Parameters and Functional Heads: Essays in Comparative Syntax (Oxford Studies in Comparative Syntax)

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 (
a. da Valere niemand (en)-kent that Valere nobody en knows 'that Valere doesn't know anybody' b. da Valere dienen boek nieverst (en)-vindt that Valere that book nowhere en finds 'that Valere doesn't find that book anywhere' c. da Valere tegen niemand (en)-klaapt that Valere against nobody en talks 'that Valere doesn't talk to anybody' d. Valere (en) klaapt tegen niemand. Valere en talks to nobody 'Valere doesn't talk to anybody.'

(23') a. da Valere geen geld (en)-eet that Valere no money en has 'that Valere doesn't have any money' b. dat Valere geen broers (en)-eet that Valere no brothers en has 'that Valere doesn't have any brothers' c. da Valere tegen geen mens (en)-klaapt that Valere against no person en talks 'that Valere doesn't talk to any person' d. Valere (en)-klaapt tegen geen mens. Valere en talks to no person 'Valere doesn't talk to anybody,' The examples in (21'), where en co-occurs with the negative adverbial nie, are strongly reminiscent of the strategy for expressing sentential negation found in French, where the pre-verbal negative head ne optionally co-occurs with the adverbial negative marker pas.23Like French ne (and Italian nori), WF en co-occurs with other negative constituents (22, 23), and moves with the finite verb.24 WF en shares with French ne the property that it cannot be the only marker of sentential negation.25 (21") a. *da Valere dienen boek en-eet b. *Valere en-eet dienen boek. c. *da Valere dienen boek en-wilt kuopen d. *Valere en-wilt dienen boek kuopen.

NEGATIVE CONCORD IN WEST FLEMISH

129

Following the discussion of the French examples, we will assume that WF en is a scope marker for negation: it serves to signal that the negative constituent with which it is associated has sentential scope. We return to the relation between the negative head and the associated negative constituents in detail in section 6. As seen above, WF en shares some of properties which characterize negative markers of the type head in Romance, which leads us to the hypothesis that it can be adequately described as heading a functional projection (NegP) dominating IP (cf. 12). In section 6 we will see that, in fact, the restrictions on the distribution of en must be made more precise: en can occur when another negative constituent with sentential scope is present in the sentence, but only under certain structural conditions. We assume that when en is present, the sentence contains a functional projection NegP, generated in a position dominating IP. When en is absent, such a functional projection could either be absent or else there might be a functional projection with a phonologically empty head. Following the reasoning applied to Romance (more specifically languages such as French or Piedmontese), in this paper we postulate that a functional category NegP is always present in WF finite clauses, even when its head is void of phonological content. This amounts to introducing an important difference between WF and other Germanic languages such as standard German and Dutch: while the latter mark sentential negation only by means of a negative marker of the type adverb which is (in first approximation) adjoined to VP, WF expresses sentential negation by means of both a functional category NegP and a negative element adjoined to VP or a projection dominating VP (be it the negative marker nie or a negative quantifier—see section 4.1). The examples in 22,23,26b and 27b then, do not necessarily show that WF is like standard Dutch in allowing a negative marker inside VP to express sentential negation by itself. If these sentences contain a NegP whith a phonologically empty head, then their structure resembles that of Romance negative clauses, where a negative quantifier with sentential scope in a position lower than Infl always co-occurs with a NegP dominating Infl. In our introductory discussion of Romance, we noted a correlation between the presence of the functional category NegP and NC. In this language family, all the languages that express sentential negation by means of a negative marker which is the head of NegP also show NC. Having assumed that WF en is the head of NegP, it is interesting to ask whether this language shows NC. In the next section, by investigating multiple occurrences of negative quantifiers, we will show that it does.

4

Negative concord in WF

When more than one negative element (e.g., the negative adverb nie and/or one or more negative quantifiers) is present in a sentence in WF, either an NC reading or a DN reading will obtain, depending on the configuration. In this section we will examine the configurations which determine these two different readings. First, we will discuss the position of the negative adverb nie and negative quantifiers. Then we will show that NC always obtains26 when two (or more) negative elements are

130

PARAMETERS AND FUNCTIONAL HEADS

dominated by the same projection. Finally, we will discuss the mechanism that generates NC readings.

4.1

The data

Contrary to standard Dutch, where two occurrences of negative elements within VP yield DN readings, in WF the co-occurrence of the VP-adjoined negative adverb nie with one or more negative quantifiers dominated by the same maximal projection, or the co-occurrence of two or more negative quantifiers dominated by the same maximal projection often yields an NC reading, i.e., an instance of one negation only. In 28 we illustrate NC involving the negative adverb nie;27 in 29 NC affects bare negative quantifiers; in 30 it affects geen-NPs; and in 31 it affects a bare negative quantifier and a geen-NP.28 In section 5 we will return in detail to the typology of negative constituents and we will discuss the co-occurrence constraints on the NC reading of negative constituents. Examples 28d, 29d, 30b, 31c and 31d show that negative subject-NPs, which necessitate the use of existential construction with der, also enter into NC patterns: (28) a. K'(en)-een niets nie gezien. I en have nothing not seen 'I haven't seen anything.' b. K'(en)-een t niemand nie gevroagd. I en have it nobody not asked 'I didn't ask anyone.' c. K'(en)-een nieverst nie geweest. I en have nowhere not been 'I haven't been anywhere.' d. Gisteren (en)-eet-ter niemand nie geweest. yesterday en has der nobody not been 'Nobody came yesterday.' (29) a. K'(en)-eenan niemand niets gezeid. I en have to nobody nothing said 'I didn't say anything to anyone.' b. Valere (en)-ee nieverst niemand gezien. Valere en has nowhere nobody seen 'Valere did not see anyone anywhere.' c. Valere (en)-ging nooit nieverst noatoe. Valere en went never nowhere to 'Valere never went anywhere.' d. Gisteren (en)-eet-ter niemand niets gezeid. yesterday en has der nobody nothing said 'Nobody said anything yesterday.'

NEGATIVE CONCORD IN WEST FLEMISH

131

(30) a. Valere (en)-wilt an geen mens geen geld geven. Valere en wants to no person no money give 'Valere doesn't want to give any money to anyone.' b. Gisteren (en)-oat-ter ier geen mens geen tyd. yesterday en had der here no person no time 'Yesterday nobody had any time at all.' (31) a. Valere (en)-leest nooit geen boeken. Valere en reads never no books 'Valere never reads any books.' b. Valere (en)-wilt an niemand geen geld geven. Valere en wants to nobody no money give 'Valere doesn't want to give any money to anyone.' c. Gisteren (en)-oat-ter ier niemand geen geld. yesterday en had der here nobody no money 'Yesterday nobody had any money.' d. Gisteren (en)-eet-ter hier geen mens niets gezeid. yesterday en has der here no person nothing said 'Nobody said anything round here yesterday.'

4.2

Scope, NC and the position of negative constituents29

4.2.1

The scope of negative VP constituents

Consider first the following examples, which contain a single negative quantifier (niets) and a reason adverbial (doavuoreri): (32) a. da Valere niets doavuoren gedoan eet that Valere nothing for that reason done has 'there was nothing which Valere did for that reason' b. da Valere doavuoren niets gedoan eet 'that because of this Valere did nothing' Both sentences are grammatical, but with a different interpretation. In 32a the negative quantifier niets has been scrambled out of its base position and it must have wide scope with respect to the reason adjunct doavuoren: 'there is nothing which Valere did for that particular reason', i.e., 'he did everything for another reason'. In the unmarked interpretation of 32b30 niets has narrow scope with respect to the reason adjunct: 'Valere's failure to act is due to that reason'. We propose that in order to take wide scope, negative constituents must be scrambled; when negative VP constituents remain in their base position, their scope is strictly local. Negative constituents cannot precede sentential adverbs such as woarschijnlijk, which are adjoined to a projection dominating VP:31 (33) a. da Valere woarschijnlijk geen geld eet that Valere probably no money has 'that Valere probably has no money'

1 32

PARAMETERS AND FUNCTIONAL HEADS

b. *da Valere geen geld woarschijnlijk eet (34) a.

da Valere woarschijnlijk niemand gezien eet that Valere probably nobody seen has 'that Valere probably did not. see anyone'

b. *da Valere niemand woarschijnlijk gezien eet A first conclusion is that though negative VP constituents are scrambled out of their base position, they cannot be adjoined too high in the tree. They are scrambled to a position outside VP but remain lower than the position to which sentential adverbials are adjoined. 4.2.2

Me and VP constituents

Let us also examine the position of nie with respect to VP complements. As a first approximation we assume that nie is adjoined to VP or to a projection dominating VP and that nie itself cannot move. Observe the position of a non-negative constituent with respect to nie, e.g., the PP doarover in the following examples: (35) a. da Valere woarschijnlijk nie doarover klaapt that Valere probably not thereabout talks 'that Valere probably does not talk about that.' b. da Valere woarschijnlijk doarover nie klaapt c. da Valere doarover woarschijnlijk nie klaapt In 35a doarover is taken to be in its base position; in 35b it is scrambled and precedes nie; in 35c it is scrambled and adjoined to a higher projection, say IP. Extraction of the R-pronoun doar is possible from the base position in 35a, but it is not possible from the adjoined positions in either 35b or 35c: (36) a. de dingen woaj da Valere woarschijnlijk nie t; over klaapt the things where that Valere probably not about talks 'the things that Valere probably does not talk about' b. *de dingen woaj da Valere woarschijnlijk t; over nie klaapt c. *de dingen woa^ da Valere tj over woarschijnlijk nie klaapt The data in 36 are not new. Koster (1987:174-181) shows that extraction of an R-pronoun from a PP in a derived position leads to ungrammaticality.32 The ungrammaticality follows if we assume that the moved PP in the adjoined position is not L-marked, hence a barrier. We conclude that the PP doarover in 35a is in its base position, since it allows extraction, while it is in a derived position in 35b and 35c, blocking extraction. Extending the analysis we will propose that whenever a VP constituent precedes nie it has been scrambled out of the VP and adjoined to a higher projection.

NEGATIVE CONCORD IN WEST FLEMISH

4.2.3

133

NC with me

We observe that in order to be able to enter into an NC relation with the sentential negator, nie, negative quantifiers must precede it: (37) a. da Valere niets nie kuopt that Valere nothing not buys 'that Valere doesn't buy anything' b. da Valere nie niets kuopt that Valere not nothing buys 'that Valere does not buy nothing' Example 37b cannot have the NC reading but it may have a DN interpretation: the wide scope negation in nie cancels the narrow scope negation in niets. Recall that nie is a marker of sentential negation — as such it takes wide scope. Recall also that if a negative VP complement needs to take wide scope it must be scrambled (cf. 32). Let us postulate a first condition on NC readings: in order for two elements to enter into an NC relation, they must have identical scope. A negative VP constituent which enters into an NC relation with the sentential negation nie must itself be able to take wide scope, hence it must scramble.33 Example 38 illustrates NC with three constituents: (38) a. da Valere an niemand niets nie gezeid eet (NC) that Valere to nobody nothing not said has 'that Valere didn't say anything to anyone' b. da Valere an niemand nie niets gezeid eet34 (DN) c. da Valere nie an niemand niets gezeid eet (DN) Again, the NC reading can only be achieved if all the relevant negative constituents have been scrambled out of their base position and precede nie. On the basis of the above considerations, we propose that nie is adjoined to VP (or to a projection dominating VP) and that scrambled constituents, including those that are negative, are adjoined to the maximal projection dominating nie. A negative constituent which is not scrambled out of its base position cannot take wide scope. Such a negative constituent fails to generate NC readings in relation with the sentential negator nie or with scrambled negative constituents preceding nie, though it may give rise to NC readings in relation with other negative constituents in their base position (cf. 38c). Since negative constituents that originate inside VP usually precede nie, and since extraction from inside VP complements preceding nie is impossible, we assume that negative constituents preceding nie occupy a derived position resulting from adjunction to VP (or a projection dominating VP). We assume that negative constituents with sentential scope occur in a position adjoined to VP in all cases, i.e., even when nie is not present: (39) a. da Valere woarschijnlijk [ vp mewandj [ vp m'e [ v p tj kent]] that Valere probably nobody not knows b. da Valere woarschijnlijk [vp niemand [vp tj kent] ]

134

4.2.4

PARAMETERS AND FUNCTIONAL HEADS

LF-movement

We assume that negative constituents may be scrambled out of their base position at S-structure in order to enable them to take scope outside VP. However, the data suggest that the interpretation of negative constituents involves further movement at LF. As a first piece of evidence for LF-movement, consider 40, in which the negative constituent nieverst is the complement of a preposition. Example 40a is grammatical (with an NC reading), whereas 40b is ungrammatical. (40) a.

da Valere woarschijnlijk [nieverst^ nie tj over klaapt

b. *da Valere woarschijnlijk [pp nieverst over]; nie klaapt We observe that the negative R-pronoun nieverst must scramble on its own and must strand the P over, Pied-Piping of over, i.e., scrambling of the entire PP containing nieverst, leads to ungrammaticality. This contrast would follow if we were to assume that the negative constituent nieverst would have to move at LF to be assigned scope. Recall that scrambled PPs are islands for movement. Based on this observation, the ungrammaticality of 40b can be related to that of 36b. If the scope of nieverst is determined by LF-raising, nieverst in 40b would have to be extracted out of the PP which is itself in an adjoined position. We have seen that extraction out of such scrambled PPs is generally banned. The ungrammaticality of 40b would follow. The data in 41 contrast with those in 40: (41) a.

da Valere woarschijnlijk [over niets] nie klaapt that Valere probably about nothing talks

b. *da Valere woarschijnlijk [Np. niets] nie over tj klaapt c.

da Valere woarschijnlijk [pp. over niets] nie tj klaapt

In 41b, the negative constituent niets on its own cannot precede nie; we conclude that it cannot be scrambled without the preposition over. This is due to the general prohibition against stranding prepositions to the left of the trace of their complement in Dutch (cf. the discussion of 35 and 36, above). Pied-Piping is obligatory, as seen in 41c. If we assume that NC readings are achieved via LF-movement, then we must assume that in 41c LF-movement must be able to affect the PP over niets, since, if only niets were extracted on its own, the grammaticality of 41c, in contrast with the ungrammaticality of 40b and 40c, would be surprising. We assume that in PPs such as overniets 'about nothing', tegen niemand 'against nobody' etc., the negative feature on the NP can percolate to the PP-node.

4.3

Island effects and NC

In this section we present evidence that the interpretation of negative constituents is subject to island constraints: when one negative element is inside an island for movement, and another is outside it, they cannot enter into an NC relation. In section 6 we will see that the relation between the negative head en and negative constituents is also subject to island constraints.35

NEGATIVE CONCORD IN WEST FLEMISH

135

There are two processes in WF whereby a verb projection can be moved, creating an island for movement: in the literature they are referred to as VP topicalization and Verb Projection Raising (VPR).36 In sentences with scrambled negative constituents these negative constituents can be either affected by VPR or by VP topicalization or may be left behind. Both in the case of VP constituents moved by VP topicalization and in the case of VP constituents moved by VPR, we observe the following: a. the negative constituents within the raised or topicalized VP enter into an NC relation with each other; b. VPs subject to VPR or to topicalization are scope islands: negative constituents within the raised or topicalized VP have their scope restricted to the moved projection; c. even though the negative constituents within the raised or topicalized VP enter into an NC relation with each other (a, above), they give rise to DN readings in interaction with another negative constituent in the sentence which is unaffected by VPR or VP topicalization, respectively. 4.3.1

VP topicalization

NC readings obtain between two (or more) bare negative quantifiers and between two (or more) geen-NPs occurring within a topicalized VP: (42) a. [vp. An niemand [VP niets [vp zeggen] ] ] { durven-k ik tj ook to nobody nothing say dare-I I also (i) 'I also dare [to not say anything to anyone] (ii) *'I don't dare to say anything to anyone either.' As the translation suggests, the scope of negation in this example is restricted to the topicalized VP and cannot extend to the finite verb durven. This is consistent with the findings discussed in Haegeman (1992) that the moved VP is an island for movement and for scope phenomena. Thus, 42a shows that the behavior of negative constituents is parallel to that of other scope-bearing elements within a topicalized VP: they cannot take scope outside the dominating VP.37 When VP topicalization strands one (or more) negative element(s) which have been scrambled prior to the topicalization, there cannot be an NC relation between negative constituents in the topicalized VP and those left behind. However, NC is possible among the negative constituents within the topicalized VP, or between the stranded negative constituents: (42) b. [VP Niets [vp zeggen] ]; durven-k ik [vp an niemand tj] nothing say dare 11 to nobody (i) 'I don't dare [to not say anything to anyone]' (ii) *'I don't dare to say anything to anyone.' c. [ VP An niemand [vp niets [ vp zeggen] ] ] durft-ter [vp niemand tj] to nobody nothing say dares der nobody 'Nobody dares to not say anything to anyone.'

1 36

PARAMETERS AND FUNCTIONAL HEADS

d. [ vp An niemand [vp niets [ vp zeggen] ] ] durft-ter [ vp nooit [ vp niemand tj] ] to nobody nothing say dares der never nobody 'Nobody ever dares to not say anything to anyone.' In 42b, the negative constituent niets is affected by VP topicalization, while the negative constituent an niemand is not. In this case the two negative constituents cannot enter into an NC relation. On the contrary, as the translation indicates, they yield a DN reading: they both contribute their own negative meaning to the proposition and the two instances of negation cancel each other out. In 42c niets and an niemand enter into an NC relation and the single negation has the topicalized VP as its domain. Niemand, which is stranded, does not enter into the NC relation and contributes its own negative meaning to the sentence, hence there is a DN intepretation. The same obtains in 42d where an niemand and niets enter into an NC relation, and similarly nooit and niemand, but each pair contributes one instance of negation to the clause, resulting in DN. The data in the examples above follow straightforwardly from the hypothesis that in order for two negative constituents to enter into NC they must be able to take scope over the same domain. It is established independently that the topicalized VP is a scope island, hence any scope-bearing element contained inside, including negative elements, will have its scope delimited to the topicalized VP. On the other hand, negative elements which are stranded by VP topicalization are, as we have argued, scrambled out of VP and take wide scope. They cannot enter into an NC relation with negative elements inside scope islands. 4.3.2

Verb Projection Raising (VPR)

As is well-known, non-finite complements in Dutch and German are subject to Verb Raising.38 In addition to Verb Raising, WF also has VPR, whereby a non-finite verb and one (or more) of its complements are reordered with respect to the finite higher verb (cf. Haegernan and van Riemsdijk 1986; Haegeman (1990b, 1992). Example 43a provides the underlying structure of 43b-d. In 43b the infinitive geven 'give' and its direct object nen boek 'a book' are reordered with respect to the finite verb ziet 'sees' ; 43c shows that the infinitive geven and both its complements can be reordered with respect to the finite verb; 43d is an example of Verb Raising: only geven is reordered. (43) a. da Jan [Marie an Valere nen boek geven] ziet that Jan Marie to Valere a book give sees 'that Jan sees Mary give a book to Valere' b. da Jan Marie an Valere tj ziet [nen boek geven]; c. da Jan Marie tj ziet [an Valere nen boek gevenjj d. da Jan Marie an Valere nen boek t ziet giveni Following Haegeman (1990b, 1992) we assume that VPR is the adjunction of the lower VP to the right (presumably the VP ultimately ends up adjoined to IP). VP

NEGATIVE CONCORD IN WEST FLEMISH

137

constituents which are not affected by VPR, such as an Valere in 43b, are scrambled out of their dominating VP prior to VPR. Haegeman and van Riemsdijk (1986) and Haegeman (1992) show that the VP affected by VPR is an island for movement both at S-structure and at LF. Hence, a scope-bearing element which is affected by VPR cannot take scope outside of the raised VP. Let us first illustrate this with a sentence containing one negative constituent only. In 44a, the negative quantifier niets is not affected by VPR (witness the fact that it precedes the matrix verb durft) and it may take either wide scope or narrow scope with respect to the matrix verb. In 44b, on the other hand, niets has been affected by VPR: it can no longer take scope over the matrix verb, but it can only negate the lower verb: (44) a. da Valere [vp niets t;] durft gevenj that Valere nothing dares give (i) 'that Valere dares to not give anything' (ii) 'that Valere doesn't dare to give anything' b. da Valere t; durft [Vp niets geven]j 'that Valere dares to not give anything' In sentences with multiple negative constituents, the raised VP is an island for NC. Compare 45a-c. In 45a Verb Raising has moved the non-finite verb only, and, crucially, the two negative constituents are not affected by the movement of the verb; in 45b VPR has affected both negative constituents; in 45c one negative constituent is affected by VPR, the other is stranded: (45) a. da Valere [vp an niemand [vp niets tj ] durft zeggenj that Valere to nobody nothing dares say (i) 'that Valere doesn't dare to say anything to anyone' (ii) 'that Valere dares to not say anything to anyone' b. da Valere tj durft [vp an niemand [yp niets zeggen] ] that Valere dares to nobody nothing say 'that Valere dares to not say anything to anyone' c. da Valere [vp an niemand [vp t; durft [VP niets zeggen] ] that Valere to nobody dares nothing say 'that Valere doesn't dare to not say anything to anyone' In 45a the two negative constituents enter into an NC relation and together they yield one instance of negation, with either narrow or wide scope with respect to the matrix verb durft. In 45b the negative constituents are both within the VP affected by VPR and yield an NC reading. The scope of negation is limited to the VP and cannot range over the matrix verb. In 45c an niemand is not affected by VPR; niets, on the other hand, is affected by VPR and dominated by the moved VP. As we can see from the translation, the NC relation cannot be established. An niemand and niets each contribute one instance of negation to the proposition and these two occurrences of negation yield DN. What is particularly interesting about the data of VP topicalization and VPR is that they show that NC, the phenomenon by which multiple negative elements yield one

13 8

PARAMETERS AND FUNCTIONAL HEADS

instance of negation only, is not something which is either available to the language or is not, as one might be led to think by a superficial comparison of languages such as standard Italian and English. On the contrary, the WF data show clearly that it is a phenomenon which is sensitive to syntactic constraints.

4.4

Topicalized or extraposed constituents and NC

In 4.3. we saw that NC cannot be established between a negative constituent which is within a scope island and another negative element outside the scope island. The following examples reveal further structural constraints on NC: (46) a. [CpNiemand] [ Ip eet-er niets ge/eid], nobody has there nothing said 'Nobody said nothing.' b. da-se nie wilde meegoan [pp me niemand] that she not wanted with-go with nobody 'that she did not want to go, with nobody' In (46a) the negative constituent niemand occupies the [Spec.CP] position. The NC reading is not available,39 In 46b the PP me niemand has been extraposed to the right. Since it follows the finite V, which we assume is dominated by I, the PP presumably is adjoined to IP. Again an NC reading is not available. There is a strong tendency to interpret the sentence-final negative PP as an afterthought: it will be separated from the sentence by a heavy intonation break and a pause. We will return to a discussion of the examples in section 5.4.

4.5

The mechanism which generates NC

In this section we will give an informal description of the mechanism which we propose to generate NC readings. We will assume that each negative quantifier has the logical representation of a universal quantifier followed by negation: Vcc-i. Since WF negative quantifiers can be modified by oast 'almost' we assume that they are indeed best represented as universally quantified rather than existentially quantified elements (cf. Hoeksema 1983;Zanuttinil989). 40 (47) a. dan-ze oast niemand nie gezien een that they almost nobody not seen have 'that they hardly saw anyone' b. dat-ter oast niemand dienen boek nie gelezen oat that ter almost nobody that book not read had 'that hardly anyone had read that book' We assume that each quantifier raises at LF. Though universal quantifiers normally bind one variable, i.e., they are unary quantifiers, they can become n-ary quantifiers, i.e., bind two or more variables. 41 In languages that show NC, when two negative

NEGATIVE CONCORD IN WEST FLEMISH

139

quantifiers raise, they undergo a process which we will informally call factorization (Zanuttini 1989): instead of creating two (or more) consecutive instances of a universal quantifier each followed by an instance of negation, negation is factored out and the two (or more) universal quantifiers become one binary (or n-ary) quantifier: (48) a. [VaH[VH([VH) = [Vz,2/(,2)b In some (but not all42) languages such a process also affects the marker of sentential negation. WF is a case in point: nie, which is presumably not a quantifier but simply a negative operator, also enters into NC. When nie and a negative quantifier raise, negation is factored out: (48) b. [VaHH = [^h In the remainder of the paper, we will refer to the process which generates NC as the process of factorization. We assume that the negative constituents which are to enter into an NC relation raise at LF to form a complex quantifier from which the negation is factored out. Negative factorization, resulting in NC, obtains every time negative constituents are adjoined to the same maximal projection; it therefore obtains, for instance, within VPs in situ, VPs affected by VP topicalization and by VPR.43 But because VPs which have been moved by VP topicalization or VPR are islands for scope phenomena, i.e., prevent movement of any scope-bearing element outside their boundaries, the complex quantifier which results from the process of factorization cannot raise past its VP boundaries. It follows, then, that if another negative constituent occurs in the sentence, it won't be able to enter into an NC relation with the complex quantifier created within the moved VP and it will contribute its own independent negative force, hence DN will result. Two questions arise concerning the process of factorization: a. whether the formation of the complex quantifier is achieved globally over all negative constituents in an across-the-board fashion, or whether the negative factorization is done in a step-wise fashion, affecting consecutive pairs of negative quantifiers; b. whether the complex quantifier which has been created at the VP level moves further up in the structure. In the next section we will address the first of these questions; then, in section 6, where we discuss the relation between negative constituents and the negative head en, the second question will be discussed.

5

Constraints on the process of factorization

In this section we will first describe some restrictions holding on the co-occurrence of negative constituents which enter into an NC relation in WF (5.1). We will then explain them by (a) providing a typology of negative constituents in terms of a head feature system (5.2) and by (b) arguing that negative constituents enter into an NC

140

PARAMETERS AND FUNCTIONAL HEADS

relation through step-by-step LF-raising, subject to the condition that two consecutive negative constituents can enter in the NC relation only if their heads share at least one relevant feature (5.3). This discussion thus answers the first question left open in section 4: negative constituents raise step-by-step at LF to form the complex quantifier subject to negative factorization. Finally, in section 5.4, we will begin addressing the second question left open in section 4, namely, whether the complex negative operator which results from the process of factorization can raise any further and affect other negative elements in the sentence. This issue will be discussed in section 6.

5.1 Adjacency constraints on negative constituents While discussing the position of negative quantifiers we have already mentioned that they can optionally co-occur with the negative marker nie (cf. 28) and yield an NC reading. Though this is always true of bare negative quantifiers,44 there is an important restriction concerning the co-occurrence of nie and geen-NPs: when a geen-NP is adjacent to nie the NC reading is not available and a double negation interpretation results: notice the contrast in interpretation between 49 and 50: (49) a. da Valere niemand nie kent (NC) that Valere nobody not knows 'that Valere doesn't know anyone' b. da Valere dienen boek nieverst nie vindt (NC) that Valere that book nowhere not finds 'that Valere doesn't find that book anywhere' c. da Valere tegen niemand nie klaapt (NC) that Valere against nobody not talks 'that Valere doesn't talk to anyone' d. dat-ter bier niemand nie geweest eet (NC) that der nobody not been has 'that nobody has been here.' (50) a. da Valere geen boeken nie gelezen eet45 (DN) that Valere no books not read has 'that there are no books such that Valere has not read them' b. dat-ter hier geen mens nie geweest eet that der here no person not been has 'that everybody has been here'

(DN)

As the English translations suggest, only the DN reading is available in 50. While this reading is also marginally available in 49, the unmarked interpretation of the examples in 49 is the NC reading. Interestingly, the NC reading is available if we replace nie in 50 by nie meer 'no more' with a slight semantic shift:46 (51) a. da Valere geen boeken nie meer gelezen eet47 that Valere no books no more read has 'that Valere hasn't read any books any more'

(NC)

NEGATIVE CONCORD IN WEST FLEMISH

b. dat-ter hier geen mens nie meer geweest eet that der no person here no more been has 'that nobody has been here anymore'

141

(NC)

The adjacency of nie and geen-NPs leads to a DN reading if there are only two negative constituents. It leads to ungrammaticality in sentences with more than two negative constituents, such as 52. We assume that the DN reading is unavailable with more than two negative constituents because of processing problems: double negation readings are easily accessible with only two negative constituents, but cannot be obtained when there are more than two constituents involved. What is important for our purposes though, is that the NC reading is unavailable every time nie is adjacent to a geen-NP, as shown in 52, even if a bare negative constituent is present.48 But nie and geen-NPs can co-occur in the same clause if a bare negative constituent intervenes, as in 53: (52) a. *dat-ter niemand geen geld nie oat (DN) that der nobody no money not had b. *da Valere an niemand geen boeken nie geeft that Valere to nobody no books not gives c. *da Valere nooit geen boeken nie leest that Valere never no books not reads

(DN)

(DN)

(53) a. dat-ter geen mens niemand nie gevroagd oat that der no person nobody not asked had 'that nobody had invited anyone'

(NC)

b. da Valere an geen mens niets nie gegeven oat (NC) that Valere to nobody nothing not given had 'that Valere had not given anything to anyone' Note that the restriction is not simply one of adjacency: it does not suffice for a geen-NP and nie to be separated by any random constituent; rather, the requirement is that a bare negative constituent should intervene: (54) a. dan' k an geneen student da boek nie goan geven (DN) that I to no student that book not go give 'that there is no student to whom I won't give the book' b. dan' k an geneen student niets nie goan geven that I to no student nothing not go give 'that I won't give anything to any student'

(NC)

c. dan'k an niemand da geld nie goan geven (NC) that I to nobody that money not go give 'that I'm not going to give that money to anyone' In 54a the definite NP da boek intervenes but again, as in 50, only the double negation reading is available. In 54c we see that definite NPs intervening between bare negative constituents are compatible with NC chains. Example 54b shows that

142

PARAMETERS AND FUNCTIONAL HEADS

what is essential is that between the geen-NP and nie there be an intervening bare negative constituent. 49 In addition to the ban on the co-occurrence of nie adjacent to a geen-NP with an NC reading, nie also must not occur adjacent to nie meer or to other negative constituents whose negative force is carried by nie. If it does, only the double negation reading is available. Such negative constituents are themselves compatible with geen-NPs: (55) a. da Valere doar nie meer nie geweest eet (DN) that Valere there no more not been has 'that Valere did not remain absent from that place more often' b. da Valere hier nie lange nie gewerkt eet50 that Valere here not long not worked has 'that Valere was not out of a job for long'

(DN)

c. da Valere hier nie lange geen werk net eet (NC) that Valere here not long no work had has 'that Valere was not employed here for long' A rash conclusion might be to say that only negative constituents which consist of just a single negative word are compatible with nie,51 but this cannot be correct given the grammaticality of 56, where the string niemand zen geld is one constituent:52 (56) a. da Valere [niemand zen geld] nie wilt (NC) that Valere nobody his money not wants 'that Valere doesn't want anyone's money' b. K'een an [niemand zen geld] nie geweest. I have to nobody his money not been 'I haven't touched anybody's money.'

(NC)

c. dat-ter [niemand zen werk] nie gereed was that der nobody his work not ready was 'that nobody's work was finished.'

(NC)

On the other hand, if the possessive NP in the construction illustrated in 56 is a geenNP rather than a bare negative constituent the sentence will have the DN reading with nie (57a) and the NC reading is saved by nie meer (57b): (57) a. da Valere neu [geen mens zen hulpe] nie wilt that Valere now no man his help not wants 'that Valere does not refuse anybody's help' (i.e., that he accepts anyone's help)

(DN)

b. da Valere neu [geen mens zen hulpe] nie meer wilt (NC) that Valere now no person his help no more wants 'that now Valere doesn't want anybody's help any more' The table in 58 summarizes the co-occurrence restrictions on adjacent negative constituents entering into an NC relation (bare quantifiers are referred to as "Bare Q"):

NEGATIVE CONCORD IN WEST FLEMISH

58

143

Co-occurrence restrictions on adjacent negative constituents BareQ e.g., Geen-NP e.g., Nie meer e.g.,

BareQ

Geen-NP

Nie

yes niemand niets yes geen mens niemand yes niemand nie meer

yes niemand geen geld yes geen mens geen tyd yes geen mens nie meer

yes niemand nie no *geen mens nie no *nie meer nie

We believe that these constraints on the order in which negative constituents occur are related to the LF process of factorization we have described above. In particular, they can be accounted for by assuming that negative constituents raise at LF in a step-wise fashion. At each step, a negative constituent will adjoin to the next higher negative constituent in the structure and undergo a process of factorization with this constituent, under the condition that the heads of the two negative constituents which are being factorized are elements of the same type, in a sense that will be specified in the next section. If the heads of the two constituents don't match, the sentence will be ungrammatical under the NC reading, but will allow a DN reading. In order to show how this raising mechanism works or fails to work, we first provide a typology of negative constituents in WF based provisionally on the features on the heads.

5.2

Typology of negative constituents

We will characterize the difference between negative constituents in WF by capitalizing on their nature as quantificational and as negative elements. We will refer to these properties by means of the features [neg] (negation) and [Q] (quantifier). We do not pretend to be able to define these features more precisely at present. In future research (Haegeman and Zanuttini 1995) we will further elaborate our analysis in terms of the DP analysis of NPs (Abney 1987). 5.2.1

Bare negative quantifiers

Bare negative quantifiers of the type ofniemandare both quantificational and negative. We will represent these properties as features on the head of the phrase (X° = [Neg; Q]), which percolate up to the maximal projection. 5.2.2

Geen-NPs

Geen-NPs are also both quantificational and negative, but differ from bare negative quantifiers in that these features are not both instantiated on the head of the phrase. In the spirit of Abney's (1987) DP analysis we propose that geen-NPs have an empty head D° which carries the quantificational feature [Q], while geen occupies the specifier position and carries the negative feature [neg]: (59) [Dp geen [D<> O [ NP - . . ] ] ] some text [neg] [Q]

144

PARAMETERS AND FUNCTIONAL HEADS

A first piece of evidence for this analysis is that in the singular geen-NP (as in 60), the negative element can be distinguished morphologically from the quantificational element: (60) gen-eenen boek no book Geneenen is composed of the quantificational eenen and the negative prefixe gen.5J The singular indefinite article eenen corresponds to a zero quantifier in the plural (6la); likewise, we assume that singular geneenen, composed of a negative gen and singular eenen, corresponds to the plural geen, in its turn composed of negative gen and a plural zero quantifier (61b): (61)

Singular

[neg] a. b.

gen-

[Q] eenen boek eenen boek

Plural

[neg] geen

[Q] 0 boeken 0 boeken

A second piece of support for the analysis in 59 comes from the existence in WF of constructions such as 62, where the quantificational force of the phrase is expressed by the quantifier nie vele and geen simply acts as a negative element entering into a type of NC relation with nie vele:54 (62) nie vele geen boeken not many no books 'not many books'

5.2.3

Nie

The negative marker nie is not quantificational but only negative (it is a negative operator, but not a negative quantifier): its head carries the feature [neg], which percolates up to the maximal projection, while the feature [Q] is not part of the constituent at all. 5.2.4

Nie meer

Nie meer is both negative and quantificational and its structure parallels that of geenNPs: the feature [Q] is carried by the head meer, a quantificational element,55 and the feature [neg] is carried by the negative element nie in the specifier position.56

5.3

Step-wise movement

The typology of negative constituents introduced in section 5.2 will help us account for the adjacency constraints on their co-occurrence. We suggest that in order to achieve NC, negative constituents enter a process of negation factorization by raising step-by-step and adjoining to the left of the next constituent in the structure. This step-wise factorization movement is constrained by a head-matching requirement: at least one of the head features of the negative constituent which is raising must

NEGATIVE CONCORD IN WEST FLEMISH

145

match one of the head features of the negative constituent to which it adjoins. If this condition is not met, negation factorization cannot apply and the NC reading is unavailable. 5.3.1

Bare quantifiers

Two bare negative quantifiers, e.g., niemand and niets, can occur adjacent to one another because their heads have the same features, both [neg] and [Q]. Moreover, a bare negative quantifier can occur adjacent to a geen-NP, since the feature [Q] is shared by the head of both constituents. It can also occur adjacent to the negative marker nie, since the head of the latter is specified with the feature [neg] which is shared by the head of the bare quantifier. 5.3.2

Geen-NPs

Two geen-NPs, e.g., geen mens and geen tyd, can occur adjacent to one another, because their heads match for the feature [Q]. But a geen-NP cannot occur adjacent to the negative marker nie, because the head of the latter is specified with the feature [neg] only. Since the heads of the geen-NP and of nie do not share any features, the lower element, nie, cannot raise to the higher element, the geen-NP, and undergo factorization. NC is therefore not available and a double negation reading will obtain.

5.3.3 Me meer Nie meer, whose head carries only the feature [Q], can be adjacent both to a bare negative quantifier and to a geen-NP, since the heads of both carry the feature [Q]. But it cannot be adjacent to the negative marker nie, which lacks the feature [Q] completely. 5.3.4

Summary

The typology of negative constituents sketched above is summarized in 63, where the feature of the head is specified (cf. 58): (63)

Head features on negative elements and co-occurrence restrictions

BareQ BareQ [neg;Q] Geen-NP [Q] Nie meer

[Q]

[neg;Q]

Geen-NP [Q]

Nie [neg]

yes niemand niets yes geen mens niemand yes niemand nie meer

yes niemand geen geld yes geen mens geen tyd yes geen mens nie meer

yes niemand nie no *geen mens nie no *nie meer nie

Evidence in favor of the step-wise movement of factorization is provided by the fact that, even though two negative constituents may not be allowed to occur adjacent

146

PARAMETERS AND FUNCTIONAL HEADS

to one another, they may be allowed to occur in the same negative chain as long as another negative element of the right type intervenes (as we have seen in 52 and 53): (64) a. *an niemand geen geld nie to nobody no money not b.

geen geld an niemand nie no money to nobody not

This follows straightforwardly from an analysis which assumes that each negative element raises step-by-step, and not in one across-the-board movement. In 64a the lowest negative constituent, nie, raises and adjoins to the left of geen geld; but the features of their heads do not match, hence the two negative constituents cannot undergo factorization. In 64b, on the other hand, nie raises and adjoins to the left of an niemand: since both constituents share the feature [neg], they can undergo factorization and form a complex element which now has both the features [neg] and [Q]. When this element raises in turn and adjoins to the left of geen geld, factorization will be possible, thanks to the presence of [Q] in its head.57

5.4 Negative elements in [Spec,CP] One question we have left open is whether the complex operator which is formed, for example, at the level of VP can further raise after negative factorization has affected all the relative constituents in a particular scope domain. We have already observed that negative elements adjoined to a VP in situ or the projection immediately dominating it do not yield NC with negative elements dominated by a topicalized VP or by a VP affected by VPR. This is most likely due to the fact that the projections affected by VP topicalization and VPR are islands for scope phenomena, as we have repeatedly pointed out above, and cannot interact with other scope-bearing elements in the sentence. If this is an independent property of these constructions, then these data cannot be used to determine whether the complex operator resulting from factorization at the level of VP can raise any further. In this regard, it is interesting to notice that the complex quantifier formed at the level of VP cannot raise and undergo factorization with a negative quantifier in [Spec,CP] in WE We have seen (46a) that when a sentence contains a negative quantifier within the VP in situ and one in fSpec,CP], negative concord is not possible and double negation arises instead: (65) a. [cp An niemand [c een-k] niets gezeid (DN) to nobody have-I nothing said 'To nobody I said nothing.' (i.e., I said something to everyone) b. [cp Niemand [c eet-er] geen boeken gelezen (DN) Nobody has er no books read 'Nobody did not read any books.' (i.e., Everybody read some books) We interpret the lack of NC and the presence of a DN reading as an indication that the complex operator formed at the VP level cannot raise and undergo factorization with the negative quantifier in [Spec.CP].

NEGATIVE CONCORD IN WEST FLEMISH

147

This is an interesting and somewhat puzzling difference between WF and Romance, where VP-internal elements always seem to be able to yield NC with negative constituents which have been topicalized. At present, we cannot evaluate this difference and say whether it should be related to general differences between Romance and Germanic, whether it is a peculiarity of WF which might distinguish it from other varieties of Flemish, or even whether it might be a dialectal feature on which WF speakers differ among themselves. We will simply take it as an indication that the complex operator which results from the process of factorization cannot raise to the level of CP in these cases. We have also seen (46b) that extraposed negative constituents cannot enter into an NC reading with non-extraposed ones: in 66a the PP over niemand occurs to the right of the finite verb, and we assume it is adjoined to the highest functional projection into whose head V has moved (e.g., IP): (66) a. dan-ze nooit niets gezeid een over niemand (DN) that they never nothing said have about nobody 'that they never said anything — about nobody' b. dan-ze nooit over niemand niets gezeid een (NC) that-they never about nobody nothing said have 'that they never said anything about anyone' Example 66a is grammatical with the extraposed PP over niemand interpreted as an afterthought, separated from the finite verb by a pause and a heavy intonation break. We assume that niets and nooit in 66a will undergo the negative factorization process described above, in the same way as will happen in 66b. The resulting cluster of negative elements apparently cannot raise and move rightward to adjoin to the extraposed PP. We will return to this point below. On the other hand, we must also assume that the extraposed negative constituent itself cannot move to undergo factorization with another negative constituent. This is probably to be interpreted in terms of the ECP.58 We also need to return to the relation between the negative constituents (simple or complex, as in the case of NC) and the negative head en which we assume is a scope marker for negation. We will address this question in the next section.

6 The negative head en So far, we have discussed the interaction of negative constituents with each other and with the sentential negation nie. We have concluded that they yield NC by raising at LF, where they undergo a process of factorization of the negation. On the basis of the co-occurrence restrictions on adjacent negative elements we have suggested that they move step-by-step, each time adjoining to the next higher negative constituents, and that this operation is constrained by the requirement that the constituents share at least one head feature. We have also seen that the complex quantifier formed by the process of factorization in a given VP domain cannot interact with negative elements in [Spec,CP], and have taken that as indication that it cannot raise up to that level. In

148

PARAMETERS AND FUNCTIONAL HEADS

this section, we will examine the interaction between negative constituents and the preverbal negative marker en, the head of NegP. First we v/ill describe the distribution of en with respect to other negative constituents (6.1); then we will discuss two competing analyses of the data, a movement analysis and a co-indexation analysis; we will present evidence in favor of the movement analysis (6.2). In section 6.3 we will see that though the movement analysis is preferable for one type of phenomena, the co-indexation analysis is needed to account for the link between en and negative constituents in [Spec.CP]. We will propose that the relation between the negative head and the negative constituents can be captured by the Neg-Criterion, a well-formedness condition on the relation between negative constituents and the head of NegP. One proviso is in order here: we will be assuming throughout the discussion that what is valid for the overt negative head en applies without modification to its non-overt variant.

6.1

The distribution of the negative marker en

6.1.1 En must be licensed by another negative constituent We have already mentioned in section 3.3 that en cannot be the sole marker of sentential negation: unless it co-occurs with another negative element, its presence gives rise to ungrammaticality (see 21"). Let us informally say that en must be licensed by another negative element. En is not sensitive to the type of negative constituent with which it co-occurs: it can be licensed by the negative marker nie, by a bare negative quantifier, and by a geen-NP when they are adjoined to a VP in situ (as shown in 21', 22' and 23', respectively). En is the head of NegP, which, by assumption, dominates IP. The negative constituent which licenses en is in a position from which it can take sentential scope. Recall that for a negative VP constituent to take scope outside VP it must scramble: in 67a niets has been scrambled, as we can conclude from the fact that it precedes the reason adverbial doavuoren, and it takes wide scope. In 67b niets occupies its VP-internal position and it takes narrow scope with respect to the adverbial: (67) a. Valere ee niets doavuoren gedoan. Valere has nothing for that reason done 'There's nothing that Valere did for that reason.' b. Valere ee doarvuoren niets gedoan. Valere has for that reason nothing done 'For that reason Valere did nothing.' (i.e., Valere gave up too soon) In 67'a niets takes wide scope, and en is licensed; in 67'b en cannot be licensed by niets because the latter remains in its base position and has its scope limited to VP:59 (67') a. Valere en-ee niets doavuoren gedoan. b. * Valere en-ee doavuoren niets gedoan.

NEGATIVE CONCORD IN WEST FLEMISH

149

6.1.2 Island effects and the licensing of en As mentioned already, there are no restrictions on the licensing of en by specific types of negative constituents in any way similar to the adjacency restrictions we described in section 5.1 for NC readings of other negative constituents. On the other hand, there are configurational constraints on the relation between en and the licensing negative constituent: en cannot be licensed by a negative constituent which itself is inside a scope island. As was the case for NC readings, island effects can be observed with respect to topicalized VPs and VPs affected by VPR. En cannot be licensed by a negative element contained in a topicalized VP, as shown by 68b, in contrast with 68a: (68) a. da-ze niets e«-durft zeggen that-she nothing e«-dares say (i) 'that she doesn't dare to say anything' (ii) 'that she dares to not say anything' b. [vp Niets zeggen]j (*en) durft-ze ook [vp tL nothing say en dares she also 'She also dares to say nothing.' The problem is not some sort of incompatibility between the presence of en and VP topicalization, but the structural configuration of en with respect to the negative element which licenses it. In fact, the presence of en in itself will not create ungrammaticality if VP topicalization takes place, as long as it leaves at least one negative constituent unaffected, regardless of whether it affects any other negative constituents (as in 69b): (69) a. [vp Den woarheid zeggen]; en-durft-ze [VP an niemand [vp t;] ] the truth tell en dares she to nobody 'She doesn't dare to tell the truth to anyone.' b. [vp Niets zeggen]j en durft-ze [vp an niemand [vp [t];] ] 'To nobody does she dare to say nothing.' (i.e., she talks to everybody) Similarly, en cannot be licensed by a negative element which has been affected by VPR, as illustrated by 70b, in contrast with 70a: (70) a. da Valere [niets en-durft geven] that Valere nothing en dares give (i) 'that Valere doesn't dare to give anything' (ii) 'that Valere dares to not give anything' b. da Valere t; (*en)-durft [vp niets geven]; that Valere en dares nothing give 'that Valere dares to not give anything' Again, as in the case of VP topicalization, what is at issue is not some sort of incompatibility between en and VPR per se, but rather the relative positions of en and the negative constituent which licenses it. In fact, as long as there is at least one

150

PARAMETERS AND FUNCTIONAL HEADS

negative constituent in the sentence which is not affected by VPR then the presence of en is acceptable, regardless of the fact that another negative constituent may have been affected by VPR (71): (71) a. da Valere m'ett tj en-durft [vp an Jan zeggen]j that Valere nothing en dares to Jan say (i) 'that Valere doesn't dare to tell Jan anything' (ii) 'that Valere dares to not tell Jan anything' b. da Valere an niemand t-t en durft [vp niets gevenjj that Valere to nobody en dares nothing give 'that Valere dares to not give anything to anyone' 6.1.3

Preposed and extraposed negative constituents

An interesting contrast between the relation between constituents which enter into an NC relation on the one hand and the licensing configurations for en on the other is that en can co-occur with a negative element in the position [Spec.CP] (72a). Recall that a negative constituent in [Spec.CP] could not enter into an NC relation (72b): (72) a. [cp Niets] en ee-se gedoan nothing en has she done 'Nothing has she done.' b. [cp Niets] en-ee ze nie gedoan (DN) nothing en-has she not done 'There was nothing which she failed to do.' (i.e., she did everything) In contrast, en cannot be licensed by an extraposed negative constituent (cf. 46b): (73) a. Z-(*eri) willen klapen me niemand. they en want talk with nobody They don't want to talk to anyone.' Recall that extraposed negative constituents also do not enter into an NC relation with non-extraposed ones: (73) b. Ze willen nooit klapen me niemand (NC) they want never talk with nobody 'They never want to talk — with nobody.' In 73b it would appear that the normal interpretation is one in which the extraposed constituent is an afterthought. There is a clear intonation break after klapen.

6.2

A movement analysis

In analogy with the discussion of Romance languages in sections 1 and 2, we have proposed that en is the head of the functional projection NegP, which dominates IP.60 We have taken the fact that en cannot occur as the sole negative element in the sentence to express sentential negation as an indication that it is not a real negative

NEGATIVE CONCORD IN WEST FLEMISH

151

element, but rather a scope marker, i.e., an element which marks the scope of negation. Strictly speaking then, en does not enter in an NC relation in the same way that other negative constituents enter into an NC relation: all these negative constituents can express sentential negation by themselves and do not need to be licensed. What is then the relationship of en with other negative elements, without which it cannot occur? En being a scope marker, it must always be in construction with the negative element whose scope it marks. It is clear that en also is sensitive to syntactic configuration, as shown by the contrasts discussed in section 6.1 : en can occur with a negative element adjoined to VP, but not with elements which have been moved by VP topicalization, by VPR or by extraposition. There are at least two plausible ways of establishing the relationship between en and other negative elements: they are either related by co-indexation or by movement. 6.2.1

Co-indexation

Suppose that en must be co-indexed with another negative element in the structure, and that such co-indexation is sensitive to islands. Then the constraints illustrated in 68b and 70b will follow: en cannot co-occur with negative elements affected by VPR or VP topicalization because the moved VP is an island. The problem, though, is that co-indexation does not seem subject to island effects: (75) dan-ze woarschijnlijk doarom aan Valere zoun willen [vp zen buro geven] that they probably therefore to Valere would want his own office give 'that for that reason they would probably want to give Valere his own office' In 75 the anaphor zen eigen inside the VP which is affected by VPR can be co-indexed with the NP Valere. We assume that the VP is adjoined to IP, given that it follows the finite verb goan. Valere, the antecedent of the anaphor, is presumably not in a (strict) c-command relation with vu zen eigen at S-structure. But in spite of this, co-indexation is possible. If the relationship between en and the negative constituent were established by a similar type of co-indexation, it would not be possible to explain why en could not co-index with VP-internal elements or with elements affected by VPR or VP topicalization. Similarly an extraposed anaphor may be co-indexed with an argument which has not been extraposed: (76) dan-ze Valere; goan loaten klapen over zen eigen; werk that they Valere go let talk about his own work 'that they'll let Valere talk about his own work'

1 52

PARAMETERS AND FUNCTIONAL HEADS

Again if co-indexation is used to establish the link between en and the licensing negative constituent then it is surprising that en cannot be licensed by an extraposed negative constituent. 6.2.2

Movement

On the other hand, suppose that the link between the negative scope marker en and the licensing negative constituent is established by means of movement in the sense that the negative complex resulting from the LF factorization must raise to NegP in a way to be made more precise in section 6.2.4. Then we predict straightforwardly that en cannot be related to negative elements dominated by a moved (i.e., topicalized or raised) VP. As already mentioned several times, such VPs are islands with respect to LF movement, hence the negative complex dominated by such moved VP cannot raise to NegP at LF.61 A movement analysis will also be able to account for the observation that extraposed negative constituents cannot enter into an NC relation with non-extraposed ones and cannot license en. Recall that, although intuitions are not sharp, it seems that the natural interpretation of extraposed constituents containing negative elements is to treat them as afterthoughts. If the negative constituent which licenses en must move to NegP, we predict that extraposed constituents which are by assumption right-adjoined to IP cannot be moved to NegP. We propose that this is due to an ECP effect: traces of extraposed constituents cannot be properly head-governed since they are not in the canonical government relation with V. Hence the ungrammaticality of 77a: (77) a. *dan-ze en-willen klapen me niemand that they en want talk with nobody Recall that 66a (repeated here as 77b), cannot have an NC reading involving the extraposed negative constituent over niemand: (77) b. dan-ze nooit niets gezeid een over niemand that they never nothing said have about nobody 'that they never said anything— about nobody' As mentioned already, over niemand could not be moved out of its extraposed position to be adjoined to the cluster created by the factorization of nooit and niets: we assume that the trace of over niemand will not be properly governed. Our analysis of these extraposed constituents is tentative. More research is needed to establish the scopal properties of extraposed elements. 6.2.3

Co-indexation in CP

Consider the relationship between the negative constituent in [Spec,CP] in 72 (repeated below as 78) and en. For the earlier examples we have proposed that the link between en and VP-adjoined negative constituents is established by LF-raising of the VP-negation which will end up in a position c-commanding en. In fact, the c-command relation between en and the negation in [Spcc,CP] is available already in

NEGATIVE CONCORD IN WEST FLEMISH

15 3

78 and we assume that in such cases co-indexation of en in the head of C and niets in [Spec,CP] suffices. The co-indexation is an instantiation of Spec-head agreement.62

To sum up our discussion so far: en appears to be licensed in two ways, either by means of a negative constituent which raises to NegP at LF or by a negative constituent in [Spec.CP]. In the next section we will try to replace this disjunction on the licensing of e. 6.2.4

The Neg-Criterion63

We have seen that the relationship between en and the negative constituent is achieved in two ways: either by LF-movement of the negative constituent to the NegP level or by Spec-head agreement at S-structure. We may generalize this one step further and eliminate the disjunction by arguing that en generally needs to be licensed by being in a Spec-head relation with a negative constituent. This relation is established either at S-structure (as in 78) or at LF, by moving the negative constituent to the appropriate specifier position. This generalization concerning the licensing of en is interesting since it is reminiscent of a similar proposal concerning w/z-constituents. Rizzi (1990b:378) proposes that the occurrence and position of wft-elements at LF is determined by the following principle: The M-Criterion. 1. Each +WH X° must be in a Spec-Head relation with a WH-phrase 2. Each WH phrase must be in a Spec-Head relation with a +WH X° [The Wft-Criterion] can be looked upon as a criterion of well-formedness on LF expressing the way in which wA-expressions are assigned scope. We might consider extending the type of well-formedness criterion for w/z-constituents (as stated by Rizzi) to other operators and suggest the following principle for the scope of negative elements: (79) The Neg-Criterion: 1. Each Neg X° must be in a Spec-head relation with a Negative phrase 2. Each Negative phrase must be in a Spec-head relation with a Neg X° The relationship between the negative head en and the negative constituent licensing it is then an instantiation of 79: the correct configuration is either achieved at S-structure or at LF.

154

PARAMETERS AND FUNCTIONAL HEADS

Let us speculate for a moment on the Italian data discussed in the introduction (section 1). We have pointed out that negation was expressed either by a negative head c-commanding Infl or by a negative constituent higher than Infl. The disjunction noticed there may also be related to the Neg-Criterion. It is tempting to say that the negative head in Italian is non-overt precisely when it is already in a Spec-head configuration with a negative constituent (or its trace) at S-structure. We hope to elaborate upon this proposal in later work. 6.2.5

Summary

In this section we have explored an analysis of the relation between the negative head (overtly realized by en) and the licensing negative constituent(s) in terms of the Neg-Criterion (79). The essential idea is that the negative head and the negative constituent(s) with sentential scope must be in a Spec-head relation, established either at S-structure or by LF-movement. If a movement analysis is the correct way to analyse the interaction between the negative head and the licensing negative constituent(s), then we have provided at least a partial answer to the question of whether the complex quantifier created by the process of factorization at the level of VP can raise any further: it can raise to the level of NegP, allowing the Neg-Criterion to be satisfied.

7

Issues for further research

In this section we will briefly discuss two issues which require further study. In section 7.1 we examine further constraints on the formation of negative chains, focusing on intervening non-negative elements. In section 7.2 we provide some data on NC in Bavarian, based on work by Bayer (1990).

7.1

Intervening elements

So far we have discussed cases of co-occurrence of negative constituents where all of them were contiguous, i.e., no non-negative material intervened between two negative elements. In this section, we will examine cases in which a non-negative constituent intervenes between two negative elements and show that if the intervening constituent is a scope-bearing element, negative concord cannot be established across it. When a definite NP or a deictic expression intervenes between two negative constituents, negative concord obtains irrespective of the intervening constituent: (80) a. dat er niemand dienen boek nie wilt kuopen that there nobody that book not wants buy 'that nobody wants to buy that book' b. dan'k an geen mens dienen boek nie meer goan geven that I to no person that book no more go give 'that I'm not going to give that book to anyone anymore' c. dat-ter niemand gunter niets gezeid oat that der nobody there nothing said had 'that nobody told anything there'

NEGATIVE CONCORD IN WEST FLEMISH

155

d. dat-ter niemand noa Gent nie wilt goan that der nobody to Gent not wants go 'that nobody wants to go to Gent' On the contrary, when the intervening NP is a scope-bearing element, i.e., either a quantifier or an indefinite NP, NC between the two negative constituents is impossible. We have tried to indicate in the translations of these examples that the only available reading of the sentences is one with two instances of negation. (81) a. dat-ter niemand entwoa niets gekocht eet (DN) that der nobody anywhere nothing bought has 'that nobody bought nothing anywhere' b. da Valere an niemand eentwa nie gegeven eet (DN) that Valere to nobody something not given has 'that Valere didn't not give something to someone' (hence 'that Valere did not withhold something of anyone') c. da Valere an niemand al da geld nie geeft64 (DN) that Valere to nobody all that money not gives 'that Valere doesn't withhold the money from anyone' d. dat-ter niemand al d'joengers niets gegeven eet (DN) that der nobody all the kids nothing given has 'that nobody gave nothing to all the kids' e. dat-ter niemand overal niets gekocht eet (DN) that der nobody everywhere nothing bought has 'that nobody bought nothing everywhere' Scope-bearing elements do not have the same blocking effect on NC when they occur either after or before the negative constituents, that is, when they do not interrupt the chain of negative constituents: (82) a. dat-ter niemand niets an entwien gezeid eet (NC) that der nobody nothing to someone said has 'that nobody told anyone anything' b. dat-ter niemand nieverst eentwa gekocht eet (NC) that der nobody nowhere something bought has 'that nobody bought anything anywhere' c. dat-ter entwien an niemand niets gezeid eet (NC) that der somebody to nobody nothing said has 'that somebody didn't tell anyone anything' Two questions arise then: 1. Why should quantificational elements block NC when they intervene between negative elements, thus differing from definite NPs? 2. Why should they have no effect on NC when they do not intervene?

156

PARAMETERS AND FUNCTIONAL HEADS

We can answer the latter question by suggesting that the process of NC is triggered by the presence of the feature [neg] on the phrase, either on its head (as is the case for nie and niemand), or on its specifier (as is the case in geen-NPs and in nie meer). Thus, when scope-bearing elements which lack the negative feature precede or follow negative constituents, they simply won't take part in the process of factorization which creates NC readings. A similar explanation applies to the intervening definite NPs: since they lack quantificational force, they don't raise at LF, hence don't interfere with the process of factorization. On the other hand, when scope-bearing elements intervene between two negative constituents, they do interfere with the process of factorization which generates NC. At least two different explanations come to mind with respect to their blocking effect on NC. We have proposed that the negation is factored out of the various negative constituents by a step-wise movement, where each negative constituent adjoins to the next one up, as discussed in section 5. If a non-negative quantifier intervenes between negative constituents, then at some stage of the step-wise factorization movement, a (lower) negative constituent will raise and adjoin to the left of the (adjacent and higher) intervening scope-bearing element. Because of the absence of the negative feature on the latter element, the process of factorization will be inhibited. Hence, when the next step in raising takes place, the quantifiers will combine without forming a complex quantifier with a single factorized negation. Alternatively, the blocking effect of non-negative scope-bearing elements may be accounted for in terms of some minimality condition on the process of LF-raising. Suppose that all quantifiers raise at LF and leave a trace; moreover, suppose that negative quantifiers, though moving in a step-wise fashion, skip intervening elements which lack the negative feature. Take a construction in which a negative quantifier (call it Q ) precedes a non-negative element (Q_neg) which in turn precedes another negative quantifier (Qneg?).

When raising and factorization takes place at LF the following situation will arise: Qneg2 will raise and adjoin to the left of Qneg., forming a complex quantifier through the process of factorization, say Qneg2. The non-negative element Q_ neg will itself also have to move to take scope. Two possibilities arise: first, it might raise to IP (83b), i.e., to a projection dominating Qneg2 If we adopt Rizzi's Relativized Minimality Theory (1990a) a problem will then arise with respect to government of the trace of the non-negative element, since the first available operator for its identification will be the negative element O nct> ,. which is not the appropriate antecedent:

NEGATIVE CONCORD IN WEST FLEMISH

157

Alternatively one might imagine that the non-negative quantifier lowers to VP, but again its trace will then not be governed by an appropriate antecedent. With respect to the role of intervening constituents two observations are in order. On the one hand, the blocking role of intervening quantifiers is not a property exclusive to West Flemish NC.65 Similar effects can be observed with respect to NC in Italian (84): (84) a. Vogliamo tutti non fare niente. we want all non do nothing 'We all want not to do anything.' b. *Non vogliamo tutti fare niente? non we want all do nothing In 84a non and niente may enter into an NC relation, whereas in 84b, with the intervening quantifier tutti they cannot. Similarly, compare English (85a) and Italian (85b). The intervention of tutti in 85b disallows niente to raise to non: (85) a. They've all done nothing. b. *Non hanno tutti fatto niente. non they-have all done nothing So far we have noticed intervention constraints on NC. However, one should not treat the intervention problem as one characteristic of NC only. Observe, for instance, that the relation between a negation and the polarity items anything, anybody, etc. in English is interrupted by the presence of the quantifier: (86) a. John didn't show Mary anything. b. John didn't show everybody his book. c. *John didn't show everybody anything. d. John didn't show anybody anything. We leave the issue of the intervening constituents for future research. In section 7.2 we turn to some data from Bavarian which show that the role of intervening nonnegative constituents in the negative chain is subject to cross-linguistic variation.

7.2

Negative concord in Bavarian

In a recent paper, Bayer (1990) describes the phenomenon of negative concord in Bavarian, a German dialect. While the phenomena in WF are similar to Bavarian, there are a number of interesting differences which we will briefly discuss here.

158

PARAMETERS AND FUNCTIONAL HEADS

7.2.1 Negative chains with kein We have seen that geen-NPs cannot enter into a negative chain with niet in WE This constraint does not apply to Bavarian, as illustrated in 87 (Bayer 1990:15-16, ex. 5a, 5d, 7a, etpassim): (87) a. Ich bin froh, das ich keine Rwede nicht halden brauch. I am glad that I no talk not give need 'I am glad that I don't have to give a talk.' b. und keinen andern nichd leihden wiel and no other not tolerate wants 'and does not want to tolerate another' c. das keine Unanstendiheit nichd bassirt isd that no indecency not happened is 'that no indecency has occurred' An interesting contrast between Bavarian and WF is that kein-NPs in Bavarian behave exactly like bare negative quantifiers. Perhaps a way of accounting for the difference is to argue that kein in Bavarian is the head of the constituent — a DP — and is indeed both quantificational and negative, so that such negative constituents are of the same type as unanalysed quantifiers:

If this is true, then kein-NPs will be compatible with both nichd and with bare negatives. Bayer does not discuss the question whether nichd is compatible with nicht mehr. If nicht mehr is analyzed as being headed by the quantificational element, then we expect it should not be compatible with nichd. We leave this question open. 7.2.2

Intervening elements

Bayer's examples show that the negative chain is based on a strict c-command requirement between its links. The following examples (from Bayer 1990:15, ex. 6ac) show that even definite interveners interrupt the chain. We have seen in section 7.1 that definite interveners do not break the negative chain in WF: (89) a. *das koa Mensch des Zeich ned ooschaung woid that no man this stuff not look-at wants 'that no-one wanted to look at this stuff

NEGATIVE CONCORD IN WEST FLEMISH

159

b. *das koana in des Audo ned eig'schdieng is that no one in that car not entered is 'that no-one got in this car' c.

7.2.3

das neamads da Soffi ned wohdua woidd that no-one the Sophie not hurt wanted 'that no-one wanted to hurt Sophie'

Subjects and NC

Bayer shows that it is possible for subject NPs in the [NP,IP] position to enter into a negative concord relation with VP-internal negated constituents. This is illustrated in 90a-c (Bayer 1990:18-19, ex. 13a-13b, 15): (90) a. Zu der weldlichen Obrikeid had kein Mentsch kein Ferdrauen. to the mundane authority has no man no trust 'No one trusts mundane authority.' b. Gozeidank had keine Zeitung nichz erfarren. thank God has no paper nothing experienced 'Thank God has no paper nothing experienced.' c. das koa Mensch de Jager koa Bier ned zoid hod that no man the hunter no beer not paid has 'that no-one paid the hunter a beer' Bayer's rule of NC (1990:18) applies at the IP level. The WF parallels to 90 are ungrammatical because subject-negative constituents count as indefinite and are thus excluded from the canonical subject position as are all indefinite constituents. Such subjects necessitate the use of the existential construction. Bavarian (90b), for instance, would correspond to WF (91b): (91) a. Godzijdank eet-ter geneen gezette niets geweten. thank god has der no newspaper nothing known b. *Godzijdank eet geen gezette niets geweten. 7.2.4

[Spec,CP] and NC

Bayer (1990:23, n. 4) also gives the following example: (92) Keinen Wiederspruch gibd es nicht. no objection exists there not 'There is no objection.' The example shows that NC between a negative constituent in [Spec.CP] and IPinternal nicht is allowed in Bavarian. We have seen that this is not allowed in WF.

160

PARAMETERS AND FUNCTIONAL HEADS

8

Conclusions

The findings of the present article can be summarized as follows: a. Unlike Germanic-type languages — but like Romance-type languages — WF has a negative head projecting a NegP-dominating Infl. b. Unlike Germanic-type languages — but like Romance-type languages — WF has negative concord readings when two or more negative constituents cooccur. c. The negative sentential adverbial nie in WF adjoins to VP or a projection immediately dominating VP. Other negative constituents which take wide scope are scrambled and adjoined to the projection dominating nie. d. Negative concord relations can only be established between two or more negative constituents which have the same scope domains. e. The negative concord interpretation is obtained by a factorization process at LF which extracts the negation from the negative constituents and forms one complex operator binding several variables. f. Factorization of multiple negative constituents is a step-wise process which consists of LF-raising of the negative constituent and adjunction to the next higher negative constituent. The process takes place in a certain domain. When factorization cannot apply, the multiple negative constituents will be interpreted independently, each contributing an instance of negation, and so-called double negation readings will result. g. Negative constituents are classified according to the features of their heads, the relevant features being [neg] and [OJ. h. In order for two negative elements to enter into an NC relation, i.e., to be subject to factorization, their heads must share at least one relevant feature. i. The head of NegP in WF is en, which serves as a scope marker. In itself era does not carry negative meaning — it requires the presence of at least one negative constituent. j. The occurrence and position of negative elements at LF is determined by the Neg-Criterion, a well-formedness principle parallel to Rizzi's Mi-Criterion (1990b): The Neg-Criterion: i. Each Neg X° must be in a Spec-head relation with a Negative phrase ii. Each Negative phrase must be in a Spec-head relation with a Neg X° The correct configuration is either achieved at S-structure (for the negative constituent in [Spec,CP]) or at LF. Negative elements inside scope islands cannot be associated with the negative head.

NEGATIVE CONCORD IN WEST FLEMISH

161

Notes 1. This paper was presented in 1990-91 at various research seminars and we wish to thank the following people for their advice: Bob Fiengo, Janet Fodor, Jacqueline Gueron, Tony Kroch, Hans Obenauer, Jean-Yves Pollock, Ken Safir, Beatrice Santorini, the participants of the Seminaire interdepartmental de recherche en linguistique in Geneva, and the members of the CNRS Working Group on Comparative Syntax in Paris. Special thanks are due Richard Kayne and Luigi Rizzi for their comments on a previous draft. Needless to say none of those mentioned can be held responsible for the way we have used their comments. Together with Rizzi (1995) the paper is the basis for the research project on operator scope funded by the Fond National de Recherche Scientiflque (Switzerland), No. 1133542.92. With respect to sections 3, 4, 5 and 6, more recent research in the domain of negation has led us to formulate further proposals which differ from those proposed in the current paper. We refer the reader to Haegeman and Zanuttini (1991), Haegeman (1991) and Zanuttini (1991). L. Haegeman is essentially responsible for sections 2, 3, 4 and 8; R. Zanuttini is essentially responsible for sections 5, 6, 7 and 8. 2. For the pragmatic differences between the NP containing two instances of negation which cancel each other out (3) and the corresponding NP without any instance of negation (i) see Horn (1989). (i) A friendly man walked into the room. 3. For arguments that the postverbal elements such as nessuno in 4b, etc. are not to be treated as polarity items (as in Longobardi 1987) the reader is referred to Zanuttini (1987, 1989, 1991). Polarity items are usually seen as existential quantifiers. One syntactic test which distinguishes universal from existential quantifiers is that while the former can take the modifier almost, or its equivalent in the language studied, the latter cannot (cf. Hoeksema 1983). Take for example the universal quantifiers all and everyone in English and compare them with the existential quantifiers some and someone with respect to this property: as shown in (i) the contrast is extremely sharp: (i) a. All of my friends came. Almost all of my friends came. b. Everyone came. Almost everyone came. c. Some of my friends came. *Almost some of my friends came. d. Someone came. *''Almost someone came. One piece of evidence against the interpretation of the postverbal negative quantifier in Italian as a polarity item is that such negative constituents, like universal quantifiers, may be modified by quasi (ii), a possibility which is excluded for polarity items, which are usually taken to be existential quantifiers (cf. Zanuttini 1989 for discussion and examples). In French, negative elements such as personne contrast with polarity items such as grand chose in that the former are incompatible with pas (iia) while the latter require pas (iib): (ii) a. II n'a (*pas) fait rien. he ne has not done nothing 'He has done nothing.' b. II n'a *(pas) fait grand chose he ne has not done big thing 'He didn't do much.'

162

PARAMETERS AND FUNCTIONAL HEADS

We return to the status of negative constituents with respect to the typology of quantifiers below. We will see (section 3.2) that the distinction between existential and universal quantifiers is not sufficient to capture the behavior of the negative quantifiers. 4. The dialect we will be concerned with is spoken in the province of West Flanders, situated in the west of Belgium. The data discussed in this paper are taken from the dialect spoken in the Damme area, to the East of Bruges, including the villages Lapscheure, Hoeke, Sysele and Moerkerke. For some more details on the dialect see Haegeman (1992). Some form of NC is fairly widespread in Belgian and Dutch dialects of Dutch, with considerable cross-dialectal variation. Pauwels (1958) gives the following examples from the Belgian dialect of Aarschot (cited in den Beslen 1986:207): (i)

(ii)

(iii)

De dokter is nog niet geweest nie. the doctor has not yet been not 'The doctor has not been here yet.' 'k Heb niemand nie gezien nie. I have nobody not seen not 'I haven't seen anybody.' Daar had niemand geen hoge schoenen aan nie. there had nobody no high shoes on nie 'Nobody was wearing boots there.'

Observe that nie in these examples is absolutely sentence-final and follows participles (i, ii) or particles (iii). This suggests that nie in fact also follows the sentence-final I-node. In WF, examples (i-iii), with absolutely sentence-final nie, would not be grammatical. The corresponding WF sentences would be: (iv) (v) (vi)

Den dokteur is nog nie geweest (*nie). the doctor has not yet been 'k Een niemand nie gezien (*nie). I have nobody not seen Doar oat-ter niemand geen hoge schoenen aan (*nie). there had der nobody no high shoes on

We will see that in contrast with the Aarschot dialect, WF nie precedes sentence-final participles or particles. We discuss the distribution of the negative markers in WF in sections 3 and following. An early discussion of multiple negation phenomena in Dutch and in Flemish is found in Blancquaert (1923), who also points out the Afrikaans variant (cf. den Besten 1986):

(vii)

Ik kan nie kom nie I can not come not

Blancquaert's hypothesis is that the doubling of nie as in (vii) is only found in dialects which lack the negative clitic en. We will turn to a discussion of en in section 3.3. To our knowledge Blancquaert's hypothesis has not been tested. 5. Cf. Kayne (1989), Pollock (1989), Belletti (1990), Zanuttini (1990), among others. 6. For a discussion of the distribution of ne in modern French, see Ashby (1981), Moritz (1989). 7. Cf. Zanuttini (1990, 1992). In this paper we will not discuss the relative position of NegP in the split-Infl framework proposed, for example, by Pollock (1989), and developed in Belletti (1990), and Moritz (1989), for instance. 8. In 12 the subject position would be [Spec.NegPl. 9. We will argue that the descriptive generalization applies to Piedmontese (7b) at a more abstract level.

NEGATIVE CONCORD IN WEST FLEMISH

163

10. Recall that we assume here that the correct representation of negative quantifiers of the type of nessuno is as universal quantifiers followed by negation. See note 2 above, and section 4.3 for some problems. 11. We leave the status of English not for future research. If not counts as a Negative head then it is surprising that standard English lacks NC. 12. At this stage of our enquiry, our proposals differ from Pollock's (1989), in that we postulate that the presence of a NegP dominating Infl is subject to parametric variation. 13. We use the phrase "associated with the head" to cover two slightly different situations. Either negation is overtly expressed by the negative head on its own as in Italian, or the negative head marks the sentential scope of another negative constituent. In section 6.2.4, we will return to the disjunction in our descriptive generalization—that negation either is associated with the head Neg or c-commands it. 14. Bayer (1990) discusses data from Bavarian, another instance of a Germanic language with NC. We compare the Bavarian data with the WF data in section 7.2. 15. For a general description of Verb Second and a survey of the literature the reader is referred toVikner(1990). 16. We leave the exact position of nie for future research. It is conceivable that in a split-Infl framework with AgrP dominating TP (cf. Belletti 1990) nie is TP-adjoined. Our analysis differs sharply from Bayer's analysis of NC in Bavarian, in which it is proposed that "nicht is a syncategorematic expression . . . which in the unmarked case adjoins to V°" (1990:17). This is difficult to reconcile with data such as the following: (i) a. da Valere nie over dienen boek geklaapt eet that Valere nie about that book talked has 'that Valere has not talked about that book' b. Woari ee Valere nie ti over geklaapt? where has Valere not about talked 'What did Valere omit?' As pointed out by Bayer himself (1990:22, n. 2), nie precedes PPs which are subcategorized by V. He stipulates that the negative adverb in those cases adjoins to [PP+V]. In our analysis nie is adjoined to VP or the projection dominating it immediately and precedes VP-internal material. Whenever VP constituents precede nie we assume that they have been scrambled out of their base position. 17. We will return to the status of geen below, where we will in fact deny that it is a quantificational element itself. In section 5.2 we will provide a first sketch for the analysis of geen-NPs. We will elaborate this analysis in future research (Haegeman and Zanuttini 1995). 18. Let us provisionally assume that given a clause structure as in 12 above, the canonical subject position in non-negative sentences is [NP,IP] and in negative sentences it will be [NP,NegP]. 19. Der assimilates to ter when preceded by a voiceless consonant. Following Haegeman (1992), we assume that der cliticizes onto C°, as is the case for other subject clitics, and that the canonical subject position in these constructions is occupied by an expletive pro. 20. When it is subject, the negative constituent behaves like an existential quantifier and unlike a universal quantifier, which is admitted in the [NP,IP] construction. Universally quantified NPs are incompatible with the existential construction: (i) a. dan al de studenten dienen boek al een that all the students that book already have b. *dan-der al de studenten dienen boek al een that there all the students that book already have

164

PARAMETERS AND FUNCTIONAL HEADS

We conclude that universally quantified NPs count as definite, while negative NPs, indefinite NPs, etc., count as indefinite. Similarly, an NP such as de meeste studenten, literally translated as 'the most students' must occur in the canonical (definite) subject position, as a function of the presence of the definite article de: (ii)

dan de meeste studenten dienen boek gelezen een that the most students that book read have 'that most students have read that book'

The data in 26 and 27 are problematic for our assumption that negative constituents count as universal quantifiers. Note, though, that even when they are subjects in the existential construction the negative quantifiers admit modification by oast 'almost': (iii) a. dat-ter oast niemand dienen boek gelezen eet that-fe?- almost nobody that book read has 'that almost nobody has read that book' b. dan-der oast geen studenten dienen boek gelezen een that-rfer almost no students that book read have 'that almost no students have read that book.' The data suggest that the distinction between existential and universal quantifiers will not suffice to capture the behavior of the negative quantifiers in natural language. 21. The WF definiteness restriction on the canonical subject position is possibly related to the fact that this position is governed by a C with overt agreement: (i) a. da Marie werkt that Marie works 3sg b. dan die venten werken that those men work 3pl In Haegeman (1992) it is proposed that in WF, Agr in C assigns NOMINATIVE case under government to the subject position while in standard Dutch NOMINATIVE case is assigned by I under Spec-head agreement. 22. There is idiolectal variation among users. En is used to a larger extent by older speakers of the dialect, but younger users also may use en or have intuitions about it. Wim de Geest (personal communication) informs us that en is used frequently by speakers of East Flemish. For a discussion of the distribution of en in Flemish dialects the reader is referred to the following works, all written in Dutch: Jongen (1972), Koelmans (1967) and Tavernier (1959). We leave the analysis of dialects other than WF for future research. Stoops (1972) shows that in the Antwerp dialect en was already in decline in the 17th century, when it tended to be used more frequently in subordinate clauses than in main clauses. 23. As in French (iii, below), the WF negative clitic is also optionally present in sentences where the negative constituent is a subject, as in 26 and 27: (i) (ii) (iii)

dat-ter niemand dienen boek gelezen (en)-eet (=26b) dan-der geen studenten dienen boek gelezen (en) een (=27b) Personne (n')a lu ce livre. Nobody has read this book.'

In this respect WF and French differ from Italian, where non does not co-occur with the pre-verbal negative subject (iv):

NEGATIVE CONCORD IN WEST FLEMISH (iv)

165

Nessuno (*non) ha letto questo libro. Nobody non has read this book.

Note that the incorporation of en and the finite V i s a case of strong incorporation: unlike the Verb-Particle construction (v), or Verb Raising (vi), for instance (cf. Haegeman and van Riemsdijk 1986; Haegeman 1992), the finite V does not "excorporate" when it moves toC: (v) a. da-se da boek ip-[v stiert] that she that book up sends 'that she sends that book' b. ze stiert da boek ip. she sends that book up 'She sends the book.' (vi) a. da ze morgen dienen boek [v wilt] lezen that she tomorrow that book wants read 'that she wants to read that book tomorrow' b. Ze wilt morgen dienen boek lezen. she wants tomorrow that book read 'She wants to read that book tomorrow.' 24. Like French ne, WF en does not co-occur on past participles. (i)

*Valere ee nie en-geklaapt Valere has not en talked

But the negative clitic may be present on the finite auxiliary associated with the perfect participle: (ii)

Valere (en-)ee nie geklaapt

WF en, however, differs from negative markers of the type head in Romance in not being able to occur with infinitives, at least not for some speakers (including L. Haegeman): (iii) a. dan-k durven geen geld (*?en)-geven that I dare no money en give 'that I dare not to give any money' b. dan-k liever zoun niemand (*?en)-zien that I rather would nobody en see 'that I'd rather not see anyone' c. dan-k zoun willen nieverst noatoe (*?en) goan that I would want nowhere to en go 'that I'd rather not go anywhere' de Geest informs us that the corresponding sentences are grammatical in East Flemish with en. Koelmans, on the other hand, suggests that the use of en is restricted to co-occurring with the finite form of the verb. He cites one counterexample which he says "can be neglected" (1967:15): (iv)

Gaje vandaag niet en kaarten? go you today not en play cards

Since judgements on these data are quite delicate, we leave this issue open for future investigation. Another apparent discrepancy between WF en and negative markers of type head in Romance is that it can occur with imperatives:

166

PARAMETERS AND FUNCTIONAL HEADS (v)

(£rt)-komt nieverst an. en come nothing to 'Don't touch anything.'

Notice that in WF, contrary to Romance, imperatives do not employ a verbal form unique to their paradigm, but are morphologically indistinguishable from the present indicative. Since the constraint against negative heads occurring with imperatives in Romance affects only those forms which are unique to the paradigm of the imperative (see Rivero 1994; Kayne 1992; Zanuttini 1991), the occurrence of WF en with imperatives is not surprising. 25. In fact this statement needs to be qualified. Absolute negative en is found in the following example: (i) a. K'en doen. I en do 'I don't.' b. Z'en doet. she en does Example (i) illustrates the sole use (to our knowledge) of en with negative meaning on its own. It also illustrates one of the rare uses in WF of do as a substitute for a deleted VP. The sentence is quite frozen, though. Witness the fact that it is far less acceptable, if at all, with a past tense, or with a non-pronominal subject: (ii) a. ?? k'en degen I en did b. Z'en doet. she en does 'She doesn't.' c. ??? Marie en doet En is also found as a component of tenwoare 'unless', which introduces a counterfactual conditional clause. However this may well be a fossilized form, given that we also find it in the standard Dutch variant tenzij, where en is otherwise not admitted: (iii) a. tenwoare da-se tus bleef (WF) it en were that she home stayed 'unless she were to stay at home' b. Tenzij ze thuis bleef. (standard Dutch) Stoops (1972) shows that the use of en as the sole marker of negation was already rare in the 17th century Flemish dialect of Antwerp. En also co-occurs with restrictive moar. (iv)

K'en een moar drie boeken. I en have but three books

We assume that the negative component of moar is relevant here. Tavernier (1959) cites other examples of so-called "expletive en" in the dialect of Ghent. These are occurrences of en in contexts which are not strictly negative: - in clauses of comparison: (v)

'k goa ' tu zeggen gelakovda 't en es I go it to you say like it en is 'I'll tell it to you just as it is.'

- in temporal clauses introduced by voordat:

NEGATIVE CONCORD IN WEST FLEMISH

167

(vi)

Zie dadier wig zat vuur dat a op u kappe en komt. see that you here away are before that he on your hood comes 'Clear out before he gets you.' -after 'tenwoare 'unless': (vii)

't en woare da me tuis en bleeve it en be that we at home en stayed 'unless we stayed at home' The West Flemish parallels of (v-vii) sound unacceptable to L. Haegeman, though older speakers might admit them. Parallel uses of "expletive ne" are found in French. We do not go into theseu ses of en here. 26. In fact, this is not quite correct. When two negative constituents are dominated by the same VP projection, the NC reading is possible, and so is the double negation reading. However, to obtain the latter special emphasis is needed on both negative elements with a marked pause separating them: (i) K'en een an niemand niets gegeven. I en have to nobody nothing given (i) 'I gave nothing to nobody.' (ii) 'I did not give anything to anyone.' Often, in fact, such double negation readings have an echoic ring to them, being denials of preceding affirmations. Whenever we refer to NC readings between two negative constituents, the reader might wish to bear in mind that the alternative double negation reading is possible but is the marked interpretation. We assume that the double negation reading is obtained as in the examples in 2: i.e., by independent LF-raising of each negative constituent, which then contributes both its quantificational force and its negative meaning to the sentence interpretation. When more than two negative elements are present the double negation reading is difficult, even impossible to get. This is presumably because of processing reasons: (ii)

K'en an niemand niets nie gegeven. I have to nobody nothing not given 'I did't give anything to anyone.' In (ii) the double negation reading is inaccessible. 27. As pointed out already, WF differs from French with respect to the behavior of the negative adverb. Whereas WF nie enters in NC, French pas does not: (i) a. Je ne vois pas personne. I ne see not nobody b. Je ne vois personne. 'I don't see anyone.' (ii) K'en zien niemand nie. I en see nobody not (i). 'I don't see anyone.' (ii). 'I do not see nobody. In WF (example ii), a DN reading is also marginally possible. This reading is not available in French. Speakers of French accept the DN reading in (iii), which has an idiomatic ring to it: (iii)

Ce n'estpas rien. it is not nothing 'It is quite something.'

168

PARAMETERS AND FUNCTIONAL HEADS

In a discussion of Early Middle English negative concord, Jack (1978:299) says: "A general rule, common to all the texts that I have examined, is that ne ... nawt is not normally used in clauses containing other negative forms, e.g. nan, 'no', never 'never', nowher 'nowhere'." This suggests that NC in Early English is like that in French in that the negative marker nawt does not participate in it. In further research (Haegeman 1995) we will try to establish to what extent Old and Middle English negative concord is subject to any of the constraints observed for West Flemish. Shanklin (1990) discusses the development of negation in English, focusing especially on the interaction of negation and polarity items. 28. Notice that, given two (or more) quantifiers, it is not necessary that all of them be negative. It is possible to have a combination of a negative quantifier with one or more non-negative quantifiers, as shown by the following examples, corresponding to the (a) and (b) sentences in 29-30: (i) a. K'(en) een an niemand eentwa gezeid. I en have to nobody something said 'I didn't say anything to anyone.' b. Valere (en) ee nieverst eentwien gezien. Valere en has nowhere somebody seen 'Valere did not see anyone anywhere.' (ii) a. Valere (en) wilt an geen me.ns geld geven. Valere en wants to no person money give 'Valere doesn't want to give any money to anyone.' b. Gisteren (en)-oat er ier geen mens tyd. yesterday en had er here no person time 'Yesterday nobody had any time at all.' (iii) a. Valere (en)-leest nooit boeken. Valere en reads never books 'Valere never reads any books.' b. Valere (en) wilt an niemand geld geven. Valere en wants to nobody money give 'Valere doesn't want to give any money to anyone.' In (iv) two non-negative quantifiers combine with a negative element: (iv)

da Valere nooit entwien eentwa gezeid eet that Valere never anybody anything said has 'that Valere never told anyone anything'

We assume that in these cases the indefinite or quantificational elements act as polarity items in the scope of negation. In Romance also, the spreading of negation across all VPinternal elements is not obligatory, and a polarity item might be substituted for a negative element, as we can see in these examples from Italian: (v) a. Nessuno ha comprato alcun libra. nobody has bought any book 'Nobody has bought any book.' b. Non ho comprato alcun libra, non have bought any book 'I haven't bought any book.' In (va), the negative quantifier nessuno licenses the polarity item alcun. Observe that in (vb), the negative head non can also license the polarity item alcun. In WF the negative

NEGATIVE CONCORD IN WEST FLEMISH

169

head en cannot license a polarity item—the presence of another negative constituent is required: (vi) a. *K-en een eentwa gezeid. I en have something said b. K'en een an niemand eentwa gezeid. I en have to nobody something said 'I did not say anything to anyone.' The fact that en on its own cannot license a polarity item supports our proposal that it does not independently mark negation, but rather that it is a scope marker for negative constituents. In this respect, en is like French ne, which also does not suffice to license a polarity item: (vii)

II n'a *(pas) dit grand chose. he ne has not said big thing 'He didn't say much.' 29. As the reader will observe, we remain, in fact, rather vague about the position to which moved negative constituents adjoin. We assume that they scramble out of the VP and adjoin to the projection dominating nie. This may be VP, as suggested elsewhere in the text as a first approximation, or a higher projection such as TP in the split-INFL framework. We follow den Besten and Webelhuth (1987) in assuming that scrambling involves adjunction to a projection dominating VP. Not all VP complements are subject to scrambling; while definite NPs tend to scramble, indefinite NPs tend to stay VP-internal, and predicates of small clauses cannot scramble at all. This is not the place to go into a discussion of scrambling. For an introduction to the phenomenon see Haegeman (1990b, ch. 11, 1992, ch. 3). For more detailed discussion of scrambling see Grewendorf and Sternefeld (1990). 30. The interpretation with niets having wide scope is possible but it requires a slight intonation break after doavuoren and emphasis on niets. The sentence will have an echoic ring to it. We assume that the marked wide scope reading of niets will be obtained if niets scrambles out of the dominating VP and adjoins to a projection lower than that dominating the reason adverb. 31. Den Besten and Webelhuth (1987) assume that such adverbials are I'-adjoined. In a split-Infl framework in which AgrP dominates TP, Haegeman (1992) proposes that such adverbials are TP-adjoined. For our purposes it suffices to assume that sentential adverbials such as woarschijnlijk 'probably' are VP-external. 32. Following Huybreghts" "Antecedency Binding Constraint", Koster (1987:181) proposes that "categories in derived positions (that bind a trace) are islands". Similarly, Baker (1987) proposes that extraction from within a constituent which is in a derived A'-position is not possible. For a discussion of the extraction possibilities from Dutch PPs in general the reader is referred to van Riemsdijk (1978), Koster (1987:174-181) and to the discussion in Rizzi (1990a: 106-110). Cf. also den Besten and Webelhuth (1990) and Haider (1990). 33. This hypothesis predicts correctly that a constituent negation such as nie in (ia), which negates the universal quantifier al, cannot enter into an NC relation with the sentential negator nie (ib) or with a negative constituent which has wide scope (ic): (i) a. dan [nie al djoengers] da boek gekocht een that not all the children that book bought have 'that not all the children have bought that book' b. dan [nie al djoengers] da boek nie gekocht een that not all the children that book nie bought have 'that not all the children didn't buy that book'

170

PARAMETERS AND FUNCTIONAL HEADS c. dan [nie al djoengers] niets gezeid een that not all the children nothing said have 'that not all the children didn't say anything' (i.e. that not all the children kept silent, some children did talk)

Note that geen-NPs such as (ii) enter into an NC relation. We return to such NPs in section 5. (ii) a. dan-der doa geen studenten nie meer geweest een that-der there no students no more been have 'that no students went there any more' b. dan-ze neu nooit geen studenten nie meer zien that they now never no students no more see 'that they don't see any students any more' 34. Examples 38b and 38c will have a DN reading. In 38b an niemand will be taken to enter into an NC reading with nie. Niets will be interpreted independently. In 38c the NC chain will consist of niemand and niets', nie will have independent negative force. 35. Longobardi (1991) shows that negative quantifiers in Italian are sensitive to island effects. 36. Cf. Haegeman and van Riemsdijk (1986) for a description of VPR in WF and Zurich German. See also Haegeman (1992) for further discussion and the adjunction analysis of VPR adopted in the present paper. In these papers it is shown that VPR and VP topicalization create islands for movement both at S-structure and at LF. An issue that will remain undiscussed here for reasons of space is the exact category of the constituent that moves under VP topicalization and VPR. Perhaps rather than a VP, a projection of a functional head (say TP) is moved. We might then argue that the negative constituents must scramble because they need to take scope over this higher projection. Given that we have left the split-INFL hypothesis and the issue of functional heads composing INFL almost entirely out of the discussion in this paper, we do not go into this issue any further here. 37. As pointed out by Rizzi (personal communication) the VP topicalization data can be reproduced for Italian: (i) a. Vedere Maria non vorrei. see Maria non I-would want 'To see Mary, I wouldn't want.' b. *Vedere nessuno non vorrei. see nobody non I-would want 'Not to see anyone, I would want.' c. Non vedere nessuno vorrei. non see nobody I-would want 'Not to see anyone, I would want.' (ii) a. Detto questo non ho di certo. said this non I-have certainly 'Said this, I certainly have not.' b. *Detto niente non ho di certo. said nothing non I-have certainly 'Not said anything, I certainly have not.' The data above show that the link between non and the post-verbal negative element (nessuno, niente) is sensitive to the island effects due to VP topicalization. Various explanations have been offered for the island effects created by VP topicalization and by VPR, to be discussed in section 4.3.2. In Haegeman (1992, based on Baker 1987),

NEGATIVE CONCORD IN WEST FLEMISH

171

it is argued that extraction out of the VP affected by VP topicalization or VPR is banned because the VP is moved in an adjoined position, an A'-position. Koster (1987:181) invokes Huybregts' Antecedency Binding Constraint. In terms of the Barriers framework we can say that the VP affected by VP topicalization or by VPR is in an adjoined position where it will constitute a barrier since it is not L-marked. 38. Verb Raising is illustrated in (i). The non-finite verb kopen 'buy' is reordered with respect to ziet 'sees', the finite verb of which it is a complement: (i) (ii)

dat Jan [Marie dat boek kopen] ziet that Jan Marie that book buy sees dat Jan [Marie dat boek t;] ziet kopenj

Example (i) is the multi-clausal D-structure, in which the perception verb ziet takes as a complement the infinitival VP headed by kopen. Example (ii) has the properties of a mono-clausal structure as a result of Verb Raising, the infinitive kopen has been moved to the right of the higher verb. Various analyses have been proposed in the literature (cf. Haegeman and van Riemsdijk 1986 for a survey of some of the literature; also Haegeman 1992). In this paper we assume that Verb Raising is a form of incorporation (cf. Baker 1988), where the matrix verb incorporates the head of its complement VP. 39. This is so at least for some speakers, including L. Haegeman. The DN reading is available but has a strong echoic ring to it. 40. A complication for this analysis is that, as seen in section 3.2, negative quantifiers act like existentially quantified NPs, in that they cannot occur in the canonical subject position and necessitate the existential pattern: (i) a. *da niemand dienen boek nie gelezen oat that nobody that boek not read had b. dat-ter dienen boek niemand nie gelezen oat that there that book nobody not had read that 'That nobody had read that book." Note, though, that in (ib) niemand can still be modified by oast 'almost' (47b, iia), in contrast with existentially quantified NPs (iia-b): (ii) a. Oast niemand oat-ter dienen boek gelezen. almost nobody bad-there that book read 'Hardly anyone had read that book.' b. *Oast enigte studenten oan-der da gelezen almost some students had there that read c. *Oast eentwien oat-ter da gelezen

almost someone had there that read

41. Cf. Higginbotham and May (1981) for a formal discussion of quantifiers which bind two or more variables. 42. French is a case in point. We have seen before that pas does not enter into an NC reading: (i) (*) Je n'ai pas vu personne. I ne have not seen nobody (a) 'I have not seen nobody.' (b) *'I haven't seen anybody.' Pas and personne in (i) each contribute their own negative meaning to the sentence, leading to double negation. NC between pas and personne is not possible. Jack (1978) suggests that in Early Modern English nawt also did not co-occur with other negative elements.

172

PARAMETERS AND FUNCTIONAL HEADS

43. Various people, including Hans Obenauer, have pointed out the following problem for the strong correlation which we have posited in the introduction of our paper between NC and the presence of a Negative Head. If NC is possible within raised or topicalized VPs then we cannot maintain that NC is dependent on the presence of a Negative Head, since we assume that NegP dominates IP. Two possible answers can be envisaged to this objection: (i) Perhaps we should simply weaken the correlation between NC and a Negative head. We might perhaps say that while whenever there is a negative head, NC obtains, the opposite does not necessarily hold and NC is not dependent on a Negative Head. Or else we might say that NC within VP is not exactly the same process as NC at the IP level and that only the latter depends on the presence of a negative head, (ii) Alternatively, and perhaps more interestingly, we should take seriously the possibility that VPR and VP topicalization involve TPs rather than VPs. If this were the case we could indeed propose that these non-finite TPs also are dominated by a NegP. However, we would then have to stipulate that the Neg Head of the NegP dominating the non-finite TP must be non-overt. 44. If they have the appropriate configurational relation with nie, of course. 45. Cf. Vanacker (1965) for a discussion (in Dutch) of the negative paradigm in Flemish dialects. 46. Substituting nie meer for nie yields grammatical examples in the case of 49 as well, as shown by the following examples: (i) a. da Valere niemand nie meer kent that Valere nobody no more knows "that Valere doesn't know anyone any more' b. da Valere dienen boek nieverst nie meer vindt that Valere that book nowhere no more finds 'that Valere doesn't find that book anywhere any more' c. da Valere tegen niemand nie meer klaapt that Valere against nobody no more talks 'that Valere doesn't talk to anyone any more' d. dat-ter hier niemand nie meer geweest eet that der nobody no more been has 'that nobody has been here any more' 47. Cf. Vanacker (1975) for a discussion of the distribution of this pattern. Vanacker signals that geen-NPs also cannot co-occur with nie in the Antwerp dialect. 48. As was the case with a single geen-NP, the ungrammaticality of 52 can be remedied by substituting nie meer for nie (with a slight semantic shift): (i) a. dat-ter niemand geen geld nie meer oat that der nobody no money no more had 'that nobody had any money left' b. da Valere an niemand geen boeken nie meer geeft that Valere to nobody no books no more gives 'that Valere doesn't give any books to any one any more' c, da Valere nooit geen boeken nie meer leest that Valere never no hooks no more reads 'that Valere doesn't read books any more' 49. For further discussion of inlervening non-negative constituents the reader is referred to section 7.1.

NEGATIVE CONCORD IN WEST FLEMISH

173

50. In 55b the double negation reading is available. The NC reading is available when nie is replaced by nie meer. (i)

da Valere nie lange nie meer gewerkt eet that Valere not long no more worked has 'that Valere did not continue working for long'

In 55a the replacement of nie by nie meer is not possible presumably because one occurrence of nie meer would be redundant. The translation suggests that when 55a has the double negation reading, meer is interpreted as an adverb of frequency. 51. Note in passing that it also will not do to say that the negative element geen in geen-NPs cannot take scope outside the constituent and that this explains why it cannot enter into an NC relation with elements which have sentential scope. It is clear that the negative element in what we have been calling geen-NPs can take sentential scope: (i)

(ii)

dat-ter da geen mens gezeid eet that der that no person said has 'that nobody has told that' dan-k geen boeken gekocht een that I no books bought have 'that I did not buy any books'

That geen-NPs can take scope outside VP is also supported by the fact that they co-occur with en: (i') (ii')

dat-ter da geen mens gezeid en-eet dan-k geen boeken gekocht en-een

The negation contributed by the negative specifier geen (whose structural position we will return to presently) determines the negative status of the entire constituent, which may take sentential scope. Geen-NPs can also take scope over reason adverbs: (iii)

dat-ter geen mens doavuoren weg-goat that ter no person therefore away-goes 'that there is nobody who leaves for that reason'

52. Consider the following examples which illustrate the possessor construction in WF: (i)

(ii)

(iii)

K'een [Valere zenen boek] gelezen. I have Valere his book read 'I have read Valere's book.' K'een [Valere zenen boek] nog nie gelezen. I have Valere his book not yet read 'I haven't read Valere's book yet.' [Valere zenen boek] een-k nog nie gelezen. Valere his book have I not yet read.

Examples (ii) and (iii) provide evidence that the bracketed string "NP-possessive-NP" is one constituent, since it is subject to scrambling and can be moved to [Spec,CP], The NP Valere cannot scramble or move to [Spec,CP] on its own: (iv) (v)

*K'een Valere zeker [zenen boek] gelezen I have Valere certainly his book read *Valere een-k [zenen boek] gelezen.

Consider also the following examples:

174

PARAMETERS AND FUNCTIONAL HEADS (vi) a. dat-ter [eentwien zenen velo] kapot was that der somebody his bicycle broken was 'that somebody's bicycle was broken' b. *dat [eentwien zenen velo] kapot was (vii) a. dat-ter gisteren [niemand zen werk] gereed was that there yesterday nobody his work finished was 'that nobody's work was ready yesterday' b. *? da [niemand zen werk] gisteren gereed was

The NP-possessive-NPs in (vi) and (vii) necessitate the der construction, showing that they behave like indefinite NPs. The indefiniteness is due to the presence of the quantificational or negative element. Example (viib) might be marginally acceptable with an echoic reading. Further research has to establish the properties of such constituents (Haegeman 1992). The -s genitive which is available in standard Dutch (viiia) is not available in West Flemish (viiib): (viii) a. Ik heb niemands geld genomen. I have nobody's money taken 'I haven't taken anybody's money.' b. *k-een niemands geld gepakt 53. Geen een 'no one' is found in many Dutch dialects (cf. Koelmans 1971). For 60, we tentatively suggest the following structure:

We assume that the rightward incorporation of gen by eenen is a PF process (cf. Haegeman 1992, ch. 2, for evidence that rightward procliticization of subject clitics is a PF phenomenon in WF). Dialectal geneen corresponds to standard Dutch geen enkel: (ii)

Ik heb geen enkel boek. I have no single book 'I don't have a single book.'

For further discussion and alternative analyses see Haegeman and Zanuttini (1995). 54. NC in NPs is not a general property of Flemish. Vanacker (1975) points out that the construction is relatively new. It is found in the Flemish spoken in the north of France, and in coastal WF. The construction does not occur in earlier Dutch or its dialects. As a first approximation, 62 would have the structure in (i). We assume that the relation between nie vele and geen is established by Spec-head agreement. In section 6 we will see that such a mechanism is independently needed to account for the relation between a negative constituent in [Spec,CP] and en, the head of NegP in C (cf. also Haegeman and Zanuttini 1995).

NEGATIVE CONCORD IN WEST FLEMISH

175

55. The quantification in nie meer is over times. 56. The same analysis would apply to constituents such as nie vele 'not much', nie genoeg 'not enough', nie dikkerst 'not often', etc. These are all incompatible with nie in an NC reading but enter an NC relation with nie meer on the one hand and with geen-NPs on the other: (i) a. Z'een doar nie vele nie gekocht. they have there not much not bought They rarely did not buy anything there.' (DN) b. Z'een doar nie vele nie meer noatoe geweest. they have there not much no more to been 'They did not often go there any more.' (NC) (i.e., they stopped going there frequently c. Z'een doa nie vele geen boeken gekocht. they have there not much no books bought They did not often buy books there.' (NC) 57. Schematically we assume that the raising will work as follows. Starting from the structure in (i):

LF-raising of m'ek is possible since it shares one feature ([neg]) with the higher quantifier. This results in (ii):

176

PARAMETERS AND FUNCTIONAL HEADS

The complex constituent created in (ii) will raise to adjoin to the higher negative constituent, provided [Fx] is either [neg] or [F,]. Note that this analysis presupposes that nlets- heads the complex negative quantifier resulting from the adjunction. Ultimately the negation will be factored out of the complex constituent resulting from the LF-movement. 58. As a first explanation we propose the following. Based on Rizzi (1990a:87) let us assume that traces must be properly head-governed, and that proper head government is subject to a directionality requirement: the governing head must be in the canonical configuration with the governee. We might also consider the possibility that two negative constituents are extraposed: (i)

dan-ze geklaapt een tegen niemand over niets that-they talked have against nobody about nothing

The intuitions on this example are not as sharp as one might wish, but it seems (to L. Haegeman) that the natural interpretation is one where the two negative constituents are interpreted as (parallel) afterthoughts: 'they did not talk to anyone, they did not talk about, anything.' An NC reading is not available, it seems. Note that the type of PP which is extraposed in (i) can regularly be subject to extraposition: (ii) a. dan ze dikkerst geklaapt een tegen Valere that they often talked have against Valere 'that they have often talked to Valere' b. dan ze dikkerst geklaapt een over den verus that they often talked have about the removal 'that they have often talked about the removal' c. dan ze dikkerst geklaapt een tegen Valere over den verus The absence of the NC reading suggests that the LF-movement involved in the negative factorization is not possible in the extraposition site, i.e., adjoined to IP. This is perhaps again an ECP effect: the traces resulting from the adjunction of the negative constituents will not be properly governed. 59. This statement needs modification. Indeed, just as it is possible to assign sentential scope to niets in 67b (repeated here as (ia)), as a marked option, it is possible, marginally to find 67b withew, as in (ib): (i) a. Valere ee doavuoren niets gedoan b. Valere en-ee doavuoren niets gedoan In order for (ia) to be assigned a wide scope reading of niets, emphasis is needed on niets. The same applies to (ib). Indeed, the sentences will have an echoic ring to them. 60. As mentioned before the position of Neg in the Split-INFL framework of Pollock (1989) remains to be established. 61. It is less obvious why en cannot be related to the negative constituent in its VP-internal base position: (i)

*Valere en-ee doavuoren [ vp niets gedoan]

In fact, the question needs to be worded slightly differently: why do VP constituents which have wide scope scramble obligatorily? One possibility is to say that such constituents have to be governed by the negative head (en or its non-overt counterpart) at S-structure, hence must escape the VP barrier. The details of this proposal need to be worked out, especially in light of the interaction between the split-Infl analysis and our proposals. 62. In 71 b and 75a the finite V ee with the procliticized en has moved under C by Verb Second.

NEGATIVE CONCORD IN WEST FLEMISH

177

63. We thank Luigi Rizzi for pointing out the relevance of the Mi-Criterion to us. Obviously we ourselves are responsible for the way in which we have used his suggestion. 64. Notice that in 81c the intervening element al da geld is definite. The problem though is that it is a scope-bearing element. The same observation applies to 8Id. 65. As pointed out to us by Richard Kayne (personal communication).

References Abney, S. 1987. The English Noun Phrase in its Sentential Aspects. MIT dissertation. Ashby, W. 1981. "The Loss of the Negative Particle ne in French." Language 57:674-687. Baker, M. 1985. "The Mirror Principle and Morphosyntactic Explanation." Linguistic Inquiry 16:373-415. . 1987. "Against Reanalysis of Heads," ms, McGill University. . 1988. Incorporation: A Theory of Grammatical Function Changing. Chicago: University of Chicago Press. Bayer, J. 1990. "What Bavarian Negative Concord Reveals about the Syntactic Structure of German," in J. Mascaro and M. Nespor (eds.), Grammar in Progress. GLOW Essays for HenkvanRiemsdijk. Foris: Dordrecht. 13-24. 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. 1986. "Double Negation and the Genesis of Afrikaans," in P. Muysken and N. Smith (eds.), Substrata versus Universals in Creole Genesis. Amsterdam: John Benjamins. 185-230. den Besten, H. and G. Webelhuth. 1987. "Remnant Topicalization and the Constituent Structure of the VP in the Germanic SOV Languages," paper presented at the GLOW Workshop on Dialectology, GLOW Colloquium, Venice. . 1990. "Stranding," in G. Grewendorf and W. Sternefeld (eds.), Scrambling and Barriers. Amsterdam: John Benjamins. 77-92. Blancquaert E. 1923. "Over dubbele ontkenningen en nog wat," Handelingen 6de Vlaamsch Filologencongres, 60-69. Burridge, K. 1983. "On the Development of Dutch Negation," in H. Bennis and W. van Lessen Kloeke (eds.), Linguistics in the Netherlands. Dordrecht: Foris. 31-40. Cardinaletti, A. 1990a. Impersonal Constructions and Sentential Arguments in German. Padua: Unipress. . 1990b. "Pronomi nulli e pleonastic! nelle lingue germaniche e romanze," tesi di dottorato, Padova. Evers, A. 1975. The Transformational Cycle in Dutch and German. Bloomington, Indiana: Indiana University Linguistics Circle. Goodall, G. 1987. Parallel Structures in Syntax. Cambridge: Cambridge University Press. Grewendorf, G. and W. Sternefeld (eds.) 1990. Scrambling and Barriers. Amsterdam: John Benjamins. Haegeman, L. 1990a. "Subject Pronouns and Subject Clitics in West Flemish," The Linguistic Review 7:333-363. . 1990b. "The Syntax of Motional goan in West Flemish," in R.B. Bennema and P. Coopmans (eds.), Linguistics in the Netherlands. Foris: Dordrecht. 61-71. . 1991. "Negative Concord, Negative Heads," in D. Delfitto, M. Everaert, A. Evers and F. Stuurman (eds.), Going Romance and Beyond. Utrecht: OTS Working Papers. 46-81.

178

PARAMETERS AND FUNCTIONAL HEADS

. 1992. Generative Syntax. A Case Study in West Flemish. Cambridge: Cambridge University Press. . 1995. "The Development of the Indefinite Pronouns in the West Flemish Dialect," ms, University of Geneva. [In preparation.] Haegeman, L. and H. van Riemsdijk. 1986. "Verb Projection Raising, Scope and the Typology of Rules Affecting Verbs," Linguistic Inquiry 17:417-466. Haegeman, L. and R. Zanuttini. 1991. "Negative Heads and the Neg Criterion," The Linguistic Review 8:23 3-251. . 1995. "The Structure of Negative Quantifiers in West Flemish," ms, University of Geneva. [In preparation.] Haider, H. 1990. "Topicalization and Other Puzzles of German Syntax," in in G. Grewendorf and W. Sternefeld (eds.), Scrambling and Barriers. Amsterdam: John Benjamins. 93112. Higginbotham J. and R. May. 1981. "Questions, Quantifiers and Crossing," The Linguistic Review 1:41-80. Hoeksema, J. 1983. "Negative Polarity and the Comparative," Natural Language and Linguistic Theory 1:403-434. . 1988. "A Constraint on Governors in the West Germanic Verb Cluster," in M. Everaert, A. Evers and M. Trommelen (eds.), Morphology and Modularity, in Honour of Henk Schultinck. PLS29. Foris: Dordrecht. Horn, L. 1989. A Natural History of Negation. Chicago: University of Chicago Press. Jack, G.B. 1978. "Negative Adverbs in Early Middle English," English Studies 69:294-309. Jespersen, O. 1965. The Philosophy of Grammar. New York: Norton and Company. [Originally published 1924.] Jongen, R. 1972. "Nog eens over de verbreiding van het ontkennende en of: welke is de bruikbaardheid van de RNDA," De Nieuwe Taalgids 65:51-57. Kayne, R. 1984. Connectedness and Binary Branching. Dordrecht: Foris. . 1989. "Notes on English Agreement," CIEFL Bulletin 1:40-67. . 1992. "Italian Negative Imperatives and Clitic Climing," in L. Tasmowsky and A. ZribiHertz (eds.), Hommages a Nicolas Ruwet. Ghent: Communication and Cognition. 300-312. Koelmans, L. 1967. "Over de verbreiding van het ontkennende en" De Nieuwe Taalgids 60:12-18. . 1971. "Enkele opmerkingen over het gebruik van geen," Taal en Tongval 23:52-55. Koster, J. 1975. "Dutch as an SOV language," Linguistic Analysis 1:111-136. . 1987. Domains and Dynasties. Dordrecht: Foris. Longobardi, G. 1987. "Negation Parameters in Romance Languages and Dialects," paper presented at the GLOW Workshop on Dialectology, GLOW Colloquium, Venice. . 1991. "In Defense of the Correspondence Hypothesis: Island Effects and Parasitic Constructions in LF," in J. Huang and R. May (eds.), Logical Structure and Linguistic Structure. Dordrecht: Reidel. 149-196. Mascaro, J. and M. Nespor (eds.) 1990. Grammar in Progress. GLOW Essays for Henk van Riemsdijk. Dordrecht: Foris. May, R. 1985. Logical Form: Its Structure and Derivation. Cambridge, Mass.: MIT Press. Mitchell, B. 1985. Old English. Oxford: Clarendon Press. 2 vols. Moritz, L. 1989. "Syntaxe de la negation de phrase en francais et en anglais," Memoire de licence, Universite de Geneve. Ouhalla, J. 1990. "Sentential Negation, Relativised Minimality and the Aspectual Status of Auxiliaries," The Linguistic Review 7:183-231. Parry, M. 1985, "The Dialect of Cairo Montenotte, Ph.D., University of Wales.

NEGATIVE CONCORD IN WEST FLEMISH

179

Pauwels, J.L. 1958. "De expletieve ontkenning nie(t) aan het einde van de zin in het Zuidnederlands en het Afrikaans." [Appendix in J.L. Pauwels, Het dialect van Aarschot en omstreken. Published by het Belgisch Interuniversitair Centrum voor Neerlandistiek. Bouwstoffen en Studien voor de geschiedenis en lexicografie van het Nederlands 5:1, 2, 434_477. Pollock, J.-Y. 1989. "Verb Movement, Universal Grammar, and the Structure of IP," Linguistic Inquiry 20:365^24. Reinhart, T. 1976. "The Syntactic Domain of Anaphora," Ph.D., MIT. Riemsdijk, H. van. 1978. A Case Study in Syntactic Markedness. Dordrecht: Foris. Rivero, M.-L. 1994. "Clause Structure and V-Movement in the Languages of the Balkans," Natural Language and Linguistic Theory 12:63-120. Rizzi, L. 1990a. Relativized Minimality. Cambridge, Mass.: MIT Press. . 1990b. "Speculations on Verb Second," in J. Mascaro and M. Nespor (eds.), Grammar in Progress. GLOW Essays for Henk van Riemsdijk. Dordrecht: Foris. 375-386. . 1995. "Residual Verb Second and the W/t-Criterion." In L. Rizzi and A. Belletti (eds.), Parameters and Functional Heads. New York: Oxford University Press. 63-90. [This volume.] Rizzi, L. and I. Roberts. 1989. "Complex Inversion in French," Probus 1:1-30. [Also published in this volume, 91-116.] Shanklin, M.T. 1990. "The Grammar of Negation in Middle English," Ph.D., University of Southern California. Sportiche, D. 1988. "A Theory of Floating Quantifiers and Its Corollaries for Constituent Structure," Linguistic Inquiry 19:425—449. Stoops, Y. 1972. "lets over de dubbele ontkenning bij Willem Ogier," Taal en Tongval 23:142152. Tavernier, C. 1959. "Over Negatie en expletief en in het Gents dialect," Taal en Tongval 11:245-252. Vanacker, V.F. 1965. "Tegenstellingen bij een negatiesyntagma in de Znl dialekten," Taal en Tongval 17:41-50. . 1975. "Substantiefgroepen met dubbele ontkenning in zuidwestelijke dialekten," Taalen Letterkundig gastenboek voor Prof. Dr. G.A. Van Es. 127-133. Vikner, S. 1990. "Verb Movement and the Licensing of NP-Positions in the Germanic Languages," Ph.D., University of Geneva. Webelhuth, G. 1990. "Diagnostics for Structure," in G. Grewendorf and W. Sternefeld (eds.), Scrambling and Barriers. Amsterdam: John Benjamins. 3-39. Zanuttini, R. 1987. "Negazione e concordanza negative in italiano e in piemontese," Rivista di Grammatica Generativa 12:135-149. . 1989. "Two Strategies for Negation: Evidence From Romance," in J. Powers and K. de Jong (eds.), Proceedings ofESCOL '88. Columbia: Ohio State University. 535-546. . 1990. "Two Types of Negative Markers," Proceedings of NELS 20. 2:517-530. . 1991. Syntactic Properties of Sentential Negation: A Comparative Study of Romance Languages," Ph.D., University of Pennsylvania. . 1995. "On the Relevance of Tense for Sentential Negation," in L. Rizzi and A. Belletti (eds.), Parameters and Functional Heads. New York: Oxford University Press. 181207. [This volume.]

This page intentionally left blank

5

On the Relevance of Tense for Sentential Negation Raffaella Zanuttini

Much recent work has focused on the issue of which functional categories are present in a sentential structure. Both Moro (1987) and Pollock (1989), independently and on the basis of quite different considerations, suggest that in Romance and in English the Infl node should be split into two components, a Tense Phrase and an Agreement Phrase. Kitagawa (1986, section 2.4.3) argues for the existence of a negative head in Japanese, an idea extended to Romance by Kayne (1989a:242) and to English by Pollock (1989), who suggest that sentential negative markers can be analyzed as belonging to a Negative Phrase. Rivera (1994) proposes the existence of a functional projection of a modal or aspectual nature in the Balkan languages. Shlonsky (1989), based on facts from Semitic languages, suggests that the functional category Agreement Phrase should be further split into three distinct components— PersonP, NumberP and GenderP. These proposals raise interesting questions concerning the relations holding among functional categories. For instance, does the presence of one particular functional category imply the presence of another? Further, can we identify conceptual relations among functional categories which could shed light on their distribution and hierarchical structure? In this paper,1 I address these two questions, focusing on the projections of the functional categories negation and tense (henceforth, NegP and TP respectively). The languages I will consider are the Romance languages and English. I will show that in both Romance and English the functional category NegP is parasitic on the functional category TP, i.e., NegP can only occur in the sentence if TP is present. This can be expressed as a one-way implication: if NegP, then TP. I will express this relation by means of the notion of selection?' I will argue that in these language groups the head of the functional category NegP takes TP as its obligatory complement, and that this relation determines the ordering of these two categories in the structure: Given a TP, the NegP will be generated to its left so as to satisfy the selectional requirements of its head. No other constraint needs to be specified 181

182

PARAMETERS AND FUNCTIONAL HEADS

in the grammar to explain the structural position of NegP: A NegP can be generated anywhere it can find a TP which the head Neg° can take as its complement.3 The paper has the following structure. First I will discuss the case of Romance: I will argue that two syntactically different types of negative marker should be distinguished in this language family. One type of negative marker is the head of a NegP, while the other does not appear in the functional category NegP but is an adverbial element in a lower position in the structure. Focusing on the negative marker which is the head of NegP, I will show that the constraints on its distribution can be explained by saying that it selects TP as its obligatory complement. Then I will discuss the case of English: I will argue that also in English we must distinguish two syntactically different types of negative markers: one the head of NegP and the other an adverbial element. I will show that in English as well as in Romance the distribution of NegP can be adequately accounted for in terms of its relation to the functional category TP. My approach will lead me to propose an analysis of English imperatives according to which they are not forms belonging to a morphological paradigm of their own. I will suggest, instead, that they are instances of present subjunctives in root contexts.4

1 Negation and Tense in Romance As a preliminary to the discussion of the relation between sentential negation and tense in Romance, I will present some reasons why we need to distinguish between two different syntactic types of negative marker in this language family. I will then present the data on the basis of which I will argue for a dependence of the head of NegP on the functional category TP.5

1.1

Two Types of Negative Markers

In Romance, as is well known, some negative markers precede the finite verb, while others follow it. In Italian, for example, the sentential negative marker non immediately precedes the finite verb, as shown in la, and cannot follow it, as shown in Ib. On the other hand, in Piedmontese, a dialect spoken in northern Italy, the sentential negative marker nen follows the finite verb, as shown in 2a, and cannot precede it, as in2b: (1) a. Maria non parla molto. (Italian) Mary NEC talks much 'Mary doesn't talk much.' b. *Maria parla non molto. (2) a. Maria a parla nen tant. (Piedmontese) Mary CL talks NEC much 'Mary doesn't talk much.' b. *Maria a nen parla tant.

ON THE RELEVANCE OF TENSE

183

When the verb is an auxiliary, followed by a past participle, Italian non precedes the auxiliary, as shown in 3a and cannot follow it, as in 3b, while Piedmontese nen follows the auxiliary and precedes the past participle, as in 4: (3) a. Maria non ha parlato molto. (Italian) Mary NEC has talked much 'Mary hasn't talked much.' b. *Maria ha non parlato molto. (4) a. Maria a 1'ha nen parla tant. (Piedmontese) Mary CL has NEC talked much 'Mary hasn't talked much.' b. *Maria a nen 1'ha parla. These contrasts are parallel to the contrast we find in French between preverbal ne on the one hand and postverbal/ras on the other: French ne patterns with Italian non, while French pas patterns with Piedmontese nen. We can make a first approximation that groups the negative markers in Romance into two categories: negative markers which pattern like Italian non, such as the ones of Spanish, Catalan, Romanian and Portuguese, for example; and negative markers which pattern like Piedmontese nen, such as those of Occitan, Franco-Provencal and Milanese, to mention only a few languages of this type.6 I will argue that these two types of negative markers do not appear in different positions with respect to the verb because of independent differences between the two groups of languages I mentioned, but rather because they belong to distinct syntactic types. I will provide evidence for this distinction by drawing examples from Italian and Piedmontese mainly, for ease of exposition; but the same pattern can be reproduced with examples from any other pairs of languages representative of the two different groups. The distribution of the Piedmontese negative marker nen as a sentential operator overlaps with some adverbs, for example gia 'already', anco 'yet', mai 'never', sempe 'always',pi 'no more'. As we can see in 5, these adverbs follow the finite verb, just like the negative marker nen: (5) Maria a parla sempe/anco/gia/pl. Mary CL talks always/still/already/no more 'Mary always/still/already/no more talks.' Nen can occur in combination with some of these adverbs, as we see in 6 and 7, in which case it must follow them: (6) Maria a canta pi nen. Mary CL sings more not 'Mary doesn't sing anymore.' (7) Maria a parla anco nen. Mary CL talks yet not 'Mary isn't talking yet.'

184

PARAMETERS AND FUNCTIONAL HEADS

These examples also show that nen need not be immediately adjacent to the verb: Lexical material can intervene to separate it from the verb with which it is associated. Italian non differs from Piedmontese nen in several respects. First, it does not share its distribution with VP adverbs: non occurs between the subject (when one is present) and the verb, while the adverbs in question follow the verb, as shown in 8. Example 9 shows that these adverbs in Italian cannot occur in the same position as non: (8)

Maria non ha ancoralmai telefonato. Mary NEC has yet/never called 'Mary hasn't yet called.'/'Mary has never called.'

(9) *Maria gia/mai ha telefonato. Mary already/never has called 'Mary has already/never called.' Furthermore, in contrast to Piedmontese nen, the Italian negative marker non must be immediately adjacent to the verb; even if we take sentential adverbs which can precede the finite verb, they cannot intervene between non and the verb, as shown by 10 and 11: (10) *Maria non ieri cantava. Mary NEG yesterday sang 'Mary wasn't singing yesterday.' (11) *Maria non probabilmente canta. Mary NEC probably sings 'Mary probably isn't singing.' The only lexical items which can intervene between non and the verb are pronominal clitics, as illustrated in 12: (12)

Maria non gli parla. Mary NEG to-him talks 'Mary doesn't talk to him.'

A further difference between the Italian and Piedmontese negative markers is that in Piedmontese, in purpose clauses the negative marker nen can occur to the left of the complementizer of the dependent clause: (13) A 1'ha fait parej per nen ch'a se stofieissa. CL has done so for NEG that CL self bored 'He did it that way so that he wouldn't get bored.' This property of nen, like the distributional properties described before, is not unique to the negative marker of Piedmontese, but is shared by other postverbal negative markers in Romance, as shown in 14:

ON THE RELEVANCE OF TENSE

185

(14) a. Je lui ai menti pour (ne) pas qu'il parte. (French) I to-him have lied for NEC that he leaves 'I lied to him so that he wouldn't leave.' b.

Li mentigueri per pas'que se'n anesse. (Occitan) to-him I lied for NEG that he leave 'I lied to him so that he wouldn't leave.'

Indeed, in Quebecois, as shown by Daoust-Blais and Kemp (1979), this position to the left of the complementizer can be filled not only by the postverbal negative marker pas, but also by the adverbs jamais and plus (as well as by the quantifiers rien and personne), as shown in 15: (15) a. J'ai cache lesciseaux pour plus que ma fille seblesse. (Quebecois) 'I hid the scissors so that my daughter would no longer hurt herself.' b. J'ai cache les ciseaux pour jamais que pareille chose arrive a ma fille. 'I hid the scissors so that such a thing would never happen to my daughter.' Leaving aside the question of the exact position of these elements in such a construction, let us simply observe that these data point in the same general direction as the data discussed above. That is, we see that the postverbal negative markers can occur in the same positions in which adverbs can occur. Italian non, on the other hand, and—to the best of my knowledge—all of the preverbal negative markers which pattern like it in other Romance languages, cannot occur in such a position, to the left of Comp, as shown in 16a—instead, it has to occur inside the clause, as in 16b:7 (16) a. *Ha fatto cosi per non che si stufasse. has done so for NEG that himself bored 'He did it that way so that he wouldn't get bored.' b.

Ha fatto cosi perche non si stufasse.

A final difference between the Italian and Piedmontese negative markers concerns their interaction with movement. Piedmontese nen does not interfere with clitic movement: Pronominal clitics can move to the left of the finite verb, as shown in 17a. Parallel facts are found in the other languages with postverbal negative markers of the type of Piedmontese nen, as exemplified in 17b: (17) a. A-mlodanen.

(Piedmontese)

CL

subj-CLdat CLobj Sives NEG 'He/she won't give that to me.'

b. Me lo donna pa. (Franco-Provencal) CL dat CLacc 8ives NEG 'He/she won't give that to me.' Italian non, on the other hand, does interfere with clitic movement (as do the negative markers of the type of Italian non in other Romance languages). This is shown

186

PARAMETERS AND FUNCTIONAL HEADS

by cases of long clitic climbing, exemplified in 18. Certain Italian verbs, given an infinitival embedded clause,8 allow the clitic to be attached either to the embedded V, as in 18a, or to the matrix V, as in 18b: (18) a. Devoparlarri. must talk-to-you 'I must talk to you.' b. TIj devo parlare e;. To-you must talk 'I must talk to you.' As has been noted in the literature (in particular, Kayne 1989a and references therein), the presence of the negative marker non in the embedded clause interferes with the possibility of long clitic climbing, decreasing the acceptability of the sentence, as shown in 19: (19) a. Devo non parlarff. must NEC talk-to~you 'I must not talk to you.' b. ??Tij devo non parlare e,. To-you must NEC talk 'I must not talk to you.' On the basis of these differences between Italian non and Piedmontese nen, I suggest that the two negative markers need two distinct syntactic characterizations. In accord with many suggestions in the literature, I will take the fact that Italian non resembles the pronominal clitics in always adjoining to V, and the observation that its presence interacts with long clitic climbing as indications that it is an element of the same type as the pronominal clitics in terms of X-bar Theory, namely a head. On the other hand, because the distribution of Piedmontese nen overlaps with that of adverbs, and because its presence does not interact with clitic movement, I will suggest that Piedmontese nen is not a functional head, but that it is an adverb which occurs in an adjoined position.9 In what follows, I will relate the different properties of these negative markers to the fact that one of them is a functional head which shows selectional restrictions, while the other is a lexical element in an adjoined position.

1.2 Negation and Tense The crucial aspect of this proposal, as far as the interaction of negation and tense is concerned, is the following. The preverbal negative marker, here exemplified by Italian non, is a head; as such, it selects a complement. The complement which the head of the NegP selects is the maximal projection TP; in particular, expressing selectional properties as relationships from one head to another, the preverbal negative marker which is the head of the NegP selects the head of the TP. By contrast, the postverbal negative marker, here exemplified by Piedmontese nen, is not the head of

ON THE RELEVANCE OF TENSE

187

a NegP, but rather an adverbial element in an adjoined position. As such, it is not selected by anything nor does it select any complement of its own. In the next two sections I discuss two sets of data that support this hypothesis: imperatives and past participles in the presence of negation. 1.2.1

Imperatives

If we examine the forms which a language employs for imperatives, we notice that only some of them are morphologically unique to this paradigm; others coincide with verbal forms of the present indicative or subjunctive, or the infinitive. This has recently been discussed in the literature, in particular in Rivero (1994), who refers to the form which is morphologically unique to the imperative as a true imperative and to the other as a surrogate imperative; and by Kayne (1992). Standard Italian has a true imperative only in the second person singular, which does not coincide with any other second singular form. The form employed for the second person plural, on the other hand, is formally indistinguishable from the corresponding form in the indicative:10 (20) a. Telefona! b. Telefonate! 'Call!'

(2ndsg) (2nd pi)

It is also possible to have an imperative in the first person plural, and this also is morphologically indistinguishable from the indicative, as shown in 21. Evidence that this is an imperative is provided by the position of pronominal clitics: While in indicative clauses the clitics precede the verb (22), in imperatives they follow it (23): (21) Telefoniamo subito! 'Let's call right away!' (22) Le telefoniamo tutti i giorni. 'We call her every day.'

(Indicative)

(23) Telefoniamo/e subito! (Imperative) 'Let's call her right away!' Having said that true imperatives are morphologically distinct from all other verbal forms, notice now that the peculiarity of their morphology is that it is extremely reduced: In the cases we examined, it is limited to the vowel a, the so-called thematic vowel, i.e., the vowel which occurs between the root and the inflectional morphology in finite verbs (e.g., parl-a-v-o, where -v- is the tense morpheme and -o the agreement morpheme, for a form of the first singular imperfect indicative). It is an interesting question whether such a vowel has any syntactic function, or whether it is present merely for the purposes of syllabification." I will not address this issue now nor will I speculate on the nature of the vowel; I will simply say that the way imperatives differ from other verbal forms is by lacking (some of) the functional projections that other verbs have. I will suggest that what they lack is the category TP and, in some languages, AgrP.12

188

PARAMETERS AND FUNCTIONAL HEADS

In the presence of negation, the Italian negative marker non cannot occur with the true imperative form, but can occur with the verbal forms which are identical to those of the indicative, as shown in 24: (24) a. *Non telefona!

(2nd sg)

b. Non telefonate! (2nd pi) 'Don't call!' c. Non telefoniamogli subito! 'Let's not call him right away!' The negative counterpart of the second person singular imperative is expressed by means of another suppletive form, the infinitive:13 (25)

Non telefonare! 'Don't call!'

(2nd sg)

Notice that the infinitive, even though not a finite form, does show inflectional morphology: The thematic vowel a is followed by the suffix -re. I suggest that this suffix is the tense morpheme, while the agreement morpheme is missing.14 An analogous pattern is found in Catalan, as illustrated by 26-28: (26) a. Parla!

(2ndsg — true imperative)

(Catalan)

b. Parleu! (2nd pi — same as indicative) (27) a. */Voparla! b. No parleu! (28) Afoparlis!

(2ndsg — subjunctive)

I will propose that the pattern just described with respect to the occurrence of the negative marker should be explained by relating the presence of the negative marker to the presence of a certain type of inflectional morphology. Since the negative marker in Italian can occur with an infinitive, which lacks agreement, my proposal will be that the TP is the functional category whose absence determines the impossibility of the negative marker.15 In this view, then, non cannot occur with a true imperative form because it lacks a TP. On the other hand, it can occur with the suppletive form borrowed from the paradigm of the indicative, or of the infinitive, because these have a TP projection.16 The present account is further supported by the fact that Spanish, which has true imperative forms for both second person singular and second person plural, does not allow the presence of the negative marker no with either form, and expresses the negative imperative by negating the subjunctive—a tensed form: (29) a. Habla! (2nd sg — true imperative)/ *No habla! (Spanish) b. Hablad! (2nd pi — true imperative) / *No hablad! (30) a. Afohables!

(2nd sg —subjunctive)

ON THE RELEVANCE OF TENSE

b. Afohableis!

189

(2nd pi — subjunctive)

On the other hand, Piedmontese—which has a true imperative form—does not have the same constraint on the co-occurrence of the negative marker nen with the imperative: The true imperative formparla in 3la can be negated by the negative marker nen, as in 31b, with no need to employ a suppletive form: (31) a. Parla! (2ndsg—true imperative) (Piedmontese) 'Talk!' b. Parla nenl 'Don't talk!' This is precisely what we would expect if the postverbal negative marker nen is not a functional head which takes TP as its obligatory complement, but rather an adverbial element. Being an adverbial element, nen does not select a complement, and in particular is not affected by the absence of a TP projection. Parallel data are found in other Romance languages which have a postverbal negative marker, as illustrated in 32 and 33, supporting the hypothesis that the incompatibility between negation and imperatives is limited to those languages which have apreverbal negative marker of type X°. (32) a. Geina-tepa. (2ndsg—true imperative) (Franco-Provencal) bother-you NEG 'Don't worry.' b. Fee-me pa rire. Make-me NEG laugh 'Don't make me laugh.' (33) a. Guardammgai tosann! (2ndsg — true imperative) (Milanese) Look NEG the girls 'Don't look at the girls!' b. Venmmgasenzadanee! Come NEG without money 'Don't come without money!' 1.2.2

Past Participles

The dependence of the preverbal negative markers on the category TP is also shown by the case of past participles.17 Take a pair of sentences like 34a and 34b: (34) a. Mary hasn'f always paid taxes, b. Mary has always not paid taxes. We see that negation can be either on the auxiliary, as in 34a, or on the past participle, as in 34b. The propositions expressed by these two sentences have different truth conditions (as discussed in Stockwell, Schachter and Partee 1973). Imagine a

190

PARAMETERS AND FUNCTIONAL HEADS

situation in which Mary has been working for the past 10 years and thus has had ten occasions in which she should have paid taxes. Then 34a is true if she paid taxes on some of those ten occasions but not in all of them, while 34b is false in such a situation. On the other hand, 34b is true only if she failed to pay taxes on all those ten occasions, while 34a would be false in such a situation. (35) a.

Maria non ha sempre pagato le tasse. Maria NEG has always paid taxes

b. *Maria ha sempre non pagato le tasse. Maria has always NEG paid taxes In Italian, while these two propositions can obviously be expressed, they cannot be rendered by expressing negation once on the auxiliary and once on the past participle, as we can see in 35: 35a is good—the negative marker is on the auxiliary—and corresponds to English 34a. But 35b, which would correspond to English 34b with negation on the past participle, is impossible. The proposition which is expressed in English by 34b cannot be expressed in Italian by having the negative marker non precede the past participle. Why should Italian have such a constraint on the occurrence of the negative marker on the past participle? From the perspective of the current proposal, it is because the past participle does not have a TP associated with it (the TP in this type of sentence is presumably higher, where the auxiliary is). Hence non cannot occur to its left, since a Neg° can only be generated in a position where it can take a TP as its complement. Another situation which exhibits the constraint against having a Neg° on a past participle is that of so-called absolute constructions. Absolute constructions are non-matrix clauses which consist of a past participle followed by a lexical NP and, possibly, some other complement of the verb. In Italian, as noted by Belletti (1981, 1992), they are acceptable with unaccusatives and with transitive verbs. Two examples of this construction are given in 36 and 37: (36) Arrivata in ritardo, Maria non ha piu trovato posto a sedere. arrived late, Maria NEG has more found place to sit 'Having arrived late, Mary couldn't find a place to sit.' (37) Passato queU'esame, Maria e poi andata avanti sen/a problemi. passed that exam, Maria is then gone ahead without problems 'Having passed that exam, Mary then went on without any problem.' Belletti (1992) suggests that absolute constructions should be viewed as being AgrPs which lack a temporal specification and have no TP at all. Kayne (1989b), though differing from Belletti in his account of some of the properties of these constructions, also suggests that they should be represented as simple AgrPs.18 Under the present account of sentential negation, non should not be able to occur within these structures, since its obligatory complement, TP, is not present.19 This prediction is correct: As already noted by Belletti, absolute constructions cannot be negated by non:

ON THE RELEVANCE OF TENSE

191

(38) *Non arrivata in tempo, Maria non ha piu trovato posto a sedere. NEG arrived in time, Maria NEC has more found place to sit 'Not having arrived in time, Mary couldn't find a place to sit.' (39) *Non passato quell'esame, Maria ha avuto mold problemi. NEG passed that exam, Maria has had many problems 'Not having passed that exam, Mary then had many problems.' By assuming that non is the head of the NegP and that it takes TP as its obligatory complement, we can account for the impossibility of the presence of non in these constructions by deriving it from the absence of a TP.20 Summarizing up to this point, I have shown that at least two syntactically different types of negative markers must be distinguished in Romance. These two types can be exemplified by the Italian preverbal negative marker non and by the Piedmontese postverbal negative marker nen. The preverbal negative marker is the head of NegP and depends on the presence of TP; I expressed this relation by saying that it takes TP as its obligatory complement. From this it follows that the NegP can occur only in the presence of the functional category TP, so that Neg° can satisfy its selectional requirements.21 The postverbal negative marker, on the other hand, is not part of the NegP but is an adverbial element and its distribution is not affected by the presence or absence of any functional category.

2 Negation and Tense in English The question that I would like to ask now is whether the properties that I have attributed to the head of NegP in Romance hold for the head of NegP in other languages as well. This question amounts to asking whether the distribution of the functional category for negation can be explained in terms of its relation with the functional category for tense in other languages as well, or whether this is simply a peculiarity of Romance. The previous discussion has suggested that the grammar of a language such as Italian does not need to specify where NegP can be generated: All it says is that the head of NegP selects the head of TP as its complement; from this it follows that NegP can be generated in the structure only where a TP projection is licensed. Many Germanic languages lack an (overt) negative marker of type head: German, Dutch and the Scandinavian languages only have the negative marker which corresponds to Piedmontese nen in our previous discussion, i.e., the adverbial type negative marker. Flemish, a dialect of Dutch, has both a negative marker of type X° and an adverbial negative marker, en and niet, respectively. The data provided to me by Liliane Haegeman for West Flemish and Wim de Geest for East Flemish indicate that the distribution of the negative marker of type head, en, is comparable to that of the negative markers of type head in Romance (abstracting from differences related to the V2 nature of Flemish). In particular, en cannot occur on past participles or on imperatives, except for imperatives which employ suppletive forms from the paradigm of the indicative.22 I will not discuss the Flemish data further, because

192

PARAMETERS AND FUNCTIONAL HEADS

I'd like to turn to another Germanic language which has two syntactically distinct negative markers, namely, English. I will suggest, following Kayne (1989c), that in a sentence such as 40 the reduced form of the negative marker, n't, and the full form, not, differ not only in their phonological shape, but also in their syntactic category: (40) She couldn 't not have noticed it. The negative element n't is a functional head (see also Pollock 1989:397) and is a bound morpheme, which needs to be supported (cf. the arguments of Zwicky 1970: 328 and Zwicky and Pullum 1983). I will assume that the verb raises and adjoins to the negative marker. The negative element not, on the other hand, is not part of a functional category, but rather is an adverbial element which can be adjoined to, or perhaps occur in the specifier position of, any maximal projection. This distinction is parallel to the one I drew in the previous section between Italian non on the one hand and Piedmontese nen on the other. In this section I will argue that English n 't, the head of NegP, is subject to the same constraints I described for Italian non, and that these constraints can be adequately accounted for by assuming that in English, as in Romance, the head of NegP selects TP as its complement.23 I will not go into the details of the arguments in favor of viewing n't as a functional head. What I will do, instead, is test whether n't shows the same constraints characterizing the negative heads in Romance and whether its syntactic behavior can be explained along the lines presented in the previous section. If the head of NegP in English selects the head of TP, just as its Romance counterparts do, then we would expect to find that the functional category NegP in English also can only be generated in a position immediately to the left of the functional category TP. This prediction is borne out by the data. N't cannot occur on a past participle, as shown in 41b and 42b: (41) a. We haven'? had lunch yet. b. * We have hadn 't lunch yet. c.

We hadn't had lunch yet.

(42) a. Mary hasn 't always paid taxes, b. *Mary has always paidn't taxes. Example 41b shows that n't cannot occur on had as a past participle, even though it clearly can occur on that lexical item when it is not a past participle, as in 41c. The contrast in 42 is parallel to the case of Romance that we discussed above in 35b, where the negative marker of type head could not be on the past participle. It could be argued that the impossibility of 41b and 42b in English stems from the fact that past participles cannot raise and adjoin to the negative affix. Thus, 41b and 42b would be impossible because the past participle had and paid cannot adjoin to the affix n't, leaving it unbound and thus creating a configuration which is unacceptable because it violates a morphological requirement on bound affixes. But notice that the same restriction we observe with past participles is also manifested in examples such as 43, which involve an infinitival auxiliary:

ON THE RELEVANCE OF TENSE

(43) a.

193

She could [not have noticed it.]

b. *She could [haven't noticed it.] c.

She can't [no; have noticed it]

d. *She can't [haven't noticed it.] One cannot maintain that the ungrammaticality of haven't in 43b and 43d is due to the inability of have to raise and adjoin to the affix n 't. Lobeck (1986:107-119) notes the auxiliary-like behavior of non-finite perfective have and argues for an analysis in which it is base generated in Infl.24 Johnson (1988) departs from Lobeck's basegeneration proposal in favor of a movement analysis, but maintains the claim that perfective have is in Infl at S-structure.25 In addition to the facts pointed out by Lobeck, he describes a non-standard variety of English which allows inversion of both a modal and have around the subject, as in 44, and suggests viewing have as a bound morpheme which is base-generated inside VP and then raises in order to affix onto a head position: (44) a. Shouldn't have Pam remembered her name?

(Johnson 1988)

b. What could have Mary bought? c.

Couldn 't have Sara understood the implications of her actions?

Leaving aside the issue of base-generation in Infl vs. movement to Infl, I think Lobeck and Johnson convincingly show that perfective have occurs in a position higher than VP. Yet note that neither the cases which clearly suggest that have is in Infl in the standard language, nor the examples in the non-standard variety, ever allow the negative head n't to be adjoined to non-finite have. Example 45 is sharply ungrammatical even in the non-standard variety described by Johnson (personal communication): (45) a. * Should haven'/Pam remembered her name? b. *Could haven't Sara understood the implications of her actions? I therefore conclude that the impossibility of n't on an infinitival auxiliary is not due to the fact that have is not high enough in the structure, but rather to the fact that NegP can only be generated to the left of a maximal projection TP. Since neither past participles nor infinitival auxiliaries have a TP projection, they cannot license an instance of NegP. Thus, the ungrammaticality of sentences 41-45 should be given a unitary account: Because n't is always the head of NegP and takes TP as its complement, it cannot occur in these examples, where its selectional requirements are not met due to the absence of a TP projection.

2.1

Imperatives

So far the parallelism between the Romance negative markers of type head and English n't holds. But an apparent difference between the two can be observed in the case of imperatives. I have said that, in Romance, true imperatives cannot be negated

194

PARAMETERS AND FUNCTIONAL HEADS

with a negative marker of type head (for example Italian non), and suggested that this constraint can be explained by viewing imperatives as structures which lack the functional category TP. If we now turn to English imperatives, we might expect them to also lack the option of expressing negation by means of the negative marker of type head, n 't. What we find, though, is that English imperatives do in fact express negation by means of n't, as shown in 46: (46) a. Don't (you) be shy! b. Don't (you) do that! At first sight, these sentences seem to constitute a counterexample to our generalization on the distribution of negative markers of type head. However, I will suggest that the English imperatives exemplified in 46 are not cases of true imperatives, in the sense that they do not employ a verbal form morphologically unique to the imperative (as was the case with Italian parla discussed above). Rather, the verbal forms employed in 46 correspond to the suppletive forms that are found in Romance. My proposal here is that the verbal forms used in English imperatives are forms of the present subjunctive. The idea that imperatives and present subjunctives might be the same is not completely new. In the chapter on the English imperative in Stockwell, Schachter and Partee (1973:634), rules are given that "account directly for plain imperatives and subjunctives (which are here regarded as equivalent to embedded imperatives)".26 Similarly, Beukema and Coopmans (1989) give the imperative and the subjunctive identical characterizations in their system: Both of them are specified as [—Tense; +Agr]. Moreover, all discussions of the English imperative up until the recent article by Beukema and Coopmans treat a set of sentences of the type exemplified in 47,48 and 49 as the prototypical cases of English imperative: (47) a. Go there! (Stockwell, Schachter and Partee 1973) b. Don't go there! (48) a. You go there! b. Don't you go there! (49) a. Somebody go there! b. Don't anybody go there! Some studies also notice that English imperatives can not only have a second person subject, as in 47 and 48, and a quantificational subject, as in 49, but also a definite NP subject, as in 50:27 (50) a. Don't that boy over there dare move! b. Don't the rest of you say anything!

(Beukema and Coopmans 1989)

ON THE RELEVANCE OF TENSE

195

I think it is correct to say that examples such as 49 and 50 are examples of English imperative, since their syntactic behavior is indistinguishable from that of examples such as 47 and 48. But notice that in all the other Germanic languages, as well as in all of Romance, the equivalents of 49 are never expressed by means of the verbal form unique to the imperative. Rather, they are expressed by means of constructions with the illocutionary force of an imperative, but which employ a verbal form from the paradigm of the subjunctive, when available, or of the infinitive (German), as illustrated in 51: (51) a. Que personne ne bouge! (French—subjunctive) 'Nobody move!' b. Alle aufstehen! (German—infinitive) 'Everybody stand up!' Thus, it seems that we are faced with three choices in dealing with what have been considered the core cases of imperative in English: a. Either we say that all the examples in 47-50 employ imperative forms, and we set English aside from all other Germanic and Romance languages, in attributing to it a special form of the imperative for the third person; or b. We decide to set aside sentences such as 49 and 50 from the other examples of imperative, even though we do not find any reason to do so based on the observation of their syntactic properties; or, c. We treat all the examples in 47-50 in a unitary fashion and say that they are imperative in their illocutionary force but that the form they employ is a suppletive form borrowed from another paradigm. The first and second strategies require that we maintain a distinction among verb forms which has no morphological or syntactic consequences. I will therefore reject them and choose instead the third strategy which argues that English imperatives are not true imperatives in the sense specified above, i.e., forms morphologically unique to this verbal paradigm. In a more speculative vein, I will propose to view the English imperative as an instance of present subjunctive in a matrix context.28

2.2 Imperatives as Subjunctives Let us first observe a similarity between subjunctives and imperatives in English. Both in the case of embedded present subjunctives, as in 52, and in the case of imperatives, as in 53, the verb appears in its bare form, without any inflectional morphology:29 (52) a. I insist that she stay. b. I demand that everybody be admitted. (53) a. Be good! b. Don't anybody move!

196

PARAMETERS AND FUNCTIONAL HEADS

In the case of lexical verbs, no affix lowering takes place, and in the case of have and be, no raising of V into a projection in which they would pick up inflectional morphology. Given current assumptions on the position of adverbs and floating quantifiers, it seems that have and be may raise in embedded subjunctives, yielding slightly marginal results: (54) a. ?I insist that he have already left by the time I come home. b. ?The director requires that they have all learned it by heart before Sunday. Since it does not affect the morphological shape of the verb, I will assume it is raising to an abstract functional category FP.30 The lack of inflectional morphology is perhaps the most striking similarity between subjunctives and imperatives. Let us now turn to considering some differences. One difference is that subjects must be lexically realized in the case of embedded subjunctives, while they can be omitted in the case of imperatives, if they are second person pronouns. Although this is an interesting discrepancy between the two constructions, it does not prove that imperatives cannot be subjunctive forms. We know that in French, a language which does not allow null subjects, forms which are unmistakably subjunctive can be used with imperative force and, in that case only, can lack a subject, as in 55: (55) a. Soyonsprudents! Let's be cautious! b. Ayez confiance! Be confident! Therefore, a subjunctive form used with imperative function can lack the subject even in languages which are not null-subject languages. Yet, the case of French clearly shows that such forms need not be, morphologically, imperative forms, but can be forms of the present subjunctive. Let me therefore set aside this discrepancy between embedded subjunctives and imperatives, since it does not bear on the present discussion, and concentrate on two other differences between these constructions: a. Embedded subjunctives occur in the presence of a complementizer, that, while imperatives are matrix sentences which lack an overt complementizer, as we can see by comparing 52 and 53. b. In embedded subjunctives, negation is expressed by the negative marker not, without the option of do-support, and it occurs between the subject and the bare form of the verb, as in 56. In imperatives, on the other hand, as we can see in (57), negation is expressed with the use of cfo-support, obligatorily, and it occurs to the left of the subject, when one is present: (56) a. I insist that she not stay, b. *I insist that she don't stay. (57) a. Don't you say that!

ON THE RELEVANCE OF TENSE

197

b. * You don 7 say that! c. *You not say that! d. *Not you say that! I believe that these two differences, i.e., presence or absence of a complementizer and the strategy employed for negation, are related and should be reduced to the well-known difference between matrix and embedded contexts. 2.2.1

Subjunctives

It is always the case in English that main clauses lack an overt complementizer, while complement clauses have one. Moreover, we know that the content of Comp is determined on the one hand by the finiteness of the verb in the complement clause, and on the other by the selectional properties of the matrix verb. For example, the verb wonder selects an interrogative clause, and such selection is manifested by the obligatory presence of a w/z-constituent in Comp: either a moved w/z-phrase, or the lexical item whether. My proposal here is that, as there is a class of verbs which select a w/z-clause as a complement, there is also a class of verbs in English which select a subjunctive clause as their complement (verbs such as require, demand, insist), and that in both cases selection of the clausal complement is manifested through the obligatory presence of certain features in C°.31 Just as embedded w/z-clauses are said to have w/z-features in C, embedded subjunctive clauses must contain in C the features selected by the matrix verb. I will argue that the complementizer that which introduces subjunctive clauses is different from the homophonous complementizer which introduces indicative clauses: It carries the features of modality selected by the matrix verb, as well as the tense features, which are transmitted from the matrix to the embedded clause. The idea that, in certain types of embedded clauses, the complement clause should play a role in the expression of tense is not new.32 In addition to adopting this idea, I suggest that the complementizer in subjunctives expresses subjunctive modality, i.e., it is able to play a role normally reserved to lexical elements (e.g., modal verbs) or functional categories.33 This line of reasoning leads me to suggest that the role of the modal is here played by the lexical item that and, consequently, that there is no modal element occupying the Infl position in embedded subjunctives.34 In this view, then, the complement clause of an embedded subjunctive has an impoverished structure, which lacks certain functional projections normally present in other types of finite clauses, i.e., the functional projections containing tense and modal elements, and contains an abstract functional category PP.35 Now we can examine how the properties that we have seen to characterize subjunctive clauses follow: First, the lack of inflectional morphology on the verbs; secondly, the behavior of negation. Lexical verbs in subjunctive clauses do not show any inflectional morphology, as illustrated in 52, because the functional categories from which affix lowering takes place are absent. Moreover, neither main verb be nor auxiliary be and have can raise and pick up inflectional morphology; when raising takes place (as in 54), it is to a projection that does not contain overt inflectional morphology (perhaps the abstract functional category suggested in n. 30). The

198

PARAMETERS AND FUNCTIONAL HEADS

negative morpheme n 't cannot occur in this structure because the functional category TP, which is necessary for the presence of NegP, is missing. The question then arises why negation can be expressed via the negative marker not, as in 56: If not is the head of NegP (cf. example (i), n. 23), its occurrence should obey the same constraints as the occurrence of n't?6 If not is the adverbial element, its presence is surprising in a different way: Normally, in finite clauses, adverbial not cannot occur at the same time as affix lowering: (58) a. b. c.

She has always not agreed with you. She DID always not agree with you. *She always not agreed with you.

As we can see in these examples, not can occur when an auxiliary has raised (58a) and in the presence of emphatic do (58b) ; but it cannot occur when affix lowering has taken place (58c).37 Why do subjunctives differ from indicatives in this respect? Under the present proposal, this follows straightforwardly from the fact that subjunctives have an impoverished structure and lack the functional projections from which affix lowering takes place. Hence, whatever the nature of the problem illustrated by 58, it will arise in the case of indicatives but not in the case of subjunctives. 2.2.2

Imperatives

Having described the structure of embedded subjunctives, let us now turn to imperatives. In the present proposal, imperatives are instances of present subjunctive in root contexts. They lack the complementizer that, because they are matrix clauses, but modality and tense are expressed in C° all the same. Let us conjecture that the CP of imperatives is similar to the CP of subjunctives, but with one crucial difference: What in subjunctives is expressed via features on the lexical item that, in imperatives is expressed via features on the abstract functional head C°. Now we can examine how the two properties characterizing imperatives are derived: the lack of inflectional morphology on the verb and the behavior of negation. In imperatives, similarly to embedded subjunctives, verbs do not show any inflectional morphology because Infl only contains an abstract functional category, devoid of lexical content; therefore affix lowering does not take place and verb raising, to the extent to which it occurs, does not result in the acquisition of inflectional endings. Whereas in embedded subjunctives NegP cannot occur because the functional category for tense is not present, in matrix subjunctives, i.e., imperatives, NegP can occur where the functional category carrying the tense features is licensed. I suggest that NegP is generated to the left of CP, where the selectional requirements of Neg° are satisfied by C° containing the features for tense (59).38 The difference between embedded subjunctives and imperatives should now be clear. In the former, the tense features are on the lexical item that, hence NegP cannot be generated at all.39 In the latter, on the contrary, the features for tense are realized on the functional head C°; hence NegP can be generated to the left of CP, and do-support can take place, inserting do in C°.40 Do will then adjoin to n't, yielding the sequence don't; in this way, n't, a bound morpheme, will have the required morphological support. 41

ON THE RELEVANCE OF TENSE

199

The analysis advocated here argues that in both embedded subjunctives and imperatives tense and modality are expressed in C° and that no movement from I to C is involved. If I-to-C movement were involved in the case of imperatives, we would expect them to show some of the properties of interrogative clauses, where the verb is assumed to raise from (V to) I to C. Interrogatives have the option of raising do and leaving the negative marker behind, as shown in 60: (60)

Do you not read the newspaper?

But, as shown in 61, raising do and leaving not behind is not an option for negative imperatives, not even marginally: (61) *Do you not read the newspaper! This would be surprising if we assumed that the mechanism generating negative imperatives were the same as that generating negative questions. If, on the contrary, we assume that imperatives differ from other constructions due to the special nature of their C°, then no such parallelism is expected.42 Before concluding, let us now examine briefly how an analysis of subjunctives which advocates the presence of an empty modal in Infl would handle the same set of facts. The absence of verbal morphology would be considered parallel to the absence of inflectional morphology in the presence of an overt modal (e.g., 'She must be happy' )• The impossibility of the negative marker n 't would be related to the fact that the empty modal cannot provide the lexical support required by a bound morpheme; the presence of not, on the contrary, would be expected, since it is not affected by the lexical content of the element in Infl. Nothing would be said about the nature of the complementizers in embedded subjunctives, since under this view subjunctives differ from other finite clauses only in containing a non-overt modal. But if we extend this approach to subjunctives to the case of imperatives, then the question arises of why do-support in the presence of negation should be available in the latter but not in the former. If do cannot occur in embedded subjunctives (e.g., *'I require that she don't move') because incompatible with the presence of an empty modal, why should it be possible—and in fact, required — in imperatives?43 One could say that the empty modal is different in the two cases: It is such that it blocks insertion of do

200

PARAMETERS AND FUNCTIONAL HEADS

in subjunctives but not in imperatives. But it seems difficult to make precise what exactly this difference could be. Hence I have chosen to say that subjunctives have indeed an abstract functional category, as suggested by the raising of perfective have, but that this category does not correspond to an empty modal; I have suggested in note 30 that it might be an abstract agreement projection. Such an abstract functional category is also present in imperatives and is compatible with the occurrence of do,44

3

Conclusions

In conclusion, in this paper I have discussed the syntactic relevance of tense for sentential negation in Romance and English, showing that certain negative elements can only occur in the presence of tense. I have expressed this dependence by analyzing these negative elements as the head of the functional category NegP and arguing that NegP depends on the presence of a functional category expressing tense. I have chosen to express this relation in terms of selectional restrictions, by saying that the head of NegP selects TP as its complement, though a precise characterization of this relation awaits further study. This allowed me to account for the constraints on the distribution of the negative markers of type head in these two language groups. This approach also suggests that the grammar does not need any other principle to constrain the structural position of NegP: NegP can be generated where there is a functional category which can satisfy the selectional requirements of its head.

Notes 1. This paper was written in 1990 and represents my thoughts on this topic as of that time. Since then 1 have developed this work further. See Zanuttini (1991,1995) for details. This paper was first presented at the conference "Time in Language" held at MIT in March 1990; revised versions were then presented at the University of Geneva, University of Venice, Groupe de Syntaxe Comparee (Paris) and University of Delaware. I am indebted to those audiences for many helpful comments. I would especially like to thank the linguists at the University of Geneva; in particular, R. Clark and I. Roberts. I am also grateful to T. Guasti, R. Kayne, A. Kroch, M. Moser, C. Poletto, B. Santorini and A. Tomaselli for their valuable help and support. 2. The choice of this notion is not to be seen as a claim that functional categories select one another in exactly the same way as lexical categories (an issue which needs further investigation). At this point, it is simply a choice of convenience, one possible way of expressing the observed relation. 3. As will become clear in the course of the discussion, this does not amount to claiming that all negative markers need to co-occur with tense, since, in the present proposal, not all negative markers are the head of NegP. Those which are not can be analyzed either as specifiers of NegP or as adverbial elements adjoined to a maximal projection. I will choose the latter strategy in this work, though the former also deserves serious consideration. 4. This claim has two independent parts. One is that English imperatives do not correspond to what will be called 'true imperatives' in Romance, that is, forms which are morphologically unique to the imperative. The other is that they are instances of present subjunctive. The second proposal is rather speculative at this point, but its validity has no bearing on that of the first.

ON THE RELEVANCE OF TENSE

201

5. The content of this section is taken from Zanuttini (1990). 6. The languages that have postverbal negative markers are spoken in geographically contiguous areas which cover the eastern part of southern France, the western and central part of northern Italy, and those parts of Switzerland where varieties of Franco-Provengal and Raetho-Romance are spoken. 7. Even though earlier I suggested that French ne patterns with the preverbal negative markers of Italian, Catalan and Spanish, there are in fact several points on which it differs from them. One of them we just observed: While the negative markers in these languages cannot occur to the left of Comp (as shown in 16), French ne can, as shown in 14a. At present I cannot provide an account for these differences. 8. In Romance, contrary to English, infinitivals following modals are taken to be control constructions. 9. A precise proposal on the structural position of Piedmontese nen is found in Kayne (1990: 261). 10.1 chose as examples verbs of the first conjugation, in -are, where it is clear that the second singular imperative does not correspond to any other second person form. The distinction is not formally obvious in the case of verbs of the second and third conjugation, but it will be assumed to hold for them as well. 11. That is, if we wanted to extend the proposal of Harris (1991) concerning the 'word marker' present in Romance nouns. 12. The question here is whether the morpheme present in the second person plural in Spanish imperatives should be considered an agreement morpheme. 13. There is another type of imperative for the second singular polite form, which employs a third person pronoun (optionally) and a verbal form from the subjunctive; the presence of the negative morpheme is not incompatible with this, as shown in (ii): (i) Dica tutto cio che pensa! Say (3rd sg subjunctive) all that you think 'Feel free to say everything you think.' (ii) Non dica tutto cio che pensa. But these forms do not show the order Verb-clitic which is characteristic of imperatives; hence, I will consider them forms of the subjunctive used with imperative force and will not discuss them any further. 14. Recent work (Kayne 1990:259 and Guasti 1991) has suggested characterizing the morpheme -re as the head of an InfP (Infinitival Phrase); this is not incompatible with the present proposal that it is the functional element expressing tense. 15. The question arises whether infinitives in Italian could have an abstract agreement, which could be the category licensing negation in infinitives. This is a possibility which I will not investigate at this moment. 16. The question then is why there should be a special relation (or any relation at all) between negation and tense. One answer that immediately comes to mind is that both sentential negation and tense are operators which take scope over the same domain; thus their association in the syntax could be seen as a reflection of their similar semantic properties. Although intuitively appealing, this line of reasoning is weakened by the fact that other negative markers, which also take scope over the same domain as tense, do not show any syntactic dependence on it (e.g., the negative marker of Piedmontese discussed in this paper). Another possible answer is to relate the property of depending on tense to another property of this class of negative markers, namely their being weak phonologically (although they cannot be assimilated to pronominal clitics, they do typically show a reduced vowel). Then this class of negative markers could be described as depending on tense at all

202

PARAMETERS AND FUNCTIONAL HEADS

levels, i.e., phonological, syntactic and semantic (in the sense that their scopal properties can never be distinguished from those of tense). 17. Pollock (1989:413) also relates the impossibility of French ne with past participles to the absence of Tense. He notes that the contrast between infinitives and past participles with respect to the presence of sentential negation was first pointed out by Kayne (1975:199): (i) Jen'allais pas ne pas venirchez toi. (Kayne's ex. 386) 'I wasn't going not to come to your house.' (ii) *Je n'ai pas ne pas dit cela. 'I didn't not say that.' 18. See also Vinet (1989:177, and her fn. 7). 19. An alternative explanation is proposed by Vinet (1989:179). She notes the impossibility of negation in parallel constructions in French (her examples: * Unefois Marie pas prete, ... , *Aussitot les ballons pas creves, . . . ) and relates it to a semantic incompatibility between negation and the aspectual properties of these constructions. 20. The case of past participles differs from that of imperatives in two main respects. First, while the constraint on the presence of a preverbal negative marker with true imperatives is extremely clear (in the sense that the grammaticality judgments it generates are solid and consistent), in the case of past participles different degrees of acceptability arise with different classes of verbs. To a first approximation, the presence of the negative marker seems to be completely ruled out with intransitives, extremely marginal with unaccusatives, and marginally acceptable (at least in some cases) with transitive verbs. Secondly, in the case of imperatives the preverbal negative markers here described as Neg" contrast sharply with the postverbal ones, the adverbial elements, which can indeed negate a true imperative; in the case of past participles and absolute constructions, on the other hand, the contrast between the two types of negative markers is not nearly as sharp. I will leave open the issue of what exactly the source of these differences is. 21. A similar proposal, relating the presence of negation and tense, is found in Pearce (1990), who suggests that certain negative markers did not occur in Old French due to the absence of the functional projection TP, but appeared in Middle French when a Tense projection was developed. 22. One interesting difference between Flemish en and the Romance negative markers discussed in the previous section is that en cannot occur with infinitives (see Haegeman and Zanuttini, 1995). This property sets it aside from negative markers of the type of Italian non, Catalan and Spanish no, and Romanian nu, but not from all preverbal negative markers in Romance. Some of the languages which have both a preverbal and a postverbal negative marker seem to show the same constraint; for example, the variety of Piedmontese spoken in Cairo Montenotte (described in Parry 1985) does not allow the presence of the preverbal negative n before infinitives. Though it is tempting to relate this property to that of having a discontinous negative constituent, notice that such a correlation does not hold for standard French, which has a discontinous negative constituent and still allows the preverbal negative marker to co-occur with an infinitive. 23. Following Pollock (1989) and Chomsky (1989), I take the not which has sentential scope (as in i) to have the same syntactic properties as n't: (i) He does not like linguistics. For ease of exposition, 1 will refer to the negative clement in question as n't and will use not only to refer to the adverbial element exemplified in 40. 24. First, in contrast to main verb have, perfective have can occur either to the left or to the right of preverbal adverbs (the numbering of examples is Lobeck's):

ON THE RELEVANCE OF TENSE (27) a. b. (28) a. b.

203

Louise must have always liked Bizet, Louise must always have liked Bizet. You should always have coffee in the morning, *You should have always coffee in the morning.

Secondly, it must remain outside an ellipted VP: (29) Mary could have worked hard and Bill a. might have too. b. *might too. Thirdly, it can contract with a modal, a possibility which, according to Selkirk (1984: section 7.2.2.3), is given only when both the modal and have are dominated by the same non-phrasal node: (40) (42)

John [ In fl could've] watched Miami Vice. *John [Infl could] [VP *ve] a Miami Vice T-shirt.

25. Proposing that have is in Infl amounts to saying, in our terms, that it has raised to at least the lowest of the functional projections which constitute Infl. 26. Jacobs (1981) and Huntley (1984) argue that subjunctive complements to volitional predicates are embedded imperatives, an idea extended to Romance in Kempchinsky (1984: 98). 27. Interestingly, only with the negative form of the imperative. 28. The line of reasoning I have adopted here leads me to argue that negative imperatives and imperatives with a third person subject are always instances of suppletive imperatives; but it does not necessarily lead me to conclude that the second person affirmative imperative is also a suppletive form. It is conceivable that, similar to the case of Italian, in English some forms are also true imperatives while others are suppletive forms. I will leave this issue open. 29.1 would like to emphasize that the parallelism I am describing holds between imperatives and present, not past subjunctives. Past subjunctives do not correspond to the bare form of the verb and show verb movement to C: (i) a. Had I done that earlier,... b. Were I able to speak Chinese,... 30. The idea of an abstract functional category is reminiscent of Roberts' (1985:40-41) suggestion that subjunctives have an empty modal in Infl. What is crucial for the present account is that such a projection does not correspond to an abstract TP. I could take it to be an abstract modal phrase and subscribe to the view that modals in English are not generated under TP, contrary to what is often believed (a recent proposal along these lines is provided by Kayne 1989c, section VII). Alternatively, I could take it to be an abstract agreement phrase, void of lexical content. At present I do not have empirical reasons that can help me choose between these two alternatives. In the remainder of the paper I will adopt the latter solution in order to more easily pursue the idea of the parallelism between embedded subjunctives and imperatives. The reasons that motivate this choice will be briefly discussed in the last paragraph of section 2.2.2. I will assume that such an abstract agreement projection provides a position for the subject of the embedded subjunctive clause. 31. Embedded interrogatives are assumed to have vWi-features in C°, even though in English w/z-elements always appear in [Spec.CP]. See Rizzi (1990) for a proposal relating these two properties. 32. Similar, though not identical proposals are found in Stowell (1981, 1982), Den Bcsten (1983), En? (1987), Raposo (1987), among others.

204

PARAMETERS AND FUNCTIONAL HEADS

33. In a similar vein, Kempchinsky (1986:102-106) argues for the presence of a subjunctive operator in Comp. 34. A similar proposal is found in Emonds (1976:198, n. 17), where it is suggested that subjunctives lack TENSE and the modal auxiliaries and that their complementizer contains the feature SUBJUNCTIVE at deep structure. The proposal that the complementizer of subjunctives is different from that of indicative complement clauses is supported by the fact that speakers seem to delete the complementizer less easily in the former than in the latter case. Robin Clark (personal communication) also pointed out to me that in the case of subject extraction, whereas in indicative clauses the presence of the complementizer yields ungrammaticality, in subjunctive clauses it actually improves the status of the sentence, as shown in the following examples: (i) a. *Who did you insist see John? b. ??Who did you insist that see John? 35. Cf. n. 30. 36. Notice that not in subjunctives can take scope wider than the subject and is not restricted to the VP:

(i) She requested that everybody not leave at once. 37. Unless it is an instance of metalinguistic negation, in which case it occurs within the not ... but template: (i) She always not only liked, but loved you. The facts in 58 strongly suggest that the not which, on the basis of its position with respect to the verb, appears at first sight to be a VP-adjoined adverbial element, is in fact an element of type X° intervening between the inflectional heads and the V. If we assume that not is a head of this type, then we can explain why it blocks affix lowering along the lines of Pollock's (1989) or Chomsky (1989) proposals; or we could adopt an explanation along the lines of Laka's (1990) Tense C-command Condition, which argues that the functional head corresponding to Tense must c-command all other heads in the clause which are prepositional operators. 38. Another way to express this idea would be to say that CP in imperatives is in fact a complex category, CP/TP. 39. If the only constraint on the occurrence of negation is that it is dependent on the presence of tense, then we might expect it to appear associated with the constituent through which tense is expressed: CP in the case of subjunctives under the approach advocated here. What is important to keep in mind is that the relation between negation and tense that I have been describing refers to the functional categories NegP and TP and does not necessarily apply to cases when either negation or tense are expressed by a lexical element which does not belong to these functional categories (e.g., a quantifier, an adverb, or a complementizer, as in this case). Negation could be expressed in C in subjunctives only if English had a negative complementizer corresponding to that. Notice that such was the strategy employed by Latin, in exactly these contexts, where the complementizer ut alternated with ne when the complement was negative: (i) a. Precor ut me vidas. 'I beg that you visit me.' b. Quaeso ne me deseras. 'I insist that you not leave me'. (i) c. . . . milites implorabant, ne se in servitute Romanis traderent. . . . soldiers begged (3pl), ne themselves in slavery to the Romans give 'They begged the soldiers not to hand themselves as slaves to the Romans.'

ON THE RELEVANCE OF TENSE

205

Classical Greek also employed the same strategy. English has the archaic negative complementizer lest and still has the negative complementizer unless, but only for unselected contexts; there is no negative complementizer for clauses which are arguments. 40. This aspect of the proposal is reminiscent of Lasnik (1981:167-168): In order to explain the possibility of do-support in imperatives, he argues for the existence of a null counterpart of Tense, under which do can be inserted. 41. Pollock (1989:402) argues that the do which is found in imperatives is a main verb, a "living fossil" closely related to Old and Middle English causative do. His reasoning is based on the fact that do cannot normally co-occur with be, since it would be semantically empty (400). In the account advocated here, the exceptional co-occurrence of do and be is attributed to the impossibility of be to raise to C; this in turn might be related to the defective nature of Infl. At present I am not in a position to provide a precise account of these two properties of imperatives. 42. The distribution of the uncontracted form not in imperatives constitutes a puzzle for which no complete solution has yet been provided, to the best of my knowledge. To state it simply, not can occur in a negative imperative only if the subject is not overt: (i) a. Do not move! b. *Do you not move! c. *Do not you move! d. *You do not move! e. *Not you move! f. * You not move! 43. The analysis proposed in this paper does not run into the same difficulty since the structure of the clause is different in the two cases: With embedded subjunctives, there is no inflectional category where do could be inserted, whereas in imperatives, do can be inserted under C°. 44. Suggesting the presence of an abstract agreement projection in Infl is consistent with the fact that, when do is inserted, it never shows agreement morphology: (i) 'Don't/*Doesn't that boy over there move! The absence of TP in Infl might also be related to the absence of V-to-C movement.

References Belletti, A. 1981. "Frasi Ridotte Assolute," Rivista di Grammatica Genemtiva 6:3-32. . 1992. "Agreement and Case in Past Participial Clauses in Italian," in T. Stowell and W. Wehrli (eds.), Syntax and the Lexicon. New York: Academic Press. 21-44. 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.] Beukema, F. and P. Coopmans. 1989. "A Government-Binding Perspective on the Imperative in English," Journal of Linguistics 25:417^4-36. Chomsky, N. 1989. "Some Notes on the Economy of Derivations and Representations," MIT Working Papers in Linguistics 10:43-74. Daoust-Blais, D. and W. Kemp. 1979. "Pour pas tout que ca se perde: pour as a 'quantifier raising' subordinator in Quebec French," ms. Emonds, J. 1976. A Transformational Approach to English Syntax. New York: Academic Press. Enc, Murvet. !987. "Anchoring Conditions for Tense," Linguistic Inquiry 18:633-657.

206

PARAMETERS AND FUNCTIONAL HEADS

Guasti, M.T. 1991 '"Thefaire-par construction in Romance and in Germanic," Proceedings of WCCFL 9:205-218. Haegeman, L. and R. Zanuttini. 1995. "Negative Concord in West Flemish," in L. Rizzi and A. Belletti (eds.), Parameters and Functional Heads. New York: Oxford University Press. 117-179, [This volume.] Harris, J. 1991. "The Exponence of Gender in Spanish," Linguistic Inquiry 22:27-62. Horn, L. 1989. A Natural History of Negation. Chicago: University of Chicago Press. Huntley, M. 1984. "The Semantics of English Imperatives," Linguistics and Philosophy 7:103-134. Jacobs, R. 1981. "On Being Hypothetical," in R. Hendrick, C. Masek and M. Miller (eds.), Papers from the 17th Regional Meeting, Chicago Linguistic Society. Chicago: University of Chicago Press. 99-107 Johnson, K. 1988. "Verb Raising and 'Have'," in McGill Working Papers in Linguistics. Special. Issue on Comparative. Germanic Syntax, Montreal. 156-167. Kayne, R.S. 1975. French Syntax. Cambridge, Mass.: MIT Press. . 1989a. "Null Subjects and Clitic Climbing," in O. Jaeggli and K. Safir (eds.), The Null Subject Parameter. Dordrecht: Kluwer. 239-269. . 1989b. "Facets of Romance Past Participle Agreement," in P. Beninca (ed.), Dialect Variation and the Theory of'Grammar. Dordrecht: Foris. 85-103. . 1989c. "Notes on English Agreement," CIEFL Bulletin \ :40-67. . 1990. "Romance Clitics and PRO," Proceedings ofNELS 20. 2:255-302. . 1992. "Italian Negative Imperatives and Clitic Climing," in L. Tasmowsky and A. ZribiHertz (eds.), Hommages a Nicolas Ruwet. Ghent: Communication and Cognition. 300-312. Kempchinsky, P. 1986. "Romance Subjunctive Clauses and Logical Form," Ph.D., UCLA. Kitagawa, Y. 1986. "Subjects in Japanese and English," Ph.D., University of Massachusetts, Amherst. Laka, I. 1990. "Negation in Syntax: On the Nature of Functional Categories and Projections," Ph.D., MIT. Lasnik, H. 1981. "Restricting the Theory of Transformations: A Case Study," in N. Hornstein and D. Lightfoot (eds.), Explanations in Linguistics: The Logical Problem of Language Acquisition. London: Longman. 152-173. Lobeck, A. 1986. "Syntactic Constraints on VP Ellipsis," Ph.D., University of Washington. Moro, A. 1987. "Tempo e predicazione nella sintassi delle frasi copulari," Tesi di Laurea, Universita di Pavia. Parry, M. 1985. "The Dialect of Cairo Montenotte, Ph.D., University of Wales. Pearce, E. 1990. "An Analysis of Negated Infinitives in Middle French," Wellington Working Papers in Linguistics 2:31-45. Pollock, J.-Y. 1989. "Verb Movement, Universal Grammar, and the Structure of IP," Linguistic Inquiry 20:365-424. Raposo, E. 1987. "Case Theory and Infl-to-Comp: The Inflected Infinitive in European Portuguese," Linguistic Inquiry 18:85-110. Rivero, M.-L. 1994. "Clause Structure and V-Movement in the Languages of the Balkans," Natural Language and Linguistic Theory 12:63-120. Rizzi, L. 1990. "Speculations on Verb Second," in J. Mascaro and M. Nespor (eds.), Grammar in Progress. GLOW Essays for Henk van Riemsdijk. Dordrecht: Foris. 375-386. Roberts, I. 1985. "Agreement Parameters and the Development of English Modal Auxiliaries," Natural Language and Linguistic Theory 3:21-58. Shlonsky, U. 1989. "The Hierarchical Representation of Subject Verb Agreement," ms, University of Haifa.

ON THE RELEVANCE OF TENSE

207

Stowell, T. 1981. "Origins of Phrase Structure," Ph.D., MIT. . 1982. "The Tense of Infinitives," Linguistic Inquiry 13:561-570. Stockwell, R.P., P. Schachter and B.H. Partee. 1973. The Major Syntactic Structures of English. New York: Holt, Rinehart and Winston. Vinet, M.-T. 1989. "Des petites propositions a valeur aspectuelle," Revue canadienne de linguistique 34:171-192. Zanuttini, R. 1990. "Two Types of Negative Markers," Proceedings ofNELS 20. 2:517-530. . 1991. "Syntactic Properties of Sentential Negation: A Comparative Study of Romance Languages." Ph.D., University of Pennsylvania. . 1995. Syntactic Properties of Sentential Negation: A Comparative Study of Romance Languages. Oxford: Oxford University Press. [In preparation.] Zwicky, A. 1970. "Auxiliary Reduction in English," Linguistic Inquiry 1:323-336. Zwicky, A. and O.K. Pullum. 1983. "Cliticization vs. Inflection: English n't," Language 59:502-513.

This page intentionally left blank

6

A Cross-Linguistic Study of Romance and Arberesh Causatives Maria Teresa Guasti

Languages vary in the way of expressing causative constructions. There are languages that have morphological causatives. These are obtained by combining a verb root that expresses some event with a bound morpheme—the causative affix—to form a morphologically complex verb (cf. Baker 1988; Comrie 1985, among others). Other languages have analytical causatives. Unlike the preceding case, the causative verb and the verb expressing the caused event are morphologically distinct. Despite this morphological difference, one finds languages in which causative constructions are syntactically alike, for example, as far as the Case pattern is concerned. In this paper,1 I propose a unified account of morphological and analytical causatives which develops Baker's (1988) fundamental insight that morphological causatives are derived by Verb Incorporation, i.e., syntactic head-to-head movement. The occurrence of the process of syntactic incorporation is meant to account for the similarities between the two types of causatives. The morphological difference displayed by the two classes of causatives is dealt with in terms of different lexical properties of the causative verbs. Combining this hypothesis with a refinement of Rizzi's (1990) Relativized Minimality Condition, I will argue that the formation of analytical causatives—but not that of morphological causatives — requires the additional process of excorporation of the causative verb from the previously formed complex verb. Thus, two morphologically independent verbs are present in the former case, but not in the latter. I will illustrate my proposal by analyzing causatives in two groups of languages: Romance and Arberesh (Albanian dialects spoken in southern Italy). Romance languages, specifically Italian and French, have analytical causatives. The causative verb takes a complement containing an infinitive verb. The two dialects of Arberesh that I study, the one spoken in Barile and the one spoken in San Nicola dell'Alto, have, respectively, morphological and analytical causatives (cf. Brandi and Savoia 1990; Savoia 1989; Turano 1989). Unlike Romance, the causative complement in San Nicola contains a subjunctive verb. Leaving aside this property, causatives in this language are very much similar to causatives in Italian and French. 209

210

PARAMETERS AND FUNCTIONAL HEADS

Beyond being interesting in itself, the study of causatives in the four languages mentioned offers the possibility of viewing some asymmetries between Italian and French causatives in a new light. I will show that French causatives share not only some properties with Italian causatives, but surprisingly, with causatives found in San Nicola. Based on this fact, I will suggest that French causative complements can enter two different structures, the one found in Italian and the one found in San Nicola. This structural ambiguity is not exhibited by Italian, Barile or San Nicola causative complements.

1

Morphologica! and Analytical Causatives

According to Baker (1988), in languages that have morphological causatives, the causative affix is syntactically like a verb heading the VP that is associated with the matrix clause. The basic verb root is generated under a different verbal head, the one present in the causative complement. This verb root is raised by head-to-head movement and is incorporated into the causative morpheme, thus forming a morphologically complex verb. A major advantage of this view is that it reduces certain constraints on morphological operations to independently motivated principles, notably the Empty Category Principle (ECP), which states that a trace must be properly governed. The derivation of morphological causatives can be illustrated from an Arberesh dialect spoken in southern Italy, specifically from the variant spoken in Barile (Potenza). Consider the example in 1: (1) 'Diymbret 'peta'tsum'bun 'cenen.2

(Savoia 1989:408, ex. 136)

the children CAUS-jump-PERF-3PL dog-ACC The children made the dog jump.' The causative morpheme 'peta (or its allomorphs beta, 'paid) heads the matrix VP. The basic verb root, tsum'b- (jump), is generated in the verbal head of the causative complement. At S-structure, the two verbal elements are joined together, via head-to-head movement, thus forming a single complex verb which is subsequently amalgamated with the inflectional morphemes of tense and agreement (cf. Brandi and Savoia 1990). The structure of 1 is given in 2: Baker assumes that the node indicated as XP is a CP; later, I will revise this assumption. Baker argues that a number of properties manifested by causative constructions are a side effect of incorporation. The morphologically complex verb yielded by incorporation becomes the governor of the D-structure arguments included in the maximal projection of the incorporee, as stated by the Government Transparency Corollary (GTC): Government Transparency Corollary: A lexical category which has an item incorporated into it governs everything which the incorporated item governed in its original structural position. (Baker 1988:64)

ROMANCE AND ARBERESH CAUSATIVES

211

Under standard assumptions that Case assignment occurs under government, the GTC accounts for the Case pattern displayed by causative constructions. The complex verb governs and assigns Case to the arguments associated with the basic verb root. In causatives of intransitive verbs (as in example 1), the subject associated with the lower verb, the causee, acts invariably as a direct object of the complex verb.3 It becomes the subject when the complex verb is passivized, as illustrated in 3, and can be represented by a proclitic on the complex verb, as in 4: (3) 'Ceni 'kye 'pata-tsum'bur njga 'diymbret.

(Savoia 1989:408, ex. 137)

dog was CAUS-jump-PAST PART by children The dog was made to jump by the children.' (4) 'Diymbret a 'peta'tsum'bun. the children it CAUS-jump-PERF-3PL 'The children made it jump.' In causatives of transitive verbs, the base subject or the causee is marked in the dative Case, as in 5, as indirect objects usually are in this language, and can be expressed by a dative proclitic, as in 6.4 The direct object of the basic verb behaves as a direct object of the complex verb. It becomes the subject when the complex verb is passivized, as in 7, and can be represented by a proclitic on the complex verb, as in 8: (5) 'Mishtra'pe'ta'zjyo5i'yibrenA:n'fl > ?Mrve'r.

(Savoia 1989:402, ex. 107)

teacher CAUS-read-PERF-3so book-ACC children-DAT 'The teacher had the children read the book.'

212

PARAMETERS AND FUNCTIONAL HEADS

(6) 'Mishtl'a i 'peta'zjyoSi 'yibre'n. teacher to+them CAUS-read-PERF-3SG book-ACC 'The teacher had them read the book.' (7) 'Sedza 'kys 'peta'yagur njga 'ujt.

(Savoia 1989:408, ex. 139)

chair was CAUS-wet-PAST PART by water 'The chair was made to be wetted by the water.' (8) 'Mishtl'aya 'peta'zjyo'3i.

(Savoia 1989:402, ex. 106)

teacher it-to+them CAUS-read-PERF-3SG 'The teacher had them read it.' From a comparative point of view, it is interesting to note that analytical causatives, such as those found in Italian, are quite similar to morphological causatives of the type present in the Barile dialect of Arberesh. The subject of an infinitive intransitive verb, the causee, becomes the direct object of the complex verb (e.g., in 9). As in Barile, this can be seen in that the causee becomes the subject when the causative verb is passivized (illustrated in 10), and can be represented by a direct object clitic which has to appear on the causative verb (shown in 11): (9) I bambini facevano saltare il cane. the children made jump the dog 'The children made the dog jump.' (10) II cane e stato fatto saltare dai bambini. the dog has been made to jump by the children (11)1 bambini lo facevano saltare. the children it made jump In causatives of transitive verbs, the causee is preceded by the preposition a, as indirect objects are in Italian (12), and can be expressed by a dative proclitic on the causative verb (13). This time it is the direct object associated with the infinitive that behaves as the object of the complex verb. It becomes the subject when the causative verb is passivized (14), and can be expressed by an accusative clitic on the main verb (15): (12) II maestro faceva leggere il libro al bambino. the teacher made read the book to the child 'The teacher made the child read the book.' (13) II maestro gli faceva leggere il libro. the teacher to+him made read the book

ROMANCE AND ARBERESH CAUSATIVES

213

(14) II libro e stato fatto leggere al bambino. the book has been made to be read to the child (15) II maestro lo faceva leggere al bambino, the teacher it made read to the child The properties exhibited by both Barile morphological causatives and Italian analytical causatives indicate that the object and the subject of the basic verb have undergone a Grammatical Function (GF) changing process. The NP which at Dstructure has the grammatical function of subject of the infinitive verb acts either as a direct object or as an indirect object of the complex verb. The NP with the GF of direct object of the basic verb becomes the object of the complex verb. Assuming with Baker that GF changing phenomena are a side effect of incorporation, on the basis of the close similarity between Barile and Italian causatives, it is tempting to conclude that syntactic incorporation occurs not only in Barile, as shown in 2, but also in Italian. However, as the surface output in the two cases is different—no morphologically complex verb is formed in Italian—the question which immediately arises is how the two cases can be discerned. Before addressing this question, I will examine Baker's proposal for Italian and point out some problematic facts for his analysis.

2 Analytical Causatives as the Output of Abstract Incorporation Baker proposes that analytical causatives are derived by head movement in the syntax of LF, rather than at S-structure as morphological causatives are. In other words, analytical causatives are instances of abstract incorporation. This solution, however, seems to be inadequate both for theoretical and empirical reasons. Case assignment takes place at S-structure. But, as Baker mentions (1988:462, n. 37), if abstract incorporation occurs at LF, a level which follows S-structure, it is unclear how one could derive the Case assignment properties that crucially depend on the occurrence of incorporation.

Furthermore, empirical evidence strongly suggests that syntactic incorporation occurs also in Italian. Consider example 17 (adapted from Belletti 1990:136, n. 56): (17) /professorl facevano commentare tutti- quel libro a Ugo. the professors made comment all that book to Ugo 'All the professors made Ugo comment on that book.'

214

PARAMETERS AND FUNCTIONAL HEADS

Based on the hypothesis that subjects are base-generated inside the VP (see Giorgi and Longobardi 1991; Koopman and Sportiche 1991, among others), Sportiche (1988) proposes that Floating Quantifiers (FQs) occur in positions which the subject has moved through on its way from its base position to the specifier of the higher projection, IP. In 17 iprofessori 'the professors', the matrix subject, has stranded the FQ tutti 'all' in a position which, for concreteness, I assume to be in the vicinity of the matrix VP. From the order in 17 — causative-infinitive-FQ—the inference can be drawn that the infinitive has been raised from the causative complement to a position inside the matrix sentence, as displayed in 18. This can be directly accounted for if one admits that incorporation of the infinitive by the causative verb takes place at the syntactic level and not at LF, as Baker proposed. (18) [Ip I professorij [facevano commentarej]k [ VP tuttij tk [ tj quel libro a Ugo] ] Under this view, both analytical and morphological causatives are derived by syntactic incorporation, hence, the same behavior is expected. Still, the morphological result is not the same in the two cases. In Italian, different from Barile, the causative verb and the infinitive are independent words. This fact is also supported by the possibility of inserting adverbial elements between the causative verb and the infinitive, as shown in 19:5 (19) I professori: non fanno piu commentare (tutti:) quel libro a Ugo. the professors NEC make anymore comment (all) that book to Ugo 'All the professors do not make Ugo comment on that book anymore.' I attribute the morphological differences to the fact that Italian causatives, unlike Barile causatives, also involve excorporation of the causative verb from the complex verb previously formed. In other words, Italian causatives are the result of two processes: incorporation and excorporation. In the next section, I introduce the background on which my proposal is based.

3 Obligatory Head Movement as Selected Substitution My proposal is a development of Rizzi and Roberts' (1989) view of verb movement, further elaborated in Roberts (1993b). Specifically, it exploits the idea that heads (X°) project negative bar level heads (X~~ ] ). According to Rizzi and Roberts three types of head-to-head movements can be distinguished: (1) free substitution, (2) selected substitution, and (3) adjunction. Obligatory movement of the verb to 1° to pick up tense and agreement morphemes is an instance of selected substitution. Roberts, based on Selkirk (1982), represents a head that triggers selected substitution as an X""1. Negative bar level heads are filled by affixes, e.g., by inflectional affixes. Affixes cannot stand alone as they are not independent words — to survive they have to be attached to other lexical material. Thus, whenever an X"1 is projected, a morphological subcategorization slot is

ROMANCE AND ARBERESH CAUSATIVES

215

automatically created to host the incorporee. Subcategorization properties have to be satisfied, so, whenever a morphological subcategorization slot is projected, movement of the required element into it must occur, thereby deriving the obligatoriness of verb movement. This view includes the hypothesis that in languages where there is obligatory verb movement, or where inflectional heads project X""1 heads, verb roots are best analyzed as V"1. This proposal is schematically represented in 20, where it is shown that the inflectional head 1° projects an I"1 which subcategorizes for a V~':

4 Morphological Causatives as X -1 Subcategorization I assume with Baker that the causative morpheme in morphological causatives is syntactically a verb exhibiting all the properties of verbs (e.g., thematic properties, syntactic subcategorization requirements). Its morphology uses a lexical affix. Unlike the Barile causative morpheme, the Italian causative verb is morphologically a verb root. I represent these two entities as V""1 and distinguish them as follows. The lexical affix of causation, similar to inflectional affixes (cf. section 3), is a typical incorporation trigger, that is, it is a V"1 that has morphological subcategorization requirements. This statement implies that it is projected in the syntax with a morphological subcategorization slot for hosting the incorporee. A root is also a V"1. But, unlike an affix, it does not have any morphological subcategorization requirements. Instead, roots are selected by affixes. Specifically, I argue that the Barile causative affix morphologically selects a V"1 root. Thus, it is projected in the syntax with a slot, [V~' ], triggering substitution of the lower verb root into it, and forming a complex head of category V""1 with it. This complex verb is subsequently raised and amalgamated with tense and agreement morphemes. I propose that the causative complement of Barile morphological causatives is a VP shell structure; namely, it includes two VP projections as proposed by Larson (1988) for double objects and further extended to all VPs by Hale and Kayser (1991) and Chomsky (1989). Morphological causatives are derived as schematically represented in 21 (cf. Hoyt 1989; Li 1990). Incorporation in morphological causatives is selected substitution of a V~' basic verb root into the slot associated with the V~' lexical affix.6 As indicated above, the notion of V~' is not unique to lexical affixes: it also represents verb roots.

216

PARAMETERS AND FUNCTIONAL HEADS

5 Analytical Causatives as X° Subcategorization I assume that the Italian causative complement is also a VP shell. In other words, no inflectional projection of tense and agreement is associated with the infinitive in Italian causative complements (the reader is referred to Guasti 1993 for evidence supporting this claim). As shown in 22, the infinitival morpheme -re is inserted under the head of VP?,. The verb root, generated in the head of VPi, is raised to the higher verbal head to be amalgamated with the infinitival morpheme.7 To this view I add the hypothesis that the subject is base-generated inside this complex VP, specifically in VPi, in a position which in Italian I assume to be to the right of the verbal head (cf. Giorgi and Longobardi 1991; Koopman and Sportiche 1991). I will argue that this is also the position where the causee surfaces in causatives.8 As noted earlier, the Italian causative head, unlike the one in Barile, does not have the morphology of an affix — it does not subcategorize for a V~'. However, this does not prevent it from having other subcategorization properties. The proposal I advance is that it subcategorizes for a morphologically well-formed word, specifically a V°, triggering incorporation of the infinitive into it as displayed in 23. Under this view, (Barile) morphological causatives and (Italian) analytical causatives are similar in that they are both derived by syntactic incorporation. They differ in the morphological nature of the head triggering incorporation: while this head is a lexical affix in Barile, it is a root in Italian. This property is the core of my

ROMANCE AND ARBERESH CAUSATIVES

217

hproposal and is responsible for the difference between morphological and analytical causatives. This brings me to the issue of excorporation, which is addressed in the next section.

6

Excorporation and the ECP

Excorporation cannot occur in morphological causatives whereas it can and must take place in analytical causatives. This divergence can be formally expressed by developing Baker's suggestion that excorporation must be prohibited as a violation of the ECP (see Roberts 1993a) and more precisely, as a violation of Rizzi's Minimality Condition (see also Guasti 1992a, 1993).9 The Relativized Minimality Condition (Rizzi 1990:27) states that a certain type a of government of a trace by its antecedent is blocked by the intervention of an element endowed with the same a-government

218

PARAMETERS AND FUNCTIONAL HEADS

capabilities. In 24, an intervening governing Z (an A- or A' -specifier, an X°) blocks government of the trace t by its real antecedent X (respectively, an A- or A'-specifier, an X°); that is, Z is a closer potential antecedent-governor for the trace t than X itself. (24) X . . . [ . . . Z ... t] I refine Rizzi's system so that it is sensitive to the X^X"1 distinction introduced earlier by including X"" ' heads among the possible antecedent-governors. The Relativized Minimality, modified along these lines, is given in 25: (25) Relativized Minimality: X a-governs Y iff there is no Z such that: (i) Z is a base-generated position (ii) Z is a-GT compatible with Y (iii) Z c-commands Y but it does not c-command X

In the case of morphological causatives, excorporation is prohibited. The lower verb root, a V"1, is incorporated into the causative affix, also a V"1, forming a complex head. Now, if the causative affix is excorporated and raised alone to the inflectional head, a violation of the Minimality Condition, thus of the ECP, results. As illustrated by the diagram in 26, the causative affix raised to Infl is not able to properly govern its trace t- because a closer potential antecedent-governor, the head V~' of the incorporee, intervenes. Hence, proper government of the trace of the excorporee is not satisfied, leading to an ECP violation.

The situation is quite different in the case of analytical causatives. After incorporation of the V°, the causative root fa- 'make', which I have represented as a

ROMANCE AND ARBERESH CAUSATIVES

219

V"1, is excorporated and raised to Infl where it is amalgamated with the inflectional morphemes. In terms of the definition given in 25, the head intervening between the causative root in Infl and its trace t- is a V° head. This head does not qualify as a potential antecedent-governor for the V"1 trace of the excorporee. Contrary to what happens with morphological causatives, minimality does not block antecedentgovernment and the ECP is satisfied.10 This situation is shown in 27:

The availability of excorporation in analytical causatives accounts for the fact that the causative and the lower verb act as two morphologically independent words. Under this approach the causative verb can, and in fact must, be excorporated. Roots are typical incorporees. They become morphologically well-formed words only by combining with inflectional morphemes. The causative verb root has to amalgamate with inflectional affixes and this implies that it has to be excoiporated from the complex head previously yielded. Note that the triggering factors for incorporation and excorporation are distinct: incorporation is triggered by the syntactic properties of the causative head, while excorporation is forced by the subcategorization properties of the inflectional affixes and ultimately by the morphological properties of the causative head. This proposal offers an explanation for the presence of adverbial material between the causative and the infinitive verb in analytical causatives. It accounts for the relative order causative-infinitive-FQ in the following way (I assume that the FQ is left-adjoined to the matrix VP). First, the infinitive is incorporated by the causative verb. The complex verb thus formed makes a preliminary move and adjoins to 1°. From there, the causative verb root, a V"1, is excorporated and amalgamated with the inflectional morphemes, as seen in 28. 11>12 Recapitulating, analytical and morphological causatives are derived by syntactic incorporation. This explains why the same properties are manifested by both types of causatives. However, the elements involved in the incorporation process have different morphological and lexical properties. This fact, combined with independently motivated principles, namely the ECP, accounts for the surface differences between the two types of causative constructions.

220

PARAMETERS AND FUNCTIONAL HEADS

7 French Causatives and Syntactic Incorporation Causatives in other varieties of Romance manifest a number of similarities with their Italian counterparts. This can be seen by looking at French (cf. the Italian examples in section 1): (29) Les enfants faisaient sauter le chien. the children made jump the dog 'The children made the dog jump.' (30) Le maitre faisait lire le livre a Venfant. the teacher made read the book to the child 'The teacher made the child read the book.' (31) Les enfants le faisaient sauter. the children it made jump 'The children made it jump.' (32) Le maitre lui faisait lire le livre. the teacher to+him has made read the book 'The teacher made him read the book.' (33) Le maitre le faisait lire a 1'enfant. the teacher it has made read to the child 'The teacher made the child read it.' Examples 29 and 31, on the one hand, and example 33 on the other, respectively show that the lower subject of an intransitive verb and the lower object of a transitive verb behave as direct objects of the complex verb. Examples 30 and 32 prove that the lower subject of a transitive verb behaves as an indirect object of the complex

ROMANCE AND ARBERESH CAUSATIVES

221

verb.13 The close similarity between Italian and French causatives calls for a unified account. I propose to extend to French the analysis proposed in section 5 for Italian. This amounts to saying that syntactic incorporation of the infinitive also occurs in French. This extension is supported by the distribution of FQs (example from Quicoli 1976:585): (34) Mes amiSj [feront manger^ [vp tous^ tk [t; de la salade a ce gargon] ] 'All my friends will make this boy eat salad.' In 34, the FQ stranded by the matrix subject appears to the right of the infinitive. As argued by den Dikken (1990), this is a sign that the infinitive has been incorporated into the causative verb. Under previous assumptions, this conclusion implies that French fai-(re), analogous to Italian fa-(re), selects a V°. Thus, the infinitive verb is incorporated into the causative verb and with it, forms a complex verb. As in Italian, excorporation of the causative verb takes place, thereby accounting for the presence of two verbs from the morphological point of view. Although this proposal is correct, it does not exhaust the discussion about French causatives. There are some striking asymmetries between Italian and French causatives that deserve to be treated separately.

8

Differences between Italian and French Causatives

French and Italian causatives differ with respect to the distribution of the anaphoric clitic se. Consider the reflexive verbs 'se reveiller' and svegliarsi 'to wake up' in French and Italian, respectively. When these verbs are embedded under a causative verb, the anaphoric clitic se must be present in French causative complements, as shown by the contrast between 35 and 38, whereas, in the same context, si must be absent in Italian, as illustrated by the contrast between 36 and 37:14 (35)

II faisait se^ reveiller la fille;. he made se wake up the girl 'He made the girl wake herself up.'

(36) *Egli faceva svegliam'j la ragazza;. he made wake-5/ up the girl 'He made the girl wake herself up.' (37)

Egli faceva svegliare la ragazza.15 he made wake up the girl 'He made the girl wake herself up.'

(38) *I1 faisait reveiller la fille.16 'He made the girl wake herself up.'

222

PARAMETERS AND FUNCTIONAL HEADS

Leaving aside Italian for the moment, it can be proven that French causative complements containing the anaphoric clitic do not behave as those causative complements discussed in section 7. One major difference is that they do not involve incorporation of the infinitive. In fact, an FQ stranded by the causer cannot surface to the right of the infinitive, as in 39, but only to its left, as in 40. Notice, crucially, that if se is absent, the FQ can appear to the right of the infinitive, in 41, as expected: (39) *Les professeurs^ font s'acheter tous^ ce livre aux etudiants. the professors make se buy all this book to the students (40)

Les professeurSi font tous^ s'acheter ce livre aux etudiants. the professors make all se buy this book to the students

(41) Les professeurSi font acheter tous-t ce livre aux etudiants. the professors make buy all this book to the students In section 7, it was shown that clitics corresponding to the arguments of the infinitive appear on the causative verb, a result which ensues from incorporation of the infinitive. In the causatives under investigation, not only the anaphoric clitic, but also clitics representing the complements of the infinitive cannot appear on the matrix verb. They must surface on the infinitive, as demonstrated by the contrast between 42 and 43.17 Crucially, if se is absent these clitics must appear on the causative verb, as displayed by the contrast between 44 and 45 (cf. also 33): (42)

*Sa mere les faisait s'acheter a la jeune fille.

(43) ??Sa mere faisait se les acheter a la jeune fille. 'Her mother made the young girl buy them to herself.' (44)

Sa mere les faisait acheter a la jeune fille.

(45)

*Sa mere faisait les acheter a la jeune fille. 'Her mother made the young girl buy them.'

In contrast, a clitic corresponding to the causee continues to appear on the matrix verb. In this respect, 46 and 47 do not differ from the causative constructions discussed in section 7 (see 31 and 32). (46)

Sa mere lui a fait s' acheter les chaussures. 'Her mother made her buy herself the shoes.'

(47)

II /' a fait se reveiller. 'He made her wake herself up.'

ROMANCE AND ARBERESH CAUSATIVES

223

These facts lead to the following conclusions. Unlike Italian, French causative complements can contain the anaphoric clitic se. When this happens, syntactic incorporation of the infinitive does not take place. Moreover, while the object clitic has to appear on the infinitive, the clitic representing the causee continues to surface on the causative verb. The most obvious explanation would be that, in French causatives, syntactic incorporation is only optional contrary, to the previous claim that causatives are derived by a process of obligatory syntactic incorporation. This solution, however, leaves some facts unexplained. First, if incorporation is optional, it is unclear why it cannot take place precisely when the clitic se is present. Second, it does not account for the impossibility of si in Italian causatives. I must therefore dismiss this solution and elaborate an alternative which is based on the results of an investigation of causatives in another variety of Arberesh, that spoken in San Nicola (Catanzaro). Unlike Barile (section 1), this dialect has analytical causatives which are similar to Italian and French (section 7) in terms of Case. Additionally, they manifest the properties of French causatives containing the anaphoric clitic se. So, for the moment, French must be put aside.

9 9.1

Subjunctive Causative Complements in San Nicola

The Data

Analytical causatives in the Arberesh dialect of San Nicola are formed by two verbs. Unlike Italian and French, both are inflected for tense and agreement. The verb in the complement is in the subjunctive mood and is introduced by the subjunctive particle (SP) te. As in Italian and French, the causee, appearing invariably at the end of the sentence, is morphologically marked in the accusative or dative, according to the transitivity of the verb (examples 48 and 49 from Turano 1989).18 (48) Maria bon te shurbenj Frankun. Maria makes SP works Frank-ACC 'Maria makes Frank work.' (49) Maria i bon te ghojirnj ghibrin ghajarellit. Maria to+him makes SP reads book-ACC child-DAT 'Maria makes the child read the book.' The causee can be expressed by an accusative or dative clitic appearing on the matrix verb, as displayed in 50 and 51. In causatives of intransitive verbs, the causee becomes the subject when the causative verb is passivized, as in 52. In 51 the dative NP is doubled by a clitic, as it is usually the case in Arberesh. (50) Maria e bon te shurbenj. Maria him makes SP works 'Maria makes him work.'

(Turano 1989:117, ex. 30)

224

PARAMETERS AND FUNCTIONAL HEADS

(51) Maria i bon te dobarnj makinin.

(Turano 1989:174, ex. 80)

Maria to+him makes SP repairs car-ACC 'Maria makes him repair the car.' (52)

Franku osht i bon te shurbenj.19

(Turano 1989:117, ex. 29)

Franku is DET made SP works 'Frank is made to work.' The situation becomes more puzzling when one considers the lower object. This cannot become the matrix subject when the causative verb is passivized, as shown by the ungrammaticality of 53 (Turano 1989:136, ex. 55). This option is, however, available when not just the causative but also the embedded verb is passivized, as in 54 (Turano 1989:137, ex. 56): (53) *Gibri osht i bon te ghojirj ghajarellit. book is DET made reads child-DAT (54)

Gibri osht i bon te jet i ghojirtur ka ghajarelli. book is DET made SP is DET read by children 'The book is made to be read by the children.'

Moreover, a clitic corresponding to the lower object must appear both on the causative verb as well as on the subjunctive verb, as in 55. The example in 56 is ungrammatical because the clitic is present only on the causative verb: (55)

Maria ja bon t' e dobarnj Frankut. Maria to+him-it makes SP it repairs Frank-DAT 'Maria makes Frank repair it.'

(56) *Mariaj'a bon te dobarnj Frankut. Maria to+him-it makes SP repairs Frank-DAT If the Case pattern and the phenomena related to it are a side effect of incorporation, as I have assumed throughout, there is but one way to interpret the above evidence; namely, that incorporation must apply also in this type of causatives. The differences between San Nicola and Italian (and French, to some extent) causatives must then be ascribed to the structural differences between causative complements in the two languages and ultimately to the nature of the incorporee.

9.2

The Analysis of Subjunctive Causative Complements

I assume that subjunctive complements in San Nicola causatives have the structure in 57. The subjunctive marker heads its own projection, a Mood Phrase (MP) (see Motapanyane 1991; Rivero 1991; Terzi 1991). The M° takes an IP as complement.

ROMANCE AND ARBERESH CAUSATIVES

225

The subjunctive particle contributes to the inflectional features of the same clause containing the verb. I represent this by coindexing M° with 1°, an instance of headto-head agreement. This coindexation makes the subjunctive marker a member of the verbal chain containing the inflected verb. In 57, the causee is located in the base-generated position for the subject, i.e., the thematic position which like Italian, is to the right of the verb. Subjunctive clauses under causative verbs are like all clauses; in particular, they include agreement features. Therefore, an expletive pro is inserted in Spec-IP, in compliance with the Extended Projection Principle (EPP), according to which all clauses have a structural subject position.20'21 I attribute the GF changing process, and consequently the Case pattern exhibited by San Nicola causative constructions, to the occurrence of incorporation. Specifically, I maintain that the causative verb is a root that selects an X°, as in Italian and French, and unlike Barile. I propose that in San Nicola, this X° is the subjunctive particle rilling the M° head. This particle is incorporated into the V"1 head of the causative verb forming with it a complex head. In turn, the V""1 causative verb is excorporated from the complex head and raised to tense and agreement. As in Italian, excorporation results in a well-formed structure, as can be easily checked. Given that the subjunctive verb has not undergone incorporation, one expects that an FQ stranded by the matrix subject cannot appear to the right of the subjunctive verb. This prediction is indeed borne out:22 (58) *Professort-l i kan bon te kommentarj gjithe^ gibrin Maries, professors to+her have made SP comment all book Mari-DAT 'All the professors made Marie comment on the book.' In section 9.1, a number of properties of San Nicola causatives that are not found in Italian causatives were presented. These differences must be ascribed to the fact that in San Nicola the causative complement is a full clause or, equally, that the incorporee is not the lower verb, but the subjunctive particle, the M° head. This brings me to examine more closely the diagnostic properties for the GF changing process advocated earlier, namely cliticization and NP-movement.

226

9.3

PARAMETERS AND FUNCTIONAL HEADS

Incorporation ofM° and Government

Incorporation of M° automatically changes the governing properties of the structure and interferes with the normal Case-marking relationships. As stated by the GTC (section 1), a complex head governs everything that was governed by its members, the causative verb and the subjunctive particle. Investigating Italian causatives, I have shown that incorporation affects all the arguments of the lower verb, placing them in the government domain of the complex head. This does not hold in the language under inspection: only the lower subject is governed by the complex head, whereas the lower object remains in the government domain of the lower verb and thus receives Case from it. Let us consider the facts in greater detail. The lower post-verbal subject is in a chain with the expletive pro located in the SpecIP. By the definition of chain given in 59 as a sequence of indexed positions among which antecedent-government holds, the relation between pro and the postverbal subject is one of government (cf. Chomsky 1986; Rizzi 1990:92): (59)

(a t , . . . , an) is a chain only if, for 1 < i < n a4 antecedent-governs ai+1

The expletive pro, in turn, is governed by M°, given Chomsky's (1986:162) definition of government. After incorporation of M° the null element—and by transitivity the postverbal subject — comes to be governed by the complex head; thus, the causee behaves like an argument of the complex head. This justifies the placement of the clitic on the causative verb, as in 50 and 51. The same explanation can be easily extended to NP-movement in 52, a topic to which I will return in a moment. Unlike the situation with the causee, the embedded object is governed and assigned Case by the subjunctive verb. On the assumption that a clitic is adjoined to a head that has the relevant Case features (Burzio 1986, among others), one expects that a clitic referring to the lower object appears on the subjunctive verb, as in 55 and not on the causative verb, as in 56.23 The ungrammaticality of 56 can be regarded as a violation of the Subject Specified Condition (SSC) with the expletive pro in Spec-IP blocking extraction of the object clitic. (60) * Maria ja; bon te'k [Mp tk [ IP pro [j<> dobarnj tj] Frankut] 'Maria makes Frank repair it.' Let me now turn to the NP-movement of the causee in causatives of intransitive verbs. I assume that NP-movement in 52 proceeds as in 61. The NP is raised from the base position leaving a trace f there. On its way to the matrix clause, it transits through Spec-IP where a second trace, t", is left. The expletive pro, which normally is in Spec-IP, is replaced by this trace. The chain which is obtained is well-formed, in that it has both a Case and a thematic role.24 (61) Franku; osht i bon tek [MP tk [IP t"j shurbenj t'j] ] In 53 (repeated in 62), the object moves out of the subjunctive clause crossing the expletive pro coindexed with the causee, thus yielding an SSC violation: (62) *Gibri: osht i bon te'k [ MP tk [ IP pro( ghojirj tj ghajarellit;]

ROMANCE AND ARBERESH CAUSATIVES

227

'The book is made to be read by the child.' In causatives, the lower object becomes the matrix subject if the lower verb is also in the passive. In this case, the object becomes a structural subject and is coindexed with the expletive pro in Spec-IP. This time, the D-structure object can move through Spec-IP, erasing pro and leaving a trace there. The chain connecting the raised NP and its trace is well-formed—each link of the chain is antecedent-governed: (63)

Gibri osht i bon tek [MP tk [Ip t"j jet i ghojirtur tj] ka ghajarelli The book is made to be read by the children.'

There is a last point that should be dealt with concerning the Case pattern. The causee is marked accusative or dative depending on whether the subjunctive verb has been already assigned accusative or not. The question is how Case on the causee can hinge on the Case features of the subjunctive verb, given that it has not undergone incorporation. Baker (1988) proposes that a complex head inherits the Case features from its members. Elaborating on this view, I propose that a complex head can inherit only the Case features that have not been already assigned by at most one of its members. I implement this proposal as follows. I assume that structural Case is assigned by a verb in conjunction with the inflectional heads typically present in a clause (Roberts 1987). Thus, in subjunctive clauses, the three heads M°, 1°, V° are responsible for Case assignment. This implies that M° counts as a Case assignor, although not by itself. In San Nicola causatives, the complex [V"1 + M°] acquires the Case capacities from the causative verb and from M°. Assume now that the causative verb can assign accusative and dative Case.25 Under these assumptions, the inheritance mechanism works as follows: a. In causatives of intransitive verbs, the complex [V""1 + M°] inherits accusative Case from the causative verb which is then assigned to the causee. b. In causatives of transitive verbs, M° is part of the chain associated with the lower verb and is involved in the assignment of the accusative feature to the lower object. Therefore, the complex [V""1 + M°] cannot inherit accusative Case because it has already been assigned by one of its member, M°. Under these circumstances, the causee receives dative Case, a Case contributed by the causative verb itself (see Guasti 1993).26 This implementation explains why the Case pattern of San Nicola causatives looks like that of Italian causatives. In summary, incorporation applies in the Arberesh variety of San Nicola, as manifested by the GF changing phenomena also displayed by this type of causatives. Unlike Italian, the causative head has different lexical properties. In Italian, it selects a morphologically well-formed V°, the infinitive verb; in San Nicola, it selects the M° containing the subjunctive particle te. This hypothesis is the basis for the explanation of the asymmetry between the lower subject and the lower object manifested by NP-movement and cliticization.

228

PARAMETERS AND FUNCTIONAL HEADS

10 Nora-Active Voice in Arberesh and Anaphoric Clitics Causative complements in San Nicola exhibit some similarities with those French causative complements that contain the anaphoric clitic se. In both cases, an asymmetry between the subject and the object of the embedded verb was found. Furthermore, neither the infinitive nor the subjunctive verb have been incorporated, as shown by the distribution of the FQ in 39 and 58. The resemblance between San Nicola and French causatives becomes closer once the non-active voice is considered. In Balkan languages and in Arberesh, the non-active morphology or the medium voice (M) can be employed in reflexive and reciprocal constructions, as in 64, 65 and 66: (64)

Ai ziohet (San Nicola) he wake-M-PRES-3SG

(65)

Ato vrehen (San Nicola) They look-M-PRES-3pL 'They look at each other.'

(66)

A'je 'zfo'net. (Barile)

(Brandi and Savoia 1990:89, ex. lOla)

he wake-M-PRES-3sG 'He wakes himself up.' It is reasonable to treat the non-active morphology in the same manner as the clitic se, given that both perform the same functions. For our purposes, it is interesting to examine the interactions between non-active voice and causatives. As in French, the non-active voice must appear on the lower verb in the San Nicola variety. Compare the example below (67) with the French sentence in 35: (67)

Ai bon te ziohet Giuvanin. he makes SP wake-M-PRES-3SG Giuvan-ACC 'He makes Giuvan wake himself up.'

At this point, the examination of causatives in the dialect of Barile becomes particularly telling. In section 4,1 had proposed that causative complements in this dialect are VP shells, as they are in Italian. Interestingly enough, the non-active voice cannot appear in Barile causative complements. In this respect, the contrast between 68 and 69 is parallel to the one displayed by the Italian examples in 36 and 37: (68) *A'je 'beta'zfo'net 'vajzen. he CAUS-wake-M-pre.s-3.sg girl-ACC 'He makes the girl wake herself up.' (69)

A'je 'beta'zfo'n 'vajzen. he CAUS-wake-PRES-3SGgirl-ACC 'He makes the girl wake up.'

(Brandi and Savoia 1990:89, ex. 102a)

ROMANCE AND ARBERESH CAUSATIVES

229

From the facts examined so far the following conclusions can be drawn. Leaving aside morphological properties, Barile and Italian causatives are structurally and syntactically similar: the causative complement is a VP shell. The distribution of clitics and of the FQ stranded by the matrix subject (as shown in section 7) supports the view that French causatives can enter the VP shell structure as well. French and San Nicola (though not Italian and Barile) admit the anaphoric clitic and the nonactive voice, respectively, in their causative complements. It is therefore tempting to regard the availability of the non-active voice as a property of the MP structure. Given the similarity between se and non-active voice, it is natural to extend the same treatment to se and say that it must also be regarded as a property of the MP structure. This implies that French can enter the MP structure as well. Thus, French causative complements are structurally ambiguous: they can be either VP shells or MPs.27 French causative complements containing the anaphoric clitic receive the structural analysis shown in 70. Unlike San Nicola, the modal head contains a phonologically null modal particle in French: (70) II a fait [M° [IP se{ reveiller la 'He made the girl wake herself up.' With this approach, some infinitives are assimilated to subjunctive forms of the verb (see Bianchi 1991 for a proposal according to which some infinitive clauses include a modal projection).28 Consider the structure in 70 again. As usual, a GF changing process is also manifested by MP French causative complements: the causee behaves like an indirect object or a direct object, depending on the transitivity of the lower verb (see 46, for example). To account for this Case pattern, I assume that incorporation takes place here. I suggest that the incorporee is the phonological null modal particle in M°, analogous to what has been seen in San Nicola causatives. This proposal is supported by the distribution of FQs in 39 and 40. Given that only the null modal particle is incorporated, in no way can the FQ appear to the right of the infinitive. This contrasts sharply with what was found in VP shell causative complements (e.g., in 34). As in San Nicola, an expletive pro occupies Spec-IP. I assume that this pro is licensed by M°. Given the assimilation, implied by my proposal, between infinitive and subjunctive verbs, this conjecture does not appear implausible. Subjunctive forms of the verb can license an expletive pro in Stylistic Inversion, as argued by Pollock (1986; see also Kayne and Pollock 1978): (71) II faudrait que proi viennentplus de linguisteSj a nos reunions. we need that come more linguists to our meetings With this background, it is possible to derive all the properties manifested by the French MP causative complements, in particular the asymmetries between the embedded object and the embedded subject with respect to cliticization. This amounts to saying that after incorporation of M°, only the causee is governed by the complex [V~' + M°], as has been extensively discussed in the case of San Nicola causatives. Consequently, a clitic staying for the causee appears on the complex verb, as in 46, whereas a clitic representing the embedded object appears on the infinitive, as in 43.

230

PARAMETERS AND FUNCTIONAL HEADS

A last aspect I would like to address more explicitly concerns the distribution of the non-active voice and of the anaphoric clitic. Along with Rivero (1990), I propose that the non-active voice heads its own maximal projection (M) and that this projection dominates VP. If the causative complements in Barile are VP shells, as proposed in this paper, the non-active voice can never be present. In other words, the causative complements are structurally too poor to include such a projection. In San Nicola, causative complements are MPs; all the functional categories normally associated with the verb are present, so the non-active voice which dominates VP is also present and can show up in causatives. Assume now that anaphoric clitics selsi also head their own projection (cf. Sportiche 1992) or have to surface on a designated functional head, along Kayne's (1991) lines. Whichever hypothesis proves correct, the impossibility of having the anaphoric clitic si in Italian causative complements depends on the absence of a functional projection for the anaphoric clitic. In Italian, as in Barile, causative complements are unambiguously VP shells and in this structure anaphoric clitics cannot show up. Consider now French. Causative verbs in French can select not only a VP shell, but also a MP structure. This structure, which is not available for Italian causative complements, includes all the functional projections usually associated with verbs, so it can accommodate the anaphoric clitic se analogous to the MP structure in San Nicola causatives. Under this view, the difference between Italian and French causatives is structural. Italian has only VP shell causative complements, whereas French, beyond the VP shell, can take the MP structure, at least under certain circumstances (see also section II). 29 Although the VP shell cannot accommodate the anaphoric clitic in French (also impossible in Italian), the MP structure can. The hypothesis according to which the differences between French and Italian causatives derive from the presence in French of two causative complements is further supported by other data concerning the distribution of negation.

11 The Distribution of Negation Negation is marginally possible in French causative complements, as originally noticed by Kayne (1989:242): (72) ??Cela faisait ne pas manger la soupe a F enfant. that made NEC not eat his soup to the child Negation in 72 is an instance of sentence negation. For example, it can have scope over lexical material other than the infinitive itself. Specifically, it can be used to contrast different complements associated with the infinitive, as sentence negation usually can (see also Guasti 1993): (73) ??Cela lui faisait ne pas manger la pomme, mais la soupe. 'That made him not eat the apple, but the soup.' (74) ??Cela le faisait ne pas manger a midi, mais a une heure. 'That made him not eat at noon, but at one o' clock.'

ROMANCE AND ARBERESH CAUSATIVES

231

(75) ??Cela le faisait ne pas manger, mais jeuner. 'That made him not eat, but fast.' The same set of diagnostics demonstrates that, in Italian, sentence negation cannot be present in causative complements: (76)

*Cio gli faceva non mangiare (mica) la mela, ma la zuppa. 'That made him not eat the apple, but the soup.'

(77)

*Cio lo faceva non mangiare a mezzogiorno, ma all'una. 'That made him not eat at noon, but at one o' clock.'

(78) ??Cio lo faceva non mangiare, ma digiunare. 'That made him not eat, but fast.' Sentence 78 may be marginally acceptable, at least for some speakers, but only by interpreting non as applying to the infinitive alone. Structurally, this can be expressed by the statement that negation is directly adjoined to the infinitive verb. Under the assumption that sentence negation heads a maximal projection NegP (cf. Belletti 1990; Pollock 1989, among others), one is led to conclude that French causatives may include a NegP, whereas this is not possible in Italian causatives. I assume here that NegP lies in between IP and VP and that ne is its head and pas its specifier. The negative head ne, due to its clitic nature, must move from Neg° to 1° via head-to-head movement, thus preceding pas at S-structure. This implies that negated causative complements also have to include an IP. I have already noted that French causative complements can enter a structure containing such a functional projection, i.e., the MP complement. It is then tempting to say that negated causative complements in French are actually MPs. This hypothesis makes certain interesting predictions. I proposed that in the MP structure the incorporee is not the infinitive, but the M°. If causative complements containing negation are also MPs, an FQ stranded by the causer should not surface to the right of the infinitive, but only to its left. As shown below, this prediction is fulfilled. Compare 79 and 80 with 39 and 40, respectively:30 (79)

*Mes amiSj faisaient ne pas manger tous^ de la salade a 1' enfant. my friends made NEC not eat all some salad to the child

(80) HMes amis^ faisaient tous^ ne pas manger de la salade a 1'enfant. my friends made all NEC not eat some salad to the child Secondly, one would expect that a clitic corresponding to the object of the infinitive could not appear on the matrix verb. This expectation is also borne out. When negation is present in causative complements, the clitic must surface on the infinitive. Compare 81 and 82 with 42 and 43. The same judgments are reported in Li (1990, crediting Dell):

232

PARAMETERS AND FUNCTIONAL HEADS

(81) *Cela la faisait ne pas manger t a 1'enfant. that it made NEC not eat to the child (82) ??Cela faisait ne pas la manger t a 1'enfant. that made NEC not it eat to the child In contrast, a clitic corresponding to the lower subject must appear on the matrix verb; sentence 83 is parallel to 46: (83) ??Cela lui faisait ne pas manger la soupe t. that to+him made NEC not eat the soup These facts lend additional support to the hypothesis that French causative complements can enter the MP structure. Causative complements containing negation display the same cluster of properties manifested by causative complements containing the anaphoric clitic se. In both cases, the infinitive does not undergo syntactic incorporation. Thus, an FQ stranded by the matrix subject cannot appear to the right of the infinitive and an asymmetry between the behavior of the lower subject and of the lower object is exhibited. The impossibility of having sentence negation in Italian causative complements is also expected, given that in this language the MP structure is unavailable. In summary, French causatives share properties of both Italian (and Barile) and San Nicola causatives. This observation has been formalized by proposing that French causatives can enter two structures, a VP shell and an MP. Although this hypothesis is well-grounded on cross-linguistic evidence, it should be noticed that the MP structure has a limited and marked status in French. If the MP structure were available without restrictions, one should be able to find causative sentences with the accusative clitic either on the infinitive or on the causative verb freely. But leaving aside causative complements containing the anaphoric clitic se and negation, a clitic corresponding to an argument of the infinitive does not appear on the infinitive itself, as shown in 84, but has to surface on the causative verb, as in 85: (84) *Cela faisait la manger t a 1'enfant. that made it eat to the child (85)

Cela la faisait manger t a 1'enfant. that it made eat to the child

The MP complement is only a marked structure in French. This conclusion adequately accounts for the observation made at the beginning of the section that for some speakers negated causative complements are marginal or even impossible. Moreover, my informants judged causative complements containing the anaphoric clitic se to be not very natural either, if acceptable at all. Other speakers found them grammatical, but they would never use them. They would rather employ a tensed sentence: (86)

II a fait en sorte que la fille se reveille, he has acted so that the girl wake up

ROMANCE AND ARBERESH CAUSATIVES

233

A similar remark is made by Seuren (1973:60, n. 7), who notes that "native speakers of French do feel a certain strain when a reflexive pronoun is used this way [in a causative complement."31 This complies with the marked status of the MP causative complements in French. It is not surprising that one finds variations among French speakers precisely when one considers these types of marked causative sentences.

12 Conclusions The investigation presented in this paper has shown that in closely related languages causative complements may vary minimally. I have contrasted two Romance languages, Italian and French, and two varieties of Arberesh, the Barile and San Nicola varieties. At one end of the scale there is Barile morphological causatives and at the other there is San Nicola analytical causatives. In between are Italian and French causatives. Italian and Barile causatives look quite similar, the morphological nature of the two causative verbs aside. The causative complement is a VP shell and in both languages the embedded verb undergoes syntactic incorporation. In San Nicola, the causative complement has the structure of an MP and only the subjunctive particle in M° is subject to syntactic incorporation. The different nature of the incorporee justifies the asymmetries between causatives in the two sets of languages. French represents an intermediate stage, sharing properties both with Italian and, at least marginally, with San Nicola causatives. I have formalized this situation by proposing that French causative complements can enter two different structures: a VP shell and an MP structure. The MP structure which is present in San Nicola has only a marked status in French. This remark accords with the observation that causative sentences analyzed as MPs in French have a marginal status. In this framework, differences between French and Italian causatives conceal the fact that French, unlike Italian, has two structurally distinct causative complements.

Notes 1. This article has benefited from discussions with Adriana Belletti, Luciana Brandi, Liliane Haegeman, Richard Kayne, Luigi Rizzi, Leonardo M. Savoia, Giuseppina Turano, and Eric Wehrli. Parts of this paper were presented at the University of Paris VIII (1990), at the Incontro di Grammatica Generativa held in Trieste (1991), at the GLOW Colloquium held in Leiden (1991), and at the CUNY Syntax Lunch in New York (1992). I am grateful to the audiences at these events. 2. Given the fact that the conjugation markings on verbs frequently result from vowel and consonant alternation or elimination in the verbal stem, it is not always possible to take apart the various inflectional morphemes. 3. More accurately, Case is determined by the GTC combined with the Case Frame Preservation Principle, which states that in a given language the maximal case-marking abilities of a complex verb do not exceed those of a simple verb (cf. Baker 1988:122). In Barile, simple verbs may assign accusative and dative Case; thus, complex verbs do the same. 4. In this paper, I only deal with the causative construction referred to as Faire-Infinitif (see Kayne 1977; Burzio 1986; Guasti 1991 and 1993 for a discussion of the other causative construction, Faire-par).

234

PARAMETERS AND FUNCTIONAL HEADS

5. In Barile causatives, adverbial material can never appear in between the causative affix and the lower verb. I am grateful to Leonardo M. Savoia for this information. 6. The representation in the text wherein the complex V~ ' is formed by V~ ' plus V~ ' , though reminiscent of an adjunction structure, is in fact obtained by substitution. I propose that in this structure each V"1 is a distinct category. 7. 1 propose that in both non-finite and finite clauses, verbs start off in a VP shell structure. The higher verbal head, which is filled by the infinitival affix, is empty in finite clauses. 8. The hypothesis that in Italian the causative verb and the infinitive head are two distinct projections is supported by the distribution of adverbials. Consider the example in (i) (cf. Ruwet 1972:138): (i) Adele ha fatto cuocere il maiale con un limone in bocca 'Adele had the pork cooked with a lemon in its mouth.' 'Adele had a lemon in its mouth when she cooked the pork.' The adverbial con un limone in bocca can either refer to the way Adele did the cooking (she was keeping a lemon in her mouth), or to the way of cooking (the pig had a lemon in its mouth). In other words, the adverbial can modify either the causative verb or the infinitive. Hoyt (1989) develops the same argument for morphological causatives in Lebanese Arabic. 9. 1 assume here that antecedent-governement is part of ECP. In Rizzi (1990), antecedentgovernment is viewed as a condition on chain formation. 10. To be more precise, I have to point out that when the infinitive is incorporated into the causative verb the complete configuration is as follows:

The infinitive verb is a V° that contains a V""1. One may wonder why this V"1 does not count as an intervener for the trace of the "V~ l when the causative verb is excorporated. Notice that even if the incorporee contains a V~~ ' it is a V°. This observation suggests that whenever the V" level has been reached (i.e., when a well-formed word is obtained, the V""1 level is not visible anymore. 1 1. To ensure that the causative head governs its trace in the adjoined V°, I have to assume the following definition of c-command: a commands ft iff at least one segment of a does not dominate ft and for every X' projection 5, if S dominates a then 5 dominates ft. 12. It has been claimed that every verb must be identified by tense (cf. Gueron and Hoekstra 1988; Higginbotham 1985; Zagona 1988), a requirement that may be regarded as a variant of the Case Filter. The verb, in causative complements, does not have its own tense, but due to incorporation, it is identified by the tense associated with the matrix verb. In morphological causatives, the complex verb formed by the basic verb root plus the causative morpheme is incorporated into Infl where tense features are located. As noticed in the text, in Italian analytical causatives, tense identification of the infinitive can be achieved by using two strategies: either the causative root is raised directly to Infl or there is a preliminary movement of the complex verb to Infl, followed by excorporation of the causative verb. 13. Some problems remain, for which I have no explanation. Unlike Italian, neither the lower subject of an intransitive, nor the object of a transitive verb can be passivized in French. Compare the examples below with their Italian counterparts (see also Guasti 1993):

ROMANCE AND ARBERESH CAUSATIVES

235

(i) *Le chien a etc fait sauter par les enfants. the dog has been made to jump by the children (ii) *Le livre a etc fait lire a 1'enfant. the book has been made read to the child. 14. According to the Binding Theory, the anaphoric clitics se and si must be coindexed with the subject of the clause which, in causative complements, is the causee. 15. Besides the reflexive reading, this sentence has another reading that can be paraphrased as follows: 'he had the girl be woken up by someone'. 16. This sentence is unacceptable in the reflexive reading but is correct if it means that someone woke up the girl. 17. The French speakers I consulted agree with the judgments reported in Rouveret and Vergnaud (1980:154). Kayne (1977:400-401) stars both 42 and 43, but notes an improvement when the lower subject is cliticized. 18. Despite the fact that it is not in the nominative, the lower subject always agrees with the lower verb. 19. The (' before the past participle ban is a masculine article accompanying adjectives, pronouns and past participles, but only in the context of the auxiliary be. 20.1 assume that SpecMP is never projected in these causative complements. I explain the absence of SpecMP by calling on the notion that a specifier position is not given gratuitously but must be licensed. One way to accomplish this is through agreement features. A head endowed with agreement features triggers the projection of a specifier to host an NP on which the agreement features are discharged. As M° lacks agreement features, it is unable to license a specifier. 21. The EPP can be viewed as a condition stating that agreement features must be discharged on an NP (see Guasti 1993; Roberts 1992). 22. The FQ cannot appear between the subjunctive particle and the subjunctive verb either. One reason may be that the subjunctive marker is a phonological clitic that must be cliticized onto the subjunctive verb at PF and this can only happen if it is adjacent to it. Alternatively, this may be due to the fact that there is no available position for it in this environment, a question which is beyond the scope of this paper. 23. In sentence 55, the clitic corresponding to the embedded object appears on the subjunctive verb as expected. However, it remains unexplained why it has also to appear on the causative verb. 24.1 assume that each trace of an A-chain must be properly head-governed. The set of headgovernors includes lexical heads, and functional heads that contain inflectional affixes typically associated with the verb, hence 1° and M°. This is sufficient to rule out raising out of finite clauses or of infinitive clauses introduced by a complementizer: (i) Gianni: sembra trjp di [jp t': volare t":] ] Gianni seems to fly In (i), the trace t': in Spec-IP is not properly head-governed, because C° does not qualify in this way. The reader can refer to Guasti (1993) for an extension of this account to NP-movement in the finite raising structures found in languages such as Romanian and Greek. 25. As in Italian (i), the verb make can take a direct and an indirect object (ii): (i) Ho fatto un regalo a Gianni. (ii) I kam bon nje rigagh Giuvanit. (I) have made a present-ACC Giuvan-DAT 26. The view that structural Case is a property of a lexical transitive verb combined with functional heads entails that in Barile and in Italian causatives (and in French ones, to some

236

PARAMETERS AND FUNCTIONAL HEADS

extent), Case can only be assigned by the complex verb, no functional heads being present in these causative complements. 27. Notice that the facts illustrated here are compatible with a less strong hypothesis according to which French causatives can enter an additional structure which includes functional projections, although this may not necessarily be an MR 28. This assimilation is not surprising. In fact, beyond Arberesh, various languages lack infinitives, for example Greek and Albanian (Tosk), and use subjunctive forms in those contexts where Italian, for example, would use an infinitive. 29. Notice that it is impossible to dispense with the VP shell structure and claim that causatives in French are always MPs. Since Pollock (1989), it is well known that infinitives do not move up to the higher functional head. A head cannot skip other intervening heads, because of the ECP. Hence, from the MP structure the infinitive will never have the chance to undergo incorporation into the causative verb, thus leaving the order in 34 unexplained. Only the VP shell can lend itself to syntactic incorporation of the infinitive, thus allowing the FQ to be on the right of the infinitive. 30. Negation is basically impossible in the San Nicola causative complements. Therefore, no comparison with French can be offered (see Guasti 1993). 31. As pointed out to me by Eric Wehrli (personal communication), the hypothesis that French has the MP structure at its disposal as a last resort, adequately accounts for the fact that whenever a clitic cannot be placed on the causative verb, it can appear on the infinitive. Some examples are given below: (i) *Cela nous les a fait telephoner pendant la nuit. this to+us them has made call during the night (ii) (?)Cela les a fait nous telephoner pendant la nuit. this them has made to+us call during the night This made them call us during the night.' (iii) *Jean nous lui a fait presenter les candidats. Jean to+us to+him has made introduce the candidates (iv) (?)Jean nous a fait lui presenter les candidats. Jean to+us has made to+him introduce the candidates 'Jean made us introduce the candidates to him.'

References 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. Bianchi, V. 1991. "Le relative infinitive e altri usi modali deH'infinito," tesi di laurea, Pisa. Brandi, L. and M.L. Savoia. 1990. "Proprieta morfosintattiche e assegnazione del Case nel causativo arberesh." Rivista di Grammatica Generativa 15:29-121. Burzio, L. 1986. Italian Syntax: A Government-Binding Approach. Dordrecht: Reidel. Chomsky, N. 1986. Knowledge of Language: Its Nature, Origins and Use. New York: Praeger. . 1989. "Some Notes on the Economy of Derivations and Representations," MIT Working Papers in Linguistics 10:43—74. den Dikken, M. 1990. "Verb Incoporation in French Causative Constructions," ms, Leiden. Giorgi, A. and G, Longobardi. 1991. The Syntax of Noun Phrases: Configuration, Parameters and Empty Categories. Cambridge: Cambridge University Press.

ROMANCE AND ARBERESH CAUSATIVES

237

Guasti, M.T. 1991 "Thefaire-par construction in Romance and in Germanic," Proceedings of WCCFL 9:205-218. . 1992a. "Incorporation, Excorporation and Lexical Properties of Causatives Heads," Linguistic Review 8:209-232. . 1992b. "Causative and Perception Verbs," Ph.D., Universite de Geneve. . 1993. Causative and Perception Verbs: A Comparative Study. Turin: Rosenberg and Sellier. Gueron, J. and T. Hoekstra. 1988. "T-Chains and the Constituent Structure of Auxiliary Verbs," in A. Cardinaletti, G. Cinque, and G. Giusti (eds.), Proceedings of the GLOW Conference in Venice. Dordrecht: Foris. Hale, K. and S. Keyser. 1991. "On Arguments Structure and the Lexical Expression of Syntactic Relations," ms, MIT. Higginbotham, J. 1985. "On Semantics," Linguistic Inquiry 16:547-593. Hoyt, K.E. 1989. "Verb Raising in Lebanese Arabic," Working Papers in Linguistics 11:77104. Kayne, R.S. 1977. Syntaxe du franc, ais. Paris: Edition du Seuil. . 1989. "Null Subjects and Clitic Climbing," in O. Jaeggli and K. Safir (eds.), The Null Subject Parameter. Dordrecht: Kluwer. 239-261. . 1991. "Romance Clitics, Verb Movement, and PRO," Linguistic Inquiry 22:647-686. Kayne, R.S. and J.-Y. Pollock. 1978. "Stylistic Inversion, Successive Cyclitity, Move NP in French," Linguistic Inquiry 9:595-621. Koopman, H. and D. Sportiche. 1991. "The Position of Subjects," Lingua 85:211-258. Larson, R. 1988. "On the Double Object Construction," Linguistic Inquiry 19:335-391. Li, Y. 1990. "X°-Binding and Verb Incorporation," Linguistic Inquiry 21:399-426. Motapanyane, V. 1991. "Theoretical Implications of Complementation in Romanian", Ph.D., Universite de Geneve. Pollock, J.-Y. 1986. "Sur la syntaxe de EN et le parametre du sujet mil," 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. Quicoli, A.C. 1976. "Conditions on Quantifiers Movement in French," Linguistic Inquiry 7:583-607. Rivero, M.-L. 1990. "The Location of Non-Active Voice in Albanian and Modern Greek," Linguistic Inquiry 21:135-146. . 1991. "Long Head-Movement and Negation," paper presented at the GLOW Colloquium, April, 1991, Leiden. Rizzi, L. 1990. Relativized Minimality. Cambridge, Mass.: MIT Press. Rizzi, L. and I. Roberts. 1989. "Complex Inversion in French," Probus 1:1-30. [Also published in this volume, 91-116.] Roberts, I. 1987. The Representation of Implicit and Dethematized Subjects. Dordrecht: Foris. . 1992. "Wackernagel meets the Extended Projection Principle," paper presented at the Incontro di Grammatica Generativa, February, 1992, Ferrara, Italy. . 1993a. "Excorporation and Minimality," Linguistic Review 22:209-218. . 1993b. Verbs and Diachronic Syntax. Dordrecht: Kluwer. Rouveret, A. and J.R. Vergnaud. 1980. "Specifying Reference to the Subject: French Causatives and Conditions on Representations," Linguistic Inquiry 11:97-202. Ruwet, N. 1972. Theorie syntaxique and syntaxe dufrangais. Paris: Edition du Seuil. Savoia, L.M. 1989. "Alcune caratteristiche del causative arberesh," Le minoranze etniche e linguistiche. Atti del II congresso internazionale. Comune di Piana. 363-420. Selkirk, E. 1982. The Syntax of Words. Cambridge, Mass.: MIT Press.

238

PARAMETERS AND FUNCTIONAL HEADS

Sportiche, D. 1988. "A Theory of Floating Quantifiers and Its Corollaries for Constituent Structure," Linguistic Inquiry 19:425-449. . 1992. "Clitics, Voice and Spec/Head Licensing," GLOW Newsletter 28:46-47. [Abstract of paper presented at the 1992 GLOW Conference.] Terzi, A. 1991. "Governed PRO and Finiteness," paper presented at ESCOL 1991, University of Maryland at Baltimore. Turano, G. 1989. "La costruzione causativa nell'Albanese di S. Nicola dell'Alto. Una applicazione della teoria Government e Binding," tesi di laurea, Firenze. Zagona, K. 1988. Verb Phrase Syntax. Dordrecht: Kluwer.

7

Hebrew Noun Phrases: Generalized Noun Raising Tal Siloni

1 Introduction One central issue in the investigation of noun phrases concerns their selectional properties, namely, their thematic representations (theta-grids). Putting Chomsky's (1970) lexicalist hypothesis in modern terms, it states that verbs and their related derived nouns in principle share the same theta-grids. The selectional properties of derived nouns are, in effect, hardly captured otherwise. Advancing the parallelism between verbs and nouns, Cinque (1980, 1981), and more recently Giorgi and Longobardi (1991) argue that theta-grids of nouns are systematically mapped into hierarchical syntactic structures, just like those of verbs. Obviously, this configurational approach is theoretically desirable, as it reduces the need for specific stipulations with regard to the syntax of noun phrases. Another central (though more recent) issue in the investigation of noun phrases concerns the syntactic status of the determiner. Due to the extension of X-bar Theory to functional categories, Infl (IP) and Comp (CP) (Chomsky 1986b), this issue has become a focus of intensive inquiry. Assuming that the determiner projects its own maximal projection (DP) as other functional categories do, linguists have been concerned with the question of what the syntactic relation is between NP and DP. Two logical possibilities emerge: 1. the noun phrase is the maximal projection of N (namely NP), DP being generated inside NP, as a sister of N', or 2. the noun phrase is the maximal projection of D (DP), which takes NP as its complement. These two alternatives are schematized in la and Ib, respectively: 239

240

PARAMETERS AND FUNCTIONAL HEADS

Traditional analyses of noun phrases naturally assigned to them the structure given in la. Recent work, however, argues in favor of the structure in Ib on the basis of quite various considerations. Abney (1987) argues for the existence of an inflectionlike element in D°, whose relation to the noun parallels that of Agr to the verb. Szabolcsi (1987, 1989) and Siloni (1990) draw an analogy between DP and CP, thus claiming that DP should head the noun phrase in the same way that CP heads the clause. Finally, Ritter (1987, 1988) and Fassi Fehri (1989) derive Semitic Construct States via head-to-head movement (Chomsky 1986b) of the noun, a derivation which crucially depends on the existence of a head position within the noun phrase to which the noun can raise. In this paper,1 I address these two issues, focusing on Hebrew noun phrases (mainly derived nominals). First, I present evidence that Hebrew noun phrases have a fully configurational internal structure. I then show that, given their configurational structure on the one hand and the order of constituents they exhibit on the other hand, it must be concluded that in Hebrew noun raising always applies. If so, there must exist an appropriate landing site to host the raised noun. I argue that this host can only be the head position D°, thus supporting the structure given in Ib. The paper is organized as follows. In section 2 I present two genitival constructions — the Construct State and the Free State — and discuss some distinctions between the two. In section 3 I analyze the internal structure of free states, arguing on the basis of evidence referring to the theory of Binding that they must have a configurational structure. This in turn leads me to conclude that noun raising must apply. Section 4 is devoted to analyzing construct states. First I argue that noun raising must be generalized. I then discuss the mechanism of genitive Case assignment utilized in this construction, emphasizing the relevance of D°, and account for the peculiar properties the construction shows. Finally, section 5 contains a discussion of the categorial structure of derived nouns. This issue deserves attention because derived nouns seem to show some verbal properties. Therefore, real conclusions with respect to the syntactic structure of noun phrases can be drawn from the behavior of derived nouns only if those are shown to be purely nominal.

2 Construct States vs. Free States In Hebrew, arguments of the noun (including possessives) can never appear prenominally. Genitival relations between a head noun and a noun phrase can be indicated in two distinct ways as illustrated below (2a-b). While in 2a an (abstract) genitive

HEBREW NOUN PHRASES

241

Case is assigned to the complement by shel 'of, in 2b shel does not appear and an (abstract) genitive case seems to be assigned by the head noun itself:2 (2) a. ha-bayit shel ha-'ish the-house of the-man 'the man's house' b. beyt ha-'ish house the-man 'the man's house' The second construction (2b) is known in the literature of Semitic languages as the Construct State. The head noun is said to be in the construct state: It loses stress (the main stress always falls on the complement) and it is therefore subject to phonological rules which operate in non-stressed environments.3 Hence the alternation between bayit in 2a (henceforth the free state) and beyt in the construct state in 2b. Derived nouns allow these two constructions as well: the free state (3a) and the construct state (3b): (3) a. ha-hitgalut shel ha-saxkan trie-appearance of the-actor 'the actor's appearance' b. hitgalut ha-saxkan appearance the-actor 'the actor's appearance' There have been different attempts to define a systematic semantic difference between the two constructions. Thus, for instance, Rosen (1957, adapted by Doron 1989) claims that the construct state in 2b implies inalienable possession, whereas its free state counterpart 2a does not. Herman (1978), on the other hand, reports that this observation does not seem to accord with the intuitions of the native informants she interrogated. As my intuitions do not reflect any systematic difference either between 2a and 2b or between 3a and 3b, the two constructions will be treated in this paper as essentially synonymous. There are, however, two important structural distinctions between free states and construct states, which are worthy of discussion: One distinction concerns the article, and the other the position of modifying adjectives (Herman 1978; Borer 1984, among others).4

2.1

The Definite Article

Hebrew has a definite article only (ha-), which is a prefix and does not inflect. Note that the head noun of 2b, for example, is interpreted as definite, just like that of 2a, as is clear from the glosses, though the definite article does not accompany its head. In fact, the article can never be attached to the head of a construct state; it results in ungrammaticality (4a). The [idefinite] value of the head is determined by that of its complement. A definite complement renders the head definite and an

242

PARAMETERS AND FUNCTIONAL HEADS

indefinite complement renders it indefinite, as shown, for instance, by the behavior of the accusative marker '",t. This marker appears exclusively with definite nouns, as illustrated in 4b-c. It can therefore serve as a reliable test distinguishing between definites and indefinites. Thus, when the head noun is assigned accusative Case, it is obligatorily accompanied by 'et (hence definite), if the complement is definite, as in 4d. If the complement is indefinite, 'et cannot appear (that is, the head noun is indefinite), as in 4e: (4) a. (*ha-)sifrey ha-meshorerim (the-)books the-poets 'the poets' books' b. Hu kone 'et te-sfarim bezol. he buys ACC the-books cheaply 'He buys the books cheaply.' c. Hu kone (*'et) sfarim bezol. he buys (ACC) books cheaply 'He buys books cheaply.' d. Hu kone 'et sifrey /uz-meshorerim ha-ce'irim. he buys ACC books the-poets the-young 'He buys the young poets' books.' e. Hu kone (*'et) sifrey meshorerim ce'irim. he buys (ACC) books poets young 'He buys books by young poets.' Moreover, if the complement of a construct state becomes the head of another construct state, its article can no more appear. In a string of two (or more) construct states, only the right most noun can carry the article (5). The (in)definiteness of the others depends on that of this last constituent: (5) gag (*ha-)beytha-'ish roof (the-)house the-man 'the roof of the house of the man'

2.2

The Position of the Adjective

Adjectives in Hebrew follow the noun they modify and agree with it in number, gender and (in)definiteness, as illustrated in 6: (6) a. ha-bayit ha-gadol the-house(MASC.SG)the-big(MASC.SG) 'the big house' b. biktotgdolot hut(FEM.PL) big(FEM.PL) 'big huts'

HEBREW NOUN PHRASES

243

Mostly, the adjective must immediately follow the noun it modifies. Hence a PP cannot intervene between a noun and its modifying adjective (la), nor can the shel 'of phrase in the free state intervene between the two (7b), rather it must follow the adjective (7c): (7) a. *ha-bayit'im ha-'arubaha-yafe the-house with the-chimney the-beautiful 'the beautiful house with the chimney' b. *ha-bayit shel ha-'isha ha-gadol the-house of the-woman the-big 'the woman's big house' c.

ha-bayit ha-gadol shel ha-'isha the-house the-big of the-woman 'the woman's big house'

In this respect as well, construct states constitute an exception: a modifying adjective cannot immediately follow a head in the construct state (8a). On the contrary, it must follow the genitival complement, that is, it must appear at the end of the construct state complex (8b): (8) a. *beyt (ha-)gadol ha-'isha house (the-)big the-woman 'the woman's big house' b. bey t ha-' isha ha-gadol house the-woman the-big 'the woman's big house' Naturally, an adjective appearing at the end of the complex can refer to both the head and the complement, providing that these two agree in number and gender; recall that they obligatorily share the same [± definite] value. Thus, 9 is ambiguous between the two interpretations (i and ii): (9) a. bey t ha-'ish ha-gadol house(MASC.SG) the-man(MASC.SG) the-big(MASC.SG) i. 'the big man's house' ii. 'the man's big house' Notice that the adjective is definite whether it modifies the complement or the head. This supplies another piece of evidence that the (in)definiteness of the head of the construct state is determined by that of its complement. To summarize this section, two distinct genitival constructions were presented: the free state and the construct state. The latter shows some peculiar properties which will be dealt with in section 4. I first turn to the analysis of the internal structure of the free state.

244

PARAMETERS AND FUNCTIONAL HEADS

3 Free States: Internal Structure Recall that the Hebrew article is a prefix attached to the noun. It follows that either the article cliticizes onto the noun, or the noun undergoes head-to-head movement and incorporates (in Baker's 1988 sense) with the article. I now show that syntactic evidence strongly favors the second possibility. Consider derived nouns, specifically those expressing a process, an event; in principle, they inherit the theta-grid of their associated verbs.5 Thus, for instance, the derived noun harisa 'destruction' assigns in lOa the same theta-roles assigned by the corresponding verb haras 'destroy' in lOb. In both constructions, ha-'ir 'the city' receives a Patient theta-role, while ha-cava 'the army' receives an Agent thetarole. The Agent argument is assigned nominative Case in the verbal construction and genitive Case in the nominal construction, whereas the Patient argument is assigned accusative Case in both constructions (accusative Case assignment in the noun phrase is discussed in section 5): (10) a. ha-harisa she! ha-cava 'et ha-'ir the-destruction of the-army ACC the-city 'the army's destruction of the city' b. Ha-cava haras'et ha-'ir. the-army destroyed ACC the city 'The army destroyed the city.' Verbal theta-grids are projected hierarchically (through certain mapping principles) onto syntactic structures. Thus, for instance, the Agent is always structurally more prominent than the Theme or the Patient. In light of that, the question arises whether nominal theta-grids are projected as well onto hierarchical syntactic structures. Obviously, a positive answer constitutes the null hypothesis here, since it enables the theory to generalize principles across categories. In order to shed light on the syntactic structure of noun phrases, I will utilize syntactic processes which refer to structural information. Prior to that, however, a note on word order is in place. In lOa the Agent immediately follows the noun. In fact, this is the only possible word order. The Agent (in other terms, the subject) can never be preceded by the Patient (the object) (1 la). This is reminiscent of the word order in Hebrew clauses attesting a post-verbal subject: the latter must immediately follow the verb (lib) and cannot be preceded by the object (lie): (11) a. *ha-harisa'et ha-'ir shel ha-cava the-destruction ACC the-city of the-army 'the army's destruction of the city' b.

'Etmol haras ha-cava 'et ha-'ir. yesterday destroyed the-army ACC the-city 'Yesterday the army destroyed the city.'

c. *'Etmol haras 'et ha-'ir ha-cava. yesterday destroyed ACC the-city the-army 'Yesterday the army destroyed the city.'

HEBREW NOUN PHRASES

245

This by itself already seems to reveal at least some structural similarity between noun phrases and their verbal counterparts.

3.1 Evidence from the Binding Theory The data discussed so far do not offer any crucial evidence regarding the internal syntactic structure of noun phrases. In principle, given general postulates of X-bar Theory, three possible structures can surface as 10a:6 1. a flat tripartite structure (12a) 2. a structure where the Patient is more prominent than the Agent (12b), or 3. a structure where the Agent is more prominent than the Patient (12c), as the null hypothesis would predict. The third structure, however, can surface as lOa only if the noun undergoes obligatory leftward movement in the course of the derivation, Agent lowering being excluded on theoretical grounds:

As far as the c-command relationship is concerned, structure 12a predicts that both arguments ought to c-command one another. Structures 12b-c, on the contrary, dictate an asymmetrical c-command relationship between the two arguments (assuming the original definition of Reinhart 1976): Either only the Patient c-commands the Agent (12b) or, vice versa, only the Agent c-commands the Patient (12c). As the Binding Theory crucially utilizes the c-command relationship, it offers a robust testing field for the issue at stake, as already pointed out by Giorgi and Longobardi (1991). Relying on their work, I argue for the structure in 12c on the basis of arguments referring to the Binding Theory.7 Principle A of the Binding Theory requires that anaphors be bound in their governing category. Before examining their distribution within the noun phrase, it should be pointed out that neither shel nor 'et block c-command of the argument they introduce toward another argument (as already pointed out by Borer 1984). This is illustrated below: In 13a the anaphor is bound (hence c-commanded) by the argument introduced by shel, and in 13b by the argument introduced by 'et:

246

PARAMETERS AND FUNCTIONAL HEADS

(13) a. ha-harisashel ha-cavaj 'et ' acmo the-destruction of the-army ACC itself 'the army's destruction of itself b. Hu her' a 'et ha-tinokj le-'acmoj ba-mar'a. he showed ACC the-baby to-himself in+the-mirror 'He showed the baby to himself in the mirror.' As is already clear from 1 3a, the Patient can be an anaphor. If so, it must be bound in its governing category. The only antecedent available is the Agent, which must therefore c-command the Patient. Hence, the options in 12 can already be narrowed down: 12b cannot be the right structure, because the Agent would not c-command the Patient. We are left with two options. Principle C of the Binding Theory requires that R-expressions be free. The flat structure in 12a then erroneously predicts a principle C violation in cases such as 13a. However, it has been argued that when an anaphor and its antecedent mutually c-command, no disjointness effect arises (Borer 1984; Giorgi and Longobardi 1991). Nonetheless, 12a must be dismissed since, contrary to 12c, it predicts that the Patient ought to c-command the Agent. This prediction is wrong, as shown in 14: (14) *ha-harisa shel 'acmo 'etha-cava the-destruction of itself ACC the-army Thus, the Agent c-commands the Patient, but not vice versa. Put differently, the Agent (the subject) must be structurally more prominent than the Patient (the object). Example 1 2c then is the only structure compatible with the data. The same conclusion can be drawn on the basis of other syntactic processes referring to c-command. One such process is binding of a pronoun by a quantified noun phrase. The bound pronoun must be c-commanded by the quantified noun phrase to prevent a Weak Crossover Violation (Reinhart 1976; Koopman and Sportiche 1982). As expected, the theme (the object) can function as the bound pronoun (15a), whereas the Agent (the subject) cannot (15b), for the latter c-commands the former, but not vice versa: (15) a. ha-te'ur shel kol ha-nashinij 'et ba'aley-henj the-description of all the-women ACC husbands-their 'all the women's description of their husbands' b. *ha-te'ur shel ba'aley-henj 'et kol ha-nashim; the-description of husbands-their ACC all the-women Summing up, in this section it has been shown, on the basis of empirical evidence related to the Binding Theory, that the Agent must occupy a higher structural position than the Theme or the Patient within the noun phrase. This singles out 12c as the only possible structure. This conclusion provides further support for the Thematic Correspondence Hypothesis (Giorgi and Longobardi 1991) which states that verbs and corresponding nouns identify the same theta-role as the role assigned to the subject position, available outside V in VPs (Fukui and Speas 1986; Kuroda 1988; Sportiche 1988) and outside N' in NPs (the external theta-role in Williams' 1981

HEBREW NOUN PHRASES

247

terms); the other theta-roles are assigned internally, within V and N', respectively (hence internal theta-roles). This hypothesis suggests a principled account of salient similarities between noun phrases and clauses, thus discarding construction-specific stipulations. Moreover, it is congruous with a restrictive (therefore preferable) view of grammatical processes which allows reference to structural information only (see Belletti and Rizzi 1988).

3.2

The Structure

For the sake of concreteness, consider lOa again (repeated here as 16): (16) ha-harisa shel ha-cava 'et ha-'ir the-destruction of the-army ACC the-city 'the army's destruction of the city' The external theta-role, ha-cava is generated as a sister of N' whereas the internal theta-role, ha-'ir 'the city' is generated within N'. The surface word order must be derived via leftward movement of the noun. As movement can apply to ccommanding positions only, the landing site must c-command the head position N°. If noun phrases were projections of N° (la), such a position would not be available, there being no head position higher than N° within the noun phrase. On the other hand, if noun phrases are DPs (Ib), then D° immediately emerges as a plausible landing site. Moreover, given the affixal nature of the Hebrew (definite) article, it constitutes a natural host, just like other affixes, e.g., tense or agreement. Thus, empirical evidence strongly supports the hypothesis that noun phrases are projections of the functional category D:

Note that noun raising to D° must apply whether the noun is definite or indefinite, the data presented in the previous section being valid regardless of the (in)definite nature of the noun. The fact that the Hebrew indefinite article is phonetically null is irrelevant to the operation in question in the same way that the overt/covert nature of tense is irrelevant to verb raising. 16 then has the following S-structure representation: (18) [DP [D ha-harisa;] [Np shel ha-cava [N, ts 'et ha-'ir] ] ] the-destruction of the-army ACC the-city

248

PARAMETERS AND FUNCTIONAL HEADS

The movement the noun undergoes is local, as it does not skip any head position. Thus, it does not seem to pose any problem for current theoretical approaches to movement. Notice that the raised noun must be able to govern across NP, otherwise it would not be able to govern its own trace, thus violating the Empty Category Principle.8 VSO surface order (for instance 1 Ib) has often been analyzed along similar lines (Sproat 1985; Fassi Fehri 1989, among others), as exemplified in 19:9 (19) . . . [Ip [j haras;] [ vp ha-cava [v/ 1; 'et ha-'ir] ] ] destroyed the-army ACC the-city Under the present approach, the word order of both 18 and 19 can be accounted for in a simple unified way. As the subject of both is generated in the specifier position of NP and VP, respectively (see references cited above), the fact that the object can intervene neither between the verb and its subject (as in lie) nor between the noun and its subject (as in 1 1 a) follows automatically from the location of specifiers in Hebrew. Consider now modifying adjectives. Recall that they appear immediately to the right of the noun, as illustrated again in 20: (20) ha-harisa ha-mehira shel ha-cava 'et ha-'ir the-destruction the-rapid of the-army ACC the-city 'the army's rapid destruction of the city' Consequently, they can only be base-generated in a position adjoined to NP so that they will follow the noun after it raises to D°, as demonstrated in 21 : (21) [DP [D ha-harisaj] [Np ha-mehira [Np shel ha-cava [N/ 1; . . . the-destruction the-rapid of the-army This suggestion does not entail any special assumptions. On the contrary, it is straightforward, if Chomsky (1986b) is correct in arguing that maximal projections can adjoin exclusively to maximal projections. If the noun were to remain in its base-position, we would have to stipulate (ad hoc) that modifying adjectives were generated in a position (right-) adjoined to N'. Under the approach advocated here, no such additional stipulation is needed. Furthermore, note that the fact that English manifests prenominal adjectives whereas Hebrew manifests postnominal adjectives now simply stems from the fact that in English noun raising cannot apply while in Hebrew it must apply. Again, if the noun remained in its base-position in Hebrew, we would have to assume that these two languages differed essentially with regard to the positioning of modifying adjectives.10 Finally, it should be pointed out that though noun raising was motivated here on the basis of evidence offered by derived nouns, it must be generalized to concrete nouns as well. First, it is hardly plausible to suppose that the same affix (namely the article) triggers movement of derived nouns only. Second, as modifying adjectives follow both derived and concrete nouns, a true generalization concerning their positioning can be reached only if nouns, whether derived or concrete, must raise to D°.

HEBREW NOUN PHRASES

249

In conclusion, arguments referring to the Binding Theory justify a configurational approach to Hebrew noun phrases, which captures in a natural way important generalizations across categories. Word order can then be accounted for solely in terms of noun raising, which in turn implies that noun phrases cannot be simply projections ofN°.

4

Construct States: Syntactic Analysis

Turning our attention to derived nouns occurring in construct states, we would obviously expect them to also reflect the insight expressed in the Thematic Correspondence Hypothesis. In fact, the Uniformity of Theta-Assignment Hypothesis (UTAH) put forth by Baker (1988) explicitly requires that thematic paraphrases have parallel Dstructures. Hence, 16 and its construct state paraphrase, 22a (in which the internal argument is assigned accusative Case as in 16, while the external argument is in the construct state) should share the same D-structure, the external argument being generated in the specifier position of NP and the internal argument inside N'. Consequently, surface word order should be derived via noun raising (22b), as in the case of free states: (22) a. harisat ha-cava 'et ha-'ir destruction the-army ACC the-city 'the army's destruction of the city' b. [DP [D harisatj] [Np ha-cava [N/ tj 'et ha-'ir] ] ] destruction the-army ACC the-city If that is indeed correct, we predict that construct states ought to show the same subject-object asymmetry manifested in free states. The prediction is born out: 23a is grammatical because an anaphor is licensed in object position while 23b is ruled out because it is not licensed in subject position. Similarly, 24a is grammatical because a quantified noun phrase in subject position can bind a pronoun in object position, but not vice versa (24b): (23) a. harisat ha-cava 'et 'acmo destruction the-army ACC itself 'the army's destruction of itself b. *harisat 'acmo 'et ha-cava destruction itself ACC the-army (24) a.

te'urkolha-nashinij'etba'aley-henj description all the-women ACC husbands-their 'all the women's description of their husbands' b. *te'ur ba'aley-hent 'et kol ha-nashim; description husbands-their ACC all the-women

Having established this, I turn to a more complete analysis of construct states. Three issues deserve attention: 1. the mechanism of Case assignment

250

PARAMETERS AND FUNCTIONAL HEADS

2. the irregular location of modifying adjectives 3. the impossibility to render the head noun (in)definite directly An important claim throughout this section is that genitive Case can be assigned from D° (as first suggested by Abney 1987). Part of the section is therefore devoted to justifying this claim.

4.1

Case Assignment and Adjacency

Abney (1987) suggests accommodating in D° an Agreement feature (or set of features). This Agr feature is coindexed with the possessor to which it assigns Case in the same way that Agr in 1° assigns Case to the subject of the clause. The core of his argumentation lies with languages in which the noun agrees with its possessor in the same way that the verb agrees with its subject (Hungarian, Turkish, and Yupik, a central Alaskan Eskimo language). Abney claims that these parallel phenomena must receive a parallel syntactic treatment which will capture the generalization. He therefore suggests attributing to clauses and noun phrases the following parallel structures:

Note that the position of the possessor at S-structure can in fact vary across or within languages, between the specifier position of DP and that of NP, just as the position of the subject at S-structure can vary between the specifier position of IP and its base-position, the specifier of VP (see 19). If this hypothesis is on the right track, we would expect, in principle, that certain grammars could realize the option of having abstract Agr in D° in the same way that some grammars have abstract Agr in 1°. As far as construct states are concerned, Abney's hypothesis, albeit based on different data, is strikingly explanatory, as pointed out by Ritter (1987, 1988) and Fassi Fehri (1989). Following them, I argue that in construct states the noun raises to D° and incorporates with Agr which happens to be abstract. Agr, once supported morphologically, assigns genitive Case to the argument in the specifier position of NP, as schematized in 26." Case assignment takes place here under government. According to the notion of government advocated in Chomsky (1986a), if a governs XP, it also governs the specifier position and the head of XP. D° clearly governs NP. Moreover, government from D° across NP must be possible, otherwise the raised noun would not govern

HEBREW NOUN PHRASES

251

its own trace, as already noted above. If D° governs into NP, it follows that it also governs the specifier position of NP. Hazout (1988) notes that the configuration in 26 is reminiscent of the Exceptional Case Marking (ECM) configuration. Since nouns are known to be unable to govern in that environment, he concludes that Case assignment in 26 would be problematic. Crucially, however, it is the functional head D° (containing the complex Agr+N) which governs (and assigns Case to) the complement in the specifier position of NP, and not the noun itself, thus rendering the ability of nouns to govern across a maximal projection irrelevant. Moreover, in 26 the Case assigner is thematically related to the assignee due to noun raising, in much the same way that 1° is thematically related to the nominative NP due to verb raising (see, for example, 19). In ECM configurations, on the contrary, the case-assigner is always thematically distinct from the assignee. Thus, the configuration in 26 and that of ECM are only superficially similar. If for some reason (to be discussed shortly) genitive Case cannot be assigned by Agr, it is assigned to the argument in the specifier position by shel, resulting in the free state. Consider modifying adjectives next. In section 3 it was shown that they are basegenerated in a position adjoined to NP. Consequently, we would predict that they ought to appear between the head of the construct state and its genitival complement, while in fact, they can never appear between these two. However, this may stem from a requirement that the Case assigner and the assignee be adjacent. Put differently, an adjective intervening between the two blocks Case assignment. To rescue the structure then, the complement itself must adjoin to NP higher than the adjective, yielding the order: Noun-Complement-Adjective. Strict adjacency between the Case assigner and the assignee is reported elsewhere in the literature as a requirement imposed on Case assignment under government (see, for instance, Belletti 1990, McCloskey 1990, Friedemann 1992, Haegeman 1992, Rizzi, this volume).12 Note again that though noun raising of concrete nouns cannot be motivated independently, it is hardly plausible to argue that their construct states are derived differently, since both construct states (namely of derived as well as concrete nouns) show the same syntactic properties. It remains to be considered why the head of the construct state cannot be rendered (in)definite directly.

252

4.2

PARAMETERS AND FUNCTIONAL HEADS

The "Article Constraint"

Recall that the head of the construct state cannot be rendered (in)definite directly, as was illustrated in 4a (repeated here as 27): (27) (*ha-)sifrey ha-meshorerim (the-)books the-poets 'the poets' books' It has been argued that the article preceding the complement in this construction is, in effect, the article of the head noun (Ritter 1987, 1988). This article has cliticized down onto the complement which derives from the fact that it cannot precede the head. This approach however entails several difficulties. First, it does not explain why the article cannot be attached to the head as usual but rather, must cliticize onto another constituent. Secondly, it predicts erroneously that sequences of two articles should be possible, one article being that of the head and the other that of the complement. Moreover, it predicts erroneously that in a string of two construct states, an article could cliticize onto the intermediate noun (which is both the complement of the first construct state and the head of the second). In fact, however, an article can only surface on the right most noun (see 5). Notice that this strongly suggests that the article of a head in the construct state in principle cannot occur. Finally, this approach does not offer any explanation for the obligatory agreement in (in)definiteness between the head of the construct state and its complement. Evidence that they must agree was presented above on the basis of the behavior of the accusative marker 'et (section 3.1) as well as the behavior of adjectives modifying the head (section 3.2). Thus, genitival relations between elements that do not share the same [idefinite] value cannot be expressed in the construct state (28a), and so shel must appear (28b): (28) a. beytha-'ish house the-man i. *'a house of the man' ii. 'the man's house' b. bayit shel ha-'ish house of the-man 'a house of the man' It seems then that the article constraint cannot simply be reduced to an obligatory rightward cliticization. Note that this constraint is reminiscent of the complementary distribution of possessors and articles in English (29): (29) *John's the book Under traditional analyses of noun phrases (la), both the possessor (John in 29) and the article compete for the same position, the specifier of NP; hence their complementary distribution. Under the more articulated analysis of noun phrases, the possessor in 29 appears in the specifier position of DP, while the article is generated in D°. In consequence, their complementary distribution cannot follow any more

HEBREW NOUN PHRASES

253

from their competition for the same position. To account for this complementarity, Abney (1987) suggests a constraint along the lines of 30: (30) Agr in D° cannot co-occur with a lexical article. Thus, possessors and articles cannot co-occur in English, because in this language Agr in D° cannot co-occur with an article. If the article is realized in D°, Agr cannot be realized in D° and the possessor cannot be assigned Case. On the other hand, if Agr is realized in D°, an article cannot be realized. I propose that the article constraint in Hebrew is another instance of the constraint stated in 30 (for a similar suggestion, see Fassi Fehri 1989). Thus, the head of the construct state cannot be directly marked definite, because a lexical article is incompatible with Agr in D°. If the article occurs, Agr cannot be realized and genitive Case cannot be assigned from D°; so shel must appear. On the other hand, if Agr is realized, genitive Case can be assigned from D°, but the article cannot occur; consequently, the noun cannot be specified in the usual way with regard to its [idefinite] value. This lack of specification is even more evident in standard Arabic, which has both a definite and an indefinite article; the former is a prefix (3la) while the latter is a suffix (31b). In the construct state, none of them can appear with the head noun (31c-d), as expected if 30 holds: (31) a. 'a/-bayt-u the-house-NOM 'the house' b. bayt-u-n house-NOM-a 'a house' c. bayt-u r-rajul-i house-NOM the-man-GEN 'the man's house' d. bayt-u rajul-i-n house-NOM man-GEN-a 'a man's house' To conclude, in construct states the noun is not specified in the usual way for (in)definiteness, since its D° cannot realize this specification. However it does not remain underspecified; rather, it must agree with the [idefinite] value (of D°) of its complement. Instances of agreement between a Case assigner and its assignee are of course natural, even in a structural configuration of government. Thus, for instance in Hebrew, a verb governing its postverbal subject shows agreement with it; or in Italian, the past participle in the so-called Absolute Construction agrees with the DP it governs (see Belletti 1990).13 Moreover, note that agreement in (in)definiteness is not a peculiarity of construct states, adjectives in Semitic always agree in (in)definiteness with the nouns they modify (6a-b), thus supporting the claim that in(definiteness) in Semitic is a feature (of D°) in which constituents can agree.14 As the claim that genitive Case can be assigned from D° is essential to the account suggested here, it might be useful to shed some additional light on this point.

254

PARAMETERS AND FUNCTIONAL HEADS

4.3 More on the Relation between D° andAgr I argue that Agr associated with genitive Case is a feature of D° for several reasons. First, neither under the hypothesis that genitive Case is assigned directly by the noun, nor under the hypothesis that it is assigned by a separate Agr category (along lines proposed in Szabolcsi 1987, 1989), is it clear how to derive the complementarity of articles and possessors in English, and in Semitic construct states. This complementarity was neatly derived under the present approach through the constraint in 30. It should be remarked that 30 is certainly not a unique type of constraint in grammar. It has often been argued that different instantiations of the same functional category cannot co-occur. Thus, for instance, in English the infinitival to as well as modals preclude Agr in 1°; and according to Rizzi (1990), in standard English the complementizer that is incompatible with (abstract) Agr in C°.15 Another empirical confirmation of the approach advocated here is suggested by Somali (a Cushitic language). Notice first that saying that the two features of D° (±definite and Agr) cannot co-occur in some languages is not equivalent to saying that they can never co-occur. In principle, then, we expect their co-occurrence to be possible in some languages. In Somali this seems to be the case: idefinite and Agr can co-occur. Moreover, when Agr is morphologically realized (under a certain configuration, to be discussed shortly), it is the article itself that bears the Agr morpheme, as expected if both are features of the same head. Consider first 32, which resembles the Semitic construct state, as pointed out by Lecarme (1989): The genitival complement buii- go. 'book-the' appears on the right, and the two members of the construction must be adjacent. However, contrary to construct states, the article of the head noun can appear: (32) tarjama-da buug-ga translation-the(FEM)book-the(MASC) 'the book's translation'

(Lecarme 1989:2, ex. 1)

If the present analysis is correct, it should be concluded that in Somali the two features, ±definite and Agr, the genitive Case-assigner, can co-occur in D°. This conclusion is strongly supported by the following data. In Somali the genitival complement can also appear prenominally, in Spec-head configuration with the head noun. Interestingly, in that case, a special agreeing form of the article must appear bearing the grammatical features of the possessor (33): (33) buug-ga tarjama-rffwa (Lecarme 1989:2, ex. 2) book-the(MASC)translation-the(FEM)+3MASC.SG 'the book's translation' Thus, whereas Semitic and English suggest that [idefinite] and Agr are features of the same head (D°) on the basis of their incompatibility, Somali further supports this claim, for it is precisely its article which actually agrees with the possessor.16 In conclusion, D° was characterized here by two features, idefinite and Agr, which in some languages (Semitic, English) preclude each other. In Hebrew, D° always triggers noun raising; put differently, D° always has the morphological subcategorization frame [ N], whether it realizes its [idefinite] feature or its Agr feature

HEBREW NOUN PHRASES

255

(see Lieber 1980; Rizzi and Roberts 1990, on morphological subcategorization). The properties of heads are by hypothesis subject to parametric variation. Thus, it may be that in Romance, where genitive Case is assigned only via dummy prepositions (setting aside possessives), Agr is not a feature of D°; or in Hungarian, where Agr is morphologically realized but does not assign genitive Case (see Szabolcsi 1987, 1989), it is not a feature of D° either.

5

Nouns—Not Verbs

A notorious puzzle raised by Hebrew (derived) process nominals is their apparent ability to assign accusative Case to their Theme argument, an anomalous behavior as far as nouns are concerned. Cross-linguistically, accusative Case assignment is by and large a verbal property. Moreover, Hebrew process nominals show another arguably verbal property, namely, they can be modified by adverbs. This is illustrated in 34: (34) harisat ha-cava 'et ha-'ir bi-mehirut destruction the-army ACC the-city in-quickness (= quickly) 'the army's quick destruction of the city' This may call into question the categorial structure of process nominals. On the one hand, they are clearly nominal: They have nominal distribution and bear a [idefinite] value, they can be modified by adjectives and license the dummy preposition shel, strictly limited to nominal contexts. On the other hand, they show some apparently verbal properties: They can be modified by adverbs and their direct objects can receive accusative Case. Do process nominals really have an ambiguous nature, much like gerunds, which should be echoed in syntactic structure? It has been argued (Hazout 1988, 1990; Borer 1990) that the syntactic representation of these nominals contains a verbal projection whose head V° undergoes nominalization in the course of the derivation. Obviously, the issue is of particular importance with respect to topics discussed in this paper. If the hypothesis that process nominals contain a verbal projection (call it the VP-hypothesis) were on the right track, the relative order of arguments within the noun phrase could be derived via leftward movement of the verb, thus rendering a generalized analysis in terms of noun raising unmotivated. In this section, I argue that process nominals are pure nouns. I first prove that they are not really modifiable by adverbs. I then show that there are important reasons to believe that accusative Case assigned by nouns is not the ordinary accusative of transitive verbs. Finally, I contrast process nominals with other nominal constructions which do have dual categorial structure.

5.7 Adverbs vs. PPs As mentioned above, process nominals seem to be modifiable by adverbs, which is apparently a verbal property. It may be suggested that adverbs in noun phrases are licensed by the process, the event. This explanation is plausible, as shown by the fact that adverbs are never allowed when the noun is a result nominal. Take the noun

256

PARAMETERS AND FUNCTIONAL HEADS

re'ayon 'interview', which can only denote a result, as demonstrated by the fact that the addition of the modifier tadir 'frequent', which is limited to process nominals (see Grimshaw 1990), results in ungrammatically (35a). As predicted, this noun also cannot be modified by adverbs (35b): (35) a. *ha~re'ayonha-tadir 'im ha-nasi the-interview the-frequent with the-president 'the frequent interview with the president' b. *ha-re'ayon 'im ha-nasi bi-mehirut the-interview with the-president in-quickness (= quickly) 'the quick interview with the president' However, this explanation is not adequate. If an event sufficed to license adverbs, we would expect them also to be admitted with English process nominals (36): (36) *the army's destruction of the city quickly I would like to suggest that it is the particular categorial status of the adverbs which enables them to appear in Hebrew noun phrases. Most adverbs in Hebrew are, in fact, adverbial PPs; the group of "real" adverbs (of the English type) is limited and lexically marked. As shown in 37, a real adverb cannot appear in Hebrew noun phrases in the same way that it cannot appear in the English counterparts: (37) a. *harisatha-'ir maher destruction the-city quickly 'the quick destruction of the city' b. *slilatha-kvishle'at paving the-road slowly 'the slow paving of the road' If process nominals contained a verbal projection, as suggested by the VPhypothesis, this bifurcation would be rather mysterious. But if process nouns are purely nominal, as argued here, it follows automatically that adverbial PPs are allowed while genuine adverbs are not. I now turn to the other arguably verbal characteristic of process nominals, i.e., their apparent ability to assign accusative Case. I will show that there are significant distinctions between accusative Case assigned by verbs and that found in nominal contexts.

5.2

The Accusative Case

The most salient distinction between accusative Case assignment by verbs and by nouns concerns the particle 'et. As already mentioned above, when verbs assign accusative Case to definite nouns, 'et must appear (4b); when they assign accusative Case to indefinite nouns, 'et cannot appear (4c). Interestingly, nouns can assign accusative Case only in the presence of 'et (38a) (Borer 1984); hence, the ungrammaticality of 38b:17

HEBREW NOUN PHRASES

(38) a.

257

harisat ha-cava 'etha-'ir destruction the-army ACC the-city 'the army's destruction of the city'

b. *harisat ha-cava'ir destruction the-army city 'the army's destruction of a city' As there is no reason to assume that a definiteness requirement is imposed on objects of derived nouns, it appears that 'et plays a crucial role in the assignment of this Case in nominal contexts. This would be unexpected, if in nominal contexts as well, accusative Case were assigned by V°. Second, nouns, unlike verbs, are unable to assign accusative Case to subjects of small clauses:18 (39) a. Ha-moremaca'etha-ti'unmeshaxne'a. the-teacher found ACC the-argument convincing The teacher found the argument convincing.' b. *meci'at ha-more 'et ha-ti'un meshaxne'a finding the-teacher ACC the-argument convincing c. Ha-me'amen ra'a 'et ha-yeled nofel. the-coach saw ACC the-boy falling 'The coach saw the boy falling.' d. *re'iyat ha-me'amen 'et ha-yeled nofel view the-coach ACC the-boy falling Again, if accusative Case were assigned by V° in nominal contexts, as suggested by the VP-hypothesis, this behavior would be rather mysterious. Furthermore, the discussed inability is known to be typical of nouns, either because they cannot govern across distinct maximal categories (see n. 8), or because the Case they assign is an inherent Case, which is thematically related (Chomsky 1986a). The third distinction refers to Burzio's (1986) generalization, which states that accusative Case is assigned to the object if and only if a theta-role is assigned to the subject. At first glance, it seems that derived nouns also abide by this generalization. In the three examples of 40, only the internal argument is realized. While the structure is grammatical, when it receives genitive Case (40a-b), it is excluded, when it receives accusative Case (40c): (40) a. ha-harisashelha-'ir the-destruction of the-city 'the city's destruction' b. harisat ha-'ir the-destruction the-city 'the city's destruction' c. *ha-harisa 'et ha-'ir the-destruction ACC the-city 'the city's destruction'

25 8

PARAMETERS AND FUNCTIONAL HEADS

A reasonable interpretation of these data would be that accusative Case cannot be assigned since the external theta-role is not discharged, the subject being unrealized (Ritter 1988). This, however, presupposes that the understood subject cannot be structurally realized as a null element. Is that indeed so? I now show precisely the opposite.19 Consider first the distribution of anaphors and pronouns, as stated by principles A and B of the Binding Theory: whereas anaphors must be bound in a local domain, pronouns must be free in that domain, which is the domain of the closest c-commanding subject. Consider now the examples in 41: (41) a.

Hiij nifga me-horadat-Oj be-darga. he was+offended from-lowering-his in-rank 'He was offended by his demotion.'

b. *Huj nifga me-horadat acmOj be-darga. he was+offended from-lowering himself in-rank The local domain in 41 cannot be the whole clause; if it were, we would expect 41a to be excluded, the pronoun being bound, and 41b to be grammatical, the anaphor being bound. It follows that the local domain must be the noun phrase containing the anaphor/pronoun. As such, it must contain a c-commanding subject structurally realized as an empty category (as proposed by Chomsky 1986a). The only candidate available is the Agent of horada 'lowering', which cannot be co-indexed with its Patient (the anaphor/pronoun), according to a reasonable interpretation. The judgments of 41 immediately follow.20 In light of this, the incapability of derived nouns to assign accusative Case in contexts of the type in 40c or 42b below cannot follow from Burzio's generalization, since the understood subject can be structurally realized as a null element, and the external theta-role can, of course, be assigned to a phonetically null subject, as in 42c, for instance:21 (42) a. Hu nifga me-horadat ben-o be-darga. he was+offended from-lowering son-his in-rank 'He was offended by his son's demotion.' b. *Hu nifga me-ha-horada 'et ben-o be-darga. he was+offended from-the-lowering ACC son-his in-rank c.

Horidu'et ben-o be-darga. lowered(3PL) ACC son-his in-rank 'His son was demoted.'

On the basis of the above data, it must be concluded that accusative Case assigned by derived nouns is licensed differently than that assigned by verbs. The former, it appears, can be assigned only when the subject is phonetically realized. Finally, VPs permit both word orders: dative-accusative (43a) as well as accusativedative (43b), the former being the unmarked order, as first observed by Shlonsky (cited by Rapoport 1987). It is precisely the dative-accusative word order which is ruled out in noun phrases (43d); the accusative complement must precede its dative counterpart (43c). In other words, the accusative argument must be the first complement following the (genitive) subject:

HEBREW NOUN PHRASES

259

(43) a. Ha-bank hexzir le-shula 'et ha-kesef. the-bank refunded to-shula ACC the-money 'The bank refunded the money to Shula.' b.

Ha-bank hexzir ' et ha-kesef le-shula. the-bank refunded ACC the-money to-shula The bank refunded the money to Shula.'

c. ?haxzarat ha-bank 'et ha-kesef le-shula refund the-bank ACC the-money to-shula 'the bank's refund of the money to Shula' d. *haxzarat ha-bank le-shula 'et ha-kesef refund the-bank to-shula ACC the-money 'the bank's refund of the money to Shula' This asymmetry would also be unexpected, if the syntactic representation of process nominals involved a verbal projection and the accusative Case in question were assigned by V°. Thus, it is hardly controversial to assume that accusative Case of derived nouns is not an ordinary type of accusative. First, it can be assigned only through a Case marker, 'et. Moreover, it cannot be assigned in Exceptional Case Marking environments (i.e., to subject of small clauses). In addition, its licensing conditions are particular, since it can only be assigned when the subject is phonetically realized and in a position locally related to it. Suppose now that accusative Case in nominal contexts is indeed an inherent Case (assigned by 'et), as suggested by the fact that it is blocked in ECM constructions. Other instances of inherent accusative Case have been discussed in the literature: in Italian (Belletti and Rizzi 1988), in German (van Riemsdijk 1981) and in Icelandic (Zaenen et al. 1985; SigurSsson 1989, among others). Moreover, it has been argued that inherent Case is subject to particular licensing conditions (see, for instance, SigurSsson 1989). Along these lines, it can be suggested that it is in relation with the genitive argument (the subject) that the inherent accusative is licensed in Hebrew noun phrases. It is therefore available only when the subject is phonetically realized and can only be assigned to the first argument following it. This suggestion will not be developed here; for the purposes of this paper, it suffices that accusative Case assigned in noun phrases crucially differs from that assigned by verbs, thus rendering the VP-hypothesis significantly less appealing. In sum, it seems that while 'et appearing in verbal contexts may simply be a morphological realization of accusative Case, its homophonous form in nominal contexts is a dummy preposition which actually assigns the Case. Accusative Case then can be assigned in Hebrew noun phrases precisely because this dummy preposition is available, whereas it cannot be assigned in English noun phrases, for example, because no such device is available. Consequently, one may wonder whether 'et can be shown elsewhere to be the Case assignor of a Theme argument. Some evidence in this direction is offered by passive and unaccusative configurations in colloquial Hebrew, as first observed by Shoshani (1980): When the Theme argument is definite, 'et can sometime be inserted:22

260

PARAMETERS AND FUNCTIONAL HEADS

(44) a. Nodali 'etze'etmol. was+known to+me ACC this yesterday 'I found it out yesterday.' b. Lo nimsar li 'et ha-hoda'aha-zot. NEG was+transmitted to+me ACC the-message the-this 'This message was not transmitted to me.' c. Lo hayakatuv sham 'et ha-sha'a. NEC was written there ACC the-hour 'The hour was not written there.' d. Kara li kvar 'et ha-te'una ha-zot. happened to+me already ACC the-accident the-this 'I already had this accident.' Note that when the Theme argument is indefinite, the verb must agree with it, which shows that the Case it is assigned is not accusative but rather nominative (compare 44b with 45a-b): (45) a. Lo nimserali hoda'a. NEG was+transmitted to+me message 'A message was not transmitted to me.' b. *Lo nimsar li hoda'a. NEG was+transmitted to+me message Hagit Borer (personal communication) observes that foreign words, such as konstrukcia 'construction', which are not assimilated into the system and do not have corresponding verbs, do not allow accusative Case assignment within their noun phrases. At first glance, this argues in favor of the VP-hypothesis; why could 'et not assign accusative Case in noun phrases whose head is a foreign word? However, a closer examination shows that foreign nouns of that type do not denote a process but rather a result. Thus, for instance, konstrukcia 'construction' cannot be modified by an adverbial PP (46a), contrary to its corresponding process nominal bniya 'construction' (46b). Accusative Case assignment in the former is naturally excluded on these grounds:23 (46) a. ha-konstrukcia she! ha-bayit (*bi-mehirut) the-construction of trie-house (in-quickness) 'the quick construction of the house' b. ha-bniya shel ha-bayit bi-mehirut the-construction of the-house in-quickness 'the quick construction of the house' Moreover, independently, 'et insertion seems to be rather idiosyncratic, as first noticed by Doron (1989). Take, for instance, the transitive verbs hitbi'a and tibe 'a, which both mean ' sink'. Their corresponding derived nouns are hatba 'a and tibu 'a, respectively, which can both have process reading, as shown, for instance, by the fact that they can be modified by adverbial PPs (47):

HEBREW NOUN PHRASES

261

(47) a. hatba'atha-sirotbi-mehirut sinking the-boats in-quickness 'the quick sinking of the boats' b. tibu'aha-sirotbi-mehirut sinking the-boats in-quickness 'the quick sinking of the boats' Interestingly, only hatba'a allows accusative Case assignment (48), although both corresponding verbs are transitive. This idiosyncrasy is completely unexpected under the VP-hypothesis: (48) a. hatba'at ha-cava 'etha-sirot sinking the-army ACC the-boats 'the army's sinking of the boats' b. *tibu'a ha-cava 'et ha-sirot sinking the-army ACC the-boats 'the army's sinking of the boats' To conclude, as accusative Case in noun phrases differs essentially from that of transitive verbs and as process nominals are modifiable by adverbial PPs and not by authentic adverbs, there seems to be no reason to conjecture that process nominals are not pure nouns. In the next section I will show that a brief comparison with constructions having dual categorial structure supplies further support for this conclusion.

5.3 Nominalized Infinitives Let us briefly consider the behavior of other nominal constructions which have been argued to contain a verbal projection. I have in mind Italian nominalized infinitives (see Grimshaw and Selkirk 1976, Salvi 1982, Zucchi 1989, Bottari 1990, for detailed discussion). Similarly to Hebrew process nominals, they manifest ambiguous categorial behavior. Like nouns, nominalized infinitives appear with an article, have nominal distribution (49a), can be modified by adjectives (49b) and their subject can be realized as a possessive pronoun (49c). Like verbs, they can assign accusative Case to their direct object (49d) and are modifiable by adverbs (49e): (49) a. II pensare a Maria lo rattrista. the thinklnf about Maria him makes+sad 'Thinking about Maria makes him sad.' b. il continue pensare a Maria the continuous thinklnf about Maria 'the continuous thinking about Maria' c. il suo pensare a Maria the his think lnf about Maria 'his thinking about Maria'

262

PARAMETERS AND FUNCTIONAL HEADS

d. il suo leggere i quotidiani the his readlnf the newspapers 'his reading the newspapers' e. il leggere spesso quotidiani the readlnf often newspapers 'often reading newspapers' However, whereas in Hebrew process nominals these arguably verbal properties are seriously constrained, in nominalized infinitives they are not subject to any such constraints. As already shown in 49d-e, the nominalized infinitive can assign accusative Case to both definite and indefinite nouns (evidently not via a particle). Moreover, its accusative Case is not dependent on the phonetic realization of the subject (49e) and it can be assigned to subjects of small clauses (50): (50) a. il sentir-lo cantare quella canzone the hearlnf-him singlnf that song 'hearing him singing that song' b. il veder Gianni lavorare come un pazzo the seelnf Gianni worklnf like a madman 'seeing Gianni working like a madman' Thus, contrasting Hebrew process nominals with Italian nominalized infinitives reinforces the claim that the former are purely nominal and do not contain a verbal projection.

6

Conclusion

Although Hebrew process nominals appear to share some verbal properties, it seems highly unlikely that their syntactic representation includes a verbal projection. It has been argued here that they are pure nouns which project a hierarchical syntactic structure, in concert with the lexicalist hypothesis (stemming from Chomsky 1970). The paper focuses on two Hebrew genitival constructions, examining various aspects of their syntax. In particular, it has been shown that given their configurational internal structure and the relative ordering of their arguments, it must be concluded that Hebrew noun phrases always involve noun raising. The landing site has been identified as the head D°, which allows a simple unified account of various syntactic phenomena, justifying the more articulated view of noun phrases as DPs.

Notes 1. An earlier version of this paper circulated as Siloni (1990) and a shorter version is included in MIT Working Papers in Linguistics 14 (1991). 1 thank Adriana Bellctti, Hagit Borer, Robin Clark, Marc-Ariel Friedemann, Teresa Guasti, Richard Kayne and Ur Shlonsky for helpful comments. Special thanks to Luigi Ri/./.i and Ian Roberts. 2. In addition, Hebrew has a clitic doubling construction:

HEBREW NOUN PHRASES

263

(i) beyt-Oj shel ha-'ishj house-his of the-man 'the man's house' This construction will not be dealt with in this paper; for detailed discussion see Borer (1984). 3. The phonological alternation between a head noun in the construct state and an "independent" head noun is clearly due to the absence of stress on the former. Thus, for instance, an initial vowel /a/ is reduced to a schwa when non-adjacent to stress; a suffix /t/, elided under stress, fails to delete. For phonological discussion of the construct state, see McCarthy (1979). The Hebrew transcription used throughout is merely a surface reflection of spoken Modern Hebrew. 4. Besides construct state nominals which are reasonably productive, Hebrew also has compounds. Compounds consist of exactly two nouns which result in idiosyncratic meaning. In some respects they behave like construct states; for instance, the head noun in both cases loses stress and undergoes the same phonological alternations. For discussion of compounds vs. construct state nominals, see Borer (1989). 5.1 put aside derived nouns expressing result, which have much freer relations with their complements, as well as known peculiarities of process nouns (the optionality of the subject and certain gaps in their complement paradigms), since the issue in question is how nominal theta-grids are mapped onto syntactic structures. For discussion of these other topics, which are not directly relevant here, see Kayne (1984), Lebeaux (1986), Safir (1987), Williams (1987), Clark (1990) and Grimshaw (1990), among others. 6. See Belletti and Rizzi (1988) for similar discussion with respect to "psych" verbs. 7. See Shlonsky (1988) for discussion of binding in Hebrew underived nominals. 8. This is not in contradiction with Kayne (1984), who argues that nouns cannot govern across maximal categories. While it is plausible that nouns cannot govern across distinct maximal categories, they must be able to govern into their own maximal category, once they are raised to D°. See section 4.1 for more discussion of this point. 9. A parallelism between the derivation of VSO surface order and that of NSO (Noun-SubjectObject) surface order was first suggested by Ritter (1987) in the context of construct states (discussed here in section 4). See also Fassi Fehri (1989). 10. This issue is somewhat more complicated in Romance, as both prenominal and postnominal adjectives can occur in the same language. For discussion see Longobardi (1990, 1992). 11. In Hebrew the Case is not visible; however, in Arabic construct states, which exhibit precisely the same properties, genitive Case is morphologically realized (see examples 31c-d). 12. Borer (1989) argues that construct states are words created in the syntax, thus explaining the phonological changes the head undergoes and supplying a strong reason for the strict adjacency imposed on the two members of the construction. Note, however, that construct states are clearly not words syntactically created via incorporation (as suggested by Shlonsky 1990). This is shown by examples such as (i), where the complement is a coordinate structure. The coordinate structure as a whole cannot be claimed to incorporate with the head of the construct state, nor can one member of the coordinate structure be extracted: (i) beyt ha-rabi mi-kiryat 'arba ve-ra'ayat-o ha-nixbada house the-rabbi from-Kiryat Arba and-wife-his the-honorable 'the house of the rabbi of Kiryat Arba and his honorable wife" Additionally, note that if noun phrases contain an additional functional category between DP and NP (Ritter 1991; Cinque 1992), the right word order can easily be derived. This possibility will not be pursued here.

264

PARAMETERS AND FUNCTIONAL HEADS

13. The fact that Italian past participles in the absolute construction cannot agree with the external argument is not directly relevant here. For our concerns, it suffices that agreement under the structural configuration of government is attested elsewhere. 14. See Borer (1989) for an explanation in terms of feature percolation rendered possible by her claim that construct states are words, as mentioned in n. 12. 15. Rizzi suggests that the complementizer that cannot co-occur with Agr in C°, in order to account for the contrast between (i) and (ii): (i) *WhOj do you think [cp tj [c that [lp tj came] ] ] (ii) Who; do you think [CP tj [c Agr; [Ip tj came] ] ] Example (ii) is grammatical because C° agrees with the trace in its specifier position, and turns therefore into a proper governor for the trace in subject position. Example (i), on the contrary, is excluded since that is incompatible with Agr in C°, and by itself it is inert for government, hence, the Empty Category Principle is violated. 16. The fact that this Agr is morphologically realized only under Spec-head configuration is not at all peculiar, but rather, a common phenomenon. Thus, for instance, the Hebrew quantifier kol 'all', which is argued by Shlonsky (1991) to head a QP, agrees with its complement only when the latter appears in its specifier position, yielding a Spec-head configuration. Similarly, the Hebrew existential particle yesh and its negative counterpart 'eyn can agree with their subject only when they follow it. Furthermore, in standard Arabic, verbs show full agreement (in gender, number and person) exclusively with pre-verbal subjects; with post-verbal subjects they agree in gender only. These phenomena can be analyzed along similar lines, namely, assuming that morphological Agr may sometimes be triggered only under Spec-head configuration. 17. When the direct object is heavy, it can appear without the particle 'et: (i) bikur sar ha-xakla'ut 'olim xadashim ha-mitgorerim be-parvarey yerushala'im visit minister the-agriculture comers new that-live in-suburbs Jerusalem 'the minister of agriculture's visit to newcomers who live in the suburbs of Jerusalem' This suggests that the process of Heavy NP Shift may rescue a construction which is otherwise ungrammatical because of lack of Case. 18. Exceptional Case Marking in Hebrew is rare. Judgments are subtle and may therefore vary among speakers. Nevertheless most of the speakers I have consulted agreed with the judgments in 39. 19. The present line of argumentation was suggested to me by Luigi Rizzi. 20. As shown below, it is not the case that anaphors can never appear in this configuration: (i) Hu nimna me-'ahavat 'acmo. he avoided from-loving himself 'He avoided loving himself.' 21. In addition, Agent-oriented adverbs and Relational clauses are also licensed in Hebrew noun phrases. According to Roberts (1987), these must be licensed by a structurally realized Agent. As the Agent is generated in the specifier position of NP, the Theme (the object) in examples of the type in 42a must be adjoined to NP for Case reasons, as already suggested in section 4.1 on different grounds. Further questions concerning null subjects in noun phrases will not be dealt with in this paper. 22. As noted by Richard Kayne (personal communication), the present analysis views as accidental not only that in both nominal and verbal contexts we find the same particle, but also that in both contexts it is limited to preceding a definite noun phrase. Conceivably, accusative Case in Hebrew may always be an inherent Case assigned by 'et (in verbal

HEBREW NOUN PHRASES

265

contexts, indefinite direct objects may be assigned partitive Case, as suggested by Adriana Belletti, personal communication). In that case the differences between accusative Case in VPs and DPs should follow from other differences between the two projections. Note that in any event, it must be assumed that inherent accusative in noun phrases and intransitive verbal configurations is not subject to the same licensing conditions, as in the latter, it is not dependent on the realization of any other argument. Following this line of investigation would be beyond the scope of this paper. Crucially, however, this does not weaken the claim that the particular behavior characterizing accusative Case in noun phrases renders the VP-hypothesis unlikely. 23. Foreign nouns which are assimilated into the system and have corresponding verbs can denote an event; they are therefore able to assign accusative Case and are modifiable by adverbial PPs, as illustrated in (i): (i) minpul ha-cava 'et xayal-av be-shitatiyut manipulation the-army ACC soldiers-its in-methodicalness (= systematically) 'the army's systematic manipulation of its soldiers'

References Abney, S. 1987. The English Noun Phrase in its Sentential Aspects. MIT dissertation. 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. Belletti, A. and L. Rizzi. 1988. "Psych-Verbs and Theta Theory," Natural Language and Linguistic Theory 6:291-352. Herman, R. 1978. Modem Hebrew Structure. Tel-Aviv: University Publishing Projects. Borer, H. 1984. Parametric Syntax. Dordrecht: Foris. . 1989. "On the Morphological Parallelism between Compounds and Constructs," in G. Booij eta), (eds.),Morphology Yearbook 1. Dordrecht: Foris. 45-65. . 1990. "Derived Nominals and the Causative-Inchoative Alternation: Two Case Studies in Parallel Morphology," paper presented at the Workshop on Hebrew Grammar, February, 1990, Jerusalem. Bottari, P. 1990. "Livelli di rappresentazione lessicale: complementazione nominale e complementazione frasale," Ph.D, Universita di Padova/Venezia. Burzio, L. 1986. Italian Syntax: A Government-Binding Approach. Dordrecht: Reidel. Chomsky, N. 1970. "Remarks on Nominalizations," in R. Jacobs and P. Rosenbaum (eds.), Readings in English Tranformational Grammar. Waltham, Mass.: Ginn. 184—221. . 1986a. Knowledge of Language: Its Nature, Origins and Use. New York: Praeger. . 1986b. Barriers. Cambridge, Mass.: MIT Press. Cinque, G. 1980. "On Extraction from NP in Italian," Journal of Italian Linguistics 5:47-99. ——•. 1981. "Sulla nozione di soggeto di SN in Italiano," Cultura Neolatina 41:555-570. . 1992. "Functional Projections and N-movement within the DP," GLOW Newsletter 28:12-13. [Abstract of paper presented at the 1992 GLOW Conference.] Clark, R. 1990. "Structured Theta-Grids and the Syntax of Nominals," ms, Universite de Geneve. Doron, E. 1989. "Derived nominals in Hebrew," in J. Payne (ed.), The Structure of Noun Phrases. Berlin: Mouton. [To appear.] Fassi Fehri, A. 1989. "Generalized IP Structure, Case and VS Word Order," MIT Working Papers in Linguistics 10:75-113.

266

PARAMETERS AND FUNCTIONAL HEADS

Friedemann, M.-A. 1992. "On the D-Structure Position of Subjects in French," Proceedings of the Leiden Conference for Junior Linguists (LCJL) 3:155-168. Fukui, N. and M. Speas. 1986. "Specifiers and Projection," MIT Working Papers in Linguistics 8:128-172. Giorgi, A. and G. Longobardi. 1991. The Syntax of NPs: Configuration, Parameter and Empty Categories. Cambridge: Cambridge University Press. Grimshaw, J. 1990. Argument Structure. Cambridge, Mass.: MIT Press. Grimshaw, J. and E.O. Selkirk. 1976. "Infinitival Noun Phrases in Italian," ms, University of Massachusetts, Amherst. Haegeman, L. 1992. Generative Syntax: Theory and Description—A Case Study in West Flemish. Cambridge: Cambridge University Press. Hazout, I. 1988. "Nominalization Constructions in Modern Hebrew," ms, University of Massachusetts at Amherst. . 1990. "Verbal Nouns: Theta-theoretic Studies in Hebrew and Arabic," PhD, University of Massachusetts at Amherst. Kayne, R. 1984. Connectedness and Binary Branching. Dordrecht: Foris. Koopman, H. and D. Sportiche. 1982. "Variables and the Bijection Principle," Linguistic Review 2:139-160. Kuroda, Y. 1988. "Whether We Agree or Not: A Comparative Syntax of English and Japanese," in W. Poser (ed.), Papers of the Second International Workshop on Japanese Syntax. Stanford: Center for the Study of Language and Information (CSLI). 103-143. Lebeaux, D. 1986. "The Interpretation of Derived Nominals," Proceedings ofCLS 22:231-247. Lecarme, J. 1989. "Genitive Constructions in Somali: The Notion of Internal Complement for Nominals," ms, CNRS-LLAOR, Universite de Nice. Lieber, R. 1980. "On the Organisation of the Lexicon," Ph.D., MIT. Longobardi, J. 1990. "Evidence for the Structure of Determiner Phrases and N-Movement in the Syntax and in LF," GLOW Newsletter 24:42-43. [Abstract of paper presented at the 1990 GLOW Conference.] . 1992. "Proper Names and the Theory of N-Movement in Syntax and Logical Form," ms, Universita di Venezia. McCarthy, J. 1979. "Formal Problems in Semitic Phonology and Morphology," PhD, MIT. McCloskey, J. 1991. "Clause Structure, Ellipsis and Proper Government in Irish," Lingua 85:259-302. Rapoport, T. 1987. "Copular, Nominal, and Small Clauses: A Study of Israeli Hebrew," PhD, MIT. Reinhart, T. 1976. "The Syntactic Domain of Anaphora," Ph.D., MIT. Riemsdijk, H. van. 1981. "The Case of German Adjectives," in J. Pustejovsky and V. Burke (eds.), Markedness and Learnability. University of Massachusetts Occasional Papers in Linguistics 6, Amherst. Ritter, E. 1987. "NSO Noun Phrase in Modern Hebrew," Proceedings ofNELS 17. 521-537. . 1988. "A Head-Movement Approach to Construct-State Noun Phrases," Linguistics 26:909-929. . 1991. "Two Functional Categories in Noun Phrases: Evidence from Modern Hebrew," ms, Universite du Quebec a Montreal. Rizzi, L. 1990. Relativized Minimality. Cambridge, Mass.: MIT Press. . 1991b. "Residual Verb Second and the W/i-Criterion," ms. [Also published in this volume, 63-90.J Rizzi, L. and I. Roberts. 1989. "Complex Inversion in French," Probus 1:1-30. [Also published in this volume, 91-116.] Roberts, I. 1987. The Representation of Implicit and Dethematized Subjects. Dordrecht: Foris.

HEBREW NOUN PHRASES

267

Rosen, H.B. 1957. Ivrit Tova: lyunim be-Taxbir. Jerusalem: Kiryat sefer. Safir, K. 1987. "The Syntactic Projection of Lexical Thematic Structure," Natural Language and Linguistic Theory 5:561-601. Salvi, G. 1985. "L'infinito con 1'articolo," in L.M. Savoia and A. Franchi De Bellis (eds.), Sintassi e morfologia della lingua italiana d'uso. Roma: Bulzoni. 243-268. Shlonsky, U. 1988. "Government and Binding in Hebrew Nominals," Linguistics 26:951-976. - . 1990. "Construct State Nominals and Universal Grammar," ms, University of Haifa. - . 1991. "Quantifiers as Functional Heads: A Study of Quantifier Float in Hebrew," 84:159-180. Shoshani, R. 1980. "The Object Marker in Hebrew in Intransitive Contexts," ms, Tel- Aviv University. SigurSsson, H. Armann. 1989. "Verbal Syntax and Case in Icelandic," Ph.D., University of Lund. Siloni, T. 1 990. "On the Parallelism between CP and DP: The Case of Hebrew Semi-Relatives," Proceedings ofLCJL 1, Leiden. 135-153. Sportiche, D. 1988. "A Theory of Floating Quantifiers and Its Corollaries for Constituent Structure," Linguistic Inquiry 19:425-449. Sproat, R. 1985. "Welsh Syntax and VSO Structure," Natural Language and Linguistic Theory 3:173-216. Szabolcsi, A. 1987. "Functional Categories in the Noun Phrase," in I. Kenesei (ed.) Approaches to Hungarian, vol. 2. Budapest: University of Budapest. 167-189. - . 1989. "Noun Phrases and Clauses: Is DP analogous to IP or CP?," in J. Payne (ed.), The Structure of Noun Phrases. Berlin: Mouton. [To appear.] Williams, E. 1981. "Argument Structure and Morphology," Linguistic Review 1:81-1 14. - . 1987. "English as an Ergative Language: The Theta Structure of Derived Nouns," Proceedings ofCLS 23:366-375. Zaenen A., J. Maling and H. Thrainsson. 1985. "Case and Grammatical Functions: The Icelandic Passive," Natural Language and Linguistic Theory 3:441^84. Zucchi, A. 1989. "The Language of Propositions and Events: Issues in the Syntax and Semantics of Nominals," PhD, University of Massachusetts, Amherst.

This page intentionally left blank

8

Three Kinds of Subject Clitics in Basso Polesano and the Theory of pro Cecilia Poletto

1 Introduction Subject clitics of the northern Italian dialects (NIDs) have been researched in a number of studies in the field of generative syntax, primarily because they seem to constitute a case of intermediate pro-drop, in which only some persons of the verb can have a null subject, while others need a subject clitic pronoun.1 Following recent work by Brandi and Cordin (1981, 1989) and by Rizzi (1986b), I will assume that NIDs are fully pro-drop, subject clitics being heads at S-structure and not true NPs. In this work I will argue that the presence of subject clitics is strongly connected with the pro-drop system of the dialects, which is different from standard Italian. On the basis of the data of a dialect spoken in the southeastern part of the Veneto, Basso Polesano, it will be shown that a subject clitic is the head that licences pro rather than Agreement, as in standard Italian. Assuming Rizzi's (1986a) theory of pro-drop, subject clitics will be distinguished on the basis of their relation to a null subject. In section 2.1, third person subject clitics will be characterized as argumental heads, drawing a parallel with object clitics. They absorb the subject theta-role and licence a pro through a chain, which transmits the person and number features to the null subject. Section 2.2. examines the distribution of another subject clitic (which will be described as an expletive) that licences pro but does not have any content features to assign to the null subject. A third type of element, the distribution of which is not affected by the presence of a pro subject, is the second person singular subject clitic examined in section 2.4. Basso Polesano appears to have a complex system with three different types of subject clitics, which have distinct syntactic properties with respect to a null subject. Further, this study of the pro-drop conditions of Basso Polesano will bring evidence 269

270

PARAMETERS AND FUNCTIONAL HEADS

that Rizzi's distinction between formal licencing and feature identification of a null subject is correct because they can be fulfilled by two different heads (section 2.3). It is probable that other dialects show the same partitions within the class of syntactic subject clitics, as well as the same pro-drop system as those to be seen for Basso Polesano. The field is open to further research that will give us a deeper understanding of the phenomenon of null subjects.

1.1

French vs. Northern Italian Dialects

The distribution of subject clitics in the NIDs has been studied in various work by different authors. Rizzi (1986b) in particular, starting from Kayne's observation that French subject clitics are only phonological clitics, compares them to the subject clitics of NIDs which, to the contrary, are shown to be syntactic clitics. Hence they are heads at S-structure and are realized as adjuncts to the head of the Infl projection. Rizzi assumes 1 and 2 as the S-structures of French and NIDs, respectively:

In 1 the French subject clitic is in the normal position of preverbal subject NPs, SpecIP. In 2 the subject clitic of NIDs is realized as an adjunct on the head of Inflection. Hence, French is not a pro-drop language while NIDs are. Rizzi provides various tests in order to distinguish between syntactic and phonological subject clitics. We will briefly examine three of them here, as they will be relevant to the discussion in the following sections.

1.2

The first test

If subject pronouns in French are only phonological clitics and they occupy the SpecIP position at S-structure, they must be in complementary distribution with other subject NPs. Consequently, a subject NP and a subject clitic can co-occur in French only if the subject NP has been left-dislocated:2

THREE KINDS OF SUBJECT CLITICS

(3)

271

Jean, il vien John, he comes

Evidence that in 3 the NP Jean occupies a left-dislocated position comes from the distribution of bare quantifiers, which can be neither left- nor right-dislocated, as 4 shows: (4) *Quelqu'un, je 1'ai vu. someone, I him have seen A bare quantifier cannot occupy a dislocated position presumably because the corresponding clitic pronoun inside the sentence cannot be bound as a variable by the quantifier (cf. Rizzi 1986b; Cinque 1990). The structural representation in 1 predicts that the sequence subject-QP-subject-clitic must be ungrammatical in French, because the quantifier and the subject clitic compete for the same subject position, namely Spec-Agr. (5) a. *Personne il ne vient. nobody, he not comes b. Personne ne vient. nobody not comes The prediction is borne out: a sentence such as 5a is completely ungrammatical while 5b, where there is no subject clitic, is acceptable. In the NIDs examined by Rizzi (1986b) and Brandi and Cordin (1981, 1989), on the other hand, a bare quantifier can co-occur with a subject clitic (in fact it must occur this way):3 (6) a. Tut/'e capita de not. everything, it has happened by night b. *Tut e capita de not. As quantifiers cannot be left-dislocated, a sentence such as 6a cannot be an instance of left-dislocation. The quantifier must be in the normal subject position, namely Spec-IP, where the subject clitic occupies a different position in the structure, and it is obligatory, as 6b shows. The conclusion is that subject clitics of NIDs do not occupy the Spec-IP position of the subject, but rather, they occupy the head of IP. The contrast between 5 and 6 shows that a subject NP and a subject clitic occupy different positions at S-structure in NIDs while they occupy the same position in French. The distinction between French and NIDs, formulated by adopting two different structures (examples 1 and 2), seems to be confirmed.

1.3

The second test

The second test concerns the position of the subject clitic with respect to the preverbal negative marker, which is a clitic too. In French, subject clitics always appear on the left of the negation marker ne, like any other subject NP, as in 7a. In the NIDs, some subject clitics appear on the left, some on the right of the preverbal negative marker (see 7b), depending on the dialect variety or on the person of the verb:

272

PARAMETERS AND FUNCTIONAL HEADS

(7) a. //n'a pas mange. he not has not eaten b. No/'amagna. not he has eaten Rizzi argues that the sequence negative clitic-subject clitic (7b) can be found only if the subject clitic and the negative marker are both members of the same cluster, which itself is contained in Infl. Hence, subject clitics of the NIDs are a part of Inflection. This is not expected when the subject clitic is in Spec-IP and the negative marker in Infl. The fact that in French we never find the order negative clitic-subject clitic constitutes evidence that subject clitics do not belong to the same clitic cluster as negation, hence they occupy the Specifier and not the head of IP.

1.4

The third test

The third test concerns the possibility of VP-conjunction. In French, the second occurrence of a subject clitic can be omitted in a sentence such as 8a while it must be repeated in the NIDs (cf. examples 8b, c):4 (8) a.

II mange du pain et bois du vin. he eats some bread and drinks some wine

b. *E1 magna pan e beve vin. he eats some bread and drinks some wine c. El magna pan e el beve vin. he eats some bread and he drinks some wine This contrast indicates that subject clitics in the NIDs occupy a position strictly connected with the verb while in French they are realized as independent items at S-structure (see Rizzi 1986b:400,404 for a detailed analysis). The tests illustrated here will be employed in the next sections in order to determine which position the subject clitics of Basso Polesano occupy.

2

The Split-Infl Hyphothesis

In the following discussion we will assume that the Infl node has to be split into two different projections, AgrP and TP (see Pollock 1989), and that AgrP is higher than TP (Belletti 1990), as in 9. The verb moves in Italian and French tensed clauses to the highest inflectional projection to take up inflectional morphology. Updating Rizzi's analysis, the subject clitic is realized in Spec-Agr in French, as in 10, and as an adjunct on the head of Agreement in NIDs (11). I will make use of another hypothesis that has recently been proposed by several authors (see Koopman and Sportiche 1988, among others); namely, that the subject originates inside the VP and then moves to Spec-Agr in order to get case from Agreement.

THREE KINDS OF SUBJECT CLITICS

273

The position of the subject inside the VP is the site where the subject theta-role is assigned by the verb, hence I will refer to it as basic position of the subject. In Italian and in NIDs, the subject can surface in postverbal position. This position must be an argumental one, because it is a possible site for extraction where the trace of the subject can be properly governed. Following Giorgi and Longobardi(1991), we will make the assumption that the postverbal surface position in Italian is the basic one.

274

PARAMETERS AND FUNCTIONAL HEADS

In Italian, a subject NP can be realized in two distinct argumental subject positions: the preverbal one, Spec-Agr, and the basic position inside the VP, as in 12:5 (12) a. Gianni ha mangiato qui. John has eaten here b. Qui ha mangiato Gianni. here has eaten John The possibility of a postverbal subject has been related to the Pro-drop Theory (Rizzi 1982): only when there is an expletive pro in the preverbal position can the subject stay inside the VP. In NIDs, the subject can surface in postverbal position, as in standard Italian. In the next sections we will examine Basso Polesano in order to determine which conditions of pro-drop are present in this dialect and how the distribution of the subject clitics correlates with them.

2.1 Argumental clitics In this section we will discuss the distribution of third person subject clitics in Basso Polesano. These clitics are marked for number and gender, as shown in 13, and are phonetically identical to the definite article: (13)

masc. fern.

sing

plur

el la

i le

As mentioned in section 1.2, within Rizzi's analysis, subject clitics in NIDs "spell out" Agreement, hence they are not true subject NPs, realized in Spec-Agr, but heads, which are adjoined to the head of Agreement. At first sight, Basso Polesano seems to confirm this hypothesis because subject clitics appear on the right of the negative clitic while subject NPs appear on the left: (14) a. No /a magna. not he eats b. Mario no magna. Mario not eats The different order with respect to negation in 14a and 14b shows that the subject NP Mario and the subject clitic la cannot occupy the same position. So, third person subject clitics do not occupy the Spec-Agr position as subject NPs. Another difference between NPs and clitics can be seen in 15. As in other NIDs, subject clitics in Basso Polesano have to be repeated in the second member of a conjoined structure while NPs do not: (15) a. La magna come un lupo e *(la) beve come un omo. she eats like a wolf and *(she) drinks like a man

THREE KINDS OF SUBJECT CLITICS

275

b. La Maria magna come un lupo e beve come un omo. the Mary eats like a wolf and drinks like a man In 15a the sentence is grammatical only if the subject clitic la is repeated in the second conjoint. Example 15b shows that this is not necessary for subject NPs. Once again subject clitics do not behave as NPs. The results of these two tests lead us to conclude that in Basso Polesano subject clitics are heads, and not NPs at S-structure, as is the case for other NIDs. Nevertheless, we cannot simply describe the distribution of subject clitics as "spelling out agreement morphology", as in the dialects examined by Rizzi (1986b) and Brandi and Cordin( 1981). If subject clitics are to be assimilated to agreement morphology, they should be present in every context, as is the case with agreement morphology. This is not true in Basso Polesano: sentences such as 16, where there is no subject clitic, are perfectly possible: (16) Mario magna tanto. Mario eats a lot The third person subject clitic seems to be completely optional in these structures. It can be realized or left out, without any change in the acceptability of the sentence (see Beninca and Vanelli 1984): (17) Mario el magna tanto. Mario he eats a lot In 17 the subject NP Mario co-occurs with the subject clitic el. As seen in section 1.2, 17 can be analyzed in two possible ways: the subject can be in Spec-Agr and the clitic in Agr, as in the dialects considered by Rizzi (1986b; cf. 11, above), or the subject can occupy a left-dislocated position, and the clitic in Spec-Agr, as in French (cf. 10). We know that there are constraints on left-dislocation: for instance, a quantifier cannot be left-dislocated (see 4). We can determine whether 17 is an instance of left-dislocation or not, using this peculiarity of quantifiers. On the basis of the two tests in 14 and 15 we have claimed that in Basso Polesano, as in other NIDs, third person subject clitics are heads. This means that a structure such as 11 must be correct for Basso Polesano too. If subject clitics are heads adjoined to Agr, they leave the Spec-Agr position free for a subject NP. We expect then, that all kinds of subject NPs can co-occur with a subject clitic, as is the case with referential subject NPs (cf. 17): (18) a. *Qualchedun el magna tanto. somebody he eats a lot b. Qualchedun magna tanto. somebody eats a lot What we find in 18 is exactly the opposite: if the subject NP is a quantifier, it cannot co-occur with a subject clitic. The sentence is grammatical only if the subject clitic

276

PARAMETERS AND FUNCTIONAL HEADS

is not present. In this case Basso Polesano patterns with French and not with the NIDs examined by Ri/zi. Example 18 is parallel to 5 (see above): example 18b is, in fact, perfectly acceptable, as is French 5b. The fact that in 18a the quantifier and the subject clitic cannot co-occur shows that in 17 the subject clitic must be left-dislocated, as in French. This is not predicted under an analysis that considers subject clitics of Basso Polesano as heads adjoined to the head of Agreement. A structural description such as 11 is not completely adequate to the data because it predicts 18a will be grammatical. We are left with a contradictory situation: the tests with coordinate structures and with the negative clitic show that subject clitics of Basso Polesano are not in Spec-Agr like subject NPs. In other words, we should assume that 11 is the correct structure. On the contrary, 17 shows that they are in Spec-Agr (because they compete with other subject NPs, like French subject clitics) and that the structure of Basso Polesano must be as shown in 10. In order to solve this problem, a further distinction has to be drawn inside the domain of syntactic clitics. We will argue that subject clitics of Basso Polesano must be different both from those of the dialects examined in Rizzi (1986b) and Brandi and Cordin (1981) and from French. In Basso Polesano, the subject clitics seem to be in complementary distribution with subject NPs, not only when the subject is in preverbal position, but also when it is in a postverbal one. As in standard Italian, the subject can surface in postverbal position, but this is possible only if there is no third person subject clitic on the left of the verb:6 (19) a. *El magna qualchedun, qua. he eats somebody, here b. A magna qualchedun, qua. A eats somebody, here Examples 18a, b and 19a, b show that a third person subject clitic can never co-occur with a subject NP if the subject NP is in an argumental position. Neither the preverbal nor postverbal position can be occupied by a subject NP if there is a subject clitic adjoined to Agr. This leads us to the conclusion that the ungrammatically of 18a is not due to the fact that the subject NP and the subject clitic compete for the same position, as in French; they must be incompatible for another reason. Note that in French a subject clitic is compatible with a postverbal subject (as in 20), while in Basso Polesano it is not (cf. 19a): (20) // a mange quelqu'un ici. it has eaten somebody here All the facts discussed up to now can be captured by making a very simple assumption: subject clitics in Basso Polesano behave like object clitics. An object clitic can appear only when the basic position of the object NP is phonetically empty: 7

THREE KINDS OF SUBJECT CLITICS

277

(21) a. Lovedo. it (I) see b. *Lo vedo qualcosa. it (I) see something This is true in Italian, French and NIDs as well. Within the standard theory (see Kayne 1975, 1989b), this observation is expressed in terms of movement. The object clitic moves from the basic position of the object and cliticizes to the head of Agreement. It leaves a trace in the basic position of the object bound by the clitic itself. An object NP cannot be realized in that position, because it is already filled by the trace of the clitic. We can try to explain the same effect for subject clitics using the same assumption. In Basso Polesano the subject clitic is generated in the basic position of the subject, which is inside the VP (see Koopman and Sportiche 1988 for a detailed analysis). It receives the theta-role of the subject and moves to the head of Agreement, leaving a trace inside the VP. This is the reason why a subject NP cannot be realized in the postverbal basic position. But a subject NP cannot be realized in preverbal position either, and this fact cannot be explained as above. There must be another factor that prevents the co-occurrence between the subject NP and the clitic. In the case of the object clitic, the standard analysis predicts that the chain object clitic-trace absorbs the object theta-role. There are then two reasons that prevent an object NP from appearing in object position: the fact that the trace of the clitic already fills the position, and the impossibility of receiving a theta-role which has already been absorbed by the chain clitic-trace. If a subject clitic is identical to an object clitic, the chain subject clitic-trace absorbs the subject theta-role. Hence a subject NP cannot be realized either in the preverbal position or in the postverbal one, because it would be left without a thetarole, violating the Theta Criterion. In the case of 18, the complementary distribution between subject clitic and subject NP is not due to the fact that they compete for the same position, as in French. They both absorb the subject theta-role, and if they are both realized, the Theta Criterion is violated. This hypothesis predicts that a subject clitic will be in complementary distribution with subject traces as well. Consequently a subject clitic should be ungrammatical in all cases of wft-movement of the subject. Restrictive relative clauses, cleft sentences, and topicalization are typical instances of w/z-movement (see Chomsky 1977).8 They can be used as a test to determine if the subject clitic can co-occur with a subject trace (cf. Beninca and Vanelli 1984). (22) a. la putina che vien vanti... the girl that comes along b. *la putina che la vien vanti... the girl that she comes along (23) a. MARIO, ga ciama, no Toni. MARIO, has called, not Tony b. *MARIO, el ga ciama, no Toni. MARIO, he has called, not Tony

278

PARAMETERS AND FUNCTIONAL HEADS

(24) a. A ze Mario, che ga ciama. A is Mario, that has called b. *A ze Mario, che el ga ciama. A is Mario, that he has called Examples 22, 23, and 24 are instances of restrictive relative, topicalization and clefting, respectively. In 22b, 23b, and 24b the presence of a subject clitic renders the sentence ungrammatical. Our prediction is then borne out. Subject clitics are incompatible with w/z-subject traces because both absorb the subject theta-role, leading to a violation of the Theta Criterion. We will call subject clitics of this type "argumental clitics" because they behave like arguments with respect to the Theta Theory. There is another fact that can be accounted for by making the assumption that these subject clitics are arguments; namely, the impossibility for a third person subject clitic to appear in contexts where an argumental theta-role has not been assigned; for example, with weather and raising verbs (see Beninca and Vanelli 1982 for a comparison with other dialects): (25) a. *Elpiove. he rains b. *E1 pare che Nane sia riva. he seems that John has come Again, the impossibility of 25a and b is not expected under an analysis that treats subject clitics as part of agreement morphology. On the contrary, the hypothesis that third person subject clitics are arguments predicts that they cannot occur if an argumental theta-role is not available for the subject. There is another well known fact that must be taken into account. As the contrast between 26 and 27 shows, Basso Polesano is not a pro-drop language in the sense that standard Italian is: (26) a. *Magna. eats b.

El magna. he eats

In standard Italian it is possible for the subject to be phonetically absent, as in 27: (27) Mangia. eats Under the current analysis this type of sentence is analyzed as having a null pronoun in the subject position Spec-Agr, called pro. This pro is licenced by the verbal agreement, which is intrinsecally strong enough to identify the subject as third person singular. Given that 26a (as compared with 27) is ungrammatical, we have to state that Agreement in Basso Polesano is not so "strong" as in standard Italian.

THREE KINDS OF SUBJECT CLITICS

279

The grammaticality of pro depends on rich morphology. In fact, the Italian paradigm is characterized by both person and number, as 28 shows. In Basso Polesano, on the other hand, the number feature is not marked on the agreement morphology, as illustrated in 29: (28) a. Mangia. (he) eats b. Mangiarco. (they) eat (29) a. El magna. he eats b. I magnc. they eat In 28b the third person plural morphology is distinct from the singular (28a), while in 29b the verbal endings are the same as in 29a. This direct link between morphology and syntax may be too strong, as Beninca has pointed out to me. There are, in fact, other dialects that have subject clitics and show morphological number and person distinctions on the verb all the same. We will therefore assume that Agreement in Basso Polesano is syntactically weaker than in standard Italian, and that morphology is not the cause of this syntactic "weakness" but a reflection of it. Starting from the hypothesis that Agreement in Basso Polesano lacks the syntactic property that enables it to transmit the person and number features to a null subject, we will try to explain the difference between 26 and 27. This asymmetry between standard Italian and Basso Polesano can quite easily be captured within the terms of our assumption that a subject clitic in Basso Polesano is akin to an object clitic, in the sense that it starts from the basic argumental position and absorbs a theta-role. In standard Italian, French and NIDs, a referential specific object cannot be left phonetically empty:9 (30) a. Leggo il libro di Gianni. (I) read John's book b. Lo leggo. (I) it read c. Leggo. (I) read The sentences in 30a and b can be interpreted in the same way; that is, there is something specific that I read. Example 30c is grammatical only if the verb is interpreted as an intransitive: it expresses a statement about my ability to read, but there is no particular object that I am reading. This can easily shown be with a verb that cannot have an intransitive reading: (31) a. Preparo la cena. (I) prepare the dinner

280

PARAMETERS AND FUNCTIONAL HEADS

b.

La prepare. (I) it prepare

c. *Preparo. (I) prepare Example 31c is ungrammatical because the theta-role of the object of the verb has not been absorbed by an argument. This means that a null referential pronominal element cannot be licenced in object position. This is probably due to the fact that there are no features on the verb that can identify the contents of this null referential element. In this case an object clitic must be realized as an adjunct to the head of Agreement. The same is true for subjects in Basso Polesano: the verb is not specified enough to assign the relevant features to the null pronominal subject. Hence a null pronominal subject cannot be licensed. So, a subject clitic is adjoined onto the head of Agr in order to provide the necessary features. In other words, our analysis builds a parallel between the ungrammaticality of 26a and 31c. In both cases the sentences without the clitic are impossible because the null pronominal element is not licensed. The verb has no features that can identify the empty category in the subject and object positions, respectively. Let us examine how the mechanism of pro-drop works in Basso Polesano. The subject clitic starts from the basic position of the subject, where it leaves a trace coindexed with it, and moves up to the head of Agreement. The Spec-Agr position cannot be left totally empty because it is an argumental position. We will admit then that it is filled by a pro. Basso Polesano is a pro-drop language, like standard Italian, but it has different licencing conditions for pro. If the verbal morphology itself is not strong enough to licence a pro, the subject clitic is. It transmits its gender, number, and person features to the pro in Spec-Agr, which can be correctly identified.

The structure of a sentence like 26b will be as in 32,10 In 32 the clitic adjoined to Agr has moved from its basic subject position inside VP, where the trace t2 is. The clitic in Agr builds a chain with the pro in Spec-Agr, which transmits the identification features. In Basso Polesano the chain is established between the subject clitic and

THREE KINDS OF SUBJECT CLITICS

281

the null pronoun in Spec-Agr, while in standard Italian the chain is formed between the verbal morphology and the pro in Spec-Agr. In conclusion, we have argued for a new characterization of syntactic subject clitics as arguments that absorb the subject theta-role in the basic position inside the VP, move to Agr and then transmit the identification features to pro in Spec-Agr.11

2.2 Expletive clitics In this section we will examine the distribution of a, another subject clitic in Basso Polesano, which is obligatory with first and second person singular and plural:12 (33) a. A magno. A (I) eat b. *Magno. (I) eat c. A te magni. A you eat d. *Te magni. you eat e. A magnemo. A (we) eat f. *Magnemo. (we) eat g.

A magne. A (you) eat

h. *Magne. (you) eat The distribution of a is quite complicated because it not only appears with first and second person singular and plural, but also in contexts in which there is no argumental theta-role assigned to the subject as, for instance, with raising and weather verbs: (34) a. A piove. A rains b. A pare che Nane vegna qua. A seems that John comes here c. *Piove. rains d. *Pare che Nane vegna qua. seems that John comes here The examples in 34 shows that a is obligatory in these contexts, too. The distribution of a is quite strange: this clitic seems to be both an expletive and an argument. Our

282

PARAMETERS AND FUNCTIONAL HEADS

theory has to explain why a appears in both contexts and which position it occupies at S-structure.13 Let us begin by applying the tests presented in section 1.2, in order to determine which position this clitic occupies; namely, if it is a head adjoined onto the head of Agreement or an NP in Spec-Agr at S-structure. The test with the negative marker gives the following results: (35) a. A no vegno mina. A not (I) come not b. Mario no vien mina. Mario not comes not c. *No a vegno mina. not a (I) come not The clitic a always appears at the left of the preverbal negative marker, like subject NPs. As with French subject clitics and subject NPs, it cannot occur on the right side of negation (35c). This fact suggests that the position of a is the same as that of French subject clitics, namely the argumental preverbal subject position, Spec-Agr. The test with conjoined structures gives the same result: (36) A magno come un lupo e bevo tuto el df A eat (I) as a wolf and drink the whole day long The clitic a can be left out in the second conjunct in 36. This means that a is not bound to the verb as third person clitics are (cf. 15). In 36 the clitic a behaves like the subject NP of 15b and not like the third person subject clitic in 15a. We have to state that a occupies a different position from third person subject clitics, which cannot be left out in a coordination and appear at the right of the negative marker. At first sight, a seems to parallel the distribution of French subject clitics, which are NPs at S-structure. Hence we could conclude that a is a true NP at S-structure and that it is only a phonological clitic. We can now ask if a is an argument or not. We have seen that third person subject clitics are heads, but they are arguments in the sense that they absorb the subject thetarole. In fact, they never co-occur with another subject NP, if this is not dislocated, because they would compete for the same theta-role. In order to determine if a is an expletive or an argument, we can test if a co-occurs with another subject NP in argumental position or not. If it does, this means that it does not absorb the subject theta-role. It is not easy to exclude a dislocated structure in which the subject NP is in a left peripheral position because, in the case of first and second person, the only subject NPs are full pronouns, which can be freely dislocated: (37) a. Mi, a vegno casa. as for me, I conic home b.

Mi vegno casa. I come home

THREE KINDS OF SUBJECT CLITICS

283

Example 37 shows that the subject clitic is optional when there is a phonetically realized subject NP, but the meaning of the sentence changes, as can be seen in the glosses. For first and second person it is impossible to determine if 37a is a left-dislocation of the full pronoun or not, using the test of the quantifier presented in section 1.2 because there are no first person quantifiers. Nevertheless, 37a is to be considered as a case of left-dislocation, because the meaning and the intonation of the sentence are those of left-dislocated structures: there is a pause after the dislocated NP, which is old information. Moreover, the use of a left-dislocated position is always optional (as is 37), while the co-occurrence between a subject NP in Spec-Agr and a subject clitic in Agr is obligatory, in the dialects in which this is attested (but see note 3): (38) a. Gnun/'amagnagnent. nobody he has eaten nothing b. *Gnun a magna gnent. nobody has eaten nothing We could tentatively conclude that a is exactly parallel to first and second person subject clitics in standard French: it is an argumental NP that appears in Spec-Agr at S-structure and absorbs the subject theta-role. A theoretical complication that this hypothesis entails is that the subject clitics in Basso Polesano should be described as a mixed series: third person subject clitics being heads and a being an NP. The language would be pro-drop only in the case of third person singular and plural, but not in the case of first and second person. Leaving aside the difficulties that would arise in the formulation of the pro-drop parameter for Basso Polesano, there are other facts that seem to point in the direction that a is different both from third person subject clitics examined in section 2.1 and from French subject clitics. We have already noted that the position of a is not that of third person subject clitics. The natural thing to do now is to check if French subject clitics and a have exactly the same distribution. A does not seem to be completely parallel to French subject clitics, because it never appears in interrogative sentences: (39) a. *Goa da magnarlo? have-a to eat it b.

Dois-je le manger? must-I it eat

Example 39a shows that a is not grammatical in a direct interrogative context. In the French example (39b), on the other hand, the subject clitic is perfectly acceptable. Whatever the explanation of this fact turns out to be, there is a contrast between a and French subject clitics. This will most probably have to be explained in syntactic terms. In this case a does not pattern with French clitics, but with third person subject clitics as presented in section 2.1. In fact, third person subject clitics cannot appear in an interrogative context such as that shown in 39, neither on the left nor on the right of the verb: 14

284

PARAMETERS AND FUNCTIONAL HEADS

(40) a. *£7vien? he comes? b. *Viene/? comes he? The relevant feature that puts together a and third person subject clitics on one side and all French subject clitics on the other, could be that French clitics are NPs while all Basso Polesano subject clitics are heads, even if they are not all in the same position adjoined to the head of Agr. Another difference that distinguishes a from French subject clitics is, quite evidently, the different morphological specification of a and French subject clitics: in French there is a different subject clitic for every person: (41) a. Je mange. I eat b.

Tu manges, you eat

c. Nous mangions. we eat d.

Vous mangez. you eat

In contrast, the subject clitic of Basso Polesano is always the same morpheme for first and second person singular and plural: (42) a. A magno. A (I) eat b. A te magni. A you eat c. A magnemo. A (we) eat d. A magne. A (you) eat The subject clitic of Basso Polesano is interpreted as first or second person singular or plural only by virtue of the morphological specification of the verb, not of the clitic itself. This fact is quite odd because every NP has specifications of person and number. This leads us to consider that a is not a true NP. In this sense a is different both from French subject clitics and from third person subject clitics examined in section 2.1, which all have specifications for number and person. Another context in which a does not parallel first and second person French subject clitics, is that of weather and raising verbs, as in 34 (repeated here): (34) a. A piove. A rains

THREE KINDS OF SUBJECT CLITICS

285

b. A pare che Nane vegna qua. A. seems that John comes here c. *Piove. rains d. *Pare che Nane vegna qua. seems that John comes here The contrast between 34a, b and 34c, d shows that a is obligatory with weather and raising verbs. Under the assumption that a is an argument NP that appears in Spec-Agr, similar to first and second person French subject clitics, this is completely unexpected. If a is to be defined as an expletive clitic, as 34 seems to indicate, why is it obligatory with first and second persons too? As noted, this clitic seems to have a double nature of expletive and first and second person pronoun at the same time. This strange distribution deserves an expiation. On the basis of the differences noted above with respect to French, we will tentatively assume that a is not an NP, and thus it does not occupy the Spec-Agr position, but rather, that it is a head. We will render this claim more precisely as follows: a does not occupy the Spec-Agr position like French subject clitics because it is an expletive, because it has no person and number features as NPs have, and because it does not appear in interrogative contexts (as third person subject clitics do not). A does not occupy the same position as third person subject clitics either, because, as already mentioned, it appears before negation, and it can be left out in the second conjunct of a coordinate structure while third person subject clitics cannot. We will admit that the system of subject clitics in Basso Polesano is uniform, because all subject clitics are heads in this dialect. Nevertheless they occupy different head positions. If a occupies neither Spec-Agr nor a position of adjunction to the head of Agreement, there are not many possible sites left. Considering that a is a head, and that it has to be realized on a head position, there are two plausible candidates as hosts of a. One is the C position, the other is a higher Agreement head of which a constitutes the morphological realization. We will see that both options present some disadvantages. Before taking up the question of the position of a, we will examine the conditions under which a pro can appear in Basso Polesano. This will shed some light on the distribution of all subject clitics.

2.3 Licencing pro Given that all subject clitics of Basso_ Polesano are heads, even if they have different structural positions, then Basso Polesano is a null subject language for all persons. In section 2.1 third person subject clitics were analyzed as argumental heads that build a chain with a pro in Spec-Agr and assign person and number features to it. A possible speculation that comes to mind is to also connect the clitic a to the presence of a pro. In order to explain the distribution of the clitic a as connected to a null subject, we need a refinement of the theory of pro-drop. In Rizzi (1986a) a new proposal was made concerning the licencing condition of a null subject. Rizzi assumes that a

286

PARAMETERS AND FUNCTIONAL HEADS

pro, like every other empty category, has to satisfy two distinct conditions in order to be grammatical: a formal licencing condition, and a recoverability condition of its person and number features. The formal licencing condition coincides with the domain of nominative case assignment. When case is assigned to Spec-Agr a pro is formally licenced through Spec-head agreement with the head Agr. Both expletive and argumental pro have to satisfy this requirement. In standard Italian also the condition on the identification of the semantic content of the null pronoun is met through co-indexing with the head of Agreement that has the relevant person and number features. Rizzi observes that this requirement is valid only for an argumental pro that has a semantic content to be recovered. It is irrelevant for expletives, which have none. On the basis of this revised theory of pro-drop, let us examine the distribution of a. It must be relevant for both expletive and argumental null subjects, because it is obligatory in both contexts, whether an argumental subject theta-role has been assigned or not (cf. 33 and 34). The presence of a is relevant both for an argumental and for an expletive pro. Rizzi's revised Pro-drop Theory states that both argumental and expletive null pronouns have to be formally licenced. The obligatory realization of a both with first and second person and with expletive subjects can thus be traced back to the formal licencing requirement. We could capture this fact by assuming that a is relevant for the formal licencing of a pro. That is, a, and not Agreement (as in standard Italian), is the head that licences pro in Basso Polesano. The difference between standard Italian and Basso Polesano lies in the parametrization of the head that can licence pro. In standard Italian it is Agr, in Basso Polesano it is a, namely a clitic head. The formal licencing requirement for pro in Basso Polesano can be stated as follows: (43) A pro is formally licenced by the clitic head a through co-indexing. This hypothesis predicts that every time there is a pro to licence, a must be present too. In the case of an expletive pro, a formally licences pro, which does not need to be identified because it has no semantic content to be recovered. The data in 34 are correctly explained by this analysis: in 34a, b a pro is grammatical because it is correctly licenced by a; in 34c, d it is not because a is not present. Given that pro can only be licenced if a is present, we expect a to be obligatory every time there is an expletive pro in the structure. The current theory admits that in null subject languages like Italian, where a subject NP can surface in postverbal position, there is an expletive pro in Spec-Agr that is coindexed with the postverbal NP. Then a must also be obligatory with postverbal subjects, in order to formally licence the expletive pro in preverbal position: (44) a. ArivaToni. A comes John b. *RivaToni. comes John

THREE KINDS OF SUBJECT CLITICS

287

This prediction is borne out. The example 44, in which the subject NP is realized in the postverbal subject position, is correct only if a is realized, as in 44a. On the other hand, 44b is excluded because the expletive pro in preverbal position cannot be licenced, given that a is not present. We can therefore maintain our proposal that a is a "pro formal licencer." A formally licences pro for first and second person, too, but in these cases the identification requirement must be met. Where do the identification features for the argumental pro come from? One cannot assume that they stem from a itself because, as we have seen in 42, this clitic is not specified for number or person features. In fact, it is the same morpheme for first and second person, singular and plural. In 42, the verbal agreement morphology has been set in italics. Note that it is distinct both for person and number. We can make the conjecture that the identification features for pro come from the verb. A pro of first or second person is thus formally licenced by the clitic a and identified by verbal morphology. A pro of first or second person therefore enters two different chains: a chain with a, in order to be licenced, and a chain with verbal morphology, in order to be identified. This amounts to saying that the two requirements of formal licencing and identification can be satisfied by different heads. This state of affairs is not as strange as it may first appear. On the contrary, it is predicted by Rizzi's theory of pro-drop. Once the two requirements are independently formulated, there is nothing that prevents them from being satisfied by different heads.15 Basso Polesano is such a case: this confirms the idea of the splitting of the formal licencing and the identification of pro is correct. Let us now turn to the analysis of third person subject clitics examined in section 2.1. In the case of third person clitics we saw that the presence of a subject clitic was crucial to the possibility of having a pro in Spec-Agr (example 26, above). In the revised version of the Pro-drop Theory that we are assuming, this means that subject clitics are relevant for the licencing of a pro (as a is), or for the identification of its content features, or for both. As shown in section 2.1, third person subject clitics must be relevant for the identification requirement of the argumental null pronoun because they are specified for both number and person features while verbal morphology is not. We can assume that third person subject clitics satisfy the identification condition, being co-indexed with the pro. Third person subject clitics must be relevant for the formal licencing of pro as well. In the discussion about a, condition 43 on the licencing of pro has been formulated in a way which excludes the possibility that verbal agreement morphology itself could have the ability to formally licence a pro. On this basis, we would expect the clitic a to be able to appear with third person pro too: (45) a. *A el vien. A he comes b. El vien. he comes However, in 45 a rather unexpected result is obtained: the presence of a is ungrammatical. This fact seems to conflict with the parameter proposed in 43, and a

288

PARAMETERS AND FUNCTIONAL HEADS

reformulation seems to be necessary. This can be done through a very simple observation: third person subject clitics can be relevant to the formal licencing requirement for pro too. This can quite easily be captured if the pro licencing condition in Basso Polesano is formulated in the following way: (46) A pro is formally licenced by a clitic head co-indexed with it. As a consequence of 46, every time there is no subject NP in Spec-Agr, there must be a clitic co-indexed with the Spec-Agr position, where a pro can thus be licenced. In the case of an expletive pro, a must appear. It formally licences the expletive pro which does not need identification. In the case of first and second person, a licences pro and the agreement morphology identifies it. In the case of third person, the agreement morphology is not specified enough to identify pro, and so another subject clitic is used: third person subject clitics licence pro and identify it. They build a thematic chain with pro, through which the thematic role, the identification features (and probably the nominative case) are shared. The condition in 46 explains why a is not needed when there is another subject clitic of third person. However, it does not explain why the co-occurrence between a and a third person clitic is ungrammatical (cf. 45a). This constraint on the co-occurrence of two elements that have the same function— in this case, the formal licencing of a pro — is reminescent of a general Economy Principle, proposed in its most definite form in Chomsky (1988). A syntactic application of this Economy Principle could be formulated in terms of a constraint that prevents two different heads from licencing the same item. The idea is akin to a sort of minimality, where two heads cannot govern the same element, but only the hierarchically nearer does. A solution along these lines has been independently proposed in Roberts (1990). We will not pursue this matter further here. Instead we will come back to the problem considered in the last section, namely the position of a. We examined two possibilities: a could have the same position as third person subject clitics, or it could be in Spec-Agr, like French subject clitics. Both options, however, will be discarded. The first option is rejected because a appears on the left of the preverbal negative marker while subject clitics appear on the right, and because a can be left out in the second conjoint of a coordinate structure, while third person subject clitics cannot. The second one is unworkable because a cannot appear in an interrogative context (while French subject clitics can), and because a does not have person and number specifications, which all NPs have (at least in Italian and French). The clitic a seems to be distinct both from third person subject clitics and from phonological clitics like French ones. If a is not in Spec-Agr and is not adjoined to the head of Agr, the number of possible positions that it can occupy is rather restricted. As mentioned in section 2.2,there are two more possible positions that can be considered. In recent work by Cardinaletti and Roberts (1990) it was proposed that some languages, including NIDs such as Trentino and Florentine, have two Agreement projections, as shown in 47. In their analysis the verb moves up to the head of Agr2 and subject clitics are generated in Agrj. The subject NP is realized in the Spec of Agr,. A pro can be generated both in Spec-Agr, or in Spec-Agr2.

THREE KINDS OF SUBJECT CLITICS

289

Applying this analysis to Basso Polesano, we could assume that the verb moves up to the head of Agr2 and third person subject clitics are adjoined to it. The pro identified by the third person subject clitics is in Spec-Agr2, where it builds a chain with the clitic through Spec-head agreement. A subject NP can be realized only in a left-dislocated position, probably adjoined to the highest projection of the sentence. As already mentioned it cannot be realized in Spec-Agr2 or in Spec-Agrj because the third person subject clitic already absorbs the subject theta-role. The Agrj projection remains empty — it is active only when the clitic a is present. In this case, the verb moves to Agr2 and a occupies the head of Agrj. A first or second person pro is realized in Spec-Agr2, and is licenced by a uad identified by the verb. The data in 37 show that a is in complementary distribution with a subject NP in argumental position. A full pronoun of first or second person can only occupy a left-dislocated position (cf. Beninca 1983 for a discussion on full pronouns). Thus, in the structure shown in 47, we have to ensure that both Specifier positions, Spec-Agrj and Spec-Agr2, remain phonetically empty. The fact that two Spec-Agr positions are available constitutes a potential problem, because there is no way to prevent ungrammatical sequences. In other words, 47 overgenerates. Let us examine the data. Given that a licences a pro, one of the two positions is occupied by this pro, but there is a Spec-Agr position still available to a subject NP. In 34 two examples of expletive a were presented; namely, cases where a subject theta-role had not been assigned and a was obligatory all the same. On this basis we have assumed that a can licence an expletive pro. A structure like that in 47, in which two distinct Spec positions are realized, predicts that a should be compatible with a subject NP. Example 48 shows that this is not true; 48 cannot be ruled out by the Projection Principle because a does not absorb the theta-role, which is available for the subject NP Nane: (48) *Naneavien. John A comes The structure of a sentence such as 48 corresponds to 49, where the subject NP Nane is in Spec-Agrj, a occupies the head of Agr1( an expletive pro is licenced in Spec-Agr2 and the verb is in Agr2. Under the assumption that a constitutes the head of a higher Agreement projection, there is no way to prevent a sentence like 48 from being generated.16

290

PARAMETERS AND FUNCTIONAL HEADS

A structure with two Spec-Agr positions also permits a sequence as shown in 50, in which the pro is in Spec-Agr j and the subject NP in Spec-Agr2: (50) *ANanevien. A John comes But this is excluded too. The impossibility of excluding sentences such as 48 and 50 constitutes an empirical counterargument to the analysis presented in 47. There is also a conceptual reason not to assume a to be in a higher Agreement projection. The clitic a of Basso Polesano cannot be considered an agreement morpheme (like Trentino and Fiorentino clitics) because it never shows any agreement features. As already mentioned, it is invariable in all contexts. For these reasons we cannot assume that a is generated as the head of a higher Agreement projection. Nevertheless, the intuitive idea that a is the head of a higher projection can be maintained. There is another possibility worth exploring: a can be base-generated in C. Claiming that a is in C amounts to saying that C has a nominal feature on it, similar to embedded clauses. The structure of Basso Polesano would be as in 51:

In 51 there is a pro in Spec-Agr which has to be licenced by a head. In standard Italian a pro is both formally licenced and identified through Spec-head agreement with the

THREE KINDS OF SUBJECT CLITICS

291

head of Agr. In Basso Polesano a licences pro through government from C into the Spec-Agr position.17 If the pro is an expletive, it does not need any identification. If it is a first or second person pro it needs to be identified. Consequently, it must be in a chain with the head of Agr too, in order to receive the number and person features coming from the agreement morphology. As we have noted, the first or second person pro in Spec-Agr enters two different chains, one with a, in order to be licenced through government, and another with the agreement morphology, so as to be correctly identified as first or second person. A structure like 51 does not show the disadvantages of 47. In fact, in 51 there is only one Spec-Agr position, which is already occupied by pro. It cannot be filled by a subject NP too, so sequences such as 48 and 50 are automatically excluded, given that there is only one Spec-Agr position. Assuming that a is base-generated in C, we can distinguish this clitic from other subject clitics in a crucial way: third person subject clitics are connected to the base subject position while a is not. A appears to be a morpheme specialized for the licencing of pro. Intuitively speaking, when the Agreement head does not contain an element capable of licencing pro, as third person subject clitics, the head of the Comp projection becomes relevant as pro licencer. In order to do this, C has to be visible: a is the special realization of a licencing Comp. Attractive as this may appear at first sight, even this analysis of a as a realization of Comp has some disadvantages: the C projection does not normally play any role in main declarative sentences in standard Italian. Only in V2 languages is the C-position occupied (by the verb) in main declaratives. The case of a would be the only case in which a C is active in a matrix declarative sentence (but see note 17). There is another interesting fact about the distribution of a which should be mentioned. In the case of an embedded sentence a can co-occur with a complementizer, as che 'that' or se 'if. In this case the clitic and the complementizer constitute a morphological unit: the final vowel of the complementizer, the e of che or se disappears: (52) a. A no so sa vegno. A not (I) know if+A (I) come b. El vole ca vegna. he wants that +A (I) come Both the complementizer and the clitic can survive together. This is not really expected if a is the realization of C: the C-position seems to be occupied by two elements, the complementizer and a. A possible solution to this problem is to admit that the complementizer and a form a complex morpheme, which expresses the selectional features coming from the matrix verb and at the same time has the ability to licence pro. Note, however, that even third person subject clitics, which we assumed to be in Agr, can appear on the complementizer: (53) A so chf vien. (I) know that+they come Considering that 52 and 53 appear as perfectly symmetric, the coalescence between the complementizer and a could be interpreted as a phenomenon of cliticization of

292

PARAMETERS AND FUNCTIONAL HEADS

a to Comp from a lower head. In this way both 52 and 53 would be the result of a movement of the clitic from an Agreement head (recall that third person subject clitics are adjoined to the Agr where the verb is). The symmetry between 52 and 53 seems to favor an explanation as shown in 47, in which a is the head of a higher Agreement projection with respect to an hypothesis that considers a as the realization of Comp. However, against the proposal that treats the morphemes ca and sa as a case of cliticization, there is the fact that the cliticization of third person subject clitics is not completely parallel to that of a: the cliticization of third person subject clitics to the complementizer is optional:18 (54)

A so che i vien. A (I) know that they come

Example 54, in which the subject clitic has not moved to the complementizer, is a grammatical sentence. The same is not true in the case of 55: (55) *A so che a vegnaro. A (I) know that A (!) will come The contrast between 54 and 55 cannot be captured if the cliticization of third person subject clitics and the coalescence of a and the complementizer are explained as instances of the same phenomenon, namely movement from a lower head. Neither analysis of a as head of Agreement nor as head of Comp seems to be completely satisfactory. Under the hypothesis that a is the head of Agrj, one cannot explain why a does not show agreement features, and why only one of the two Specifiers of Agr can be occupied and not both. Under the hypothesis that a is in C, one cannot explain why this is the only case in which a C is active in a declarative sentence, and how the coalescence between a complementizer and a can be grammatical. The position of a is already negatively defined as a head position, different from the position of subject NPs and third person subject clitics too. The actual theory on the structure of the sentence does not permit us to formulate a definitive solution that can capture all the data we have presented. We will therefore leave the problem of the position of a open for further research.

2,4 An inflectional clitic The third type of subject clitic we shall consider is the second person subject clitic te. On the basis of the distinction outlined in section 2.3, we expect te to be a "licencer" or an "identifier" for pro. We will first apply the tests seen in section 2.1 to determine what the position of te is. Te is undoubtly a head bound to the verb as third person subject clitics are — it cannot be left out in the second occurrence of a coordination: (56) a. *A te va e vien. A you go and come

THREE KINDS OF SUBJECT CLITICS

293

b. A te va e te vien. A you go and you come The examples in 56 show that te is not a true subject NP but rather, a morpheme bound to the verb. In fact, unlike NPs, it appears on the right of the negative marker, like third person clitics: (57) a. A no te magni. A not you eat b. *A te no magni. A you not eat These two tests suggest that te is a head in the same position as third person subject clitics — namely, adjoined to Agr. Note, however, that in both 56 and 57 the examples also present a clitic a. If a is not present the sentence is ungrammatical: (58) a. *Te magni. you eat b. A te magni. A you eat We have to admit that, in the case of the second person singular, two subject clitics are necessary: a and te. In the previous section the obligatory presence of a was expressed as deriving from a licencing condition for pro. A was defined as a pro licencer: it is necessary whenever there is no subject NP occupying the Spec-Agr position and whenever there is no third person subject clitic that can licence pro from Agr. Therefore, the fact that a is necessary in 58 means that te, contrary to third person subject clitics, is not able to licence pro (cf. 45 for third person subject clitics). If a really is the pro licencer in this context too, it should disappear when a second person full pronoun is present in Spec-Agr, because in this case there is no pro to licence: (59) a. A te magni. A you eat b. Ti te magni. YOU you eat Example 59b shows that this prediction is correct: whenever a subject NP occupies the Spec-Agr position, a is not realized anymore. The function of a is clear: it licences the second person singular pro. As in the case of first person singular and plural and second person plural, the pro cannot be identified by a, given that a does not have person and number features. The second person identification features must come from another source. Both the verbal agreement morphology (namely the vowel i that appears on the verb exactly as in standard Italian) and the clitic te are possible candidates as pro identifiers because they both express number and person features. If te were the pro

294

PARAMETERS AND FUNCTIONAL HEADS

identifier, we would predict that it would appear only when there is a pro to identify, just as a appears only when there is a pro to licence. The distribution of te is rather surprising: the second person subject clitic te is always obligatory, even if there is a full pronoun in Spec-Agr, as in 60: (60) a. *Timagni. YOU eat b.

Ti te magni. YOU you eat

The examples in 60 show that te is not sensitive to the element present in Spec-Agr: te must always be present, whether Spec-Agr is occupied by a pro or by a subject NP. Therefore, we will accept that te is not a pro identifier, because it is not affected by the presence of a pro. The pro identifier is the verbal morphology as in the cases of first person singular and plural and second person plural, and as in standard Italian. Now the problem becomes the following: if a already licences pro and verbal morphology identifies it, what is the function of te with respect to pro? We are compelled to assume that te has none — it is neither a pro licencer nor a pro identifier. Note that te is obligatory not only with a preverbal subject NP as 60 shows, but also with a postverbal one: (61) a. * A magni ti qua. A eat YOU here b.

A te magni ti qua. A you eat YOU here

Te is not affected by the element in Spec-Agr, but it is also obligatory even when the basic subject position inside the VP is occupied. The position of te is the same as that of third person subject clitics: it is an adjunct to the head of Agreement. But even if te seems to occupy the same position that third person subject clitics occupy, it must be kept distinct from them. In section 2.1 we showed that third person subject clitics are sensitive to the element in Spec-Agr or in the basic subject position inside the VP: only when these positions are empty is the subject clitic grammatical (we are always abstracting away from left-dislocation): (62) a.

Lu magna qua. HE eats here

b.

Qua magna lu. here eats HE

As the examples in 62 show, third person subject clitics are not realized when a subject NP is in the argumental pre- or postverbal position, unlike te, which is always obligatory.19 The contrast between 60/61 and 62 must be captured within our analysis. If we admit that te and third person subject clitics occupy the same position at S-structure, seeing as they all appear after the preverbal negative marker and must be repeated in a conjoined structure (cf. 56 and 57), the difference between te and third person subject clitics cannot be expressed in terms of different syntactic positions.

THREE KINDS OF SUBJECT CLITICS

295

Third person subject clitics are arguments that absorb a theta-role. As such they are in complementary distribution with other subject NPs which need the subject theta-role. Te occupies the same position that third person subject clitics occupy but it is not an argument and it does not absorb the theta-role of the subject. Hence te can always be present: whether there is no subject NP present, or whether it is in the preverbal position or in the basic postverbal subject position (cf. 60 and 61). We have assimilated the distribution of third person subject clitics to object clitics that cannot occur when an object NP is in the argumental object position. In both cases the chain clitic-trace absorbs the theta-role assigned from the verb, as all arguments do. In section 2.1 (examples 22-24, repeated here), we discussed a prediction made by the hypothesis that third person subject clitics are arguments: they are incompatible with the trace of a subject: (22) a. la putina che vien vanti... the girl that comes along b. *la putina che la vien vanti... the girl that she comes along (23) a. MARIO, ga ciama, no Toni. MARIO, has called, not Tony b. *MARIO, el ga ciama, no Toni. MARIO, he has called, not Tony (24) a. A ze Mario, che ga ciama. A is Mario, that has called b. *A ze Mario, che el ga ciama. A is Mario, that he has called In the above examples, the third person subject clitic cannot occur in a structure, in which a subject has been extracted because they compete with the subject trace for the subject theta-role. If our hypothesis that te does not absorb the theta-role of the second person subject is correct, te should be compatible with the trace of the subject, showing the opposite effect to third person subject clitics. We will again use topicalization, cleft, and restrictive relatives as tests, given that they are instances of movement: (63) a.

TI, te ga bevu el vin, no mi. YOU, you have drunk the wine, not me

b. *TI, ga bevu el vin, no mi. YOU, have drunk the wine, not me c.

Te si ti, che te ga bevu el vin. you are YOU, that you have drunk the wine

d. *Te si ti, che ga bevu el vin. you are YOU, that have drunk the wine

296

PARAMETERS AND FUNCTIONAL HEADS

e. Ti, che te ga bevu el vin, . . . YOU, that you have drunk the wine . . . f. *Ti, che ga bevu el v i n , . . . YOU, that have drunk the wine , . . The examples in 63 confirm that te is always obligatory, even when the subject has been moved, and in the subject position there is a variable. Hence our hypothesis that te is not affected by the element in subject position is confirmed. The subject NP can be a null subject pro, an NP, or a variable, and te must always be present. We have to distinguish it from both a and third person subject clitics. The difference between a and te is a matter of position, a being higher in the structure, as shown in sections 2.2 and 2.3. As we noted, the examples in 56 and 57 show that te appears in the same position that third person subject clitics occupy at S-structure, hence the difference between te and third person subject clitics cannot be captured in terms of different syntactic positions. The distribution of te seems to be similar to the subject clitics of the dialects studied in Brand! and Cordin (1989) and Rizzi (1986b). In this case the subject clitic is the expression of some inflectional feature. As such it can be base-generated in Agr as agreement features are. On the basis of this observation, we cannot assume (as we did for third person subject clitics) that te starts from the basic subject position inside the VP and then cliticizes onto Agr. The distinction between third person subject clitics and te can be captured in terms of movement versus base-generation. Third person subject clitics are generated in the basic subject position, receive the subject theta-role, and move to Agr, leaving a trace. Te, on the other hand, is base-generated in Agr. It does not originate in an NP argumental position, hence it does not absorb a theta-role. This is the reason why it is compatible with a variety of subject NPs in argumental subject position. The facts and discussion above lead us to introduce a new distinction in the group of syntactic subject clitics: movement from an argumental position versus base generation in Agr. The second person subject clitic te is base-generated in Agr, and is not moved to Agr from the basic argumental position as third person subject clitics and object clitics are. We will thus keep te distinct from other subject clitics examined here, assuming that Basso Polesano has three different types of subject clitics. Third person subject clitics are equivalent to object clitics, because they are argumental heads generated in basic argument position. A is an expletive clitic needed for the formal licencing of a null subject, and is higher than Agreement in the structure of the sentence. Te is parallel to an inflectional morpheme, is base-generated in Agr, and does not absorb the subject theta-role.

3

Conclusion

Starting from the observation that the concept of clitic is spurious, we have examined the subject clitic system of Basso Polesano. In this dialect there are three distinct types of subject clitics, distinguishable on the basis of different features: the position

THREE KINDS OF SUBJECT CLITICS

297

in which the clitic is realized, the relevance for a null subject, and the movement from an argumental basic position: a. We have seen that syntactic subject clitics can be distinguished on the basis of their position, in which case a, which appears higher in the structure, is different from te and third person clitics which are adjoined to Agr. b. They can be distinguished on the basis of their argumental status, in which case a and te do not absorb a theta-role, while third person clitics do. c. The third distinguishing feature is that between clitics that are relevant for pro (which are a and third person subject clitics) and te which reamains unaffected by a null subject. The conditions on the appearance of a null subject in Basso Polesano have led us to conclude that the implicit prediction made by Rizzi's theory concerning the division of the formal licencing from the identification conditions of pro is correct. If the formal licencing has to be kept distinct from the identification of the empty category pro, we expect the two requirements to be met by two different heads. This is exactly the situation shown by Basso Polesano: a clitic head, a, licences pro, while the verbal agreement identifies it. This view of the facts expresses an intuition about the role of verbal agreement, namely that verbal agreement in the northern Italian dialects is not as strong as in standard Italian, but not as weak as in standard French.

Notes 1.1 am deeply indebted to Adrian Battye, Adriana Belletti, Paola Beninca, Guglielmo Cinque, Maria Teresa Guasti, Richard Kayne, Luigi Rizzi, Ian Roberts, Alessandra Tomaselli, Laura Vanelli, and Raffaella Zanuttini for helpful comments and discussion. All errors are naturally my own. 2. We have italicized the subject clitics in order to render the examples more comprehensible. The full form of the pronoun is translated into English in capital letters, the clitic a is translated as "A". 3. The Trentino and Piedmontese examples in Rizzi (1986b) are all formulated with compound tenses, in which an auxiliary verb appears in Agr. There is evidence that the subject clitic of these structures is connected with the presence of an auxiliary because the sequence subject quantifier-subject clitic-verb is ungratnmatical in simple tenses, as in (i) (cf. Giupponi 1988): (i) *Nient el capita de not. It seems that auxiliaries need a clitic in Agreement: in some Valdotain varieties studied by Roberts (1990), auxiliaries present a distinct series of clitics, which is complete for all persons, but not for main verbs. In all Veneto dialects, too, a clitic is obligatory on the auxiliary verbs be and have, but it is an indirect or locative object clitic, as in (ii): (ii) El ga fame. he CL has hunger The clitic g in (ii) is clearly an object clitic because of its form, and because it alternates (in some varieties) with other object clitics, as in (iii):

298

PARAMETERS AND FUNCTIONAL HEADS (iii) Elm'avisto. he me has seen

As the form of subject clitic and object clitic in Trentino and Piedmontese is the same, namely /, the example of Veneto may suggest that the I that appears in example 6 could be an object clitic too ,and not a subject clitic. This area clearly deserves further investigation and more comparative data from other NIDs. 4. It is necessary to distinguish between VP coordination and verb coordination in which the subject clitic can be omitted, as in (i): (i) El parla e riparla sempre dela stesa roba. he speaks and re-speaks always of the same thing This is parallel to the case in which an object clitic, which is indoubtedly a head, can be omitted too: (ii) Lo leso e rileso sempre. it (I) read and re-read always For a detailed analysis of these cases, see Beninca and Cinque (1993). I will limit the tests to VP-coordination, in which a syntactic subject clitic has to be repeated. 5. The realization of a postverbal subject is constrained by some informational conditions, examined in Calabrese (1982, 1990). For a detailed analysis of postverbal subjects see also Rizzi (1982) 6. All the examples contain a subject quantifier in order to exclude a right-dislocation of the subject. In this way the subject clitic and the subject NP can co-occur: (i) El vien, Nane. he comes, John In example 19b, a clitic a appears in preverbal position. This clitic has a totally different distribution from third person subject clitics, as will be discussed in section 2.3. 7. We will not discuss the phenomenon of clitic doubling which is attested in various languages such as Spanish, Romanian, etc. This seems to be restricted to indirect objects or direct objects with a preposition (see Kayne 1989a and Jaeggli 1985). 8. We will not consider direct interrogative sentences here because in this context another series of subject clitics appears on the right of the verb. Hence interrogatives are not relevant here. Extraction of a wh, on the other hand, can constitute a test, in exactly the same way as topicalization, clefts, and restrictive relatives can: (i) Chi cridito che sia vegnii who believe+you that has come (ii) *Chi cridito che el sia vegnii. The results of the test are the same for examples 22-24: the subject clitic cannot appear if there is a variable in basic subject position. See Beninca and Vanelli (1984) for a comparative discussion of this phenomenon. 9. A null object pro is possible in standard Italian and French only if it is interpreted as an arbitrary element, not if it is a specific referential element (see Rizzi 1986a). 10. The TP projection has not been included in the structure because it is not relevant to the hypothesis considered here. 11. The distribution of third person subject clitics is the same in all central Veneto dialects, and probably in other varieties too. Basso Polesano should therefore not be considered an idiosyncratic dialect which has developed in a completely different way from other varieties, but as quite a common case. 12. For an analysis of the subject clitic te, see section 2.4

THREE KINDS OF SUBJECT CLITICS

299

13. A subject clitic a exists in other dialects, too, such as Padovano or Vicentino, but it has a different distribution. For an analysis that treats these cases see Beninca (1983), where it is argued that a is a sort of topic marker and not a true subject clitic as it is in Basso Polesano. 14. Interrogative sentences show a phenomenon of subject clitic-verb inversion, which is constrained by the presence of other clitics. We will not analyze it here because it would lead us too far from the topic of this paper. 15. Even in standard Italian it is marginally possible to find cases, in subjunctive contexts, in which a head licences pro and another identifies it. A sentence such as (i), below, is totally excluded if it is interpreted as second person singular, evidently because the verbal morphology is not sufficiently specified to identify the pro: (i) *Spero che venga. (I) hope that (you) come If there is an anaphoric object clitic on Agr, the sentence is better, even though not totally grammatical: (ii) ??Desidero che ti faccia bella per stasera. 16. Cardinaletti and Roberts (1990) formulated their hypothesis for other dialects, such as Trentino and Fiorentino, in which the order Subject NP-subject clitic-verb is grammatical. Hence their analysis does not run into the problem that a structure like 47 presents in Basso Polesano. 17. Cases of licencing of a pro under government are well known in the literature: the alreadymentioned case of a null object arbitrary pro in standard Italian; the pro licenced by C in V2 languages (cf. Tomaselli 1990 for Germanic languages; Beninca 1989 for Ancient Romance). Beninca (1986) examines a more relevant case for our discussion, namely Provenjal, a non-V2 language in which a null subject is licenced by a complementizer in assertive main clauses: (i) Que t parli. that to-you (I) speak Provencal shows that even in non-V2 languages, C can be active in assertive main clauses and it can licence a pro. 18. The sentence is grammatical with a pause after the complementizer, and with emphasis on what follows. 19. We use the pronoun in the example with third person subject clitics too, in order to render the comparison fully equal. See section 2.1 for discussion on the distribution of third person subject clitics.

References Belletti, A. 1990. Generalized Verb Movement. Aspects of Verb Syntax. Turin: Rosenberg and Sellier. Beninca, P. 1983. "II clitico a nel dialetto padovano," in P. Beninca et al. (eds.), Scritti Linguistici in onore di Giovan Battista Pellegrini. Pisa: Pacini. 25-35. . 1984. "L'interferenza sintattica: di un aspetto della sintassi ladina considerate di origine tedesca," Quaderni Patavini di Linguistica 5:3-17. . 1986. "Punti di sintassi comparata dei dialetti settentrionali," in G. Holtus and K. Rinpper (eds.), Raetia Antiqua et Moderna. Festschrift fur W. Theodor Elwcrt zum 80. Geburtstag. Tubingen: Niemeyer. 457^479. Beninca, P. and G. Cinque. 1993. "Su alcune differenze tra enclisi e proclisi," in Omaggio a Gianfranco Folena. Padova: Editoriale Programma. 2313-2326.

300

PARAMETERS AND FUNCTIONAL HEADS

Beninca, P. and L. Vanelli. 1982. "Aspetti sintattici del portogruarese tra veneto e friulano," in R. Sandron (ed.), Atti del convegno di Portogruaro il 18-19 die. 1982. Portogruaro: Tipografia portogruarese. -. 1984. "Appunti di sintassi veneta," in M. Cortelazzo (ed.), Cuida ai dialetti veneti 4. Padua: CLEUP. 7-38. Brandi, L. and P. Cordin. 1981. "Dialetti e italiano: un confronto sul parametro del soggetto nullo," Rivista di Grammatica Generativa 6:33-87. . 1989. "Two Italian Dialects and the Null Subject Parameter," in O. Jaeggli and K. Safir (eds.). The Null Subject Parameter. Dordrecht: Kluwer. 111-142. Calabrese, A. 1982. "Alcune ipotesi sulla struttura informazionale della frasein italiiano e sul suo rapporto con la struttura fonologica," Rivista di Grammatica Generativa 7:3-78. . 1990. "Some Informal Remarks on Focus and Logical Structures in Italian," ms, Harvard University. Cardinaletti, A. and I. Roberts. 1990 "Levels of Representation of Agreement," ms, University of Geneva/University of Venice. Chomsky, N. 1977. "On Wh-Movement," in P. Culicover, T. Wasow, A. Akmajian (eds.), Formal Syntax. New York: Academic Press. 71-132. . 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. . 1990. Types of A-bar Dependencies. Cambridge, Mass.: MIT Press. Giorgi, A. and G. Longobardi. 1991. The Syntax ofNPs: Configuration, Parameters and Empty Categories. Cambridge: Cambridge University Press. Giupponi, E. 1988. "Pro-drop und Restrukturierung im Trentino," tesi di Laurea, University ofTrento. Jaeggli, O. 1985. "Three Issues in the Theory of Clitics: Case, Doubled NPs and Extraction, ms., University of Southern California. Kayne, R.S. 1975. French Syntax. Cambridge, Mass.: MIT Press. . 1989a. "Facets of Romance Past Participle Agreement," in P. Beninca (ed.), Dialect Variation and the Theory of Grammar. Dordrecht: Foris. 85-103. . 1989b. "Null Subjects and Clitic Climbing," in O. Jaeggli and K. Safir (eds.), The Null Subject Parameter. Dordrecht: Kluwer. 239-269. Koopman, H. and D. Sportiche. 1988. "Subjects," ms, University of California, Los Angeles. Pollock, J.-Y. 1989. "Verb Movement, Universal Grammar, and the Structure of IP," Linguistic Inquiry 20:365-424. 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. Roberts, I. 1990. "Verbs and Diachronic Syntax," ms, University of Geneva. Tomaselli, A. 1990. IM sintassi del verbo finite nelie lingue germaniche. Padova: Unipress. Vanelli, L. 1987. "I pronomi soggetto nei dialetti italiani settentrionali dal Medio Evo a oggi," Medioevo Romanzo, Anno XII, 1:173-221.