static configuration, in which a Spec and a head are each independently endowed with a given feature, from the kind of dynamic agreement illustrated in 41, in which the specifier is able to endow the head with the relevant feature specification. If the satisfaction of the Mz-Criterion always involves static agreement, we are now claiming that the special extra option that French has is dynamic agreement, as stated in 41. We will assume that dynamic agreement can freely apply in the syntax or in the syntax of LF in French. Let us first consider 39a. At DS no clausal head has the [wh] feature, hence at SS the Mz-Criterion is not violated: clause B does not apply because there is no w/z-head, clause A does not apply because the w/z-element does not qualify as a whoperator under the functional definition of 26'. In the syntax of LF the w/z-element can be moved to the Spec of C, from where it can endow C with the feature [wh] under dynamic agreement (41); then, at LF the structure satisfies the W/z-Criterion. The corresponding derivation is not available in English, due to the lack of dynamic agreement: so, if the feature [wh] is not specified at DS under Infl, the language has no device to introduce it later on, and the structure corresponding to 39a will inevitably violate clause A of the Mi-Criterion at LF. Consider now 39b. At DS no clausal head has the feature [wh]. Mi-movement applies in the syntax, then C can be endowed with the feature [wh] through dynamic agreement. At SS (and at LF) the Mz-Criterion is satisfied. The same configuration does not arise in English due to the lack of dynamic agreement. The two extra options of French in 39a and 39b are thus reduced to a unique additional device that this grammatical system has, dynamic agreement, and illustrate applications of this device in the two components of LF and syntax proper, respectively.12 39d is analyzed exactly as in English: I is independently endowed with the [wh] feature, and I-to-C movement permits satisfaction of the W/z-Criterion at SS (and at
RESIDUAL VERB SECOND AND THE WH-CRITERION
77
LF). As for 39c, in the variant in which I is endowed with the [wh] feature at DS, it is excluded by the Wz-Criterion as SS; in the variant in which I is not intrinsically endowed with the [wh] feature, it is presumably excluded by whatever principle excludes I-to-C movement in non-w/z-constructions, e.g., in declaratives, in a non-V2 language like French (possibly a version of Chomsky's 1989 economy principle). Why is it that dynamic agreement does not increase the grammatical options in embedded contexts (see 40)? In particular, as 40a is ill-formed, we must rule out the following derivation: at DS the embedded C is not specified [wh], hence the WhCriterion is not violated at SS; at LF w/z-movement applies and C is endowed with [wh] through dynamic agreement, hence the Wz-Criterion is fulfilled at LF. How is this derivational path excluded? The answer is provided by the Projection Principle: The specification of the embedded C is determined by the lexical selectional property of the main verb. If the value [-wh] is selected at DS by the verb savoir 'know', then this specification cannot be changed at subsequent levels under the Projection Principle. Therefore dynamic agreement is irrelevant in embedded contexts, and the root nature of wh-in situ (LF w/z-movement) is successfully traced back to the Projection Principle. Lasnik and Saito (1992) point out that Japanese manifests a somewhat similar root/non root asymmetry: The interrogative particle ka (glossed as [+wh]) is obligatory in embedded questions and optional in main questions (examples from Lasnik and Saito 1992, chapter 1, ex. 24-25): (42) Mary-ga [John-ga nani-o katta *(ka)] siritagatte iru (koto) Mary John what bought [+wh] want-to-know fact 'The fact that Mary wants to know what John bought' (43) John-ga doko-ni ikimasita (ka) John where went [+wh] 'Where did John go?' In our terms, a C° endowed with the [wh] feature at S-structure is pronounced ka in Japanese (that ka is a manifestation of C° is very plausible, given the fact that it follows the inflectional elements in a rigidly head-final language). We may assume that Japanese is like French in that it also allows a C° to be endowed with the [wh] feature via dynamic agreement at LF. This gives the variant without ka of 43: C° is empty at SS; at LF w/z-movement applies, C° receives the feature [wh] via dynamic agreement, and the Wz-Criteriori is satisfied at LF. In embedded contexts, the presence or absence of [wh] is a matter of lexical selection, hence the dynamic agreement option has no effect, and only the variant of 42 with ka is well-formed, under the Projection Principle. If Japanese and French share dynamic agreement, they differ (among other things) in that the W/z-Criterion applies only at LF in Japanese, hence the w/z-element can remain in situ also in embedded questions, as well as in main questions with and without ka. The same root/non-root asymmetry, ultimately determined by the Projection Principle, thus has quite different manifestations in the two languages.
78
PARAMETERS AND FUNCTIONAL HEADS
7
I-to-C Movement in Italian Interrogatives
In Italian, as in English, the sequence w/i-element- subject -inflected verb is excluded in main interrogatives: (44) a. *Chi Maria ama? 'Who Maria loves?' b. *Che cosa il direttore ha detto? 'What the Director has said?' These structures become possible if the subject is in postverbal position, as in 45a, or is null, as in 45b: (45) a. Chi ama Maria? 'Who loves Maria?' = 'who does Maria love?' b. Che cosa ha detto? 'What has said?' The observational constraint is that the w/i-element must be left-adjacent to the inflected verb. It appears to be desirable to trace back the impossibility of 44 to the same theoretical explanation that we introduced for the obligatoriness of Subject-Aux inversion in English, hence assume that the linear adjacency of the w/z-operator and the inflected verbal element manifested by 45 actually results from the movement of the latter to C°. A straightforward extension of the analysis of English would run as follows: Suppose that [+wh] is licensed in main clauses under I, as in English. Then 44 violates the W/z-Criterion at SS. If the inflected verb moves to C°, the Wz-Criterion is met, as in 45. The plausibility of this hypothesis and the parallel with the English case is reinforced by the observation that I-to-C movement applies in Italian in hypothetical clauses, another environment in which it can apply in English. It was observed in Rizzi (1982, chapter III), that the hypothetical complementizer se 'if can be dropped only with a postverbal or null subject: (46) a. *(Se) Gianni fosse arrivato, tutti sarebbero stati contenti '(If) Gianni had arrived, everybody would have been happy' b.
(Se) fosse arrivato Gianni, tutti sarebbero stati contenti '(If) had arrived Gianni, everybody would have been happy'
c.
(Se) fosse arrivato in tempo, Gianni sarebbe stato contento '(If) had arrived in time, Gianni would have been happy'
In the reference quoted, this is analyzed via an ad hoc rule optionally deleting se when string-adjacent to the inflected verb (hence inapplicable when the subject intervenes, as in 46a). A more interesting and natural analysis would simply assume that se can be replaced by the inflected verb under I-to-C movement, as in the English (and French) counterpart of this construction. In 46a I-to-C movement has not applied, as the position of the subject shows, hence se cannot disappear.
RESIDUAL VERB SECOND AND THE WH-CRITERION
79
So, the I-to-C approach offers a promising unified analysis of different cases in English and Italian. Still, this unification raises several questions. There are some crucial properties with respect to which the Italian case differs from the English case. If the observational adjacency requirement manifested by 45 is a consequence of I-to-C movement, one would expect the subject to be allowed to appear immediately after the auxiliary, as in the corresponding English case. This is incorrect, since the subject can only appear after the past participle: (47) a. *Che cosa ha il direttore detto? 'What has the director said?' b.
Che cosa ha detto il direttore? 'What has said the director?'
Why is 47a excluded? One possible approach would be to claim that the sequence aux+past participle forms a unique constituent of level X° which is moved to C as a whole. This is quite implausible, though, in view of the fact that adverbs and floated quantifiers can intervene between the auxiliary and the past participle (see also the detailed evidence against an incorporation analysis of the past participle within the auxiliary in Belletti 1990). A more promising analysis is offered by the account given in Rizzi and Roberts (1989) of the corresponding French case, which can be straightforwardly extended (see also Roberts 1993) for relevant discussion). Consider the following contrast: (48) a. Ou est-elle allee? 'Where is she gone?' b. *Ou est Marie allee? 'Where is Marie gone?' Rizzi and Roberts (1989) propose that I-to-C movement in French destroys the context of nominative Case assignment (limited to the Spec-head configuration with Agr), hence 48b is ruled out as a violation of the Case Filter. 48a is well-formed because the clitic pronoun incorporates into the inflected verb in C, thus exploiting a different visibility option which does not rely on case assignment (as in Baker 1988; Everett 1986). So, 47a can be excluded in the same manner: If Agr in Italian only assigns nominative in the Spec-head configuration, I-to-C movement destroys the required configuration, and an overt subject cannot survive in the Spec/I position. No equivalent of 48a is possible, as Italian lacks subject clitics. As for the possibility of 47b, we must now assume that an independent assignor of nominative case is available for a postverbal subject. Assuming a split Infl analysis of the clausal structure a la Pollock (1989) with the relative order of projections of Belletti (1990), we can assume that T° (or, possibly, a lower inflectional head; see appendix) is able to assign nominative under government. According to this approach Italian has two distinct positions for nominative assignment: The Spec of Agr and the lower subject position which is governed by the first inflectional head. That the two contexts must be dissociated is clearly shown by Romanian: In infinitival clauses lacking Agr, an overt preverbal subject is excluded, while a postverbal subject is possible:
80
PARAMETERS AND FUNCTIONAL HEADS
(49) a. *Am plecat fara cineva a ma auzi 'I have left without anyone hearing me' b. Am plecat fara a ma auzi cineva 'I have left without hearing me anyone' Motapanyane (1989, 1991) interprets this as showing that the tense specification of infinitives retains its case assigning capacity in Romanian.13 The facts of 47 can thus be made compatible with the hypothesis that I-to-C movement occurs in Italian interrogatives.
8
Inversion in Embedded Interrogatives
In English I-to-C movement shows a clear asymmetry between main and embedded interrogatives. In Italian things are more complex. The embedded questions corresponding to 44 are quite marginal in the indicative, and acceptable, if still somewhat marked, in the subjunctive (also the indicative complements are fully acceptable if the subject is null or inverted, as in 5la: (50) a. ??Tutti si domandano che cosa il direttore ha detto 'Everybody wonders what the director has said' b. (51) a. b.
Tutti si domandano che cosa il direttore abbia detto 'Everybody wonders what the director have said' Tutti si domandano che cosa ha detto (il direttore) 'Everybody wonders what has said (the director)' Tutti si domandano che cosa abbia detto (il direttore) 'Everybody wonders what have said (the director)'
So, given the logic of the approach, also indirect questions in the indicative mood seem to involve I-to-C movement, even if the requirement is less strict than in main questions. Such a weakening of the root/embedded asymmetry awaits an explanation. Things are even more sharply different from English in other Romance languages, in which the root/embedded distinction tends to disappear altogether. This is the case, for instance in Spanish: According to Contreras (1989) the subject cannot intervene between the w/z-element and the inflected verb neither in main nor in embedded questions: (52) a. *Que Maria compro? 'What Maria bought?' b. *No se que Maria compro 'I don't know what Maria bought' Motapanyane (personal communication) observes the same fact in Romanian: (53) a. *UndeIons'adus? 'Where Ion has gone?'
RESIDUAL VERB SECOND AND THE WH-CRITERION
81
b. Unde s'aduslon? 'Where has gone Ion?' (54) a. *Nu ne-aspus undelon s'adus They didn't tell us where Ion has gone' b. Nu ne-a spus unde s' a dus Ion 'They didn't tell us where has gone Ion' A similar lack of contrast between main and embedded questions appears to hold in Catalan (J. Sola, personal communication). Let us start from such extreme cases. Given the logic of our approach, one seems to be lead to the conclusion that in these languages the functional head bearing the [wh] feature uniformly is the tensed I, also in embedded contexts. We may speculate that the rich tensed I of these Null Subject Languages, the strong gravity center of the clause, attracts specifications that may be more "scattered" in languages with a weaker Infl, including the specification [wh]. This idea seems to raise a technical problem involving the proper selection mechanism. If a verb like the Romance equivalent of wonder, etc. selects a w/z-complement, this specification should be borne by the immediately subjacent head C°, governed by the selector, while 1° would be too deeply embedded to bear a selected specification. But notice that the problem is not worse than the one raised by selection of subjunctive, which also is determined by the higher verb, and is morphologically manifested by the lower Infl. Two technical solutions come to mind: The selection could proceed stepwise, in that the main verb could select a C which in turn is a selector of a subjunctive I (Kempchinsky 1986); or it could be that a higher verb can directly select inflectional properties bypassing the complementizer under some kind of relativization of the minimality principle (e.g., a variety of the one argued for in Baker and Hale 1990). Whatever solution turns out to be acceptable for the subjunctive case, it should be immediately extendable to [wh].14 I will then assume that the [wh] feature is expressed on the embedded Infl in 52,54, etc. Therefore, examples like 52b and 54a are excluded by the Wz-Criterion, on a par with the corresponding main clauses. As for the well-formed examples (54b, etc.), we could assume that I-to-C movement applies, thus satisfying the Wz-Criterion. If C does not contain the feature [wh] in these Romance languages, no recoverability violation is produced (cf. n. 2).15 What about the Italian case? The subjunctive inflection is somewhat weaker than the indicative inflection, in that it contains systematic syncretisms, and does not license referential null subjects in one case (2nd person singular of the present). So, pursuing our physical metaphor, it is conceivable that it will exert a weaker gravitational attraction on other feature specifications, allowing then to appear more scattered in the structure; in particular, WH will be allowed to appear in an embedded C°. Hence embedded I-to-C movement (or w/z-movement to the Spec of I) would not be required in subjunctive complements. As for the fact that even embedded indicative complements appear to be somewhat more acceptable in Italian (more precisely, there is a main/embedded asymmetry in relative acceptability that seems to be less detectable in the other Null Subject Romance languages), this may be
82
PARAMETERS AND FUNCTIONAL HEADS
due to the influence of the subjunctive option, Italian, contrary to other Romance languages, normally allows subjunctive w/j-complements. Perhaps, a [wh] indicative complementizer is marginally permitted on analogy with the subjunctive option. No such analogy may arise in languages like Spanish, which disallow the subjunctive option.16
Appendix: Nominative Assignment to Postverbal Subjects McCloskey (1991) argues that nominative assignment under government requires adjacency between the assigner and the assignee; therefore, nothing can intervene between the inflected verb and the subject in a VSO language such as Irish. In this respect, nominative assignment under government tendentially patterns with the other instances of case assignment under government, while nominative assignment under agreement does not manifest an adjacency requirement. McCloskey's hypothesis is confirmed by the fact that adjacency is required between C and the subject in West Flemish (see Haegeman 1992, who explicitly argues that C is the nominative assigner in that language), and between Aux and the subject in the Aux to Comp construction in Italian (as pointed out by Belletti 1990).17 Consider also the adjacency requirement on genitive assignment under government in the Semitic languages (Siloni 1990). McCloskey's hypothesis and our current assumption on case assignment to postverbal subjects can now account for the somewhat variable adjacency requirement that appears to hold between the verb and the postverbal (non-dislocated) subject in Italian (Calabrese 1985): (55) a. ?Ha risolto il problema Gianni 'Has solved the problem Gianni' b. ?Ha vinto la corsa Gianni 'Has won the race Gianni' c. ??Ha parlato con Maria Gianni 'Has spoken with Maria Gianni' These examples become fully acceptable if the object does not linearly intervene between the verb and the subject, i.e., if it is cliticized: (56) a. Lo ha risolto Gianni 'It has solved Gianni' b. U ha vinta Gianni 'It has won Gianni' c. Le ha parlato Gianni To+her has spoken Gianni' This contrast follows if nominative assignment under government requires adjacency. A qualification is required by the fact that certain adverbs and quantifiers can intervene: The examples in 57 are fully acceptable, and 58b clearly contrasts with 58a:
RESIDUAL VERB SECOND AND THE WH-CRITERION
(57) a. b. (58) a.
83
Non parla piu nessuno 'Not speaks anymore anyone' Vince sempre Gianni 'Wins always Gianni' Ha fatto tutto Gianni 'Has done everything Gianni'
b. ??Ha fatto questo Gianni 'Has done this Gianni' Other adverbs cannot (naturally) intervene with a normal intonational contour: (59) a. ??Ha telefonato ieri Gianni 'Has telephoned yeasterday Gianni' b. ??Ti contattera domani Gianni 'You will contact tomorrow Gianni-bar The natural distinction between the adverbs and quantifiers of 57 and 58a and the adverbs or arguments of 58b and 59 is that the latter must be VP-internal (or final), while the former can fill a higher position to the left of the VP. This is clearly shown, for instance, by the fact that only the former can naturally precede the adverb bene 'well', presumably left-adjoined to the VP: (60) a. Gianni non parla piu bene 'Gianni does not speak anymore well' b. Gianni gioca sempre bene 'Gianni plays always well' c.
Gianni ha fatto tutto bene 'Gianni has done everything well'
(61) a. *Gianni ha fatto questo bene 'Gianni has done this well' b. *Gianni ha parlato ieri bene 'Gianni has spoken yesterday well' c. *Gianni giochera domani bene 'Gianni will play tomorrow well' Notice also that bene itself cannot naturally intervene between the verb and a postverbal subject: (62) a. ?Ha giocato bene Gianni 'Has played well Gianni b.??Ha fatto tutto bene Gianni 'Has done everything well Gianni'
84
PARAMETERS AND FUNCTIONAL HEADS
So, the adverbial elements of 60 can be in the specifier position (or adjoined to the projection) of an inflectional head which the verb moves through (see also the typology of positions arrived at in Belletti 1990): (63) . . . V+I (tutto) tj (bene) tv (Obj) (ieri) (Subj) It then appears that postverbal subjects must be linearly adjacent to the first inflectional head above the VP (indicated by tr in 63, presumably to be identified with T in simple tenses, and with the participial morphology in complex tenses), which can thus be identified as the nominative case assigner under government.18
Notes 1. This analysis, made possible by Chomsky's (1986) approach to the structure of clauses, has a clear explanatory advantage over previous transformational analysis (of this and other V2-type phenomena) in that, under natural assumptions on X-bar-Theory and structure preservation, it immediately explains 1. Why exactly two positions are involved (not one or three). 2. Why they are one head and one maximal projection (not two heads, or two maximal projections). 3. Why they are in that order. Empirical evidence that the auxiliary actually moves to C is provided by the fact that the preposed auxiliary cannot co-occur with if in hypothetical clauses *If'had he said that... (nor in yes-no indirect questions in Hiberno-English, which otherwise allows Subject-Aux inversion; see McCloskey 1991 and example 16). 2. In German, embedded V2 is possible in the declarative complement of some verbs, but. never in embedded questions: (i)
Ich weiss nicht was er gekauft hat 'I don't know what he bought has' (ii) *Ich weiss nicht was hat er gekauft 'I don't know what has he bought' This may be accounted for by assuming that [+wh] fills the C° position of the question, thus making it unavailable for I-lo-C movement. Notice that in main questions I-to-C movement is possible, and obligatory, as in English. See Tomaselli (1989) and Vikner (1990) and references cited there and, for relevant diachronic evidence, Tomaselli (1990). 3. Relatives and exclamatives, two constructions involving w/z-elements, share with (embedded) questions the fact that wfr-movement is obligatory. They differ from main questions in English and French in that they do not trigger I-to-C movement, a property that is particularly relevant in main exclamatives: (i) a. How smart is he? How smart he is! (ii) a. Combicn a-t-il mange? 'How much has he eaten?' Combicn i) a mange! 'What a lot he has eaten'
RESIDUAL VERB SECOND AND THE WH-CRITERION
85
As all these operators belong to the w/z-class, in order to capture the different cases, a refinement of the [±wh] feature system is needed (e.g., questions are [+wh] [+Q], relatives and exclamatives are [+wh] [—Q], etc.; see Rizzi 1990a, section 2.9 for discussion). I will assume that relatives and exclamatives are also in the scope of the appropriate extension of the Wz-Criterion, which accounts for the obligatoriness of w/z-movement, and that the different properties of these constructions are related to the different licensing conditions of the relevant features. 4. A w/z-phrase is a phrase containing a w/z-element; this definition should be refined by referring to the subclass of interrogative w/z-elements (e.g., in Italian chi, quale, but not cui, il quale, etc.; see Cinque 1981), with different subcases of the Wz-Criterion referring to different subclasses of opertors (see n. 2). Moreover, not every phrase containing a whelement qualifies as an operator; there are restrictions generally referred to as Pied-Piping conventions (Ross 1967). We will omit these two refinements here. 5. It is also necessary to assume that the functional definition of 26 holds at DS and SS, but not at LF, where it is superseded by a stronger principle according to which all elements endowed with intrinsic quantificational force are operators at this level, and must be moved to an appropriate scope position. See May (1985) for a proposal along these lines. We need such a principle to enforce general LF movement (hence capture ECP effects) of tv/z-elements in situ. This principle is perhaps to be restricted to non-discourse-linked w/z-phrases, along the lines of Pesetsky (1987). 6. If topicalization involves adjunction to IP, as proposed by Baltin (1982) and Lasnik and Saito (1984, 1992), or base-generation in an independent TOP node (Chomsky 1977 and Cinque 1990), this case is independent from 27; if topicalization involves movement to the Spec of C (with CP recursion in case of embedded topicalization), 27 and 28 reduce to the same case. 7. The VWz-Criterion is not violated in 31a because, given the usual Pied-Piping conventions, the w/i-element there can be the entire direct object, which is in an A-position, hence it does not count as a w/z-operator at SS. 8. McDaniel (1989) analyzes an interrogative construction in German and Romani which involves a w/z-element in a lower Comp connected to the appropriate Comp position through a chain of dummy scope markers (a kind of A-bar expletives, according to McDaniel). This case can be integrated if the W/z-Criterion is interpreted as applying on the head of the A-bar chain of the w/z-operator. We leave open here the many problems raised by some of the Slavic cases discussed in Rudin (1988). 9. The Mi-Criterion has an empirical coverage very close to Lasnik and Saito's (1992) system of filters and conditions, expressed in the traditional theory of Comp involving only one position, i.e., [Comp S]. A quick comparison between the two systems may be helpful. Lasnik and Saito introduce the following filters (we keep their numbering); (i) (13) A [+wh] Comp must have a [+wh] head (SS) (14) A [—wh] Comp must not have a [+wh] head (SS) (35) All w/z-elements must be in a [+wh] Comp at LF (53)*[... Head t ... ]j, w h e r e i / j (i 13) closely resembles clause B of the criterion, and rules out such cases as 9a (the Comp selected by a verb like wonder is [+wh]; the element moved to the unique Comp position becomes the head of Comp in Lasnik and Saito's system); (i!4) rules out such cases as 27b, it is then the closest correspondent to the independently-needed functional definition of 26 in our system; notice that (i!4) does not extend to 28b (under the IP-adjunction analysis), 31b or 32b. (i35) corresponds to clause A of the W/z-Criterion. Finally, (i53) is introduced by Lasnik and Saito to account for Baker's (1970) influential observation that
86
PARAMETERS AND FUNCTIONAL HEADS the following example ii is ambiguous between LFs (iiia) and (iiib), but it does not allow the interpretation corresponding to LF (iiic): (ii) (iii) a. b. c.
Who remembers where we bought what Who remembers where+what we bought 11' Who+what remembers where we bought 11' *Who+where remembers what we bought 11'
(iiic) would be derivable, in principle, via LF movement of where to the main Comp, and subsequent movement of what to the embedded Comp. As the moved element becomes the head of Comp in Lasnik and Saito's system, and the index percolates from a head to its projection, the change of head of the embedded Comp in the LF derivation of (iiic) would create the ill-formed configuration (i55). We can immediately translate their approach to this example within our system (avoiding assumptions incompatible with the restrictive approach to X-bar-Theory and structure preservation of Chomsky 1986) through the reasonable assumption that agreement specifications cannot be changed in the course of the derivation. So, if the embedded C° agrees with where at SS, it cannot agree with another element at LF, which excludes the derivation of (iiic). This assumption is the equivalent of filter (i53). It should also be noticed that the derivation of (iiic) violates Strict Cyclicity in the syntax of LF. 10. See Haegeman and Zanuttini (1990) for detailed discussion of this extension in the context of their analysis of negative concord in Romance and Germanic varieties. 11. If the feature [+neg] is specified in the SS representation of 35a, we must deal with the fact that clause B of the Negative Criterion is not violated at SS. One possible approach would be to say that this clause only applies at LF. Alternatively, one could assume that the feature [+neg] is not present in the SS representation, and that it can be specified on a clausal head at LF through the "dynamic agreement", the mechanism introduced in section 6 to deal with the possibility of wh-in situ in French. 12. The application of dynamic agreement in the syntax and at LF can be dissociated in a grammatical system. Spanish appears to allow wh-in situ in single questions (Torrego 1984:103), but disallows the equivalent of 36b with an overt subject (Contreras 1989): In terms of the proposed system, it has dynamic agreement applying at LF, but not in the syntax. The reciprocal case is represented by Brazilian Portuguese, which allows the equivalent of 36b but excludes wh-in situ in non-echo questions (C. Figueiredo, C. Quicoli, personal communications; Modern Hebrew patterns alike: Shlonsky 1988; Siloni, personal communication). 13. A reflex of this dissociation is found in Italian in the peculiar infinitival construction illustrated below, roughly paraphrased as a hypothetical clause, which allows a postverbal lexical subject, but not a preverbal subject: (i) Per averne parlato anche Gianni, vuol proprio dire che la cosa e di dominio pubblico 'If even Gianni spoke of it, this really means that the thing is generally known' (ii) *Per anche Gianni averne parlato, . . . (ii) violates the Case filter, while for (i) we must assume, following Motapanyane (1989), that the infinitival tense (or, possibly, a lower inflectional head) assigns nominative. We leave open the question why the option is limited to this peculiar construction in Italian. Notice also that in all infinitives pronominal intensifiers in postverbal position (roughly equivalent to the stressed reflexives in English: /, myself, etc), when construed with the subject, have the nominative form: (iii) Ho deciso di parlare anch'io di questa storia 'I decided to speak I too of this story'
RESIDUAL VERB SECOND AND THE WH-CRITERION
87
14. A clear case of dissociation between C and the feature [wh] in both main and embedded contexts is provided by Hungarian, in which [wh] is always associated to the functional head, distinct from C, whose specifier is the focus (see Puskas 1990; see also Kiss 1990 and Maracz 1990). 15. Alternatively, it could be that I to C does not apply, and the target of Wi-movement is the Spec of I position. We will not explore this alternative here. 16. In Italian there is an additional clear asymmetry between perche 'why' and the other whelements (argumental or not), in that the former does not require adjacency to the inflected verb. This is true in both main and embedded contexts with indicative: (i) a. *Dove Gianni e andato? 'Where Gianni is gone?' b. *Come Gianni ha parlato? 'How Gianni has spoken?' c. Perche Gianni e partito? 'Why Gianni is left?' (ii) a. ?Mi domando dove Gianni e andato 'I wonder where Gianni is gone' b. ??Mi domando come Gianni ha parlato 'I wonder how Gianni has spoken' c.
Mi domando perche Gianni e partito 'I wonder why Gianni is left'
Similar asymmetries are found in Spanish (Contreras 1989) and Catalan (Sola, personal communication), while no asymmetry is apparently manifested in Romanian (Motapanyane, personal communication). In English why does not manifest any comparable asymmetry with respect to the other w/j-elements, in that it obligatorily triggers subject Aux inversion in main interrogatives. Perhaps perche can be (but does not have to be) analyzed as a C°, possibly an option connected to its morphological analysis (per+che) which relates it to the complementizer. It could then be analyzed on a par with se 'if, whether', which manifests [+wh] on C, hence does not require I-to-C movement (an empty interrogative operator in the Spec of C should be assumed in both cases). It should be noticed that in different varieties of Veneto, a northern Italian dialect, a form of perche, on a par with se, is incompatible with an overt C (which generally co-occurs with tv/i-elements in its Spec in that dialect), as Poletto (1990) points out. Notice also that in Romanian why is expressed by a two-word phrase (de ce), and is not morphologically related to the complementizer, which may account for its different behavior. In Italian, per che ragione 'for what reason' also does not trigger I-to-C movement, but this may be related to the fact that the obligatoriness of I to C in interrogatives is generally weakened when a discourse-linked w/i-element is involved, for unclear reasons. 17. In gerunds and, more marginally, in some infinitive and subjunctive complements, the order Aux-Subj-past participle is possible in Italian: (i) Avendo Gianni deciso di partire,... 'Having Gianni decided to leave, . . . " This construction is analyzed in Rizzi (1982, ch. Ill) as involving movement of the auxiliary to C, hence as an instance of I-to-C movement, in current terms. To account for the wellformedness of (i) as opposed to 3a we could then assume that the gerundivai (more marginally the infinitival and the subjunctive) inflection is able to assign nominative under
88
PARAMETERS AND FUNCTIONAL HEADS
government. See Roberts (1993) for a refinement of this approach. Belletti (1990) has noticed that an adverb like pmbabilmente 'probably' can be inserted in (i) after the subject, but not between the auxiliary and the subject. 18. Why is it that a trace of the lowest inflectional head suffices to assign nominative under government, while a trace of the highest inflectional head (Agr) does not suffice to assign nominative under agreement in 47a? Following the proposal in Rizzi and Roberts (1989), 1 will assume that the moved inflectional head continues to govern the postverbal subject in 63 under Baker's (1988) Government Transparency Corollary, a principle that has no equivalent for agreement relations. So, a governing head, if moved, continues to govern its domain, but an agreeing head, if moved, ceases to be in an agreement configuration with its original Spec, hence nominative case cannot be assigned in 47a.
References Aoun, J., N. Hornstein, and D. Sportiche. 1981. "Aspects of Wide Scope Quantification," Journal of Linguistic Research 1:67-95. Baker, C.L. 1970. "Notes on the Description of English Questions," Foundations of Language 6:197-219. Baker, M. 1988. Incorporation: A Theory of Grammatical Function Changing. Chicago: University of Chicago Press. Baker, M. and K. Hale. 1990. "Relativized Minimality and the Incorporation of Pronouns," Linguistic Inquiry 21:289-297. Baltin, M. 1982. "A Landing Site Theory of Movement Rules," Linguistic Inquiry 13:1-38. Belletti, A. 1990. Generalized Verb Movement. Aspects of Verb Syntax. Turin: Rosenberg and Sellier. den Besten, H. 1977/83. "On the Interaction of Root Transformations and Lexical Deletive Rules," ms, University of Amsterdam. Published (1983) in W. Abraham (ed.), On the Formal Syntax of the Westgermania. Amsterdam: John Benjamins. 47-131. [Also part of den Besten 1989.] Brody, M. 1990. "Case Theory and Argumenthood," GLOW Newsletter 24:14-15. [Abstract of paper presented at the 1990 GLOW Conference]. Calabrese, A, 1985. "Focus and Logical Structures in Italian," ms, MIT. Chomsky, N. 1957. Syntactic Structures. The Hague: Mouton. . 1977. "On Wh-Movement," in P. Culicover, T. Wasow, A. Akmajian (eds.), Formal Syntax. New York: Academic Press. 71-132. . 1981. Lectures on Government and Binding. Dordrecht: Foris. . 1982. Some Concepts and Consequences of the Theory of Government and Binding. Cambridge, Mass.: MIT Press. . 1986. Barriers. Cambridge, Mass.: MIT Press. . 1989. "Some Notes on the Economy of Derivations and Representations," MIT Working Papers in Linguistics 10:43-74. Chung, S. 1982. "Unbounded Dependencies in Chamorro Grammar," Linguistic Inquiry 13:39-78. Cinque, G. 1981. "On the Theory of Relative Clauses and Markedness," The Linguistic Review \ -.247-294. . 1986. "Bare Quantifiers, Quantified NP's and the Notion of Operator at S-structure," Rivista di grammatica generativa 11:33-63. . 1990. Types of A-bar Dependencies. Cambridge, Mass.: MIT Press.
RESIDUAL VERB SECOND AND THE WH-CRITERION
89
Clements, G.N. 1984. "Binding Domains in Kikuyu," Studies in the Linguistic Sciences 14:37-56. Contreras, H. 1989. "Closed Domains," Probus 1:163-180. Everett, D. 1986. "Piraha Clitic Doubling and the Parametrization of Nominal Clitics," MIT Working Papers in Linguistics 8:85-127. Friedemann, M.-A. 1990. "Le pronom interrogatif que," Rivista di grammatica generativa 15:123-139. Georgopoulos, C. 1985. "Variables in Palauan Syntax," Natural Language and Linguistic Theory 3:59-94. . 1991. Syntactic Variables: Resumptive Pronouns and A-bar Binding in Palauan. Dordrecht: Kluwer. Grewendorf, G. and W. Sternefeld (eds.) 1990. Scrambling and Barriers. Amsterdam: John Benjamins. Haegeman, L. 1992. Generative Syntax: Theory and Description—A Case Study in West Flemish. Cambridge: Cambridge University Press. Haegeman, L. and R. Zanuttini. 1990. "Negative Concord in West Flemish," ms, Universite de Geneve. Hai'k, I. 1990. "Anaphoric, Pronominal and Referential Infi," Natural Language and Linguistic Theory 8:347-374. Hai'k, L, H. Koopman, and D. Sportiche. 1985. "Infl en moore et le liage dans le systeme A-bar," Rapport de recherche du groupe de linguistique africaniste, annee 1985-86, Montreal. Huang, J. 1982. "Logical Relations in Chinese and the Theory of Grammar," Ph.D., MIT. Kayne, R.S. 1984. Connectedness and Binary Branching. Dordrecht: Foris. Kempchinsky, P. 1986. "Romance Subjunctive Clauses and Logical Form," Ph.D., UCLA. Kiss, K. 1990. "Against LF Movement of Wh-Phrases," ms, University of Budapest. Kuroda, Y. 1986. "Whether we Agree or Not: Remarks on the Comparative Syntax of English and Japanese," ms, University of California, San Diego. Lasnik, H. and M. Saito. 1984. "On the Nature of Proper Government," Linguistic Inquiry 15:235-289. . 1992. Move Alpha. Cambridge, Mass.: MIT Press. Maracz, L. 1990. "V-movement in Hungarian: A Case of Minimality," in I. Kenesei (ed.), Approaches to Hungarian, vol. 3. Szeged: JATE. May, R. 1985. Logical Form: Its Structure and Derivation. Cambridge, Mass.: MIT Press. McCloskey, J. 1991. "Clause Structure, Ellipsis and Proper Government in Irish," Lingua 85:259-302. McDaniel, D. 1989. "Partial and Multiple Wh-Movement," Natural Language and Linguistic Theory 7:565-604. Moritz, L. 1989. "Syntaxe de la negation de phrase en francais et en anglais," Memoire de licence, Universite de Geneve. Motapanyane, V. 1989. "La position du sujet dans une langue a 1'ordre SVO/VSO," Rivista di grammatica generativa 14:75-103. . 1991. "Theoretical Implications of Complementation in Romanian", Ph.D., Universite de Geneve. Obenauer, H.-G. 1976. IStudes de syntaxe interrogative dufrangais. Tubingen: Niemeyer. Pesetsky, D. 1987. "Wh-in situ: Movement and Unselective Binding," in E. Reuland and A. terMeulen (eds.), The Representation of(ln)definiteness. Cambridge, Mass.: MIT Press. 98-129. Poletto, C. 1990. "Subject Clitic/Verb Inversion in North-Eastern Italian Dialects," ms, Universite de Geneve.
90
PARAMETERS AND FUNCTIONAL HEADS
Pollock, J.-Y. 1989. "Verb Movement, UG and the Structure of IP," Linguistic Inquiry 20:365424. Puskas, G. 1990. "Movement of Wh-Phrases in Hungarian," ms, Universite de Geneve. Rizzi, L. 1982. Issues in Italian Syntax. Dordrecht: Foris. . 1986. "On Chain Formation," Syntax and Semantics 19:65-95. . 1990a. Relativized Minimality. Cambridge, Mass.: MIT Press. . 1990b. "Speculations on Verb Second," in J. Mascara and M. Nespor (eds.), Grammar in Progress. GLOW Essays for Henk van Riemsdijk. Dordrecht: Foris. 375-386. Rizzi, L. and I. Roberts. 1989. "Complex Inversion in French," Probus 1:1-30. [Also published in this volume, 91-116.] Roberts, I. 1993. Verbs and Diachronic Syntax. Dordrecht: Kluwer. Ross, J.R. 1967. "Constraints on Variables in Syntax," Ph.D., MIT. Rudin, C. 1988. "On Multiple Questions and Multiple Wh-Fronting," Natural Language and Linguistic Theory 6:445-501. Saito, M. 1985. "Some Asymmetries in Japanese and their Theoretical Implications," Ph.D., MIT. Siloni, T. 1990. "Hebrew Noun Phrases: Generalized Noun Raising," ms, Universite de Geneve. Shlonsky, U. 1988. "Complementizer-cliticization in Hebrew and the ECP," Natural Language and Linguistic Theory 6:191-206. Tomaselli, A. 1989. La sintassi del verbo finite nelle lingue germaniche, tesi di dottorato, Universita di Pavia. . 1990. "Cases of V-3 in Old High German," ms, Universite de Geneve. Torrego, E. 1984. "On Inversion in Spanish and Some of its Effects," Linguistic Inquiry 15:103-129. Tuller, L. 1985. "Tense Features and Operators in Hausa," Rapport de recherche du groupe de linguistique africaniste, annee 1985-86, Montreal. 493-516. Vikner, S. 1990. "Verb Movement and the Licensing of NP-Positions in the Germanic Languages," Ph.D., University of Geneva.
3
Complex Inversion in French Luigi Rizzi and Ian Roberts
I Introduction 1
In this paper we would like to show that some recent theoretical innovations permit a principled account of complex inversion, a French construction which is in the agenda of theoretical and Romance syntacticians ever since Kayne's (1972) seminal analysis. Some properties of the construction will lead us to revise and tighten current assumptions on Case, visibility and head-to-head movement, and to propose a new hypothesis on the nature of the root/non-root distinction. The major cases of complex inversion are found in root interrogative sentences: (1) Quel livre Jean a-t-il lu? Which book John has he read? (2) Personne n'est-il venu? No-one isn't he come? 'Didn't anyone come?' A striking property of the construction is that there are apparently two subjects: a full NP, which occurs to the left of the inflected verb (after a w/z-word or initially in yes/no questions), and a pronoun to the right of the inflected verb. That the NP is not dislocated is shown by the fact that it follows the Spec of Comp in 1, and by the well-formedness of an example like 2 involving the quantified NPpersonne 'no-one', which is in general unable to appear in a dislocated position (see n. 2). The simultaneous presence of a lexical and a pronominal subject here gives the appearance of clitic-doubling, either of the kind found with objects in various dialects of Spanish, as illustrated from the River Plate dialect in 3a, or the kind found with subjects in northern Italian dialects, illustrated from Fiorentino in 3b: (3) a. Lo vf a Juan. Him I-saw to John. 'I saw John.' 91
92
PARAMETERS AND FUNCTIONAL HEADS
b. La Maria la parla sempre. The Mary she talks always. 'Mary is always talking.' However, despite apparent similarities, at least two fundamental properties distinguish the French case from those in 3. First, the French construction is highly selective in that it is restricted to direct questions and other environments featuring fronting of the inflected verb. No such construction-specific restriction is found with the ordinary cases of clitic doubling. Second, the pronominal elements in 3 have clear properties of syntactic clitics, which occur attached to the verb or under Infl, and do not occupy an NP position in the syntax. On the other hand, it appears to be the case that French unstressed subject pronouns are in NP position in the syntax, and are cliticized to the inflected verb in the phonology (for relevant evidence see Couquaux 1986;Kayne 1983;Rizzi 1986). The contrast with northern Italian dialects is revealing; while subject clitics and full subject NPs can, and in some cases must, cooccur in many dialects, the two elements are in full complementary distribution in standard French.2 If French subject pronouns manifest an NP position on syntactic levels of representation, then the kind of doubling shown in 1 and 2 must involve two NP positions, not just one, as in 3. Such a state of affairs thus raises different and more acute theoretical problems than the familiar cases of clitic doubling. The basic goal of this paper is to show that the fundamental properties of complex inversion can be properly understood if we combine elements of the thorough analysis proposed in Kayne (1983) with certain more recent proposals: a. Chomsky's (1986b) extension of X-bar Theory to non-lexical categories; b. an adaptation of Baker's (1988) approach to visibility and head-to-head movement; c. the idea, independently arrived at by a number of researchers, that subjects are base-generated in VP and raise to their surface subject position in IP. Our adaptation of Baker's theory of head-to-head movement, in conjunction with a strict interpretation of the Projection Principle, also yields a principled account of the fact that complex inversion is limited to root structures (cf. Den Besten 1983; Safir 1981/82; Safir and Pesetsky 1981), an account which can be extended to root phenomena in general (Emonds 1976). In section 2 we outline an analysis of subject-clitic inversion, a necessary prerequisite. In section 3, we address the problem posed by the presence of two subjects, which we factor into three distinct problems: a. how can the Case requirements of the two nominals be simultaneously fulfilled? (the Case problem); b. which positions do the two subjects come from in the derivation? (the source problem); c. which positions do the two subjects occupy at S-structure? (the landing-site problem). In section 4, we turn to the question of the restriction of complex inversion to root contexts and we develop a general approach to the root/non-root distinction.
COMPLEX INVERSION IN FRENCH
93
2 Subject-Clitic Inversion First of all, it is necessary to sketch an analysis of one component of complex inversion which exists as an independent construction: subject-clitic inversion. This construction involves the inversion of a pronominal subject with the inflected verb, shown in 4: (4) a. Est-il parti? 'Has he left?' b. Ou est-il alle? 'Where has he gone?' Following den Besten (1983) and Kayne (1983), we assume that this inversion process involves leftward movement of the verb over the subject rather than rightward movement of the subject over the verb. Adopting the extension of X-bar Theory to non-lexical categories proposed in Chomsky (1986b), and the theory of head-to-head movement of Baker (1988), this process can be seen as raising of the inflected verb from 1° to C°, shown in 5 (cf. Rizzi 1987b):3
This approach immediately explains why inversion is impossible if C° is filled. For instance, in the Quebec dialect of French, where an overt C° can co-occur with a w/z-element in Spec-CP, inversion is restricted to the case in which this option is not taken (Goldsmith 1981): (6) a.
b.
Qui que tu as vu? Who that you have seen?
Qui as-tu vu? Who have you seen?
c. *Qui qu'as-tu vu? Who that have you seen? In 6c C° is filled by que and hence is not available as a landing site for movement of the inflected verb.
94
PARAMETERS AND FUNCTIONAL HEADS
Standard French does not allow the co-occurrence of a w/i-element and que, but a reflex of the same phenomenon can be seen with a certain class of adverbs. These adverbs are able to either trigger inversion or to co-occur with a fto-clause. Again, these options are exclusive: (7) a. b.
Peut-etre qu'il a fait cela. Perhaps that he has done that. Peut-etre a-t-il fait cela. Perhaps has he done that.
c. *Peut-etrequ'a-t-il fait cela. Perhaps that has he done that. The natural account of 7 is to say that this class of adverbs (which includes peut-etre, a peine 'hardly,' and a few others) are able to appear in Spec-CP. This brings the paradigm in 7 into line with that in 6. Third, again in standard French, a conditional clause can be introduced either by the overt complementizer si 'if or by the inversion of a verb in the conditional mood, but not by both: (8) a.
Si tu avais fait cela . . . If you had done that... '
b. Aurais-tu fait cela . . . Had you done t h a t . . . ' c. *Si aurais/avais-tu fait cela . . . If had you done that... ' Si and the inflected verb thus appear to compete for the same position, namely C°. The analysis of subject-clitic inversion as involving I°-to-C° movement follows and updates the basic idea proposed by den Besten in that it treats inversion in French as essentially the same phenomenon as the more pervasive kinds of inversion found in Germanic languages. There is nevertheless a striking difference between the French case and the Germanic case (illustrated below by subject-aux inversion in English); namely, that the process is restricted to pronominal subjects in French, unlike in Germanic: (9) a. Has John spoken? b. *A Jean parle? (10) a.
Has he spoken?
b. A-t-il parle? Developing a suggestion by Szabolcsi (1983), we will propose that the impossibility of 9b should be accounted for in terms of Case Theory. The idea is that raising of 1° to C° destroys the context in which 1° assigns Case to the subject in French, but not in English or in other Germanic languages. A straightforward implementation of this proposal makes use of the idea of directionality of Case assignment; suppose
COMPLEX INVERSION IN FRENCH
95
that in French Nominative Case can only be assigned leftward, while in English and in other Germanic languages either direction of assignment is possible. In that case, a phonologically-realized NP will violate the Case Filter in the context created by I°-to-C° movement in French. This is precisely the context of 9. So Jean violates the Case Filter in 9b. In English, there is no Case Filter violation here because Nominative can be assigned either leftward or rightward.4 As it stands, this proposal is too strong, as it rules out the well-formed example lOb. In order to account for lOb we need to elaborate on what the Case Filter really requires. Following the general proposals of Baker (1988), we assume that the requirement that NPs be Case-marked is actually an instance of a more general requirement that nominals be associated with a Case feature. This association takes place in one of two ways: either by means of assignment of the feature from a head to the nominal, or by means of incorporation of the nominal into the head bearing the Case feature (for a precise formulation of this requirement, see Baker, Johnson and Roberts (1989:239):
One variety of incorporation is cliticization. Following Kayne (1983), we assume that the pronoun in subject position can clitici/e to the inflected verb in the syntax, once the latter has been moved to C°.5 So lOb has a representation as shown in 12:
Here the clitic escapes the effects of the strict directionality condition on Nominative assignment in French as it is associated with a Case feature (the Nominative feature borne by 1° in C°) by incorporation with C°, so that the fact that Case assignment to Spec-IP is blocked is irrelevant. To sum up, we treat subject-clitic inversion as the combination of the raising of the inflected verb to C° followed by incorporation of the subject pronoun with the inflected verb in C°. Incorporation of the pronoun is one way of associating it with a Case feature. Due to the directionality condition on Nominative assignment in French (or, alternatively, the language-specific mode of Case assignment discussed
96
PARAMETERS AND FUNCTIONAL HEADS
in n. 4), this is the only way that a subject can satisfy the requirements of Case Theory when 1° to C° takes place. The fact that 1° to C° can only occur with pronominal subjects is thus reduced to the fact that pronominals are the only elements that undergo incorporation in French (in fact, incorporation from subject position appears to be restricted to pronominals universally; see Baker and Hale 1988). With this background, we can go back to the issues raised by complex inversion.
3 The Problem of Two Subjects The existence of two apparent subjects in complex inversion constructions poses three problems. The first of these we call the "Case problem": how are the two subjects assigned Case? The second problem is the "source problem": where do these two subjects originate? The third problem is the "landing-site problem": where do these subjects, in particular the NP, appear at S-structure? In this section, we will answer each of these questions in turn, thereby arriving at an analysis of complex inversion.
3.1
The Case Problem
It is implicit in most versions of Case Theory and explicit in some (e.g., Vergnaud 1985) that there is a biunique relation between Case assigners and Case assignees. If this is so, the Case problem can be put as follows: how do both the full NP and the clitic satisfy the requirements of Case Theory in complex inversion? We will show that the analysis of subject-clitic inversion given in the previous section provides an automatic solution to this problem. Before presenting our analysis, we must make a preliminary assumption concerning the position of the full subject NP, a matter we will elaborate on below. For the moment, we simply recast Kayne's (1983) proposal in terms of the assumptions about X-bar Theory of Chomsky (1986b). As the NP apparently occupies a position immediately to the right of Spec-CP and immediately to the left of C°, we take it that this NP is left-adjoined to C'. The complete structure is thus the following: (13) [cp wh [c, NP [c, f c o I°-C1] IP] ] ] In this structure the NP is governed by 1° and is to the left of it. Therefore it is assigned Nominative Case from right to left, in the usual way operative in simple declarative clauses (and, presumably, the two elements are in a configuration sufficiently close to Spec-head agreement, if the proposal in n. 4 is to be adopted). As for the clitic, we have seen that it cannot be assigned Case in the usual,way because it is "on the wrong side" of 1°, and need not be assigned a Case because it is associated with a Case feature by incorporation. The Case requirements of the two nominals are thus satisfied independently of each other. This account is not incompatible with the idea that there is a bi-unique relation between Case features and nominals; the bi-uniqueness condition is relativized to modes of association of Case features with nominals, in that assignment of a Case to a nominal is subject to bi-uniqueness, as well as association of a nominal to a Case feature by incorporation. However, the
COMPLEX INVERSION IN FRENCH
97
two modes of association can independently associate a single Case feature with two nominals.6 This account allows us to see why complex inversion is impossible in English: (14) * Which books John has he read? Here 1° in C° could assign Nominative Case either leftwards or rightwards, but not to both nominals at the same time. Since English subject pronouns never undergo incorporation, he cannot incorporate into C°, so this means of satisfying the requirements of Case Theory is unavailable. Hence there is no way that the requirements of Case Theory can be satisfied in 14. This analysis retains the idea of Kayne (1972) that the possibility of complex inversion in French is a consequence of the existence of subject clitics in this language.
3.2
The Source Problem
The Case problem is just one of the issues raised by the presence of two subjects. Another question which must be answered is: where do the two subjects come from, i.e., which positions are they base-generated in? We begin by giving a brief summary of Kayne's (1983) answer to this question. In Kayne's terms, the derivation of a complex inversion structure is as follows (we alter the category labels so as to accord with Chomsky 1986b): (15) a. [CP [jP Jean a mange] ] b. [CP a [Ip Jean t mange ] ] c. [cp Jean a [Ip 11 mange ] ] d. [cp Jean a [Jp il t mange ] ] e. [cp Jean a-t-il [ IP 11 mange ] ] The first step is movement of the inflected verb to Comp, deriving 15b from 15a. Next, the subject left-adjoins to some projection of Comp, giving 15c. Example 15d is derived by the insertion of an expletive pronoun in subject position. Finally this pronoun cliticizes leftwards onto the inflected verb in Comp. This derivation involves two problematic steps. First, strict cyclicity is violated in 15d and 15e, in that the operations which derive these structures take place in a subdomain of the domain of operations deriving 15c. Such a violation is suspect, even if the Strict Cycle Condition does not itself turn out to be a primitive condition of the theory (see Freidin 1978); why should it hold as a theorem in general but not in this case? Second, a widely accepted if not explicitly formulated assumption concerning lexical insertion is that all phonetically realized material is present at D-structure (see Burzio 1986). This means that derivational operations can only create traces or fill empty positions by means of movement (they may also possibly delete material). This plausible constraint is violated by the insertion of il in 15d. It is fairly clear that both of these problems stem from the same cause: the fact that Kayne assumes that there is only one subject position in basic clause structure, at the time an uncontroversial assumption. This is why the same position must be the
98
PARAMETERS AND FUNCTIONAL HEADS
source of the two subjects, and thus why // must be inserted after the subject position has been vacated, leading to a violation both of strict cyclicity and of the condition on lexical insertion mentioned above. In the context of recent proposals by a number of authors (Kitagawa 1986; Koopman and Sportiche 1985, 1988; Kuroda 1986; Manzini 1986; Sportiche 1987; Zagona 1982; and others) regarding the base position of subjects, we can straightforwardly solve the source problem. We adopt a variant of these proposals according to which subjects are base-generated in the Specifier of VP and raise in the course of the derivation to the Specifier of IP. This amounts, in effect, to treating I as a raising trigger. The proposal is illustrated for a simple English sentence in 16:
As in normal cases of raising, the subject moves to Spec-IP in order to satisfy the requirements of Case Theory at S-structure. The relevance of this proposal for us is that it makes available two subject positions. We thus propose that the two subjects of the complex inversion construction each occupy one of the two subject positions at D-structure: the pronoun, which following Kayne we assume to be an expletive,7 occupies Spec -IP and the full NP occupies Spec-VP. The following is the DS representation of an example like 15e:
Here the subject argument, Jean, occupies a theta-position, and the expletive pronoun is in a non-theta-position. The Theta Criterion is thus met at D-structure. In French, the leftmost verbal element must raise to a tensed inflection (cf. Emonds 1978; Pollock 1989), so the following configuration is derived:
If no further movement takes place, the structure will be ruled out by Case Theory, since, given our assumptions, Jean will be unable to receive a Case here. In fact, if there is no interrogative or adverbial element present that activates the CP-level, this kind of configuration is ruled out by Case Theory. If the CP-level is activated by the presence of some appropriate element, I°-to-C° movement can legitimately apply, yielding the following configuration:
The pronoun is now able to incorporate with the auxiliary, since the auxiliary c-commands it. Moreover, our assumptions about Case Theory, spelled out in the previous section, mean that the inflected verb still has the capacity of assigning a Nominative Case feature leftwards to an NP which it governs. The NP can then move directly from Spec-VP to a position to the left of the auxiliary where it will be assigned Nominative Case. These operations yield a well-formed complex inversion structure, illustrated in 20:
COMPLEX INVERSION IN FRENCH
99
The structure can only arise where 1° moves to C°, because the environment in which the two subjects are both able to satisfy the requirements of Case Theory depends on the presence of the inflected verb in C°.8 A striking fact about the above derivation is that Jean raises from Spec-VP position to the pre-C° position, skipping Spec-IP. In this representation, the Caseless trace left in the Spec-VP position, t1^ is not a variable. Moreover, being non-pronominal, we must take it to be an anaphor, analogous to an NP-trace. Thus 20 is analogous in relevant respects to cases of super-raising that have been discussed in the literature (cf. Lasnik 1985; Chomsky 1986b; Baker 1988). In general, super-raising leads to severely ungrammatical sentences, of the type in 21: (21) *Johnj seems that Bill likes t; Why is it that the application of NP-movement skipping Spec-IP does not lead to ungrammaticality in 20? There are two issues to be addressed here. The first concerns the Binding Theory, and the second the intersection of the ECP and Theta Theory. Taking the bindingtheoretic question first, the problem is that NP-traces are subject to Principle A of the Binding Theory. This principle requires that anaphors be bound in their binding domain. In (20), the binding domain for z^ is the minimal category containing a governor for ^ and a subject, i.e., IP. Therefore Jean has moved to a position outside the binding domain of its trace in 20. However, the representation in 20 is saved from Principle A by the fact that Jean and il can (and must) have the same index. This ensures that /^ satisfies Principle A, as it is bound by an element which is in its binding domain, namely the trace of //, which occupies Spec-IP. Thus the derivation of 20 violates Principle A of the Binding Theory, but the representation does not. Since, under current assumptions, the binding conditions are checked on representations and not on derivations, 20 does not lead to a violation. It is well known that the Binding Theory is too weak to deal with the whole class of super-raising structures, however. In particular, what we have just said will not distinguish 20 from examples such as 22: (22) *Johni seems that he4 likes tj This sentence is very bad, despite the fact that the trace has an antecedent in its binding domain, the coindexed subject he. This leads us to the second issue mentioned above. Under current approaches, 22 is ruled out either as a violation of the ECP (Chomsky 1986b), or as a violation of Theta Theory (Rizzi 1990). Both accounts have in common that a crucial antecedent-government relation fails to obtain. We will develop here the theta-theoretic approach. In general, arguments in non-thetapositions must be connected to their theta-positions through chain-formation. The basic condition on chain formation is that each element in a chain antecedent-governs the next (see Chomsky 1986b). Moreover, well-formed theta-chains must preserve the bi-uniqueness condition imposed by the Theta Criterion in that they can contain exactly one argument, and can be assigned exactly one theta-role. Structures such as 22 violate this condition in that the only chains that would satisfy the theta-criterion violate the antecedent-government requirement. In particular, no chain can unite the NP-trace and John. So 22 is ruled out ultimately by Theta Theory.
100
PARAMETERS AND FUNCTIONAL HEADS
If 22 is ruled out in this way, why is 20 grammatical? Being base-generated in a non-theta-position, il is an expletive in 20, so that the chain (Jean, il, t, t') contains exactly one argument. Moreover, this chain is well-formed with respect to
the antecedent-government requirement since each member antecedent-governs the
next. Hence the Theta Criterion is satisfied here.9 To summarize, we propose that the D-structure for complex inversion is as in 17 and the S-structure as in 20. The derivation involves several types of movement: head-tohead movement of a, cliticization of il to a and NP-movement of Jean. All of these movements take place so that the two subjects are able to satisfy the requirements of Case Theory, outlined in the previous section. Movement of the inflected verb to C° is a necessary precondition for the satisfaction of these requirements, so that this approach explains why complex inversion can only occur in interrogatives or other constructions activating the CP-level. Raising the NP subject from Spec-VP to a position in C does not violate either the Binding Theory or other conditions on chains, despite being derivationally close to super-raising, because unlike other cases of super-raising, the NP moved across the subject lands within the same clause and the antecedent-government requirement on each link of the chain can be met.
3.3
The Landing-site Problem
Two questions fall under the landing-site problem: (i) what is the structure of the sequence WH NP V-C1? and (ii) how is the unique well-formed order to be guaranteed? Above we proposed that the natural updating of Kayne's analysis would posit that the full NP subject occupies a position left-adjoined to C'. On this proposal, there is only one CP, whose Specifier is occupied by the w/z-phrase, whose head is occupied by V-C1, and the subject NP is left-adjoined to C', as in 23:
(23) [CP wh [c, NP [c, [co V-C1] IP ] ] ] This analysis violates a putative constraint on adjunction, i.e., Chomsky's (1986b) proposal that maximal projections can only be adjoined to other maximal projections. If the proposal in 23 is correct, Chomsky's constraint should be weakened so as to allow adjunction of non-heads to non-heads. This would maintain the important restriction that non-heads cannot be adjoined to heads, and heads cannot be adjoined to non-heads. It is nevertheless worthwhile to explore some alternatives, although we shall tentatively conclude that the structure in 23 is to be kept. One alternative is immediately suggested by the guiding intuition behind the proposals made in the previous section for the underlying structure of complex inversion, i.e., that the construction involves two subjects. Pushing this intuition, we would be led to the conclusion that the NP literally is in a subject position at S-structure, as well as at D-structure. This implies that basic clause structure makes available three subject positions, not just two, as we have been assuming up to now: the source position of the NP, the source position of the pronoun, and the landing-site position of the NP. In fact, Pollock (1989) proposes just such a structure for clauses. He argues that, instead of considering there to be a single node Infl containing two kinds of features,
COMPLEX INVERSION IN FRENCH
101
Tense and Agr, these two elements should be treated as heading their own maximal projections. This proposal, motivated by facts from Verb Raising in French, leads to a considerably more articulated structure for the clause, namely that illustrated in24: 10
This structure in principle makes available three subject positions, all of which we could exploit in the following representation for complex inversion:
In the D-structure representation Jean occupies Spec-VP and il Spec-TP. The auxiliary raises to Agr°; the pronoun incorporates into Agr°; and Jean moves to Spec-AgrP. The main point in favour of this structure is that it provides a clear and simple solution to the landing-site problem by making available a sufficient number of structural positions. However, adopting this structure poses several problems in other areas. The basic problem is that the CP-level plays no role in 25. This means on the one hand that there is no obvious way to state the fact that complex inversion is characteristic of interrogatives. Nothing prevents the generation of sentences exactly like those in as declaratives. Although a sentence such as Jean a-t-il mange is grammatical in French, it must be understood as a question. This is clearly a fact that
102
PARAMETERS AND FUNCTIONAL HEADS
our analysis must capture, but which the proposal in 25 does not naturally deal with. Moreover, the fact that the CP-leve! plays no role in 25 means that it is hard to see how this approach can provide an account of the root nature of complex inversion
(see section 4 on this),
More seriously, we would be left without an account of the fact that complex inversion is incompatible with the presence of an overt C°, as in: (26) a. Peut-etre Jean est-il parti. Perhaps John has he left. b. Peut-etre que Jean est parti. Perhaps that John has left. c. *Peut-etre que Jean est-il parti. Perhaps that John has he left. The same observation holds for the Quebecois phenomenon mentioned above (examples from Safir 1981/82:461-462; thanks to Maria Teresa Guasti for drawing our attention to this fact): (27) a.
Quoi que Jean veut? What that John wants?
b. *Quoi que Jean a-t-il voulu? What that John has he wanted? If complex inversion involves movement of the inflected verb to C° these paradigms are immediately accounted for, on a par with the simple subject-clitic inversion cases discussed earlier (see examples 6 and 7). But given a structural representation such as 25, the gaps in the paradigms remain mysterious. All these problems clearly stem from the fact that movement to C° is not involved in this analysis. We therefore reject the proposal in 25. In particular we will not assume that V-C1 is in Agr°, but in C°, as the evidence reviewed forcefully argues.11 A less radical alternative to C'-adjunction is CP-adjunction of the wfo-phrase and structure-preserving movement of the subject NP to Spec-CP. This would give the structure in 28:
(28) [CP wh [CP NP [c, [co V-C1] IP] The order wh NP V-C1 would then involve assuming w^-adjunction to CP rather than NP-adjunction to C', an assumption that avoids the technical problem mentioned in connection with 23. However, the structure in 28 poses some problems of its own. These arise in part because it implies that w/z-movementin the syntax can have a landing site which is not the typical position of w/z-operators, the Spec of Comp, and in part because it involves movement of a non-operator, the subject NP, to an operator position. The second option raises the possibility of non-operator movement to Spec-CP in general, which would lead us to expect generalized Verb Second (V2), a phenomenon not found in (Modern) French. The first option raises the question of what prevents iteration of the w/!-adjunction, or the combination of w/z-movement to Spec-CP and w/!-adjunction to CP. This would give rise to clearly ungrammatical sentences such as the following:
COMPLEX INVERSION IN FRENCH
103
(29) *Ou quels livres Jean a-t-il trouves? Where which books John has he found? For these reasons, we maintain the analysis shown in 23, involving C'-adjunction of NP, and w/z-movement to Spec-CP in w/z-questions (or the presence of a null operator in this position in yes/no questions).12 Returning then to the structure in 23, it is important to see how a theory allowing C'-adjunction necessarily only gives rise to the order of elements found in complex inversion. Taking an example where a w/z-phrase is present, there are four logical possibilities to be considered: (30) a. [ CP wh[ c , NP b. [cp wh [(y wh
c. [ CP NP[ C , NP d. [ CP NP[ C , wh Clearly, all of these possibilities, except 30a, must be excluded. Example 30b violates the constraint on the distribution of w/i-elements in French which requires that they be either in operator position (i.e., Spec-CP) or in situ at S-structure. Example 30c is ruled out because a non-operator, NP, occupies an operator position, namely Spec-CP (in a non-V2 language). Finally, 30d is ruled out for both of these reasons. We must also rule out the possibility of C'-adjunction of a non-subject in 30a, as well as multiple C'-adjunction. Following the Principle of Full Interpretation of Chomsky (1986a), we take it that an element occurring in a given position at LF must be licensed in that position by an interpretation. As the C'-adjoined position is neither an operator position nor an argument position (nor a left-dislocation position, a position whose content is presumably licensed at LF by a rule of predication), an element occupying this position at LF can only be licensed by being in a well-formed theta-chain. The formation of a well-formed chain from this position is impossible for non-subject NPs, because the subject in Spec-IP will block chain-formation with any position it c-commands, since it will block antecedent-government of any such position.13'14 Thus the only way of licensing the C'-adjoined NP at LF is by linking it to a trace in subject (i.e., Spec-IP) position. Therefore the only possible candidate for C'-adjunction is the subject NP itself. The C'-adjunction option thus does not give rise to overgeneration. The above approach to the landing-site problem has the advantage that it allows us to deal with two other properties of complex inversion noted by Kayne. First, the construction does not allow questioning of the subject itself: (31) *Qui est-il parti? Who did he leave? Second, complex inversion is incompatible with stylistic inversion: (32) a. b.
Ou Jean est-il alle? Where John has he gone? Ou est alle Jean? Where has gone John?
104
PARAMETERS AND FUNCTIONAL HEADS
c. *Ou est-il alle Jean? Where is he gone John? According to the approach to the landing-site problem advocated above, the representations for the relevant parts of these examples would be as follows: (33) a. [cp Qui [c, t [est-il [t'[t"parti] ] ] ] ] b. [cp Ou [c, pro [est-il [t'[t"alle] Jean] ] ] ] Following Kayne (1983), we can straightforwardly account for the ill-formedness of these two representations by exploiting the fact that the crucial empty category is in an adjoined, hence A', position. Consider first 33a. Here t does not qualify as a variable because it is an A'-position; f does not qualify either, since it has the status of an incorporation trace (a status that we assume to be incompatible with the status of syntactic variable), and t", the trace in the base position of the subject, cannot be a variable because it is in a Caseless position. Hence there is no syntactic variable that the operator can bind, and so the structure is ruled out by the general ban on vacuous quantification. Next, consider 33b. We assume that stylistic inversion involves a pro subject licensed by a C° under certain conditions (as Pollock 1986 suggests for some cases). Recall that^ra is really an abbreviation for the feature matrix [—anaphoric, -hpronominal]. It is natural to assume that these features only classify empty categories in A-positions; in fact, the only distinction that is needed in A'-positions is that between intermediate traces and empty operators, a distinction that is not properly captured by the features [ianaphoric, ipronominal]. Hence the empty category occupying the C'-adjoined position in 33b cannot be pro. If pro is a necessary component of stylistic inversion, 33b will be ill-formed.15'16 Notice that the approach to the landing-site problem based on the representation in 25 is unable to account for the facts in 33 in an equally straightforward way, because in that approach the crucial empty category would be in an A-position non-distinct from an ordinary subject position.17
4 Root Phenomena A salient property of complex inversion is the fact that it is limited to root clauses, as the ungrammaticality of 34 shows: (34) *Je me demande qui Jean a-t-il vu. I wonder who John has he seen. In this section we will propose an account of this restriction, which we phrase in the context of a general approach to the nature of root phenomena. The root character of complex inversion is undoubtedly to be related to the root character of one component of the construction, namely subject-clitic inversion: (35) *Je me demande qui a-t-il vu. I wonder who has he seen.
COMPLEX INVERSION IN FRENCH
105
Both 34 and 35 appear to conform to a fundamental generalization concerning root phenomena: movement of the inflected verb to C° is by and large restricted to main clauses. This rough generalization subsumes, in addition to the French constructions, subject-aux inversion in English and the main types of V2 in other Germanic languages (cf. den Besten 1983). The account we want to propose relies on the idea that the correct distinction is not main vs. embedded clause, but rather selected vs. non-selected clause (see Kayne 1982). A quick survey of the relevant cases supports this hypothesis. In the first instance, we should separate independent CPs from subject, complement and adjunct CPs; the former allow verb-movement to C° while the latter do not. It is clearly true that independent CPs are not selected, and it follows from the Projection Principle, in conjunction with the Theta Criterion, that both complement and subject CPs must be selected. This leaves adjunct CPs. In typical adjuncts, for example the kind which can host a parasitic gap, CP is selected by a Preposition (in English, without, before, in order, etc.). Thus the whole adjunct is a PP containing a CP selected by the Preposition in such cases. There is, however, one class of adjunct CPs which provides evidence that the correct generalization regarding the possibility of inversion concerns the selected/nonselected distinction rather than the main/embedded distinction, namely the class of conditional protases (see Kayne 1982). Conditionals are embedded adjuncts, and they are also not selected. As 36 shows, they optionally allow inversion: (36) a. Had I the time, I'd help you. b. Aurais-je le temps, je vous aiderais. Putting these observations together with previous remarks on the incompatibility of inversion with a filled complementizer (cf. 8c and English *lfhadlthe time ...), the following generalization emerges: (37) Inversion is possible only if (i) CP is not selected, and (ii) C° is not filled. In most cases the two conditions overlap, for example in embedded that-claus&s, but there are cases of both unselected clauses with a filled C° that block inversion (cf. 6c-8c, 26c, 27b), and of selected clauses where C° is empty and inversion is blocked (e.g., 34 and 35). We have already seen that condition (ii) of 37 follows directly from the idea that inversion involves movement of the inflected verb into C°: if C° is filled movement cannot take place. The main topic of this section will be to explain what underlies condition (i) of 37. One possible approach would be to try to reduce (i) to (ii) by assuming that a selected C° is always filled in the relevant sense. This is not implausible in the case of indirect questions such as 34 and 35, as here we could claim that C° is filled by the feature [+wh] selected by the main predicate, and hence is not available as a landing site for movement. However, the drawback to this approach is that there is no good way to ensure that all selected CPs have a filled C°,
106
PARAMETERS AND FUNCTIONAL HEADS
especially in cases where C° is phonetically null. For this reason, we will explore a more principled approach. We will claim that condition (i) of 37 derives from the Projection Principle. The Projection Principle requires that selectional properties be satisfied at all levels of syntactic representation. This requirement extends to categorial selectional properties, thereby imposing a strong structure-preservation constraint on all selected contexts. We will propose that I°-to-C° movement or, more precisely, the instances of this process that concern us here, do not preserve the structure in the strong sense required by the Projection Principle, and so is banned in all selected contexts. To show how this idea can work, we must first introduce some assumptions concerning the nature of head-to-head movement. We further constrain the approach of Baker (1988:59) by assuming that head-tohead movement is always and only substitution of a head into another head position. In other words, we restrict the adjunction option to maximal projections (but see n. 18). In cases where incorporation results in a visible amalgam of the two heads, e.g., standard cases of Noun incorporation or V-to-I movement where V picks up tense and agreement marking, we assume that the incorporation host morphologically subcategorizes for the incorporee, hence a structural slot is created for the incorporee at D-structure as a function of the lexical properties of the incorporation host (cf. Lieber 1980, on morphological subcategorization). So (tensed) 1° in a language like French has the subcategorization frame [+V° ], an incorporating V° in Mohawk has the feature [+N° ], and so on. In general, where an incorporation trigger X° has the feature [+Y° ], this means that the slot for Y° is base-generated within X°, triggering substitution of Y° during the derivation, leading to the creation of a complex head with the government and Case-marking properties discussed at length by Baker (1988, ch. 2). With this kind of incorporation, the head of the complex formed by incorporation remains X°, the incorporation trigger.I8 Of course, nothing prevents an incorporation host of this kind from being selected by a higher head. Since incorporation does not alter categorial status, no problem is posed for the Projection Principle. Consider, for instance, Noun incorporation in an incorporating language. In such cases, the Verb has the morphological subcategorization feature [+N° —], creating a slot into which the Noun can be substituted. In 38, Noun incorporation is strongly structure-preserving, in the sense that it moves N° to a pre-existing slot and it does not change categories; the verb does not become a noun. If 1° selects a V-projection (cf. Chomsky 1986b), the Projection Principle is not violated since the complex head resulting from incorporation remains a verb at S-structure. On the other hand, if the potential host does not provide a structural slot via morphological subcategorization, adjunction of heads being excluded (or limited to clitici/ation; see n. 18), the only way for a lower head to incorporate is by direct substitution into the host head. Of course, in most cases this operation will be excluded by the Recoverability Principle, the content of the host head being nonrecoverably erased. There is one case, though, in which recoverability is not violated: this is when the host head is radically empty, hence there is no content to recover. Our claim is that this is precisely what happens in the familiar cases of I°-to-C° movement. This gives rise to a structure such as 39:
COMPLEX INVERSION IN FRENCH
107
Let us see how 39b can be ruled out in selected contexts. We maintain the standard assumption that selection involves properties of heads. If CP is selected in 39b, then there is a higher selecting head requiring that its complement's head be C°. This lexical requirement is met at D-structure but not at S-structure where the phrase's head is a C° and an 1° (under the standard definition of the "is-a" relation). So 39b, in a selected context, is ruled out by the Projection Principle.19 We thus derive condition (i)of37. This approach has a number of significant consequences. First, we account for the fact that V°-to-I° movement is typically not restricted to unselected domains, while I°-to-C° movement typically is.20 In our system, this difference follows from the fact that V° to 1° is usually an instance of the first type of incorporation described above, i.e., that which is triggered by a morphological subcategorization feature of an agreement or tense affix. In this case, the categorial status of the host head is not affected, and even if 1° were selected by C° (which it may or may not be) there would be no Projection Principle violation. This is why V°-to-I° movement systematically differs from I°-to-C° movement across languages. The second consequence is that 1° to C° is not necessarily excluded in all selected environments. If C° has the relevant morphological subcategorization feature, movement of 1° to C° would not involve substitution for C° and would not violate the Projection Principle. This appears to be the case in the instances of I°-to-C° movement attested in the Romance languages: Aux-to-Comp in Italian and the corresponding structure in inflected infinitives in Portuguese (cf. Rizzi 1982, Chs. 3 and 4; Raposo 1987). The Portuguese case is particularly telling: the construction
108
PARAMETERS AND FUNCTIONAL HEADS
involves an inflected verbal element in C° position in various kinds of infinitival complements, as in 40 (from Raposo 1987:98): (40) O Manel pensa terem os amigos t levado o livro. Manel thinks to-have-agr the friends taken the book 'Manel thinks that the friends have taken the book.' As this option is lexically selected (e.g., epistemic verbs allow it but volition verbs do not), a natural way to express this restriction is to say that epistemic verbs but not volition verbs select an embedded C° with an agreement morpheme, which in turn morphologically subcategorizes an 1° slot. Then movement of the inflected auxiliary to C° does not involve substitution for C° itself, and no problems arise with the Projection Principle. So this kind of I°-to-C° movement is allowed to apply in complement and other embedded contexts.21 To summarize, in this section we have proposed that the generalization underlying the restriction of complex inversion and subject-clitic inversion (and, more generally, I°-to-C° phenomena) to root contexts is 37. The second part of this generalization follows straightforwardly from the very idea that these processes involve I°-to-C° movement. We proposed that the first part is derived from the Projection Principle, once certain refinements are added to Baker's theory of head-to-head movement.22
5 Conclusion The analysis of complex inversion that we have proposed integrates a number of strands: the basic insights of Kayne's (1983) analysis, Chomsky's (1986b) extension of X-bar Theory, Baker's (1988) theory of head-to-head movement and the more elaborated proposals for the structure of clauses that have been made recently. We have shown how these strands can be drawn together so as to give a fairly complete analysis of complex inversion. Moreover, the analysis has led to a number of theoretical proposals; in particular, we have refined the theory of head-to-head movement by proposing that such movement is always substitution (perhaps with adjunction limited to cases of cliticization). Substitution can be into a slot provided by the morphological subcategorization of the host, or directly into the host head when the latter is empty. The second kind is properly restricted to root environments by a strict interpretation of the Projection Principle.
Appendix I: Embedded Subject-Aux Inversion in English Embedded Subject-Aux Inversion (SAI) is never found in indirect questions in English (*John wonders should he go to the store). However, SAI can be triggered by certain negative adverbials: (41) a. Never in my life have I been so insulted! b. Only in America could you get away with that.
COMPLEX INVERSION IN FRENCH
109
In certain embedded contexts, sentences of the type in (42) are possible (cf. Kayne 1982, 1983): (42) He said that under no circumstances would he do it. Two properties characterize this construction. First, that cannot be deleted: (43) ?*He said under no circumstances would he do it. Second, the complement is a weak island: (44) ?*What did he say that under no circumstances would he do? If we maintain that this type of inversion is an instance of I°-to-C° movement, as is clearly shown by the impossibility of SAI where if is present (see above), we have no alternative other than to treat these cases as instances of CP-recursion.23 We propose, therefore, that that has the marked property in English of selecting CP. Thus, if that is not present, a structure such as 43 can involve only one CP, where 1° to C° is excluded for the reasons we have presented. That this option is by and large restricted to that is shown by the deviance of recursion with other choices of C°. For example, the structure is impossible with a [+wh] C°: (45) *I wonder if/whether under no/any circumstances would John do that. The islandhood of these complements is explained by the CP-recursion idea, as the embedded clause in 44 would have a representation such as the following: (46) [CP t that [CP under no circumstances [c/ would [IP he t do t ] ] ] ] Extraction of the object in 46 would cross the lower tensed CP, which, in the system of Chomsky (1986b), has bounding properties akin to those of a standard w/z-island since its Specifier is filled by the negative adverbial.
Appendix II: On the Landing-site Problem The approach to head-to-head movement developed in section 4 allows us to elaborate a more principled solution to the landing site problem of complex inversion, which dispenses with the ad hoc step of C' adjunction (cf. section 3.3).24 The background is provided by the uncontroversial assumption that different kinds of heads license different kinds of specifiers: 1° licenses an A-specifier, C° licenses an A'-specifier, and so on. Let us now take seriously the idea, formulated in section 4, that the result of inversion is a clause headed by C° and by 1°. In that case, two specifier positions can be licensed: the typical specifier of C°, the landing site for w/z-movement, and the typical specifier of 1°, a subject position. Both positions are used in complex inversion: (47) Ou Jean [ [est-il] [t t alle t]
110
PARAMETERS AND FUNCTIONAL HEADS
If we look at the problem derivationally, as we have done throughout the paper, we can simply assume that, when the new head is created by I°-to-C° movement, the extra specifier position is automatically provided and made available for the lower subject to move into. Notice that this option never arises in cases involving incorporation qua substitution for a slot created via morphological subcategorization by the host head (that is, V-to-I movement does not create an extra position within IP corresponding to the V-specifier): in such cases the host head remains the only head of the construction after incorporation, and so no additional Spec position can be licensed. Only in the case in which incorporation involves substitution for the host head, i.e., I°-to-C° movement in root contexts, does the construction involve a genuine double head, and therefore a double specifier can be allowed. Moreover, this option is excluded in a language lacking subject clitics, such as English, for Case-theoretic reasons, as before (I has only one Case to assign, and so cannot Case-mark both its newly created specifier and the original specifier). The fact that the two specifiers are strictly ordered can now be related to the fact that a Case relation is involved only with one specifier: in 47, Jean must be adjacent (in the appropriate sense) to the head that assigns Case to it, hence ou cannot intervene. The C'-adjunction solution made crucial use of the A' status of the adjoined position to account for the incompatibility of complex inversion with wft-movement of the subject and stylistic inversion: (48) *Qui t est-il venu? (49) *Ou pro est-il alle Jean? This solution is no longer available within the more principled analysis that we are now adopting: if the NP position preceding the inflected verb is a legitimate I°-specifier, then it is an A-position, and 48 and 49 cannot be excluded as before because of the illicit A'-status of the variable and pro. A different approach is in order. Concerning 48, Marc-Ariel Friedemann (personal comunication) pointed out to us that this structure is independently ruled out by the ECP within the system of Relativized Minimality (Rizzi 1990), regardless of the A- or A'-status of the trace. In this system, traces must be properly head-governed, a requirement that is fulfilled for a subject trace in languages such as English or French by a C° agreeing with its Spec: (50) Qui C° [t est venu] (51) Who C°[t left] In 48 no such proper head governor can be provided for the trace of qui, as C° containing 1° is on the wrong side of the trace, hence the structure is ruled out by the ECP. As for 49, we can now elaborate on Sportiche's (1988b) approach to Case Theory presented in n. 4. If Case can be assigned under strict government or agreement, the choice of mode of assignment for each specific instance of Case being a parameter, then it is reasonable to look at the licensing of pro along the same lines. So, pro can
COMPLEX INVERSION IN FRENCH
111
be licensed under agreement from its licensing head (as is the case for subject pro in Italian) or under strict government (as is the case for object pro in Italian; cf. Rizzi 1986a). It appears that the non-argument pro responsible for stylistic inversion in French is licensed under strict government from C° (when additional conditions
are met): (52) Le jour [ou C° [pro est venu Jean] ] The day when came John But then pro cannot be licensed in a structure such as 49 where it would be, if anything, in an agreement configuration with the appropriate head, and would not be strictly governed by it. The important facts illustrated by 48 and 49 can thus be naturally reconciled with our more principled approach to the landing-site problem.
Notes 1. Thanks to Adriana Belletti, Anna Cardinaletti, and the audience at the Seminaire interdepartemental de recherche linguistique at the University of Geneva for their comments on an earlier version of this material. This paper was first published in Probus 1.1, 1989. We are grateful to Mouton de Gruyter for granting us permission to republish it here. 2. If subject pronouns occur in NP position in French, then a sentence such as: (i) Marie, elle parle toujours. Mary, she speaks always. must involve left dislocation. This is supported by the fact that quantified NPs, generally excluded in cases of left dislocation (cf. John/*Nobody, he's a nice guy), are in fact impossible in structures of this kind: (ii) *Personne, il n'est venu. No-one, he came. The corresponding case is possible in various northern Italian dialects: (iii) Gnun 1'adit gnent. (Piedmontese) No-one he has said nothing. 'No-one said anything.' This is expected: if the clitic is under Inft in (iii), gnun can appear in subject position, where quantified NPs are generally allowed to occur. See Rizzi (1986) for a detailed presentation of this argument. See also Renzi (1987) and Roberge (1986) for examples showing that certain dialectal varieties of French pattern with northern Italian dialects in this respect. 3. Pollock (1989), following Emonds (1978), shows that in French the leftmost verbal element must raise to 1° in tensed clauses. Such verb raising is impossible in (Modern) English for non-auxiliary verbs. 4. Alternatively, we could adopt the approach developed by Sportiche (1988b) (and also suggested by Jaeggli, personal communication) according to which Case can be assigned in one of two fundamentally different ways: either via government (defined in terms of strict c-command) or via Spec-head agreement. So, Objective and Oblique Cases are generally assigned via government by V or P, while Nominative Case is assigned via Spec-head agreement with 1° in declarative clauses in English and French (cf. also the earlier suggestion of Belletti and Rizzi 1981:125). As the mode of assignment for 1° must
112
PARAMETERS AND FUNCTIONAL HEADS
be subject to parametric variation in this system, one could then claim that 1° can assign Nominative Case both by agreement and by government in English, the latter mode of assignment being relevant in inverted clauses, while it can only assign Case via agreement in French. I°-to-C° movement destroys the Spec-head agreement configuration and makes Nominative assignment impossible in French in inverted clauses. One advantage of this approach is that it is relatively easy to see why a V° which has been raised to 1° (or C") may still assign Case to its object, while an l" which has been raised to C° has its Case-assignment capacity inhibited, as in French (this issue was raised by Alessandra Tomaselli, persona! communication): a raised V" still governs its object via Baker's (1988, ch. 2) Government Transparency Corollary, while a raised 1° is simply no longer in a Spec-head configuration with Spec-IP. Once raised, 1° can only Case-mark Spec-IP by government, an option which is unavailable in French. 5. According to (the obvious updating of) Kayne (1983), the cliticization of the pronominal subject to the inflected verb is allowed to apply in the syntax only when I-to-C movement takes place, as only in this case is the cliticization target higher than the subject pronoun. If the inflected verb does not move, cliticization in the syntax would be downgrading, hence the clitic trace would not be bound by the clitic. The process is then restricted to apply in the phonology in this case. Notice that even if the pronoun is cliticized in the syntax in 12, it still manifests an NP position in that it fills the subject position at D-Structure. 6. Nothing in what we have said rules out the comparable situation with objects, i.e., a structure like complex inversion involving an object pronoun and an object NP. In such a structure, the pronoun could satisfy Case Theory by incorporating with V while the NP is assigned Objective Case under government by V. We suggest that Case Theory actually allows this possibility, but that Theta Theory rules it out since V would have only one object theta-role to assign but two object arguments. The basic difference between the hypothetical object case and the attested subject case, then, is that object pronouns cannot be expletives in French (cf. Kayne 1983), while subject pronouns can. If also in River Plate Spanish, Rumanian, etc., object clitics cannot be expletives, as appears to be the case, then object-clitic doubling in these languages must involve the composition of two argument chains, in the sense of Chomsky (1986a), Rizzi (1987a). 7. On the fact that the expletive agrees with the argument here, but not in other constructions, see Kayne (1983:127-129). 8. Generating the pronoun and the NP the other way around in 17, i.e., with il in Spec-VP and Jean in Spec-IP at D-structure, gives rise to an S-structure which could satisfy Case Theory without I°-to-C() movement (the only movement needed would be incorporation of il with the inflected verb in I"). However, in such a sentence Theta Theory would be violated at D-structure, as the argumental NP occupies a rion-theta-position. 9. An example such as (i) is ruled out in English by the antecedent-government condition: (i) *A man seems that there was killed t. Here the chain (a man, there, t) is not well-formed because a man does not antecedentgovern there. The difference with the complex inversion example in 20 is that the raised NP antecedent-governs the clitic in 20. Recall that the configuration of 20 is impossible in English for Case reasons, as English pronouns do not incorporate. 10. We follow Belletti (1990) in assuming that AgrP dominates TP, while Pollock proposes that TP dominates AgrP. 11. If, because of its other virtues, we still want to adopt Pollock's proposed clause structure, we must explain why 24 is not an option for complex inversion. To get this result, it is enough to assume that one of the Spec positions in 24 is either absent or an A'-position, hence not available as the base position for il. The most plausible candidate for this is Spec-TP. If Spec-TP is not present, it obviously cannot: be occupied by il. If it is present
COMPLEX INVERSION IN FRENCH
113
but an A'-position, it could not be the base position of an expletive, since expletives belong to the A-system. So, il would have to be base-generated in the Spec-Agr position, which means that the representation in 25 could not arise since incorporation of il from Spec-Agr to Agr° would violate the ECP (see Baker 1988). 12. Another possibility which comes to mind is CP-recursion. This means that the structure of complex inversion would be as follows: (i) [CP1 WH [c, j Cl" [CP2 NP [C2 [C2» Vr] IP] ] ] ]
However, this proposal fails to account for nearly all the important properties of complex inversion. In particular, there would be no way to account for the root nature of the phenomenon (CP-recursion, if available, should be possible in both root and embedded contexts). So we reject this possibility. 13. This requires a version of the Relativized Minimality Principle (see Rizzi 1990), according to which subjects block antecedent-government not just in A-chains but in theta-chains, the latter also including some chains headed by an argument in an A'-position (cf. n. 16). The same reasoning extends to the case where the C'-adjoined position is occupied by a predicate or adjunct, assuming that such an element must be connected by a wellformed chain to its canonical functional position, and that the subject (or perhaps the main predicate; see Roberts 1988) is able to block antecedent-government in this case as well. 14. The presence of an object clitic on the verb in C° (as in *Pourquoi cela l'as-tu dit) does not save the proposed object, because object clitics are unable to be expletives in French (cf. n. 5), therefore a chain including the two arguments cela and le inevitably violates the Theta Criterion here (cf. Kayne 1983:117). 15. The fact that variables are restricted to A-positions is actually a subcase of the restriction of the features [ianaphoric, ipronominal] to A-positions, under the usual assumption that variables are defined in terms of this feature system. 16. It was proposed in Rizzi (1987a) that this approach also gives an account of the fact that pro cannot appear in Spec-CP and thereby fulfill the V2 requirement in German: (i) Gestern wurde pro getanzt. Yesterday was danced, (ii) Es wurde t getanzt. It was danced, (iii) *Pro wurde t getanzt. There is evidence that the element fulfilling the V2 requirement does not have to be phonetically realized, e.g., the empty operator involved inyes/no questions or the discoursebound empty operator discussed in Huang (1983) can fulfill the V2 requirement. Thus the phonetic emptiness of Spec-CP is not in itself the cause of the ungrammaticality of (iii). Rather, (iii) is excluded because pro cannot appear in an A'-position such as Spec-CP. 17. We allow the possibility that theta-chains can be headed by A'-positions, as is the case with the theta-chain headed by the subject NP in the C'-adjoined position in complex inversion (other cases would be clitic chains and the chains relating preposed initial arguments to their theta-positions in V2 structures). 18. What is the status of cliticization with respect to our proposals for head-to-head movement? There are two possibilities. On the one hand, we could treat cliticization on a par with Noun incorporation, by taking cliticization hosts to have an appropriate morphological subcategorization frame. For languages such as Romance, which have cliticization but not Noun incorporation, we can make the required categorial distinction by adopting the proposal made by Baker and Hale (1988) that pronouns are members of the category Determiner (D) (cf. Postal 1966). Cliticization hosts such as Romance Verbs (or perhaps Infl) would then have the specification [+D(1 —]. On the other hand, we could distinguish
114
PARAMETERS AND FUNCTIONAL HEADS
cliticization from other types of affixation by weakening the ban on head adjunction and maintaining that cliticization is the one case of head-to-head movement which involves adjunction rather than substitution. 19. We assume, with Chomsky (1965), that a positive specification of categorial selection in a lexical entry implies a negative value of all the non-occurring specifications. So [+ C°] implies, among other things, [-— I°], whence the desired result. This account further entails that there can be no operation of S'-deletion in the literal sense of elimination of the CP-level. If this were allowed, a predicate which selected CP at D-structure would select IP at S-structure and LF in a clear violation of the strong version of the Projection Principle required by our analysis. The obvious alternative is that "S'-deletion" verbs in fact select infinitival IPs at all levels. 20. For example, according to Pollock (1989),V"-to-I° movement in French takes place in both main and embedded clauses; the same is true for V°-to-l" movement in Italian (Belletti 1990), Middle English (Roberts 1985) and Vata (Koopman 1984). 21. There is another class of apparently non-selected CPs, relative clauses, pointed out by Bonnie Schwartz (personal communication). These clauses clearly strongly disallow inversions (*The man who do I know). While it may be possible to claim that restrictive relatives are in fact selected by the Determiner of the head, such an account does not seem viable for appositives, where inversion is equally impossible. This suggests that an extension of our approach is needed. The Projection Principle serves to maintain the semantics/syntax correspondence in cases of selection, but there is no doubt that this correspondence must be maintained in other cases too. In particular it is plausible to suggest that the predication function can only be fulfilled by certain categories (see the list given in Williams 1980). In that case, full relative clauses presumably must be CPs at LF in order to be licensed by predication. If this is so, then the same result obtains as in the case of selection: no substitution for C° would be possible, as the categorial status would be affected, thus preventing predication. The common factor behind relatives and indirect questions is, on this view, the fact that the Projection Principle and other well-formedness conditions on the syntax/semantics interface require that such clauses be projections of C" alone at the relevant syntactic levels. 22. A problem with this approach is posed by cases of embedded V2 in German. The usual [—wh] complementizer in German is daft. Unlike English that, daft is generally obligatory. Thus a normal case of [—wh] subordination features daft in the embedded C°, with the tensed Verb in final position in the lower clause. However, certain verbs of saying and thinking allow daft to be dropped, and this triggers V2 in the complement CP: (i) a. Ich sagte er hatte meine Frau gesehen. I said he had my wife seen, (ii) b. Ich glaube er mag tnich nicht. I think he likes me not. The CPs here are clearly complements to sagen and glauben, respectively. So we are apparently faced with an instance of I to C in a selected context. This phenomenon in fact lends prima facie support to our first suggestion concerning condition (i) of 37, in that we could claim that C" simply isn't filled here. Within the more principled approach involving the Projection Principle, we could explore the possibility that these examples involve incorporation triggered by the morphological subcategorization property of C°, as in the Romance cases discussed earlier. Alternatively, it could be the case that these structures are base-generated in extraposed position, hence the Projection Principle does not directly prevent categorial shift of an element in this position.
COMPLEX INVERSION IN FRENCH
115
23. CP recursion may also be in order to describe the colloquial varieties of French which allow subject clitic inversion in embedded interrogatives (Rene Amacker, personal communication). 24. Our proposal is conceptually close to Heider's (1987) Matching Projection approach, even if the two ideas are formally and empirically quite different.
References Baker, M. 1988. Incorporation: A Theory of Grammatical Function Changing. Chicago: University of Chicago Press. Baker, M.C. and K. Hale. 1988. "Pronoun and Anti-Noun Incorporation," ms, McGill University/MIT. Baker, M.C., K. Johnson and I. Roberts. 1989. "Passive Arguments Raised," Linguistic Inquiry 20:219-252. Belletti, A. 1990. Generalized Verb Movement. Aspects of Verb Syntax. Turin: Rosenberg and Sellier. Belletti, A. and L. Rizzi. 1981 "The Syntax of ne: Some Theoretical Implications," The Linguistic Review 1:117-154. den Besten, H. 1977/83. "On the Interaction of Root Transformations and Lexical Deletive Rules," ms, University of Amsterdam. Published (1983) in W. Abraham (ed.), On the Formal Syntax of the Westgermania. Amsterdam: John Benjamins. 47-131. Burzio, L. 1986. Italian Syntax: A Government-Binding Approach. Dordrecht: Reidel. Chomsky, N. 1965. Aspects of the Theory of Syntax. Cambridge, Mass.: MIT Press. . 1986a. Knowledge of Language: Its Nature, Origins and Use. New York: Praeger. . 1986b. Barriers. Cambridge, Mass.: MIT Press. Couquaux, D. 1986. "Les pronoms faibles sujet comme groupes nominaux," in M. Ronat and D. Couquaux (eds.), La Grammaire Modulaire. Paris: Les Editions de Minuit. 25-46. Emonds, J. 1976. A Transformational Approach to English Syntax. New York: Academic. . 1978. "The Verbal Complex of V'-V in French," Linguistic Inquiry 9:151-175. Freidin, R. 1978. "Cyclicity and the Theory of Grammar," Linguistic Inquiry 9:519-549. Goldsmith, J. 1981. "Complementizers and Root Sentences," Linguistic Inquiry 12:541-574. Huang, J. 1984. "On the Distribution and Reference of Empty Pronouns," Linguistic Inquiry 15:531-574. Kayne, R.S. 1972. "Subject Inversion in French Interrogatives," in J. Casagrande and B. Saciuk (eds.), Generative Studies in Romance Languages. Rowley, Mass.: Newbury House. 70-126. . 1982. "Predicates and Arguments, Verbs and Nouns," GLOW Newsletter 8:24. [Abstract of paper presented at the 1982 GLOW Conference.] . 1983. "Chains, Categories External to S, and French Complex Inversion," Natural Language and Linguistic Theory 1:109-137. Kayne, R.S. and J.-Y. Pollock. 1978. "Stylistic Inversion, Successive Cyclicity, and Move NP in French," Linguistic Inquiry 9:595-621. Kitagawa, Y. 1986. "Subjects in Japanese and English," Ph.D., University of Massachusetts, Amherst. Koopman, H. 1984. Verb-Movement and Universal Grammar: From the Kru Languages to Grammatical Theory. Dordrecht, Foris. Koopman, H. and D. Sportiche. 1985. "Theta Theory and Extraction," GLOW Newsletter 14:57-58. [Abstract of paper presented at the 1985 GLOW Conference.]
116
PARAMETERS AND FUNCTIONAL HEADS
. 1988. "Subjects," ms, University of California, Los Angeles. Kuroda, Y. 1986. "Whether we Agree or Not: Remarks on the Comparative Syntax of English and Japanese," ms, University of California, San Diego. Lasnik, H. 1985. "Illicit NP-movement: Locality Conditions on Chains?" Linguistic Inquiry 16:481-490. Lieber, R. 1980. "On the Organisation of the Lexicon," Ph.D., MIT. Manzini, M.-R. 1986. "Phrase Structure and Extraction," CLOW Newsletter 16:55-57. [Abstract of paper presented at the 1986 GLOW Colloquium.] Pollock, J.-Y. 1986. "Sur la syntaxe de EN et le parametre du sujet nul," in M. Ronat and D. Couquaux (eds.), La Grammaire Modulaire. Paris: Les Editions de Minuit. 211-246. . 1989. "Verb Movement, UG and the Structure of IP," Linguistic Inquiry 20:365-424. Postal, P. 1969. "On So-Called 'Pronouns' in English," in D. Reibel and S. Schane (eds.), Modern Studies in English. Englewood Cliffs, New Jersey: Prentice Hall. Raposo, E. 1987. "Case Theory and Infl-to-Comp: The Inflected Infinitive in European Portuguese," Linguistic Inquiry 18:85-110. Renzi, L. 1987. "I pronomi soggetto: un caso di parentela tipologica tra fiorentino e francese, e un capitolo poco noto di storia della lingua italiana," ms, Universita di Padova. Rizzi, L. 1982. Issues in Italian Syntax, Dordrecht: Foris. . 1986a. "Null Objects in Italian and the Theory of pro," Linguistic Inquiry 17:501-557. . 1986b. "On the Status of Subject Clitics in Romance," in O. Jaeggli and C. SilvaCorvalan (eds.), Studies in Romance Linguistics. Dordrecht: Foris. 391-420. . 1987a. "Three Issues in Romance Dialectology," talk presented at the GLOW Workshop on Dialectology, GLOW Colloquium, Venice. . 1987b. "On the Structural Uniformity of Syntactic Categories," paper presented at the Second World Basque Conference, San Sebastian, September 1987. . 1990. Relativized Minimality. Cambridge, Mass.: MIT Press. . 1991. "On the Status of Referential Indices," in A. Kasher (ed.), The Chomskian Turn. Oxford: Blackwell. 273-299. Roberge, Y. 1986. "The Syntactic Recoverability of Null Arguments," Ph.D., University of British Columbia. Roberts, I. 1985. "Agreement Parameters and the Development of English Modal Auxiliaries," Natural Language and Linguistic Theory 3:21-58. . 1988. "Thematic Minimality," ms, Universite de Geneve. Safir, K. 1981/82. "Inflection-Government and Inversion," The Linguistic Review 1:417-467. Safir, K. and D. Pesetsky. 1981. "Inflection, Inversion and Subject Clitics", Proceedings of NELSll. 331-344. Sportiche, D. 1988a. "A Theory of Floating Quantifiers and Its Corollaries for Constituent Structure," Linguistic Inquiry 19:425-449. . 1988b. "Conditions on Silent Categories," ms, University of California, Los Angeles. Szabolcsi, A. 1983. "On the Non-Unitary Nature of Verb-Second," ms, Max-Planck Insitute for Psycholinguistics, Nijmegen. Vergnaud, J.-R. 1985. Dependances et niveaux de representations en syntaxe. Amsterdam: John Benjamins. Williams, E. 1980. "Predication" Linguistic Inquiry 11:203-238. Zagona, K. 1982. "Government and Proper Government of Verbal Projections," Ph.D., University of Washington, Seattle.
4
Negative Concord in West Flemish Liliane Haegeman and Raffaella Zanuttini
1 Introduction: negative concord vs. double negation In this paper1 we will be looking at the phenomenon of negative concord. After a general description of the data of negative concord in Romance languages we will concentrate on its properties in one Germanic language, West Flemish (a Belgian dialect of Dutch). When two negative elements are present in a given syntactic domain, two different situations may arise: (i) the two negative elements may cancel each other out, or (ii) the two negative elements may constitute, together, one single instance of negation. The former case, where two negative elements cancel each other out, is referred to in logic as double negation: (1) -i[-ip] = p It is exemplified at the level of the sentence in standard English, standard German, and standard Dutch in 2: (2) a. Ididn'f say nothing, (standard English) b. Ich habe nicht nichts gesagt. (standard German) I have not nothing said c. Ik heb niet niets gezegd. (standard Dutch) I have not nothing said Double negation readings are also found in syntactic domains smaller than a sentence, e.g., within an NP, as in 3, where the negative marker not and the prefix un- cancel each other out:2 (3) [ NP A not Mnfriendly man] walked into the room. The second case, where two (or more) negative elements co-occurring in the same sentence do not cancel each other out, but together yield one single instance of 117
118
PARAMETERS AND FUNCTIONAL HEADS
negation, is typical of the Romance languages, as illustrated by the examples in 4. In these languages the negative marker (ne, non, etc.) and the negative quantifier within the VP (personne, nessuno, etc.3) do not cancel each other out, but rather yield a proposition with one instance of negation only. Similarly, as illustrated by the examples in 5, multiple occurrences of negative quantifiers (personne and rien, nessuno and niente, etc.) constitute only one instance of negation. We will call this phenomenon negative concord (NC), to distinguish it from the phenomenon of double negation (DN). (4) a. Je n'ai v\\personne. (French) b. Non o visto nessuno. (Italian) c. No he visto a nadie. (Spanish) d. No he vist ningu. (Catalan) e. Nao vi ninguem. (Portuguese) f. Nu am vazut pe nimeni. (Romanian) NEG have seen nobody 'I haven't seen anybody.' (5) a. Personne n'a rien dit. (French) b. Nessuno ha detto niente. (Italian) c. Nadie ha dicho nada. (Spanish) d. Ningu (no) ha dit res. (Catalan) e. Ninguem (nao) disse nada. (Portuguese) f. Nimeni nu a zis nimic. (Romanian) nobody has said nothing 'Nobody said anything.' In this paper we will discuss the interpretation of multiple occurrences of negative elements in West Flemish (WF).4 We will show that the co-occurrence of negative elements in this language sometimes yields double negation parallel to the examples from Germanic languages (as in 2), while other instances yield negative concord analogous to the Romance examples in 4 and 5. In our analysis we concentrate on NC and provide a characterization of the conditions which determine it, showing that: a. NC in WF presents some interesting similarities to NC in Romance; in particular, in both cases it correlates with the presence of a negative marker which is the head of NegP and precedes the finite verb; b. there is clear evidence (island effects, extraction from PP) that NC readings are derived via LF movement; c. NC readings are restricted to negative constituents in certain syntactic configurations;
NEGATIVE CONCORD IN WEST FLEMISH
119
d. there are constituent-level constraints determining which negative elements can participate in an NC reading. Such constraints will be expressed in terms of features on the heads of the constituents. e. the relation between the negative head and the negative constituents in a clause is subject to the Neg-Criterion, a well-formedness condition similar to Rizzi's M-Criterion (1990b, 1995). Before entering the discussion of WF (sections 3 and following), we will summarize the main characteristics of NC in Romance and compare Romance-type languages with Germanic-type languages (section 2). We will formulate a hypothesis about what it is that distinguishes Romance-type languages with NC from Germanic-type languages without it.
2 Negative concord in Romance 2.1
Three properties characterizing Romance
Three different strategies for the expression of sentential negation are found in Romance languages. a. Sentential negation can be expressed by means of a negative marker which precedes the finite verb in linear order (e.g., Italian non, 6a). Such a negative marker can be adequately described as the head of a functional category NegP5 and, as shown by the example, may in itself constitute the marker of sentence negation. b. Alternatively, sentential negation can be expressed by means of two negative markers, one preceding and the other following the finite verb, as in French ne and pas (6b). In this case, the pre-verbal negative marker does not suffice for the expression of sentence negation, and the post-verbal element is obligatory. Indeed, the pre-verbal negative marker can sometimes6 be omitted, in which case the postverbal negative marker on its own may express sentential negation. If no other negative constituent is present in the clause, pas is normally necessary to express sentential negation. c. Finally, sentential negation can also be expressed by means of a postverbal negative marker on its own, as in Piedmontese (6c), and in many dialects of Northern Italy, Southern France, and some Romansch dialects of Switzerland. (6) a. Non mangia. (Italian) b. II (ne) mange *(pas). (French) c. A mangia nen. (Piedmontese) 'He doesn't eat.' Let us point out two characteristics of languages which employ the strategy exemplified by Italian in 6a, that is, languages which express sentential negation by means of a pre-verbal negative marker alone.
120
PARAMETERS AND FUNCTIONAL HEADS
a. In all the Romance varieties of this type, to the best of our knowledge, the pre-verbal negative marker enters in a relation of negative concord with other negative elements present in the sentence, as exemplified in 4b-f. Note that this property is not necessarily shared by the post-verbal negative markers in languages of the second (French) or third (Piedmontese) type, as we can see in 7, where French pas is completely unacceptable and Piedmontese nen is marginal: (7) a. II n'a (*pas) vupersonne. (French) b. A l ' a ( ? ? nen) vist gnun. (Piedmontese) he has NEC seen nobody b. Every language that expresses sentential negation by means of a pre-verbal negative marker alone shows the following constraint on the expression of sentential negation: negation can take sentential scope only if it is marked in a position c-commanding Infl. In other words, a sentence in English (for example, 8a), where a negative quantifier is within the VP and assigns sentential scope to negation, is not grammatical in a language such as Italian (8b), It becomes grammatical only if the negative quantifier within VP is preceded by a negative element in a position higher than Infl, as in 8b: (8) a. I saw nothing. b. *(Non) ho visto niente. (Italian) The relevant pattern is illustrated in 9, 10 and 11, with examples from Italian. In 9 we see that a negative quantifier alone can express sentential negation when it is structurally higher than Infl: it can be in subject position (9a), or in a topic position (9b). But when it occurs in a position lower than Infl, as in 10, it cannot be the only negative element in the structure. In this case, it must co-occur with another negative element which c-commands Infl: either the pre-verbal negative marker (1 la), or a negative constituent in subject position (lib), or one in a topicalized position (1 Ic): (9) a. Nessuno ha telefonato. 'Nobody has called.' b.
Niente rni ha raccontato di tutto cio. nothing me has told of all that 'He has told me nothing about all that.'
(10) a. *Ha telefonato nessuno. has called nobody b. *Mi ha raccontato niente di tutto cio. me has told nothing of all that (11)
a. Non ha telefonato nessuno. non has called nobody 'Nobody has called.'
NEGATIVE CONCORD IN WEST FLEMISH
121
b. Nessuno mi ha raccontato niente di tutto cio. nobody me has told nothing of all that 'Nobody has told me anything about that.' c.
Mai mi ha raccontato niente di tutto cio. never me has told nothing of all that 'He has never told me anything about all that.'
These data suggest that Romance languages which are like Italian differ from the Germanic languages mentioned above in three respects: they have a pre-verbal negative marker which is a head (in terms of X-bar Theory), they have negative concord, and they must mark negation in a position c-commanding Infl for it to take sentential scope. The interesting question that arises is whether these properties go together, namely: a. the property of expressing negation via a pre-verbal negative marker, which is a head and which alone may take sentential scope; b. the property of showing NC; c. the property of requiring that negation be expressed in a position c-commanding Infl to take sentential scope. In the next section we will examine the correlation between these properties, focusing on the correlation between the presence of a Negative head and the phenomenon of NC. In a later section of the paper (6.2.4) we return briefly to the descriptive generalization raised in c, above.
2.2
Hypothesis
2.2.1
Romance-type languages
Let us suppose that in Romance languages like Italian there is a privileged locus for the expression of sentential negation, and that this is the functional projection NegP, whose head Neg c-commands Infl.7
We will try to generalize the properties identified for Italian in the previous section to all Romance languages with NC. Let us suppose that there is a configurational constraint on sentential negation in Romance: negation can only take sentential scope if it is either marked by the head of NegP itself, or is in a position c-commanding the head of NegP. It follows, then, that sentential negation can be expressed either via the pre-verbal negative marker itself—the head of NegP — or by a negative quantifier in subject position,8 or by a topicalized negative constituent. On the other
122
PARAMETERS AND FUNCTIONAL HEADS
hand, a negative constituent in a lower position in the structure will not be able to express sentential negation, since it will not meet the structural constraint concerning c-command of Infi discussed in section 2.1.9 The relevance of this configurational requirement to the presence of NC is then as follows. Once a sentence contains a negative element as the head of NegP or a negative constituent in a position c-commanding NegP, negation can take sentential scope. The presence of other—lower — negative constituents does not alter this state of affairs. Let us say that all the negative constituents which are present in the sentence are linked somehow to the head of NegP; we will refer to such linking by using the merely descriptive term "negative chain". In sentences with multiple negative elements, the negative elements enter into a negative chain with the element which marks sentential negation. The negative chain constitutes one instance of sentential negation, irrespective of how many members it contains. When the negative constituents forming a negative chain with the c-commanding sentential negative marker are subject to QR at LF, they contribute only their quantificational force to the interpretation of the sentence, not their negative meaning. So, the LF representation of a sentence such as 13 will be the one in 14:10 (13) Non ho mai detto niente a nessuno. non have never said nothing to nobody 'I've never told anybody anything.' (14) [Va;, y, t: x a thing & y a person & t a moment in time], -i [I said x to y at t] Can we generalize the properties identified for Italian to other Romance languages? If we look at languages such as French, which employ the second of the strategies described for expressing sentential negation (cf. section 2.1), we see that they show the first two properties described for languages such as Italian: they have a pre-verbal negative marker, and they have NC. There are differences between the pre-verbal negative marker in French and that in Italian, though. First, observe that while in Italian non suffices for the expression of sentential negation, in French ne does not and requires the presence of another negative constituent. We postulate therefore that ne is a scope marker: rather than carrying the negative force on its own (like non in Italian), it signals that the negative element with which it is (obligatorily) associated has sentential scope. Ne and non are similar in that they both take sentential scope: non itself is a negative constituent with sentential scope, ne is associated with a negative constituent with sentential scope. We will say that ne is a scope marker. What about the third property, the structural requirement on the position of negation? The acceptability of a sentence such as 15a alongside 15b reveals a second — surface — difference between French-type languages and Italian-type languages: when ne is optionally omitted in 15a there is no overt negative element c-commanding Infl, and yet negation has sentential scope: (15) a. J'ai rien vu. (French) I have nothing seen
NEGATIVE CONCORD IN WEST FLEMISH
123
b. Je«'ai rienvu. I NEC have nothing seen 'I have seen nothing.' However, it seems preferable to assume that languages such as French, though employing a different strategy for marking sentential negation, share the same structural properties of Italian-type languages, as well as the mechanism for NC-interpretation. In other words, we propose that they also have a negative head in a position ccommanding Infl, and the only difference is that this head has the option of being either lexically realized or empty. If we assume that the head of the NegP is always there, then we can suggest that it is its presence which triggers NC, similar to what happens in Italian. French-type languages will be argued to share the configurational constraint on the expression of sentential negation shown by Italian-type languages with one major difference: the configurational constraint on the expression of sentential negation can be satisfied by a lexically empty negative head. We now extend the reasoning adopted for French to Piedmontese-type languages, which differ from French in that they do not have an overt negative head, not even optionally (cf. example 6c). Piedmontese-type languages will also be said to have a negative head in a position c-commanding Infl, though this is at a more abstract level: the negative head is non-overt. There are in fact varieties of Piedmontese where the negative head is lexically realized and sentential negation is expressed by the discontinuous constituent n ... nent, which strongly resembles French ne ... pas. For a detailed discussion of one such variety, spoken in Cairo Montenotte, see Parry (1985). As was the case for French, we assume that the non-overt negative head is not itself the carrier of negative meaning, but its role is to assign sentential scope to the negative constituent with which it is associated. Piedmontese also has NC: (16) GnunaVhadiignente. (Piedmontese) nobody Cl has said nothing 'Nobody said anything.' 2.2.2
Germanic-type languages
Now we are equipped to characterize more precisely the difference between Romancetype languages and the Germanic-type languages such as standard German and Dutch, with respect to the interpretation of multiple occurrences of negation. We have seen that the latter differ from Romance with respect to the three properties described in section 2.1. a. They do not have a negative marker which is a head associated with Infl, whether this be a negative head which in itself expresses negation (as in Italian) or a head which assigns sentential scope to other negative constituents, i.e., a scope marker (as in French or Piedmontese). b. They do not have negative concord. c. They do not require that sentential negation be marked in a constituent ccommandinelnn 1 .
124
PARAMETERS AND FUNCTIONAL HEADS
The negative markers in Germanic languages, as illustrated by German nicht, Dutch niet, Swedish ikke, etc." will be taken to be adverbial elements. We assume that they are adjoined to VP or to a projection dominating VP. We tentatively suggest that these languages in fact lack a negative head projecting a NegP dominating Infl. 12 As a result of the absence of such a NegP, the chain formation process described for Romance languages in section 2.2.1 does not apply to the negative constituents in a sentence. When the sentential negative marker and the negative quantifiers raise at LF, they are not linked; rather, they contribute to the interpretation of the sentence both their quantificational force and their negative meaning. The LF representation of a sentence such as 2 (repeated here as 17) is given in 18: (17) a. Ich habe nicht nichts gesagt. (standard German) b. Ik heb niet niets gezegd. (standard Dutch) (18) -.[Vz-. [I said x] ] Similarly, when there are two negative quantifiers, they also contribute two instances of negation to the interpretation of the sentence, as in 19 and 20: (19) a. Niemand hat nichts gesagt. (standard German) b. Niemand heeft niets gezegd. (standard Dutch) (20) Vx(xa, person)->[Vj/(2/a thing)-ifxsaidy] ] Let us summarize our discussion so far: in languages with a negative head, Neg, projecting a NegP dominating Infl, negation can take sentential scope only if it is associated with the head of NegP,13 or if it is carried by a constituent c-commanding Neg. All the other negative constituents present in the sentence will be linked to the highest negative element which has sentential scope and they do not contribute their own negative force to the sentence. This situation yields what we have called NC readings. If, on the other hand, a language lacks a negative head projecting a NegP dominating Infl, each negative constituent raises at LF, contributing its own negative force to the interpretation of the sentence. This yields double negation readings. This is a very strong hypothesis, which captures the correlation observed in Romance between the presence of negative concord and the requirement that negation be associated with a position higher than Infl, and which offers a way to characterize the difference between Romance and Germanic-type language, or rather—to be more precise — the difference between languages with NC readings for multiple negation and those without NC readings. Our hypothesis is supported by the observation that earlier stages of English, Dutch, and Scandinavian showed both the presence of NC and the presence of a negative marker of the type head in a position preceding the finite verb (cf. Jack 1978; Jespersen 1965). WF is particularly interesting because it is a Germanic language but differs from standard German and Dutch in that it has NC readings as well as double negation readings of multiple negation. 14 An investigation of instances of multiple negation in WF will shed light on the structural conditions under which either the NC reading
NEGATIVE CONCORD IN WEST FLEMISH
125
or the DN reading is available, and will enable us to formulate the mechanism that generates NC readings more carefully. In our analysis we will first describe simple cases of sentential negation in WF (section 3). Secondly, we will show that WF allows NC, given certain syntactic configurations, and we will introduce the factorization process which generates the NC reading (section 4). Finally, we will give an informal description of the mechanism which generates the NC readings: we will establish that at LF the negative constituents which yield NC move step-wise and cluster at the level of VP or a projection immediately dominating VP, and we will argue that the negative component is factored out of the negative cluster (section 5). We will see that negative constituents with sentential scope are related to the head of NegP by means of either LF movement or by Spec-head agreement at S-structure and we will interpret this in terms of the Neg-Criterion, a well-formedness condition on negative constituents (section 6). In Section 7 further constraints on the formation of the negative chain will be examined, and we will compare the WF data with those of Bavarian. The main findings of our research will be summarized in section 8.
3 Sentential negation in West Flemish 3.1
The negative marker nit
The WF negative marker nie is at first sight parallel to the standard German negative marker nicht and the standard Dutch negative marker niet: it occurs to the left of the finite verb in embedded clauses (21a, c) and to its right in main clauses (21b, c), where Verb Second15 has moved the finite verb under C (we have italicized the finite verb): (21) a. da Valere dienen boek nie eet that Valere that book not has 'that Valere doesn't have that book' b. Valere eet dienen boek nie. Analogous to its counterparts in standard German and Dutch, we propose that WF nie can best be described as an adverbial element adjoined to VP or to a projection dominating VP.16 The direct object dienen boek precedes nie in 21 as a result of scrambling. Despite this structural similarity among the three negative markers, WF nie differs from the other two in that it co-occurs with negative constituents within VP and allows NC (as we will see in sections 4 and 5, below).
3.2 Negative constituents As is the case in standard German and Dutch, sentential negation in WF may also be expressed by means of a negative constituent on its own. We distinguish between bare negative constituents, i.e., elements such as niemand 'nobody', niets 'nothing', nieverst 'nowhere' (22), and constituents which are negative by virtue of the presence
126
PARAMETERS AND FUNCTIONAL HEADS
of the quantifier geen, 'no', i.e., elements such as geen mens 'no person' and geen geld, 'no money', which we will label "geen-NPs" for ease of reference (23):lv (22) a. da Valere niemand kent that Valere nobody knows 'that Valere doesn't know anybody' b. da Valere dienen boek nieverst vindt that Valere that book nowhere finds 'that Valere doesn't find that book anywhere' c. da Valere tegen niemand klaapt that Valere against nobody talks 'that Valere doesn't talk to anyone' (23) a. da Valere geen geld eet that Valere no money has 'that Valere doesn't have any money' b. dat Valere geen broers eet that Valere no brothers has 'that Valere doesn't have any brothers' c. da Valere tegen geen mens klaapt that Valere against no person talks 'that Valere doesn't talk to anyone' It is a general property of WF that only definite NPs may appear in the canonical subject position:18 whenever the subject NP is indefinite, the existential construction with der is obligatory:19 (24) a. *da nen student da gezeid eet that a student that said has b.
dat-ter nen student da gezeid eet that der a student that said has 'that a student said that'
(25) a. *Ee nen student da gezeid? has a student that said b.
Eet-ter nen student da gezeid? has der a student that said 'Did a student say that?'
When negative quantifiers are subjects they behave like indefinite NPs and the existential construction with der is obligatory.20 (26) a. *da niemand dienen boek gelezen eet that nobody that book read has
NEGATIVE CONCORD IN WEST FLEMISH
b.
127
dat-ter niemand dienen boek gelezen eet that der nobody that book read has 'that nobody has read that book'
(27) a. *da geen studenten dienen book gelezen een that no students that book read have b.
dan-der geen studenten dienen boek gelezen een that der no students that book read have 'that no students have read that book'
There is then a structural difference to be noted between WF negative constituents and those in Romance languages. In the Romance languages negative subjects may occur in the subject position (cf. 5); in WF they can't. However, this is due to reasons which are not related to the negative component of the subject but rather to its indefinite nature. Note also that the ban on having negative subjects in the canonical subject position is not general for Dutch: in standard Dutch the equivalents of 26a and 27a are grammatical:21 (26') dat niemand dat boek gelezen heeft that nobody that book read has 'that no one has read that book' (27') dat geen studenten dat boek gelezen hebben that no students that book read have 'that no students have read that book' 3.2.1
The negative head en
The data discussed so far seem to suggest that WF is similar to standard German and Dutch in expressing sentential negation by means of an adverbial-type negative marker and in allowing a negative VP-constituent to express sentential negation. But WF in fact differs from these languages, and resembles instead what we have called Romance-type languages, in showing the (optional) presence of a negative marker of clitic nature—en — in co-occurrence with a negative constituent.22 This element can co-occur with the negative marker nie (21'), with bare negative quantifiers (22'), and with geen-NPs (23'). En cliticizes onto the finite verb and moves with the finite verb under Verb Second (cf. 21b', 21d', 22d' and 23d'): (21') a. Valere dienen boek nie (e«)-eet that Valere that book not en has 'that Valere doesn't have that book' b. Valere (ew)-eet dienen boek nie. Valere en has that book not 'Valere doesn't have that book.' c. da Valere dienen boek nie (en)-wilt kuopen that Valere that book not en wants buy 'that Valere doesn't want to buy that book'
128
PARAMETERS AND FUNCTIONAL HEADS
d. Valere (