Subjects, Expletives, and the EPP
Recent Titles in OXFORD STUDIES IN COMPARATIVE SYNTAX Richard Kayne, General Editor Particles: On the Syntax of Verb-Particle, Triadic, and Causative Constructions Marcel den Dikken
XP-Adjunction in Universal Grammar: Scrambling and Binding in Hindi-Urdu Ayesha Kidwai
The Polysynthesis Parameter Mark C. Baker
Multiple Feature Checking and Grammatical Functions in Universal Grammar Hiroyuki Ura
The Role of Inflection in Scandinavian Syntax Anders Holmberg and Christer Platzack Clause Structure and Word Order in Hebrew and Arabic: An Essay in Comparative Semitic Syntax Ur Shlonsky Negation and Clausal Structure: A Comparative Study of Romance Languages Raffaella Zanuttini Tense and Aspect: From Semantics to Morphosyntax Alessandra Giorgi and Fabio Pianesi Coordination Janne Bondi Johannessen Adverbs and Functional Heads: A Cross-Linguistic Perspective Guglielmo Cinque The Feature Structure of Functional Categories Elabbas Benmamoun A Handbook of Slavic Clitics Steven Franks and Tracy Holloway King
The Higher Functional Field: Evidence from Northern Italian Dialects Cecilia Poletto The Syntax of Verb-Initial Languages Edited by Andrew Camie and Eithne Guilfoyle Parameters and Universals Richard Kayne Portuguese Syntax: New Comparative Studies Edited by Joao Costa Infinitive Constructions: A Syntactic Analysis of the Romance Languages Guido Mensching Subject Inversion in Romance and the Theory of Universal Grammar Edited by Aafke Hulk and Jean-Yves Pollock Subjects, Expletives, and the EP Edited by Peter Svenonius A Unified Theory of Verbal and Nominal Projections Yoshiki Ogawa
Subjects, Expletives, and the EPP
Edited by PETER SVENONIUS
OXFORD UNIVERSITY PRESS
2002
OXFORD UNIVERSITY PRESS
Oxford New York Auckland Bangkok Buenos Aires Cape Town Chennai Dar es Salaam Delhi Hong Kong Instanbul Karachi Kolkata Kuala Lumpur Madrid Melbourne Mexico City Mumbai Nairobi Sao Paulo Shanghai Singapore Taipei Tokyo Toronto
and an associated company in Berlin Copyright © 2002 by Oxford University Press, Inc. Published by Oxford University Press, Inc. 198 Madison Avenue, New York, New York 10016 www.oup.com 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 the prior permission of Oxford University Press. Library of Congress Cataloging-in-Publication Data Subjects, expletives, and the EPP / edited by Peter Svenonius. p. cm. — (Oxford studies in comparative syntax) Papers presented at a workshop held June 1997, University of Troms0. Includes bibliographical references and index. Contents: The que/qui-alternation and the distribution of expletives / Knut Tarald Taraldsen — Icelandic expletive constructions and the distribution of subject types / 0ystein Alexander Vangsnes — Expletives, subjects, and topics in Finnish / Anders Holmberg and Urpo Nikanne—The EPP in a topic-prominent language / Katalin E. Kiss — The extended projection principle as a condition on the tense dependency / Ian Roberts and Anna Roussou — Parameters of subject inflection in Italian dialects / M. Rita Manzini and Leonardo M. Savoia — Subject positions and the placement of adverbials / Peter Svenonius. ISBN 0-19-514224-1; 0-19-514225-X (pbk.) 1. Grammar, Comparative and general—Syntax—Congresses. I. Svenonius, Peter. II. Series. P291 .S83 2002 415—dc21 2001036744
1 3 5 7 9 8 6 4 2 Printed in the United States of America on acid-free paper
Contents
Contributors 1. Introduction PETER SVENONIUS
vii 3
2. The Que/Qui Alternation and the Distribution of Expletives KNUT TARALD TARALDSEN
29
3. Icelandic Expletive Constructions and the Distribution of Subject Types 0YSTEIN ALEXANDER VANGSNES
43
4. Expletives, Subjects, and Topics in Finnish ANDERS HOLMBERG AND URPO NIKANNE
71
5. The EPP in a Topic-Prominent Language KATALIN E. Kiss
107
6. The Extended Projection Principle as a Condition on the Tense Dependency 125 IAN ROBERTS AND ANNA Roussou 7. Parameters of Subject Inflection in Italian Dialects M. RITA MANZINI AND LEONARDO M. SAVOIA
157
8. Subject Positions and the Placement of Adverbials PETER SVENONIUS
201
Index
243
This page intentionally left blank
Contributors
ANDERS HOLMBERG University of Troms0 KATALIN E. Kiss Linguistic Institute of the Hungarian Academy of Sciences RITA M. MANZINI University of Florence URPO NIKANNE University of Oslo IAN ROBERTS University of Stuttgart (currently at University of Cambridge) ANNA Roussou University of Bangor (currently at University of Cyprus) LEONARDO M. SAVOIA University of Florence PETER SVENONIUS University of Troms0 KNUT TARALD TARALDSEN University of Troms0 0YSTEIN ALEXANDER VANGSNES University of Bergen
This page intentionally left blank
Subjects, Expletives, and the EPP
This page intentionally left blank
1 Introduction Peter Svenonius
The literature on the three concepts named in the title of this book is vast. In this introduction, I simply provide a sketch (in § 1-3) of the issues that set the scene for the chapters in the volume. My discussion is judiciously larded at appropriate points with brief references to those chapters, and each chapter is summarized in §4. In §5,1 discuss the chapters in relation to each other, and then summarize some of their shared assumptions.
1. Subjects The subject occupies a precarious position in generative linguistics; on the one hand it is an indispensable concept at a descriptive level, and is accorded basic status as a primitive notion in some frameworks (along with other grammatical relations); on the other hand it has defeated all attempts at a cross-linguistically valid definition, and a substantial part of the field takes it to be no more than a descriptive label for an epiphenomenal collection of properties (cf. McCloskey 1997 for a clear overview of the issues). Traditional grammarians distinguish between grammatical or formal subjects and logical or notional subjects; for example, Jespersen notes (1927:227-228, 1949:107-110) that in a raising construction like (la), the grammatical subject is he, but the notional subject of the main clause is the (discontinuous) infinitive clause he to fall; while in the most deeply embedded clause in (1b) (from Dickens), the grammatical subject is there while the logical subject is what. (1)
a. b.
He happened to fall. I don't mean to say that I know what there is particularly dead about a door-nail.
At our present level of understanding, it seems that we can deconstruct the traditional subject into three components, one thematic-aspectual (the thematically most prominent argument of a predicate), one morphosyntactic (classically identified by 3
SUBJECTS, EXPLETIVES, AND THE EPP
4
case and/or agreement), and one discourse-informational (the topical or thematic entity named in a proposition) (cf. the discussion of the V-domain, the I-domain, and the C-domain in Platzack 2000). In the most straightforward cases, all of these components converge on a single noun phrase, prominent in all three components. In other cases they diverge. In a case like (2a), the morphosyntactic and discourse-informational systems converge on an entity which is not the thematically most prominent argument. In (2b), morphosyntactic marking is split: the thematically prominent argument controls agreement, but does not occupy the canonical subject position. In the reply in (2c), the discourse-informational topic neither occupies the subject position nor shows any other morphosyntactic subject marking. (2d) challenges theories of the thematic/aspectual system to determine which argument is prominent. (2)
a. b. c. d.
Theatergoers were provoked by the performance. There were both kinds of wine at the party. Where's Odradek?—I left him under the stairs. Peach schnapps attracts bees.
Despite such permutations, English generally consistently uses position (along with what case and agreement it has) to mark the thematically most prominent argument (lexical selection and voice are commonly used to ensure that the resultant subject is topical). Different interactions among the different components give rise to different systems cross-linguistically, for example ergative systems where the transitive object patterns with the intransitive subject for case/agreement) and topic-prominent systems, where morphosyntactic marking is more robust for the discourse topic than for the thematic-aspectually highest argument (cf. Li and Thompson 1976, E. Kiss 1995b, and the other papers in E. Kiss 1995a). Although the interactions among these components vary from language to language, the components are not independent of each other: each feeds the next. That is, first, the case/agreement system is sensitive to the thematic/aspectual system; there are 'active' or 'split' case-marking systems that distinguish agentive from patientive arguments, as well as ergative and accusative systems, but in each of these case-marking is determined by the thematic role of the argument, either in isolation (in an active system) or relative to other arguments (in accusative and ergative systems) (cf. Dixon 1994:28-35 for comments on semantic-based marking, and E. Kiss 1987 and in this volume for discussion of the situation in Hungarian). Various grammatical processes, such as imperative formation, binding, causative formation, and control are part of the thematic-aspectual domain and are similarly cross-linguistically sensitive to this component (Anderson 1976), so that even in an ergative language like Inuit, the argument of a control-type predicate controls the thematic-aspectual Actor of a transitive clause1 and even a topicprominent language like Lahu can have accusative case marking on the Theme of a transitive verb (Matisoff 1973:155-158) and regularly forms causatives in which the Agent of the embedded verb is the controllee (cf. Matisoff 1973:436). In the familiar GB-based frameworks, the only way for the syntax to refer to thematic structure is indirectly, so it is important that the rules of VP projection are universally determined by properties of 0-assignment. In the alternative adopted by Manzini
INTRODUCTION
5
and Savoia (this volume), thematic roles are aspectual features, allowing 6-assignment to occur at a greater distance; this makes the actual structure of the VP irrelevant (cf. Borer 1994); characteristics of feature movement are then responsible for the observed connection between the thematic-aspectual system and such processes as control. The second correspondence I alluded to above was that the discourse-informational component may universally be fed by the Case/agreement system. It is of course difficult to show that this is universal, but it seems that processes involving the informational domain (or C-domain), including relativization, topicalization, and other A-bar-type operations, are not directly sensitive to the thematic-aspectual system but only as mediated by the morphosyntactic system (cf. the 'topic' of Dixon 1972 and 'pivot' of Dixon 1994; see also Manning 1996 for discussion), so that an ergative language may favor the absolutive, rather than the Actor, in such processes. For example, ergative languages frequently require promotion of an Actor to absolutive case (through antipassive) for A-bar extraction (Nakamura 1996:92). E. Kiss argues in this volume that nominative in Hungarian is licensed not in the Isystem as in English, but in the lexicon; consistent with this, the nominative argument shows no prominence for purposes of focusing or other movement processes. In sum, the notion of subject is a multifaceted one. In the Government-Binding, Principles and Parameters, and Minimalist frameworks, subjecthood is generally a descriptive label, without theoretical status. Yet we can discuss, with Koster (1978), whether CPs can be subjects, without devolving into incoherence; just as we can say that the subject in Icelandic can be dative (Zaenen et al. 1985), that Warlpiri subjects of transitive verbs take the ergative case (Hale 1973), that Korean can have multiple subjects (Park 1973), or that English requires a subject. This last claim is the topic of §3; but before moving on to the subject requirement itself, I turn in §2 to one of its most important diagnostics, the expletive.
2. Expletives Expletive, pleonastic, or dummy subjects have been pivotal in syntactic argumentation. They are identified by their lack of semantic content, and their resolutely grammatical nature makes them an excellent probe into the boundary between syntax and semantics. They are crucial, for example, in the identification of syntactic positions as thematic or nonthematic, so that the pattern in (3) shows that the position after promise is thematic and that the position after suspect is not (cf. Bresnan 1972: ch. 3). (3)
a. We suspect there to be a party next door b. * We promise there to be a party next door
Three types of expletive subject are recognized in traditional grammars of English: extraposition it, weather it, and impersonal there, distinct from locative there (shown in (4d)). (4)
a. b.
It is obvious where you got that hickey. It gets dark in November.
6
SUBJECTS, EXPLETIVES, AND THE EPP c. d.
There's a fly in your soup, isn't there? There's our bus (*isn't there?)
Expletives are by definition semantically empty. Chomsky (1981:323-325) has argued that weather it is not an expletive, but rather what he calls a quasi-argument. It can bind PRO in an adjunct, for example, unlike the true expletive there (cf. E. Kiss, this volume, for similar examples from Hungarian; also Bennis 1986 on extraposition it controlling PRO). (5)
a. It often clears up here right after snowing heavily, b. * There is often a party here right after being a wake.
There is also a possible argument from Burzio's Generalization, which states that all and only verbs which assign an external 6-role can also assign Case to an object (Burzio 1986:178). Compare (6a), in which a weather predicate takes an object, to (6b), in which an unaccusative verb fails to; not even a resultative predicate can save (6c), though the resultative predicate is fine when the internal argument gets Case elsewhere, as in (6d) (i.e., the noun phrase mackerel has moved from object to subject position; cf. Simpson 1983 on resultative predicates with unaccusative verbs). (6)
a. It rained mackerel. b. * Rain fell mackerel. c. * Rain fell mackerel thick on the ground. d. Mackerel fell thick on the ground.
This suggests that it gets a 0-role from rain, specifically an external 0-role. Thus, we might wish to disregard meteorological it, when considering the behavior of expletives. A positive result of this move is that it allows us to handle the most serious of Postal and Pullum's (1988) examples of apparent expletive objects; consider their examples in (7). (7)
a. b. c.
We demand it of our employees that they wear a tie. I would really dig it if you tickled my toes. Beat it!
Examples like (7a) plausibly involve extraposition from a small clause subject position, thus being analyzable as ECM (i.e., it is an embedded subject; cf. (3a) above). On the other hand, examples like (7b) more clearly suggest extraposition from object position, on which see below. What is immediately pertinent is (7c), where there is no CP. Beat it means, roughly, 'go away immediately,' and as Postal and Pullum point out, there is no word that could substitute for it and preserve that meaning. Thus, this might be another example of a quasi-argument, a subcategorized (or s-selected) pronoun with only idiomatic content (essentially following Bolinger 1973). Hoekstra (1983) argues that extraposition it is referential as well (cf. also Bennis 1986, Vikner 1995). It is always coindexed with a CP (Rosenbaum 1968), resembling in this sense a resumptive pronoun. In fact, E. Kiss (this volume) shows that in Hungarian, extraposition pronouns can even be focused, and must therefore have
INTRODUCTION
7
semantic content (presumably, as she suggests, through the 0-chain that they form with the coindexed CP). This would eliminate Postal and Pullum's (7b) above, leaving uncontested the contention that expletives can appear only in subject position. Thus the only reasonably clear case of an expletive is the impersonal one, represented by two types. One type is the historically locative there in English, er in Dutch, and der in Danish and some dialects of Norwegian and Swedish (in the latter spelled dar). In the other Germanic languages, the existential expletive is generally identical to the neuter pronoun (det in Norwegian and Swedish, pad in Icelandic, tad in Faroese, es in German and Yiddish). This has consequences for agreement, in that locative-based expletives seem to correlate strongly with agreement with the associate, whereas pronoun-based expletives vary on this score (cf. French, where expletive il controls singular agreement).2 It might be thought that not even the impersonal expletive is strictly nonreferential, since, like extraposition it, it is always linked to something, namely an associate NP; but this is true only of English. There are clear cases where the corresponding expletive is not associated with any nominal, as in the impersonal assertions in French (Kayne 1975:247, note 56), Icelandic, Swedish (from Falk 1993:74), and Danish. (8) a. II sera parle de vous par tout le monde French it will.be spoken of you by all the world 'The whole world will be talking about you' b. Pad verdur talad um big ut um allan heim. Icelandic it becomes spoken about you out of all world 'The whole world will be talking about you' c. Det kluckar i roren av avloppsvatten. Swedish it gurgles in the.pipes by drainwater 'Drainwater is gurgling in the pipes' d. Der ma ikke ryges. Danish there may not be.smoked 'Smoking is not allowed' English is quite unusual among the Germanic languages in disallowing impersonal statements without any associate NP (cf. Vikner 1995). Analyses of expletive constructions abound. An important development was that of Safir 1982, 1985, 1987 involving a Case/agreement chain between the expletive and the associate; properties of such a chain are then argued to account for the definiteness effect. This approach successfully handles the fact that when Case is not transmitted, as in (8), there is no definiteness effect (a preposition assigns Case to the postverbal DPs in (8a-c)). However, Vangsnes (this volume) notes that in Icelandic, the definiteness restriction varies with the position of the associate, not with any property of the expletive; as a consequence, he argues, a Case chain cannot be involved and the definiteness effect must receive a different analysis. He provides one, basing it on de Hoop's 1996 theory of strong and weak Case. Since the available Case licensing is predicted by that theory to vary with the position of the associate, this more naturally accounts for the observed facts in Icelandic, and extends easily to Safir's data as well (on the assumption that P assigns strong Case).
8
SUBJECTS, EXPLETIVES, AND THE EPP
Chomsky (1986) proposed that the associate moves abstractly to the position of the expletive, which bears some subset of the features associated with nominal arguments. There have been many variations on this analysis, including different proposals regarding what features the impersonal expletive bears. In this spirit, Taraldsen, this volume, argues that the expletive bears only number features, and Vangsnes, in this volume, proposes that expletives have deictic features, and may also have Case features, parameterized across languages. Holmberg and Nikanne, also in this volume, show that the division in Finnish expletives between (roughly) the impersonal type and the extraposition and weather type is overtly reflected in their case, with the impersonal type being partitive; this, they argue, is a default form, and the impersonal expletive in Finnish is actually caseless. On the abstract movement analysis, an expletive with the requisite features may obviate movement, and therefore agreement with the associate will not obtain. However, Manzini and Savoia, in this volume, point out that agreement is always strict in person,3 suggesting that the associate always raises, and that movement of the associate cannot be too closely linked to the checking of agreement features. In contrast, Vangsnes (this volume) argues that there is never associate movement, arguing for agreement at a distance. Agreement at a distance, in the form of feature movement, is also the solution opted for by Holmberg and Nikanne, in this volume. Note that Vangsnes rejects the possibility of uninterpretable features, and so the expletive, with its deictic features, must have semantic content; thus the expletive status of there is not unquestioned. In fact, Bennis (1986) argues that there is an adverbial, and Hoekstra and Mulder (1990) and Moro (1997) similarly argue for a kind of content for there (though such accounts have problems with examples like (8d) above, cf. Falk 1993: ch. 4). Consider also E. Kiss's (this volume) arguments that all-focus statements asserting the existence of an event are predicated over the event variable. On the basis of data from Hebrew and other languages, Borer (1997) similarly argued at the Troms0 conference that locative expressions could provide an anchoring for the event variable. A natural combination of these ideas would be to take the expletive as an overt expression of the event variable (cf. Kratzer 1995 and §1.5.3 below). Taking impersonal-type expletives to be contentful would eliminate the question of why there are overt expletives but raise the new question of why they are so rarely overt. Even the Germanic languages make use of null expletives, particularly in inversion contexts. In fact, the expletives used in impersonal statements are obligatorily null when not clause-initial in Icelandic, German, and Yiddish, as indicated here for Icelandic (from Thrainsson 1979:477), with Norwegian for comparison (cf. Platzack 1985, Vikner 1995: ch. 6). (9) a. I gaer voru (*bad) mys i badkerinu. Icelandic b. I gar var *(det) mus pa badekaret. Norwegian yesterday were it mice in the. bathtub 'Yesterday, there were mice in the bathtub' The expletive is strictly obligatory in initial position if no other element appears there (cf. (8b) above), consistent with the V2 requirement. However, the same expletive cannot appear, in German or Icelandic, if some other element occupies the
INTRODUCTION
9
initial position, as indicated in (9a). This pattern led Platzack (1985) to analyze Icelandic expletives as topic expletives (but see Vangsnes, this volume). Norwegian, in contrast, requires the expletive subject regardless, as indicated in (9b), much as does English. As Roberts and Roussou note in this volume, the appearance of expletives in SpecCP in V2 languages like Icelandic and in SpecIP in non-null-subject languages like English suggests a unification of the two phenomena; they propose that both are the result of an identification requirement on Tense, which appears in T in languages like English but in C in V2 languages. The question then arises whether the subject position in constructions like (9a) is occupied by any element, for example expletive pro, as in Jaeggli and Safir 1989, Christensen 1991, or Holmberg and Platzack 1995; the trace of the topicalized adverbial, as in Vangsnes, this volume; or nothing, as in Bennis 1986 (for Dutch), Haider 1987 (for German), and Roberts and Roussou, this volume (in general). The difference between pro and trace and nothing may seem like an iota,4 but what is at stake is the very universality of the EPP: if there is nothing in the subject position in (9a) or its German equivalent, then the EPP must not hold in that language, as the EPP is commonly understood. Thus it is time to turn to the EPP.
3. The EPP The Extended Projection Principle of GB/P&P/MP work of the 1980s and 1990s states that a clause must have a subject; analogous statements can be found in many other theories, but what makes this situation volatile in the MP framework and its predecessors is exactly the absence of a primitive notion of subject, as discussed in §1 above. An NP daughter was stipulated as part of the expansion for S in each of the early instantiations of the generative framework (cf. Chomsky 1957:26, Chomsky 1965:72, Chomsky 1970:216 note 7, Chomsky 1981:25; compare Jespersen 1924, inter alia, and Harris 1946). The development of a theory of specifiers (Chomsky 1970, Stowell 1981:257ff., 1983) and its extension to the clause, analyzed as IP (Stowell 1981, Chomsky 1981:111), generalized the subject-taking property of the clause to that of phrases in general, but a difference still remained. While the clause requires a subject, the various other categories do not. Chomsky 1981:40 links the subject requirement (called "Principle P" on pp. 2627 there) with the Projection Principle, a requirement for lexico-thematic isomorphism across levels of representation, and coins the term Extended Projection Principle for the combination (Chomsky 1982:10). In popular usage, the expression (usually abbreviated EPP) has come to refer to the requirement that every clause have a subject (and so it will be used here), and the connection between the EPP and the Projection Principle has remained turbid. The Projection Principle prevents, for example, 0-marked elements from disappearing during the course of a derivation, and prevents a single argument from receiving two 0-roles, but it is unclear why it should require that every clause have a subject. In order to understand the EPP it is necessary to determine whether it is crosslinguistically valid or not. On the face of it, it seems manifestly implausible that it
10
SUBJECTS, EXPLETIVES, AND THE EPP
should be; the vast majority of the world's languages are unlike English, and freely allow sentences without overt grammatical subjects. For instance, Gilligan 1987 examines a genetically and areally diverse sample of about 100 languages and finds only eleven cases where thematic subjects cannot be null, and no cases where expletive subjects cannot be (he notes, p. 136, that Icelandic has oveijt expletives in initial position, and Classical Arabic has them in subordinate clauses). Thus sentences like the Italian (lOa) (from Rizzi 1982:126) or the Northern Sami (lOb) (from Nickel 1990:397) are typical of the world's conversations. (10) a.
b.
Non voglio che venga nessuno. not \vant.lSGthatcome.3scnobody '[I] don't want anybody to come' Arvigodii. began.to.rain '[It] began to rain'
Italian
Northern Sdmi
There has long been an intuition that even clauses without overt subjects have covert ones; witness Jespersen's (1937:20) analysis 'S° V for an Italian sentence like Vive 'He lives,' where the superscripted zero indicates phonological silence. Taraldsen's Generalization (Taraldsen 1980) is the statement in the terms of generative grammar that overt agreement morphology licenses null subjects. In the formal theory of null subjects that emerged from the work of Taraldsen (1980), Chomsky (1981), and Rizzi (1982), their distribution is subject to the same factors that constrain the distribution of other phonologically empty categories, as well as to (some of) the contraints on overt noun phrases; thus, in (lla) (from Rizzi 1982:176 note 14), in the embedded finite clause, an overt subject would be licensed, and so would a subject trace. In (lib) on the other hand (Rizzi 1982:129 example (40b)), the embedded clause is an infinitival, and the licensing possibilities for subjects are much reduced. Accordingly, the null subject is impossible, and (lib) is bad, either because it violates the subject condition or because the embedded verb fails to assign its quasi-0-role (cf. the discussion in §2). (11) a.
None chiaro quando piovera. not is clear when rains 'It is not clear when it rains' b. * Non e chiaro quando piovere. not is clear when rain. INF
Italian
The result was to give a formal characterization to the 'null subject' intuition shared by Jespersen and others, based on language-internal empirical evidence. This makes it possible to contemplate the possibility that some sort of subject requirement holds for sentences like (lOa-b). In fact, even verb-subject ('Free Inversion') sentences in Italian, like the embedded clause in (lOa), might have an expletive subject in a position to the left of V, though this is controversial (see Manzini and Savoia, this volume, for a critical assessment). At any rate, it has often been assumed since the early 1980s that all languages are subject to the EPP, some licensing pro while others do not. The first serious problem faced was the existence of languages like Chinese, which do not have rich
INTRODUCTION
11
Agr, yet allow pro drop. Gilligan's (1987) cross-linguistic survey shows that of twenty-six languages with no subject agreement, fully seventeen of them allow null thematic subjects, or about 65 percent. However, Huang (1984, 1987) analyzed Chinese null arguments, showing that they could be accommodated under GB licensing theory. Interestingly, this type of pro drop is an areal feature: all of Gilligan's seventeen examples are from south and east Asia or Oceania (as Gilligan notes, p. 197), and twelve of them are also like Chinese in additionally allowing null objects, despite having no object agreement. McCloskey (1996) challenges the view of the EPP as a universal, arguing that Irish allows truly subjectless sentences, such as that in (12). (12)
Laghdaigh ar a neart. decreased on his strength 'His strength decreased'
Irish
It is possible to postulate a null expletive in (12), but McCloskey argues that there is no evidence for such an expletive element internal to Irish, and furthermore that there are reasons specifically to believe that expletive subjects (of the impersonal type) are systematically absent from Irish. For some languages, it is possible to analyze agreement morphology as pronominal, possibly directly satisfying the EPP; Irish is known for the complementary distribution of overt subjects and overt agreement morphology (McCloskey and Hale 1984), but the verb in examples like (12) is in the 'analytic' form, which does not show agreement. Nevertheless, the view that head movement is what obviates the EPP in examples like (12) is an attractive one (cf. Benedicto 1994 for an early exploration of this idea, and Svenonius 1994: ch. 3 for another). Artemis Alexiadou and Elena Anagnastopoulou, in a paper presented at the Troms0 conference on which this volume is based (published as Alexiadou and Anagnostopoulou 1998), develop such a theory; on their theory, V-raising in Irish and other languages satisfies the EPP, because the finite verb bears the requisite nominal features for checking the EPP. Versions of this analysis are adopted by several of the chapters in this volume: Manzini and Savoia, Roberts and Roussou, and Vangsnes, though each with some variation.
4. The chapters 4.1. The conference The chapters in this volume originated as papers presented under a lingering arctic sun on June 6 and 7, 1997, at the conference in Troms0 bearing the same name as this volume. There were four other papers as well, which unfortunately could not be included. They were Hagit Borer's "Licensing Events: The Role of Locatives," Halldor Armann Sigurdsson's "Stylistic Fronting," Tor A. Afarli and Kristin M. Eide's "The EPP and Predication," and Artemis Alexiadou and Elena Anagnostopoulou's "V-Movement and EPP Checking." Borer's paper is referenced here as Borer 1997, and Sigurdsson's as Sigurdsson 1997. A version of the paper
12
SUBJECTS, EXPLETIVES, AND THE EPP
presented by Afarli and Eide appears as Afarli and Eide 2000, and Alexiadou and Anagnastopoulou's paper is represented by Alexiadou and Anagnostopoulou 1998. Here I briefly summarize only the papers included in the volume, in the order of their appearance. 4.2. Taraldsen Tarald Taraldsen, in "The Que/Qui Alternation and the Distribution of Expletives," provides a unified analysis for several constructions in French in which there is no overt preverbal subject. It obviates the problematic account of Rizzi 1990 by which qui is an agreeing complementizer, and eliminates the null expletive and rightward movement needed by Kayne and Pollock 1978. Central to the analysis is the postulation of an expletive i in French, corresponding to the expletive id (i before consonants) in the Rhaeto-Romance language Vallader, seen in (13a). The expletive is argued to appear in colloquial French expressions like the one in (13b) (with a morphophonological formative /t/ to its left). (13) a.
b.
I turnaran quei temps docts it will, return those times learned Those learned times will return' Pourquoi tu dois-ti partir? why you must-Ti leave 'Why must you leave?'
Vallader
French
This element is argued to account for the que/qui alternation in French, where qui appears when a subject has been wh-moved: qui is que plus the expletive i. The distribution of qui and of ti have in common that they both require that the actual subject be in SpecCP. Taraldsen suggests that in those cases, the subject checks number features in the C projection; the strong number feature responsible for the EPP raises from Infl to C when the subject enters SpecCP (separating from V, developing a suggestion in Taraldsen 1996 that feature movement is overt rather than covert as in Chomsky 1995). The expletive / occupies the SpecIP position without checking any (interpretable) features; it is a 'pure expletive,' required because the subject has been extracted. In addition, Taraldsen treats cases of French Stylistic Inversion, where a subject may be postverbal in case some element has been wh-moved, as in (14a). In fact, there may even be no obvious candidate for subjecthood, as in (14b). (14) a.
b.
Ou crois-tu quevontse cacher les chats? French where think you that will RFX hide the cats 'Where do you think the cats will hide?' Quand crois-tu que sera precede a un reexamen de la when think you that will.be proceeded to a reexamination of the question? case 'When do you think that a reexamination of the case might occur?'
Again, this possibility hinges on there being wh-movement out of the clause. The classic analysis of Stylistic Inversion has a null expletive in SpecIP. Taraldsen ar-
INTRODUCTION
13
gues, however, that the VP, along with some functional projections, occupies SpecIP in such cases, number features being checked by virtue of the percolation of the number features of the trace of the subject. In this way, Taraldsen eliminates the need for a null expletive in the analysis of French, and along the way provides a new account for *que-trac& effects. 4.3. Vangsnes In "Icelandic Expletive Constructions and the Distribution of Subject Types," 0ystein Vangsnes discusses Icelandic expletive constructions, in particular the distribution of different types of subjects (quantified, definite, and so on) in two different positions: one, the 'intermediate' position, follows the position of the verb in second position and is identified with SpecTP (following Bobaljik and Jonas 1996); the other, the 'postverbal' position, is lower down, presumably inside the VP. Vangsnes demonstrates that the restrictions on the two positions are different. The lower, postverbal position shows the classical definiteness effects as discussed by Milsark: weakly quantified DPs are allowed, and strongly quantified DPs, as well as names, possessed DPs, definite descriptions, and partitives are excluded. However, the intermediate position shows a different type of restriction. Here, although definite descriptions, partitives, names, and possessed DPs are impossible, strongly quantified DPs are permissible. McNally (1992) pointed out that examples of strongly quantified noun phrases in existential statements in English, like that given in (2b) in §1 above, are restricted to 'kind' expressions (in Carlson's 1977 sense). Vangsnes makes it clear that this is not the case in Icelandic. As Vangsnes argues, this is a serious problem for accounts in which the definite effects noted in existential constructions are linked crucially to the presence of an expletive, since the expletive is the same whether the associate is in the intermediate or the postverbal position. There is an additional, surprising constraint on the intermediate position: 'bare' indefinites, which in Icelandic appear with no determiner at all even when singular (there is no indefinite article in Icelandic), are degraded in the intermediate position, but perfect in the postverbal position. This apparently has nothing to do with the determiner per se, as a modified noun is acceptable. For example, (15a) is degraded, but (15b) is good. (15) a. ? Pad er fiskur eldadur i eldhusinu. it is fish cooked in the. kitchen 'There has been fish cooked in the kitchen' a. Pad er uldinn fiskur eldadur i eldhusinu. it is rotten fish cooked in the.kitchen There has been rotten fish cooked in the kitchen'
Icelandic
This is reminiscent of the observation made by E. Kiss (this volume) that Hungarian bare N subjects remain VP-internal and cannot appear in higher subject positions. Vangsnes develops an account in which the definiteness effect on the low position comes from de Hoop's 1996 strong and weak Case; T can license strong Case, allowing strongly quantified elements in the high position. The restrictions on associates in the high position come from a licensing requirement that is specific to
14
SUBJECTS, EXPLETIVES, AND THE EPP
definite expressions: unlike strongly quantified noun phrases, they must check deictic features as well as person and number features. Only in SpecAgrSP can these features be checked. 4.4. Holmberg and Nikanne Anders Holmberg and Urpo Nikanne discuss "Expletives, Subjects, and Topics in Finnish." They show that Finnish has overt expletives, both of the it type and the there type, despite the fact that it is a richly agreeing pro-drop language in the sense that first- and second-person subjects are freely omissible. Furthermore, under certain conditions these expletives are obligatory. This is unexpected on several accounts of the 'pro-drop parameter.' For example, in systems descending from Taraldsen's 1980 proposal, such as that of Rizzi 1982, pro drop is possible when empty categories can be licensed in subject positions. In such a language, overt expletives, if present at all, would necessarily be optional. Similarly, in Alexiadou and Anagnostopoulou 1998, pro drop is the result of a language checking the EPP feature through verb movement; this leaves unexplained the Finnish facts. Holmberg and Nikanne carefully show that the expletive competes with various topical elements for the structural position of the subject in Finnish, a position which can be distinguished from the operator position in SpecCP, from the specifier of TP, and from various focus positions. They argue that the topic position is the specifier of AgrSP, which they call FP (following Holmberg et al. 1993). Holmberg and Nikanne propose that EPP effects in Finnish are generally due to a negatively specified Focus feature, [-Foe]. The feature is semantically defined, but by the absence of content: an element that does not contribute any new information is minus focus. This feature must be checked by a topical element or an expletive in SpecFP, when present in F. It is not present when the sentence is allfocus, in which case there are no EPP effects. Although the F head is responsible for nominative case and for subject agreement inflection, they argue that these can be assigned at a distance, so the nominative subject need not move to SpecFP if some other element serves as the topic. 4.5. E. Kiss Katalin E. Kiss's "The EPP in a Topic-Prominent Language" examines the relevance of the EPP for Hungarian, a topic-prominent language. The EPP as stated in Chomsky 1981,1995 is a condition on IP, or on Infl, the locus of verbal inflectional morphology, and affects the thematically most prominent argument (requiring it to appear in SpecIP). Hungarian seems problematic for the EPP as stated in three respects. First, not all clauses have a VP-external subject: there are topicless (active) sentences in which no argument is external to the VP; though topics generally do have to move to a VP-external position. Second, sentences in which some element moves to a VP-external position do not seem to involve SpecIP; the positions are apparently distinct for quantified elements, focused elements, and topics, and there is no obvious link to Infl (i.e., there is no connection between the fronting of these elements and verbal morphology). Third, the movement of some element to a VP-external
INTRODUCTION
15
position is not limited to the thematically highest argument, although the thematically highest argument does always show morphological nominative case. E. Kiss splits the EPP into two different parts: one, a requirement that the VP be predicated over some argument, possibly the Davidsonian event argument, and two, a constraint on argument structure, requiring the thematically most prominent argument to be morphosytactically subject-marked. E. Kiss shows that Hungarian is consistent with the first aspect of the EPP, the predication requirement, if it is assumed that (i) in Topicless sentences, the VP is predicated over the event argument, and (ii) quantification (in sentences with quantifiers or focused elements) counts as predication in the relevant sense. Hungarian is consistent with the second version of the EPP, the subject-marking of the most prominent argument requirement, if it is assumed that this subject-marking (in the form of nominative case marking) can be instantiated without movement to SpecIP. E. Kiss argues that this occurs in the lexicon in Hungarian, where nominative case is associated with the thematically most prominent argument. Consistent with this analysis, there are no grammatical function-changing processes in Hungarian, such as passive, raising, ECM, or fowg/i-movement; the surface case of the thematically most prominent argument of a verb is always nominative. Thus, Hungarian differs from English in two parameters: first, in English, predication of a VP over the event argument can occur only in there-insertion contexts, and there are language-specific constraints on there-insertion, whereas in Hungarian, an active (not stative) VP can be predicated of the event argument with no expletive; and second, in English, subject-marking is effected by movement to SpecIP, while in Hungarian, it is effected by lexical assignment of nominative case. 4.6. Roberts andRoussou Ian Roberts and Anna Roussou's contribution is "The Extended Projection Principle as a Condition on the Tense Dependency." Roberts and Roussou argue that the EPP and V2 are both manifestations of the same condition, and furthermore that this condition derives naturally from the analysis of T as containing a tense variable, as in work by En$. This unification of the EPP and V2 is motivated by a straightforward observation: they both involve the obligatory filling of the specifier of the head containing tense (T). Halldor Sigurdsson, at the Troms0 conference, independently argued on the basis of Stylistic Fronting data from Icelandic for a merger of V2 and the EPP, though he analyzed them formally in terms of the checking of an Event feature E (Sigurdsson 1997). The basic idea is that the variable in T must be bound; if it is not, then it receives an arbitrary interpretation, which they argue to be the case in VI declaratives. A variable in a head position, for Roberts and Roussou, can be bound either by a higher head or by an element in a specifier position. As an example of the former, T in a subordinate clause is bound by C. In German, a subordinate clause with an overt complementizer shows no V2 effect, and in fact even the EPP is called off, in that expletives do not appear (cf. §2 above). An example of binding by an element in a specifier is provided by German main clauses: V moves to C, because of an independent requirement, taking T along with it; but V does not have the appropriate content to bind T, in Germanic. Therefore SpecCP must be filled with something
16
SUBJECTS, EXPLETIVES, AND THE EPP
that can bind T. This is the V2 requirement. Roberts and Roussou account for a wide range of patterns of subject realization and main and embedded V2. One important proposal they put forth is that the PF-realization of functional features is a point of parametric variation, where at least AgrS, Force, Fin, and Wh are functional features (cf. Rizzi 1997). A feature X that requires PF realization is indicated thus: X*. For example, V2 languages have Fin*, requiring something with PF-features (the verb) to appear overtly in the Fin node, whereas languages like English have Fin. In the standard Minimalist Program, the difference between overt realization of features in a particular location and their absence at PF is dependent on two factors: one, whether the relevant features are 'strong' or 'weak' (in more recent parlance, whether there is an EPP feature or not), and two, whether there happen to be any elements with the appropriate features that are not phonologically pronounced. Thus, in principle, a language in which the subject remained quite low (in the VP, say) could either be one in which all D features were weak, or one in which there were strong D features but null expletives. For Roberts and Roussou, on the other hand, this difference is expressed directly, in the form of a single parameter: whether a feature requires PF realization or not. They refine this proposal in a variety of ways. For instance, it interacts with their assumption that Merge is more economical than Move: if there is an element that can realize X* simply by being inserted, it will be, but if not, some element may be moved into X* (or its specifier). Another proposal is that it is the highest F* in a dependency that must instantiate the LF feature of that dependency. In other words, it won't do just to have any pronounceable material in a node marked *; it must have inflection or content corresponding to the LF value of the functional feature. Verbal agreement can be pronominal and can have a referential interpretation, and when it does, verb movement satisfies the subject requirement directly, as in Alexiadou and Anagnostopoulou 1998, without pro (as in Manzini and Savoia, this volume). They discuss cross-linguistic variation in the satisfaction of various features in the C system, for example Force, comparing cases in which there are dedicated particles for this purpose (e.g., Welsh) with cases in which movement is forced (e.g., Danish) and cases in which the Force features are weak (e.g., German). 4.7. Manzini and Savoia Rita Manzini and Leonardo Savoia, in "Parameters of Subject Inflection in Italian Dialects," draw on a very detailed analysis of a large number of Italian dialects, especially northern ones, to propose a number of important developments of the Minimalist Program, considering in particular the empirical and theoretical status of the EPP. Northern Italian (NI) dialects show verbal agreement with the subject, as in Standard Italian, but in addition have a preverbal subject clitic which also indicates the person, number, and gender of the subject. This clitic appears regardless of whether there is also an overtly realized subject DP. Manzini and Savoia continue the deconstruction of Infl begun by Pollock 1989, splitting the traditional agreement categories into separate projections. Each pro-
INTRODUCTION
17
jection is motivated by the overt realization of clitics, for example, the realization of first- and second-person clitics following negation, while third-person clitics are above negation, gives a Number head higher than the Person head. The highest head is D, the locus of definiteness, followed by N, followed by Num[ber], followed by Pferson]—all nominal categories, yielding an inversion, of sorts, of Abney 1987, where sentential-type functional heads were introduced into the noun phrase. On Manzini and Savoia's account, D is strong universally, requiring that something appear there overtly (following Roberts and Roussou, this volume, this has to do with PF realization rather than movement prior to Spell-out). The element may be a subject clitic, in NI, or a full DP as in English, or the finite verb, as in standard Italian. 0-role assignment, following Manzini and Roussou 1999 and drawing on Borer 1994, is modeled as association of an Asp[ectual] feature on the verb with D features, which enables the element bearing D features to be interpreted as an argument. This leads to some major differences between the system proposed here and Chomsky's model. First, since features, including Asp, can move from the nodes hosting them, there is no longer any need for arguments to be projected inside VP at any level; a subject may be merged in a specifier position in the I-domain and Asp may be moved up to it (taking agreement features along with it as free riders). Second, there is no need for pro, since the functional heads may be filled by items bearing D features; Asp can move to D and associate directly with the clitic there. Manzini and Savoia also discuss the distribution of overt subjects in the varieties under discussion. Examing preverbal subject, they show reason to believe that preverbal subjects in NI are topics, occupying a higher specifier position, whereas the preverbal subject in English and the full DP subject in French are in the specifier of I. Turning to postverbal subjects, they present a detailed and original account of impersonal constructions, contrasting their account with those of Chomsky 1995 and 1998. They argue, in particular, that person agreement is obligatory in constructions with a postverbal subject (as mentioned briefly in §2. above; cf. note 3), whereas number agreement varies cross-linguistically in such constructions. On their analysis, this is because agreement is effected by the movement of Asp features to different inflectional heads. Parametrically, Asp may or may not carry with it number features; they motivate this in terms of a distinction between features with referential import (including number) and predicative features (including the basic denotation of a noun). 4.8. Svenonius In "Subject Positions and the Placement of Adverbials," I examine issues of adverbial placement and the patterns of adverb-subject ordering within IP in various Germanic languages. Adverb placement in general and adverb-subject ordering in particular have been used to motivate analyses of clause structure in many influential accounts. I take a critical look at the reliability of adverbs as a diagnostic for clause structure, arguing that their distribution warrants a more semantically based treatment than is usually offered.
18
SUBJECTS, EXPLETIVES, AND THE EPP
Specifically, I suggest that the strict relative ordering of adverbials should not be taken as an indication of separate functional heads, as in Cinque 1999. I argue that to do so requires the abandonment of the Head Movement Constraint, which is otherwise well motivated. I opt instead for a model in which multiple adjunction is permitted (contra Kayne 1994), and allow the relative ordering of adverbials to be determined by their semantics. This leads to the possibility that relative subject-adverb orders in the IP in Germanic are not the result of multiple specifier positions, as is assumed in Bobaljik and Jonas 1996, but simply the result of two alternative orders of attachment of the subject and the adverb. I show that the conditions on adverb-subject order involve a notion of topicality, grammaticized to different extents in different languages. For instance, in Norwegian, the formal topicality of an argument is determined by its discourse function. I provide an account in which topicality is linked to the EPP, in the form of a +Topic feature, and to the model-theoretic semantic interpretation of the clause. The connection of the topic feature to the model-theoretic interpretation provides the foundation for the analysis of the restrictions on adverb-subject order. The basic idea is that an adverb cannot attach to a clause that already has a topic specified, as that denotes a unit of the wrong semantic type. I link this analysis to earlier work (Svenonius 1994) on the (contextual) anchoring of propositions. I conclude that there is no evidence from adverb placement for two specifier positions in Germanic, calling into question one plank of support for Pollock's 1989 split-Infl hypothesis. This brings the chapter into a certain tension with those of Vangsnes and Manzini and Savoia, as they are built substantially around versions of that hypothesis. However, my findings do not establish the absence of multiple heads; Manzini and Savoia, in fact, even show that those heads can simultaneously be filled (as opposed to other work which demonstrates either specifiers, adjuncts, or alternate head positions). The presence of additional inflectional heads would not actually contradict the account.
5. What is the EPP? It is clear that the EPP straddles the interfaces, PF and LF—its PF nature is manifest in its requiring that a position be filled by an overt element, as is most clearly seen with expletives; and its LF nature is revealed by the connection to the topicality of the element satisfying it, or the lack of topicality in expletive constructions. Here I examine the different ways in which the authors in this volume have sought to deal with the empirical issues that subjects raise in their respective chapters. 5.1. The PF side of the EPP The EPP, as Holmberg (2000) puts it, clearly has a 'phonological half; there are many cases in which phonologically null elements are motivated for languages but where those same elements do not satisfy the EPP, for example trace in many cases.
INTRODUCTION
19
Taraldsen's article here can be taken as a case in point: on his analysis, the French complementizer qui includes an expletive, which means that cases in French which were previously analyzed as containing a subject gap don't actually have one. The usual assumption is that the EPP is satisfied by a trace bound by an overt subject or operator in an A-bar position, but cases such as these raise the possibility of reanalyzing some other cases of apparent subject traces; compare Taraldsen's 1990 analysis of Danish relative der as filling the subject position in relative clauses like manden der stjal bogen, literally 'the man there stole the.book,' with the well-known fact that the relative complementizer that in English cannot be omitted in the context of a subject gap, in, for example, the man *(that) stole the book. Conceivably, then, that fills the subject position overtly in English, and there are no real subject gaps. The reverse pattern in sentences like Who did you say left? might then be amenable to an analysis along the lines of Taraldsen's, in this volume, for Stylistic Inversion. Taraldsen also eliminates the possibility of LF satisfaction of the EPP, as he abandons covert movement, arguing that the strong/weak distinction can be recast as a question of how much material is pied-piped by feature movement (as in Taraldsen 1996). Roberts and Roussou, it will be recalled, also argue that a model which allows both a covert/overt movement distinction and a null/overt PF distinction for lexical items is inadequately restrictive. The phonological side of the EPP, on their account, is expressed by their diacritic '*,' which requires PF-interpretable (i.e., pronounceable) material and replaces the strong/weak feature distinction. One part of the EPP, in their system, is the universal requirement that some member of the T dependency bear * (i.e., be PF-interpretable). Manzini and Savoia follow the same general line of thinking, in particular in eliminating pro entirely, opting for 9-feature identification in functional heads. My account in this volume would be compatible with this; in particular, the difference between a language with overt expletives and one without them might be cast in terms of whether type-shifting were allowed without the overt insertion of a nominal element. Holmberg and Nikanne, however, do retain pro; they argue that AgrS in Finnish bears person features, and is sufficient to license first- and second-person pro, but not third. On their account, pro actually fills the EPP position in sentences like their example (7a) Olin vdsynyt, '[I] was tired.' 5.2. The LF side of the EPP Many of the chapters in this volume connect the EPP, in one way or another, to discourse notions such as topicality. Here the papers on Finno-Ugric languages are perhaps the most explicit; E. Kiss argues that part of the EPP comes from the semantic fact that declarative statements must be predications, and a predicate must be predicated of something. In Hungarian, this requirement can be satisfied by various types of quantification, or by predication over an event, or by predication over a topic. In English, licensing in SpecIP is formally associated with being the subject of predication. Manzini and
20
SUBJECTS, EXPLETIVES, AND THE EPP
Savoia also link subjects with topicality in that they argue that preverbal subjects in various Romance languages occupy a topic position. Holmberg and Nikanne connect the EPP in Finnish to the notion of focus, which is in effect the inverse of topicality. Nonfocus elements evacuate the VP, and a high functional head F attracts a nonfocus element to its specifier. For Roberts and Roussou, Tense is a kind of pronominal element and in order to get a specific reference, it must be bound. The usual binder, in a V2 language, is a topical element, as this gives the right kind of anchoring to the context of the utterance. In a non-V2 language like English, the Case system conspires to make the subject the only possible binder for T. Similarly, Vangsnes postulates a licensing requirement on AgrS that forces it to enter into a checking-type relation. This relation may be satisfied by a variety of elements; a definite subject can satisfy all of them, or an expletive or adverbial may supply deictic features while a logical subject supplies such features as Case. Finally, in my own account I link the EPP with a topic feature, but argue that that feature has become grammaticized in languages like Danish and English to the point where it no longer corresponds to discourse topicality. 5.3. The syntactic nature of the EPP The usual treatment of the EPP is syntactic. Chomsky (1995) suggested an uninterpretable categorial D feature as the formal core of the EPP, while Rothstein (1983) has modeled it as a syntactic predication requirement; Afarli and Eide, at the Troms0 conference, argued that the predication requirement is semantic, but allowed it to be satisfied in a purely formal way by expletive subjects (Afarli and Eide 2000). Various other accounts have tried to derive the EPP from Case requirements (e.g., Lasnik 1992, Martin 1999). The most serious problem for a Case-based account is the fact that EPP effects are observed in small clause environments, as I discussed in Svenonius 1996. Some examples are shown in (16). (16) a. His efforts only got him glared at. a.' * His efforts only got glared at him. b. With *(it) as cold as it is, even warm coffee freezes quickly. c. *(It) being Easter break makes *(it) likely that they won't be home. Case requirements are satisfied in the starred examples, yet they are still bad. All of these examples are plausibly handled by semantic requirements of the type discussed here. In (16a), the matrix verb s-selects for a complement denoting a State of Affairs (cf. Svenonius 1994 for discussion and references); the participial predicate glared at him in (16a') does not denote a State of Affairs, but the small clause him glared at can; compare Heycock 1992 and E. Kiss's (this volume) predication requirements. Assuming that weather predicates like cold assign quasi-0-roles, as discussed in §2 above, leaving the subject out of the absolute construction in (16b) would violate the s-selectional restrictions of the predicate. Similarly, the gerundive expression in (16c) does not contain a true expletive. The extraposition it in the comple-
INTRODUCTION
21
ment of (16c) is required by the fact that make s-selects a State of Affairs, as with (16a). However, there are small clause examples with what are likely to be true expletives, like those in (17). As with the examples in (16), the expletives are obligatory. (17) a. b.
c.
If the band cancels, we can watch *(there) be a huge brawl. leg h0rte *(det) bli snakket om deg. Norwegian I heard it become spoken about you 'I heard people talking about you' Jag sag *(det) dansas pa skeppet. Swedish I saw it be.danced on the.ship 'I saw there be dancing on the ship'
Assuming that the expletives here are 'pure' expletives, not forced by selectional requirements of the embedded predicate, and not contributing any content for predication, such examples seem to represent the irreducible core of the EPP. Note that such examples are possible only with verbal predicates; even participles are excluded, as indicated in (18). Example (18a) shows that a passive participle is a legitimate predicate in a small clause complement to have; (18b) shows that there is excluded from this type of small clause. Examples (18c-d) show that this is not due to some incompatibility between the impersonal construction and the sselectional requirements of the matrix verb. Examples (18d-e) show that gerunds are verbal in the requisite sense. (18) a. They had horns blown upon their arrival. b. * They had there blown horns upon their arrival. c. They had there be horns blown upon their arrival. d. They had there being horns blown upon their arrival. e. There being a riot shouldn't deter you. Nor are adjectives, nouns, or prepositional phrases possible impersonal predicates, in the absence of a verb.5 (19) a. * I want there available a fireman at all times. b. * We would consider there a woman the best candidate. c. * There seems a draft in this room. Scandinavian shows the same restriction. Infinitives, including s-passive infinitive forms, are possible in Scandinavian (cf. (17b-c) above and (20b) below), though just as in English, passive participles are not (cf. (20a) below). (20) a. * leg h0rte del snakket om deg. I heard it spoken about you b. leg h0rte det snakkes om deg. / heard it be. spoken about you 'I heard people talking about you'
Norwegian
Without an expletive, an example like (20a) would violate the s-selectional requirements of the matrix verb, because the complement would not denote a State
22
SUBJECTS, EXPLETIVES, AND THE EPP
of Affairs. But an expletive cannot be inserted anyway. Apparently, a verb is necessary (in the small clause) in order for an impersonal expletive to appear, in contrast to the situation for extraposition and weather expletives. One characteristic that separates verbs from other categories is the possibility of specifying them for tense; this can be represented by saying that all and only verbs carry the Davidsonian e, and that tense binds e; this connects impersonal expletives with E. Kiss's (this volume) predication over the event as well as with Roberts and Roussou's (this volume) connection between the EPP and binding of T.6
6. Conclusion At its core, the EPP stands out as a canonical example of a doughtily syntactic requirement, irreducible to any constraints of logic, discourse, or phonology; so much so that Chomsky (1998) has adopted the expression "EPP-feature" for strong features in general, those which trigger overt movement. Strikingly, the authors in this volume have nevertheless resisted postulating uninterpretable features to account for EPP effects, arguing in most cases that it is a semantic requirement at some level. The autonomy of the syntax generally resides in what makes the satisfaction of the requirement overt rather than covert, that is, why the PF material must be pied-piped, for Taraldsen, why D is strong for Manzini and Savoia, why the T dependency must be * for Roberts and Roussou, and so on. To a certain extent, the shifting of the burden of the EPP to such notions as event anchoring, categorical predication, and discourse topicality may be an act of sweeping the syntactic problems under a semantic carpet; certainly, those semantic notions need to be better understood before we can claim to have solved the riddle of the EPP. But the contributions in this volume (along with the other work referenced here, and much more that I have only alluded to) have clearly made progress not only in defining the problems but also in narrowing the range of plausible solutions.
Notes 1. Cf. Woodbury 1977 and also Dixon 1994:137; the thematic system may underdetermine the possibilities; Kibrik 1985 shows that Tindi also allows Theme controllees. 2. See Christensen and Taraldsen 1989 for the Scandinavian facts, McCloskey 1991 on agreement with it vs. there in English, Cardinaletti 1997 on the distinction in general, and Manzini and Savoia (this volume) for comment. 3. Cf. Sigurasson 1990-1991, 1992, Taraldsen 1994 for similar observations about Icelandic, and Holmberg and Nikanne, this volume, on Finnish. 4. Though blood was shed in the fourth century over whether the essences of the Father and the Son were homoousious (i.e., of the same substance) or homoiousious (i.e., of similar substance), today's academic debates rarely lead to homicide. 5. In fact, the requirement in modern spoken English is even stronger: many, if not most, speakers require be in order for impersonal there to occur, rejecting sentences like There exist three counterexamples or There emerged a gopher from its hole as irredeemably literary. Even so, there is a contrast between the literary I saw there emerge a gopher and the impossible examples in (19); furthermore, ex-
INTRODUCTION
23
amples of the There emerged a gopher are ordinary in the other Germanic languages, while those of the type in (19) are impossible. 6. See Svenonius 1994: ch. 1 on the formal dependency of verbal small clauses as opposed to nonverbal ones. If nonverbal stage-level predicates do not bear e, then e cannot be used for adverbial modification in examples like / want him off my boat immediately. Furthermore, if all verbs carry e, then it cannot be used to distinguish stage-level from individual-level predicates. These observations suggest that Kratzer's 1995 locative / for stage-level predicates should be distinguished from the verbal e.
References Abney, Steven. 1987. The English Noun Phrase in Its Sentential Aspect. Ph.D. dissertation, MIT. Afarli, Tor A., & Kristin M. Eide. 2000. Subject requirement and predication. Nordic Journal of Linguistics 23, 27-48. Alexiadou, Artemis, & Elena Anagnostopoulou. 1998. Parametrizing AGR: Word order, Vmovement and EPP-checking. Natural Language & Linguistic Theory 16, 491-539. Anderson, Stephen R. 1976. On the notion of subject in ergative languages. In Charles N. Li (ed.), Subject and Topic, 1-23. Academic Press, New York. Benedicto, Elena E. 1994. Agr, phi-features, and V-movement: Identifying pro. In Elena E. Benedicto & Jeffrey T. Runner (eds.), Functional Projections, 1-18. GLSA, University of Massachusetts, Amherst. Bennis, Hans. 1986. Gaps and Dummies. Foris, Dordrecht. Bobaljik, Jonathan David, & Dianne Jonas. 1996. Subject positions and the roles of TP. Linguistic Inquiry 27, 195-236. Bolinger, Dwight. 1973. Ambient it is meaningful too. Journal of Linguistics 9, 261-270. Borer, Hagit. 1994. The projection of arguments. In Elena E. Benedicto & Jeffrey T. Runner (eds.), Functional Projections, 19^-7. GLSA, University of Massachusetts, Amherst. Borer, Hagit. 1997. Licensing events: The role of locatives. Paper presented at the Workshop on Subjects, Expletives, and the EPP, Troms0. Bresnan, Joan. 1972. Theory of Complementation in English Syntax. Ph.D. dissertation, MIT.
Burzio, Luigi. 1986. Italian Syntax: A Government-Binding Approach. Reidel, Dordrecht. Cardinaletti, Anna. 1997. Agreement and control in expletive constructions. Linguistic Inquiry 28, 521-533. Carlson, Gregory. 1977. Reference to Kinds in English. Ph.D. dissertation, University of Massachusetts, Amherst. Chomsky, Noam. 1957. Syntactic Structures. Mouton, The Hague. Chomsky, Noam. 1965. Aspects of the Theory of Syntax. MIT Press, Cambridge, Mass. Chomsky, Noam. 1970. Remarks on nominalization. In Roderick A. Jacobs & Peter S. Rosenbaum (eds.), Readings in English Transformational Grammar, 184-221. Georgetown University Press, Washington, D.C. Chomsky, Noam. 1981. Lectures on Government and Binding. Foris, Dordrecht. Chomsky, Noam. 1982. Some Concepts and Consequences of the Theory of Government and Binding. MIT Press, Cambridge, Mass. Chomsky, Noam. 1986. Knowledge of Language: Its Nature, Origins, and Use. Praeger, New York. Chomsky, Noam. 1995. The Minimalist Program. MIT Press, Cambridge, Mass. Chomsky, Noam. 1998. Minimalist inquiries: The framework. MIT Occasional Papers in Linguistics 15, 1-56. [Reprinted 2000 in Roger Martin, David Michaels, & Juan
24
SUBJECTS, EXPLETIVES, AND THE EPP
Uriagereka (eds.), Step by Step: Minimalist Essays in Honor of Howard Lasnik, 89-155. MIT Press, Cambridge, Mass.] Christensen, Kirsti Koch. 1991. AGR, adjunction, and the structure of Scandinavian existential sentences. Lingua 84, 137-158. Christensen, Kirsti Koch, & Knut Tarald Taraldsen. 1989. Expletive chain formation and past participle agreement in Scandinavian dialects. In Paola Beninca (ed.), Dialect Variation and the Theory of Grammar, 53-83. Foris, Dordrecht. Cinque, Guglielmo. 1999. Adverbs and Functional Heads: A Cross-Linguistic Perspective. Oxford University Press, New York. Dixon, R. M. W. 1972. The Dyirbal Language of North Queensland. Cambridge University Press, Cambridge. Dixon, R. M. W. 1994. Ergativity. Cambridge University Press, Cambridge. Falk, Cecilia. 1993. Non-referential Subjects in the History of Swedish. Ph.D. dissertation, University of Lund. Gilligan, Gary. 1987. A Cross-Linguistic Approach to the Pro-drop Parameter. Ph.D. dissertation, University of Southern California. Haider, Hubert. 1987. Theta-tracking systems—Evidence from German. In Pieter Muysken & L£szlo Maracz (eds.), Configurationality, 185-206. Foris, Dordrecht. Hale, Kenneth. 1973. Person marking in Walbiri. In Stephen R. Anderson & Paul Kiparsky (eds.), A Festschrift for Morris Halle, 308-344. Holt, Rinehart, & Winston, New York. Harris, Zellig. 1946. From morpheme to utterance. Language 22, 161-183. Heycock, Caroline. 1992. Layers of Predication: The Non-lexical Syntax of Clauses. Ph.D. dissertation, University of Pennsylvania. Hoekstra, Teun. 1983. The distribution of sentential complements. In Hans Bennis & W. U. S. van Lessen Kloeke (eds.), Linguistics in the Netherlands 1983, 93-103. Foris, Dordrecht. Hoekstra, Teun, and Rene Mulder. 1990. Unergatives as copular verbs. Linguistic Review 7, 1-79. Holmberg, Anders. 2000. Scandinavian Stylistic Fronting: How any category can become an expletive. Linguistic Inquiry 31, 445—483. Holmberg, Anders, Urpo Nikanne, Irmeli Oraviita, Hannu Reime, & Trond Trosterud. 1993. The structure of INFL and the finite clause in Finnish. In Anders Holmberg & Urpo Nikanne (eds.), Case and Other Functional Categories in Finnish Syntax, 177-206. Mouton de Gruyter, Berlin. Holmberg, Anders, & Christer Platzack. 1995. The Role of Inflection in Scandinavian Syntax. Oxford University Press, New York. Hoop, Helen de. 1996. Case Configuration and Noun Phrase Interpretation. Garland, New York. Huang, C. T. James. 1984. On the distribution and reference of empty pronouns. Linguistic Inquiry 15, 531-574. Huang, C. T. James. 1987. Remarks on empty categories in Chinese. Linguistic Inquiry 18, 321-337. Jaeggli, Osvaldo, & Kenneth J. Safir. 1989. The null subject parameter and parametric theory. In Osvaldo Jaeggli and Kenneth J. Safir (eds.), The Null Subject Parameter, 1-44. Kluwer, Dordrecht. Jespersen, Otto. 1924. The Philosophy of Grammar. [Reprinted 1992 University of Chicago Press, Chicago.]
INTRODUCTION
25
Jespersen, Otto. 1927. A Modern English Grammar: On Historical Principles. Vol. 3. George Allen & Unwin, London. Jespersen, Otto. 1937. Analytic Syntax. [Reprinted 1984 University of Chicago Press, Chicago.] Jespersen, Otto. 1949. A Modern English Grammar: On Historical Principles. Vol. 7. George Allen & Unwin, London. Kayne, Richard S. 1975. French Syntax: The Transformational Cycle. MIT Press, Cambridge, Mass. Kayne, Richard S. 1994. The Antisymmetry of Syntax. MIT Press, Cambridge, Mass. Kayne, Richard S., & Jean-Yves Pollock. 1978. Stylistic Inversion, successive cyclicity, and Move NP in French. Linguistic Inquiry 9, 595-621. Kibrik, A. E. 1985. Toward a typology of ergativity. In Johanna Nichols & Anthony C. Woodbury (eds.), Grammar Inside and Outside the Clause: Some Approaches to Theory from the Field, 268-323. Cambridge University Press, Cambridge. Kiss, Katalin E. 1987. Configurationality in Hungarian. Reidel, Dordrecht. Kiss, Katalin E. (ed.). 1995a. Discourse Configurational Languages. Oxford University Press, New York. Kiss, Katalin E. 1995b. Discourse configurational languages: Introduction. In E. Kiss 1995a, 3-27. Koster, Jan. 1978. Why subject sentences don't exist. In Samuel Jay Keyser (ed.), Recent Transformational Studies in European Languages, 53-64. MIT Press, Cambridge, Mass. Kratzer, Angelika. 1995. Stage-level and individual-level predicates. In Gregory N. Carlson & Francis Jeffry Pelletier (eds.), The Generic Book, 125-175. University of Chicago Press, Chicago. Lasnik, Howard. 1992. Case and expletives: Notes toward a parametric account. Linguistic Inquiry 23, 381-405. Li, Charles N., & Sandra A. Thompson. 1976. Subject and topic: Anew typology of language. In Charles N. Li (ed.), Subject and Topic, 457-489. Academic Press, New York. Manning, Christopher D. 1996. Ergativity: Argument Structure and Grammatical Relations. CSLI, Stanford, Calif. Manzini, M. Rita, & Anna Roussou. 1999. A Minimalist theory of A-movement and control. UCL Working Papers in Linguistics 11, 403-440. Martin, Roger. 1999. Case, the Extended Projection Principle, and Minimalism. In Samuel David Epstein & Norbert Hornstein (eds.), Working Minimalism, 1-25. MIT Press, Cambridge, Mass. Matisoff, James A. 1973. The Grammar ofLahu. University of California Press, Berkeley. McCloskey, James. 1991. There, it, and agreement. Linguistic Inquiry 22, 563-567. McCloskey, James. 1996. Subjects and subject positions in Irish. In Robert D. Borsley & Ian Roberts (eds.), The Syntax of the Celtic Languages: A Comparative Perspective, 241-283. Cambridge University Press, Cambridge. McCloskey, James. 1997. Subjecthood and subject positions. In Liliane Haegeman (ed.), Elements of Grammar: Handbook in Generative Syntax, 197-235. Kluwer, Dordrecht. McCloskey, James, & Kenneth Hale. 1984. On the syntax of person-number inflection in Modern Irish. Natural Language & Linguistic Theory 1, 487-534. McNally, Louise. 1992. An Interpretation for the English Existential Construction. Ph.D. dissertation, University of California, Santa Cruz.
26
SUBJECTS, EXPLETIVES, AND THE EPP
Moro, Andrea. 1997. The Raising of Predicates: Predicative Noun Phrases and the Theory of Clause Structure. Cambridge University Press, Cambridge. Nakamura, Masanori. 1996. Economy of Chain Formation. Ph.D. dissertation, McGill University. Nickel, Klaus Peter. 1990. Samisk grammatikk. Universitetsforlaget, Oslo. Park, Byzing-Soo. 1973. Multiple subject constructions in Korean. Linguistics 100, 63-76. Platzack, Christer. 1985. The Scandinavian languages and the null subject parameter. Working Papers in Scandinavian Syntax 20. Platzack, Christer. 2000. Multiple interfaces. In Urpo Nikanne & Emile van der Zee (eds.) Cognitive Interfaces: Constraints on Linking Cognitive Information, 21-53. Oxford University Press, Oxford. Pollock, Jean-Yves. 1989. Verb movement, Universal Grammar, and the structure of IP. Linguistic Inquiry 20, 365-424. Postal, Paul M., & Geoffrey K. Pullum. 1988. Expletive noun phrases in subcategorized positions. Linguistic Inquiry 19, 635-670. Rizzi, Luigi. 1982. Issues in Italian Syntax. Foris, Dordrecht. Rizzi, Luigi. 1990. Relativized Minimality. MIT Press, Cambridge, Mass. Rizzi, Luigi. 1997. The fine structure of the left periphery. In Liliane Haegeman (ed.), Elements of Grammar: Handbook in Generative Syntax, 281-337. Kluwer, Dordrecht. Rosenbaum, Peter S. 1968. The Grammar of English Predicate Complement Constructions. MIT Press, Cambridge, Mass. Rothstein, Susan. 1983. The Syntactic Forms of Predication. Ph.D. dissertation, MIT. Safir, Kenneth J. 1982. Syntactic Chains and the Definiteness Effect. Ph.D. dissertation, MIT. Safir, Kenneth J. 1985. Syntactic Chains. Cambridge University Press, Cambridge. Safir, Kenneth J. 1987. What explains the Definiteness Effect? In Eric J. Reuland & Alice G. B. ter Meulen (eds.), The Representation of (In)definiteness, 71-97. MIT Press, Cambridge, Mass. Sigurdsson, Halld6r Armann. 1990-1991. Beygingarsamraemi. Islenskt mdl og almenn mdlfrcedi 12-13, 31-77. Sigurdsson, Halldor Armann. 1992. Um beygingarsamraemi og malkunnattu. Islenskt mdl og almenn mdlfrcedi 14, 63-87. Sigurdsson, Halldor Armann. 1997. Stylistic Fronting. Paper presented at Workshop on Subjects, Expletives, and the EPP, Troms0. Simpson, Jane. 1983. Resultatives. In Beth Levin, Malka Rappaport, & Annie Zaenen (eds.), Papers in Lexical-Functional Grammar, 143-157. Indiana University Linguistics Club, Bloomington. Stowell, Tim. 1981. Origins of Phrase Structure. Ph.D. dissertation, MIT. Stowell, Tim. 1983. Subjects across categories. Linguistic Review 2, 285-312. Svenonius, Peter. 1994. Dependent Nexus: Subordinate Predication Structures in English and the Scandinavian Languages. Ph.D. dissertation, University of California, Santa Cruz. Svenonius, Peter. 1996. Predication and functional heads. In Jose Camacho, Lina Choueiri, & Maki Watanabe (eds.), The Proceedings of the Fourteenth West Coast Conference on Formal Linguistics, 493-507. CSLI, Stanford, Calif. Taraldsen, Knut Tarald. 1980. On the Nominative Island Condition, Vacuous Application and the Thai-Trace Filter. Indiana University Linguistics Club, Bloomington.
INTRODUCTION
27
Taraldsen, Knut Tarald. 1990. Second thoughts on der. In Hans-Georg Obenauer and Anne Zribi-Hertz (eds.), Structure de la phrase et theorie du liage, 217-249. Presses Universitaires de Vincennes, Paris. Taraldsen, Knut Tarald. 1994. Reflexives, pronouns, and subject/verb agreement in Icelandic and Faroese. Working Papers in Scandinavian Syntax 54, 43-58. Taraldsen, Knut Tarald. 1996. How to keep linguists happy: Review of The Minimalist Program by Noam Chomsky. GLOTInternational 2.3, 12-15. Thrainsson, Hoskuldur. 1979. On Complementation in Icelandic. Ph.D. dissertation, Harvard University. [Published 1979, Garland, New York.] Vikner, Sten. 1995. Verb Movement and Expletive Subjects in the Germanic Languages. Oxford University Press, New York. Woodbury, Anthony C. 1977. Greenlandic Eskimo, ergativity, and Relational Grammar. In Peter Cole & Jerrold M. Sadock (eds.), Syntax and Semantics 8: Grammatical Relations, 307-336. Academic Press, New York. Zaenen, Annie, Joan Maling, & Hoskuldur Thrainsson. 1985. Case and grammatical functions: The Icelandic passive. Natural Language & Linguistic Theory 3, 441-483.
This page intentionally left blank
2 The QuelQui Alternation and the Distribution of Expletives Knut Tarald Taraldsen
1. Introduction According to Rizzi (1990), qui is an agreeing form of C replacing que in (2) to provide a 'formal licenser' for a trace in SpecIP: (1)
Quel livre crois-tu que/*qui les filles vont acheter? which book think-you that the girls will buy
(2)
Quelles filles crois-tu *que/qui vont acheter ce livre-la? which girls think-you that will buy that book-there
There are various conceptual and empirical reasons to be dissatisfied with the analysis of qui as an agreeing C as well as with Rizzi's theory of formal licensing of traces. For example, que, being invariant and incompatible with a following subject, lacks the characteristic morphosyntactic properties of agreeing C in Germanic. And there is an alternative account of (l)-(2), essentially for free, that explains all properties of qui and is consistent with theories of trace-licensing where agreeing C plays no role. This analysis, which is suggested by certain observations about Vallader, a Romance language closely related to French, has it that qui is actually que followed by an expletive i. Thus, we will end up explaining the que/qui alternation by using a Minimalist version of Kayne and Pollock 1978 to answer the seemingly independent question why the null expletive of (3) must be replaced by its overt counterpart, in French, when the extracted w/i-phrase is the subject: (3)
Ou crois-tu que vont se cacher les chats? where think-you that will themselves hide the cats
29
30
SUBJECTS, EXPLETIVES, AND THE EPP
2. Why qui is not an agreeing C Rizzi's analysis of qui, essentially rephrasing Pesetsky 1982, leads to some expectations that are not fulfilled. It is striking, for instance, that qui does not actually show any morphophonemic reflexes of the person/number features it would acquire through agreement. Of course, this could be accidental. After all, within the West Germanic group of languages, where a particularly good case can be made for an agreeing C, there is a great deal of dialect variation with respect to the morphophonemic realization, ranging from a full paradigm of six distinct forms (West Flemish) to a single invariant form (e.g., Standard Dutch). Yet, it remains striking that not a single dialect of French (or any other Romance language) has any morphophonemic reflex of agreement in its complementizer system. Qui also has a syntactic property which is not shared by agreeing complementizers in West Germanic. It cannot occur unless the subject of the clause has been wh-moved. Rizzi accounts for the contrast (1) versus (2) in terms of Relativized Minimality. Taking SpecCP as an A-position whenever C holds agreement features, he points out that (1), with qui instead of que, would be well-formed only if the object could move to an A-position across an intervening closer A-position (the subject). But that raises the question why West Germanic counterparts of (1) are perfectly fine with agreement in C,1 and the assumption about Spec of agreeing C being an A-position would seem to stand in the way of an improper movement account of (4): (4) * Les filles semblent qui vont acheter ce livre-la. the girls seem that will buy that book-there On the conceptual side, there is the question why a trace in SpecIP (AgrSP) should be allowed just in case C agrees with it (via its agreement with SpecCP). In Rizzi's analysis, agreeing C shares with the lexical heads the property of being formal licensers. But why should this be? To the extent that a subject trace governed by agreeing C is formally licensed just because the two are coindexed (assuming agreement to be represented by coindexing), we bring back a conceptual problem that Rizzi's formulation of the ECP was designed to eliminate, the one concerning the disjunction between lexical government and antecedent government in the classical statement of the ECP. Rizzi's analysis eliminates the disjunction from the statement of his Identification Requirement, but it is now seen to return in disguise in the formulation of Formal Licensing, as a direct consequence of Rizzi's analysis of qui.
3. The chalchi alternation In Vallader, a Rhaeto-Romance variety spoken in the Engadine, we find a similar alternation between cha (= que) and chi (= qui) in sentences like (5) and (6): (5)
Qual cudesch crajast cha/*chi las mattas cumpraran? which book think.you that the girls will.buy
(6)
Qualas mattas crajast chi/*cha cumpraran quel cudesch? which girls think.you that will.buy that book
Vallader
THE QUE/QUI ALTERNATION AND THE DISTRIBUTION OF EXPLETIVES
31
In the absence of further information, one might well attempt to apply Rizzi's analysis here, too. But Vallader also has chi introducing sentences from which nothing has been extracted, provided the subject is not preverbal, as well as a 'free' expletive subject i, strongly suggesting that the chi of (7) is simply elided cha followed by expletive i:2 (7) . . . la spranza chi/*cha turnaran quels temps docts Vallader the hope that \vill.return those times learned (8)
I turnaran quei temps docts it will, return those times learned
This suggestion is reinforced by the observation that i and chi share the idiosyncratic morphophonemic property of adding a -d before vowels:3 (9)
. . .il vast territori. . . pro'l qual id appartaignava eir Iowa the vast territory to which it belonged also Iowa (10) El disch chid es turna Peider he says that.it is returned Peter
Vallader
This idiosyncracy is also shared by the chi of (6): (11) Qualas mattas crajast chid ban cumpra quel cudesch? which girls think.you that.it have bought that book
Vallader
So Vallader probably w/i-moves subjects from a position lower than SpecIP, just like Italian.4 According to Rizzi 1982, an Italian sentence like (12) would have an analysis like the one in (13), with a covert expletive (pro) in SpecIP:5 (12) Quali ragazze credi che compreranno quel libro? Italian which girls think.you that will.buy that book (13) quali ragazze . . . [CP che dp pro compreranno+I [VP tpro tv quel libro. . . The corresponding Vallader example in (6) should be analyzed the same way, except that Vallader, not a null-subject language, requires an overt expletive in SpecIP: (14) qualas mattas . . . [CP cha [jp i cumpraran+I [VP t\ ?v Quel cudesch]]] As in Italian, the subject trace causes no that-trace problem because it is not in SpecIP.
4. French qui = qu(e) i Rethinking the French quelqui alternation along similar lines, we would take qui as qu(e) i, with expletive i in SpecIP:6 (15) quelles filles . . . [CP que [jP i vont+I [VP t\ fy acheter . . . The expletive / is different from the usual French expletive il in three respects: It never has a final -1 (in any register), it does not control subject/verb agreement (always 3so with il), and it does not require an indefinite associate (cf. *Quelles filles est-ce qu'il est arrive? vs. Qu'est-ce qu'il est arrive?}. An immediate advantage of this analysis is that the ungrammatically of qui in (1) becomes straightforward, provided a preverbal subject must be in SpecIP rather
32
SUBJECTS, EXPLETIVES, AND THE EPP
than in the low position held by the w/i-trace in (15), leaving no room for the expletive.7 This approach is also in a sense immanent in Kayne and Pollock's 1978 analysis of (3) as having a null expletive which fails to control verbal agreement or induce an indefiniteness requirement on its associate: (3)
Ou crois-tu que vont se cacher les chats? where think.you that will themselves hide the cats
(16) ou . . . [CP que [\ppro vont+I [VP les chats tv se cacher . . . The structure in (16) is licensed by w/i-movement out of the embedded clause. In (3), the relevant w/z-phrase is ou 'where,' an adverbial complement. If the w/i-phrase is the subject of the embedded clause, it should still be possible to have (the trace of) the subject in the low subject position, to the extent that this depends only on having a w/i-phrase passing through the embedded SpecCP. Yet, the structure in (17) must be ill-formed, since otherwise (2) (repeated below) would be expected to be possible with que: (2)
Quelles filles crois-tu *que/qui vont acheter ce livre-la? which girls think.you that will buy that book-there
(17) quelles filles . . . [CP que [\ppro vont+I [VP tpro tv acheter ce livre-la In particular, there should be no *that-trace effect. Kayne (1980) suggests that the null expletive fails to be licensed in structures like (17). But then, the natural expectation should be that an overt expletive with the same properties as the covert expletive of (16) would show up in connection with subject extraction in French, especially since Vallader shows that subject extraction from a 'low' position does not depend on the expletive being null. On the analysis of qui as qu(e) i, this expectation is fulfilled, since i is an overt expletive which, like the pro of (16), but unlike il, does not control verbal agreement or require an indefinite associate.8
5. Another manifestation of French expletive i Unlike Vallader, Standard French does not seem to have an expletive in isolation from que: (18) *Elle dit qui sont arrivees ses copines. she says that.i are arrived her girlfriends (19) * I sont arrivees ses copines. i are arrived her girlfriends But in 'complex subject inversion' environments, colloquial styles of French allow an invariant (f)i: (20) Pourquoi tu dois-ti partir? why you must-ti leave Putting aside the status of the initial t-,9 one can profitably look at this i as another manifestation of the expletive / following que in (2). In particular, it is invariably /,
THE QUE/QUI ALTERNATION AND THE DISTRIBUTION OF EXPLETIVES
33
and it does not control verbal agreement or induce an indefiniteness requirement on its associate. The classical analysis of complex inversion (Kayne 1983, Rizzi and Roberts 1989) would parse the Standard French sentence (21) as (22): (21) Pourquoi tes copines doivent-elles partir? why your girlfriends must-they leave (22) [CP pourquoi [CP tes copines [c doivent] [IP elles . . . The standard analysis of complex inversion applies to the '^'-construction' as well, once the morphological differences between i and expletive il are taken into consideration, so that the (t)i of (20) is licensed in the following context with C filled by the finite V: (23) . . . S T J B J C [ I P i I . . . f S U B J In particular, the ^-construction, just like the standard construction, is confined to main clauses, a property generally attributed to V-to-C movement.10 Unlike the standard construction, however, the ^'-construction allows the preverbal subject to be a pronoun, and (f)i, unlike standard subject clitics, does not appear unless there is an additional subject: (24) * Pourquoi tu dois-il /tu partir? why you must-he/you leave (25) *Pourquoi elle doit-il /elle partir? why she must-he/she leave (20) Pourquoi tu dois-ti partir? why you must-Ti leave (26) Pourquoi elle doit-(t)i partir? why she must-Ti leave (27) Dans quel pays habite-t-il? in which country lives-T-he (28) *Dans quel pays habite-ti? in which country lives-Ti Both contrasts follow from (t)i having a more impoverished morphology than the other subject clitics. In particular, it lacks gender and number specifications. If arguments must be endowed with such features, (28) is a straightforward violation of the 0-criterion.11 The ill-formedness of (24)-(25) has been taken to reflect a general prohibition against spelling out a morphological feature in more than one position in a chain (Kayne 1983). In (24)-(25) gender and number features are spelled out both by the preverbal subject pronoun and by the postverbal subject clitic, hence twice within the same chain, since the subject pronoun and the subject clitic are associated with the same 0-role.12 In (20)-(26), on the other hand, the gender and number of the chain are specified only by the preverbal subject pronoun, since (f)i lacks these features.
34
SUBJECTS, EXPLETIVES, AND THE EPP
6. The distribution of French expletive i According to Rizzi 1990, the w/i-moved subject in (2) moves through the embedded SpecCP: (15) quelles filles . . . [CP *SUBJ que [IP i vont+I [VP fSUBJ tv acheter . . . If so, we have that the i in (2) occurs in the same environment (16) which licenses the (t)i of colloquial complex inversion, modulo C = que versus C = V (but see note 8). Why should this be the only context licensing French expletive i! And why should French differ from Vallader? In section 5, we found reasons to maintain that French expletive / does not have (specified) gender and number features. In particular, the contrasts between i and other subject clitics exhibited by (24)-(28) follow from that hypothesis. We will now argue that the limited distribution of French expletive i follows in toto from its impoverished morphology. If i has no number feature, and finite I has an uninterpretable number feature that must be checked before Spell-out, then sentences like (18)-(19) (repeated below) have no convergent derivation: (18) *Elle dit qui sont arrivees ses copines she says that are arrived her girlfriends (19) * I sont arrivees ses copines I are arrived her girlfriends In structures subsumed under (23), however, the number feature may raise from I to C, where it is checked pre-Spell-out by the raised subject DP, as is implicit in standard accounts of complex inversion.13 This raises the question why the number feature of I couldn't be checked by raising the number feature of the subject DP before Spell-out. Chomsky (1995) stipulates that pure feature movement is possible only after Spell-out, but the analysis just proposed for structures conforming to (23) would be incompatible with that. Rather, the analysis presupposes that the dichotomy 'weak' versus 'strong' is reinterpreted according to a suggestion in Taraldsen (1996), who proposes to identify covert movement with pre-Spell-out pure feature movement, eliminating LF-movement: A strong feature is one that cannot raise without pied-piping its host category. From this point of view, our claim really is that D has a strong number feature in this new sense in French, but not in Vallader, while I in both languages has an uninterpretable number feature that can raise without pied-piping its host.14 In order to account for the obligatory presence of / in (23), we also have to assume that I has an EPP feature which does not raise to C. (29) displays the movements and checking relations relevant to licensing structures of the form (23) in French: (29) Checking:
number EPP . . . SUBJ C [IP i I . . . tS U B J • • • raising of I's number feature
In this perspective, the limited distribution of French expletive i follows from its being a 'pure expletive,' in Chomsky's (1995: ch. 4) sense, in a language where a
THE QUE/QUI ALTERNATION AND THE DISTRIBUTION OF EXPLETIVES
35
number feature cannot raise from D without pied-piping the DP. This might make it tempting to account for the contrast between French and Vallader i by identifying French i with y 'there' versus Vallader i(d) = 'it,' an expletive endowed with a number feature. This temptation should probably be resisted, however. If Vallader i(d) had a number feature like it, it would presumably be singular. But the number feature of I in Vallader sentences with expletive i(d) is not always singular, but corresponds to the number of the associate of the expletive, as in the case of 'pure expletives,' as we already saw in (7)-(8) (repeated below), where the verb occurs in a plural form, agreeing with the postverbal subject: (7)
. . . la spranza chi/*cha turnaran quels temps docts the hope that will, return.PL those times learned
(8)
I turnaran quei temps docts it will, return. PL those times learned
Vallader
Thus, Vallader i(d) does not have a specified number feature either. To account for the contrast between French and Vallader, we therefore need the assumption that in Vallader, the number feature of a postverbal subject DP can raise to I without pied-piping the DP.15 In Vallader, there is no counterpart to French v, that is, no other numberless form to which i(d) might be assimilated. Rather, we take it that i(d) descends from Latin id 'it,' but has lost its gender/number features in the transition from the Latin three-gender system to the Romance two-gender system. More specifically, we suggest that the Latin neuter pronoun lost its gender feature altogether by refusing to become masculine, and that the number feature is dependent upon gender, and therefore was lost, too. If this is correct, there is now no reason not to consider the French expletive a result of the same development.
7. French Stylistic Inversion The claim that French does not allow a number feature to raise from postverbal subject DP without pied-piping the DP appears to be contradicted by the existence of sentences with Stylistic Inversion, such as (3): (3)
Oii crois-tu que vont se cacher les chats? where think.you that will themselves hide the cats
According to the classical analysis of Stylistic Inversion (Kayne and Pollock 1978), (3) would have a structure like (16) (repeated below), where pro would be a 'true expletive,' like i, and the number feature of I would be checked by covert raising of the associate (les chats) or its number feature: (16) ou . . . [CP que [IP pro vont+I [Vp les chats ?v se cacher . . . The classical analysis, however, encounters certain difficulties that lead us to think that it should be discarded for independent reasons. Most notably, it fails to explain why a null expletive should be possible in structures where a w/i-phrase has been raised to the immediately higher SpecCP (possibly as a step in successive cyclic w/i-movement), but not elsewhere.16 Kayne (1980) suggests that the raised
36
SUBJECTS, EXPLETIVES, AND THE EPP
w/z-phrase triggers covert movement of the expletive's associate, which in turn licenses pro. In French, according to this analysis, the null expletive would otherwise be illicit, since French is not a null-subject language. But how can post-Spellout movement be relevant to whether an expletive is pronounced or not, a PF-property? (On the approach taken in section 5, where covert movement is replaced by pre-Spell-out feature movement, this problem does not necessarily arise, as pointed out by P. Svenonius (p.c.), but there would instead be another problem: Why should formal features of the postverbal DP be able to rise to I without pied-piping the DP only in Stylistic Inversion contexts?) Besides, if Kayne 1994 is correct, we need to be able to account for the fact that the whole VP (excluding a V-movement account) precedes the postverbal subject in sentences like (3) without appealing to rightward movement of the subject. We will assume that Stylistic Inversion is in fact as required by Kayne's 1994 theory: The subject moves out of the VP to a relatively low Spec-position, and the VP subsequently crosses over it to a higher Spec-position.17 In particular, one may assume that VP-raising can be scrambling to a position lower than I: (30) . . . Spec I . . . [VP t . . . ] . . . SUBJ . . . rvp . . . In that case, SpecIP must be filled by an expletive which can be covert in a nullsubject language like Italian, but not in French or Vallader. Whereas Vallader can always use i(d), since the uninterpretable number feature of I can be checked by pure feature movement from the postverbal subject in Vallader, French can resort to i only if the subject raises to SpecCP before Spell-out. Thus, (30) represents a legitimate stage in the derivation of the structures subsumed under (23), and only those. Sentences like (3) must be analyzed as in (31), with the VP raised to SpecIP: (31) . . . [ CP que [IP [VP tSUBI vont se cacher] I . . . les chats . . . fvp . . . In (31), the number feature of I is checked by the number feature of the subject trace in SpecVP: On Kayne's 1994 theory, the Spec of the Spec of XP is in the checking domain of X.18 Since the trace is a copy of the subject, the appearance of agreement with the low subject is created. Alternatively, SpecVP licenses a number feature on the head of VP, and this number feature projects to VP and licenses the number feature of I in (31). Again, I will appear to agree with the low subject. When the VP is scrambled, as in (30), the chain linking the low subject to its trace in SpecVP is recoverable via reconstruction, since scrambling generally allows for reconstruction. Movement to SpecIP, an A-position, should not reconstruct, however.19 Therefore, structures like (31) end up violating the ©-criterion unless the low subject eventually raises to a position c-commanding the trace inside the VP in SpecIP, if we adopt the traditional assumption that the head of a chain must c-command all other members of the chain at LF. Like Kayne 1980, we may assume that the subject raises (covertly) just in case a w/i-phrase has been raised to the immediately higher SpecCP.20 But unlike Kayne 1994, we succeed in accounting for the restrictions on Stylistic Inversion without having LF-movement sanction a PF-property.
THE QUE/QUI ALTERNATION AND THE DISTRIBUTION OF EXPLETIVES
37
Even though we have so far analyzed French Stylistic Inversion as VPmovement to SpecIP, it is clear that what raises must be an extended projection of the V including a number of functional heads above the bare VP. In particular, the constituent raised to SpecIP must contain the position(s) that clitic pronouns move to, and therefore, also all adverbial positions occurring in between the clitic position and the VP:21 (32) Pourquoi se sont caches les chats? why themselves are hidden the cats (33) *Pourquoi sont caches les chats se? why are hidden the cats themselves Hence, a French finite clause must in general be as in (34), where either a DP or the complement of F raises overtly to SpecIP (unless an expletive is inserted): (34) . . . [IP Spec I [pp Spec F [xp . . . Cl. . . VP Taking the second option after the subject DP has moved to SpecFP gives rise to Stylistic Inversion.22 In (34), I and a functional head F' in the complement of F must both bear agreement features, F1 licensing the corresponding features of the V. The existence of F' follows from the standard Minimalist assumption (modulo Chomsky 1995: ch. 4) that the agreement features of the V must be checked against the features of a functional head to which V adjoins, plus the fact that if the XP containing the V raises to SpecIP, V cannot adjoin to I (or F). Taking F1 to be X, the head of the complement of F, we may exploit this property to find a way around an outstanding problem for the analysis outlined above: (35) Quand crois-tu que sera precede a un reexamen de la when thinLyou that will.be proceeded to a reexamination of the question? case The distribution of impersonal sentences without expletive il, like the embedded clause in (35), is subject to the same limitations as are sentences with Stylistic Inversion (cf. Kayne and Pollock 1978). Our analysis of Stylistic Inversion should therefore extend to (35), but (35) seems to lack both a (visible) subject DP that could raise to SpecFP, and a trace of that subject in SpecXP that would license the number feature of I after raising of XP to SpecIP. We might now say that the number feature of I in Stylistic Inversion contexts is not licensed by the SpecXP, but rather by the number feature of X projected onto XP (the second of the two possibilities mentioned above), and allow X, but still not I, to be singular by default. Assuming that F attracts PP (a un reexamen de la question in the case of (35)) as well as DP, to SpecFP, the similarity with the Stylistic Inversion paradigm will be accounted for. Thus, we end up with an analysis according to which a finite clause contains two functional heads, I and F, associated with EPP features, whose exact nature remains to be investigated further. Similarly, two distinct functional heads, I and X, carry number features. Unlike X, however, I does not allow default licensing of its num-
38
SUBJECTS, EXPLETIVES, AND THE EPP
her feature. Like all uninterpretable features, the number feature of I must therefore be checked before Spell-out, that is, SpecIP must be filled by a YP bearing a number feature, DP or XP, or the number feature of I raises to C and is checked by an appropriate phrase in SpecCP, as indicated in (29). Given the ungrammaticality of Stylistic Inversion with transitive verbs in French, a property illustrated by (36) below, it is perhaps possible to identify the functional head X with AgrO: (36)* Ou crois-tu que vont cacher leurs jouets les chats? where think-you that will hide their toys the cats If AgrO, like I, can Case-license a nominative DP checking its number feature, as suggested by Icelandic sentences with oblique subjects, like (37), the ungrammaticality of (36) could be attributed to the postverbal subject failing to have its Case licensed. It is neither in SpecIP nor in SpecAgrOP: (37) Okkur lika hestarnir. we. DATIVE like. 3PL horses, the
Icelandic
8. The *que-trace effect In section 4, we saw that Kayne and Pollock's (1978) analysis of Stylistic Inversion raises the question why the position of the postverbal subject of a sentence like (3) cannot be filled by a w/i-trace so that (38) would be analyzable as (39), which does not violate the ECP/*that-trace filter: (3)
Ou crois-tu que vont se cacher les chats? where think.you that will themselves hide the cats
(38) *Quels chats crois-tu que vont se cacher? which cats think.you that will themselves hide (39) quels chats . . . [CP que [\ppro vont+I [VP t V se cacher . . . The analysis of Stylistic Inversion in section 7 raises a similar question. Why cannot (38) receive the analysis in (40)? (40) quels chats...[cp que [ip {vp tsSUBj vont se cacher] I ...t...tvp... Again, we assume that the ECP/*that-trace filter would not be violated. If (40) is to come out well-formed at LF, the subject trace inside the raised VP must have an antecedent in a c-commanding A-position. Since the w/z-phrase itself is not likely to be in an A-position or to have left behind traces in A-positions c-commanding the embedded IP, the trace in the low subject position must move covertly to a IP-external A-position, a movement triggered by the presence of (the trace of) a w/i-phrase in SpecCP. If it cannot, (40) has no well-formed LF-representation. For the required movement to take place in (40), (a trace of) the w/z-phrase must attract the low subject the same way it does in the grammatical instances of Stylistic Inversion discussed above. But if attraction ought to culminate in 'quantifier absorption,' as suggested by Kayne (1980), one would not expect to see its ef-
THE QUE/QUI ALTERNATION AND THE DISTRIBUTION OF EXPLETIVES
39
fects in (40), where the low subject is not interpretable as a separate quantifier to be absorbed, but rather as the variable to be bound by the w/i-quantifier. More generally, whatever turns out to be the correct answer to the independent question why (40) cannot converge will combine with the ECP/*that-trace filter to explain the *
9. Summary This article started out from a critique of Rizzi's 1990 account of the French quelqui alternation. Although the analysis proposed here endorses Rizzi's hypothesis that a w/i-phrase in SpecCP may agree with a feature in C (a number feature raised from I, in our analysis), it does not consider qui as a spell-out of [c Agr]. Nor do we need to assume that SpecCP is ever an A-position, or that functional heads containing agreement have a privileged status, alongside (certain) lexical categories, with respect to the licensing of traces. The analysis of the qui in examples like (2) as the usual complementizer que followed by expletive, on the model of the Vallader data discussed in section 3, explains the characteristic properties of this qui, and draws support from the existence of expletive (f)i in complex inversion structures in colloquial French. If our characterization of the properties of the latter is correct, our analysis of the former is essentially for free, as promised in section 1, and different analyses introduce unwarranted complications over the minimum required to account for the empirical facts. The analysis presented in this chapter rests on certain assumptions concerning the distribution of formal features over functional heads and the way these are licensed. In particular, number features are taken to be associated with two distinct functional heads in the clausal skeleton, a proposal that may be assimilated to the classical distinction between AgrO and AgrS (represented as I above). EPP features also occur in different positions, that is, on I and F (the functional head intermediate between I and XP (perhaps = AgrOP) according to the analysis proposed in section 7. It remains unclear exactly what the content of the EPP features are: the EPP feature of F can be licensed both by DP or PP, if the analysis given in section 7 is correct, and the EPP feature of I is licensed indifferently by DP (including i) or XP (= AgrOP). In the latter case, a common denominator can perhaps be extracted, since D and AgrO may share certain features, including nominative case, to the extent that AgrO may license nominative case on its Spec in Stylistic Inversion contexts. As for the licensing requirements, a departure from Chomsky's (1995: ch. 4) theory has been proposed: all uninterpretable features must be checked before Spell-out, but pure feature movement is allowed before Spell-out, suggesting the elimination of post-Spell-out movement. An uninterpretable feature of a functional head will behave as a 'strong' feature in Chomsky's sense in contexts where the feature needed to check it cannot raise from its host without pied-piping it. Otherwise, it corresponds to a 'weak' feature in Chomsky's framework. Thus, the outcome is determined by the feature architecture of the host of the attracted feature, not by the attractor. In French, but not in Vallader, the number feature of D
40
SUBJECTS, EXPLETIVES, AND THE EPP
must pied-pipe the DP. Needless to say, more research on feature architecture is needed to give this proposal more substance.
Notes This material has been presented in talks at NELS 27 (McGill University) and the University of Venice as well as at the Colloquium on the EPP and Expletives at the University of Troms0. I am grateful to the respective audiences for questions and comments. I am particularly indebted to Anders Holmberg and Peter Svenonius for their comments on a preliminary drafts. I take responsibility for all remaining errors. 1. West Germanic agreeing complementizers are also possible when nothing at all has been moved to the corresponding SpecCP. 2. See Linder 1987, which (7) is taken from. The other Vallader examples are constructed on the basis of Linder 1987 or Liver 1991, except when otherwise noted. 3. The -d is perhaps a reflex of i(d) being the descendant of Latin id 'it.' Example (9) is from Linder 1987. 4. Notice that Vallader i(d), unlike the subject clitics of North Italian dialects, never co-occurs with preverbal 'full' subjects; cf. Linder 1987, Manzini and Savoia 1998. Thus, taking the w/z-phrase to extract via a preverbal subject position in (6) would complicate the analysis of the general pattern. 5.1 take the lower subject position to be SpecVP. 6. The possibility that the French qui of (2) should be analyzed as que followed by an expletive, on the model of Rhaeto-Romance, is suggested independently by Manzini and Savoia (1998), who also point out that Rohlfs ([1949] 1968) proposes que il as the historical source of il. Manzini and Savoia do not, however, discuss the status of French i vs il. I. In other words, the subject may surface in a position lower than SpecIP only in sentences with 'Stylistic Inversion,' e.g,. (3). This will follow from the analysis presented in section 8. 8. In section 7, we will consider a different analysis of Stylistic Inversion, which does not make use of null expletives. Rather, the (extended) VP raises to SpecIP. When this movement is blocked, an (overt) expletive appears in SpecIP. 9. In view of the analysis given in section 6, it is important that this t- not be identified as the 3so verbal agreement inflection. Such an interpretation would not be straightforward, since t- combines with non-3so verbal inflection in examples like (i): (i) Pourquoi nous devrions-ti partir? The contrast *il at ete . . . vs Us ont ete . . . , with pronounced final t, may also indicate that not even the t- of a-t-il ete . . . ? is a part of the verbal agreement inflection escaping phonological deletion in liaison contexts. 10. Sportiche (n.d.) and Kayne (1994) propose that the verb does not move to C in the complex inversion construction, and that the preverbal subject is not in SpecCP (or adjoined to CP). As will become evident in section 6, the choice between their analysis of complex inversion and the one reproduced in (23) ultimately has no bearing on our account of French i, the relevant generalization being in either case that i occurs only in contexts where some other element can license the number feature of I. II. In section 6, we will see that (28) also is excluded for another reason. The number feature on AgrS fails to be licensed, since i lacks a number feature. Therefore, *Pourquoi est/sont-(t)i arrive tant de touristes? is also excluded, even though no violation of the 0-criterion arises. 12. Since Pourquoi la copine de Jean part-elle? is grammatical, although the gender and number of the chain are spelled out both by the postverbal subject clitic and by the the determiner la heading the prenominal subject (according to the DP-hypothesis), the general principle invoked in the text must be refined. Perhaps it is relevant that the gender and number features of D are checked by those of an N in la copine de Jean. The analysis by Sportiche (n.d.) and Kayne (1994) mentioned in note 10 might account for (24)-(25) vs (20)-(26) by assuming, as suggested by Cardinaletti (1996) and Manzini and Savoia (1998), that there are different (preverbal) subject positions specialized for different types of subject: ['strong' subject X ['weak' subject pronoun Y [ / . . . The fact that i should be restricted to a position lower than that of the i'/-series should still be attributed to a morphological difference between the
THE QUE/QUI ALTERNATION AND THE DISTRIBUTION OF EXPLETIVES
41
two, in fact the one assumed in the text, if, as proposed by Manzini and Savoia and by Cardinaletti, the elements occurring in the intermediate subject position must license number/person features of Y. 13. See Kayne 1989 and Taraldsen 1992, where it is likewise proposed that the number feature of I may raise to C and be licensed by SpecCP. Our analysis is incompatible with a strict version of Chomsky's (1995) idea that all strong features of X must be checked before XP is embedded in a larger structure. Nor is it consistent with Zwart's (1993) claim that features of X raised to Y cannot be checked against SpecYP. 14. This view of 'strong' vs 'weak' is not required for the case of complex inversion with i, if complex inversion is analyzed as suggested by Sportiche (n.d.) and Kayne (1994). Then, the number feature is checked in its base position either by a subject clitic in the intermediate subject position or by a 'strong' subject moving through that position on its way to the highest subject position (in the cases with 0. 15. This predicts that Vallader i(d) should never occur without an associate, a prediction borne out by the data we have had access to so far. 16. Similarly, pro is licensed in certain subjunctive clauses. 17. More accurately, an extended projection of V raises; see end of this section for discussion. Stylistic Inversion could not easily be analyzed as involving only movement of the finite V, since the postverbal subject also follows the past participle in the compound past tenses and certain complements of the V; cf. Kayne 1972. For the same reason, the postverbal subject cannot simply be taken to remain in SpecVP. 18. This presupposes that 'minimal domain' is defined on the basis of dominance rather than inclusion: The minimal domain of X contains just those categories that are not dominated by some other category in the domain of X. According to Kayne (1994), SpecXP is dominated only by a segment of XP. Overt agreement with the subject's Spec is seen in Aleut; cf. Bergsland and Dirks 1981. 19. Phrases in an A-position do not reconstruct with respect to A-binding, although they may do so with respect to quantifier scope; cf. May 1985 on 'quantifier lowering' effects. 20. Covert raising of the subject to a position c-commanding the raised V-projection must also be countenanced in certain subjunctive clauses; cf. Kayne and Pollock 1978. The required extension would not be more straightforward if we took an alternative approach to Stylistic Inversion licensed by w/z-movement, taking the subject DP to raise to SpecIP (licensing number agreement directly), while taking the complement ZP of I to have an initial Spec much like SpecCP, such that a w/z-phrase filling SpecZP pied-pipes XP when raising to the 'external' SpecCP. On this alternative analysis, the dependence of Stylistic Inversion on w/z-movement would reduce to pied-piping under w/z-movement, with the proviso that the XP must be stranded in the first external SpecCP under successive cyclic w/z-movement. 21. We agree with Kayne 1991 that clitics adjoin to a functional head, not to V itself. This functional head might be lower than the adverbial positions, though, if subject-initial sentences could be analyzed as having XP containing the clitic position(s) raised to SpecIP (above the adverbials) with the subject DP still in SpecXP rather than previously raised to SpecFP, as in the Stylistic Inversion cases. 22. If the subject DP remains inside XP raising to SpecIP, the outcome will show the same linear order as if the subject had raised to SpecIP alone. The ungrammaticality of (i) remains an embarrassment to the text analysis, or any other account of Stylistic Inversion in terms of 'remnant' VP raising: (i) * Ou se sont tous caches les chats? where themselves are all hidden the cats
References Bergsland, Knut, & Moses Dirks. 1981. Atkan Aleut School Grammar. University of Alaska, Anchorage. Cardinaletti, Anna. 1996. Subjects and clause structure. University of Venice Working Papers in Linguistics 6.1, 55-95. Chomsky, Noam. 1995. The Minimalist Program. MIT Press, Cambridge, Mass.
42
SUBJECTS, EXPLETIVES, AND THE EPP
Kayne, Richard S. 1972. Subject inversion in French interrogatives. In J. Casagrande & B. Sacluk (eds.), Generative Studies in Romance Languages, 70-126. Newbury House, Rowley, Mass. Kayne, Richard S. 1980. Extensions of Binding and Case-marking. Linguistic Inquiry 11, 75-96. Kayne, Richard S. 1983. Chains, categories external to S, and French Complex Inversion. Natural Language and Linguistic Theory 1, 107-139. Kayne, Richard S. 1989. Notes on English agreement. CIEFL Bulletin 1, 41-67. Kayne, Richard S. 1991. Romance clitics, verb movement and PRO. Linguistic Inquiry 22, 647-686. Kayne, Richard S. 1994. The Antisymmetry of Syntax. MIT Press, Cambridge, Mass. Kayne, Richard S., & Jean-Yves Pollock. 1978. Stylistic Inversion, successive cyclicity, and Move NP in French. Linguistic Inquiry 9, 595-621. Linder, K. P. 1987. Grammatische Untersuchungen zur Charakteristik des Ratoromanischen in Graubiinden. Gunter Narr Verlag, Tubingen. Liver, R. 1991. Mannel pratique de romanche. Sursilvan—Vallader. Ediziun Lia Rumantscha, Cuira. Manzini, M. Rita, & Leonardo M. Savoia. 1998. Varieta romanze a verbo secondo. In Gaetano Chiappini (ed.), Studi in memoria di Giorgio Chiarini, 459-492. Editore Alinea, Florence. May, Robert. 1985. Logical Form: Its Structure and Derivation. MIT Press, Cambridge, Mass. Pesetsky, David. 1982. Paths and Categories. Ph.D. dissertation, MIT. Rizzi, Luigi. 1982. Issues in Italian Syntax. Foris, Dordrecht. Rizzi, Luigi. 1990. Relativized Minimality. MIT Press, Cambridge, Mass. Rizzi, Luigi, & Ian Roberts. 1989. Complex Inversion in French. Probus 1, 1-30. Rohlfs, G. 1968 [1949]. Grammatica storica delta lingua italiana e dei suoi dialetti. Morfologia. Einaudi, Torino. Sportiche, Dominique, (n.d.). Subject clitics in French and Romance: Complex Inversion and clitic doubling. Ms., UCLA. Taraldsen, Knut Tarald. 1992. Second thoughts on der. In Hans-Georg Obenauer & Anne Zribi-Hertz (eds.), Structure de la Phrase et Theorie du Liage, 217-249. Presses Universitaires de Vincennes, Paris. Taraldsen, Knut Tarald. 1996. Review of The Minimalist Program by Noam Chomsky. GLOT International 2.3, 12-15. Zwart, Jan-Wouter. 1993. Dutch Syntax: A Minimalist Approach. Ph.D. dissertation, University of Groningen.
3 Icelandic Expletive Constructions and the Distribution of Subject Types 0ystein Alexander Vangsnes
1. Introduction Icelandic expletive-associate1 constructions show that the widespread view that the expletive enforces the so-called Definiteness Effect on the associate, or the 'logical subject,' cannot be correct. Uncontroversial data presented below tell us this, notably the fact that universally quantifying noun phrases are allowed as associates in Icelandic expletive constructions. The prime concern of this chapter will be to discuss this fact along with other issues pertaining to the distribution of subject types, as well as the role played by the expletive. The organization of the chapter is as follows. In section 2 I present the fact that the associate of Icelandic expletive constructions may occur in two positions, either postverbally or in a position between the finite and the nonfinite verb (henceforth 'the intermediate position'). In this respect Icelandic contrasts with Mainland Scandinavian, where the associate can occur in a postverbal position only (unless topicalized or w/i-moved). Moreover, it seems quite reasonable that the existence of transitive expletive constructions (TECs) in Icelandic, but not in Mainland Scandinavian, is related to the general availability of the intermediate position for the associate, since the associate of a TEC is the external argument (i.e., the subject), and thus 'needs' a preverbal position. In section 31 show that the two positions differ with respect to what kind of noun phrases are allowed to occur: whereas noun phrases in the postverbal position are subject to a Definiteness Effect similar to that observed in English and Mainland Scandinavian, the intermediate position is less restrictive in that it allows both universally quantifying and partitive noun phrases in addition to ordinary indefinites. On the other hand, the intermediate position is more restrictive in that bare noun phrases are awkward there, whereas they are fine in the postverbal position.
43
44
SUBJECTS, EXPLETIVES, AND THE EPP
The evidence presented in section 3 clearly suggests that the fairly well-established idea that the Definiteness Effect is triggered by the presence of the expletive as such cannot be correct, and in section 4 I briefly consider two such expletivebased approaches, those of Milsark 1974, 1977 and Safir 1985, 1987, before I develop an alternative approach based on the Case theories of Belletti 1988 and de Hoop 1992. Given that expletive-based accounts of the Definiteness Effect will be rejected, there are also reasons to believe that the idea of covert raising of the associate to the position of the expletive is on the wrong track. However, such raising has been motivated on the grounds of the existence of agreement between the verb and the associate, and in section 5.1 I therefore first present an approach to agreement which renders raising of the associate unnecessary. In section 5.2 I discuss the Extended Projection Principle, and by unifying certain ideas about subject-verb agreement in Alexiadou and Anagnastopoulou (1998) and Holmberg and Platzack (1995) I develop a theory of (Agr)SP in section 5.3 where (i) semantic relevance is attributed to that functional domain, (ii) the presence versus absence of expletives is attributed to the differing status of subject-verb agreement across languages, and (iii) the differing status of expletives across languages (notably in Icelandic versus Mainland Scandinavian) also pertains to issues concerning subject-verb agreement. Section 6 concludes the chapter.
2. Two positions for the associate in Icelandic As can be seen in sentences with both a finite and a nonfinite verb (at the same clausal level), the associate in Icelandic expletive constructions may occur either in a position after the nonfinite verb (i.e., 'postverbally'), or in a position between the finite and the nonfinite verb (i.e. 'verb intermediately'). In the recent literature on Scandinavian syntax (e.g., Bobaljik and Jonas 1996, Holmberg and Platzack 1995, Jonas 1994, Koeneman 2000, Koeneman and Neeleman 1997, Vikner 1995), this fact has been given considerable attention since it is one of the properties that distinguishes Icelandic from the closely related Mainland Scandinavian languages: in the latter the associate of expletive constructions may normally occur only in a position after the nonfinite verb.2 Compare the Icelandic sentence pair in (1) with the Norwegian one in (2). (1)
a. b.
(2)
a.
Pad hefur verid einhver kottur i eldhusinu. EXPL has been some cat in kitchen, the Pad hefur einhver kottur verid i eldhusinu. Both: There has been some cat in the kitchen'
Det bar vore ein katt pa kj0kenet. EXPL has been a cat on kitchen, the 'There has been a cat in the kitchen' b. * Det har ein katt vore pa kj0kenet.
Icelandic
Norwegian
The fact that Icelandic has so-called transitive expletive constructions (henceforth TECs), whereas the Mainland Scandinavian languages do not, can be taken to
ICELANDIC EXPLETIVE CONSTRUCTIONS
45
follow from the availability of the intermediate position for the associate in Icelandic. Generally, in Mainland Scandinavian only unaccusative and passive verbs may occur in there-type expletive constructions. In Icelandic there is no general restriction on the type of verb that may occur, but if the verb is not unaccusative or passive, such as a transitive verb involving an agent subject and a theme object, the associate can occur only in the intermediate position. Consider (3). The Norwegian examples in (4) illustrate that TECs are not possible in Mainland Scandinavian. (3)
a.
Pad hefur einhver kottur etid mysnar. EXPL has some cat eaten mice, the 'Some cat has eaten the mice' b. * Pad hefur etid einhver kottur mysnar.
(4)
a. * Det har ein katt ete mysene. EXPL has a cat eaten mice, the b. * Det har ete ein katt mysene.
Icelandic
Norwegian
Jonas (1994) and Bobaljik and Jonas (1996) propose that the intermediate position is SpecTP, and not SpecVP, which had been the general view earlier (cf., e.g., Holmberg and Platzack 1995; Sigurdsson 1989; Vikner 1995). Bobaljik and Jonas's claim is based on certain facts about the placement of the associate relative to sentential adverbs and shifted objects. Consider first the following example where the pronominal object is intended to be coreferent with 'school milk.' (Mjolk, 'milk,' is a feminine noun in Icelandic.)3 (5)
? Pad drekka sennilega sumir krakkar hana aldrei. EXPL drink probably some kids it.FEM never 'Some kids probably never drink it'
The judgments vary considerably here, and most informants react to the sentence in (5), hence the question mark, but given a suitable context (e.g., 'All schoolchildren are given free milk every day, but. . .'), the sentence turns out as not all that bad, and importantly, certain other word order patterns, to be presented shortly, cannot be saved by such contextual construction. In the sentence in (5) the object occurs in a position preceding the adverb aldrei 'never.' Such object shift is generally possible if the object is definite and the main verb has raised to the functional domain of the sentence (V-to-I). If the main verb stays in VP, as it does when we have a complex verb form, object shift is not possible. Moreover, in such cases we see that the adverb aldrei must precede the main verb. Consider the examples in (6). (Again the question mark on (6a) should be taken to indicate the general dislike of the sentence.) (6)
a. ? Pad hafa sennilega sumir krakkar aldrei drukkid hana EXPL have probably some kids never drunk U.FEM
b. * Pad hafa sennilega sumir krakkar hana aldrei drukkid. EXPL have probably some kids it.FEM never drunk c. * Pad hafa sennilega sumir krakkar drukkid hana aldrei. EXPL have probably some kids drunk it.FEM never
46
SUBJECTS, EXPLETIVES, AND THE EPP
If aldrei is generated in the same position in (5) and (6a), it means that object shift is movement of the object to a preverbal position. This in turn makes it unlikely that the associate (sumir krakkar) is in SpecVP. The view that Bobaljik and Jonas would take, and which I will adhere to, is that aldrei is uniformly adjoined to VP and sennilega 'probably' is uniformly adjoined to TP. Hence, given the clause structure CP—AgrSP—TP—VP, the associate occurs in SpecTP, whereas the shifted object must occur in a projection dominating VP. Bobaljik and Jonas (1996) claim that SpecTP is generally unavailable in Mainland Scandinavian, that is, also in nonexpletive constructions. However, Holmberg (1993) has argued convincingly that such a view is difficult to maintain, since subjects of nonexpletive constructions may both precede and follow sentence adverbs assumed to be adjoined to TP (see Svenonius, this volume, for examples and discussion). The question we thus need to ask with respect to the difference between Icelandic and Mainland Scandinavian is why SpecTP is available in Icelandic expletive constructions but not in Mainland Scandinavian ones. Although the question is far from trivial, it will be given a fairly trivial answer in the present account. I will assume that the difference in availability of SpecTP stems from the assumption that the Mainland Scandinavian expletive carries Case whereas the Icelandic one does not. That in turn raises the question why the expletive should differ with respect to this property in the two language types; a tentative answer to that will be given in section 5.3. Let us next consider the types of noun phrases that may occur in the postverbal and intermediate positions in Icelandic.
3. The distribution of noun phrase types in the two positions 3.1. The postverbal position Icelandic associates in the postverbal position are subject to the Defmiteness Effect more or less in the same way as associates are in English and Mainland Scandinavian expletive constructions.4 The examples in (7) show that definite descriptions (boldface),5 universally quantifying noun phrases (italicized), and partitive noun phrases (underlined) are not allowed in the postverbal position of Icelandic expletive constructions. (7)
a. b. c. d. e.
Pad hefur verid *kotturinn/*Petur i eldhusinu. EXPLhas been cat.the I Peter in kitchen.the Pad hefur verid *J)essi kottur/*kottur Peturs i eldhusinu. EXPLhas been this cat I cat Peter's in kitchen.the Pad hefur verid *serhver kottur /*einn af kottunum i eldhusinu. EXPLhas been each cat I one of cats.the in kitchen.the Pad hafa verid *allir kettirnirl*bddir kettirnir i eldhusinu. EXPL have been all cats.the I both cats.the in kitchen.the Pad hafa verid nokkrir kettir/*sumir kettir i eldhusinu. EXPL have been some cats I some.of. the cats in kitchen, the
ICELANDIC EXPLETIVE CONSTRUCTIONS
47
It is worth noticing that this restriction can also be observed when no overt expletive is present. A property distinguishing the Icelandic expletive from its Mainland Scandinavian counterpart is that it occurs only sentence initially. Still, if the single argument noun phrase of an intransitive sentence occurs postverbally and there is a nonexpletive constituent in sentence-initial position, it cannot be a definite description, a universally quantifying noun phrase, or a partitive. Consider the examples in (8). (8)
a. b. c.
d. e.
I dag hefur (*pad) verid *kb'tturinn/*Petur i eldhusinu. today has EXPL been cat. the I Peter in kitchen, the f dag hefur (*pad) verid *bessi kottur/*kottur Peturs i eldhusinu. today has EXPL been this cat I cat Peter's in kitchen.the I dag hefur (*pad) verid *serhver kb'ttur /*einn af kbttunum i today has EXPL been each cat lone of cats, the in eldhusinu. kitchen, the I dag hafa (*pad) vend *allir kettirnir/*badir kettirnir i eldhusinu. today have EXPL been all cats, the/ both cats, the in kitchen, the I dag hafa (*bad) verid nokkrir kettir/*sumir kettir i today have EXPL been some cats I some. of. the cats in eldhusinu. kitchen, the
The fact that the expletive cannot occur overtly in a noninitial position has been taken as evidence for claiming that the Icelandic (there-type) expletive actually never occurs in SpecAgrSP, but rather in SpecCP only (cf. Christensen 1991, Svenonius, this volume). However, there is one instance where the expletive can be claimed to surface in a position to the right of C, namely in embedded clauses following wh-type complementizers. Consider the following examples. (9)
Professorinn langadi ad vita . . . professor.the wanted to know a. . . . hvort bad hefdi einhver lokid ritgerdinni i gaer. if EXPL had someone finished thesis, the yesterday b. * ... hvort i gaer hefdi einhver lokid ritgerdinni. if yesterday had someone finished thesis, the c. * ... hvort i gaer hefdi pad einhver lokid ritgerdinni. if yesterday had EXPL someone finished thesis, the
(10)
Professorinn sagdi. . . professor.the said a. . . . ad pad hefdi einhver lokid ritgerdinni i gaer. that EXPL had someone finished thesis, the yesterday b. . . . ad i gaer hefdi einhver lokid ritgerdinni. that yesterday had someone finished thesis, the c. * ... ad i gaer hefdi bad einhver lokid ritgerdinni. that yesterday had EXPL someone finished thesis, the
48
SUBJECTS, EXPLETIVES, AND THE EPP
In embedded clauses introduced by the complementizer ad 'that,' topicalization is sometimes allowed (cf. (10b)). This has been analyzed as CP-recursion (cf. Vikner 1995 and references cited there). However, w/z-type complementizers such as hvort 'if/whether' generally do not allow such embedded topicalization. Accordingly, one may argue that CP-recursion is not allowed in embedded clauses. Nevertheless, expletives do occur, and if we assume the associate in the embedded TEC in (lOa) to occupy SpecTP, it seems quite straightforward to conclude that the expletive occupies SpecAgrSP. I will take this stand: the expletive is never base generated in SpecCP, but must at some point in the derivation occupy SpecAgrSP. This will be crucial to the discussion in section 5, and in 5.3 we will return to the question why the expletive does not surface in SpecAgrSP in main clauses. 3.2. The intermediate position Although the Definiteness Effect applies in cases where there is no overt expletive, there are cases where the restriction does not apply although the expletive is overt. In fact, associates in the intermediate position are not subject to the Definiteness Effect: although they cannot be unembedded definite descriptions, they may be both universally quantifying and partitive noun phrases, and importantly a universally quantifying noun phrase may contain an embedded definite. Consider the intransitive expletive constructions in (11) and the TECs in (12). (Again definites are boldface, universally quantifying noun phrases are italicized, and partitive noun phrases are underlined.) (11)
a. b. c. d. e.
(12) a. b. c. d. e.
Pad EXPL Pad EXPL Pad EXPL Pad EXPL Pad EXPL
hefur *k6tturinn/*Petur verid i eldhusinu. has cat. the I Peter been in kite hen. the hefur *{>essi k6ttur/*kottur Peturs verid i eldhusinu. has this cat I cat Peter's been in kitchen.the hefur serhver kottur/einn af ko'ttunum verid i eldhusinu. has each cat lone of cats, the been in kitchen, the hafa allir kettirnir/bddir kettirnir verid i eldhusinu. have all cats, the/both cats, the been in kitchen, the hafa nokkrir kettir/sumir kettir verid i eldhusinu. have some cats I some.of.the cats been in kitchen.the
Pad EXPL Pad EXPL Pad EXPL Pad EXPL Pad EXPL
hefur k6ttur/*kotturinn/*Kalli etid mysnar. has cat I cat.the I Kalli eaten mice.the hefur *j>essi kottur/*kottur Peturs etid mysnar. has this cat I cat Peter's eaten mice.the hefur serhver kottur/einn af kottunum etid mys. has each cat lone of cats, the eaten mice hafa allir kettirnir/bddir kettirnir etid mys. have all cats, the/both cats, the eaten mice hafa nokkrir kettir/sumir kettir etid mys. have some cats I some.of.the cats eaten mice
As we see from (lid) and (12d), the associate may contain the definite article, and it is important to notice that the associate may very well be contextually
ICELANDIC EXPLETIVE CONSTRUCTIONS
49
anaphoric. Consider the following authentic example taken from an advertisement: the text accompanied a drawing of personified condoms (i.e., with small faces and arms) performing various 'human' activities, such as smoking cigars, wearing sunglasses, and so on. (13)
Pad eru ekki allir smokkarnir profadir af RFSU. EXPL are not all condoms.the tested by RFSU 'Not all the condoms have been tested by RFSU'
The associate in this example is contextually anaphoric in that its referent is the totality of condoms in the drawing. If the definite article were left out of the associate, such a reading would not be salient. Rather, the resulting noun phrase, allir smokkar, would quantify over the universal set of condoms. (Cf. the corresponding difference between the English noun phrases all the condoms versus all condoms.} Whether or not universally quantifying noun phrases carry the definite article is of no consequence for the possibility of their occurring in the intermediate position— in the example in (14), also authentic (this time the title of a rock song), the universally quantifying noun phrase lacks the definite article and consequently denotes the totality of a universal set. (14)
Pad eru allir dagar eins i sveitinni. EXPL are all days alike in country side.the 'All days are alike out in the countryside'
The sentence in (14) furthermore illustrates a trivial, but important, fact: as long as an expletive sentence does not contain a nonfinite verb or a VP-adverbial, we cannot distinguish between the intermediate and the postverbal positions. However, we can legitimately conjecture that the associate in (14) must be in the so-called intermediate position (i.e., by hypothesis SpecTP), since if we were to convert the sentence into a variety that contains a complex tense form, the associate could occur only to the left of the nonfinite verb. This is illustrated in (15). (15)
Pad hafa (allir dagar) verid (*allir dagar) eins i sveitinni. EXPL have all days been all days alike in country side.the
3.3. Further facts It is furthermore worth noticing that generic readings are not available for the associate. Consider the sentence pair in (16) involving an individual-level predicate (ILP) which requires a generic reading of the bare plural subject—as we see, only the nonexpletive version of the sentence is grammatical. (16) a.
Fidluleikarar eru greindir. violists are intelligent b. * Pad eru fidluleikarar greindir. EXPL are violists intelligent
Moreover, the generic pronoun madur 'one' (literally 'man') cannot occur as associate in expletive constructions either. Compare the transitive sentence in (17a) with its TEC counterpart in (17b).
50
SUBJECTS, EXPLETIVES, AND THE EPP
(17) a.
b.
Madur/serhver krakki a ad taka ly si a hverjum degi. one leach kid ought to take cod.liver.oil on every day 'One/each kid should take cod liver oil every day' Pad a serhver krakki/*madur ad taka lysi a hverjum EXPL ought each kid I one to take cod.liver.oil on every degi. day
Importantly, however, the ungrammatically of (16b) does not hinge on the predicate being individual level. Milsark (1977) observed that ILPs are restricted from occurrence in existential sentences in English (see also Diesing 1992). However, ILPs are perfectly fine in Icelandic expletive constructions as long as the associate has a quantifier. (18)
Pad eru margir/allir fidluleikarar greindir. EXPL are many I all violists intelligent
In fact, with stage-level predicates (SLPs), which Milsark observed to freely occur in English existential sentences, we find the same contrast: bare plurals are not possible, whereas noun phrases containing a quantifier are fine. (19) a. * Pad EXPL b. Pad EXPL
eru fidluleikarar breyttir a tonleikunum. are violists tired at concert, the eru margir fidluleikarar preyttir a tonleikunum. are many violists tired at concert, the
Moreover, it is clear that with both ILPs and SLPs the associate must occur in the intermediate position. This is illustrated in (20) and (21). (20) a.
Pad EXPL b. * Pad EXPL
hafa margir fidluleikarar verid greindir i pessari hljomsveit. have many violists been intelligent in this orchestra hafa verid margir fidluleikarar greindir i pessari hljomsveit. have been many violists intelligent in this orchestra
(21) a.
hafa margir fidluleikarar verid preyttir a tonleikunum. have many violists been tired at concert, the hafa verid margir fidluleikarar preyttir a tonleikunum. have been many violists tired at concert, the
Pad EXPL b. * Pad EXPL
The conclusion, then, is that the ILP/SLP distinction plays no role with respect to Icelandic expletive constructions: with both kinds of predicates the associate must occur in the intermediate position, and it must contain a quantifier. The latter observation ties in with the observation that bare indefinites quite generally are awkward in the intermediate position (cf. Sigurdsson 1989, Jonas 1994, Bobaljik and Jonas 1996, Vangsnes 1995).6 Similar judgments are not reported for the postverbal position. Consider the examples in (22)-(24) not involving either ILPs or SLPs. (22) a.
Pad er eldadur fiskur i eldhusinu. EXPL is cooked fish in kitchen, the 'There has been cooked fish in the kitchen'
ICELANDIC EXPLETIVE CONSTRUCTIONS
51
b. ? Pad er fiskur eldadur i eldhusinu. EXPL is fish cooked in kitchen, the (23) a.
Pad hefurhellst bjor a golfid. EXPL has been.poured beer onfloor.the 'Beer has been poured on the floor' b. ? Pad hefur bjor hellst a golfid. EXPL has beer been.poured on floor, the
(24) a.
Pad hafa verid mys i badkerinu. EXPL have been mice in bathtub, the There have been mice in the bathtub' b. ? Pad hafa mys verid i badkerinu. EXPL have mice been in bathtub, the
Again, judgments vary, and it should be noted that many speakers find the (b) examples considerably more awkward than is indicated by the question mark. There do, however, seem to be at least two ways in which the associates in (22)-(24) may be more felicitous in the intermediate position, either by being interpreted contrastively or when they contain some descriptive attribute, for instance, an adjective. If the bare indefinites in (22b), (23b), and (24b) are stressed, the sentences are more acceptable, and moreover the following sentences are fine. (25) a.
b.
Pad hafa mys en ekki rottur verid i badkerinu. EXPL have mice but not rats been in bathtub, the There have been mice but not rats in the bathtub' Pad hefur uldinn fiskur verid eldadur i eldhusinu. EXPL has rotten fish been cooked in kitchen, the 'Rotten fish has been cooked in the kitchen'
Thus, it seems that for a bare indefinite to be felicitous in the intermediate position, it must have a nontrivial or marked interpretation. Such nontrivial interpretations are not salient for the isolated sentences in (22b), (23b), and (24b), but the reason why they are just odd, and not downright ungrammatical, is probably that the speaker with some effort is able to construct an appropriate interpretation. In some sense adding more descriptive material amounts to the same thing as allowing a contrastive interpretation: rotten fish is intrinsically contrasted with nonrotten fish, as it were. It then seems that the correct characterization of bare indefinites in the intermediate position is that they must allow a contrastive reading. We may in turn take contrast to involve a kind of quantification: in order to have a contrast there must be a set consisting of at least two members, one being the denoted entity, the other, the entity or entities with which the denoted entity is contrasted.7 Having said that, we can establish the following generalization: a referential noun phrase occupying the intermediate position of an Icelandic expletive construction must be quantificational. 3.4. Summary We may now summarize the findings as to what types of noun phrases are allowed as associates in the two positions, as shown in table 3.1.
SUBJECTS, EXPLETIVES, AND THE EPP
52
Table 3.1: The distribution of subject types in Icelandic expletive constructions Expletive
Intermediate position (SpecTP)
Postverbal position
Pad Pad Pad Pad Pad
*unembedded definite *generic V/partitive indefinite *non-Q bare indefinite
*unembedded definite *generic * V/partitive indefinite non-Q bare indefinite
Although they note the awkwardness of bare indefinites in the intermediate position, Bobaljik and Jonas (1996), like most authors before them (e.g., Sigurdsson 1989, Vikner 1995, Holmberg and Platzack 1995), fail to observe the difference between the two positions with regard to strong noun phrases (i.e., definite descriptions, generics, and universally quantifying and partitive noun phrases), and they discuss Icelandic TECs as if intermediate associates were subject to the Definiteness Effect in the same manner as postverbal associates.8 Partially on that basis Bobaljik and Jonas (1996) argue that the Mapping Hypothesis of Diesing 1992 must be revised.9 The core idea of the Mapping Hypothesis is that indefinite subjects in VP receive an existential/nonspecific reading ("mapped into the nuclear scope"), whereas indefinite subjects in IP receive either a generic or specific reading ("mapped into a restrictive clause"). If associates in the intermediate position were subject to the same restriction as associates in the postverbal position, that would not be in line with the Mapping Hypothesis, given that the intermediate position is SpecTP, an IP specifier. Bobaljik and Jonas (1996) therefore suggest modifying the Mapping Hypothesis so that subjects in SpecTP are mapped onto a semantic interpretation in the same way as VP subjects, and that only SpecAgrSP counts as the SpecIP position in Diesing's original formulation of the Mapping Hypothesis. It is, however, unclear how the Mapping Hypothesis can accommodate the Icelandic data. For one thing, there does not seem to be any difference in the semantic interpretation of the indefinite subject in the example set in (26): most informants agree that it can receive only an existential interpretation. (Accordingly, the determiner could appropriately have been glossed as 'sm,' that is, unstressed 'some.') (26) a. b. c.
Einhver nemandi hafdi fallid a profinu. some student had flunked on exam, the Pad hafdi einhver nemandi fallid a profinu. EXPLhad some student flunked on exam, the Pad hafdi fallid einhver nemandi a profinu. EXPL had flunked some student on exam, the
On the other hand, a noun phrase which necessarily must receive a specific (presuppositional) reading may occur in the very same positions as the subject in (26). Consider the following sentence where the subject is an intensifying noun phrase entailing presuppositionality on a par with indefinite-tfiw noun phrases in English (cf. Vangsnes 1999:75; also see note 4).
ICELANDIC EXPLETIVE CONSTRUCTIONS
(27) a. b. c.
53
Pessi lika gullfallega stulka hafdi fallid a prbfmu. this PRT gold.beautiful girl had flunked on exam.the Pad hafdi pessi lika gullfallega stulka fallid a profmu. EXPL had this PRT gold.beautiful girl flunked on exam.the Pad hafdi fallid pessi lika gullfallega stulka a profinu. EXPL had flunked this PRT gold.beautiful girl on exam.the
Accordingly, there is some uncertainty as to how the Icelandic data relate to the Mapping Hypothesis, and the issue awaits further investigation. However, the data presented clearly show that there are correlations between syntactic positions and types of noun phrases in Icelandic expletive constructions, and, what is furthermore important, in nonexpletive constructions, all the types may appear both in an initial position (i.e., SpecCP) and a noninitial position to the left of the intermediate position. Consider the following examples where different types of noun phrases appear sentence initially (i.e., in the position filled by the expletive in expletive-associate constructions). (28) a. b. c. d. e.
Stulkan hefur fallid a profinu. girl, the has flunked on exam, the Allir nemendurnir hafa fallid a profinu. all students, the have flunked on exam, the Einhver nemandir hefur fallid a profinu. some student has flunked on exam, the Nokkrir af nemendunum hafa fallid a profinu. some of students, the have flunked on exam, the Bjor hefur hellst a golfid. beer has been.poured on floor, the
The evidence from Icelandic then tells us the following: (i) the idea that the expletive triggers the Definiteness Effect (through coindexation, unselective binding, Case transmission etc.; cf. Lasnik 1995 for references and an overview) cannot be correct, (ii) the idea that the associate raises to the position of the expletive cannot be correct. In the following section I will explicate why these conclusions can be drawn and, moreover, in the remainder of the chapter I will sketch an alternative approach to expletive constructions which can better deal with the Icelandic facts. Let me first give the basic intuition behind my objections against these fairly wellestablished ideas. First of all, if some explicit expletive-associate relation were the trigger for the Definiteness Effect, and thus served to explain why universally quantifying noun phrases (along with defmites) are not allowed in the postverbal position, we should expect such noun phrases to be barred from occurrence in the intermediate position as well. If in turn the expletive does not trigger the semantic restrictions seen on the associate of expletive constructions, it seems entirely counterintuitive that there should be a difference between overt and covert movement of the associate to the position of the expletive: there are no semantic restrictions on subjects occurring to the left of SpecTP in nonexpletive constructions, so why should there be any if the associate moves to this (or these) LF-position(s) after Spell-out?
54
SUBJECTS, EXPLETIVES, AND THE EPP
The alternative approach that I will outline assumes that no covert movement of the associate takes place—in fact, associate raising to the position of the expletive cannot take place, and the semantic effects appear for exactly that reason. The analysis that I develop will rely on certain ideas in Alexiadou and Anagnostopoulou 1998 (henceforth A&A) as to how the Extended Projection Principle is satisfied, though rethought in a particular framework in which the EPP has a more explicit semantic foundation than standardly assumed. As I will show, the fact that Icelandic has expletives can be taken to follow from the language not being a nullsubject/pro-drop language (in line with A&A), and the fact that SpecTP is available in expletive constructions can be taken to follow from Icelandic nevertheless having 'strong AGR' (contra A&A, but in line with Holmberg and Platzack 1995). The difference between Icelandic and Mainland Scandinavian (and English) will in turn follow from the latter having 'weak AGR.'
4. The Definiteness Effect and Case checking 4.1. Expletive-based approaches The semantic restrictions seen on noun phrases in the postverbal position of Icelandic expletive constructions are clearly the same as found in, for instance, English and Mainland Scandinavian languages, and it is of course desirable to give a unified account for all the languages. Also, whatever analysis we propose for English and Mainland Scandinavian should capture the full range of Icelandic facts. The purpose of this section is to accomplish that goal. Let us first consider some accounts that are inadequate. Milsark (1974, 1977) proposes that the expletive introduces an existential quantifier that must bind a variable, the variable being the associate. For this reason universally quantifying noun phrases cannot occur as associates of the expletive: their universal quantification would simply clash with the existential quantifier introduced by the expletive. Milsark further treats definiteness as a special kind of universal quantification, and accordingly he gives a unified account of why both definite descriptions and universally quantifying noun phrases are not possible as associates. Safir (1985,1987) argues that the Definiteness Effect occurs in what he calls unbalanced '0-chains,' which obtain, among other things, between the expletive and the associate in an expletive construction. Safir argues that the associate does not receive Case from the verb, and that it instead inherits it from the position of the expletive through coindexation. After modifying Principle C by substituting 'names' for 'R-expressions' (hence "names must be free") and defining the denotation of 'names' to be 'strong noun phrases' (i.e., definite descriptions and universally quantifying noun phrases), he argues that only indefinites can inherit Case from the expletive. With various modifications and extensions, Safir's general idea (i.e., that a relation between the expletive and the associate triggers the Definiteness Effect) has been applied to Scandinavian expletive constructions by Sigurdsson (1989:292ff.),
ICELANDIC EXPLETIVE CONSTRUCTIONS
55
Afarli (1992:70f.), Holmberg (1994:220, note 4), and Holmberg and Platzack (1995). Moreover, Safir's idea was also adopted in Chomsky 1986 and is, to my mind, implicit in Chomsky 1995. As they stand, neither the Milsarkian nor the Safirian approach to the Definiteness Effect can account for the Icelandic data, notably the fact that universally quantifying noun phrases may occur in the intermediate position. A way of saving Safir's analysis would be to say that when the associate occurs in the intermediate position, it does indeed get Case, say from T. Hence, no expletive-associate chain is established, and no Definiteness Effect occurs either. However, the question that would then arise is why the expletive is there in the first place, and the question seems unanswerable. Milsark's account cannot be saved either, since the expletive must bind a variable, and the variable is unequivocally the associate. 4.2. A Case-based approach Another approach to the Definiteness Effect can be attributed to Belletti (1988), who argues that be and other unaccusative verbs assign and license a particular Case, 'partitive Case,' to their postverbal argument, and that this Case is compatible only with indefiniteness, hence accounting for why the associate is subject to the Definiteness Effect when it occurs in the postverbal position. This approach is followed by Lasnik (1992, 1995). Concerning the Icelandic facts, a Case-based approach may appear more promising since it makes no direct reference to the presence of expletives, and we can locate the Definiteness Effect to the postverbal position as such. However, there are certain problems with Belletti's original theory. First, her choice of terminology is rather unfortunate, as it is quite clear that there is no direct relation between morphological partitive case (in Finnish) and Belletti's abstract partitive Case.10 Second, Belletti argues that (abstract) partitive Case is inherent, but there are both conceptual (cf. Lasnik 1992, 1995) and empirical (cf. Vainikka and Maling 1996) arguments in favor of regarding partitive Case as a structural Case. In fact, it seems clear that the content of the notion 'partitive Case' must relate to the referential properties of the noun phrase which carries it rather than to morphological case and even 9-roles: in Icelandic the postverbal argument of an expletive construction may very well carry a lexical/inherent/quirky morphological case. Consider the example in (29) from Sigurdsson 1989:288, where the associate carries dative. (29)
Pad hafdi kolnad premur malfraedingum. EXPL had cooled three.DAT linguists.DAT 'Three linguists had got colder'
It thus seems quite clear that the general idea we should pursue is that partitive Case represents a particular type of semantic property which is licensed in the postverbal position of unaccusatives, and that it opposes some other semantic property which is not licensed in that position. In that respect the theory developed by de Hoop (1992) seems appropriate: based on Milsark's (1974,1977) distinction between weak and strong noun phrases, she labels the semantic property licensed in the postverbal position of unaccusatives 'weak Case' and the property not licensed, 'strong Case.'
56
SUBJECTS, EXPLETIVES, AND THE EPP
Let us now make some assumptions regarding the checking of Case. First of all we acknowledge a complex VP-structure as in (30), where external arguments ('subjects') are base-generated in SpecvP and internal arguments ('(direct) objects') in the lower SpecVP. (30)
[VP DPSUBJ [v, v [VP DPOBJ [V]]]]
Given this structure, we take the difference between unaccusative and unergative predicates to be that the former assign a 0-role to the internal argument position, whereas the latter assign a 6-role to the external position. Similarly, regular transitive verbs assign their external, subjective 0-role to SpecvP and the internal, objective 9-role to SpecVP. These assumptions concerning 0-role assignment should be quite straightforward. Next we dispense with the idea that Case is checked in AGR projections, but instead assume that Case checking is done in local spec-head relations with the verb and tense. In particular we consider weak Case a default Case which can always be checked in SpecVP, and furthermore, for obvious reasons, we firmly acknowledge a distinction between morphological case and abstract Case (cf. above). Moreover, along with the first claim (repeated as (i)) we assume the following, where (ii) is intended to capture Burzio's Generalization:11 Checking of Case^f. (i) Weak Case can be checked in SpecVP against the verb in V°. (ii) Strong Case can be checked in SpecVP against the verb in V°, iff the verb assigns a 0-role to SpecvP. (iii) Case, both weak and strong, can be checked in SpecTP against [tense] in T°. (iv) No Case can be checked in SpecvP. (v) An X° can check only one Case. For the issues under consideration here this definition is sufficient—we are not concerned with indirect objects and prepositional complements. The effect of the assumptions about Case checking above is the following: (i) if the argument of an unaccusative verb carries strong Case (i.e., if it is definite, universally quantifying, partitive), it must raise to SpecTP, (ii) the argument of an unergative verb must always raise to SpecTP, (iii) an object of a transitive verb carrying strong Case can remain in situ in SpecVP.12 Notice that the assumptions regarding T° capture the basic intuition that subjective Case (in general nominative) is related to tense. Given that SpecTP is available in Icelandic expletive constructions but not in other languages such as English and Mainland Scandinavian, the fact that universally quantifying and partitive noun phrases are allowed in this position now follows. We are then left with the questions why the availability of SpecTP varies and why definites are not allowed in this position even in Icelandic expletive constructions. As for the availability of SpecTP, I will relate that to the classical insight that the Mainland Scandinavian expletive is a subject, whereas the Icelandic one is an empty topic (see Platzack 1983, Christensen 1991, Falk 1993). One way of captur-
ICELANDIC EXPLETIVE CONSTRUCTIONS
57
ing that difference is to say that the Mainland Scandinavian expletive carries Case, whereas the Icelandic one does not. Accordingly, we take it that the Mainland Scandinavian expletive must check its Case, and the only position where that can be done is in SpecTP. On that account SpecTP is never available for the associate in Mainland Scandinavian. Why the expletive differs in this way is a question with no obvious answer, but a tentative answer will be given in section 5.3 along with an account of why definites are not allowed in SpecTP in Icelandic expletive constructions.
5. The content of (Agr)SP 5.1. Agreement The purpose of this section will be to outline a theory which enables us to capture the facts presented in this chapter while at the same time of course being suitable to account for other phenomena. The account of the Definiteness Effect in the preceding section has paved the ground for eliminating the idea that the associate raises to the position of the expletive at the level of Logical Form. However, the main empirical reason (apart from the conceptual ones in Chomsky 1995) for assuming that such LF-raising of the associate takes place is the fact that the verb agrees with the associate in number. Consider the English examples in (31) and the corresponding Icelandic ones in (32). (31) a. b.
There are/*is linguists in the room. There seem/*seems to be linguists in the room.
(32) a.
Pad EXPL Pad EXPL
b.
eru/*er malfraedingar i herberginu. are/ is linguists in room.the virdast/*vir5ist vera malfraedingar i herberginu. seem I seems be linguists in room.the
The particularly interesting cases are the ones in the (b) examples, where there is agreement between the matrix raising verb and the associate in the embedded infinitival. In a system where the associate would raise to an agreement-checking position at LF, these facts receive a straightforward account, but if we are to dispense with LF-raising of the associate, we need an alternative account. The alternative account is the following. Agreement is a lexically defined matter. Constituents share morphological features if they are lexically defined to do so and if either of the following structural relations hold between them: (i) they are overt heads in the same local binding domain, (ii) they are specifiers of a head which is either specified for the features in question or shares them with another head. As argued in Vangsnes 1999:109f., such an approach to agreement is needed in order to account for noun phrase internal agreement where features may be shared among a relatively high number of constituents. Take, for instance, the following Icelandic example from Sigurdsson 1993:181, where agreement features 'spread' from the feminine plural noun to five adnominal constituents.
58 (33)
SUBJECTS, EXPLETIVES, AND THE EPP allar hinar brjar nyju kenningar binar all.F.PL the.F.PL three.F new.PL theories your.F.PL
Returning to the expletive constructions in (31) and (32), I will assume that an agreement relation is established between the associate and the verb heading V°. If this verb is the finite verb, it will exhibit an overt correlate of the agreement. If, on the other hand, the verb is a nonfinite verb, it will not show any overt correlate, but it will nevertheless agree, and in turn the finite verb, occupying a higher X° position in the clause, will agree with the nonfinite verb as long as no finite C° intervenes.13 Under this approach to agreement we need not assume that the associate moves to the position of the expletive at LF in order to account for why the associate agrees with the verb. Moreover, we do not need to assume any Agr-projections solely for the purpose of agreement checking. Let us next consider certain other issues pertaining to the functional domain AgrSP. 5.2. The Extended Projection Principle In Chomsky 1995 the Extended Projection Principle (EPP) is related to checking of a 'categorial D-feature' in IP, and more specifically, given a split-IP, in AgrSP, the functional domain where the subject-related (^-features are checked. The D-feature is by hypothesis a noninterpretable nominal feature (on a par with ^-features) and must be checked off before LF. This follows from the principle of Full Interpretation, which requires that only interpretable features be present at LF, the interface level with the conceptual-intentional component of the brain. Checking of the D-feature can be met either by raising a referential subject to SpecAgrSP or by merging an expletive in that position.14 Languages differ as to the strength of the D-feature: in languages with (overt) expletives, the D-feature is strong and must therefore be checked before Spell-out, whereas in languages without (overt) expletives the D-feature is weak and can therefore be checked covertly (possibly by a covert 'null' expletive). The presence versus absence of expletives is thus related to the strength of the D-feature. However, A&A (i.e., Alexiadou and Anagnastopoulou 1998) argue that the Dfeature in AgrSP is universally strong (i.e., that it must be checked prior to Spellout in all languages). They suggest that in addition to moving/merging an XP in SpecAgrSP, checking of the D-feature in AgrSP can also be accomplished by head movement of the verb to AgrS°. The latter checking is, however, possible only in null-subject (pro-drop) languages, on their account, because the finite verb in such languages carries nominal features on a par with pronouns. The presence of nominal features on the verb is in turn the standard account of why these languages allow pro drop: the presence of pronouns is superfluous since their denotation is expressed by the verb. As A&A acknowledge, the theory they develop requires that a distinction be drawn between overt subject/verb agreement and the presence of the nominal features which are suitable for checking the D-feature in AgrSP. Icelandic, for example, has overt subject/verb agreement, but nevertheless also has expletives, indi-
ICELANDIC EXPLETIVE CONSTRUCTIONS
59
eating that EPP-checking cannot be achieved by head movement of the verb to AgrS°. On the other hand, in a series of works, starting with Holmberg and Platzack 1988 and culminating with Holmberg and Platzack 1995, Anders Holmberg and Christer Platzack have worked toward relating several of the differences between Icelandic (and to some extent Faroese) and Mainland Scandinavian (e.g. [±V-to-Imovement], [± oblique subjects] etc.) to a single parameter which has its basis in the simple fact that Icelandic has rich verbal inflection with both tense and person affixes, whereas Mainland Scandinavian (with a few exceptional dialects) has only tense inflection. The well-known difference between Icelandic and Mainland Scandinavian V-to-I-movement is illustrated by the examples in (34) and (35), Norwegian representing Mainland Scandinavian in general. (34) a.
ad Jon las aldrei bokina. that Jon read never booLthe b. * ad Jon aldrei las bokina.
Icelandic
(35) a.
Norwegian
at Jon aldri las boka that Jon read never book.the b. * at Jon las aldri boka
Holmberg and Platzack's approach appears highly successful, indicating that overt subject/verb agreement does play an important syntactic role in Icelandic, but this then is potentially in conflict with the theory developed by A&A (and in fact with any theory which attempts to relate the presence/absence of expletives to the nullsubject parameter). However, it seems that the insights of both theories can be brought together in a highly fruitful way if we exploit the idea in A&A that subject/verb agreement either involves a pronominal, [+interpretable] feature, or not. In other words, subject/verb agreement may either be a mere morphological phenomenon, the requirement that two (or more) constituents share formal features, or it may also have a semantic impact in that the feature shared has a semantic content. That would allow us to conjecture that there are two parameters related to subject/verb agreement: the null-subject parameter and the AGR parameter. In the next subsection I will capitalize on the distinction between two types of agreement and outline a theory which captures both the syntactic relevance of the overt subject/verb agreement in Icelandic and the fact that expletives are present in the language. The difference between Icelandic and other languages with respect to the availability of SpecTP in expletive constructions will also be accounted for. 5.3. Checking versus identification In Chomsky 1995 AGR is related to the checking of uninterpretable features such as Case, the assumed D-feature, and in particular agreement features. AGR is not relevant for the interpretation of a sentence—it serves only as the locus where uninterpretable features are checked and deleted before LF. As a consequence, in Chomsky 1998, where checking of (person) agreement is achieved through the operation AGREE, which makes no reference to AGR projections, AGR projections
60
SUBJECTS, EXPLETIVES, AND THE EPP
are dispensed with altogether, and the functional domain of the clause is reduced correspondingly (cf. note 14). However, let us consider the possibility that there is in fact a functional category between T and C which does have semantic relevance, specifically that of anchoring the state-of-affairs with respect to the participants in the act, in particular the relation between the state-of-affairs and the most prominent participant: the subject. This functional category can then be identified with AgrSP, but we may label it '(Agr)SP' since we will tone down the role played by agreement as inducing purely formal mechanisms, and the existence of the functional domain will instead be motivated on semantic grounds. The function of TP is to anchor the state-of-affairs denoted by the verb with respect to time, and (Agr)S and T are then 'extensional' functional categories: both are relevant for determining the extensional interpretation of the clause. Correspondingly, the VP-domain will be the 'intensional' domain, the domain where the denotative properties of the clause are determined. Following Vangsnes 1999, where a similar referential split into an extensional and an intensional domain is argued for in the case of the noun phrase, we assume that the presence of T is necessary for the clause to receive a temporal interpretation (of its own), and correspondingly the presence of (Agr)S is necessary in order to anchor the state-ofaffairs with respect to the subject (i.e., the argument carrying the 0-role highest on the 6-hierarchy—cf. Jackendoff 1972—hence the most prominent participant). CP differs from (Agr)SP and TP in being a 'categorial' rather than an extensional functional projection: it defines a syntactic object as a 'clause.' Moreover, also following Vangsnes 1999, we assume that functional projections are headed by abstract heads—CP, (Agr)SP, and TP by K, o, and T, respectively— and we assume that these abstract heads must be identified, stated as the 'Identification Principle': The Identification Principle A functional category must be identified. Proper identification is defined as follows, where 'extended projection' is understood in the sense of Grimshaw (1991). Identification fef. A functional category F in an extended projection P is identified iff a constituent containing at least one feature relevant for F is merged in the head or specifier position of F. Furthermore, we acknowledge two types of morphological features, 'lexical' features and 'agreement' features, and importantly we assume the following preference principle: Preferred identifier'^ef. When there are several candidate constituents for identifying a functional category F, the preferred identifier for F will be: (i) the constituent containing the largest number of agreement features relevant for F; or else, if there is no such constituent,
ICELANDIC EXPLETIVE CONSTRUCTIONS
61
(ii)
the constituent containing the largest number of lexical features relevant for F; or else, if there is no such constituent, (iii) the constituent containing the smallest number of irrelevant lexical features; or else, if there is no such constituent, (iv) anX°. Issues where this preference principle plays an important role are discussed in some detail in Vangsnes 1999. What is important to notice here is that the principle implies that a potential identifier carrying one relevant agreement feature more than some other identifier will always be the preferred one regardless of how many lexical features the other carries. Moreover, the preference for an X° identifier over an XP identifier is argued for by A&A (519-520) with reference to a notion of 'Economy of Projection,' the basic idea being that when head movement takes place, projection of the full phrase need not apply. (See also Nunes 1998.) As for the features relevant for the identification of a and t, we assume the following: (36)
a must be identified by an element containing at least one of the following features: [deixis], [Case], [person].
(37)
T must be identified by an element containing at least one of the following features: [tense], [number].
As for the distinction between lexical features and agreement features, we take [deixis], [Case], and [tense] to be lexical features, and [person] and [number] to be agreement features. Taking [person] to be an agreement feature implies that this feature is not present on pronouns—in fact, we will assume that the person specification of pronouns is a value of the lexical feature [deixis], which is a feature carried by pronouns and definites. Relating this to the discussion of subject/verb agreement in the previous subsection, the kind of agreement which licenses pro drop (i.e., in null-subject languages) involves both an agreement feature and a lexical feature (the pronominal [+interpretable] feature), whereas subject/verb agreement in languages such as Icelandic involves only an agreement feature. The lexical feature in question we take to be [deixis], and the agreement feature, [person]. This then means that we can represent the difference between null-subject languages, Icelandic, and Mainland Scandinavian as in (38)—the latter has neither of the two features relevant for the identification of o. (38) a. b. c.
Vfin[deixis, person, tense] Vfm[person, tense] Vfin [tense]
Null-subject languages Icelandic Mainland Scandinavian
As for nonfinite verbs, the feature [tense] is not present in any of the three language types.15 Let us now assume that in addition to the identification requirement on a, there is a requirement that a referential expression for the subject must be represented in the (Agr)SP domain in order for the denotations of the verb and the subject to be successfully anchored to the same state-of-affairs—in effect this then represents our understanding of the Extended Projection Principle.
62
SUBJECTS, EXPLETIVES, AND THE EPP
However, in its strictest form we take this requirement to apply only to definite subjects which by hypothesis carry the lexical feature [deixis], the intuition behind this being that only uniquely identifiable referents need such an explicit anchoring. In the case of nondefinite subjects we assume that only a default, nonspecific anchoring is required, and this is achieved as long as some relevant lexical feature is present in either the specifier or head position, in other words, if either of the features [deixis] and [Case] (or both) is there. The system now works in the following way. In null-subject languages o is identified by raising of the verb to (Agr)S°—the verb carries two features relevant for the identification, an agreement feature ([person]) and a lexical feature ([deixis]). Since the verb in these languages carries the lexical feature [deixis], there is no need for an overt XP in Spec(Agr)SP—the feature will contain the same information as pronouns and, furthermore, provide a default interpretation in other cases. In Icelandic o is identified by head movement of the verb to (Agr)S° just as in null-subject languages, since the verb carries an agreement feature relevant for the identification ([person]). However, since the verb does not carry the feature [deixis], an overt XP must occur in Spec(Agr)SP—the verb cannot provide any anchoring for the subject, neither a specific nor a default, nonspecific one. Accordingly, pronominal subjects must be overt, and moreover, in the case of nondefinite subjects an element containing either of the features [deixis] or [Case] must occur in Spec(Agr)SP. A definite subject will necessarily move to Spec(Agr)SP. Moreover, a nondefinite subject will meet the requirement for a successful anchoring if it moves to Spec(Agr)SP as it will carry the feature [Case]. Finally, merging of an expletive in the position will also accommodate the requirement on the assumption that it contains the feature [deixis]. And the last assumption seems quite plausible. The etymological origins of expletives are typically either pronouns (it, pad) or deictic adverbs (there), and since pronouns by hypothesis carry the feature [deixis] and deictic adverbs arguably do so, too, the conjecture that expletives carry this feature appears unproblematic. In fact, the present theory offers an explicit explanation for why exactly pronouns and deictic adverbials serve as the 'etymological pool' for expletives. Turning now to Mainland Scandinavian, verb raising does not suffice to identify a since the verb does not carry any relevant feature. Unlike the situation in Icelandic, the verb does not raise to (Agr)S° in Mainland Scandinavian (in embedded clauses and infinitivals—in main clauses the verb raises to C° due to the V2 effect, which will not be discussed here). This captures the general insight of Holmberg and Platzack (1995) with respect to verb raising. Now, since the verb cannot identify a in Mainland Scandinavian, the abstract head must be identified by an XP carrying either of the lexical features [deixis] or [Case]. A subject noun phrase raised from a lower position will suffice since it will always be Case-marked, and if it is definite it will carry the feature [deixis], too. Moreover, an expletive in Spec(Agr)SP would also suffice since it presumably carries the feature [deixis] as well as the feature [Case] (cf. section 4.2). Pro drop is obviously not allowed. This follows for reasons pertaining both to the identification of o and to the anchoring of the subject, and not just for the latter reason, as in Icelandic.
ICELANDIC EXPLETIVE CONSTRUCTIONS
63
Table 3.2: Subject anchoring and identification of <J across languages NSL Subject anchoring Identification of a
V° V°
Icelandic Mainland Sc. XP
v°
XP XP
The differences between null-subject languages, Icelandic, and Mainland Scandinavian can now be summarized as in Table 3.2.16 Returning to the question why the Mainland Scandinavian expletive carries Case whereas the Icelandic one does not (cf. section 4.2), the key to understanding this difference could lie in the different ways in which identification of a is achieved in the two language types. In Mainland Scandinavian nonexpletive constructions the identifier will always be a noun phrase which carries Case, and identification by the feature [Case] is therefore possibly recognized as a 'generalized strategy' for identification of G. That is, from an acquisitional point of view, since what the identifiers of a in nonexpletive constructions have in common is that they all carry [Case], definites and indefinites alike, the child infers from this that the identifier of o in expletive constructions (i.e., the expletive) does, too. In Icelandic the identifier is always the verb, so the feature [Case] is never involved in the identification and there is no evidence for inferring the presence of the feature [Case] from its presence on subjects of nonexpletive constructions. The recognition of the verb as the generalized identifier, on the other hand, would serve to explain why there is verb raising in control infinitives (on the assumption that (Agr)SP is present and TP is not) (cf. Vangsnes 1999: 86ff., 2000). Moreover, returning to the view stated in section 3.1 that the expletive is never base-generated in SpecCP in Icelandic, we may argue that other constituent types than expletives and noun phrases may provide a default subject anchoring. Consider, for example, the sentences in (8d) and (8e) repeated here. (8)
d. e.
I dag hafa (*pad) vend *allir kettirnir/*badir kettirnir i eldhusinu. today have EXPLbeen all cats, the/ both cats.the in kitchen, the I dag hafa (*bad) veriQ nokkrir kettir/*sumir kettir i today have EXPL been some cats I some. of. the cats in eldhusinu. kitchen, the
These sentences exhibit the Definiteness Effect although no overt expletive is present. If we assume that the fronted temporal adverbial moves via Spec(Agr)SP and thereby provides a default subject anchoring, and, moreover, that this applies to all cases where constituents are fronted, there is no reason to assume the existence of a covert expletive in examples like these. On the additional assumption that SpecCP must always be filled in main clauses, it will follow that the expletive always occurs sentence initially, that is, that it always moves from Spec(Agr)SP to SpecCP.17 The notion of 'generalized strategy' obviously deserves some further discussion, but that would lead too far here. In Vangsnes 1999 the relevance of the notion is
64
SUBJECTS, EXPLETIVES, AND THE EPP
discussed in the light of several issues pertaining to cross-dialectal differences in the Scandinavian noun phrase. With reference to the present discussion, an interesting point is that the present line of reasoning predicts that in the history of Mainland Scandinavian, Case-marked expletives should evolve after the verb is no longer recognized as the identifier of o. That appears to be compatible with the findings of Falk (1993) for Swedish.18 All in all, the theory outlined here may appear somewhat unorthodox and incompatible with established beliefs, say, within Minimalist syntax. The incompatibility is only slight, however. The notion of 'identification' has a clear parallel in the Minimalist notion of 'checking' as conceived of in Chomsky 1995. It is a central trigger for movement, and it involves morphological features. However, the morphological features involved in checking are highly stipulative and bear only a faint relation to visible morphology—if a language exhibits a particular kind of movement, a feature is involved regardless of whether there is an overt correlate of the feature. On the other hand, a language may very well possess overt correlates of a feature which triggers movement although the movement in question is not visible in overt syntax. Although there seems to be considerable evidence in favor of making a distinction between overt morphology and syntactically relevant morphology, one runs the risk of reducing the explanatory force of the theory considerably. In that respect it seems that if the notion of 'checking' were to be replaced by 'identification,' that risk is reduced. By making a distinction between agreement features and lexical features, it appears easier to evaluate the syntactic role of morphology from both a synchronic and a diachronic point of view. A full discussion of these matters is unfortunately beyond the scope of this chapter.
6. Conclusion Empirical evidence from Icelandic tells us that the expletive as such cannot be the trigger of the Definiteness Effect, and that in turn suggests that the idea of expletive-associate replacement at LF is on the wrong track. The present chapter has outlined a syntactic theory which can do without these ideas. The Definiteness Effect is accounted for by exploiting the insights of the Case theories of Belletti (1988) and de Hoop (1992), where the presence of the expletive as such is irrelevant, agreement is accounted for in a way which renders covert raising of the associate to the position of the expletive unnecessary, and the presence versus absence of expletives, as well as their differing status, across languages is accounted for by attributing to (Agr)SP a semantic function which in turn requires that certain formal conditions are met. Although the theory outlined is speculative to some extent, it is clear that it can account for the Icelandic facts which the chapter set out to explain while at the same time keeping related facts about other languages in sight. Whether the basic ideas underlying the theory prove successful for a wider range of phenomena remains to be seen, but there are no a priori reasons that they will not.
ICELANDIC EXPLETIVE CONSTRUCTIONS
65
Notes For judgments and discussion of the Icelandic data presented here I am indebted to Porhallur Eyborsson, Maria Anna Gardarsdottir, Gunnar Olafur Hansson, Petur Helgason, Porbjorg Hroarsdottir, Porsteinn G. Indridason, Kjartan G. Ottosson, Eirikur Rognvaldsson, and Halldor Armann Sigurdsson. Moreover, I thank Kirsti Koch Christensen, Anders Holmberg, Natalia Sanchez-Lefebvre, and Corel Sandstrom for useful comments. Peter Svenonius deserves special thanks for scrutinizing an earlier version of the chapter, and I am grateful to an anonymous reviewer for critical remarks which led to considerable improvement. I of course take full responsibility for remaining errors and inadequacies. 1. In the recent literature, in particular Minimalist syntax, it has been customary to use the term 'associate' to refer to the referential noun phrase which earlier had been referred to as 'logical subject' (i.e., a man in a sentence like There is a man in the garden). Although this terminology reflects a particular theoretical idea that I will argue against, roughly that the noun phrase in question is 'associated' with the expletive, I will adhere to the new established terminology. 2. There are exceptions to this. In Swedish the associate may occur between the finite and the nonfinite verbs in passive constructions formed with the auxiliary bli (cf. (ib)). (i)
a.
Det blev skjutet nagra algar Swedish EXPL was shot some moose. PL b. Det blev nagra algar skjutna EXPL was some moose. PL shot. PL Notice that the participle agrees with the associate when the latter appears to the left of it (as in (ib)), but not when it appears to the right (as in (ia)). This is an interesting fact. See note 13 for further discussion. Moreover, expletive sentences with a quantified NP in the intermediate position have been reported in traditional descriptions of all the Mainland Scandinavian languages (cf. Christensen 1991:148-156 for references), and although such sentences are nowadays generally considered archaic and/or stylistically highly marked, this is of course an interesting fact from a diachronic point of view—in fact, possibly also from a dialectal perspective. The following impersonal passive, uttered by a speaker of a western Norwegian dialect, was heard by the author on the radio in March 2000. (ii)
No har det vore mange b0ker skrivne om det emnet. Western Norwegian now has EXPL been many books written.PL about that subject (See note 18 for further discussion.) 3. The example Bobaljik and Jonas (1996:213) provide is the following (their (22a)). (i)
Pad lauk einhver verkefninu alveg. there finished someone assignment.the completely 'Someone completely finished the assignment.' However, several speakers of Icelandic consider this sentence highly deviant, and I have therefore chosen not to copy the example. Still, Bobaljik and Jonas's point is probably valid, given that the sentences I provide are fairly acceptable. 4. There are a number of apparent counterexamples to the Defmiteness Effect in both Scandinavian and English. For discussion and accounts of several of such alleged counterexamples, see Vangsnes 1994, 1999:74ff. and references cited there. Icelandic also allows certain instances of grammatically definite noun phrases in expletive constructions (cf. Rognvaldsson 1984: 365, Sigurdsson 1989:294ff.), such as the following example taken from Rognvaldsson 1984: 365: (i)
Pad skin alltaf blessud solin. EXPL shines always blessed sun.the 'There shines always the blessed sun' Examples like this have been taken as evidence for a pragmatic treatment of the Defmiteness Effect, but in my opinion they can be accounted for in much the same way as the cases discussed in Vangsnes 1994, and I will not discuss them further here. The same goes for a particular kind of intensifying noun phrases involving the proximal demonstrative jxssi, noted in Vangsnes 1995, 1999:75 and due to Gunnar Olafur Hansson (p.c.), which do not involve unique reference, and whose occurrence as associates is therefore not surprising.
66
SUBJECTS, EXPLETIVES, AND THE EPP (ii)
Pad var komin pessi Ifka gullfallega stelpa i partyid. EXPL was come this PRT gold.beautiful girl inparty.the 'This really good-looking girl had come to the party' These noun phrases are similar (apart from the intensifying effect) to indefinite-//!/.? noun phrases in English. 5. By 'definite descriptions' I understand proper names and noun phrases with the definite article, a demonstrative, and/or a possessor. 6. Anders Holmberg (p.c.) points out that bare NPs are also awkward in the VP-extemal position of Swedish expletive passives (cf. note 2). (i)
Det blev nagra a'lgar /??algar skjutna Swedish EXPL became some moose. PL/ moose. PL shot Similar judgments have been reported about VP-external bare indefinites in Finnish expletive constructions, cf. Holmberg and Nikanne 1994. 7. Cf. the 'Principle of Contrastiveness' of de Hoop (1992:166). 8. It is also worth noticing that Chomsky (1995:340ff.), who bases his discussion of Icelandic TECs on work by Bobaljik and Jonas, also contributes to spreading this incorrect information when he states about Icelandic TECs that "[t]he usual properties of expletive constructions hold: [Spec, T], the associate of the expletive in [Spec, AgrS], is nonspecific . .." (p. 341), and he furthermore (p. 342) makes a suggestion as to why associates in both VP and SpecTP are subject to the Defmiteness Effect. That suggestion is of course pointless, given the evidence presented here. 9. Bobaljik and Jonas (1996) also base their revision of the Mapping Hypothesis on the (valid) observation that definite descriptions of nonexpletive constructions precede a TP-adverbial. Cf. (i), which is Bobaljik and Jonas's (3a) and (30a). (i)
I gser klaruSu (pessar mys) sennilega (*bessar mys) ostinn. yesterday finished these mice probably these mice cheese, the 'These mice probably finished the cheese yesterday' They suggest that the reason why definites occur in a position preceding the adverbial (i.e., in SpecAgrSP) can be related to the Mapping Hypothesis in its revised version. 10. Taking Finnish as her starting point, Belletti (1988) shows that the postverbal argument of Finnish existentials can carry partitive morphological case and, moreover, that the alternation between accusative and partitive objects in certain cases coincides with the distinction between a definite and an indefinite reading, Finnish otherwise lacking both definite and indefinite articles. However, partitive as a morphological case is far from strictly related to indefiniteness in Finnish (cf. de Hoop 1992, Kiparsky 1998, Vainikka and Maling 1996, Vangsnes 1994), suggesting that there is no direct relation between partitive morphology and the abstract partitive Case that she proposes. 11. The classical GB formulation of Burzio's Generalization is as follows (cf. Van Riemsdijk and Williams 1986:250): "If a verb assigns case to its object, then it assigns a 6-role to its subject." 12. It is worth noticing that the proposed definition of Case checking also accounts for certain instances of the Defmiteness Effect in nonexpletive constructions in Norwegian, notably the fact that the postverbal argument of pseudo-passives is subject to the restriction. Consider the examples in (i): (ia) a regular passive, (ib) a corresponding impersonal passive, and (ic) a pseudo-passive where the raised noun phrase is the argument of a stranded preposition. (i)
a.
Rotne egg /dei rotne egga vart kasta pi politikaren. Norwegian rotten eggs/the rotten eggs.the was thrown on politician.the 'Rotten eggs/the rotten eggs were thrown on the politician' b. Det vart kasta rotne egg /*dei rotne egga pa politikaren. EXPL was thrown rotten eggs/ the rotten eggs.the on politician.the c. Politikarenj vart kasta rotne egg/*dei rotne egga pa fj. politician.the was thrown rotten eggs/the rotten eggs.the on We see that the postverbal argument is subject to the Defmiteness Effect in the pseudo-passive just as it is in the impersonal passive. This is accounted for since no 0-role is assigned to SpecvP in either case, the raised noun phrase in the pseudo-passive having received its 6-role from the stranded preposition. See Vangsnes 1994:46ff. for discussion of this and other instances of the Defmiteness Effect in nonex-
ICELANDIC EXPLETIVE CONSTRUCTIONS
67
pletive contexts, notably on the direct objects of certain double object constructions in Norwegian. The latter cases presumably can also be accounted for within the present approach if Burzio's Generalization is formulated so as to apply to the relation holding between the indirect and the direct object—in the cases in question, the verb arguably does not assign any 0-role to the indirect objects. 13. It is not immediately clear how the participle agreement in Swedish expletive passives can be accounted for under this approach. As shown in note 2, the participle will agree in number with an. associate which precedes it but not with an associate which follows it. In Icelandic, on the other hand, there will be number agreement in both cases. Cf. the following examples. (i)
a.
Pad voru skotnir /*skoti5 nokkrir elgir. EXPL were shot.PL/ shot.PL some moose.PL b. Pad voru nokkrir elgir skotnir/*skotid. The reason may simply be that agreement is lexically defined in a different way in Swedish than in Icelandic, perhaps so that number agreement can be established between the verb and a noun phrase only in the external subject position. The issue deserves further investigation and speculation, but I will leave that to future research. 14. In Chomsky 1998 the functional domain AgrSP is dispensed with along with other Agr-projections as checking of agreement no longer is argued to take place in designated functional projections. The EPP is now attributed to the presence of an EPP-feature in T°, corresponding to the D-feature in the earlier account, but the basic machinery remains the same: the checking of the feature is achieved either by raising of a subject to SpecTP or by merging an expletive in the position. 15. An important deviation from standard assumptions, discussed in Vangsnes 1999:86ff. and Vangsnes 2000, is that T° is never present in infmitivals. (Agr)SP, on the other hand, is projected in for instance control infinitivals, and the difference between Icelandic and Mainland Scandinavian with respect to verb raising in such infinitivals is accounted for in a uniform way with verb raising in (embedded) finite clauses: the verb raises in Icelandic, and that follows from the AGR-parameter since the verb is a suitable identifier of a in Icelandic but not in Mainland Scandinavian. 16. This system gives us a fourth possible language type, one where subject anchoring is provided by the verb and identification of a by an XP. However, notice how the preference principle for identifiers interferes in that respect: the XP would have to carry either the same number of relevant agreement features and at least one more lexical feature, or at least one more agreement feature than the verb. Given that the verb at least would carry the feature [deix] (since it accommodates subject anchoring), we expect such languages to be rare, if existent at all. 17. This line of reasoning appears to be compatible with Holmberg's (2000) discussion of Stylistic Fronting (SF) in Icelandic. In SF a sentence adverbial, participle, or particle is fronted to first position, there is always a subject gap (i.e., no argument noun phrase in the intermediate position), and, importantly, a Definiteness Effect can be observed on the postverbal argument (if there is one). Cf. (ia) and compare with the expletive construction in (ib). (i)
a.
Tekin hefur verid _ erfid akv6rdun/*erfida akvordunin. taken has been difficult decision I difficult decision.the 'A difficult decision/The difficult decision has been made' b. Pad hefur verid tekin erfid akvordun/*erfida akvordunin. EXPL has been taken difficult decision I difficult decision.the Holmberg argues that the fronted constituent in SF functions as an expletive, and although his theoretical interpretations differ somewhat from the present ones, that is in effect what I argue for the fronted adverbial in the examples in (8). 18. On the other hand, it is not entirely inconceivable that non-Case-marked expletives would exist after the verb no longer is the identifier of a, say in a transition period before identification by the feature [Case] is recognized as the generalized strategy. That could possibly serve to explain the remnant cases noted by Christensen (1991) where Mainland Scandinavian expletive constructions have the associate in the intermediate position (cf. note 2).
68
SUBJECTS, EXPLETIVES, AND THE EPP
References Afarli, Tor A. 1992. The Syntax of Norwegian Passive Constructions. John Benjamins, Amsterdam. Alexiadou, Artemis, & Elena Anagnostopoulou. 1998. Parametrizing AGR: Word order, Vmovement and EPP-checking. Natural Language & Linguistic Theory 16,491-539. Belletti, Adriana. 1988. The Case of unaccusatives. Linguistic Inquiry 19, 1-34. Bobaljik, Jonathan D., & Dianne Jonas. 1996. Subject positions and the roles of TP. Linguistic Inquiry 27,195-236. Chomsky, Noam. 1986. Knowledge of Language: Its Nature, Origin, and Use. Praeger, New York. Chomsky, Noam. 1995. The Minimalist Program. MIT Press, Cambridge, Mass. Chomsky, Noam. 1998. Minimalist inquiries: The framework. MIT Occasional Papers in Linguistics 15, 1-56. Christensen, Kirsti Koch. 1991. AGR, adjunction, and the structure of Scandinavian existential sentences. Lingua 84, 137-158. Collins, Chris, & Hoskuldur Thr&nsson. 1993. Object Shift in double object constructions and the theory of Case. MIT Working Papers in Linguistics 19,131-174. Diesing, Molly. 1992. Indefinites. MIT Press, Cambridge, Mass. Enc, Miirvet. 1991. The semantics of specificity. Linguistic Inquiry 22,1-25. Falk, Cecilia. 1993. Non-referential Subjects in the History of Swedish. Ph.D. dissertation, University of Lund. Grimshaw, Jane. 1991. Extended projections. Ms., Brandeis University. Holmberg, Anders. 1993. Two subject positions in IP in Mainland Scandinavian. Working Papers in Scandinavian Syntax 52, 29-41. Holmberg, Anders. 1994. The pros and cons of agreement in Scandinavian impersonals. In Guglielmo Cinque et al. (eds.), Paths Towards Universal Grammar: Studies in Honor of Richard S. Kayne, 217-236. Georgetown University Press, Washington D.C. Holmberg, Anders. 2000. Scandinavian Stylistic Fronting: How any category can become an expletive. Linguistic Inquiry 31, 445-483. Holmberg, Anders, & Urpo Nikanne. 1994. Expletives and subject positions in Finnish. In M. Gonzalez (ed.), Proceedings of the North East Linguistics Society 24, 173-187. GLSA, Amherst, Mass. Holmberg, Anders, & Christer Platzack. 1988. On the role of inflection in Scandinavian syntax. Working Papers in Scandinavian Syntax 42, 25-42. Holmberg, Anders, & Christer Platzack. 1995. The Role of Inflection in Scandinavian Syntax. Oxford University Press, New York. Hoop, Helen de. 1992. Case Configuration and Noun Phrase Interpretation. Ph.D. dissertation, University of Groningen. [Published 1996. Garland, New York.] Jackendoff, Ray S. 1972. Semantic Interpretation in Generative Grammar. MIT Press, Cambridge, Mass. Jonas, Dianne. 1994. The TP parameter in Scandinavian syntax. In Cecilia Hedlund & Anders Holmberg (eds.), Proceedings of the XlVth Scandinavian Conference of Linguistics and the VIHth Conference of Nordic and General Linguistics, Special Session on Scandinavian Syntax, 33-59. Gothenburg Papers in Theoretical Linguistics, Gothenburg University. Jonas, Dianne, & Jonathan D. Bobaljik. 1993. Specs for subjects: The role of TP in +Icelandic. MIT Working Papers in Linguistics 16, 99-148.
ICELANDIC EXPLETIVE CONSTRUCTIONS
69
Kiparsky, Paul. 1998. Partitive case and aspect. In Miriam Butt & Wilhelm Geuder (eds.), The Projection of Arguments, 265-307. CSLI, Stanford, Calif. Koeneman, Olaf. 2000. The Flexible Nature of Verb Movement. Ph.D. dissertation, University of Utrecht. Koeneman, Olaf, & Ad Neeleman. 1997. Transitive expletive constructions in Flexible Syntax. In Kimary N. Shahin, Susan Blake, and Eun-Sook Kim (eds.), Proceedings ofWCCFL 17. CSLI, Stanford, Calif. Lasnik, Howard. 1992. Case and expletives: Notes toward a parametric account. Linguistic Inquiry 23, 381-405. Lasnik, Howard. 1995. Case and expletives revisited: On Greed and other human failings. Linguistic Inquiry 26, 615-633. Milsark, Gary L. 1974. Existential Sentences in English. Ph.D. dissertation, MIT. Published in 1979. Garland, New York. Milsark, Gary L. 1977. Toward an explanation of certain peculiarities of the existential construction in English. Linguistic Analysis 3, 1-29. Nunes, Jairo. 1998. Bare X-bar theory and structures formed by movement. Linguistic Inquiry 29, 160-168. Platzack, Christen 1983. Existential sentences in English, German, Icelandic, and Swedish. In Fred Karlsson (ed.), Papers from the 7th Scandinavian Conference of Linguistics, 80-100. Department of General Linguistics, University of Helsinki. Riemsdijk, Henk van, & Edwin Williams. 1986. Introduction to the Theory of Grammar. MIT Press, Cambridge, Mass. Rognvaldsson, Eirikur. 1984. Rightward displacements of NPs in Icelandic. In K. Ringgaard & V. S0rensen (eds.), The Nordic Languages and Modern Linguistics 5, 362-368. University of Aarhus. Safir, Kenneth J. 1985. Syntactic Chains. Cambridge University Press, Cambridge. Safir, Kenneth J. 1987. What explains the Definiteness Effect? In Eric Reuland & Alice ter Meulen (eds.), The Representation of(In)definiteness, 71-97. MIT Press, Cambridge, Mass. SigurSsson, Halldor Armann. 1989. Verbal Syntax and Case in Icelandic: In a Comparative GB Approach. Ph.D. dissertation, University of Lund. Sigurflsson, Halld6r Armann. 1990-1991. Beygingarsamraemi. Islenskt mdl og almenn mdlfrcedi 12-13, 31-77. Sigurdsson, Halldor Armann. 1993. The structure of the Icelandic NP. Studia Linguistica 47, 177-197. Vainikka, Anne, & Joan Maling. 1996. Is partitive case inherent or structural? In Jacob Hoeksema (ed.), Partitives: Studies on the Syntax and Semantics of Partitive and Related Constructions, 179-208. Mouton de Gruyter, Berlin. Vangsnes, 0ystein Alexander. 1994. The Syntactic Source and Semantic Reflexes of the Definiteness Effect. M.A. thesis, University of Bergen. [Published as Skriftserie fra Institutt for fonetikk og lingvistikk 48. Department of Linguistics and Phonetics, University of Bergen.] Vangsnes, 0ystein Alexander. 1995. Referentiality and argument positions in Icelandic. Working Papers in Scandinavian Syntax 55, 89-109. Vangsnes, 0ystein Alexander. 1999. The Identification of Functional Architecture. Ph.D. dissertation, University of Bergen.
70
SUBJECTS, EXPLETIVES, AND THE EPP
Vangsnes, 0ystein Alexander. 2000. Scandinavian infinitivals and the AGR parameter. Paper presented at the 15th Comparative Germanic Syntax Workshop, University of Groningen. Vikner, Sten. 1995. Verb Movement and Expletive Subjects in the Germanic Languages. Oxford University Press, New York.
4 Expletives, Subjects, and Topics in Finnish Anders Holmberg and Urpo Nikanne
1. Introduction Finnish exhibits an unusual combination of properties relating to 'the subject': (a) it is a null-subject language, (b) it has obligatory expletives, (c) it is topic-prominent in the sense that the external argument need not be the subject but can be any category which can serve as the topic of the sentence. The Finnish facts to be discussed here therefore have consequences for the formulation of parameters concerning null subjects and the EPP, among other things. For instance, according to the standard view of null subjects, following Rizzi 1982, 1986, null-subject languages do not have overt expletives, since any language which can license empty referential subjects will be capable of licensing empty nonreferential subjects. More recently Alexiadou and Anagnostopoulou (1998) have argued that null-subject languages satisfy the EPP just by V-movement to AgrS, and therefore have no need for expletives, overt or covert. Finnish shows that neither theory can be entirely correct. As shown by (1), Finnish can license a referential or a 'quasi-referential' null subject, but not a nonreferential one. (1)
a.
Olen vasynyt. be.lsc tired Tm tired' b. Sataa (vetta). rains water c. * Leikkii lapsia kadulla. play children in. street
Sentences (2a-c) are well-formed variants of (Ic) (EXPL = expletive pronoun): (2)
a.
Sita leikkii lapsia kadulla. EXPL play children in. street
71
72
SUBJECTS, EXPLETIVES, AND THE EPP b. c.
Kadulla leikkii lapsia. in. street play children Lapsia leikkii kadulla. children play in. street 'Children are playing in the street'/There are children playing in the street'
According to E. Kiss (1995, 1997) a language is topic-prominent if syntactic structure reflects topic-comment structure so that an argument is externalized if and only if it is the sentence topic. In such languages externalization of an argument (movement to a predicate-external A-position) is not required for reasons of Case or a formal EPP, but is, in a sense, semantically triggered. Consequently we do not expect to find obligatory expletive pronouns in such languages. Again Finnish shows that this cannot be the whole truth. Multiple Subject Constructions (MSCs) have attracted some attention recently, especially following Bobaljik and Jonas 1996 and Chomsky 1995. An MSC is a construction which has more than one grammatical subject realized in VP-external positions. Finnish has two types of MSCs. One, exemplified in (3a), has an expletive pronoun in a higher position and a lexical argument DP in a lower position, where both positions are VP-external. The other, exemplified in (3b), has a seemingly referential subject pronoun in the higher position, coupled with a lexical argument DP in the lower, VP-external position. In this chapter we shall deal with the first type only. (3)
a.
b.
Sita ovat nama lapset jo oppineet uimaan. EXPL have these children already learned to. swim These children have already learned to swim' Ne ovat ministerit ostaneet uusia autoja. they have ministers bought new cars 'The ministers have bought new cars'
The existence of MSCs has consequences for the architecture of the sentence. How many sentential functional heads do we need to assume, and how many spec positions? What is the status of Agr? These questions have gained new interest in the wake of Chomsky 1995: ch. 4, where the status of Agr as a head is called into question. Chomsky argues that the two specifiers of MSCs are specifiers of the same head, namely T. We will show that this is not the case in Finnish. Finnish exhibits strong evidence that the higher of the two specifiers in the MSCs crucially occupies the spec of the sentential head bearing the subject agreement features, that is, AgrS in traditional terminology, but not necessarily tense or any other inflectional features. The lower specifier, on the other hand, will be shown to be in the complement of AgrS, inside what we take to be the predicate phrase. We will show, however, that the spec of the head usually hosting the subject agreement inflection, somewhat surprisingly, is not the privileged position of the subject in Finnish, but of the sentence topic, which may be, for instance, an object. Even 'MSC' in fact is a misnomer, since the argument which is 'multiplied' need not be the subject, but may be an object. All non-English sentences in this paper are Finnish, unless indicated otherwise.
EXPLETIVES, SUBJECTS, AND TOPICS IN FINNISH
73
2. The finite clause in Finnish The diagram in (4) shows the structure of the Finnish finite clause, according to the theory of Holmberg et al. 1993. (4)
The finite clause in Finnish
The label F corresponds roughly to what is standardly called AgrS. In Holmberg et al. it is mnemonic for 'finite.' The reason why they use the category label F instead of AgrS is, primarily, that a certain finite verb form in Finnish, the passive, is marked by a suffix -Vn which corresponds in terms of its morphological distribution to subject agreement (it is the outermost verbal inflection), but which is invariant, showing no agreement. According to Holmberg et al., it is a realization of finiteness, of which AgrS is another realization.1A further reason for preferring the label F is that although the spec position of the head in question is the default position for the nominative subject, other categories may fill this position, the generalization being that a category in this position, when referential (i.e., except when it is an expletive), has the discourse function of topic; this will be discussed below. That is to say, AgrS is associated with this head but is not a defining property of it The theory we elaborate in this paper is consistent with the assumption that agreement features (<j)-features) are not represented anywhere in the syntax except as part of nominal categories and the agreement inflection base-generated on the verb or other predicative head. This is not crucial, though; we may equally well assume that F is inherently provided with <|>-features which are checked off in the manner sketched in Chomsky 1993.2
74
SUBJECTS, EXPLETIVES, AND THE EPP
Only F and T(ense) are obligatory in the structure of the finite clause. All the other functional categories are optional. Following Holmberg et al. we assume that Mood (Potential -ne-, Conditional -isi-) is a feature of the category T in Finnish. Here is an example of a sentence with a maximal structure of inflectional categories; In the example structure (5) we illustrate T with the conditional form for reasons of exposition: the conditional is morphologically more visible than the indicative.
'that (the) children wouldn't have eaten (the) sausage' A characteristic property of Finnish is that I is visibly split into F and T in one construction, namely negative finite sentences: The negation is inflected for subject agreement while the next head down, either the auxiliary or the main verb, is inflected for Tense and Mood. Following Mitchell 1991 and Holmberg et al. 1993, we assume that Neg is merged with TP, and raises from Neg to F.
3. Subjects, topics, and the EPP in Finnish 3.1. Null subjects We will begin by considering some parameters involving the subject or, more generally, the highest spec positions in the sentence. First, as is well known, languages differ with regard to whether they allow null subjects. For instance, Italian does, but English does not:
EXPLETIVES, SUBJECTS, AND TOPICS IN FINNISH (6)
a. b. c.
Sono stanco. am tired Piove. rains E venuto un uomo della finestre. has come a man through, the window
75 Italian
The received view is that the Italian constructions above all have a small pro subject, licensed by AgrS. The reason why, for instance, English does not allow null subjects is that AgrS lacks the features required to license pro. Consideration of Finnish shows that this cannot be the whole truth. It shows, in particular, that a construction like (la), featuring a null referential subject, and a construction like (Ic), which according to the received view features an expletive null subject, are not licensed by the same mechanism: Finnish is a null-subject language, licensing referential subject pro and "quasi-referential" pro (as in weather constructions), yet it does not allow verb-initial impersonal constructions. We repeat examples (1) and (2): (1)
a.
Olen vasynyt. be. ISG tired 'I'm tired' b. Sataa (vetta). rains water c. * Leikkii lapsia kadulla. play children in. street
Sentences (2a-c) are well-formed variants of (Ic) (EXPL = expletive pronoun): (2)
a. b. c.
Sita leikkii lapsia kadulla. EXPL play children in. street Kadulla leikkii lapsia. in. street play children Lapsia leikkii kadulla. children play in. street 'Children are playing in the street'/'There are children playing in the street'
Finnish has rich subject agreement morphology, distinguishing three persons and two numbers in (almost) all tenses and moods. It seems clear enough that the null subject in (la) is licensed by virtue of the features of AgrS. Yet these features apparently do not help in the case of (Ic). A possible objection at this point is that Finnish is not a full null-subject language, since third-person referential pronouns cannot be freely dropped. Consider the paradigm in (7): the pronouns are optional in first and second persons, but obligatory in the third persons.
76 (7)
SUBJECTS, EXPLETIVES, AND THE EPP a.
(Mina) ol-i-n vasynyt. / be-PAST-lsc tired
(Me) ol-i-mme vasyneita. we be-PAST-lPL tired.PL
b.
(Sina) ol-i-t
(Te) ol-i-tte
c.
thou be-PAST-2sc tired Han ol-i-0 vasynyt. he/she be-PAST-3SG tired
vasynyt.
vasyneita.
you be-PAST-2pi tired. PL He ol-i-vat vasyneita. they be-PAST-3pL tired. PL
This does not, however, explain why the overt expletive is required in (Ic). Note that the subject pronoun in construction with the weather verb in (Ib) can be null. In colloquial Finnish the pronoun can optionally be overt:3 (8)
Nyt (se) taas sataa. now it again rains 'Now it's raining again'
This is also true of the pronoun in construction with an extraposed sentence: it is only optionally overt. (9)
(Se) oli hauskaa etta tulit kaymaan. it was nice that [you] came visiting
The weather pronoun is arguably not expletive in the same sense as the pronoun in existential and ergative constructions (the weather pronoun is sometimes characterized as 'quasi-referential'). As for the extraposition pronoun, it is generally agreed that it is referential (cf. Bennis 1986). It may still be the case that the weather pronoun and the extraposition pronoun are both somehow defective, lacking some specification(s) that standard referential pronouns have. Rizzi (1986) suggests that they have number but not person. Assume that it is, rather, the other way around: the weather pronoun and the extraposition pronoun have specification for person, with the default value 3, but not number. If we then postulate that Finnish third-person AgrS is not specified for number, it follows that it can identify a null numberless third-person pronoun (i.e., the null extraposition pronoun or a null weather pronoun), but not an ordinary referential pronoun specified for person and number. This hypothesis can be supported by the observation that in spoken Finnish, verb agreement does not distinguish morphologically between third-person singular and third-person plural (the 3so form in (7c) is used for both).4 Now if Finnish AgrS can license a null pronoun as long as it is not marked for number, there is no reason, within a theory along the lines of Rizzi 1986, why it should not license a null expletive in existential, ergative, impersonal passive, and other constructions. Furthermore, there is another language which displays the same partial referential pro drop as Finnish, namely, Hebrew (see Shlonsky 1988). But unlike Finnish, Hebrew has no expletive pronouns. So apparently the Finnish-Hebrew type of partial pro drop does not necessarily correlate with overt expletives. In conclusion, the fact that Finnish is only 'partially pro drop' does not explain why it cannot license a null expletive pronoun in (Ic), assuming that null expletive pronouns exist. Alexiadou and Anagnostopoulou (1998; henceforth A&A) present another theory of null subjects and the EPP, according to which
EXPLETIVES, SUBJECTS, AND TOPICS IN FINNISH
77
(a) EPP is universal: a strong D-feature in I (following Chomsky 1995: ch. 4); (b) In some languages the AgrS features on the finite verb are nominal enough to check the EPP feature. In these languages V-movement to I is sufficient to satisfy the EPP. In other languages either a nominal argument must move to SpecIP, or a nominal expletive must be inserted there, to check the EPP feature. (c) Null-subject languages (NSLs) have subject agreement features which can check the EPP feature. This is why they do not have overt expletives. A&A argue that they do not have covert expletives either. (d) Since A-movement to SpecIP is not needed in NSLs, they have no such movement. Instead, all movement to preverbal position is A-bar movement. More precisely, A&A argue that preverbal arguments in finite clauses are left-dislocated in the languages they discuss. Their theory of NSLs is empirically supported mainly with data from Greek and Spanish, but other languages are discussed as well. The theory makes the same prediction as Rizzi's (1986) theory: NSLs do not have expletives. As shown above, the prediction fails in the case of Finnish: Finnish has rich subject agreement morphology, with overt obligatory movement of the head bearing the agreement features to I, where the features are capable of licensing a null referential or quasi-referential subject, yet Finnish does not allow verb-initial impersonal sentences as in (Ic). Matters are complicated by the fact that Finnish does allow some verb-initial impersonal sentences. The following are some examples: (10) a.
Sattui onnettomuus. occurred [an] accident
b.
On ilmennyt ongelmia. have appeared problems
c.
Tuli kiire. came haste 'One is in a hurry'
In these examples the expletive sita is optional. We will return to these examples below, showing what it is that distinguishes them from the (more common) cases where an expletive is obligatory. At this point, just note that in terms of A&A, Vmovement is sufficient to "check the D-feature in I" in Finnish, judging from examples like (10), but this does not eliminate the need for an overt expletive in other cases. This indicates that there are at least two distinct parameters involved, or, in terms of checking theory, two distinct features. There is one feature which requires nominal features in SpecIP or in I; this feature can be checked by nominal features on a verb moved to I, and arguably correlates with the NS-property. But there is also another feature, active in some languages but not others, which requires the filling of SpecIP, and which is not checked by features of the finite verb moved to I. In the following we will try to flesh out this hypothesis. But first we need to discuss some additional parameters which also concern the subject position, broadly speaking.
78
SUBJECTS, EXPLETIVES, AND THE EPP
3.2. Subject versus topic-prominent languages The distinction between "subject-prominent" and "topic-prominent" languages has been discussed in the literature in various frameworks and in relation to various languages; see E. Kiss 1995, 1997, this volume, for some recent discussion. A language is subject-prominent when the argument which is externalized in an active sentence is always the subject in the sense of the thematically highest argument in VP.5 A language is topic-prominent when the argument which is externalized need not be the subject, but can be any category capable of functioning as topic. English is generally taken as the perfect representative of subject-prominent languages, while representatives of topic-prominent languages include Chinese (Li and Thompson 1976), Tagalog and related languages (Guilfoyle et al. 1992), and Hungarian (E. Kiss 1995, 1997, this volume). Greek and Spanish, discussed by A&A, presumably also qualify as topic-prominent languages in this sense. Consider the following sentences: (11) a. Graham Greene has written this book, b. * This book has written Graham Greene. (12) a. b.
Graham Greene on kirjoittanut taman kirjan. Graham Greene has written this book Taman kirjan on kirjoittanut Graham Greene. this book has written Graham Greene
The word order in (12b), an active construction, is perfectly grammatical in Finnish but sharply ungrammatical in English (if Graham Greene is the author). The closest English translation of (12b) is the passive (13) (13)
This book is written by Graham Greene.
The passive is the standard way of externalizing a nonsubject topic in English and other subject-prominent languages. (12b) may be used in any situation where the book has been previously introduced in the discourse while the identity of the author is new information. Consider, for example, a situation where a teacher is sitting with a stack of books in front of him, presenting one book at a time, telling the students facts about the author. Taking a new book from the stack and holding it up to the students, the teacher might utter (12b). In this situation the existence of the book is old information, while Graham Greene's role in the depicted event is new information (cf. Vallduvf and Engdahl 1996). In this situation the passive (13) would be the preferred word order in English, while (12b) would be the preferred order in Finnish. The possibility of expressing this meaning with an active construction structured basically as in (12b), the object c-commanding the subject, is characteristic of topic-prominent languages, as we see it. By this test, for instance, French is subject-prominent, while Russian is topic-prominent.6 (14)
* Ce livre a ecrit Graham Greene. this book has written Graham Greene
French
(15)
Etu knigu napisal Graham Greene. this book has. written Graham Greene
Russian
EXPLETIVES, SUBJECTS, AND TOPICS IN FINNISH
79
As an informal characterization, the topic is the sentence constituent which refers to the entity which the sentence is about (cf. E. Kiss 1995, 1997, this volume). Therefore, typically it has specific reference. The most prototypical topic, therefore, is an unstressed definite pronoun. 3.3. Two triggers for externalization The following is a formal account of the variation discussed in the two preceding sections. First, we assume that arguments are all provided with a feature [±Foc], which is either inherent (e.g., weak pronouns and clitics are inherently [-Foe]) or assigned to the argument when it is merged with a syntactic tree. The semantic effect of the feature [+Foc] is that the argument is interpreted as the information focus, in the sense of Vallduvi and Engdahl 1996. The semantic effect of the feature [-Foe] is that the argument is interpreted as part of the presupposition or 'ground' (cf. Vallduvi and Engdahl 1996).7 There may be distinctions among the [-Foc]-marked arguments in terms of salience or foregrounding, and so on, but we disregard such differences here. Second, we assume that a feature [-Foe] must move to a position outside the predicate phrase. The feature [-Foe] is an uninterpretable feature, in the sense of Chomsky 1995: ch. 4. As such it must be checked, and thereby eliminated, before LF.8 The feature is checked by a feature of F, the 'EPP feature.' In addition to the EPP, checking [-Foe] is presumably crucial in processes such as Scrambling and Object Shift (cf. Holmberg 1997); however, in this work we will consider only EPP-related [-Foe]-movement. Broadly following Diesing 1992, we assume that the sentence is divided into three domains: the focus domain, the presupposition domain, and the operator domain. In Diesing 1992 the focus domain is VP. We assume it is TP, the maximal projection of the predicate; this will be substantiated below. The presupposition domain is then FP, while the operator domain is CP. Arguments are merged in VP, where they receive their thematic roles, and so on, but arguments which are not part of the information focus must ultimately be moved out of the focus domain. This idea has been implemented in various ways in the literature. We do it here in terms of movement of the postulated feature [-Foe], attracted by a feature in F. Adopting in essence the theory of movement in Chomsky 1995: ch. 4, we assume that syntactic features can be interpretable or uninterpretable. The uninterpretable features have to be checked, and thereby eliminated, before the LF-interface. This is the mechanism which drives some or even all syntactic movement. Furthermore, movement specifically affects features. A feature selected for movement may move alone, or it may pied-pipe other features. If the phonological features are among the features pied-piped, we have a case of overt movement. If not, the movement is covert (cf. also Bobaljik 1995, Groat and O'Neil 1996). The feature [-Foe] is uninterpretable, and as such must be checked. The checker is a feature of F, which we will henceforth refer to as the EPP feature. This feature is also uninterpretable, and hence must also be eliminated in the course of the syntactic derivation. We assume that this feature attracts all [-Foe] features in its do+main to F.9 Languages vary with regard to whether [-Foe]-movement is overt o
80
SUBJECTS, EXPLETIVES, AND THE EPP
not, that is, whether [-Foe] does or does not pied-pipe other features, specifically the phonological features of the category dominating it. In Finnish the EPP feature is 'strong' in the sense that it can be eliminated only by a phonologically visible category—in fact, only by a phrasal category. Consequently one of the moved [-Foe] features must move overtly, that is, it must pied-pipe the minimal maximal category dominating it, with its phonological matrix, to SpecFP. Given the overarching economy principle according to which movement should be restricted to a minimum, which in terms of feature-movement theory means that as few features as possible should move (cf. Chomsky 1995: ch. 4), only one of the moved [-Foc]features can pied-pipe anything. We thus derive the result that in Finnish sentences with one or more [-Foc]-marked categories, one and only one of these categories will move overtly to SpecFP. In a language where the EPP feature is weak (in the sense of Chomsky 1995: ch. 4 all [-Foe]-movement will be covert. Note the analogy to vv/z-movement. As is well known, languages differ with regard to whether w/z-movement is overt or not. One way of formally expressing this variation is to assume that w/z-phrases have a w/i-feature which is attracted by an uninterpretable WH-feature in C, which, as in the case of the [-Foe] feature just discussed, attracts all the w/i-features in its domain. The cross-linguistic variation is due to whether WH in C is strong or not (i.e., whether it can be checked only by an overt category). If it is strong, one of the attracted features must pied-pipe its category with a phonological matrix; if weak, all wft-movement can be covert.10 So movement of [-Foe] out of VP is one of the mechanisms behind externalization of arguments. There are other mechanisms, however. To begin with, there is a feature in I which is checked by an overt nominative category. We leave open at this point whether the feature is properly located in T, F, or even C. We will discuss this matter in some detail in subsequent sections, but in the present section we shall refer to it simply as IP. In Holmberg and Platzack 1995 the feature is identified as a finiteness feature, present in finite clauses only. They argue that the feature in question can be checked either by a nominative DP moved to or (in the case of expletive pronouns) inserted in SpecIP, or by AgrS. The latter is possible in languages where AgrS counts the feature [nominative] among its features, typically in null-subject languages and certain other languages with morphologically rich AgrS (for instance Icelandic). Note that this feature is similar to the [Defeature postulated by A&A: a feature of I checked either by a DP in SpecIP or by a verb in I with nominal agreement features.11 Let us regard the relevant feature as a nominative Case feature of I. The feature triggers movement of the finite verb to I in languages with rich enough subject agreement morphology. It triggers movement of a nominative DP to SpecIP in languages where subject agreement is not rich enough, lacking a nominative feature in particular. Alternatively, it triggers insertion of a nominative expletive in SpecIP. This accounts, in part, for the distinction between subject- and topic-prominent languages. In English, for instance, the nominative Case feature is not checked by the finite verb, which entails that an overt nominative DP must move to SpecIP, regardless of the value of its focus feature. Alternatively an expletive pronoun is inserted in SpecIP. Consequently no nonsubject argument will ever get a chance of moving to SpecIP.12
EXPLETIVES, SUBJECTS, AND TOPICS IN FINNISH
81
The scenario depicted above is almost certainly too simple to account for all the variation found among the languages of the world regarding the EPP (loosely speaking). To begin with, the appearance of an expletive pronoun in English infinitival clauses is not predicted by the theory outlined so far:13 (16)
I believe *(there) to be a dead man in the cellar.
Second, there are languages where V has agreement features which can check [D] in I but where nevertheless a nonsubject argument cannot in general move to SpecIP across the subject. Third, Scandinavian Stylistic Fronting (cf. Holmberg 2000b) shows that there is variation among the languages where [D] is checked by the finite verb regarding what categories can satisfy the EPP. Scandinavian Stylistic Fronting is an operation which fills SpecIP with an overt category in constructions where that position is not filled by the subject, namely, in subject relatives, embedded subject questions, and various impersonal constructions. As discussed by Holmberg (2000b) the filler can be virtually any category, head or phrase. In Finnish, the generalization seems to be that the EPP can be satisfied only by categories which are referential in a broad sense, including locative and temporal adverbials but excluding sentence adverbials and manner adverbials. (17) a. b. c.
d. e.
Tana'an leikkii lapsia kadulla. today play children in. street Tromssassa leikkii lapsia kadulla. in.Troms0 play children in.street Kirveen avulla murtautuu helposti taloon. axe. GEN with, help break easily into, house 'One can easily break into the house with the help of an axe' * Ehka leikkii lapsia kadulla. perhaps play children in.street * Helposti murtautuu taloon kirveen avulla. easily break into.house axe.GEN with.help
In addition, the EPP can be satisfied by an expletive, se or sita. Note that referential small pro also can check the EPP feature. This is shown by the fact that the expletive cannot co-occur with a referential null subject: (18)
Miksi (*sita) leikitte kadulla? why EXPL play. 2PL on. street 'Why are you playing in the street?'
Although SpecIP is phonetically empty in (18), the EPP feature is, apparently, checked, since the expletive is not needed to check it (and therefore is excluded by the principle of Full Interpretation). We have opted for postulation of a feature [-Foe] as the 'EPP feature' in Finnish. 3.4. Verb-initial sentences in Finnish As mentioned, Finnish does exhibit certain V-initial impersonal constructions. They complicate the scenario sketched in the preceding section in an interesting 00000
82
SUBJECTS, EXPLETIVES, AND THE EPP
(19) a. b. c.
d. e.
On ilmennyt ongelmia. have appeared problems Sattui onnettomuus. occurred [an] accident Tuli kiire. came haste 'We/they are in a hurry' Sataa vetta. rains water Oli hauskaa etta tulit kaymaan. was nice that came.2sc visiting 'It was nice that you came by'
In these constructions the expletive sita is optional, except in (19e), where the subject pronoun, if there is one, must be se (the nominative form of 'it'). Compare these with (20), all of which require either that an argument is moved to sentenceinitial position, or that the expletive sita is added. (20) a. * On leikkinyt lapsia kadulla. have played children on. street b. * Astui mies huoneeseen. entered man into.room c. * Kay vat monet elakelaiset paivittain uimassa. go many pensioners daily swimming Concerning (19) note the following generalizations. First, if a locative, goal, or other argument is added to the sentence, the result is generally bad, unless the added argument is moved to initial position. (21) a. * Ilmeni ongelmia pian. appeared problems soon b. Pian ilmeni ongelmia. (22) a. * Sattui onnettomuus minulle. occurred accident to.me b. Minulle sattui onnettomuus. (23) a. * Tuli kiire sille. came haste to.him b. Sille tuli kiire. (24) a. * Sataa vetta nyt. rains water now b. Nyt sataa vetta. (25) a. * Oli hauskaa minusta etta tulit kaymaan. was nice I.ABL that came.2sc visiting b. Minusta oli hauskaa etta tulit kaymaan. 'I thought it was nice that you came by'
EXPLETIVES, SUBJECTS, AND TOPICS IN FINNISH
83
Second, although the constructions in (19) contain a complement, that complement cannot move to sentence-initial position: (26) a. * Ongelmia ilmeni. problems appeared
b. * Onnettomuus sattui. [an] accident occurred c. * Kiire tuli. haste came
d. * Vetta sataa.14 water rains e. * Etta tulit kaymaan oli hauskaa. that came. 2sc visiting was nice The generalization is that a V-initial sentence is allowed in Finnish just in those cases where the sentence does not contain a category which can move to preverbal position (a 'potential topic'). As noted in the previous section, this confirms that Vmovement alone is capable of checking the Case-feature of I. The following stipulation will go a long way toward explaining this generalization: (27) The EPP feature in F is optional in Finnish. Consider what happens if F is merged without the EPP feature. The sentences in (19) will not violate the EPP. The argument, if there is one, is [+Foc], and therefore need not move. Apart from (19e) the sentences all assert the coming into existence of a phenomenon, so a presuppositional reading of the argument is out of the question. As for (19e), it may be the case that clausal complements are inherently [+Foc]. Now what happens if an additional argument (in the wide sense) is added, as in (21)-(25)? Recall that all arguments have to be supplied with a [-Foe] or a [+Foc] feature. Assume that at most one argument can be [+Foc]. If so, the added argument has a [-Foe] feature which must be checked. But if F is merged without the EPP feature, the [-Foe] feature cannot be checked, and the derivation crashes at LF, on account of containing an uninterpretable feature. This is why the (a)-sentences in (21)-(25) are starred under the no-EPP-feature option. For the same reason the sentences in (20) are starred. The sentences in (26) are also starred, but for a different reason: they violate the last resort condition on movement (Chomsky 1993, 1995): The [+Foc] arguments have moved although there is no movement trigger. Consider what happens if F is merged with the EPP feature. The sentences in (19) are ruled out: the derivation crashes due to an unchecked uninterpretable feature. Merging an expletive will save the derivation. As mentioned, an expletive (sita in (19a-d), se in (19e)) is optional in these constructions. The (a) sentences in (21)-(25) are also ruled out, now containing two unchecked uninterpretable features: the EPP feature and the [-Foe] feature of the complement. The same is true of (20). The (b)-sentences of (21)-(25) are well formed, the [-Foe] feature of the fronted argument and the EPP feature both erased as a result of overt movement of the argument to SpecFP. The sentences in (26) are ruled out with or without an EPP feature: a [+Foc] argument cannot check the EPP feature.15
84
SUBJECTS, EXPLETIVES, AND THE EPP
4. The topic position is SpecFP As mentioned in section 3.3, we claim that subject and nonsubject topics land in the same position in Finnish. In this section we will argue that the position in question is SpecFP. Consider again the two alternative ways of saying in Finnish that the writer of this book is Graham Greene. (28) a. b.
Graham Greene on kirjoittanut taman kirjan. Graham Greene has written this book Taman kirjan on kirjoittanut Graham Greene. this book has written Graham Greene
Let us first assume that the structure of (28a) is (29): (29) [FP Graham Greenet [F on: [TP t{ [T fj [AuxP t-} [PrtP kirjoittanutk [VP t{ [v, /k tama'n kirjan ]]]]]]]] This should be uncontroversial as far as the position of the subject goes; it has moved to the highest position in the IP-domain, arguably via SpecTP. We now claim that (28b) has the structure (30): (30) [pp [taman kirjan]j [F on; [TP /• [T- /• [AuxP t-} [PrtP kirjoittanutk [VP Graham Greene [v, t-} t{ ]]]]]]]] In the following we will argue specifically that the final landing site of the fronted object in (28b) is the same as that of the fronted subject in (28a). In section 6.2 we will show that the subject may occur in a lower VP-external position in the socalled MSC, and that likewise the object, when externalized, may occur in this position. This lends support to the general view that the subject has no privileged status with regard to the EPP in Finnish. Consider what the alternatives are to the hypothesis that the topicalized object in (28b) is in SpecFP: A. It is in SpecCP; B. It is between SpecCP and SpecFP (e.g., adjoined to FP); C. It is lower than SpecFP. The following facts rule out alternative A: (a) As noted, a phrase in SpecCP is contrastive, but the topicalized object in (28b) need not be contrastive any more than does the subject in (28a). (b) A subject topic can occur embedded, preceded by a complementizer. This holds true of object topics, too. (31) a.
Se etta Graham Greene on kirjoittanut taman kirjan, ei merkitse it that Graham Greene has written this book not means etta se olisi helppo lukea. that it be.CQND easy read.INF 'The fact that Graham Greene has written this book does not mean that it's easy to read'
EXPLETIVES, SUBJECTS, AND TOPICS IN FINNISH b.
85
Se etta taman kirjan on kirjoittanut Graham Greene, ei merkitse it that this book has written Graham Greene not means etta se olisi helppo lukea. that it be.CQND easy
read. INF
'The fact that this book is written by Graham Greene . . .' (c) The topic, whether subject or object, can be preceded by the finite head when that head is focused by movement to C, as typically in yes-no questions, where the head is affixed with a question particle: (32) a.
b.
Onko Graham Greene tosiaan kirjoittanut taman kirjan? has.Q Graham Greene really written this book 'Has Graham Greene really written this book?' Onko taman kirjan tosiaan kirjoittanut Graham Greene? has. Q this book really written Graham Greene 'Is this book really written by Graham Greene?'
The following fact speaks against alternative B (the topic is between C and the subject): A nonsubject topic phrase cannot precede the subject in the initial portion of the sentence, when we control for the possibility that the nonsubject phrase is in SpecCP. (33)
Miksi (*tata kirjaa) Graham Greene (tata kirjaa) ei (tata kirjaa) why this book Graham Greene not olisi (tata kirjaa) voinut (tata kirjaa) kirjoittaa (tata kirjaa)? have.CQND
been.able.to
write
'Why could Graham Greene not have written this book?' The freedom of placement that the object enjoys in this construction is due to the possibility of scrambling verb complements in sentences which have a filled SpecCP (to be discussed below). As can be seen, the object may scramble leftward all the way up to, but not past, the subject, if the subject is a topic (i.e., if it does not remain in VP with focus): there is no place for the object between the subject in SpecFP and SpecCP.16 The following fact speaks against alternative C (the nonsubject topic is lower than the subject): A subject topic precedes the highest head in IP, that is, F, the head bearing subject agreement (i.e., the negation, the auxiliary, or the finite verb) except if F is moved to C. Movement of F to C is always accompanied by some form of focus on the head, realized as a focus particle or as contrastive intonation. (34) a.
Ilmeisesti Graham Greene on kirjoittanut taman kirjan. evidently Graham Greene has written this book b. On Graham Greene kirjoittanut taman kirjan. has Graham Greene written this book 'Surely Graham Greene has written this book' c. * Ilmeisesti on Graham Greene kirjoittanut taman kirjan. evidently has Graham Greene written this book
In (34c) the auxiliary cannot have a focused reading. This follows if (a) a focused auxiliary must move to C, and (b) the adverb cannot be adjoined higher than to FP
86
SUBJECTS, EXPLETIVES, AND THE EPP
(as is generally the case in Finnish; cf. Holmberg 1997). The example thus shows that the finite, AgrS-bearing head H can precede the subject if and only if H is in C. The same holds true of the nonsubject topic. (35) a.
Ilmeisesti taman kirjan on kirjoittanut Graham Greene. evidently this book has written Graham Greene b. On taman kirjan kirjoittanut Graham Greene. has this book written Graham Greene 'Surely this book is written by Graham Greene' c. * Ilmeisesti on taman kirjan kirjoittanut Graham Greene. evidently has this book written Graham Greene
We conclude that the topic position in Finnish is SpecFP, whether the topic is a subject or not. Below we will show that the subject may occur in other positions, too, below F but preceding VP, namely, when SpecFP is filled by an expletive. We will show that, again, this holds not only for subjects but also for other sentence constituents, which can function as topics.
5. Differences between subject and nonsubject topics What we have established at this point is that the nonsubject topic occupies the same structural position as the subject topic, namely, SpecFP. In certain other respects the nonsubject topic does not behave like a subject, though. Thus, with regard to case, agreement, binding, and reconstruction, subjects and nonsubjects are distinct, whether they are topics or not. Consider case. We have already seen several examples illustrating the fact that an object topic retains objective case (usually accusative or partitive), while the subject left behind in VP has nominative case; see, for instance, (Ib). The generalization is that an object topic always bears the case that is assigned to it (or checked) downstairs. Consider (36): the verbpitda 'hold, like' selects ablative case. (36b) shows that this case is preserved under fronting to topic position. (36) a.
Lapset
pitavat tasta kirjasta.
children like
b.
this book.ABL
Tasta kirjasta pitavat lapset.
Consider (37). In Finnish an object in the scope of negation gets partitive case. (37b) shows that the partitive case is retained under fronting to topic position, while the subject which is left behind in VP, hence presumably c-commanded by negation, is nominative, unaffected by negation. (37) a.
Graham Greene
ei ole kirjoittanut ta'ta kirjaa /*taman kirjan.
Graham Greene.NOM not has written
b.
this book.PARTlthis book.ACC
Tata kirjaa ei ole kirjoittanut Graham Greene. this book. PART not has written Graham Greene.NOM
Consider agreement. Example (38) shows that the nominative subject, regardless of its position, triggers verb agreement. A nonsubject topic never triggers verb agreement.
EXPLETIVES, SUBJECTS, AND TOPICS IN FINNISH (38)
87
Taman kirjan ovat lukeneet monet opiskelijat. this book.SG.ACC have.PL read.PL many students.PL.NOM
Consider binding. Example (39) shows that a nonsubject topic cannot control anaphora: (39) a.
b.
Myohemmin ehdottaja hylkasi ehdotuksensa. later proposer abandoned proposal POSS. 3 'Later the proposer abandoned his proposal' Myohemmin ehdotuksensa hylkasi sen ehdottaja/*ehdottajansa. later proposal abandoned its proposerlproposer.poss.3
The Finnish third-person possessive suffix (-nsa or -Vri) is an anaphor, essentially falling under Principle A of the Binding Theory (see Trosterud 1993). As shown by (39b), the object cannot bind an anaphor in the subject even though the object in this construction is (by hypothesis) in SpecFP and the subject in SpecVP, so that the structural conditions required by Principle A are met. Finally, consider reconstruction. (40) a.
b.
Ilmeisesti itseaan aanesti vain Jussi. apparently for.himself voted only Jussi 'Apparently the only person who voted for himself was Jussi' Anna on hukannut kenkansa, Liisa on hukannut kenkansa, Anna has lost shoes.POSS.3, Liisa has lost shoes.poss.3, ja nyt kenkansa on hukannut Arnekin. and now shoes.POSS.3 has lost Arne.too 'Anna has lost her shoes, Liisa has lost her shoes, and now Arne, too, has lost his shoes'
The fronted object in (40a) is the reflexive pronoun itse 'self plus a possessive suffix. Given standard binding theory, the object must be reconstructed for binding to be possible. The same holds true of the object with a possessive suffix in (40b). The sentence-initial adverbial rules out an analysis where the object is fronted to SpecCP (which is further confirmed by the noncontrastive interpretation of the object in both cases). We seem to be led to the conclusion that the object topic heads an A-bar chain from SpecFP; the properties listed above as characteristic of object topics are shared by fronted wft-phrases or constituents fronted to SpecCP for contrast. On the other hand, when the subject topic occupies SpecFP, it exhibits all the signs of heading an A-chain: its case and agreement features are checked in this position, it binds anaphora, and it is not subject to reconstruction.17 If so, SpecFP is a "mixed position": an A-position when filled by the subject, but an A-bar position when filled by a nonsubject. The notion that there are positions which are open as regards A/A-bar status has been proposed for various languages (cf. Diesing 1990, Rognvaldsson and Thrainsson 1990, and Rizzi 1991, among others). On the other hand, even when filled by a nonsubject, SpecFP has certain properties which are not characteristic of A-bar positions. For one thing, it does not interfere with canonical A-bar movement to SpecCP. Consider (41).
SUBJECTS, EXPLETIVES, AND THE EPP (41)
Kuka taman kirjan on kirjoittanut? who this book has written '(What about this book:) Who has written this book?'
The initial w/i-phrase is in SpecCP, uncontroversially an A-bar position, as a result of movement. If the object topic is in SpecFP and if SpecFP is an A-bar-position, we predict, on fairly standard assumptions, that the construction should be ungrammatical due to a violation of Relativized Minimality. There is a slight problem, though, in that the construction may be derived by scrambling of the object to some lower VP-external position, rather than fronting it to SpecFP. As mentioned earlier, scrambling is always an option when SpecCP is filled. Consider, for instance, (42). SpecCP is filled by a w/i-phrase, SpecFP is occupied by the subject, and the locative complement may optionally scramble to a lower preverbal position, either preceding the auxiliary or between the auxiliary and the main verb. (42)
Milloin Marja (Pariisissa) on (Pariisissa) kaynyt (Pariisissa)? when Marja (in.Paris) has (in.Paris) been (in.Paris) 'When has Marja been to Paris?'
However, the position preceding the negation is too distant for scrambling; as discussed in section 4, the position immediately preceding the negation is unambiguously SpecFP. Consequently, we can be reasonably certain in (43) that the object is in SpecFP, yet no Relativized Minimality effect can be discerned. (43) a.
b.
Miksi pihalla ei leiki lapsia? why on.yard not play children 'Why aren't there children playing in the yard?' Kuka minusta ei pida? who me not like 'Who doesn't like me?'
Another property which the nonsubject topic has which is unexpected if it heads an A-bar chain is that it can control a floated subject quantifier. (44) a. b.
c.
Ilmeisesti kriitikot ovat (kaikki) ylistaneet tata kirjaa. evidently the.critics have all praised this book Ilmeisesti nama kirjat on (kaikki) kirjoittanut Graham Greene. evidently these books has all written Graham Greene 'Evidently Graham Greene has written all these books' Nama kirjat Graham Greene on (*kaikki) kirjoittanut (kaikki).
In (44a) a subject topic controls a preverbal floated quantifier. In (44b) an object topic controls a floated quantifier in what looks like the same position. (44c) shows that a phrase moved to SpecCP cannot control a quantifier in that position, although it can control a postverbal quantifier.18 We suggest the following explanation of why object topics exhibit this particular mix of so called A- and A-bar properties. Consider first agreement, arguably the smallest problem for the present theory. In Finnish, as in many other languages, the verb can agree only with a nominative argument, and hence can never agree with an object even if that object is in SpecFP. We may assume, following Holmberg and
EXPLETIVES, SUBJECTS, AND TOPICS IN FINNISH
89
Platzack 1995, that the reason is that AgrS is inherently nominative in the languages in question, and hence can be checked only by a nominative category. We return later to the mechanism which ensures that the (^-features and the Case of the subject are checked even when the subject remains in VP. Now consider binding and reconstruction. What object movement to SpecFP has in common with w/i-movement or contrast movement to SpecCP is that they are all movements from a Case-position to a non-Case-position. In other words, the Case of the moved category is checked in the launching site, not in the landing site. By contrast, movement of the nominative subject to SpecFP is movement to a Case-checking position.19 Now assume that the only member of an A-chain which is visible for Binding Theory is the member in the Case-checking position.20 It follows that the subject will bind an anaphor from SpecFP (or SpecTP; see note 21), and is not reconstructive from SpecFP (or SpecTP), while the object can bind an anaphor only from its Case-checking position inside VP, and must be reconstructed to that position if it contains an anaphor. Consider Quantifier Float. The object in SpecFP can control a floated quantifier because that relation falls not under Binding Theory, but movement. Following Sportiche 1988 and much subsequent work, we assume that Quantifier Float is derived by movement of a DP out of a QP headed by the quantifier. The position of the quantifer is thus either the base position of the QP or a position to which the QP has moved before being stranded by the DP. We will see below that an object moved out of VP need not move all the way to SpecFP, but may be realized in a position between SpecFP and VP. We are therefore not surprised to find that a floated quantifier may occur in such a position: We assume the quantifier and its complement DP have moved to a position outside VP but lower than SpecFP, where the quantifier is stranded while the DP moves on to SpecFP. Why movement to SpecCP cannot proceed in two steps in this way remains an open question, though.21 Finally, the absence of any interaction between movement to SpecCP and movement to SpecFP follows within the movement theory of Chomsky 1995: ch. 4. According to Chomsky, movement is triggered by Attraction: a feature F attracts a matching feature F' from downstairs, triggering movement of F' (pied piping an entire category in the case of overt movement) to the checking domain of F (with subsequent checking and erasure of either F or F1, as the case may be). As dictated by the MLC, the attractor F will always attract the closest instance of F' in its c-command domain, and will not see any other features. A number of standard Relativized Minimality effects follow from this condition, including superiority, w/z-island effects, and at least some HMC-effects. Now assume that w/i-movement is triggered by a [WH]-feature in C attracting a [WH] feature from downstairs. The presence of an object (or subject) in SpecFP will have no effect, except if the object (or subject) has the feature [WH], in which case it will itself be moved, blocking movement of a lower category. Assume that movement of a contrasted category to SpecCP is likewise triggered by a feature [Contrast] in C. Again presence of an object (or subject) in SpecFP will have no effect unless the object (or subject) is marked [Contrast].
90
SUBJECTS, EXPLETIVES, AND THE EPP
6. Expletives and argument structure In this section we will discuss the expletive sita, in particular as it appears in the 'Multiple Subject Construction' (MSC). We will show (a) That sita is a pure expletive in Chomsky's (1995) sense. In other words, it has no <j)-features and is thus not directly involved in any Case or agreement checking. It is also not a placeholder for the subject. Its function is just to check the EPP feature. (b) The position of sita is SpecFP (with one minor exception). 6.1. Sita is not nominative The following is a list of constructions exhibiting the expletive sita: (45) The Partitive Construction (see Nikanne 1994) Sita leikkii lapsia pihalla. EXPL plays children. PART in.yard There are children playing in the yard' (46) Passive Sita vasytaan nykyaan helpommin kuin ennen. EXPL get.tired.PASS nowadays easier than before 'These days people get tired easier than before' (47) The Generic Subject Construction Sita va'syy nykyaan helpommin kuin ennen. EXPL gets, tired nowadays easier than before 'One gets tired these days easier than before' (48) The Multiple Subject Construction Sita ovat nama lapset jo oppineet uimaan. EXPL have these children already learned to. read 'These children have already learned to read' Morphologically sita is the partitive form of the third-person singular pronoun, the nominative form of which is se. As a referential pronoun it is used to refer to nonhuman as well as (colloquially) human referents of both genders. We claim that sita in the expletive constructions is a pure expletive in the sense of Chomsky 1995, which is to say that it does not head a chain, does not have nominative case, and (hence) does not control agreement. It also does not have any privileged relation to the subject. Its function is just to overtly fill the SpecFP position in constructions where there is no lexical filler of that position, that is, formally to check the EPP feature in F. All the constructions (45)-(48) are ungrammatical if the expletive is left out, unless an argument or adverbial is moved to the initial position, as discussed in section 3. According to an influential tradition the function of expletives such as there in English is to check nominative case; see Groat 1995 for a recent exposition of this view. The obligatory insertion of the expletive in SpecIP (that is, SpecTP in most recent theories) in constructions without a lexical nominative subject in SpecIP would then at least in part follow from Case theory: nominative must be checked
EXPLETIVES, SUBJECTS, AND TOPICS IN FINNISH
91
in SpecIP. In the case of Finnish there are extremely good reasons to reject this theory:22 (a) As we have seen, nominative case need not be checked in SpecIP; instead SpecIP, or more precisely SpecFP, may host a nonnominative category. Hence the position of sita is no indication that it is nominative (b) Sita does not control agreement. See, for instance, (48), where the finite verb is plural, agreeing with the lower subject. As discussed in the preceding section, only a nominative NP can control agreement. (c) The morphological form of sita indicates that it is not nominative. In fact, the nominative form se of the same pronoun is used as an expletive, too, in weather constructions (colloquially). (49)
Nyt se taas sataa. now it again rains
This form is impossible in the expletive constructions listed above. (50)
* Se leikkii lapsia pihalla. it plays children.PART in.yard
(51)
* Se vasytaan nykyaan helpommin kuin ennen. it get.tired.PASS nowadays easier than before
(52)
(*) Se vasyy nykyaan helpommin kuin ennen. it gets, tired nowadays easier than before
(53)
* Se ovat nama lapset jo oppineet uimaan. it have these children already learned to.read
Example (52) is grammatical if the pronoun is interpreted referentially, meaning 'It/he/she gets tired more easily than before.' If the function of the expletive in (45)-(48) is to check nominative case in SpecIP, why not use the nominative form of the pronoun? Rather, the choice of form indicates that the expletive crucially should not be nominative, checking nominative case in SpecIP. A nominative pronoun in that position apparently can be interpreted only referentially (shown most clearly by (52)); consequently it competes with the other arguments for the 6-roles assigned by the predicate, leading to a violation of the 0-criterion. The nonnominative partitive pronoun, on the other hand, can be interpreted nonreferentially, and hence does not impinge on 0-role assignment. We claim that sita as an expletive, in spite of its partitive form, is formally caseless: the form is partitive because partitive is the default oblique (i.e., nonnominative) case in Finnish.23 The reason why the nominative pronoun can be used in (49), we contend, is that in that construction no 9-role is assigned. Therefore, although the subject is nominative, it does not compete with any other argument for a 0-role. Effectively, it heads a vacuous chain (i.e., a chain with Case but no 0-role).24 In conclusion, there is virtually no empirical support for the hypothesis that expletive sita checks nominative Case, and that its obligatory character can be explained by Case theory. The fact that the expletive pronoun is morphologically partitive is thought-provoking against the background of the Partitive Hypothesis, according to which a
92
SUBJECTS, EXPLETIVES, AND THE EPP
postverbal argument in existential and ergative constructions has (abstract) partitive case; see Belletti 1988. Lasnik 1995 presents a version of the Partitive Hypothesis in which the English expletive there crucially takes a partitive nominal as its associate, serving as a host for this argument in LF. Lasnik does not explicitly propose that there itself is partitive, but such an analysis would in fact be fully consistent with his theory. Plausibly, if the expletive is in a chain with a partitive argument, it must itself be either partitive or caseless, since otherwise a case clash would ensue. Now in Finnish there is a common construction where the expletive co-occurs with a postverbal partitive argument, namely (45), the so-called Partitive Construction. However, there are other constructions where the expletive sita cooccurs with arguments which are not partitive. For instance, in (48) the 'associate' of the expletive (i.e., the argument whose place in the overt structure is in a sense occupied by sita) is nominative, and, as will be seen below, the associate may in principle have any case at all. 6.2. Argument positions in IP and the position of the expletive In this subsection we will argue for the hypotheses in (54). (54) a. b.
The expletive sita is always in SpecFP. If SpecFP is occupied by an expletive, a lexical argument (subject or nonsubject) can stay in the spec of any available category between F and VP.
If sita. is in SpecFP, it should be the case that it can be preceded by at most one constituent, which should have a contrast!ve reading. The following examples show that this is the case. (55) a.
b.
c.
d.
Pihalla sita leikkii lapsia. in.yard EXPL plays children.PART '!N THE YARD, it seems, there are children playing' Nykyaan sita vasytaan helpommin kuin ennen. nowadays EXPL get.tired.PASS easier than before THESE DAYS people get tired easier than before' Helpommin sita nykyaan vasyy kuin ennen. easier EXPL gets.tired nowadays than before 'I tell you, these days one gets tired MUCH EASIER than before' Nama lapset sita ovat jo oppineet uimaan. These children EXPL have already learned to. swim 'Look at these children, they have already learned to swim'
Almost any constituent can be moved to the left of the expletive, but—apart from those sentence adverbials that can adjoin to FP (or, in the case of certain adverbs, CP)—only one constituent, and that constituent always has a contrastive reading. Furthermore, if sita is in SpecFP, and if our theory of Finnish sentence structure in section 2 is basically correct, then sita should always immediately precede the head bearing subject agreement (or, in the case of passives, the head bearing the Faffix; see section 2), except if the head is itself moved to C. The following examples show that this is indeed the case. For expository purposes we use the same per-
EXPLETIVES, SUBJECTS, AND TOPICS IN FINNISH
93
son (3PL, ending -vAf) and same mood (conditional, ending -isi) in all the examples. We will notate the symbol of the relevant syntactic position (e.g., F, SpecCP, etc.) above the word in that position. The following examples illustrate the generalization that the expletive always immediately precedes the element (negation, auxiliary, or verb) that is inflected for subject agreement, whether or not that element is also inflected for tense.
(56)
F T Ptc Sita eivat [nama lapset]j olisi ikina oppineet t± uimaan. EXPL not. 3PL these children have.cOND ever learn. PTC to. swim 'These children would never have learned to swim' SpecCP F T Ptc b. [Nama lapset]j sita eivat ^ olisi ikina oppineet t± uimaan. these children EXPL not.3PL have.cOND ever learn.PTC to.swim These children, they would never have learned to swim' SpecCP F T Ptc c. Uimaanj sita eivat [nama lapset]j olisi ikina oppineet t| tj to.swim EXPL not.3PL these children have.cOND ever learn.PTC To SWIM, at least, these children would never have learned' F T Ptc d. * Sita [nama lapset] eivat olisi ikina oppineet uimaan. EXPL these children not.3PL have.cOND ever learn.PTC to.swim F T Ptc e. * [Nama lapset] eivat sita olisi ikina oppineet uimaan. these children not.SpL EXPL have.cOND ever learn.PTC to.swim f. SpecCP F T Ptc Uimaan sita eivat olisi [nama lapset] ikina oppineet. to.swim EXPL not.3PL have.cOND these children ever learn.PTC g. F T P t c Sita eivat olisi [nama lapset] ikina oppineet uimaan. EXPL not.3PL have.cOND these children ever learn.PTC to.swim F T h. ? Sita eivat oppisi [nama lapset] uimaan. EXPL not. 3PL learn. COND these children to. swim F T i. Uimaan sita eivat [nama lapset] oppisi. to. swim EXPL not. 3PL these children learn. COND To SWIM, at least, these children would not learn' F T j. ? Uimaan sita eivat oppisi [nama lapset]. to.swim EXPL not.3pL learn.CQND these children F Ptc k. Sita'olisivat [nama lapset] oppineet uimaan. EXPL have.COND.3PL these children learn.PTC to.swim These children would have learned to swim'
a.
94
SUBJECTS, EXPLETIVES, AND THE EPP
1.
SpecCP F Ptc [Nama lapset] sita olisivat oppineet uimaan. these children EXPL have.COND.3pL learn.PTC to.swim THESE CHiLDREN/These children would have learned to swim'
C F Ptc Olisivatj sita t-v [nama lapset] oppineet uimaan. have.COND.3PL EXPL these children learn.PTC to.swim 'These children WOULD have learned to swim' SpecCP F Ptc n. * Olisivatj sita fj oppineet [nama lapset] uimaan. have.coND.3pL EXPL learn.PTC these children to.swim
m.
If the expletive is itself unambiguously in the contrast position (as in (56d)) or below F (as in (56e)), the sentence is ungrammatical. The conclusion is, almost inevitably, that the expletive can only be in SpecFP. In section 3 we mentioned that an expletive cannot co-occur with a referential null subject (as observed by Auli Hakulinen (1975), who gave the following example): (57)
* Sita uskon vallankumoukseen. EXPL believe. ISG revolution. ILL 'I believe in revolution'
This is explained if subject pro and the expletive sita are both confined to SpecFP. In that case pro is impossible in (57), and the sentence is ruled out by the ©-criterion. That subject pro is confined to SpecFP is a consequence of the identification requirement on small pro. The features of pro can be identified only in a spec-head relation with F, since F is the head hosting the required AgrS features. The following is an important exception to the generalization that the expletive always precedes the head bearing subject agreement: When SpecCP or C is filled, V-movement, otherwise obligatory, becomes optional (see Vilkuna 1989, 1995, Holmberg 1997, 2000a). As a result such sentences are often verb-final. This is exemplified in (58). (58) a.
Nahtavasti Pekka matkustaa huomenna Pariisiin. apparently Pekka travel. 3SG tomorrow to. Paris b. * Nahtavasti Pekka huomenna Pariisiin matkustaa. c.
Huomennako Pekka Pariisiin matkustaa? tomorrow. Q Pekka to. Paris travel. 3SG 'Is it tomorrow that Pekka is going to Paris?'
The initial adverb, which is adjoined to FP, is added in (58a-b) in order to prevent an analysis where the subject is moved to SpecCP, which would make possible leaving the verb in situ. According to Holmberg 1997, the order in (58c) is the result of nonmovement of the main verb, made possible by filling of C; although the initial adverbial is by hypothesis in SpecCP, C is filled in this construction by a focus feature realized (or checked) by the question affix -kO. Not only the main verb, but also the negation and the auxiliary, can be left in situ when C is thus filled; see Holmberg 1997 and Holmberg 2000a for two quite different ways of explain-
EXPLETIVES, SUBJECTS, AND TOPICS IN FINNISH
95
ing this phenomenon. As a result, when C is filled, we may find material intervening between the expletive sita and the head bearing subject agreement. We have already seen one example in this section, namely (56c). The following is another example: (59)
Milloin sita nama lapset olisivat uimaan oppineet? when EXPL these children would, have swim learned 'When would these children have learned to swim?'
This is significant in the present context, since it shows that the expletive need not be in the spec of an overt head bearing subject agreement. In other words, there is no PF-condition which requires a spec-head relation or an adjacency relation between sita and the head containing AgrS. However, when the expletive does immediately precede a head in PF, the head is always F, containing AgrS (or in passives, the F-affix). Therefore we maintain that sita is is in SpecFP in the verb-final constructions, too, except that F in those cases is phonetically empty. There is one case where the expletive sita is (arguably) in a position higher than SpecFP. In some cases sita may serve as a carrier of the question particle -kO or the 'focus-particle' -hAn. (60) a.
b.
Sitako ovat teidan lapset jo kaikki kayneet uimassa? EXPL. Q have your children already all been swimming 'Have your children already all been swimming?' Sitahan ei nykyaan puhuta vakoilusta. EXPL.PRTCL not nowadays talk.PASS espionage.ABL We don't talk about espionage these days, do we?
Yes-no questions in Finnish involve a question affix -kO, affixed onto a constituent moved to SpecCP, which thereby becomes the focus of the question. In the unmarked case the finite verb is fronted, to serve as host of the affix. In that case the event itself is, in a sense, the question focus. However, if the sentence is headed by an expletive, the expletive may serve as the host, in which case the reading is equally unmarked: the event is focus. Thus, while (61a) can be pragmatically equivalent to (60a), (61b) and (61c) cannot; the latter two have focus on the subject and the complement, respectively. (61) a. b. c.
Ovatko sita teidan lapset jo kaikki kayneet uimassa? have. Q EXPL your children already all been swimming Teidan lapsetko sita ovat jo kaikki kayneet uimassa? your children.Q EXPL have already all been swimming Uimassako sita teidan lapset ovat jo kaikki kayneet? swimming.Q EXPL your children have already all been
In (66) fronting of sita, and in (6la) fronting of the finite verb, is triggered by (some version of) the 'Stranded Affix Filter' of Lasnik 1981. We conclude that the expletive sita occupies SpecFP, the topic position, in constructions which do not have an overt topic. Assuming, as we do, that the function of the expletive is to check the EPP feature, we are led to conclude that the EPP feature is in F and not, for instance, in T. The EPP feature in F is checked by an argument which is usually, but not always, the subject, or by the expletive sita. In this
96
SUBJECTS, EXPLETIVES, AND THE EPP
sense sita is not an expletive subject but an expletive topic. It satisfies the requirement that a given sentence-initial spec-position, namely SpecFP, be filled in sentences where, for whatever reasons, that position is not filled by an argument. That position is not a privileged position for the subject, though, but for the topic. This claim will be further substantiated in the following section. 6.3 More on the MSC As we have seen, there is a construction where the expletive sita co-occurs with a VP-external argument lower in the structure. In the examples discussed so far the lower argument is the subject, so that the structure looks very much like the Multiple Subject Constructions found in some Germanic languages, for example, Icelandic. (62) Pad hafa margir studentar lesid pessa bok. Icelandic EXPL have many students read this book Bobaljik and Jonas (1996) have argued that the expletive in this construction is in SpecAgrS and the lexical subject is in SpecTP. Crucially, Bobaljik and Jonas show that the lexical subject is not in SpecVP, but has raised to a higher spec-position in the IP-domain, which they identify as SpecTP. Following Chomsky 1993, Bobaljik and Jonas assume that T checks nominative Case. Hence the nominative subject has to enter into a checking relation with T at some point in the derivation. In some languages (for instance, English), T raises to AgrS, and the subject raises to the spec of the T-AgrS complex checking nominative Case (all in overt syntax). In other languages (for instance, Icelandic), the subject moves via SpecTP to SpecAgrS, checking nominative on the way. In Icelandic only the subject can move to SpecTP in the MSC; Icelandic is subject-prominent, in spite of having subject agreement which is rich enough to check nominative case (according to Holmberg and Platzack 1995).25 What about Finnish, a topic-prominent language with a MSC? It turns out that the lexical argument in the Finnish MSC need not be the subject—which is to say that MSC is actually a misnomer. (63) a.
b.
c.
Sita en niita lapsia ole nahnyt minakaan. EXPL not.lsc those kids have seen I.even 'Even I haven't seen those kids' Sita ei tallaista kirjaa olisi voinut kirjoittaa kukaan EXPL not such book would.have been.able.to write anyone muu kuin Graham Greene. else but G. G. 'No one but Graham Greene could have written a book like this' Nyt sita ovat taman kirjan lukeneet kaikki neljannen luokan now EXPL have this book read all fourth grade oppilaat. students 'Now all the fourth grade students have read this book'
In these constructions the object is raised out of VP while the nominative subject remains in VP. The expletive is in its usual position, that is, SpecFP, immediately
EXPLETIVES, SUBJECTS, AND TOPICS IN FINNISH
97
preceding the head bearing subject agreement (the negation or auxiliary). That the position of the expletive is indeed SpecFP can be established by applying the same tests as above in (56). The object is thus in a lower spec-position in the IP-domain. The verb agrees with the nominative subject, not with the object. Let us assume, provisionally, that the position of the preverbal argument in the Finnish MSC is SpecTP. In that case we have to accept that SpecTP, just like SpecFP, is not a privileged position for the subject. Some languages have the option of moving other categories than the subject to SpecTP. The question is, which feature triggers movement of the object to SpecTP in (63), and the subject in (61), on the natural assumption that the trigger in both cases is the same? The nominative case-feature of T (assuming, for the sake of argument, that it is in T) cannot attract a nonnominative object. The EPP feature, which attracts a [-Foe] argument, cannot be responsible either, for two reasons: (a) it is satisfied in these constructions by the expletive in SpecFP, and (b) the lexical subject in (61) and the object in (63) are not [-Foe]. A [-Foe] argument is referential and specific, refers to an entity previously introduced in the discourse, and is not information focus. Hence the most typical linguistic expression defined by the feature [-Foe] is an unstressed definite pronoun. But the argument in SpecTP in the Finnish MSC cannot be an unstressed definite pronoun. A pronoun in that position either has to be stressed or (preferably) construed with a focus particle.26 (64) a.
b.
Sita olen *mina/MiNA /minakin/jopaminajo monta EXPL have. ISG I II(stress)II.too I even I already many kertaa ka'ynyt Pariisissa. times been in. Paris Sita ovat *sen/??SEN/senkin jo lukeneet kaikki neljannen EXPL have it I IT lit.too already read all fourth luokan oppilaat. grade students
The fronted argument need also not be specific. (65)
Sita voi kuka tahansa heti tulla puhumaan meille. EXPL can who ever at. once come talk to. us 'Anyone can come at once and talk to us'
The Finnish MSC differs markedly from the Icelandic MSC in that there is no definiteness effect in the Finnish MSC of the sort exhibited by the Icelandic MSC (see Vangsnes 1995, this volume); in Finnish the argument in SpecTP can be definite. However, like the corresponding argument in Icelandic it must be focused, either by contrastive focus or by virtue of being information focus, with or without focus particles. So we are talking about movement to a preverbal focus position. This is reminiscent of movement to the preverbal focus position in Hungarian and certain other languages (cf. E. Kiss 1995), although there are obvious differences; for one thing this movement is optional in Finnish. Consider now the question of landing site for the argument moved in the MSC. We have provisionally assumed, following Bobaljik and Jonas's (1996) analysis of
98
SUBJECTS, EXPLETIVES, AND THE EPP
the Icelandic MSC, that the argument lands in SpecTP. In fact the position of the argument relative to other constituents in the IP-domain is quite free. Compare, for instance, (66a-d).
(66) a.
b.
c.
d.
e.
F T Ptc Sita eivat nama lapset olisi ikina oppineet uimaan. EXPL not.3pL these children have.coND ever learn.PTC to.swim These children would never have learned to swim' F T P t c Sita eivat olisi nama lapset ikina oppineet uimaan. EXPL not.3PL have.CQND these children ever learn.PTC to.swim F T Ptc Sita eivat olisi ikina nama lapset oppineet uimaan. EXPL not.3PL have.coND ever these children learn.PTC to.swim F T Ptc Sita eivat olisi oppineet nama lapset ikina uimaan. EXPL not.3PL have.COND learn.PTC these children ever to.swim F T Ptc Sita eivat olisi oppineet ikina nama lapset uimaan. EXPL not.3PL have.COND learn.PTC ever these children to.swim
Crucially, in each of these positions the lexical argument must be focus: a nonfocused argument is excluded. But apparently there is no structurally defined focus position to the right of F in Finnish; instead there appears to be a focus domain, stretching from F down to the bottom of VP. That is to say, the focus domain is not VP (as we provisionally assume earlier), but TP, F demarcating what we might call the topic domain and C, as before, the operator domain. It looks like the distribution of focused arguments inside the focus domain is essentially free. As long as the EPP feature and the nominative Case feature in F are checked, the former by movement to SpecFP or merge of an expletive, and the latter by head-adjunction to F, and as long as the order of the functional heads (F-Neg-T-Aux-Prc-Pass) is respected, as well as the order of adverbs (cf. Holmberg et al. 1993), the positioning of focused arguments inside TP is free.27 What is the exact structural position of the argument in each case? The analysis of the Finnish finite clause given in (4) is rich enough to accommodate (66a-c), assuming that every head has a spec-position available for a focused argument. For (66d) we would have to postulate one more head between Ptc and V. This could be v, as in Chomsky 1995, or the head Act/Pass in Holmberg and Platzack 1995. We leave open the exact analysis of the constructions in (66); the main point is that the position of the focused preverbal argument relative to the other preverbal constituents is essentially free between F and V.
7. The MSC and multiple specifiers Chomsky (1995: ch. 4) argues that a head can have as many specifiers as it has features licensing them: the so-called Multiple Spec Hypothesis. In this way Chomsky can abolish Agr as a separate head projecting its own phrase. He notes that the Icelandic MSC looks like a case where multiple specs are not allowed, since the
EXPLETIVES, SUBJECTS, AND TOPICS IN FINNISH
99
two subjects are always separated by a head (the finite verb), indicating that they must each be licensed by a separate head—where, furthermore, the higher head is arguably AgrS. Chomsky suggests that the head movement to the position between the two subjects in the Icelandic MSC is an effect of the V2 condition, which, he speculates, is a PF condition. In the Finnish MSC, too, a head always intervenes between the two 'subjects' (i.e., the expletive and the preverbal lexical argument), where, furthermore, the head always bears the finite inflection, either subject agreement or, in passives, the corresponding invariant finite inflection, but not necessarily Tense or any other verbal inflection. Finnish is not a V2 language (for instance, w/z-movement does not induce V-movement to C), and there is no particular reason to think that the obligatory movement of the head bearing the finite inflection to the position right-adjacent to the higher subject is a PF-rule. The finite head does not move to the position right-adjacent to the higher specifier in the MSC because it is an MSC, but because it always moves to that position (except in the case discussed in section 6.2, where the otherwise obligatorily overt movement of the finite head to F is rendered optional, conditioned by the contents of C). Furthermore, if we are right, the two specifiers are in distinct syntactic domains: the higher one is in the topic-domain, outside the predicate phrase (i.e., TP), while the lower one is always inside the predicate phrase, and is therefore [+Foc]-marked. Given that specifiers of the same head are necessarily in the same syntactic domain, the two specifiers in the MSC cannot be specifiers of the same head. One question which we have not touched on yet is how agreement is ensured between the finite verb and the subject when the subject does not itself move to SpecFP. Following Chomsky 1995: ch. 4, we assume that the ^-features of the subject move covertly (i.e., without pied-piping the phonological and other features of the subject), adjoining to F (the binding facts mentioned in note 17 can be construed as evidence in favor of this analysis), entering a checking relation with the features of the finite inflection (leaving open the precise character of this relation). Why, however, does the finite verb agree with the subject in that case? Why does it not agree with the object topic moved to SpecFP? In other words, why is (67) illformed? (67)
* Nama kirjat ovat lukeneet minakin. these books have.3PL read.PL I.too.NOM
We hypothesize that the reason is that (a) Case and (^-features are part of the same, indivisible feature complex in Finnish (and probably in Germanic, too), and (b) F has a nominative-checking feature. Therefore F will never attract a nonnominative verb, and the finite verb will therefore never agree with a nonnominative argument. There are topic-prominent languages where the finite verb regularly agrees with the topic, as is apparently the case in some Bantu languages (cf. Jang 1997, Bresnan 1994). We suggest that agreement is a different sort of category in these languages, in particular that it is dissociated from Case. Many questions are still left unanswered. For instance, why don't other languages, such as English, French, and Swedish, have MSCs? It is tempting to see the possibility of MSCs in Finnish as a consequence of the visible Agr-T split. See
100
SUBJECTS, EXPLETIVES, AND THE EPP
Thrainsson 1996 for arguments to the effect that some languages but not others have I split in Agr and T, and that MSCs can be expected only in languages where I is split. One subject would then be licensed by AgrS and the other by T. In Finnish the higher subject does bear a special relation to subject agreement, as we have seen, although the relation is indirect. On the other hand, there seems to be no obvious connection between T and the lower subject; we have shown that the position of the lower subject is essentially free, in the predicate-internal IP-domain.
Notes Thanks to Maria Vilkuna, Anne Vainikka, Trond Trosterud, Tarald Taraldsen, Peter Svenonius, and Shigeru Miyagawa for comments and suggestions. Thanks also to the audiences at the EPP workshop in Troms0 in the spring of 1997 and all other workshops and seminars where parts of this work have been presented. 1. The following example illustrates the finite inflection in the passive; note that the form of the inflection does not vary according to the number, case, or any other property of the arguments of the sentence. (i)
Kirja
/kirjat
/kirjoja
tila-t-i-in
Englannista.
book.NOM. solbook. NOM. PL/book. PART. PL order-PASS-PAST-F from. England "The book/the books/books were, ordered from England' 2. Uriagereka (1995) postulates a category in Romance languages which he calls F, and which seems to correspond quite closely to the Finnish category we call F. Although the labels are only accidentally identical, Uriagereka's F being short for 'functional,' this encourages us to use this label. 3. The form of the expletive in (2a) is sitd, which is the partitive form of the third-person singular nonhuman pronoun, while the form in the weather expression (8) is se, which is the nominative form of the same pronoun, or sitd. The extraposition pronoun can only be se. In the text below we will argue that the form sitd is a default, syntactically caseless form, while se is nominative. 4. Furthermore, Finnish has pronouns which are clearly specified for number but not person, namely, the pronouns in the MSC which we call the Split Subject Construction, exemplified in (3b). The contrast between (i) (= 3b) and (ii) shows that the pronoun is marked for number. The contrast between (ii) and (iii) shows that it is not marked for person. (i)
Ne ovat ministerit ostaneet uusia autoja. they have.3PL ministers bought new cars (ii) Se on Jussi kaynyt monta kertaa Pariisissa. it have.3sc Jussi been many times in. Paris (iii) Se olen minakin kaynyt monta kertaa Pariisissa. it have. 1 so me.too been many times in.Paris This pronoun does not seem to have a null counterpart. (iv) (*)On Jussi kaynyt monta kertaa Pariisissa.
has Jussi been many times in.Paris This expression is grammatical only with the verb focus reading (roughly "Jussi HAS been . . .") derived by V-movement to C; see text below. 5. If there is an Agent, it is always the highest argument in the VP; that much is uncontroversial. With regard to the rest of the thematic hierarchy and its syntactic projection there is more disagreement (cf. Speas 1990). The following is a more cautious definition of subject-prominence: a language where the externalized argument in an active sentence containing an Agent can only be the Agent is subjectprominent. 6. Obviously, care must be taken when applying this crude test. For instance, this particular order may be ruled out for independent reasons even in a topic-prominent language (for instance, if it is strictly verb-final). Or the order may be coincidentally permitted in a subject-prominent language, but with another interpretation, as in the Swedish sentence (i).
EXPLETIVES, SUBJECTS, AND TOPICS IN FINNISH
101
(i)
Den bar boken skrev Graham Greene nar ban var tjugofem ar. Swedish this here book wrote Graham Greene when he was twenty-five years In this case the order is a result of object movement to SpecCP, V-movement to C (Swedish being a V2 language), and subject movement to SpecIP. On the most natural reading Graham Greene does not represent new information here, but part of the old information. 7. Contrastiveness is a different feature; an argument can be [-Foe] and yet be contrastive. In the construction (i) (i) John, I like (but I can't stand his wife). John is (normally) [-Foe], that is, part of the presupposition, but contrastive. On the syntax of contrastiveness in Finnish, see Vilkuna 1995 and the text below. 8. A logical possibility is that there is an unmarked value of [Foe] in addition to + and -. We do not, however, make use of this possibility. An argument is either [+Foc] or [-Foe], where in the latter case you have to do something with it in the syntax to eliminate the [-Foe] feature. In this sense [+Foc] represents the unmarked case in the theory assumed here. 9. Cf. Miyagawa's 1998 'feature uniformity.' 10. There are also languages where all w/z-movement is overt: the Slavic languages. We do not know whether there are languages where all [-Foe]-movement is overt. We predict that there could be, though. 11. Holmberg and Platzack (1995) argue that the feature is actually located in C in some languages (notably V2 languages), but in I in other languages. 12. Although even English has a somewhat marginal nonsubject topic construction, namely, Locative Inversion: Into the bar strolled three drunken sailors. 13. But cf. Martin 1997. The presence of there would follow if the matrix verb has a Case-feature which is uninterpretable, and hence has to be checked. The question is whether this holds true of all predicates selecting a sentential complement in the manner of believe. Martin (1997) argues that it does. 14. This word order is fine if another predicate is added. For instance: (i)
Vetta sataa kaatamalla. water rains by.pouring 'It's pouring down rain' 15. Obviously, admitting the possibility that the EPP can be optional entails a weakening of the theory. An anonymous referee suggests that the verb-initial sentences represent the case when T/MP itself (or possibly some lower sentential projection) is marked [-Foe], and consequently moves to SpecFP. Semantically it makes sense: those sentences are not about somebody or something, but about the event or state itself, depicted by T/MP. Attractive though the idea is, we cannot think of any empirical support for it. The prediction is that categories which are normally situated between T/MP and FP should appear in sentence-final position, in verb-initial impersonal sentences. An obvious candidate is the negation. As shown by the following examples, the prediction fails. (i)
* Ole ilmennyt ongelmiaei. have appeared problems NEC (ii) Ei ole ilmennyt ongelmia NEC have appeared problems (iii) * Tullut kiketta ei. come haste NEG (iv) Ei tullut kiketta. NEC come haste The negation is invariably the initial constituent in such cases. Likewise modal adverbs, which in the unmarked case appear between the finite verb or auxiliary and the nonfinite verb, that is, between F and T/M, do not appear sentence-finally in verb-initial sentences (except with a clear prosodic break). (iv)
* On sattunut onnettomuus kai. has occurred accident presumably (v) On kai sattunut onnettomuus. has presumably occurred accident As illustrated by (v), such adverbs occur between the finite auxiliary and the nonfinite portion of the sentence in verb-initial sentences, too, consistent with the analysis that those sentences are derived by
102
SUBJECTS, EXPLETIVES, AND THE EPP
the same movements as personal finite sentences, including movement of the finite verb, auxiliary, or negation to F, but without movement of an argument to SpecFP, or insertion of an expletive in SpecFP. 16. This is a reflection of the generalization that Finnish allows only one fronted topic per sentence. 17. In fact it is hard to demonstrate that the subject ever binds anaphora from its derived position. A sentence such as (i) would appear to require binding of the anaphoric possessive suffix from the derived subject position. 000000000000000000000000000000000000000000000000000000000 Pekka not knowing. POSS. 3 has ever seen live cow 'As far as he knows, Pekka has never seen a live cow' But there is another possibility: that the possessive anaphor is bound not by the subject itself, but by AgrS. In (ii) this is, indeed, the only possibility: (ii)
Itseaan aanesti tieta'akseen vain Jussi. onMmself voted knowing.POSS.3 only Jussi 'As far as he (i.e. Jussi) knows, only Jussi voted for himself Here the topic is itself an anaphor, and is therefore reconstructed. Hence the only binder of the possessive suffix is AgrS (i.e., the inflection on the verb). Yet another possibility is that the possessive anaphor in the participial adverbial is bound by a phrase-internal pro. However, in that case this pro itself falls under Principle A; cf. Trosterud 1993. 18. Another test for A/A-bar-hood familiar from the literature is whether the category in question licenses a parasitic gap. Unfortunately this test cannot be used in Finnish since Finnish has null objects, and there is no simple way of distinguishing a parasitic gap from a null object. 19. This is so on the assumption that nominative case is checked by F. Alternatively it is checked by T. Subject movement to SpecFP is then formally distinct from object movement to SpecFP in that the former is movement to a <|)-feature-checking position. The explanation of the binding and reconstruction effects in the text below still holds. 20. This follows if Binding Theory sees only ^-features and case is a <j)-feature, and hence always moves together with the ^-features. 21. Guilfoyle et al. observe a similar cluster of properties associated with the topic in a group of related Austronesian languages (Malagasy, Cebuano, Tagalog, and Bahasa Indonesia and Malaysia), notorious for their topic-orientedness. They argue that the topic NP, whether subject or object, is in SpecIP in these languages. They observe that a topic can always control Quantifier Float, but only a subject can bind anaphora. They assume that while Quantifier Float is controlled from SpecIP (i.e., it is determined by the S-structure configuration), anaphor binding is controlled strictly from SpecVP. In other words, no movement, not even A-movement, will affect binding relations. There are classical arguments against the latter hypothesis (as a universal), though, for instance (i), where binding presupposes raising. (i) They seem to each other [ t to have gone mad]. 22. If the AgrS features in English are not capable of checking nominative case, then presumably there checks nominative case in English existential and other constructions with there. That is to say, we do not agree with Chomsky (1995: ch. 4) that English there is a pure expletive. 23. See Vainikka 1993 for arguments that partitive is the default object case. 24. This does not rule out use of sitd in weather constructions, correctly, since (i) is a viable alternative to (49): (i)
Nyt sita taas sataa. now EXPL again rains 25. Icelandic is not 'nominative-prominent,' though, since a nonnominative argument may move to SpecTP and SpecFP, as long as it is a subject in the sense of being the highest argument; cf. Holmberg and Platzack 1995. 26. Finnish unstressed pronouns do not have any special distributional characteristics, such as occurring in designated clitic or weak pronoun positions. 27. Cf. Vilkuna's (1989) claim that while SpecCP (Vilkuna's K-position) and SpecIP (Vilkuna's Tposition) are fixed, constituent order in the VP-domain is free, where Vilkuna's VP-domain corresponds at least in part to our TP-domain.
EXPLETIVES, SUBJECTS, AND TOPICS IN FINNISH
103
The position of nonfocused arguments inside the focus domain is not free, however. The generalization concerning arguments marked [-Foe] is that they must either move to SpecFP or be governed by (be a complement of) the main verb or some other lexical head (such as an adposition). This entails that a subject marked [-Foe] must always move to SpecFP, while an object marked [-Foe] has the option of staying in situ; cf. Holmberg (1999).
References Alexiadou, Artemis, & Elena Anagnostopoulou. 1998. Parametrizing AGR: Word order, Vmovement and EPP-checking. Natural Language & Linguistic Theory 16,491-539. Belletti, Adriana. 1988. The case of unaccusatives. Linguistic Inquiry 19, 1-34. Bennis, Hans. 1986. Gaps and Dummies. Foris, Dordrecht. Bobaljik, Jonathan. 1995. Morphosyntax: The Syntax of Verbal Inflection. Ph.D. dissertation, MIT. Bobaljik, Jonathan, & Dianne Jonas. 1996. Subject positions and the roles of TP. Linguistic Inquiry 27, 195-236. Bresnan, Joan. 1994. Locative Inversion and the architecture of Universal Grammar. Language 70, 72-131. Chomsky, Noam. 1993. A Minimalist program for linguistic theory. In Ken Hale & Samuel J. Keyser (eds.), The View from Building 20, 1-52. MIT Press, Cambridge, Mass. Chomsky, Noam. 1995. The Minimalist Program. MIT Press, Cambridge, Mass. Diesing, Molly. 1990. Verb movement and the subject position in Yiddish. Natural Language & Linguistic Theory 8, 41-80. Diesing, Molly. 1992. Indefinites. MIT Press, Cambridge, Mass. Groat, Erich. 1995. English expletives: A Minimalist approach. Linguistic Inquiry 26, 354-365. Groat, Erich, & John O'Neil. 1996. Spell-out at the LF interface. In Werner Abraham et al. (eds.), Minimal Ideas. John Benjamins, Amsterdam. Guilfoyle, Eithne, Henrietta Hung, & Lisa Travis. 1992. Spec of IP and spec of VP: Two subjects in Austronesian languages. Natural Language & Linguistic Theory 10, 345-374. Hakulinen, A. 1976. Suomen sitd: Pragmatiikan heijastuma syntaksissa. Sananjalka 17, 25^1. Holmberg, Anders. 1997. Word order variation in some European SVO languages: A parametric approach. In Anna Siewierska (ed.), Constituent Order in the Languages of Europe, 553-598. Mouton de Gruyter, Berlin. Holmberg, Anders. 1999. Remarks on Holmberg's Generalization. Studia Linguistica 5 1-39. Holmberg, Anders. 2000a. Deriving OV order in Finnish. In Peter Svenonius (ed.), The Derivation ofVO and OV, 123-152. John Benjamins, Amsterdam. Holmberg, Anders. 2000b. Scandinavian Stylistic Fronting: How any category can become an expletive. Linguistic Inquiry 31, 445^84. Holmberg, Anders, & Urpo Nikanne. 1993. Introduction. In Anders Holmberg & Urp Nikanne (eds.), Case and Other Functional Categories in Finnish Syntax, 1-20. Mouton de Gruyter, Berlin. Holmberg, Anders, Urpo Nikanne, Irmeli Oraviita, Hannu Reime, & Trond Trosterud. 1993 The structure of INFL and the finite clause in Finnish. In Anders Holmberg & Urpo Nikanne (eds.), Case and Other Functional Categories in Finnish Syntax, 177-206. Mouton de Gruyter, Berlin.
104
SUBJECTS, EXPLETIVES, AND THE EPP
Holmberg, Anders, & Christer Platzack. 1995. The Role of Inflection in Scandinavian Syntax, Oxford University Press, New York. Jang, Youngjun. 1997. Strength of T and the object position. GLOW Newsletter 38, 71-72. Kiss, Katalin E. 1995. Introduction. In Katalin E. Kiss (ed.), Discourse Configurational Languages, 3-27. Oxford University Press, New York. Kiss, Katalin E. 1997. Discourse-configurationality in the languages of Europe. In Anna Siewierska (ed.), Constituent Order in the Languages of Europe, 681-727. Mouton de Gruyter, Berlin. Lasnik, Howard. 1981. Restricting the theory of transformations: A case study. In Norbert Hornstein & David Lightfoot (eds.), Explanation in Linguistics, 152-173. Longman, London/New York. Lasnik, Howard. 1995. Case and expletives revisited: On Greed and other human failings. Linguistic Inquiry 26, 615-633. Li, Charles, & Sarah Thompson. 1976. Subject and topic: A new typology of language. In Charles Li (ed.), Subject and Topic, 457-489. Academic Press, New York. Martin, Roger. 1997. Uninterpretable features. Research Report 1. Graduate School of Language Sciences, Kanda University of International Studies. Mitchell, Erika. 1991. Evidence from Finnish for Pollock's theory of IP. Linguistic Inquiry 22, 373-379. Miyagawa, Shigeru. 1998. EPP and w/z-questions. Ms., Department of Linguistics and Philosophy, MIT. Nikanne, Urpo. 1994. On movement to the Spec(IP) position in Finnish. In Norbert Corver & Henk van Riemsdijk (eds.), Studies on Scrambling, 431^57. Mouton de Gruyter, Berlin. Nilsen, 0ystein. 1997. Adverbs and A-shift. Working Papers in Scandinavian Syntax 59, 1-31. Rizzi, Luigi. 1982. Issues in Italian Syntax. Foris, Dordrecht. Rizzi, Luigi. 1986. Null object and the theory of pro. Linguistic Inquiry 17, 501-558. Rizzi, Luigi. 1990. Relativized Minimality. MIT Press, Cambridge, Mass. Rizzi, Luigi. 1991. Proper head government and the definition of A-positions. GLOW Newsletter 26, 46^7. Rognvaldsson, Eirikur, & Hoskuldur Thrdinsson. 1990. On Icelandic word order once more. In Joan Maling & Annie Zaenen (eds.), Syntax and Semantics 24: Modem Icelandic Syntax, 3-40. Academic Press, New York. Shlonsky, Ur. 1988. Rich INFL and the licensing of pro. In Anna Cardinaletti, Guglielmo Cinque, & Giuliana Giusti (eds.), Constituent Structure, 331-349. Foris, Dordrecht. Speas, Margaret. 1990. Phrase Structure in Natural Language. Kluwer, Dordrecht. Sportiche, Dominique. 1988. A theory of floating quantifiers and its corollaries for constituent structure. Linguistic Inquiry 19, 425-449. Thrainsson, Hoskuldur. 1996. On the (non-)universality of functional categories. In Werner Abraham et al. (eds.), Minimal Ideas, 253-281. John Benjamins, Amsterdam. Trosterud, Trond. 1993. Anaphors and binding domains in Finnish. In Anders Holmberg & Urpo Nikanne (eds.), Case and Other Functional Categories in Finnish Syntax, 225-243. Mouton de Gruyter, Berlin. Uriagereka, Juan. 1995. Aspects of the syntax of clitic placement in Western Romance. Linguistic Inquiry 26, 79-123. Vainikka, Anne. 1989. Deriving Syntactic Representations in Finnish. Ph.D. dissertation, University of Massachusetts, Amherst.
EXPLETIVES, SUBJECTS, AND TOPICS IN FINNISH
105
Vainikka, Anne. 1993. The three structural cases in Finnish. In Anders Holmberg & Urpo Nikanne (eds.), Case and Other Functional Categories in Finnish Syntax, 129-159. Mouton de Gruyter, Berlin. Vallduvf, Enric, & Elisabet Engdahl. 1996. The linguistic realization of information packaging. Linguistics 34, 459-519. Vangsnes, 0ystein. 1995. Referentiality and argument positions in Icelandic. Working Papers in Scandinavian Syntax 55, 89-109. Vilkuna, Maria. 1989. Free Word Order in Finnish: Its Syntax and Discourse Functions. Finnish Literature Society, Helsinki. Vilkuna, Maria. 1995. Discourse configurationality in Finnish. In Katalin E. Kiss (ed.), Discourse Configurational Languages, 244-268. Oxford University Press, New York.
This page intentionally left blank
5 The EPP in a Topic-Prominent Language Katalin E. Kiss
1. Introduction The standard version of the EPP, requiring essentially that the subject position be filled in every sentence, is clearly violated in such so-called free word order languages as Hungarian, in which the VP-external position(s) can be filled by any argument of the V or can also remain empty. In an attempt to maintain the universality of the EPP, Alexiadou and Anagnostopoulou (1998) claim that in 'free word order' languages displaying rich verbal inflection and pro drop, the EPP is actually automatically satisfied by V movement to AgrS. This chapter will argue that the EPP, if slightly reformulated, expresses generalizations which make intuitively contentful predictions about possible and nonexistent sentence structures also in the 'free word order' language type represented by Hungarian. Section 2 presents certain basic facts of Hungarian contradicting the EPP. Section 3 discusses the possibility of interpreting the EPP as the requirement that sentences instantiate a predication relation. This proposal will also bear on the analysis of the there is/there are construction in English. Section 4 examines the possibility of interpreting the EPP as a constraint on the argument structure of predicates, requiring that the most prominent argument of a predicate be realized as a subject.
2. The EPP as a predication requirement 2.1. Subject positions in Hungarian In his formulation of the EPP, Chomsky (1995:55) states that SpecIP is obligatorily realized (perhaps by an empty category) or, in a different terminology, I has a strong D-feature. If the EPP universally holds, then the constituent which agrees with I, the grammatical subject, is expected to appear invariably in SpecIP; or if it is barred from SpecIP for some reason, or if the verb has no subject argument,
107
108
SUBJECTS, EXPLETIVES, AND THE EPP
SpecIP is expected to be filled by an expletive. Languages like Hungarian, however, apparently do not bear out the predictions of the EPP. The problem is that a subject can stand in any sentence position, and at the same time a non-pre-VP subject is not accompanied by a visible expletive, and there is no evidence of an expletive pro appearing in SpecIP, either. It is not the case that one and the same subject can occupy any position in the Hungarian sentence. The position of a subject depends on its referential properties and on its operator feature, if any.1 This is established in points (i-v). (i) A determinerless and quantifierless, nonreferential subject is located in the VP, in a VP-initial, immediately preverbal position. A determinerless and quantifierless, nonreferential (i.e., semantically incorporated) subject is allowed only by intransitive presentative verbs, which assert the existence, or availability, or coming into being of their subject; hence it is always of a theme function (as shown by Szabolcsi 1986). For example, (1)
a.
Vendeg erkezett. guest arrived 'Some guest(s) arrived'
In the major part of the syntactic literature on Hungarian the position of a bare nominal subject is analyzed as left-adjoined to the V (cf. Ackerman 1984, Horvath 1986, Szabolcsi 1986, Brody 1990, and E. Kiss 1998a), although there is no agreement on whether the preverbal bare nominal itself is a head or a phrase, and whether it is base-generated adjoined to the V, or is moved there by incorporation/head movement or by phrase adjunction. The claim that it is inside the VP can be proven by showing that it must follow adverbials of manner, which are assumed to be left-adjoined to the VP. (2)
a.
Fokozatosan [Vp to keletkezett a melyedesben] gradually lake arose the depression, in 'Gradually a lake arose in the depression' b. * To fokozatosan keletkezett a melyedesben.2 lake gradually arose the depression.in
It is further evidence of the VP-internal position of the bare nominal subject that it necessarily follows the focus and the negative particle, which are operators preceding the VP (cf. Brody 1990, 1995, E. Kiss 1998a). In the case of negation and focus the bare nominal subject is realized in postverbal position—presumably because the V has moved forward across the subject into the F or Neg head, respectively (cf. Brody 1990): (3) a. JANOSHOZ erkezettj [Vp vendeg ^ ] John, to arrived guest 'It was to John that some guest(s) arrived' b. Nem erkezettj [Vp vendeg t± ] not arrived guest 'No guest(s) arrived' The claim that a preverbal bare nominal subject is adjoined to the V is based on the fact, among others, that the subject+V complex can undergo coordination.
THE EPP IN A TOPIC-PROMINENT LANGUAGE (4)
109
Janosnak [VP [ajandek kesziil] es [torta siil] a John. DAT present prepares(iNTR) and cake bakes(iNTR) his sziiletesnapjara] birthday-for 'Some present is in preparation and some cake is baking for John on his birthday'
(ii) Nonspecific indefinite subjects—whether of a theme, an agent, or an experiencer role—must remain in the VP in postverbal position.3 Their order with respect to the other postverbal arguments and adjuncts does not appear to be grammatically constrained. For example, (5)
(6)
[yp Telefonalt egy vendeg] telephoned a guest 'A guest has telephoned' a.
b.
Eliitott egy auto egy jarokelot. hit a car.NOM a pedestrian.ACC 'A car hit a pedestrian' Eliitott egy jardkelot egy auto. hit a pedestrian.ACC a car.NOM
Postverbal nonspecific indefinite subjects can have any 0-role; any verb that allows a nonspecific indefinite subject allows its subject to be realized in the VP. The fact that in (6) and (7) neither the V nor the postverbal subject has left the VP can be proven by showing that they follow VP-adverbials: (7)
a.
Azonnal [Vp telefonalt egy vendeg a tuzoltoknak] immediately telephoned a guest the fire, guard, to 'A guest telephoned to the fire department immediately' b. ?? Telefonalt egy vendeg azonnal a tuzoltoknak. telephoned a guest immediately the fire, guard.to
(iii) Subjects with a [+identificational focus] feature appear in the pre-VP focus position, whatever their referential properties are.4 I follow Brody 1990 in assuming that a [+identificational focus] constituent occupies the specifier of a Focus Phrase, a functional projection dominating VP, and the filling of SpecFP goes together with V movement into F. (8) Bizonyara [FP A /EGY DIAK nyertej [VP meg az elso dijat t-v ]] certainly the/a student won PREF the first prize 'It was certainly THE/A STUDENT who won the first prize' Bizonyara 'certainly' and similar sentence adverbials can either precede or follow the topic in the Hungarian sentence. They are, however, always external to the predicate phrase (i.e., the VP optionally extended by the functional projections FP, QP (quantifier phrase), and NegP). Thus bizonydra cannot follow a focused subject. (9)
a. * A the b. ? A the
DIAK / EGY DIAK bizonyara nyerte meg az elso dfjat. student/a student certainly won PREF the first prize DIAK / EGY DIAK nyerte bizonyara meg az elso dfjat. student/a student won certainly PREF the first prize
110
SUBJECTS, EXPLETIVES, AND THE EPP
(iv) Universally quantified subjects (as well as subjects modified by is 'also' and meg . . . is 'even') are confined to a prefocus quantifier position, located in the specifier of a quantifier projection dominating VP and FP (cf. E. Kiss 1998a). (10)
Szerintem [QP minden diak [FP A SZINTAXIST szereti legjobban]] in.my.opinion every student the syntax likes best 'In my opinion every student likes SYNTAX the best'
The sentence adverbial cannot follow the QP, which indicates that the QP is in a predicate-phrase-internal position: (11)
? Minden diak szerintem [ppA SZINTAXIST szereti legjobban] every student in.my.opinion the syntax likes best
(v) Definite and specific indefinite subjects can appear either postverbally (see (12a)) or in a topic position, presumably in the specifier of a topic projection (12b). In the latter case they can freely precede sentence adverbials. (12) a.
b-
A szintaxist szerintem jol [VP tudjak a diakok] the syntax.ACC in.my.opinion well know the students.NOM 'Syntax, in my opinion, the students know well' tlopP A diakok szerintem jol [Vp tudjak a szintaxist]] the students.NOM in.my.opinion well know the syntax 'In my opinion the students know syntax well'
Let us summarize the possible surface realizations of subjects with different referentiality and operator features: (13)
There is no empirical evidence suggesting that any of the surface positions that the subject can occupy in visible syntax is identical with SpecIP. Similarly, there is no evidence indicating that the subjects appearing in various surface positions are linked to one and the same specifier slot by movement or via a chain. 2.2. Object positions in Hungarian In the Hungarian sentence every surface position that is available for the subject is also available for nonsubjects with similar referential properties. Thus in (14a-f), the very same positions that are taken by the subjects in (2)-(12) are occupied by objects.
THE EPP IN A TOPIC-PROMINENT LANGUAGE
111
VP-internal, V-adjoined position for a [-ref] object (14) a.
Peter [Vp [y- hazat vesz]] Peter house.ACC buys 'Peter buys a house'
VP-internal, postverbal position for a [+ref,-spec] indefinite object b. Peter [Vp vesz egy hazat] Peter buys a house.ACC 'Peter buys a house' SpecFP for a focused object c.
Peter valoszmuleg [FP A SAROKHAZAT veszij [VP meg t|]] Peter probably the house.on. the.corner. ACC buys PREF 'It is probably the house on the corner that Peter buys'
SpecQP for a quantified object d. Peter valosziniileg [QP mindket hazat [VP meg veszi]] Peter probably both house.ACC PREF buys 'Probably Peter buys both houses' VP-internal, postverbal position or topic position for a [+ref,+spec] object e.
f.
Peter valosziniileg [VP meg veszi a hazat] Peter probably PREF buys the house.ACC 'Probably Peter buys the house' [TopP A hazat valoszmiileg [VP meg veszi Peter]] the house.ACC probably PREF buys Peter 'Probably Peter buys the house'
We can assume that the constituents extracted from the VP into operator positions participate in feature checking. The focus operator is moved to SpecFP to have its [+focus] feature checked (cf. Brody 1990, 1995), whereas the universal quantifier is moved to SpecQP to have its [+distributive quantifier] feature checked (cf. Szabolcsi 1997, E. Kiss 1998a). Beghelli and Stowell (1994) appear to suggest that the topic is also moved to SpecTopP to have its preferential, +specific] features checked in SpecTopP. Given that preferential, +specific] constituents can also stand postverbally, Beghelli and Stowell allow these features to be checked at LF. There are, of course, other descriptive options, as well. For instance, we can assume that a postverbal referential, specific constituent is stranded in the specifier of a lower TopP projection, with the rest of the clause preposed into a higher functional projection—a la Kayne 1998. In any case, the functional projections in which these features are checked are, to all appearances, unrelated to verbal inflection (tense and agreement), and hence to the standard formulation of the EPP. 2.3. An expletive in Hungarian?
Actually, Hungarian does make use of an expletive-like demonstrative pronoun associated with a clause (whether extraposed or in situ); however, this pronoun, too appears not only with subject clauses but with complement clauses of any function as well.
112
SUBJECTS, EXPLETIVES, AND THE EPP
(15) a.
b.
(16) a.
b.
Az nyilvanvalo [CP hogy a teszt tul nehez volt a diakok that.NOM obvious that the test too difficult was the students szamara] for That is obvious that the test was too difficult for the students' Az [CP hogy a teszt tul nehez volt a diakok szamara] that.NOM that the test too difficult was the students for nyilvanvalo. obvious Azt hiszem [CP hogy a teszt tul nehez volt a diakok that.ACC think.1 that the test too difficult was the students szamara] far 'I think that the test was too difficult for the students' Attol tartok [CP hogy a teszt tul nehez volt a diakok that.ABL fear. I that the test too difficult was the students szamara] for 'I am afraid that the test was too difficult for the students'
The pronominal element associated with a clausal complement is not restricted to a particular sentence position. In (15)-(16) it appears in SpecTopP, but it can also stand in SpecFP (17a), in SpecQP (17b), or in VP-internal position (17c). (The VPinternal nominative and accusative pronominals, which have no operator function, can also be dropped.) (17) a.
b.
c.
Szamomra [qp az is [Vp nyilvanvald volt [CP hogy Eva for.me that.NOM too obvious was that Eve megbukik]]] fails 'That, too, was obvious for me that Eve would fail' Szamomra [FP csak az [VP volt nyilvanvalo [CP hogy Eva for.me only that.NOM was obvious that Eve megbukik]]] fails 'Only that was obvious for me that Eve would fail' Szamomra nem [VP volt nyilvanvalo az /pro [CP hogy Eva for.me not was obvious that.NOM/pro that Eve megbukik]] fails 'That was not obvious for me that Eve would fail'
As is also clear from the set of facts above, the pronominal element associated with a clause is not a regular expletive but represents the head of a complex NP. In Hungarian, complement clauses appear subordinated to a DP node (cf. E. Kiss 1990, Kenesei 1994). The function of the pronominal head of the DP is to take up the case assigned to the sentential complement, and to represent it in the clause-internal positions (e.g., SpecFP) of the matrix clause that are not available for complex constituents. Crucially, this pronominal is not merely a placeholder for the
THE EPP IN A TOPIC-PROMINENT LANGUAGE
113
complement clause but also appears together with it (see (15b) and (17c)). Furthermore, it accompanies not only subject clauses but also complement clauses of any function—hence its existence in complex Hungarian sentences cannot be a consequence of the EPP. 2.4. The EPP as a requirement to fill SpecTopP If we intend the EPP to formulate generalization(s) about possible and impossible structures in all human languages, including the 'free word order' type, we must abstract away from its standard wording, and must attempt to reinterpret the insights that it seeks to capture in more general terms. It has been well known since at least Kuroda 1972-1973 that in such apparently free word order languages as Japanese and Hungarian, the sentence structure displays the same subject-predicate dichotomy logically as it does in English, except that the logical subject of predication (also called topic) or, in structural terms, the external argument, does not necessarily coincide with the grammatical subject. In view of this, a natural way of extending the EPP to the 'free word order,' or topicprominent, language type would be to interpret it as the requirement that the sentence contain a subject of predication. But is it really the case that SpecTopP must always be realized in the Hungarian sentence? In fact, Hungarian allows apparently topicless, verb-initial sentences as well, in which all the arguments are located in the VP (as was demonstrated in (5)-(6))—however, their use is constrained. Compare the examples in (18), and (19)-(20). (18a-c) are sentences describing an event, whereas (19a-c) and (20a, b) are sentences describing a state (or, in a different terminology, (18a-c) contain a stage-level predicate, and (19a-c) and (20a, b), an individual-level predicate). (18) a
b. c.
[Topp Janos [VP meghivta Marit vacsorara]] John.NOM invited Mary. ACC for. dinner 'John invited Mary for dinner' [-Topp Marit [Vp meghivta Janos vacsorara]] Mary.ACC invited John.NOMfor.dinner [Vp Meghivta Janos Marit vacsorara] invited John.NOM Mary.ACC for.dinner
(19) a.
[Topp A diakok [Vp utalnak ket tantargyat]] the students.NOM hate two subjects.ACC 'The students hate two subjects' b. [jopp Ket tantargyat [Vp utalnak a diakok]] two subjects.ACC hate the students c. ?? [VP Utalnak a diakok ket tantargyat] hate the students two subjects.ACC
(20) a.
Janos kekszemu volt. John blue.eyedwas 'John had blue eyes' b. ?? Kekszemu volt Janos. blue, eyed was John
114
SUBJECTS, EXPLETIVES, AND THE EPP
(18a-c) describe one and the same event. (18a), containing a subject topic, formulates it as a statement about the referent of the grammatical subject; (18b), containing an object topic, formulates it as a statement about the referent of the object; whereas the topicless (18c) describes it without formulating it as a statement about any of the participants. That is, the presence of a visible topic is not obligatory in a sentence describing an event. A sentence describing a state, such as that in (19), on the other hand, is marginal without a topic; it is felicitous only if the predicate is emphatic (presumably preposed into the focus projection), with everything else deaccented. (Alexiadou and Anagnostopoulou (1998:494) report a similar fact from Greek, saying that "postverbal subjects occur with all eventive predicates"). Consider also the following examples: (21)
Koltoznek a golyak. migrate the storks 'The storks are migrating'
(22) a. b.
A golyak koltoznek. the storks migrate 'Storks migrate' "The storks are migrating'
a. b.
Szeptemberben koltoznek a golyak. September.in migrate the storks 'Storks migrate in September' 'In September, the storks are migrating'
(23)
Whereas the topic-predicate structures in (22) and (23) can be assigned both an eventive and a generic interpretation, the topicless (21) only expresses an event. The facts illustrated in (18)-(23) do not follow from either the standard version of the EPP, or its reinterpretations by Alexiadou and Anagnostopoulou (1998) and Holmberg and Nikanne (this volume). In Alexiadou and Anagnostopoulou's theory the EPP is satisfied if the V is merged with an inflectional suffix that is specific enough to license pro drop. The inflectional suffix with which the V merges is of the same type both in (18a-c) and in (19)-(20); hence Alexiadou and Anagnostopoulou's theory does not predict any difference between the two cases. According to Holmberg and Nikanne (this volume), movement to SpecTopP in Hungarian is triggered by the [-(information-)focus] feature of arguments. However, none of the arguments in (18)-(23) are [+(information-)focus], since they refer to familiar unique individuals (Mary, John, September, and the kind storks), which are present in the universe of discourse. I propose to derive the grammaticality difference between topicless sentences describing events, and topicless sentences describing states along the lines proposed by Kratzer (1989) and Erteschik-Shir (1997) on the basis of Davidson 1967. Namely, I assume that eventive verbs, and hence eventive sentences, have an event argument serving for spatiotemporal specification, typically represented by a variable bound by an invisible existential quantifier. Stative verbs, and hence stative sentences, on the other hand, do not have an event argument.5 As shown by Maleczki (1999), generic sentences pattern with statives even if they contain a non-
THE EPP IN A TOPIC-PROMINENT LANGUAGE
115
stative verb selecting an event argument: their event argument slot must remain empty (generic sentences are not anchored spatiotemporally). In eventive sentences the optional temporal and locative expressions are licensed by the invisible event variable; they function as restrictions on it. These expressions tend to appear in topic position. For example: (24)
[jopP Tegnap [VP meghivta Marit Janos vacsorara]] yesterday invited Mary.ACC John for.dinner 'Yesterday John invited Mary for dinner'
It has often been claimed that eventive sentences whose topic position (SpecTopP) is filled with a temporal or locative phrase restricting the event variable predicate about this temporal or locative element; thus (24) makes a statement about the previous day. Or, in the terminology of Erteschik-Shir 1997, the truth-value of (24) is evaluated with respect to the referent of tegnap. Let us attempt to maintain that the EPP indeed requires that sentences express predication about a constituent in SpecTopP. In accordance with this claim, let us assume that eventive sentences with no visible constituent in SpecTopP predicate about the situational or contextual restriction on their event variable. Then in the case of sentences like (18c) the subject of predication is the situationally or contextually determined invisible restriction on the event variable (meaning 'here and now,' or 'there and then'). (For a similar view, see Erteschik-Shir 1997.) If we maintain that the EPP is a predication requirement on sentences, then (19c) and (20b) are marginal as neutral sentences because they have no visible topic, and we cannot attribute to them an event argument restricted by a topicalized invisible element. Then (21) has no generic interpretation because the EPP requires that we attribute to it an invisible topic, meaning 'here and now,' restricting the interpretation of the event variable. 2.5. Restricting the domain of EPP as a predication requirement Even though in the case of the eventive sentence type illustrated in (18c) it appears to be both intuitively appealing and theoretically legitimate to assume an invisible element in SpecTopP, this analysis cannot be extended to all topicless sentence types. The (b) sentences in (25)-(29) below contain a stative predicate which selects no event argument, so their SpecTopP position must be genuinely empty; still, they are perfectly grammatical. These sentences include declarative sentences involving focusing (25a, b) or universal quantification (26a, b),6 as well as interrogative (27a, b and 28a, b) and optative sentences (29a, b). (25) a.
b. (26) a.
A diakok [FP A TEREPMUNKAT szeretikj [VP ?j a legjobban]] the students the fieldwork.ACC like the best 'The students like FIELDWORK the best' [Fp A TEREPMUNKAT szeretikj [Vp a diakok ^ a legjobban]] the fieldwork.ACC like the students the best A professzor [QP minden diakot [VP ismer]] the professor every student.ACC knows 'The professor knows every student'
116
SUBJECTS, EXPLETIVES, AND THE EPP b.
(27) a.
b. (28) a.
b.
[QP Minden diakot [yp ismer a professzor]] every student.ACC knows the professor Marit [FP ki ismeri [vp a legjobban]]? Mary.ACC who knows the best 'Who knows Mary the best?' [pp Ki ismeri [vp Marit a legjobban]]? who knows Mary.ACC the best Janos ismeri Marit? John knows Mary.ACC 'Does John know Mary?' Ismeri Janos Marit? knows John Mary.ACC
(29) a. ? Janos barcsak szeret-n-e Marit! John if.only love-COND-3SG Mary.ACC 'If only John loved Mary!' b. Barcsak [FP szeretne Janos Marit]]! if.only love.cond.3sg John Mary.ACC According to the evidence of these sentences, nondeclarative sentences do not require a topic, though they can have one. Declarative sentences (i.e., sentences expressing a statement) are also exempted from having a topic if they involve quantification or focusing (i.e., exhaustive identification, which is also a kind of quantification in the analysis of e.g., Szabolcsi 1983). That is, the EPP can be interpreted as a general predication requirement only if we restrict it to nonquantificational statements, as follows: (30) The domain of predication Statements express predication, or quantification, or both.
(31) EPP1 A sentence expressing predication must contain a topic. In other words, a sentence expressing predication projects a TopP, with Top having strong [+ref, +spec] features. The motivation for these generalizations is intuitively clear: sentences which serve to describe individuals or situations (i.e., whose logical function is to express predication) need a subject of predication (i.e., a topic) also on the grammatical level. Sentences which express such logical operations as quantification or interrogation, on the other hand, are also grammatical without a subject of predication. (The class of logical operators which make the presence of a topic optional will, naturally, have to be made more precise.) Under this interpretation, EPP1 is an output condition forced upon a set of syntactic structures by semantic considerations. 2.6. The case of English The nonspecific subject of be and marginally of certain other verbs meaning existence, availability, or coming into being can be VP-internal in English, as well. The
THE EPP IN A TOPIC-PROMINENT LANGUAGE
117
subjects of all other verbs must be preposed into a VP-external but predicatephrase-internal subject position, to have their nominative case checked. This subject position is presumably the same as the SpecTenseP position identified by Bobaljik and Jonas (1996), so I will refer to it as SpecTP. As I argued elsewhere (E. Kiss 1996), specific subjects are moved further on from SpecTP to SpecTopP. The claim that specific and nonspecific subjects occupy different positions at S-structure also in English is based on evidence of the following type: (i) Only specific subjects can be followed by sentence adverbials without a comma intonation; nonspecific subjects must be preceded by them: (32) a. * [TP A baby boy luckily was born] b. Luckily [TP a baby boy was born] c. John luckily [TP was born on time] (ii) Only specific subjects can be followed by negation; nonspecific subjects can at best be preceded by it: (33) a. * [Tp A baby wasn't born] b. Not [TP a baby was born] (iii) VP-deletion—which is argued in E. Kiss 1996 to be actually TP-deletion—is not possible in the presence of a nonspecific subject (see (34)); it can take place only if the subject to be left over is a TP-external specific subject. (34)
* [TP A storm occurred], and then a flood did.
(iv) Only and even phrases, whose maximal scope is the predicate phrase (TP), take scope over the whole sentence when preceding a nonspecific (i.e., TP-internal) subject; however, they take scope over the subject DP only when preceding a specific (i.e., TP-external) subject: (35) a. Only [TP a baby was born], nothing else happened, b. * Only [John] [TP gave a talk], nothing else happened. If our assumptions made in sections 2.3-2.4 are tenable, an English sentence like (32b) represents a TopP projection the specifier of which is filled by a contextually determined invisible element restricting the event variable. If English does not require the spelling out of this element, then the appearance of there in the SpecTopP position of sentences with a VP-internal nonspecific subject needs a special explanation. Szabolcsi (1986) argues that the verbs which allow a there expletive in subject position can be regarded as existential quantifiers. It seems reasonable to assume that when these verbs are used in the capacity of existential quantifiers, they must c-command and precede their subjects, which they bind as variables. It is presumably due to a constraint on Case assignment that the V-initial construction resulting from this binding configuration is illegitimate in English, and must be remedied by a there expletive, spelling out the otherwise invisible restrictor of the event variable. Namely, in English nominative case is assigned to the SpecTP position, and it needs a carrier there. (It is argued in E. Kiss 1996 that there is generated in SpecTP, and is moved on to SpecTopP.) It is not ob-
118
SUBJECTS, EXPLETIVES, AND THE EPP
vious whether the nominative case assigned to there needs to be transmitted to the VP-internal subject at all, which, according to Szabolcsi (1986), has no 0-role, and hence needs no case. (According to Belletti 1988, on the other hand, it receives partitive case from the V.) This approach is compatible with the analysis of expletives in Chomsky 1995:272-276, in which there and the VP-internal subject have different referential indices. The verb agrees with the postverbal subject because the ())features of the subject are covertly adjoined to I. Under these assumptions the function of there is significantly different from that of it, whose role is to represent an extraposed clausal complement first in SpecTP (the site of nominative case assignment), then in SpecTopP (the site of the subject of predication). In constructions of the following type: (36) a. b.
Itj is believed/obvious [that. . . ]j Itj seems [that. . .\
the expletive and the clause, having identical indices, satisfy EPP1 together (i.e., the extraposed clause shares the topic role of if). There, on the other hand, satisfies EPP1 on its own; it functions as a spatiotemporal topic in a sentence in which the nonspecific grammatical subject is not suitable for the topic role. That is, it and there have different functions with respect to EPP1—even though they share the role of serving as targets of nominative case assignment. A further EPP 1-related difference between English and Hungarian is that in English the subject always takes precedence over other arguments in filling SpecTopP. In Hungarian sentences SpecTopP can be filled with the visible or invisible restrictor of the event variable, or with a preferential, +specific] nonsubject argument also when the subject is preferential, +specific]; hence it would be available for the topic role. The source of this difference will be discussed in section 3, where it will be argued that in Hungarian, unlike English, subject marking is a lexical-morphological process which does not involve SpecTP. As I argued at book length in E. Kiss 1987, the Hungarian subject has no structurally prominent position in the VP, either; it is on a par with the other arguments of the V—hence the proposing of the subject into SpecTopP is not more economical than the preposing of any other argument. In English, on the other hand, subjects must pass through SpecTP in the course of the derivation. A subject in SpecTP takes precedence over nonsubject arguments as the target of movement into SpecTopP owing to the Minimal Link Condition of Chomsky 1995 because it is closer to SpecTopP than the other arguments. It remains to be answered by further research why a [+specific] subject cannot be kept in SpecTP by base-generating an adverbial or expletive restricting the event variable in SpecTopP, that is, why SpecTP cannot host a [+specific] constituent. (In Icelandic, too, SpecTP is available for specific subjects only if they are focused or quantified—see Vangsnes 1995.)
3. EPP as a universal principle of argument-ordering In a subject-prominent language the predicate-phrase-external argument comprises two functions: the function of the subject of predication (in other words, the topic) and the function of the grammatical subject (the argument which agrees with the
THE EPP IN A TOPIC-PROMINENT LANGUAGE
119
finite V; which bears nominative in the unmarked case; which can bind but cannot be bound by a coargument; which can control; etc.). In accordance with this fact, the standard formulation of the EPP also comprises two requirements: in addition to requiring that the clause instantiate a predication structure, it also requires that the clause contain a grammatical subject. The subject function is the function of the most prominent argument of a predicate (cf. Alberti 1997a, Grimshaw 1990). In the language type represented by English the subject function is associated with a particular position (SpecTP); a constituent assumes subject function by being moved to SpecTP, or being linked to it via a chain. Hence in these languages the requirement that the predicate have a subject amounts to the requirement that the specifier of the TP projection extending the predicate be filled. The statement that the most prominent argument of a predicate must be realized as a subject—in other words, every predicate that has at least one argument must have a subject—expresses a significant generalization also about Hungarian. In Hungarian, the subject is encoded as the carrier of nominative case; hence it follows that there can be no one-place predicate whose argument is other than the subject, and predicate-argument structures of the following types are impossible: (37) a. * Janosnak fel. John. DAT fears b. * Kinyilik az ajtot. / *Az ajtot kinyflik. opens the door.ACdthe door.ACC opens The only apparent counterexample, presented in (38) below, is spurious: (38)
Neki mar befellegzett. he. DAT already clouded It is all over with him' [Lit. 'It has already clouded for him']
The predicate of (38) is a weather verb. Weather verbs have an invisible third person singular pro subject in the nominative, whose presence is proven by the fact that it can control: (39) a.
b.
[pro [pro Korai sziirkuletet okozva] befelhosodott] early twilight.ACC causing clouded 'It clouded, causing early twilight' [pro [pro Az ablakokat megrezegtetve] dorgott es villamlott]7 the windows.ACC shaking thundered and lightninged 'It thundered and there was lightning, shaking the windows'
With the apparent counterexample eliminated, the generalization in question, implicit in the standard wording of the EPP, can be formulated as follows: (40) EPP2 Of the arguments of a predicate, one must be marked as a subject. I will argue below that in Hungarian the most prominent argument of every predicate is encoded as a subject (i.e., as the bearer of nominative case) in the lexicon. Consequently, EPP2 is satisfied in the lexicon already; hence it does not impose any requirement on the surface position of the grammatical subject. In actual mono- and bilingual dictionaries predicates are, indeed, stored as represented in
120
SUBJECTS, EXPLETIVES, AND THE EPP
(41). This way of representation also reflects the intuition on how predicates are recorded in the mental lexicon. The representation consists of the verb bearing a present indicative third-person singular suffix, as well as each of its arguments represented by a dummy pronoun and a case ending. The tense marker apparently plays a licensing role in Hungarian, as well, as infinitival phrases contain a PRO subject. This fact, however, does not affect the satisfaction of EPP2 as formulated in (40); an argument of each verb is marked as a subject in the lexicon, irrespective of whether its phonological realization will be licensed in syntax. (41) a. b.
esz-ik valaki valami-t eat-s somebody something-ACC ad-0 valaki valaki-nek valami-t gives somebody somebody-DAT something-ACC
If EPP2 is satisfied in the lexicon already, then it is expected that an argument cannot assume subject function in the course of the derivation. Indeed, the syntactic operations changing the grammatical functions of constituents are missing from Hungarian grammar. Thus there is no grammatical passive. If the subject of a transitive verb is not known, or cannot be spelled out for some other reason, it is represented by a third-person plural pro: (42)
A javaslatot [Vp elfogadtak pro] the proposal.ACC accepted they 'The proposal was accepted'
The absorption of the agent or experiencer 0-role always takes place in the lexicon; it is denoted by idiosyncratic derivational suffixes. Thus a common English verb usually corresponds to a group of Hungarian verbs derived from the same (often obsolete) stem by idiosyncratic suffixes. The verbs in this group often have different argument structures, and the most prominent argument of each verb, whatever its 0-role, is invariably identified in the lexicon as the bearer of nominative case. Thus, instead of open or turn, the Hungarian lexicon contains the following transitive and intransitive verbs, derived from the same obsolete stem by different (nonproductive) suffixes: (43) a. nyit valaki valami-t opens somebody something-ACC nyilik valami opens something b. fordit valaki valami-t turns somebody something-ACC fordul valami turns something Verbs of middle meaning are also formed by derivational suffixes in the lexicon. In the case of megveszteget 'bribe,' for example, the agent 6-role is absorbed by adding the reflexive suffix -d-\ (44) a.
Konnyii a biirokratakat megveszteget-ni. easy the bureaucrats.ACC bribe-iNFiNiTlVE 'It is easy to bribe bureaucrats'
THE EPP IN A TOPIC-PROMINENT LANGUAGE
b.
121
A biirokratak konnyen megvesztegetodnek. the bureaucrats easily bribe.REFL.3pL 'Bureaucrats bribe easily'
The accusative with infinitive construction is also missing in Hungarian. The construction in (45a), which occurs only with the verbs lot 'see,' hall 'hear,' and erez 'feel,' is analyzed as an object control structure: (45) a.
cf. b.
Lattam a napotj [PROj felkelni] saw. I the sun.ACC to. rise 'I saw the sun rise' Lattam a napot, hogy felkelt. saw. I the sun that rose 'I saw the sun that/as it rose'
What argues for analyzing (45a) as containing an infinitival adjunct clause whose PRO subject is controlled by the matrix object is the fact that the theme of the V meaning 'see' is clearly napot 'sun.ACC,' and not the clause. (45a) can be used only if the speaker actually saw the sun in the sky; it cannot be used when what the speaker saw was the fact that the sun was rising (e.g., when he was inside and only saw the sun's rays appearing on the wall). That is, (45a) is not synonymous with Lattam, hogy felkelt a nap 'I saw that the sun rose,' or Lattam, hogy keltfel a nap 'I saw that the sun was rising'; its meaning is the same as that of (45b), containing an object and an adjunct clause. The object control analysis of (45a) is also supported by the fact that subject idioms cannot be embedded in this construction: (46) a.
Lattam, hogy a ver meghiil benne. saw. I that the blood freezes in. him 'I saw that his blood ran cold' b. * Ldttam a vert meghiilni benne. saw. I the blood freeze in. him 'I saw his blood run cold'
Structures which involve a small clause with an exceptionally case-marked subject in English are realized in Hungarian as complex predicates (with the head of the small clause perhaps incorporated into the V) taking an object complement. For instance, (47) a.
b.
Marij [vp [V' tolvajnak tartja] tj Janost] Mary thief.DAT considers John.ACC 'Mary considers John a thief Marij [Vp [yr boldogga tette] tj Janost] Mary happy.TRANSLAT made John.ACC 'Mary made John happy'
Even though ECM does not exist in Hungarian, there exists an ECM-like phenomenon: a subject extracted from an object clause via operator movement receives accusative case in the course of the derivation. (48) Kitj akartok, [CP tj hogy [elnok legyen tj]] who.ACC want.you that president become 'Whom do you suggest should become president?'
122
SUBJECTS, EXPLETIVES, AND THE EPP
In the standard analysis of this construction (cf. Chomsky 1981:174, with reference to Kayne 1980), it is claimed that the accusative case is assigned by the matrix V to the intermediate trace of the moved constituent in SpecCP. I assume that the target of accusative assignment by the V akar 'want,' selected as such in the lexicon, is the CP complement. Since the CP cannot pick up the case assigned to it, the case will percolate on the first suitable target of accusative assignment in CP: the NP in SpecCP (i.e., the embedded subject passing through SpecCP at the intermediate stage of the derivation). That is, the accusative case appearing on the subject extracted from the embedded clause in (48) is a phonological phenomenon; it is the case assigned to the object clause; it does not affect the embedded subject status of the constituent. Thus, the assumption that EPP2 is satisfied in the lexicon in Hungarian can be maintained.
4. Summary The EPP in its standard version was formulated on the basis of so-called subjectprominent languages, in which the external argument fulfills two roles simultaneously in the unmarked case: it represents both the logical subject of predication and the grammatical subject (i.e., the thematically most prominent argument, which agrees with the V and is the target of nominative assignment). This chapter has argued that in accordance with this fact, the standard EPP comprises requirements on both functions. In languages in which the function 'logical subject/topic' and the function 'grammatical subject' are associated with possibly different constituents, these two requirements do not necessarily coincide, either. It has been claimed that one of the requirements implicit in the standard version of the EPP (called here EPP1) is the requirement that sentences realize a predication relation. This requirement is valid in a restricted domain: it does not hold of sentences containing logical operators of a sentential scope. The other requirement implicit in the standard version of EPP (called here EPP2) is the requirement that one of the arguments of a predicate be marked as a subject. It was argued that spechead agreement with the Tense head, resulting in the obligatory filling of SpecTP, is not the only way of subject marking. Subjects can also be identified in the lexicon and be marked morphologically, in which case EPP2 is satisfied in the lexicon already.
Notes 1. This situation is not unlike that found in Icelandic by Vangsnes (1995). 2. Bare nominals can occur VP-externally as contrastive topics in sentences displaying a particular rise-fall intonation contour. Under a contrastive intonation and interpretation (2b) is also marginally acceptable. 3. E. Kiss (1996) argues that the "internal" position associated with nonspecific subjects by Diesing (1992) is a predicate-phrase-internal position across languages. 4. In the theory of Alberti 1997b referentiality features are neutralized by the operator features [+focus], [+quantifier], and [+contrastive topic]. Observations of Vangsnes (1995) concerning Icelandic also confirm Alberti's claim.
THE EPP IN A TOPIC-PROMINENT LANGUAGE
123
5. Kratzer (1989) actually claims that it is the class of stage level predicates that have an event argument. The class of stage level predicates is slightly wider than the class of nonstative verbal predicates; it also includes such adjectival predicates as sick and hungry. E. Kiss 1998b, however, shows on the basis of evidence put forth in Emonds 1976 that sentences containing the V be and the adjective sick or hungry share the behavior of nonstative sentences only when sick and hungry are not predicative adjectives but function as adverbs. 6. As Kuroda (1972-1973) reports, sentences with a universally quantified subject are also analyzed as topicless thetic judgements in the logical theory of Anton Marty. 7. Similar structures involving a dative experiencer and an "impersonal" verb in pro-drop languages like Russian (e.g., that in (i)) must also involve a pro quasi-argument in the nominative. (i)
Mne bylo holodno. to.me be.pA.ST.3sc.NEUT cold 'I was cold' [Lit. 'it was cold to me']
Russian
References Ackerman, Parrel. 1984. Verbal modifiers as argument-taking predicates: Complex verbs as predicate complexes in Hungarian. Groninger Arbeiten zur Germanistischen Linguistik 25, 23-71. Alberti, Gabor. 1997a. Argument Selection. Peter Lang, Frankfurt. Alberti, Gabor. 1997b. Restrictions on the degree of referentiality of arguments in Hungarian sentences. Acta Linguistica Hungarica 44, 341-362. Alexiadou, Artemis, & Elena Anagnostopoulou. 1998. Parametrizing AGR: Word order, Vmovement, and EPP-checking. Natural Language & Linguistic Theory 16, 491-539. Beghelli, Filippo, and Timothy Stowell. 1994. The direction of Quantifier Movement. GLOW Newsletter 32, 56-57. Belletti, Adriana. 1988. The Case of unaccusatives. Linguistic Inquiry 19,1-34. Bobaljik, Jonathan D., & Dianne Jonas. 1996. Subject positions and the role of TP. Linguistic Inquiry 27, 195-236. Brody, Michael. 1990. Some remarks on the Focus Field in Hungarian. UCL Working Papers in Linguistics 2, 201-225. University College London. Brody, Michael. 1995. Focus and checking theory. In Istvan Kenesei (ed.), Approaches to Hungarian V, 29-44. JATE, Szeged. Chomsky, Noam. 1981. Lectures on Government and Binding. Foris, Dordrecht. Chomsky, Noam. 1995. The Minimalist Program. MIT Press, Cambridge, Mass. Davidson, Donald. 1967. The Logical Form of action sentences. In Nicholas Rescher (ed.), The Logic of Decision and Action, 81-95. University of Pittsburgh Press, Pittsburgh. Diesing, Molly. 1992. Bare plural subjects and the derivation of logical representations. Linguistic Inquiry 23, 353-380. Emonds, John. 1976. A Transformational Approach to English Syntax. Academic Press, New York. Erteschik-Shir, Nomi. 1997. The Dynamics of Focus Structure. Cambridge University Press, Cambridge. Grimshaw, Jane. 1990. Argument Structure. MIT Press, Cambridge, Mass. Holmberg, Anders. 1997. Scandinavian Stylistic Fronting: Phonological feature movement in the syntax. Working Papers in Scandinavian Syntax 60, 81-124. Horvath, Julia. 1986. Focus in the Theory of Grammar and the Syntax of Hungarian. Foris, Dordrecht.
124
SUBJECTS, EXPLETIVES, AND THE EPP
Kayne, Richard S. 1980. Extensions of Binding and Case marking. Linguistic Inquiry 11, 75-96. Kayne, Richard S. 1998. Overt vs. covert movement. Syntax 1, 128-191. Kenesei, Istvan. 1994. Subordinate clauses. In Ferenc Kiefer & Katalin E. Kiss (eds.), The Syntactic Structure of Hungarian, 275-354. Academic Press, San Diego. Kiss, Katalin E\ 1987. Configurationality in Hungarian. Reidel, Dordrecht. Kiss, Katalin E. 1990. Why noun-complement clauses are Barriers. In Joan Mascaro & Marina Nespor (eds.), Grammar in Progress, 265-277. Foris, Dordrecht. Kiss, Katalin E. 1996. Two subject positions in English. The Linguistic Review 13,119-142. Kiss, Katalin E. 1998a. Mondattan. In Katalin E. Kiss, Ferenc Kiefer, & Peter Siptar (eds.), Uj magyar nyelvtan, 15-184. Osiris, Budapest. Kiss, Katalin E. 1998b. On generic and existential bare plurals and the classification of predicates. In Susan Rothstein (ed.), Events, 145-162. Kluwer, Dordrecht. Kratzer, Angelika. 1989. Stage and Individual Level predicates. In Papers on Quantification. NSF Grant Report. Dept. of Linguistics, University of Massachusetts, Amherst. Kuroda, S.-Y. 1972-1973. The categorical and the thetic judgement. Foundations of Language 9,153-185. Maleczki, Marta. 1999. Weak subjects in fixed space. Acta Linguistica Hungarica 46, 95-118. Szabolcsi, Anna. 1983. Focusing properties, or the trap of first order. Theoretical Linguistics 10, 125-145. Szabolcsi, Anna. 1986. From the Definiteness Effect to Lexical Integrity. In Werner Abraham & Sjaak de Meij (eds.), Topic, Focus and Configurationality, 321-348. John Benjamins, Amsterdam. Szabolcsi, Anna. 1997. Strategies for scope taking. In Anna Szabolcsi (ed.), Ways of Scope Taking, 109-154. Kluwer, Dordrecht. Vangsnes, 0ystein. 1995. Referentiality and argument positions in Icelandic. Working Papers in Scandinavian Syntax 55, 89-109.
6 The Extended Projection Principle as a Condition on the Tense Dependency Ian Roberts and Anna Roussou
1. Introduction The purpose of this chapter is to provide a unified account of two seemingly unrelated phenomena: what we call the 'subject requirement,' which corresponds to the Extended Projection Principle (EPP) of Chomsky 1982, 1995, and the 'V2 requirement.' We argue that they both reduce to properties of T, and, arguably, to a single property of T in connection to the position where T is spelled out. In particular, the 'subject requirement' arises when T is realized in the IP domain, while the 'V2 requirement' arises when T is spelled out in the CP domain. In the first case the features associated with AgrS identify the T-dependency, and when AgrS is absent, as in some infinitivals, identification comes directly from C. The mode in which the subject requirement is satisfied further depends on how AgrS (i.e., the subject position) is parametrically realized (if at all), that is, either as a head or as a specifier, depending on the morphological properties of the particular language. In the second case, identification of the T-dependency follows from the presence of an XP in SpecCP. This is the essence of our account of the V2 requirement. We take identification to refer to PF-recoverability of LF-features, as these are instantiated by dependency formation. We will investigate the nature of two widely assumed, but little-explicated, constraints on clause structure. These are (i) the requirement that all clauses have a subject, namely the EPP in its standard sense, as introduced in Chomsky 1982:10, and (ii) the requirement that a declarative C which attracts the inflected verb in a V2 language must have a filled Specifier. Henceforth, we will refer to (i) as the 'subject requirement' and (ii) as the 'V2 requirement.' Neither the subject requirement nor the V2 requirement can be straightforwardly explained in terms of selection or checking. Following the standard idea
125
126
SUBJECTS, EXPLETIVES, AND THE EPP
that selection is connected to 9-role assignment, only argumental subjects can be selected, and so expletive subjects are unaccounted for. Chomsky (1995: ch. 4) proposes that the subject requirement can reduce to obligatory checking of the D-feature of T by the subject. In the context of the system developed there, this amounts to a simple restatement of the subject requirement rather than an explanation of it.1 Moreover, since the V2 requirement can be satisfied by any XP, neither selection nor feature-checking seems able to offer even a descriptively adequate account. The two requirements share two important properties: they both involve specpositions (IP and CP, respectively), and they both can be satisfied by an expletive, as (1) shows: (1)
a.
b.
Es wurde getrunken. it was drunk. 'People were drinking' There's a fly in my soup.
German
These observations suggest that the two requirements may be instances of a single property. This idea is supported by the fact that we can account for the obvious difference between them, namely, that the subject requirement forces a DP to occupy SpecIP while the V2 requirement allows an XP of any category to occupy SpecCP, by taking into consideration that the SpecIP is an A-position, while SpecCP is not. In feature-theoretic terms, an A-position is a position associated with nominal features, while an A'-position isn't. It is standardly assumed that only DPs can occupy A-positions. Thus we can view our two requirements as manifestations of a single property in different positions: since SpecIP is an A-position, it will have to be filled by a DP.2 SpecCP, being an A'-position, can be filled by any XP, and must be filled by some XP (cf. Roberts 1993 for a preliminary formulation). We can summarize this idea schematically as follows: (2)
a.
[IP DP[+Agr] /*XPt.Agr] I[+Agr]
(subject)
00000000000000000 The goal of this chapter is to formulate a principle which unifies these two requirements in the desired fashion: this principle can be thought of as a generalized EPP. We will argue that this principle relates to Tense. The proposed analysis then attempts to shed some light on the role of Tense in clause structure with reference to these two requirements. The structure of the chapter is as follows. In section 1 we discuss the similarities between the subject and the V2 requirement. In section 2 we provide a provisional unification of these two requirements under a T-criterion. In section 3 we provide a discussion of the relation between C and T, and outline our ideas about parametric variation, which accounts for the principal differences between V2 languages and non-V2 languages. We next consider in more detail the subject requirement and how it is manifested cross-linguistically, depending on the properties of AgrS. Following the same line of reasoning we consider the obligatoriness of XP-fronting in V2 clauses in V2 languages in connection with the realization of T in C. Finally, we consider the implications of our analysis for a number of other phenomena, such as expletives, embedded V2, VSO orders, and Long Head
THE EPP AS A CONDITION ON THE TENSE DEPENDENCY
127
Movement in languages like Breton, concluding with some speculations about the origin of V2.
2. A condition on T? As suggested above, the two requirements essentially reduce to two different positions with different feature specifications, described in terms of the A- vs. A'-distinction, modified appropriately. On this basis we further suggest that both requirements relate to a property of T, giving rise to the following preliminary formulation of the EPP: (3)
The head containing T must have a filled specifier.
The statement in (3) is intended to refer to T itself (where T does not move), or the position T raises to, or the position which some head H containing T raises to. In its sensitivity to overt (i.e., S-structure, or Spell-out) positions and in its formulation, (3) is similar to 'criteria' of the kind proposed by Rizzi (1991) for WH, by Haegeman (1995) for Negation, and by Brody (1990) for Focus, among others. We might think of (3), then, as a half of a putative 'Tense-criterion.'3 Condition (3) derives the subject requirement in an example like (Ib) as follows. In declarative clauses, T doesn't move in English, and so (3) directly requires that SpecTP be filled. Hence, if there is no argument available to fill SpecTP, an expletive must be merged in that position. Alternatively, if TP is dominated by AgrSP, then, following Chomsky (1995: ch. 3), we can assume that T raises to AgrS in finite clauses in English. In that case, (3) applies to AgrS (i.e., the head containing T), and thus it is SpecAgrSP that must be filled. On either analysis, the relevant spec is an A-position, in the sense that it carries nominal features, and as such can be filled only by a DP, as discussed above. Now let us see how (3) derives the V2 requirement in (la). We assume, following a long series of studies going back to den Besten 1983, that T, containing V, raises to C in V2 clauses. Thus C is the position to which (3) applies and so SpecCP must be filled. Since SpecCP is an A'position, any XP can be moved to it. This distinction regarding the different positions realizing T will become more apparent as we discuss the subject and V2 requirements in more detail. The approach based on (3) has the advantage that the subject requirement and the V2 requirement are distinguished purely by the independent property of V (or, more precisely, T) movement. Another advantage of this approach is that we predict the absence of EPP effects in nominals, on the basis that Tense is not present in these constructions, as the following contrasts show: (4)
a. It appears that a fly is in my soup. b. * Its appearance that a fly is in my soup.
(5)
a. [For there to be a white Christmas] would be nice. b. ? [There being a white Christmas] is nice. c. * [There's being a white Christmas] is nice.
The contrast between (5b) and (5c) is particularly revealing. The bracketed constituent in (5b) is an ACC-ing gerund, standardly analyzed as an IP (see Reuland
128
SUBJECTS, EXPLETIVES, AND THE EPP
1983). The corresponding constituent in (5c) is a POSS-ing gerund, standardly a DP (Abney 1987). The EPP is required in (5b), but not in (5c); (5b) contains T while (5c) does not.4 A further advantage of (3) is that it predicts that the subject requirement doesn't hold (or, to be more precise, it is not relevant) in V2 clauses in V2 languages, since it is stated to hold of the head containing T. For example, Haider (1993) argues in detail that subjects can remain VP-internal in German, which is consistent with (3). Moreover, it is well known that the expletive es cannot appear in SpecIP in V2 clauses in Dutch and German or in embedded clauses (cf. Bennis 1986): (6)
a.
b.
Gestern wurde (*es) getrunken. yesterday was it drunk 'Yesterday people were drinking' Es wurde getrunken. it was drunk. 'People were drinking'
German
(3) requires that the specifier of the head containing T be filled. Assuming, as mentioned above, that this head is C in V2 clauses, the requirement is met by gestern in (6a). No such requirement is imposed on SpecIP, and so es cannot appear. In (6b), es appears in SpecCP, thereby satisfying (3). Although the formulation in (3) has some clear advantages, it also raises some problems. The first point is purely conceptual. Like other 'criteria' which have been proposed in the literature, (3) looks like an S-structure condition. If S-structure is not defined as a separate level of representation in the theory, this is clearly a problem. An analysis which keeps the insight of (3) but makes it a requirement holding at either interface would clearly be preferable in the context of Minimalist theories. Moreover, like other 'criteria' it is mainly descriptive: it is not clear why such a requirement should hold of Tense in the first place. As such, (3) remains a restatement of the EPP observation, although it has the additional advantage that it extends the EPP to the V2 phenomenon, providing a (descriptive) unification of the subject and V2 requirements. Second, the above account of expletive distribution in V2 languages runs into various empirical difficulties. For German, (3) predicts that where T does not raise to C (i.e., in typical embedded clauses), an expletive will be obligatorily present. However, expletives are ruled out here just as they are in (matrix) V2 clauses (cf. Cardinaletti 1990, Vikner 1995, among others): (7)
. . . daB (*es) getrunken wurde that it drunk was '. . . that people were drinking'
German
Moreover, in many V2 Germanic languages (including West Flemish and the Mainland Scandinavian languages), expletives are required in SpecIP in V2 clauses, despite the XP fronting in SpecCP. We illustrate this with a Danish example from Vikner 1995:185:
THE EPP AS A CONDITION ON THE TENSE DEPENDENCY (8)
Igar er der kommet en dreng. Yesterday is there come a boy 'Yesterday a boy came'
129 Danish
It is obvious that the account of (6) given above, based on (3), runs into problems with respect to (8). A third problem concerns VSO languages: it is usually argued that the finite verb is at least as high in the clause structure as T (i.e., T, AgrS, or C) and the specifier of the position where the verb appears is obligatorily empty, giving rise to the basic 'verb-initial' phenomenology (see McCloskey 1996 and the papers in Borsley and Roberts 1996 for variants of this account of VSO order). Schematically, then, the structure is as in (9): (9)
[pp (*XP) [F V+T] Subject...]
In such structures, (3) appears to be directly violated, since the head that contains T, denoted as F in (9), does not have a filled specifier. To conclude this section, we see that (3) can unify the subject and the V2 requirement in a natural way, reducing the difference between these two to the position where T is realized. Moreover, formulating the EPP as a 'T-criterion' has the advantage of automatically accounting for the absence of EPP effects in nominals. It also accounts for the impossibility of es in SpecIP in V2 clauses in German. On the other hand, the 'T-criterion' shares with other criteria the disadvantage of being a parameterized S-structure requirement. We also saw that the account of expletive distribution in V2 languages has a number of empirical problems. Finally, VSO orders present a challenge to (3). In the next section, we will retain the idea that a condition on T can unify the subject and V2 requirements, but reformulate (3) in such a way as to avoid the difficulties we have mentioned here.
3. The nature of the 'T-Criterion' 3.1. General considerations We consider that an understanding of the 'T-criterion' depends on an understanding of the syntactic conditions on Tense (this idea is also developed in partially similar ways for V2 by Holmberg and Platzack 1995). Building on ideas of Partee (1984) and Eng (1987), among others, we take Tense to have semantic properties that are akin to those of pronominals. The denotation of a pronoun must be fixed in relation to something else; it is not intrinsically determined by properties of the pronoun. For Tense, this amounts to fixing the denotation of certain times or intervals. One way in which denotations can be fixed is by syntactic binding. Let us then propose the following syntactic condition for the grammatical properties of Tense: (10)
T must be bound.
We take the formal instantiation of a binding relation to be co-membership in a chain/dependency. We define dependencies as follows, adopting and adapting the
130
SUBJECTS, EXPLETIVES, AND THE EPP
definition given in Manzini 1995. Condition (llaii) essentially means that a attracts P; (llaiii) incorporates the Minimal Link Condition (MLC): (11) a.
b.
(ot, p) is a well-formed dependency iff: i. a asymmetrically c-commands p; ii. there is some feature F such that oc and p share F; iii. there is no y such that y asymmetrically c-commands P but not a. If ((Xj . . . Ojj) is a well-formed dependency and (Pi ... pm) is a well-formed dependency and (o^, Pj) satisfies (lla), then (ttj . . . Pm) is a dependency.
More generally, suppose that the temporal properties of a clause are derived from the interaction of C, T, and V, roughly as follows.5 A simple sentence like (12) has the approximate logical representation in (13): (12)
John left.
(13)
3t[t > IQ] (AT (e, t) & leave (e, j))
The formula in (13) states that there is a time t which precedes the speech time fy, and the event of John leaving took place at t. In neo-Reichenbachian terms we can take T to provide the Reference Time which gives the restriction on the temporal quantifier. A naive first approximation is thus that the Past feature of T is interpreted as [t > IQ]. The Speech Time is contextually given, presumably via C, as in Eng's (1987) theory of T-anchoring. The temporal quantifier is associated with C. AT may correspond to an aspectual head, giving a punctual reading here. AT relates the Event Time to the Speech Time and the Reference Time; the Event Time is an argument of the predicate (cf. Higginbotham 1985). (13) relates to the structure of (12) as follows: (14)
3t[t > to] (AT (e,t) & leave (e j)) C T Asp V
In order for the interpretation in (13) to be derived from the structure of (12), then, C must bind T. That is, there must exist a (C . . . T) dependency which satisfies the conditions in (11) (there must also be a (T ... Asp) and an (Asp . . . V) dependency, but we leave these aside here). In other words, the requirement in (10) follows from simple compositionality: C semantically binds T, and so must syntactically bind it. As is well known, it is quite common to find a morphosyntactic correlate of the (C, T) dependency. Indeed, since the earliest studies of complementation in generative grammar (Rosenbaum 1967), it has been observed that the C-system is correlated with finiteness. Very simple facts about English clearly show this: the complementizer/or must be followed by an infinitival, while that must be followed by a finite clause, etc. (cf. Stowell 1981, Pesetsky 1982). (15) a. b.
I think that John left. I hope for John to leave.
THE EPP AS A CONDITION ON THE TENSE DEPENDENCY
131
Moreover, in languages like Irish this correlation between C and T is also morphologically reflected on C according to the Tense specification of the embedded clause (Cottell 1995):
(16) a.
AFFIRMATIVE NON-PAST
PAST
go
gur ar ar cdr sular
a an cd sula
b.
(declarative C) (indirect relative clause) (interrogative) (interrogative of place) (subordinator of time/purpose)
NEGATIVE NON-PAST
PAST
nach mura
ndr murar
(negative C) (conditional negative C)
Note that the correlation between C and T is (to a large extent) parallel to that found between D and N in nominals. Longobardi (1994) proposes that DPs, but not NPs, can be arguments, and that D therefore is the 'argumentizer' of a nominal. Moreover, N must raise either overtly or covertly to D (i.e., it must form a dependency with D) in order for the nominal's referential property to be interpretable at LF. This points to a deep unity in the nature of the functional heads D and C, as has often been observed. The above considerations show why T must be bound by C. Now, in order for the (C . . . T) dependency to be formed, AgrS, which asymmetrically c-commands T and is asymmetrically c-commanded by C, must join the dependency (cf. (llaiii)).6 We now have a dependency (C, AgrS, T), formed for interpretive reasons, which contains the heads relevant to the V2 requirement (C) and the subject requirement (AgrS). Our basic idea is that these requirements derive from different ways in which (C, AgrS, T) is overtly realized. In order to see this in detail, we need to introduce one further principle and some simple assumptions about parameters. We account for cross-linguistic variation in terms of the simple idea that PF-interpretation differs from LF-interpretation in being language-specific. Parameterization is variation in PF realization, that is, variation in morphophonological properties of functional heads. More specifically, we assume that all functional features can be associated with a diacritic *, which requires PF-realization. PF-realization can be achieved by Move or by Merge (where Merge may include morphological properties of a language, i.e. its inflectional system). Which option is taken depends on the lexicon, but the most economical is always preferred. For this reason, Merge is always preferred over Move. If the lexicon provides a phonological matrix for F*, then this matrix will be F*'s realization, and Move is unavailable. Conversely, if the lexicon has no phonological matrix for F*, material
132
SUBJECTS, EXPLETIVES, AND THE EPP
from elsewhere must be moved to F* (subject to the usual constraints on movement) (cf. Roberts and Roussou 1997).7 As far as the (C, AgrS, T) dependency is concerned, then, the ways in which it is parameterized depend on the differing cross-linguistic PF-realizations of C, AgrS, and T. In these terms, we can see the connection with the T-criterion' described in the previous section: in a language where C* triggers movement, V+T raises to C and an XP must appear in SpecCP. The V2 requirement is thus connected to C* and the fact that T raises to C. Also, in many languages with AgrS*, T appears in AgrS and SpecAgrSP must contain a DP (aside from the cases mentioned in note 2). So the following generalization emerges: (17)
The highest F* in the dependency (C, AgrS, T) must have a filled specifier.
(Here 'highest' is defined, standardly, in terms of asymmetric c-command.) We will see in later sections that (17) is not quite accurate. However, as things stand, it can replace the T-criterion of the previous section with no empirical loss. Roberts and Roussou (1997) formulate the following recoverability condition on dependencies: (18) Recoverability of dependencies: In a dependency DEP = ((Xj . . . an), where [ai F*] asymmetrically ccommands all [aj>i G*], DEP must be interpreted as an F-dependency. The statement in (18) relates PF realization to LF interpretation in a fairly straightforward way. Let us illustrate this with w/i-movement: we assume that it is (18) that causes overt movement of exactly one w/i-phrase to the Specifier position of a +wh C in languages like English. For this account to go through, we must allow for dependency formation between a head and its spec. In fact, the definition of dependency given in (11) allows this straightforwardly. Condition (18) also derives Cheng's (1991:30) generalization that languages with dedicated w/i-particles lack wft-movement. In such languages (which include Chinese and Japanese), the whparticle realizes the WH* under Merge. Since the particle is by definition a w/z-element, (18) is also satisfied by the particle being merged in C. Thus w/i-movement is not required and hence, for economy reasons, impossible. So what needs to be done, in order to properly understand (17), is to apply (18) to the dependency (C, AgrS, T). Thus the V2 requirement must be seen as identifying the C-dependency and the subject requirement must be seen as identifying the AgrS-dependency. What seems to hold, and what we take to be the essence of the EPP, is the following: (19)
In the dependency (a} .. . o^ . .. T), oq (#T) must be *.
The basic case of (19) is, then, that T must be in a dependency identified as an AgrS-dependency or as a C-dependency. Intuitively speaking, (19) corresponds to the two possible positions where T can be realized: either in the IP domain, as part of the AgrS dependency, or in the CP domain, as part of the C-dependency. In the next sections, we will show how (19) provides an account of the subject requirement and the V2 requirement.
THE EPP AS A CONDITION ON THE TENSE DEPENDENCY
133
3.2. The subject requirement Let us first consider the subject requirement. Our contention is that this requirement is the case of (19) with AgrS*, in finite clauses. In nonfinite clauses, as we shall see below, the subject requirement does not hold (in the sense that there is no requirement for SpecAgrSP to be filled). We assume that all features are LF-interpreted (i.e., there are no uninterpretable features in Chomsky's 1995 sense; see also Brody 1997). LF-interpretation simply implies a mapping to semantic representation; categorial features, for example, are presumably mapped to semantic types (the mapping is not necessarily one-to-one). It then follows that agreement features as well are interpreted. In itself, this is unproblematic: it is clear that person and number (as well as natural gender) have semantic content. Similarly the D-feature of I, which for Chomsky (1995: ch. 4) is uninterpretable, corresponds to an interpretable property of AgrS, given that this position is associated with the realization of nominal subjects, in a way that will become precise shortly. Moreover, it is natural to assume that in null-subject languages the (^-features of AgrS realized on the verb are interpreted in the same way that the (^-features of subject pronouns are interpreted.8 Let us generalize this idea in terms of the principle in (20): (20)
In DEP = ( . .. D . . . <|>. . . ) , and only here, the argument bearing <|> can have a referential/specific interpretation.
The condition in (20) forces the formation of a dependency between AgrS and a head in the thematic domain, that is, V: DEP = (AgrS . . . V) (cf. Manzini and Roussou 2000 for further details and elaboration). In these terms, we can phrase the null-subject parameter as follows: (21) Null-subject parameter. If AgrS has fully specified nominal features, no DP is required for the argument's interpretation. Here 'fully specified nominal features' is intended to include D- and ^-features. Another way of phrasing (21) is to say that null-subject languages are those with the dependency Dep = (AgrS* . . . V).9 This relates to the recoverability of dependencies given in (18): under (18), AgrS* requires its dependency to be interpreted as an AgrS-dependency. We take this to mean that the dependency must be interpreted as containing a referential argument, as AgrS's features are inherently those of referential arguments. The structurally closest referential argument to AgrS will always be the subject argument (the external argument where there is one; otherwise —in passives and unaccusatives—the derived subject). In such a system, overt subjects are not required and in fact, by economy, not allowed. This is in fact another case of Cheng's (1991) generalization extended to AgrS: where AgrS* satisfies (18), nothing can appear in its specifier. Two important properties of null-subject languages follow directly from this. First, overt expletives are not found (cf. Rizzi 1982). Second, when there is an overt subject in a null-subject language, it cannot occupy AgrS. This derives the classical observation due to Rizzi (1982) that relates null-subject languages and 'free inversion,' as well as the general fact that in null-subject languages, placement of overt subjects is syntactically relatively free, and determined largely by discourse conditions such as topic and
134
SUBJECTS, EXPLETIVES, AND THE EPP
focus (cf. Philippaki-Warburton 1985, Tsimpli 1995, Cardinaletti 1994, Alexiadou and Anagnostopoulou 1998, Manzini and Savoia, this volume, among others).10 Note also that (18) in its application to AgrS* and (20) together create the biuniqueness requirement of the 0-criterion. In null-subject languages, then, the subject requirement is effectively satisfied by AgrS (cf. Borer 1986). We see that in this case, the generalization in (17) regarding T does not, and in fact cannot, hold. This is because AgrS* is fully specified. The subject requirement boils down to the application of (18) to AgrS* in this case. In non-null-subject languages such as English, a subject must appear in SpecAgrSP to satisfy (18). Merger/movement of a DP to SpecAgrSP satisfies (18) in that the DP provides AgrS with the relevant 'full specification' (if the DP is not an expletive, this in fact provides the AgrS-dependency directly with an argument). In this kind of language, we also have AgrS*; however, in this case the requirement of phonological realization is met in a different way, since the morphological system does not provide a realization of AgrS's nominal features as a head. Thus the only option is DP-insertion in SpecAgrSP (cf. also Manzini and Savoia, this volume). For this reason, a DP must appear in SpecAgrSP. When there is no argumental DP available, and when discourse and quantificational conditions on expletive-associate pairs are met, expletives are required. It is clear from the binding facts discussed in Chomsky 1995:274 that the expletive 'inherits' the agreement features of the associate, as (22a) shows. (22b) shows that this relation extends to expletives in infinitives: (22) a. b.
There arrived three men without identifying themselves. [For there to have arrived three men without identifying themselves] is unlikely.
In general, then, the subject requirement derives from (18) applying to AgrS*, which fits with the generalization made in (19). Turning to nonfinite clauses, we follow Manzini and Roussou (2000) in assuming that control and raising infinitives lack AgrS (or that AgrS is inactive). Condition (20) forces the creation of a dependency between the lower clause and the higher clause. Still following the Minimal Link Condition (see (11)), this implies that the lower C forms a dependency with T. According to the nature of the upper predicate, this gives rise to a dependency with one DP and two 6-roles in (23a), and one DP and one 0-role in (23b). The former is control and the latter is raising:11 (23) a. b.
John [tried+62+9!] + I to tH [02] leave, John [seemed+Gj] +1 to tH [6j] leave.
In infinitives with overtly realized subjects—/
THE EPP AS A CONDITION ON THE TENS (24) a. b.
135
For there to be life on Mars would be nice, We believe [Cfor there to be life on Mars].
The infinitival complementizers here (for or Cyor) realize a temporal/modal property of the clause. This property, however it is to be characterized in semantic terms, is incompatible with the presence of the operator which licenses 'arbitrary subjects' (cf. Chierchia's 1995 analysis of Italian impersonal si, and Manzini and Roussou 2000 on arbitrary control). It seems that it is for this reason that arbitrary subjects are not found in this kind of infinitival clause: (25) a. * For to be life on Mars is nice. b. * We believe [Cfor to be life on Mars].13 When the infinitival complementizers are absent, the Arb-licensing operator can be present, and the relevant interpretation arises: (26)
[To win the lottery] would be nice.
Note, though, that the ungrammaticality of (25) relates to the use of an existential sentence along with the presence of for. The Cfor does not in general block the arbitrary interpretation: instead it gives rise to an obviation effect so that a subject must be present. In fact (26) is possible with/0/; but in this case the arbitrary subject must be overtly realized by the pronominal one: (27)
[For one to win the lottery] would be nice.
We tentatively assume that the presence of for or Ctor blocks clause union; their presence creates a separate clause domain, triggering the presence/activation of AgrSP. Given that AgrS is * in English, it follows that the subject has to be overtly realized, as is indeed the case in (27). The above remarks on infinitives hold for both null-subject and non-null-subject languages, to the extent that infinitives in both types of language are broadly similar (cf. Roussou, in press, for a discussion of control and raising in subjunctive clauses in Balkan languages). The same account extends to matrix clauses with an arbitrary reading: (28) a. * Finishes papers in the summer. b. One finishes papers in the summer.
(with arbitrary interpretation)
(28a) is ungrammatical simply because AgrS is present in a finite context and it requires Spell-out in English by virtue of its * property. Indeed, if the pronoun one is present the result is grammatical (or you with arbitrary reference in some cases). Essentially, there are three kinds of (C, T) dependency at work as far as infinitival clauses are concerned: (i) (C, T) is a subpart of a control/raising dependency, (ii) C is for or Cfor, and (iii) C contains the Arb-operator. To summarize so far, we have shown how the subject requirement follows from the fact that (19) is satisfied by AgrS* in finite clauses (in differing ways in nullsubject and non-null-subject languages) and by C in nonfinite clauses. The (C, T) dependency has three forms in nonfinite clauses. In the cases where the (C, T) dependency is headed by the Arb-operator, (19) is not satisfied.14 However, in precisely these cases (20) does not hold. There are no D-(and <()-) features associated
136
SUBJECTS, EXPLETIVES, AND THE EPP
with the external argument/derived subject of the predicate and this argument lacks a referential/specific interpretation. It seems, then, that the subject argument is a variable unselectively bound by the Arb-operator in C. The same reading can extend to T as well. For the subject, this is a default case—what happens when all other ways of identifying the subject fail. Perhaps we can say the same for T: identification by the Arb-operator is the default licensing mechanism when (19) fails to hold. What the discussion shows so far is that there is a correlation between AgrS and T as part of the dependency formed with C. In particular, they both relate to operators (or properties) associated with C. In fact, both the temporal variable provided by T and one of the features of AgrS, namely person, are evaluated with respect to the Speech Time. We already mentioned that temporal interpretation is relative to Speech Time, what we identify as the 'now' of speaking. Similarly, person is interpreted relatively to Speech Time, since it provides the basic speaker (I) vs. hearer (you) distinction (i.e., the participants in the discourse). Davis (1998) independently argues that the EPP reduces to an identification of a temporal head by (()features. Under his approach first and second person are intrinsically associated with T. In our terms we take person to be part of the feature specification of a distinct head, namely AgrS, which selects T. Note that the relation between T and AgrS is quite often morphologically expressed. For example, agreement features may be realized as inflectional endings associated with the predicate (which in turn relates to T; recall that the predicate provides the Event Time and the thematic properties which are ultimately attracted by AgrS, and T provides the Reference Time which binds/attracts E). Thus the realization of T within the IP domain is crucially linked to the features of AgrS. In infinitival complements where AgrS might be missing, T is associated with the AgrS in the matrix clause (control and raising constructions); this is compatible with the idea of clause-union that we mentioned before. If control (and raising) is blocked, we get an arbitrary reading. These relations are directly captured under the formation of a (C, AgrS, T) dependency. The question that remains open, of course, is why does (19) hold? Condition (19) is in fact a generalization of Chomsky's (1995) proposal that the EPP reduces to a strong D-feature of T. Equating * with a strong feature, what (19) says is that the T-dependency must be associated with a strong feature; this feature can be at the AgrS or C level. Although this is empirically superior to Chomsky's approach (and our discussion in this section encompasses the results of Alexiadou and Anagnostopoulou's (1998) extension of Chomsky's approach), it arguably retains the conceptual problem associated with the EPP: why is there a special condition of this kind? The clue to an answer to this question arguably comes from the cases where (19) fails to hold. Consider in particular the case where neither AgrS nor C has an overt realization, in infinitives with an arbitrary subject, as in (26). As mentioned earlier, the arbitrary interpretation, both of Tense and of the subject, can be seen as the default case. In other words, (19) is not a condition on representations, but a condition on representations with a nonarbitrary interpretation. If Tense or the subject is to have a nonarbitrary interpretation, then the T-dependency, as defined in (19), must be identified in the relevant way. Viewed in this light, (19) follows from Full Interpretation (representational economy) combined with the general
THE EPP AS A CONDITION ON THE TENSE DEPENDENCY
137
idea that dependencies must be identified by some F* which is interpreted structurally and semantically as a member of the dependency. All (19) really contributes is a definition of the T-dependency. The fact that the T-dependency always includes at least AgrS and C follows from the way in which T is interpreted combined with the MLC as in (llaiii), as sketched in section 3.1. If we add that identification is irreflexive, we derive the result that the identifier of the T-dependency cannot be a finite element, which, as we will see shortly, is relevant to the analysis of Germanic V2 (and the Breton Long Head Movement data).
3.3. The EPP and V2 3.3.1. The V2 requirement In this section we show how (19)—or, to be more precise, the recoverability condition in (18)—can derive the V2 requirement. We first discuss full V2, on the basis of German data, and then its implications for the problematic cases that we mentioned in section 2. To see how our account works, let us compare some simple clauses in English and German. The simplest relevant cases are root declaratives introduced by an adverb: (29)
a. b.
Yesterday John danced. Gestern hat Johann getanzt. yesterday has John danced c. * Gestern Johann hat getanzt.
German
In (29a), neither V nor T moves. That V does not move is a consequence of the well-known lack of movement of main verbs out of VP in English (cf. Emonds 1978, Pollock 1989). That T does not move in declaratives is what distinguishes English from the other Germanic languages. As far as the position of the adverbial yesterday in (29a) is concerned, Kayne (1994:28) points out that adverbs here pattern like negative phrases which are usually thought to occupy SpecCP in requiring the presence of an overt C in an embedded clause:15 (30) a. b.
I didn't know *(that) yesterday John danced. I didn't know *(that) never had John danced so well.
We conclude, therefore, that in (29a) the adverb occupies SpecCP. T-movement is absent here (this is not, strictly speaking, observable in (29a) without auxiliary hypotheses about do-insertion, but it is clear from sentences containing other auxiliaries, e.g., *Tomorrow will John dance/Tomorrow John will dance, that T does not move to C). In German, on the other hand, movement of the finite verb over the subject, when the adverb is present, is obligatory, as the contrast between (29b) and (29c) shows. Where a complement is topicalized, the same contrast between German and English arises: (31) a. b.
That book, I've already read. Das Buch habe ich schon gelesen. the book have I already read.
German
138
SUBJECTS, EXPLETIVES, AND THE EPP
We see that German and English are exactly alike, except for the movement of the finite verb/auxiliary (Miiller and Sternefeld (1993:481) also make this point). This variation is a fundamental element in the account of V2. In other words, V2 is essentially the result of the fact that the dependency between C and T is realized overtly on C in languages like German (in finite clauses). In terms of the parametric system outlined above, German has Cfin* (a finite C must have a PF-realization). In other words, parametric variation arises as to which position in the T-dependency is spelled out (i.e., C or T). Although this approach can provide an account of the surface differences between German and English as far as the position of the verb/auxiliary is concerned, it does not explain why XP movement must take place in German but not in English, that is, why it is necessary in the former, but not in the latter case to fill SpecCP (as we continue to call the landing site of V2 topicalization for now, pace Miiller and Sternefeld 1993—see note 15). This correlation between XP and T-toC movement is central to the concerns of this chapter, in that we want to derive the V2 requirement from (19) (as a consequence of (18) regarding T). Clearly, (19) is satisfied in languages with Cfin*. But postulating Cfin* only gives the result that the finite verb appears in C (in the relevant kinds of clauses—see below on the rootembedded asymmetry). The real question is what forces XP-fronting. According to Zwart 1997, XP-movement is the result of checking a topic feature, triggering T-to-C movement as well, in a way similar to \vA-questions in German or English: (32) a.
Was hast du gelesen? what have you read b. * Was du hast gelesen? c. What have you read? d. * What you have read?
German
The difference between English and German is that while in both languages \vhfeatures are strong, topic features are strong in German only; hence the difference in (29a) and (29b) above. Therefore, according to Zwart's analysis topicalization is the trigger for V2, essentially reducing V2 to a Topic Criterion (this idea is arguably implicit in Miiller and Sternefeld's 1993 analysis, too). Note that this approach differs crucially from the approach that takes V2 to be contingent on the correlation of Inflection and C (as developed by den Besten 1983, Platzack 1987, Tomaselli 1989 and others). Moreover, the Topic Criterion seems problematic to us. The various criteria that have been proposed in the literature (wh-, Focus, Negative Criterion) are basically formulated to account for the movement of phrases with quantificational properties, that is, XPs that need to satisfy scope requirements and receive a special interpretation. Topics, however, do not seem to function like quantifiers/operators (see the papers in E. Kiss 1995 and more recently Rizzi 1997 for detailed argumentation to this effect). If this is correct, then the idea that they are subject to a Topic Criterion, which has the same properties as the other criteria, turns out to be dubious.16 In order to understand the nature of XP-fronting in V2 clauses, let us proceed in a slightly different way and consider the contexts where XP-fronting does not take
THE EPP AS A CONDITION ON THE TENSE DEPENDENCY
139
place, although the finite verb has moved to C. The relevant contexts are the following: (33) a.
b.
c.
Yes/no questions: Hat Johann das Buch gelesen? has John the book read 'Has John read the book?' Conditionals: Hatte ich mehr Zeit gehabt, . . . had I more time had 000000000000000000 Narratives: [. . . ] Kommt eine Frau herein und . . . comes a woman in and 00000000000000000000000000
German
In all the above constructions the position of the finite verb in relation to the subject indicates that it has moved to C. In conditionals there is one more piece of evidence, namely that when the conditional C is present, inversion is blocked: (34) a.
Wenn ich mehr Zeit gehabt hatte, . . . when I more time had had b. * Wenn hatte ich mehr Zeit gehabt,. ..
German
What the constructions in (33) have in common is T-to-C movement in the absence of an overt XP preceding the auxiliary. One possibility is to assume that the above constructions as well are subject to some type of Operator Criterion, which in this case is satisfied by a null operator (cf., e.g., Rizzi 1991, Haegeman 1995). Although this approach has the advantage of bringing V2 in declaratives as well as in all the other contexts under a criterion-type constraint, it turns out to be problematic for a number of reasons. First, there is no syntactic motivation—beyond regularizing the pattern to V2— for postulating an XP-operator in SpecCP. There is no doubt that conditionals and yes/no questions, at least, must be characterized by an interpretable feature (see the references in Heim 1982:168ff. on conditionals and Higginbotham 1993 on yes/no questions). This feature is most naturally associated with C (and, indeed, is crosslinguistically realized by Move or Merge applying to C). There is no reason to link this feature to an empty operator which then imposes its own criterial licensing condition. Such a move adds nothing to the semantic characterization of these clause types. The only syntactic motivation for it is to generalize the spec-head configuration, but this is precisely the configuration whose nature in V2 clauses we wish to understand. The second problem is more specific to V2: if the possibility of having a null operator to satisfy the various criteria is in principle available, what rules out the same strategy in declaratives as well, under the assumption that there is a null topic operator? In other words, the question is why (35) is ungrammatical with a null operator:17
140 (35)
SUBJECTS, EXPLETIVES, AND THE EPP * [CP Op [c Hat][IP Johann das Buch gelesen]]
If null operators can satisfy the other criteria, then a null topic operator should in principle satisfy the Topic Criterion, giving rise to VI. The same reasoning applies even if we assume that V2 is not the result of topicalization but of some 'TCriterion,' as mentioned in section 2: why can't some null operator satisfy the TCriterion? On the basis of the above discussion, then, let us assume that there is no need to postulate a null operator in the above contexts and that the appropriate features are directly associated with C (cf. Manzini 1996 on how to account for locality effects on A'-dependencies under this approach). In other words C is specified as +Q in yes/no questions, and with the relevant feature in conditionals (see below for a proposal regarding narratives like (33c)). The exact nature of the features involved in these constructions is not, strictly speaking, relevant to what follows. What is important is that in yes/no questions, conditionals, etc., C has one more feature (apart from +Fin) which is essentially responsible for the specific interpretation that is derived. For ease of exposition we will refer to these features as indicators of clause type. Let us now go back to V2. What distinguishes V2 declaratives from the VI constructions (as well as the residual V2 cases, like Neg-preposing, focusing, w/i-questions, etc.) is that the latter include an additional feature that marks clause-typing. We now account for the V2 requirement in a straightforward way based on the idea that finite C attracts T, combined with (18) (the requirement for the recoverability of dependencies) and the idea that a declarative C lacks clause-type features, as just said. As we have already mentioned, the attraction property of finite C in V2 languages satisfies (19). To satisfy (18), the C-dependency must be identified. This is achieved by the clause-typing feature of C in the VI clauses we have looked at (see also Platzack 1998 for a more recent discussion). In w/i-questions, this is achieved by the w/i-feature,18 and in focus and Neg-preposing constructions by the relevant feature. If there is no 'additional' feature present, as the above discussion suggests, then the question is how the C-dependency is identified in V2 declaratives, where C has no independent content apart from finiteness. Our proposal is precisely that XP-fronting takes place in order to satisfy (18). The C-dependency is headed by the category which marks the Speech Time (see section 3.1), and so, effectively, it is the Speech Time which must be identified. XP-fronting gives a topic-interpretation because the XP must be interpreted as old information: the content of the XP is interpreted as part of the Speech Time in the same way that the features of AgrS are. When the T-dependency is realized in the IP domain, the presence of AgrS suffices for identification; when it is realized in the CP domain, however, this information can be provided only by some other property relevant to the C system, triggering XP-fronting. More generally, then, we can see C and SpecCP in a V2 declarative as together specifying contextually given information (see also During 1995, Reinhart 1995).19 The relevant interpretation is achieved without postulating a Topic Criterion. Note that our analysis goes through even if we assume that the XP moves into a separate Topic projection along the lines proposed by Miiller and Sternefeld (1993).20 Our alternative requires a topic feature, but no checking of that feature, and so no Topic
THE EPP AS A CONDITION ON THE TENSE DEPENDENCY
141
Criterion. Furthermore, it is the realization of C*fin that triggers topicalization and not vice versa. Our approach is supported by examples like (33c), cases of 'lively narrative style.' What is going on here is that there is no further specification of the Speech Time, and so the information given in the clause is interpreted as highly novel. The discourse conditions which favor this sort of order include the beginnings of narratives; this is quite consistent with our analysis, in that the beginning of a narrative does contain old information which can specify the Speech Time. On this view, then, the 'liveliness' of the style has to do with the novelty of the information contained in the clause. A further advantage of considering that XP-fronting is not triggered by a feature-checking or criterial requirement comes from markedness considerations. We can think of markedness relations among features of functional heads (of the kind discussed in detail by Cinque (1999)) in terms of a feature hierarchy. Functional heads, as features F, G, H . . . , can come with various further feature specifications /, g, h . . . (we write the subfeatures lowercase and potentially autonomous functional features uppercase). We can then treat unmarked values of functional heads as simply the autonomous functional feature F, while the marked value will have a further subfeature, giving F+/. So C means 'declarative,' while C [-declarative] means nondeclarative. More precisely, neither [-declarative] nor [+declarative] exists; what exist are clause-type features: Q, Exclamative, Conditional, etc. These are all subfeatures of C. In other words, instead of saying that we have C ^declarative], we have C = Declarative by default, where no subfeature is present, and C = Q, Exclamative, and so on, as marked subfeatures. Now, if parameterization applies indifferently to all types of features, F+/will have two chances of PF-realization (because both F and/can be assigned *), while F will have only one. Let us make the further natural assumption that where F* must be realized, then so must all subfeatures of F*. Then marked feature values are more likely to be overtly realized than unmarked ones, and we derive implicational statements of the form 'if a language has a declarative particle, then it has an interrogative particle,' etc. (this would be the application of the approach to C*Me • as far as we are aware, this statement is empirically correct). This approach carries over to the F*j^ove case, which seems right. In the case of C, we derive that C*+/,/a marked clause type, may trigger movement ('residual V2') or that C* may trigger movement (full V2), or neither, but we predict that there are no residual V2 languages where V2 takes place only in declaratives (this would be the illicit feature combination C*+/, but we have just postulated that C* implies/* for all/= subfeatures of C). As far as we are aware, this is correct. Incidentally, note that an approach like Zwart's (1997) cannot explain why there are no languages with strong Top-features but weak whfeatures, and so on.21 To summarise the discussion so far, we have accounted for the two basic movements involved in V2: T-to-C movement and XP-to-SpecCP movement. T-to-C is triggered by C*. XP-movement is triggered by the need to satisfy (18), i.e. to identify the C-dependency in root declaratives; since these clauses are unmarked for clause type, the only content of C is marking of Speech Time; XP-topicalisation is required in order to identify the Speech Time.
142
SUBJECTS, EXPLETIVES, AND THE EPP
3.3.2. V2: Implications and extensions Let us begin our elaboration of the basic analysis by considering how our proposals look if we adopt a split-C system of the type proposed by Rizzi (1997). Rizzi splits the traditional C into (among others) Force and Fin, with the possible interpolation of Topic and Focus in between. Force contains clause-type features (i.e., Q, etc.). In matrix declaratives, Force contains no feature. Fin is connected to the tense/mood system of the clause; therefore Fin binds T. In a split-C system our discussion about markedness translates as follows. We take the various subfeatures associated with C to correspond to different C positions. The default case is having Fin and some higher C position, which would roughly correspond to Rizzi's Force (and could ultimately be taken as the position that interacts with the Speech Time). The generalization is then that when Fin is *, as in declaratives, the other C-positions must also be * when activated. For example, if we have a system where Fin is spelled out in the default case, then there must be some C-realization in the marked cases such as interrogatives, exclamatives, etc. In V2 languages T is phonologically realized in Fin*. Given that identification is irreflexive, as we already mentioned, another C position is activated to identify the T-dependency, giving rise to XP-fronting. In an articulated CP-structure, XP-fronting targets a TopicP which is situated above FinP (in this respect we adopt ideas from Miiller and Sternefeld 1993; see note 15). The analysis that we have given so far about V2 and markedness is retained. In the light of these modifications, let us now consider the problematic cases for what we called the T-criterion' in section 2, starting with VSO languages. In fact a closer inspection of at least the Celtic languages reveals that VSO orders have differing properties in different languages. One interesting variant of the basic VSO typology is found in Literary Welsh. Here there is a system of clause-initial particles. (36) illustrates the situation: (36) a.
b.
c.
Root affirmative fe/mi/y Fe/mi welais i Megan. PRT saw I Megan 'I saw Megan' Interrogative a A welaist ti Megan. PRT saw you Megan 'Did you see Megan?' Root negative ni(d) Ni welais i ddim Megan. PRT saw I not Megan 'I didn't see Megan'
Literary Welsh
We can analyze this as C*Merge. Even if the particle is merged in the lower C position, namely Fin, we don't see the V2-effect, just as in languages with WH-C*Merge there is no overt WH-movement (cf. the discussion of Chinese and Japanese above). The distinct properties of the particle identify the relevant dependency in a straightforward way. Thus the problem arises only in the C*Move case.
THE EPP AS A CONDITION ON THE TENSE DEPENDENCY
143
Another possibility, which is instantiated by Old Irish, is that 'pure' verb-initial orders are found, but that in initial position the verb has special morphology. The morphology is an instantiation of inflectional features associated with some C head. This is arguably another instance of C*Merge, where what is merged are inflectional features (these features might be thought of as comparable to the marked clause-type features described above). (37) (from Carnie et al. 1995) illustrates Old Irish: (37) a.
b.
Beirid in fer in claideb. carries(ABS) the man the sword 'The man carries the sword' Nf beir /*beirid in fer in claideb. not carries(coNJUNCT)/carries(ABS) the man the sword 'The man does not carry the sword'
Old Irish
When negation is present, as in (37b), the morphological realization of the verb changes; in this case the negative particle realizes a distinct position above Fin, and the verb remains in a lower position, presumably Fin (cf. Carnie et al. 1995 for detailed argumentation that V moves to C in Old Irish). A third possible VSO structure is that T occurs in AgrS in a null-subject system, where the rich agreement satisfies (19). This is arguably what we find in Modern Irish; this gives 'pure' VSO orders (on Irish word order, see McCloskey 1996 and the references given there). Crucially, though, the C (Fin) position is not overtly realized (for further discussion see Roberts 2000). A fourth, and extremely interesting, situation is instantiated by Breton. In this language, although the principal word order is VSO, finite verbs are unable to appear in first position (this example, like all the Breton ones to follow, is taken from Borsley et al. 1996, henceforth BRS): (38)
* Lenn Anna al levr. reads Anna the book
Breton (BRS's (37), p. 62)
Instead, if no XP is fronted to a preverbal focus position, a nonfmite verb form must precede the finite form. This can trigger a process comparable to English dosupport in otherwise simple tenses, where the finite form is a dummy auxiliary and the nonfinite form is the main verb. These properties of Breton are illustrated in (39). (39a) shows XP-fronting; (39b) the order participle-Aux (the equivalent of 'have,' i.e., this is a periphrastic perfect), and (39c) the 'do-support' case: (39) a.
b.
c.
Lennet en deus Yann al levr. readJsc.M has Yann the book 'Yann has read the book' Al levr a lavaras Yann e lennas. the book PRT said Yann PRT read 'Yann said that he read the book' Lenn a ra Anna al levr. read PRT does Anna the book 'Anna reads the book'
Breton (BRS's (1), p. 53)
(BRS's (6), p. 55)
(BRS's (3), p. 54)
144
SUBJECTS, EXPLETIVES, AND THE EPP
BRS show in detail that the movement of the nonfinite verb over the finite verb is not XP-movement (topicalization, focusing or scrambling), but rather long head movement (LHM). They also show that the movement is subject to the constraints on LHM that have been identified in the literature (see in particular Rivero 1991, Roberts 1994). We can account for these facts by saying that Breton has C*Move; in this respect, it resembles a V2 language (as has often been noted). C* also satisfies (19) for T, whatever position we take regarding the position of the finite verb, the particles, and the subject (all rather intricate questions which would take us too far afield here—see Stephens 1982, Schafer 1994). Moving a nonfinite element into C* is clearly enough to identify the C-dependency (i.e., to satisfy (18)). (The C position here can correspond to Fin exactly as in German.) This suggests that a condition on the C-dependency is that it cannot be identified by T (i.e., by a finite element), given the irreflexive nature of identification that we already mentioned. Why does Breton allow LHM of a nonfinite verb to satisfy C*Move and at the same time identify the C-dependency, while German and other V2 languages require movement of a finite verb to this position followed by XP-fronting? In this we can follow BRS in assuming that the nature of the Breton auxiliary plays a crucial role in licensing the trace of LHM (BRS formulate the condition in terms of head-government, following Roberts 1994). Breton auxiliaries are able to license an LHM-trace; German ones are not (in BRS's terminology, Breton has functional auxiliaries while German does not—see their paper and the references given there for a general characterization of functional auxiliaries). Thus in German, in the absence of sentential particles, the only element able to raise to C (i.e., Fin) is the finite verb. When the finite verb raises to C, the C-dependency has to be identified by XP-movement, as described in the previous section. The Breton system is more economical (as it involves one movement—LHM—rather than two); hence the absence of V2 in this language.22 There is another possible explanation for the LHM that we find in Breton, namely to assume that Breton is exactly like German in that the finite verb moves to C* (Fin). If this is correct, then the C-dependency in Breton is subject to the same identification requirement as its German counterpart. One possibility is to satisfy the XP-fronting requirement by means of focusing, as in (39b). The other possibility is to resort to LHM of the nonfinite verb. In this case the nonfinite element moves to a position higher than Fin; that could be the head of a Topic/Focus projection or even Force itself. The implications of this analysis are that Breton has to be treated differently from the other Celtic languages for which VSO order cannot be derived by movement to C. There are further implications regarding the position of the subject which we leave open for future research. It is worth noting, though, that this is a viable option, which can be further explored in the light of a split-C system. In any case our approach also subsumes Rivero's 1993 system of Tense-licensing, which BRS assume. BRS propose that Tense may be licensed in main clauses by (i) V-adj unction, (ii) government by a filled C, (iii) a C with a filled spec or by heading an L-marked C. Note in particular that (ii) and (iii) (which derive LHM
THE EPP AS A CONDITION ON THE TENSE DEPENDENCY
145
and V2 clauses, respectively) are derived by our (18) and (19) as described here. Assuming that where V moves to T it moves on to AgrS, (i) also falls under (19). Next, let us consider the Germanic expletives that, as we noted in section 2, pose a problem for a simple formulation of the 'T-criterion' as in (3). For simplicity, let us compare German and Danish. Recall that the difference between these two languages concerns the presence of the expletive in V2 clauses where some XP, other than the expletive, is in SpecCP: (40) a.
b.
Gestern wurde (*es) getrunken. yesterday was it drunk 'Yesterday people were drinking' Igar er *(der) kommet en dreng. yesterday is there come a boy 'Yesterday a boy came'
German; =(6a)
Danish; =(8)
V2 operates as described earlier in both of these languages. The difference in the distribution of expletives must be attributed to the different properties of AgrS in the two languages (cf. Holmberg and Platzack (1995), who relate these differences to Nominative Case). Danish has AgrS* which is satisfied in the same way as in English—in the absence of a head realising AgrS*, a DP must always appear in its specifier. German AgrS, on the other hand, has no PF-interpretation requirement, in agreement with what has been proposed by Haider (1993) (see also den Besten 1985, Grewendorf 1989, Diesing 1992). For this reason, German appears to have multiple subject positions, since no DP is required to move to SpecAgrSP. In this respect, German is similar to null-subject languages, as discussed in section 3.2 above, with the crucial difference that null referential subjects are not allowed.23 Let us next briefly comment on the nature of the root-embedded asymmetry. In finite embedded clauses, there is a dependency between C and the selecting predicate; also, complementizers typically appear. Complementizers are able to satisfy C* in V2 languages, rendering V-movement unnecessary. Since T is realized lower in the clause on V, it clearly satisfies (19). This is the basic account of the root-embedded asymmetry. However, in WH-complements where no overt complementizer is present, something more must be said. Here we can capitalize on an observation by Stowell (1981:422) to the effect that selection for +WH neutralizes selection for Fin. This can be illustrated by paradigms such as the following: (41) a. I explained how to fix the sink. b. I explained how we should fix the sink. c. I explained that we should fix the sink. d. * I explained to fix the sink.
[+WH, -Fin] [+WH, +Fin] [-WH, +Fin] [-WH, -Fin]
In Rizzi's system, this is straightforwardly accounted for by the fact that both Force and Foe (the positions of w/i-elements) are structurally higher than Fin. Suppose, concretely, that if Force is selected as +Q, this activates Foe as +WH (but not conversely; cf. the brief discussion of exclamatives in note 18), but no feature of Fin is selected. More generally, we can say that the presence of Foe blocks the selectional relation between Force and Fin—this follows from standard assumptions about the locality of selection.
146
SUBJECTS, EXPLETIVES, AND THE EPP
Embedded V2 in asymmetric V2 languages appears in two variants: in German, a complementizer is impossible, while in other languages, such as Danish, it is required, as in (42a) and (42b), respectively (Vikner 1995:66-67): (42) a.
b.
Er glaubt [diesen Film haben die Kinder gesehen]. he thinks this film have the children seen. 'He thinks that the children have seen this film' Vi vet [at denne bog har Bo ikke laest]. we know that this book has Bo not read. 'We know that Bo has not read this book'
German
Danish
The bracketed constituents in (42) are embedded V2 clauses: the Danish one is introduced by the complementiser at, 'that'; in German no complementizer appears. We analyse the German case exactly as a matrix V2 clause: the C* is satisfied by V-movement, which triggers XP-movement in conformity with the analysis given in the previous section.24 In Danish, we must posit a higher complement projection which acts like an embedded clause, and a lower one which acts like a main clause. We assume that the higher one is Force and the lower one is Fin. This is similar to the CP-recursion analysis argued for by Vikner (1995) and Holmberg and Platzack (1995), with the difference that we attribute distinct feature specification to the relevant C heads. Embedded V2 is also found in the so-called symmetric V2 languages like Icelandic and Yiddish. Here all kinds of embedded clauses can be V2; moreover, as opposed to German and Danish, embedded V2 is not sensitive to the nature of the selecting predicate. We illustrate this with Icelandic, as follows (from Vikner 1995:91): (43) a. b.
Hann veil [ad kannski las Jon aldrei (*J6n) bokina]. Icelandic he knows that maybe read Jon never book, the Hann veit [ad kannski hefur Jon ekki (*J6n) lesid bokina]. he knows that maybe has Jon not read book, the
As the above examples show, the definite subject cannot occur in a lower position, suggesting that V movement to Fin* has taken place, exactly as in matrix clauses, as already argued for the German example in (42a). Where (43) differs from (42), though, is that both Force (i.e., the higher C head) and Fin are spelled out.25 Here Force is always realised by a complementiser and Fin is always realised by Vmovement. Finally, we would like to close the discussion with some speculations on the origin of V2. Recall that we argued, adopting and adapting standard analyses in the literature, that V/Aux movement is triggered under the need to satisfy C*. Recall also that movement in that respect is a rather residual case: V/Aux is attracted in the absence of any other designated head that could satisfy C* (Fin). If, for example, the language in question makes available a specialized lexical item that can be merged in C*, then movement is barred under Economy (in this connection, more needs to be said about the complements to German verbs like glauben, which can be either V2 or do/?-clauses—see note 24). The Move option, then, understood in those terms, has an altogether residual property, possibly reflecting earlier stages in
THE EPP AS A CONDITION ON THE TENSE DEPENDENCY
147
the language where there was indeed a distinct morpheme to spell out the relevant feature(s). We would like to suggest that this is in fact the case with regard to Germanic V2. As has been argued in the literature (Ferraresi 1997), Gothic made available a number of first and second position particles (this was also a characteristic of other Indo-European languages); while the former are associated with clause typing, the latter are either emphatic or correlate with realis (like the clitic uh) or irrealis mood (like the nonclitic frau). An example of the latter is given below (from Ferraresi 1997:131): (44)
ip is qap-uh. . . PRT he said-PRT 'But he said
Gothic
According to Ferraresi, ip occurs in Force, while the definite (pronominal) subject is realized as a topic, and -uh is a Fin particle occuring with realis mood. (Other elements can attach to uh in addition to V.) What is relevant to our discussion is that the presence of particles in C (Fin) already instantiate the * option; once the particles become phonologically weak elements, equivalent to affixes, V-movement takes place to satisfy the affixal nature of these elements, giving rise to the V2 phenomenon. Thus we have a change in the instantiation of C* (i.e., Fin* and/or Force*) from C*Merge to C*Move, with an intermediate stage of C*Merge+Move (the affixal stage). Although speculative, this approach reflects the residual nature of Vmovement to C argued above. A more detailed analysis of the diachronic development of V2 is open to future research (cf. also Kiparsky 1994 for a correlation between the development of subordination and V2). To summarize, we have argued that an analysis that relates V2 to the conditions in (18) and (19) is sufficient to derive the 'V2 constraint' on principled grounds: XP-fronting takes place to identify the C-dependency. We showed how this approach is compatible with the observed variation among VSO languages. We also accounted for the distribution of expletives on the basis of the properties not only of C but of AgrS as well. Finally, we discussed briefly the root-embedded asymmetry, made further use of the split-C system of Rizzi 1997, and speculated on the origin of V2.
4. Conclusion In this chapter we argued for a unified account of the 'subject requirement' and the ' V2 requirement' as instances of the EPP, which reduces to a property of the T-dependency. In particular, we argued that T is bound by Fin (the lower C position in Rizzi's 1997 terms). Furthermore, we posited the constraints in (18) and (19) as follows: (18) Recoverability of dependencies: In a dependency DEP = (al . . . (\), where [ai F*] asymmetrically ccommands all [aj>i G*], DEP must be interpreted as an F-dependency.
148
SUBJECTS, EXPLETIVES, AND THE EPP
(19) In the dependency (<Xi .. . a,,.. . T), cq must be *. The V2 requirement must be seen as identifying the C-dependency, and the subject requirement must be seen as identifying the AgrS-dependency. In the IP domain, whether there is an overt subject or not depends on the properties of AgrS. If AgrS*, it can be realized as either a head or an XP, triggering a specifier in the latter case, which is prototypically found in non-null-subject languages like English and Danish. In the CP domain (i.e., C*, as in the Germanic languages), fronting of an XP takes places to identify the C-dependency. The fronted XP is interpreted as old information, and as such identifies the C (=Speech Time) dependency. We also saw that (19) may in fact fail to hold, and when this happens, specific interpretations arise. In the IP domain, (19) does not hold in the case of arbitrary subjects. In the CP domain, it fails to hold in cases of VI 'lively narrative style' in V2 languages. We also considered the implications of our analysis with respect to the different distribution of expletives in the various Germanic languages, the root-embedded asymmetry, VSO languages, and the diachronic development of V2: a more elaborate analysis of these topics along the lines suggested here is left open to future research.
Notes Earlier versions of this chapter were presented at the LAGB Autumn Meeting in Cardiff (September 1996), the Bangor Research Seminar (October 1996, February 1998), the University of Stuttgart (October 1996), the Workshop on Subjects, Expletives, and the EPP in Troms0 (June 1997), and the Research Seminar at the University of Geneva (November 1997). We are grateful to the audiences for comments. We would also like to thank Bob Borsley, Liliane Haegeman, Ursel Luhde, Gereon Miiller, Peter Svenonius, and Tarald Taraldsen for useful comments and discussions. This chapter forms part of a larger research project supported by the British Academy (Grant no. APN 2989), which we gratefully acknowledge. 1. Chomsky (1998) suggests that the EPP reduces to a property of functional categories, namely that they require a specifier. Since it is an invariant property, the EPP does not resemble true lexical selection, which is by definition selection among a number of options. Moreover, this account is subject to the same criticism as the feature-checking one: it simply restates the observation. 2. Bob Borsley (p.c.) points out that SpecIP is not necessarily filled by a DP, given examples like the following: (i)
a. Under the bed is a good place to hide. b. Very cool is the best way to keep it. c. Very slowly is the best way to do it. The following suggest that the initial non-DPs are in SpecIP: (ii)
a. Is under the bed a good place to hide? b. Is very cool the best way to keep it? c. Is very slowly the best way to do it? In terms of the agreement requirement on SpecIP, roughly stated in (2a), we could perhaps consider that PPs and APs of the type seen here are able to be associated with agreement features (i.e., 3sc). Note that this is more natural than stipulating that they are able to be DPs, which is what is required by Chomsky's 1995 account of the subject requirement. 3. The other half of the Tense criterion might be formulated as follows:
THE EPP AS A CONDITION ON THE TENSE DEPENDENCY
149
(i) XP in SpecV[_iexj must specify T. Assuming, following Grimshaw 1991, that clausal functional projections are in the relevant sense verbal, (i) would require that any head with a filled specifier contain T. We will ignore this in what follows, especially as (3) will be substantially reformulated. 4. We thus expect that ACC-ing gerunds lacking an overt subject have an interpretation with an arbitrary null subject ('arbitrary PRO' in standard terms): (i) [ PRO^ being a genius ] is nice. This, too, follows from (3), as we will show in the following section. We also expect that POSS-ing gerunds, being DPs and as such not subject to (3), tolerate the absence of any subject. In fact, this is not true: (i)
a. [It raining all the time] depresses me. ACC-ing b. * [Its raining all the time] depresses me. POSS-ing c. * [(The) raining all the time] depresses me. POSS-ing There appears to be no way that verbs lacking an external argument can appear in POSS-ing gerunds. Contrary to what is proposed in the text, this may be due to the subject requirement combined with the lack of genitive forms of expletives (although note that referential its exists). 5. This passage owes much to discussions with Gereon Miiller, for which we are very grateful. The interpretation of his remarks given here is our own, however. 6. We continue to refer to the various projections above T in the IP domain as AgrS, bearing in mind that this is not simply a checking position as in Chomsky 1995: ch. 3 but the head that realizes the subject's nominal properties. For the purposes of this chapter we will refer to a single Agr-projection, namely the one licensing the subject, even though there is evidence that there is more than one such projection between C and T. This idea is developed in various ways in Cardinaletti and Roberts 1991, Manzini and Savoia, this volume, and Sportiche 1992. 7. This approach derives the effects of Last Resort directly from Full Interpretation, in that movement takes place only to satisfy interface conditions: to give F* a PF-realization, or to recover F*'s content (see below). 8. Of course, there are well-known differences between the interpretation of overt subject pronouns and null subjects (see Montalbetti 1984). 9. Null-object languages with 'rich' object agreement presumably have AgrO*. An example may be Pashto, as discussed by Huang (1984:535-536). In certain tenses of Pashto, agreement is with the object (it is a split-ergative language). In just these tenses, null objects are licensed: (i)
ma mana we-xwar-a. Pashto I apple pRF-eat-3F.se 'I ate the apple' (ii) ma — we-xwar-a. / PRF-eat-3F.se 'I ate it(FEM)' Leaving aside the analysis of split ergativity, then, Pashto illustrates the possiblity of 'rich' AgrO licensing null objects. 10. Null-subject languages differ, of course, in imposing various independent conditions on placement of subjects. In Italian, for example, it is well known that VSO order is not possible, while in Greek and Spanish this order is readily found. We illustrate this point with a Greek example: (i)
Efaje o Yannis ta mila. Greek ate. 3s the John the apples 'John ate the apples' 11. We take the lack of AgrS (inactive AgrS) to relate to the phenomenon of clause-union. Full clause-union, giving rise, for example, to the 'restructuring' phenomena of Italian, arises when all features of lower T are associated with the higher T—cf. Kayne 1989, Roberts 1997. 12. For present purposes we assume, following Kayne 1991:666-668, that English whether, French de, and Italian di are in SpecCP, or in some other position which does not block control.
150
SUBJECTS, EXPLETIVES, AND THE EPP
13. The presence ofCfor is specific to English. As Kayne (1984) showed, in French this complementizer is absent and, as a consequence, control interpretation with believe-type verbs is possible: (i)
Nous croyons [etre les meillieurs]. French we believe be the best 'We believe ourselves to be the best' This follows straightforwardly from the assumptions made in the text (cf. Boskovic 1997 for an alternative account). 14. In the case of Ctor (19) is satisfied, although C is null. As described above, the nature of Cfor— the fact that it blocks control—forces the realization of a subject. 15. For a number of very detailed and convincing arguments against topicalization as adjunction to IP, see Miiller and Sternefeld 1993:479^84. Miiller and Sternefeld (1993:484-486) further assume that topicalization has its own landing site in a split-C system (cf. also Rizzi 1997) and that this position is the target of XP-fronting in V2. We will not adopt precisely this analysis of V2, although we accept that w/z-movement and topicalization, including XP-fronting in V2, are distinct movement types. See below for a sketch of how a Rizzi-style split-C system can be integrated with our approach. (Rizzi's approach to topicalization in English assumes the presence of a null operator; we'll gloss over this matter here.) 16. Also, topics differ from w/z-phrases and other XPs which can be fronted in that they cannot remain in situ (in fact, the notion of an in-situ topic is hard to understand—this suggests that the property of being a topic does not inhere in a given constituent, but is attributed to certain structural positions; the postulation and checking of topic features misses this point). While languages vary in allowing w/zmovement of exactly one w/z-phrase (French), requiring movement of exactly one w/z-phrase (English), requiring movement of all w/z-phrases (Slavic), or not allowing overt w/z-movement (Chinese), no such cross-linguistic variation is found with regard to topicalization. We take these as further reasons not to assimilate topicalization to w/z-movement in the way Zwart (1997) does. 17. (35) is not a case of topic-drop or 'pronoun zap' (see Ross 1982, Huang 1984). The latter is illustrated by (i) and (ii): (i)
Hat Johann gelesen. German has John read 'John has read it' (ii) Hat das Buch gelesen. has the book read 'He/she has read the book' Our point concerns a putative null sentential topic operator, comparable to the sentential operators found in conditionals and yes/no questions. Such an operator might be invoked for 'lively narrative style' examples like (33c) (see Roberts 1993:57). However, the question then arises as to the characterization of this element. Note also that the general objection to null operators in (33a, b) would also apply to this operator. We have nothing to say here about German topic-drop (see Cardinaletti 1990). 18. The w/z-feature arguably combines with Q to give the relevant interpretation. Note that a w/z-DP can front independently of Q. In English, this shows up where wh fronts without subject-aux inversion (in matrix clauses): (i) How nice he is! So we can suppose that w/z-fronting takes place freely, as long as it can be licensed by an interpretation. (Certain types of) w/z-phrases can be licensed independently of Q, in which case they have an exclamative interpretation (for reasons which need not concern us). 19. We assume that initial expletives in V2 clauses are interpreted as their associates. 20. We expect that the same analysis extends to subject-initial V2 sentences, in agreement with the claim made in Schwartz and Vikner 1997 and contra Zwart 1997. 21. One could extend this line of reasoning, following proposals by Giorgi and Pianesi (1998), and say that F can be entirely absent from the representation, but will be 'read in' at LF by convention. On the other hand, F+/has to be syntactically present in order to be interpreted. Once syntactically present, F+/is parameterized, and so might be PF-realized. Cinque (1999:133) criticizes the Giorgi and Pianesi approach on the grounds that it leads to two ways of giving a default value for F: F is either present with the default value or absent and interpreted with a default value. In terms of the proposals being made
THE EPP AS A CONDITION ON THE TENSE DEPENDENCY
151
here, though, we could think that F can be present with a default value only if PF-realized. The maximally unmarked case is then where F has no PF-realization and the default LF interpretation. It is natural to think of this as the absence of F from the numeration. Below we suggest that this might be true for the declarative C head in languages like English. What this approach requires is a theory of LF which can tell us how the defaults are filled in. 22. This analysis of LHM extends to the other cases mentioned by Rivero (1991), Roberts (1994), and BRS: Bulgarian, Czech, Slovak, Serbo-Croatian, Romanian, Old Spanish, and European Portuguese (see BRS, p. 58, and the references given there). As a last remark on VSO languages, we comment briefly on the VSO order itself. At least in Irish and Welsh, it seems clear that V is in AgrS and the subject is in SpecTP in these orders (see McCloskey 1996 on the former and the introduction to Borsley and Roberts 1996 on the latter). Roberts and Shlonsky (1996:188f.) argue that in Welsh (and by implication in other VSO languages), following ideas of Koopman (1992), Cardinaletti and Starke (1994), and others, weak pronouns have the inherent property of being attracted to Agr. In VSO systems, where by hypothesis AgrS is unable to attract D-features, SpecAgrS is not available for weak pronouns (or anything else). Instead, weak pronouns appear in AgrS itself. In other words, in a VSO language like Welsh, the subject agreement endings are weak pronouns; they appear in AgrS as this is the only way in which they can satisfy the requirement of being associated with AgrS. Roberts and Shlonsky show that this account can capture the well-known antiagreement effect of these languages. Assuming this analysis to be basically correct, we can understand why the subject must move to SpecTP: it moves there in order to be in a position from which headmovement (or perhaps feature-movement) to AgrS can take place. On this view, then, the position of the subject in SpecTP follows from the nature of AgrS. (We will not speculate here on how, but see Roberts 2000 for an account). 23. In this context, French is an interesting case. One possibility is to assume that French has AgrS* satisfied by V-movement (cf. Pollock 1989). Hence we expect to find null expletives. These are clearly found in Stylistic Inversion, illustrated in (i): (i)
a.
Ou est alle Jean? French where is gone John b. Je veux que soit precede au reexamen de cette question. / want that be proceeded to. the reexamination of this question (cf. Kayne and Pollock 1978, Pollock 1989, Taraldsen, this volume) The possibility of null expletives of this kind is predicted by our approach, since AgrS* is satisfied by V-movement. What remains unclear, however, is the precise nature of their distribution in relation to the overt expletive il. The alternative possibility is to assume that AgrS* in French is satisfied in the same way as in English—by merging a DP in the specifier. The presence of the overt expletive il, then is predicted, in the same way as the presence of there is in English. On this approach, (i) becomes problematic, unless we assume that Stylistic Inversion creates a context similar to V2, which allows for an alternative satisfaction of AgrS*. We leave this issue open to future research. 24. Miiller and Sternefeld (1993:490) observe that extraction is possible from exactly those clauses which are able to have embedded V2; this is true both for the V2 clauses themselves and for the nonV2 counterparts introduced by daft: (i)
Wen glaubst du [daB [jeder magf]]? German who believe you that everyone likes 'Who do you believe that everyone likes?' (ii) Wen glaubst du [mag [jeder t ]]? who believe you likes everyone 'Who do you believe everyone likes?' In Miiller and Sternefeld's terms the XP-fronting requirement is clearly satisfied by the (intermediate) trace of wen in (ii). What (i) shows is that C* can be satisfied by a complementizer or by V-movement exactly where CP is not an island for extraction. See Miiller and Sternefeld and the references given there for more discussion. We assume that nonbridge contexts are 'marked' in the sense that there is some operator (e.g., factive, negative, etc.) involved, blocking embedded V2 (which relates only to Fin) as well as extraction.
152
SUBJECTS, EXPLETIVES, AND THE EPP
25. Holmberg and Platzack (1995) assume that under a CP-recursion approach, symmetric V2 is derived on the basis that both C positions are specified as [+Fin]: the lower one is lexicalized under Vmovement, while the upper is lexicalized by means of an overt C. Our account is essentially a translation of theirs into our framework.
References Abney, Steven. 1987. The English Noun Phrase in Its Sentential Aspect. Ph.D. dissertation, MIT. Alexiadou, Artemis, & Elena Anagnostopoulou. 1998. Parametrizing AGR: Word order, Vmovement and EPP-checking. Natural Language & Linguistic Theory 16, 491-539. Bennis, Hans. 1986. Gaps and Dummies. Foris, Dordrecht. Besten, Hans den. 1983. On the interaction of Root Transformations and Lexical Deletive Rules. In Werner Abraham (ed.), On the Formal Syntax of Westgermania, John Benjamins, Amsterdam. Besten, Hans den. 1985. Some remarks on the Ergative Hypothesis. In Werner Abraham (ed.), Erkarende Syntax des Deutschen, 53-74. Narr, Tubingen. Borer, Hagit. 1986.1-subjects. Linguistic Inquiry 17, 375-416. Borsley, Robert, Maria-Luisa Rivero, & Janig Stephens. 1996. Long Head Movement in Breton. In Borsley & Roberts 1996, 53-74. Borsley, Robert, & Ian Roberts (eds.). 1996. The Syntax of the Celtic Languages. Cambridge University Press, Cambridge. BoSkovic, Zelko. 1997. The Syntax ofNonfinite Complementation: An Economy Approach. MIT Press, Cambridge, Mass. Brody, Michael. 1990. Some remarks on the Focus Field in Hungarian. UCL Working Papers in Linguistics 2, 201-225. Brody, Michael. 1997. Perfect chains. In Liliane Haegeman (ed.), Elements of Grammar: Handbook of Generative Syntax, 139-167. Kluwer, Dordrecht. Billing, Daniel. 1995. The 59th Street Bridge Accent: On the Meaning of Topic and Focus. Ph.D. dissertation, University of Tubingen. Cardinaletti, Anna. 1990. Impersonal Constructions and Sentential Arguments in German. Unipress, Padua. Cardinaletti, Anna. 1994. Subject positions. GenGenP 2.1, 64-78. Cardinaletti, Anna, & Ian Roberts. 1991. Clause structure and X-second. Ms., University of Venice and University of Geneva. Cardinaletti, Anna, & Michael Starke. 1994. The typology of structural deficiency. Ms., University of Venice and University of Geneva. Carnie, Andrew, Heidi Harley, & Elisabeth Pyatt. 1995. The resurrection: Raising to Comp? Some evidence from Old Irish. Proceedings of the 5th Annual Conference of the Formal Linguistics Society of Mid-America, 85-100. Indiana University Linguistics Club, Bloomington. Cheng, Lisa. 1991. On the Typology ofWh-questions. Ph.D. dissertation, MIT. Chierchia, Gennaro. 1995. The variability of impersonal subjects. In Emmon Bach, Eloise Jelinek, Angelika Kratzer, & Barbara Partee (eds.), Quantification in Natural Languages, 107-143. Kluwer, Dordrecht. Chomsky, Noam. 1982. Some Concepts and Consequences of the Theory of Government and Binding. MIT Press, Cambridge, Mass. Chomsky, Noam. 1995. The Minimalist Program. MIT Press, Cambridge, Mass.
THE EPP AS A CONDITION ON THE TENSE DEPENDENCY
153
Chomsky, Noam. 1998. Minimalist inquiries: The framework. MIT Occasional Papers in Linguistics 15, 1-56. Cinque, Guglielmo. 1999. Adverbs and Functional Heads: A Cross-Linguistic Perspective. Oxford University Press, New York. Cottell, Siobhan. 1995. The representation of tense in Modern Irish. GenGenP 3, 105-124. Davis, Henry. 1998. Person splits, ^-features and temporal architecture. GLOW Newsletter 40, 83-84. Diesing, Molly. 1992. Indefinites. MIT Press, Cambridge, Mass. Emonds, Joseph. 1978. The verbal complex V'-V in French. Linguistic Inquiry 9, 151-175. En?, Miirvet. 1987. Anchoring conditions for Tense. Linguistic Inquiry 18, 633-657. Ferraresi, Gisella. 1997. Word Order and Phrase Structure in Gothic. Ph.D. dissertation, University of Stuttgart. Giorgi, Alessandra & Fabio Pianesi. 1998. Tense and Aspect: From Semantics to Morphosyntax. Oxford University Press, New York. Grewendorf, Gunther. 1989. Ergativity in German. Foris, Dordrecht. Grimshaw, Jane. 1990. Argument Structure. MIT Press, Cambridge, Mass. Grimshaw, Jane. 1991. Extended Projection. Ms., Brandeis University. Haegeman, Liliane. 1995. The Syntax of Negation. Cambridge University Press, Cambridge. Haider, Hubert. 1993. Deutsche Syntax Generativ. Gunter Narr Verlag, Tubingen. Haider, Hubert, & Martin Prinzhorn. 1985. Verb Second Phenomena in Germanic Languages. Foris, Dordrecht. Heim, Irene. 1982. The Semantics of Definite and Indefinite Noun Phrases. Ph.D. dissertation, University of Massachusetts, Amherst. Higginbotham, James. 1985. On semantics. Linguistic Inquiry 16, 547-593. Higginbotham, James. 1993. Interrogatives. In Kenneth Hale & Samuel Jay Keyser (eds.), The View from Building 20, 195-228. MIT Press, Cambridge, Mass. Holmberg, Anders, & Christer Platzack. 1995. The Role of Inflection in Scandinavian Syntax. Oxford University Press, New York. Huang, C.-T. James. 1984. On the distribution and reference of empty pronouns. Linguistic Inquiry 15, 531-574. Kayne, Richard S. 1984. Connectedness and Binary Branching. Foris, Dordrecht. Kayne, Richard S. 1989. Null subjects and clitic climbing. In Osvaldo Jaeggli & Ken Safir (eds.), The Null Subject Parameter, 239-261. Kluwer, Dordrecht. Kayne, Richard S. 1991. Romance clitics, verb movement and PRO. Linguistic Inquiry 22, 647-686. Kayne, Richard S. 1994. The Antisymmetry of Syntax. MIT Press, Cambridge, Mass. Kayne, Richard S., & Jean-Yves Pollock. 1978. Stylistic Inversion, successive cyclicity, and Move NP in French. Linguistic Inquiry 9, 595-621. Kiparsky, Paul. 1994. Indo-European origins of Germanic Syntax. In Adrian Battye & Ian Roberts (eds.), Clause Structure and Language Change, 140-169. Oxford University Press, New York. Kiss, Katalin E. (ed.). 1995. Discourse Configurational Languages. Oxford University Press, New York. Koopman, Hilda. 1992. The internal and external distribution of pronominal DPs. Ms., UCLA. Lebeaux, David. 1986. The interpretation of derived nominals. In A. M. Farley, P. T. Farley, & K.-E. McCullogh (eds.), Papers from the 22nd Annual Regional Meeting of the Chicago Linguistics Society 22, 231-147. Chicago Linguistics Society, Chicago.
154
SUBJECTS, EXPLETIVES, AND THE EPP
Longobardi, Giuseppe. 1994. Reference and proper names. Linguistic Inquiry 25, 609-665. Manzini, M. Rita. 1995. From Merge and Move to Form Dependency. UCL Working Papers in Linguistics 7, 323-346. Manzini, M. Rita. 1996. Syntactic dependencies and their properties: Weak islands. In Artemis Alexiadou & T. Allan Hall (eds.), Studies on Universal Grammar and Typological Variation, Mouton de Gruyter, Berlin. Manzini, M. Rita, & Anna Roussou. 2000. A Minimalist theory of control and A-movement. Lingua 110, 409-447. McCloskey, James. 1996. On the scope of V-movement in Irish. Natural Language & Linguistic Theory 14, 47-104. Montalbetti, Mario. 1984. After Binding: The Interpretation of Pronouns. Ph.D. dissertation, MIT. Miiller, Gereon, & Wolfgang Stemefeld. 1993. Improper movement and unambiguous Binding. Linguistic Inquiry 24, 461-507. Partee, Barbara. 1984. Nominal and temporal anaphora. Linguistics and Philosophy 7, 243-286. Pesetsky, David. 1982. Paths and Categories. Ph.D. dissertation, MIT. Pesetsky, David. 1989. Language particular processes and the Earliness Principle. Ms., MIT. Pesetsky, David. 1991. Zero Syntax II: An essay on infinitives. Ms., MIT. Philippaki-Warburton, I. 1985. Word order in Modern Greek. Transactions of the Philological Society 2. Platzack, Christer. 1987. The Scandinavian languages and the Null Subject Parameter. Natural Language & Linguistic Theory 5, 377-401. Platzack, Christer. 1998. A visibility condition for the C-domain. Working Papers in Scandinavian Syntax 61, 53-99. Pollock, Jean-Yves. 1986. Sur la syntaxe de en et le parametre du suject nul. In M. Ronat & D. Couquaux (eds.), La Grammaire Modulaire. Editions de Minuit, Paris. Pollock, Jean-Yves. 1989. Verb movement, Universal Grammar, and the structure of IP. Linguistic Inquiry 20, 365-424. Reinhart, Tanya. 1995. Interface strategies. OTS Working Papers 95-002, Utrecht University. Reuland, Eric. 1983. Governing -ing. Linguistic Inquiry 14, 101-136. Rivero, Maria-Luisa. 1991. Long head movement and negation: Serbo-Croatian vs Slovak and Czech. The Linguistic Review 8, 319-351. Rivero, Maria-Luisa. 1993. Finiteness and second position in long head movement languages: Breton and Slavic. Ms., University of Ottawa. Rizzi, Luigi. 1982. Issues in Italian Syntax. Foris, Dordrecht. Rizzi, Luigi. 1991. Residual Verb Second and the w/i-Criterion. Technical Reports in Formal and Computation Linguistics 2, University of Geneva. [Reprinted in Adriana Belletti & Luigi Rizzi 1997 (eds.), Parameters and Functional Heads, 63-90. Oxford University Press, New York.] Rizzi, Luigi. 1997. The fine structure of the left periphery. In Liliane Haegeman (ed.), Elements of Grammar: Handbook of Generative Syntax, 281-337. Kluwer, Dordrecht. Roberts, Ian. 1993. Verbs and Diachronic Syntax. Kluwer, Dordrecht. Roberts, Ian. 1994. Two types of head movement in Romance. In David Lightfoot & Norbert Hornstein (eds.), Verb Movement, 207-242. Cambridge University Press, Cambridge.
THE EPP AS A CONDITION ON THE TENSE DEPENDENCY
155
Roberts, Ian. 1997. Restructuring, Head Movement and Locality. Linguistic Inquiry 28, 423-460. Roberts, Ian. 2000. Principles & Parameters in a VSO language: A case study in Welsh. Ms. University of Stuttgart. Roberts, Ian, & Anna Roussou. 1997. Interface interpretation. Ms., University of Stuttgart and University of Wales, Bangor. Roberts, Ian, & Ur Shlonsky. 1996. Pronominal enclisis in VSO languages. In Robert Borsley & Ian Roberts (eds.), The Syntax of the Celtic Languages, 171-199. Cambridge University Press, Cambridge. Rosenbaum, Peter. 1967. The Grammar of English Predicate Complement Constructions. MIT Press, Cambridge, Mass. Ross, John Robert. 1982. Pronoun deleting processes in German. Paper presented at the annual meeting of the Linguistics Society of America, San Diego. Roussou, Anna. 1994. The Syntax of Complementisers. Ph.D. dissertation, University College London. Roussou, Anna. In press. Control and Raising in and out of subjunctive clauses. In Brian Joseph, Angela Ralli, and Maria-Luisa Rivero (eds.), Comparative Syntax of Balkan Languages, 74-104. Oxford University Press, New York. Schafer, Robin. 1994. Nonfinite Predicate Initial Constructions in Modern Breton. Ph.D. dissertation, University of California, Santa Cruz. Schwartz, Bonnie, & Sten Vikner. 1997. The verb always leaves IP in V2 clauses. In Adriana Belletti & Luigi Rizzi (eds.), Parameters and Functional Heads, 11-62. Oxford University Press, New York. Sportiche, Dominique. 1992. Clitic constructions. Ms., UCLA. Stephens, Janig. 1982. Word Order in Breton. Ph.D. dissertation, University College London. Stowell, Tim. 1981. The Origins of Phrase Structure. Ph.D. dissertation, MIT. Terzi, Arhonto. 1992. PRO in Finite Clauses: A Study of the Inflectional Heads of the Balkan Languages. Ph.D. dissertation, CUNY. Tomaselli, Alessandra. 1989. La sintassi del verbofinito nelle lingue germaniche. Ph.D. dissertation, University of Pavia. Tsimpli, lanthi-Maria. 1995. Focusing in Modem Greek. In Katalin E. Kiss (ed.), Discourse Configurational Languages, 176-206. Oxford University Press, New York. Vikner, Sten. 1995. Verb Movement and Expletive Subjects in the Germanic Languages. Oxford University Press, New York. Willis, David. 1996. The Loss of Verb Second in Welsh: A Study of Syntactic Change. Ph.D. dissertation, Oxford. Zwart, C. Jan-Wouter. 1997. Morphosyntax of Verb Movement: A Minimalist Approach to the Syntax of Dutch. Kluwer, Dordrecht.
This page intentionally left blank
7 Parameters of Subject Inflection in Italian Dialects M. Rita Manzini and Leonardo M. Savoia
1. The null-subject parameter In this chapter we consider the null-subject parameter, with the implications it has for Chomsky's 1995, 1998 theory of the EPP and agreement. We take the view that null-subject languages lexicalize the EPP feature through the finite verb (cf. Pollock 1996), while non-null-subject languages lexicalize the EPP feature through a D head or a DP. We further devise a derivation whereby a thematic feature (or, to be precise, an aspectual one) is attracted by D in accordance with the model of Manzini and Roussou 1999. This allows us to dispense with the empty category pro. We argue that in expletive constructions without agreement in Number and Gender, the associate nevertheless agrees in Person; this points to the conclusion that the associate always agrees with the verb and with the expletive. Therefore we propose a new model of agreement which relies on the same aspectual feature being attracted by several positions endowed with nominal features. Sharing of Person, Number, and Gender features in this model is a reflex of the sharing of the aspectual feature. Contra Chomsky 1995, 1998, we do not have recourse to noninterpretable features (cf. Brody 1997). Furthermore, while the classical conception of the null-subject parameter provides the canonical model of a macroparametric view (Baker 1996), the view that emerges from our study is strictly microparametric, in the sense that underlying structures and derivations are universal and only lexicalization choices can differ from language to language. 1.1. Background Classical discussions of the so-called pro-drop or null-subject parameter (Taraldsen 1978, Chomsky 1981,1982, Rizzi 1982) take as their starting point the behavior of subjects in two different types of languages: on the one hand, languages like English; on the other hand, languages like Italian. According to Rizzi 1982, there
157
158
SUBJECTS, EXPLETIVES, AND THE EPP
are three properties concerning the syntax of the subject that differentiate these two types of languages, as summarized in (1). (1)
a.
b.
c.
Optional vs. obligatory presence of a preverbal subject Italian: Gianni dorme. Gianni sleeps Dorme. sleeps Piove. rains English: John is sleeping. * (He) is sleeping. * (It) is raining. Possibility vs. impossibility of so-called free inversion of the subject Italian: Arriva Gianni. arrives Gianni English: * Is coming John. Possibility vs. impossibility of extracting a subject across a thattype complementizer: Italian: Chi pensi che telefoni? who think that telephones English: Who do you think (*that) is calling?
In the empirical domain that we are focusing on here, that of Italian varieties, the properties just observed for Standard Italian hold for all Southern Italian dialects, as exemplified in (2): (2)
a.
b.
c.
'r addsa arra'piorta it [I] have opened a ddur'mu:ts [s/he] has slept mi Vijturu me [they] saw ma fau 'mob ra 'pista to. me make pain the feet 'My feet hurt' a vs'nuits 'fijpte has come your.child 'ruammunu i pittfi'ridjdj sleep the children 'rornrni u pittji'ric|c|u sleeps the child ki tta 'kri:5a ka 've? who you believe that comes a kka tts 'kre99S3 ka VinaSs? to who you believe that comes
Gallo Taranto Catenanuova Gallo
Taranto Modica
Gallo Rocca Imperiale
PARAMETERS OF SUBJECT INFLECTION IN ITALIAN DIALECTS
ku 'kriri ka 'veni? who [you] believe that comes
159 Modica
Within the framework defined by the parameter in (1), Northern Italian dialects pose some questions of considerable interest, as initially pointed out by Brandi and Cordin (1981, 1989). On the one hand, these dialects require the subject to be present, since in the counterpart to examples such as (2a) the subject is obligatorily realized in the shape of a subject. On the other hand, a comparison with non-nullsubject languages such as English reveals at least one important difference with respect to them. Indeed, in Northern Italian dialects the subject clitic systematically doubles a preverbal lexical subject, as illustrated in (3), where each dialect corresponds to a major geographical region: (3) 1 to'zat 1 'mondsa 'dut Andraz the child CiS eats all il ka'nai al 'duar 'puak Vito d'Asio the child CLS sleeps little Casaccia-Val Bregaglia la Ma'riala 'marjga the Mary CiS eats Casorezzo kel 'la al 'do:rmi 'semper that one CiS sleeps always kli 'doni 'li li m 'tfama Revere those women CiS me call Alfonsine kal ta'baki a/( e av'nudi a ka 'mi those children CiS have come to home my MEzzenile la fi'0ta e 'dysrt the child CiS sleeps Castellazzo Bormida 'kil ir 'beif he CiS drinks a Ma'ria a 'duorme 'puoku Fontanigorda the Mary CiS sleeps little la mi fiX'Xola la 'dorme 'Groppo Firenze the my daughter CiS sleeps too.much According to Chomsky 1995,1998 the basic structure of the sentence is as in (4): (4)
In such a structure, a lexical subject fills the position of Sped. In Italian, according to Rizzi 1982, Spec! can be occupied by an empty category licensed by the rich
160
SUBJECTS, EXPLETIVES, AND THE EPP
agreement morphology of the verb. Since this empty category has pronominal properties, Chomsky (1982) suggests the term pro to refer to it. By contrast, in a language like English the subject position must be lexically filled in that the (J)-features of I are not sufficient to identify pro. Under these assumptions, the classical analysis of the null-subject parameter derives properties (la) and (Ib). Indeed, to the extent thai pro is available in Italian but not in English, it is available to express both a subject pronoun with argumental properties, as in (la), and a subject pronoun with nonargumental or expletive properties, as in (Ib). For Rizzi (1982), furthermore, (Ic) is just a by-product of (Ib). In particular the grammatical Italian sentence is just a concealed expletive sentence, where the subject is extracted not from the preverbal but from the postverbal position. Brandi and Cordin (1981, 1989) and Rizzi (1986) propose that Northern Italian dialects are essentially null-subject languages. The differences between languages like Standard Italian and these dialects are imputed to the existence of additional inflectional material realized by the subject clitic. Cardinaletti and Roberts (1991) identify the relevant inflectional position as an independent head AgrS. Chomsky (1995) criticizes the postulation of a category Agr whose sole contribution to the derivation is mediating an agreement relation between two other categories (i.e., the finite verb and the subject in the case of AgrS). Nevertheless, even taking this criticism of the category Agr into account, a number of facts point to the conclusion that the subject clitic of Northern Italian dialects is realized as an independent head. Perhaps the strongest argument is that in questions and similar V2 environments, raising of the finite V to C strands the subject clitic to its right (Manzini and Savoia 1998a and references quoted there). This means that the subject clitic cannot simply be adjoined to I, at least if we assume that excorporation is banned on grounds of restrictiveness of the theory. Taking the evidence just mentioned into account, we postulate the existence of a functional category which immediately dominates I and is morphologically realized by the subject clitic; following Manzini and Savoia 1997 we refer to this category simply as D, as in (5). Note therefore that we do not assume the existence of a special inflectional category to host the D feature; on the contrary, D is the same category as we find in the highest position of nominals. Any differences between the two can be taken to be a consequence of the different environments they find themselves in.1 (5)
kulo 'dono ro 'drwom that woman CiS.SpSF sleeps.
Castellazzo Bonnida
PARAMETERS OF SUBJECT INFLECTION IN ITALIAN DIALECTS
161
In (5) the subject clitic appears as an independent inflectional head, while the lexical subject appears in SpecD. For the time being, the head-Spec relation can be taken to explain agreement in Number, Gender, and Person of the clitic with the lexical subject; we shall, however, return to the position of the lexical subject in section 1.3. As usual, the verb in (5) moves to the inflectional position I, attracted by a strong V feature. As for the agreement relation between the two inflectional heads, D and I, we can assume that it is the by-product of movement of the lexical subject through the Spec! position. Again we shall return to this mechanism in section 2. The null-subject character of Northern Italian dialects corresponds to the possibility of having sentences similar to (5) in all respects, except that the SpecD position is not lexically realized, as for instance in (6): (6)
ro 'dr worn CiS. 3 PSF sleeps
Castellazzo Bormida
1.2. Elimination o/pro and formulation of the parameter In terms of the classical analysis of the null-subject parameter, in (6) the SpecD position is filled by the empty category pro. Note, however, that in the case of Northern Italian dialects pro doubles the features Person, Number, and Gender independently present on the subject clitic. Assuming that the subject clitic, being an inflectional category, is unable to be associated with a thematic role, pro still has the function of bearing the thematic role itself; but even this is not true of expletive pro sentences (cf. Borer 1986). In fact, proposals in Manzini and Roussou 1999 and Manzini and Savoia 1997 call into question this last remaining role for pro, as a conveyor of thematic properties. Remember that in the framework of Chomsky 1981, a level of syntactic representation, D-S, was postulated which was conceived as a pure representation of thematic relations, prior to any movement. Thus in such a framework, the architecture of the grammar itself required that an argument be inserted from the lexicon into each thematic position; from there it would be taken to its surface position by a subsequent operation of movement, to eventually produce another level of representation, S-S. Within the Minimalist model, however, there is no obstacle in the architecture of the grammar itself to assuming that arguments are inserted not in thematic position, but directly in the position where they surface. Indeed, abstract movement of a thematic feature to the domain of the argument is sufficient to ensure the correct interpretation of the sentence. Concretely, within the framework of Chomsky 1981, 1986, the derivation of a simple sentence like John sleeps could proceed only as in (7a) for reasons inherent to the architecture of the system. At a first stage of the derivation, which does not include movement, there is a pairing of the argument with a vP-internal 0-position; at a second stage, the argument is moved to Spec! (i.e., copied into this position, given a copy theory of movement). Within the Minimalist framework, an alternative becomes available, as detailed in (7b). The argument is merged directly into the Spec! position in overt syntax, while an operation on features associates the thematic feature of V with the argument. We provisionally notate this operation as a feature movement operation.
162 (7)
SUBJECTS, EXPLETIVES, AND THE EPP a.
[IP I [vP John [VP slept]]] [IP John I [vP Jefeft [Vp slept]]]
b.
[IP John I [Vp slept (0)]] [IP 6-John I [VP slept (0)]]
In Chomsky's 1995, 1998 grammar the derivation in (7b) is barred by additional assumptions, in particular the assumption that 0-roles are not features, but configurations; this in turn forces the principle that "Pure Merge in 0-position is required of (and restricted to) arguments" (Chomsky's 1998 (6)). But there is no empirical loss in assuming that 0-roles are features (Hornstein 1999) and that they can in principle be associated by an operation on features with an argument generated in 0-position; on the contrary, there is considerable empirical gain in making this assumption. Indeed, Manzini and Roussou (1999) argue that the difference between (7a) and (7b) can easily be tested at the interfaces, where the lexical material in (7a) is predicted to give rise to reconstruction effects at LF, and to the blocking of phonosyntactic processes at PR That A-movement does not reconstruct is explicitly recognized in Chomsky 1995; that traces do not block phonosyntactic processes is a point variously made as part of the debate on wanna-contraction (cf. van Riemsdijk and Williams 1984 and references quoted there). Thus the available evidence argues in favor of (7b). In short, the choice between (7a) and (7b) was forced by the overall architecture of the theory within the first Principles and Parameters framework. However, it becomes a wholly empirical matter within the Minimalist framework. Following Manzini and Roussou (1999), we can conclude that empirical considerations favor (7b). Following Borer 1994 and Arad 1998, we assume that thematic properties are aspectual in nature. In particular, we indicate with MEAS the Measure of the event, corresponding in thematic terms to the theme/patient, and with OR the Originator of the event, corresponding in thematic terms to the agent. With this background, we can then return to pro. On the basis of the model just proposed, the doubling of D by pro is not necessary to obtain an argumental interpretation for the inflectional head. To such end it is sufficient to postulate that an aspectual feature MEAS/OR is transformationally associated with D. Technically, the feature is ATTRACTED by D. As defined in Manzini and Roussou 1999 and Manzini and Savoia, forthcoming, the operation of attraction belongs to the same family as Chomsky's 1995 ATTRACT or Chomsky's 1998 AGREE, in that it is an operation at a distance on features. At the same time it differs from those rules in that we assume that it consists wholly in the formation of an ordered set of features, in the case at hand
PARAMETERS OF SUBJECT INFLECTION IN ITALIAN DIALECTS
163
(8)
ro 'drwom CiS sleeps
Castellazzo Bormida
Northern Italian dialects do not differ on this point from classical null-subject languages like Standard Italian or Southern Italian dialects. Indeed, if we assume that the basic functional structure of the sentence is universal and does not vary across languages, as independently argued by Cinque (1999), then Southern Italian varieties also include a D position, though not an overtly realized one. If so, a standard null-subject sentence will be derived by attraction of an aspectual feature to D, as illustrated in (9):
(9)
Dorme. sleeps
Italian
We have already seen that English also presents derivations of the type of (7b), in which the lexical subject is merged directly in its surface position and the aspectual feature is attracted to its domain. If the basic structure of the sentence is universal and D is therefore present in classical non-null-subject languages like English as well, then the derivation in (7b) need only be modified so that John is merged in SpecD, as in (10): (10)
[DP OR-John D [IP I [VP slept (Oft)]]]
In terms of Chomsky's 1995 theory of parameters, it is now possible to propose that D is strong in Northern Italian dialects as well as in English; by contrast, it is weak in Standard Italian and Southern Italian dialects, essentially as proposed by Platzack (1995). Classical non-null-subject languages such as English have a strong D but do not possess a specialized D element to check it. Therefore the strong D implies that a DP is merged into SpecD. In Northern Italian dialects, on the other hand, the strong D feature is checked by merger of the subject clitic (i.e.,
164
SUBJECTS, EXPLETIVES, AND THE EPP
by a specialized D head). Suppose in turn that the strong/weak parameter of Chomsky 1995 reduces to whether D is realized at the PF interface or not (cf. Roberts and Roussou, this volume). As before, we predict the existence of languages that must realize D but do not possess a specialized D element to realize it. Classical non-null-subject languages such as English fall into this category. In such languages DP is merged into SpecD to realize D. In Northern Italian dialects, on the other hand, D is realized by merger of the subject clitic (i.e., by a specialized D head). Obviously enough, if the content of the D position reduces to the D feature, which is in fact the conclusion we reach in section 2, the parameter concerning the realization or not of D at PF can be stated for the position or for the feature equivalently. As suggested by Platzack (1995), classical null-subject languages such as Standard Italian and Southern Italian dialects represent the case in which D is weak or, as we can now also say, they need not be realized at the PF interface. This means, on the one hand, that DP need not be merged into SpecD position, contrary to English and similar languages; it also means that, contrary to Northern Italian dialects, there are no specialized subject clitics merged in D. An alternative to the view just proposed, which also implies the elimination of pro, is set out in recent work by Pollock (1996), Nash and Rouveret (1997), and Alexiadou and Anagnostopoulou (1998). These authors assume, with Chomsky (1995), that the D feature of I is universally strong, or, in other terms, that the EPP is universal. The difference between languages like English and Italian is that in Italian raising of V to I is sufficient to check its strong D feature, while in English this is not the case. As Pollock (1996) acknowledges, this analysis represents a translation into Minimalist terms of the traditional idea that the null-subject parameter depends on the richness of verbal inflection. Classical theories of the nullsubject parameter had obvious difficulties in maintaining the desired correlation between the morphological richness of I and the lack of realization of the subject (Jaeggli and Safir 1989). Pollock (1996) derives a different generalization from his theory, namely that only languages where V moves to I belong to the null-subject typology. Within our theory, overt raising of I to D for purposes of feature checking can be excluded on empirical grounds. To see this, consider the negation clitic, which is generally agreed to correspond to an independent head NEC (Manzini and Savoia 1998b, and references quoted there). The Neg clitic clearly appears to the left of I and to the right of D in dialects which realize both (see (29)-(30) below for some examples); hence, in dialects which do not independently realize D, the fact that the finite verb appears to the right of Neg can be taken to mean that the verb has not raised to D. One way to maintain the universal EPP is to say that D is a category peculiar to Northern Italian dialects. But this would give up the highly restrictive assumption that the grid of functional categories is universal, for no apparent empirical gain. There is, however, a way of maintaining what amounts to the traditional construal of the pro-drop parameter that we intend to adopt. At the same time we intend to put forward a view of parameters as lexicalization choices even more restrictive than the one sketched in adapting the strong/weak parameter of Chomsky 1995. There we suggested that the same position/feature may be lexicalized in one
PARAMETERS OF SUBJECT INFLECTION IN ITALIAN DIALECTS
165
language but not in another. We now suggest instead that two languages can differ only as to how they lexicalize a given feature. We can assume that the null-subject parameter, as defined between Standard Italian and English, reduces to whether the D feature is lexicalized by the finite verb (Italian) or not (English). In the first case, normal economy principles dictate that the same feature should not be lexicalized again; we shall return immediately below (section 1.3) to the special problems posed by the apparently optional preverbal subject. In the second case, the D feature necessitates independent realization, in which case a language like English has recourse to the merger of a DP in SpecD. The similarity between Standard Italian and Northern Italian dialects suggested to early theorists that they should be treated alike from the point of view of the null-subject parameter. In the terms being suggested here, this amounts to saying that in Northern Italian dialects it is the finite verb that lexicalizes the D feature; but this directly contradicts the idea suggested so far that the subject clitic in these languages is nothing but a pure lexicalization of the D feature. Let us then continue to assume that subject clitics are as realizations of the D feature. Thus, contrary to null-subject languages like standard Italian, Northern Italian dialects do not lexicalize D through the finite verb; and contrary to classical non-null-subject languages they have a specialized element, namely the clitic head, to realize D. Again we shall return in section 1.3 to the problem posed by the doubling of the clitic by the lexical subject. An interesting question that arises in connection with the null-subject parameter concerns the status of a language like (Standard) French, which is classically treated as a non-null-subject language of the same type as English, that is, in present terms, as a language where D is lexicalized by a DP in SpecD, be it lexical or pronominal. In reality, there are a number of reasons why this characterization should be doubted; in particular, French does have a set of elements that are at least superficially characterizable as subject clitics. Kayne (1975) proposes that the pronominal subjects of French are in fact clitics, but only at the PF interface; in other words, they are syntactically represented as DP and cliticize only in the phonology. The main obstacle to construing them as syntactic clitics (i.e., as pronominal heads rather than as pronominal DPs) is represented by their inability to be doubled by lexical DPs, in contrast with the behavior just described for Northern Italian dialects. A careful analysis of Northern Italian dialects, however, undermines this classical approach. In particular, it is interesting to note that a number of Ladin dialects of the Dolomitic area of Italy are characterized by conditions intermediate between those of typical Northern Italian dialects and those of French, as detailed in Savoia and Manzini, forthcoming, and as noted in Vanelli 1998:118. Thus, at Colfosco in (11) we find a different realization for subject clitic forms and full pronominal DPs in the first and second persons (though in the plural there is simply nonlexicalization of the subject clitic3). As expected, in the first and second persons the clitic and full forms can co-occur. In this respect, therefore, Colfosco is like a Northern Italian dialect. In the third person, however, we find a single form of pronominal subject. On the one hand, this does not double a lexical DP; on the other hand, it is obligatory in the absence of one, exactly as happens in French:
166
SUBJECTS, EXPLETIVES, AND THE EPP
(11) a.
c.
i 'dorme b. ju (i) 'dorme Colfosco - Val Badia ts 'donnas tu (ta) 'dormas al/ala 'dorm al/ala 'dorm dor'mjon nos dor'mjon dor'mirze os dor'mi:ze al/ala 'dorm al/ala 'dorm CiS sleep(s) PN (CiS)sleep(s) i mi'torjs (*al) Vaji de'do: the boys (CiS) come afterward
In other dialects of the same area, for instance at La Pli de Mareo, there are two different series of pronominal subjects, namely a subject clitic form and a full pronominal form; the two never co-occur, as can be seen by the comparison between (a) and (b) in (IT). What is more, the subject clitic never co-occurs with a lexical DP subject, as in (ll'c); in questions, subject clitics appear to the immediate right of the verb in C while lexical DPs are in peripheral position, as in (ll'd). In other words, this dialect reproduces the same conditions as French, except that the two series of pronominal forms for subjects unambiguously identify subject clitics with respect to full pronominal DPs. (11') a.
c. d.
i 'dormi b. ju te 'dormes t0 al/ara 'dorm el/era i dor'mjurj nos i dor'mi:s vos ai/ares 'dorm ei/eres CiS sleep(s) PN i mi'turjs 'dormia de la the children sleep in there a-i 8or'mi i mi'turjs? have-CiS slept the children
'dormi 'dormes 'dorm dor'mjurj dor'mi:s 'dorm sleep(s)
La Pli de MareoVal Badia
This set of observations leads us to conclude that languages like (Standard) French or the Ladin dialects in (ll)-(ir) are characterized by a realized D at PF. However, they represent the case in which this can be realized either by a specialized subject clitic in the D head or by a full DP in SpecD. On grounds of economy, we can assume that the PF realization of D by both the specialized clitic and by a full DP is not allowed. Indeed, in section 1.3 we shall modify our analysis of typical Northern Italian dialects to the effect that head-Spec configurations involving a subject clitic and a lexical DP are excluded. Thus, we shall uphold Koopman's 1996 Generalized Doubly Filled Comp Filter, according to which either a head X or its Spec can be realized, but not both. Summarizing so far, our theory allows for four different types of languages, according to whether the D feature is lexicalized by the finite verb (Italian), by a specialized head (Northern Italian dialects), by a full DP (English), or by either a specialized head or a full DP (French). This theory can therefore be schematized as in (12):
PARAMETERS OF SUBJECT INFLECTION IN ITALIAN DIALECTS
167
(12) D lexicalized by: V Italian D Northern Italian dialects D, DP Ladin dialects, French DP English Manzini and Savoia, forthcoming, develops a finer-grained system of subject clitic projections (for a different system see also Poletto 1993, 1997), which we shall sketch in section 2. It is in terms of this system, which replaces D with a set of positions including P(erson) and N(umber), that the null-subject parameter is ultimately stated; thus the D feature can eventually be lexicalized (or not lexicalized) in each of the finer-grained inflectional positions in any given language (cf. note 3). 1.3. The position of the preverbal and postverbal subject The systematization of the null-subject parameter in (12) leaves a number of questions open. The first one concerns the position of the lexical subject. Indeed, in English and similar languages the preverbal subject lexicalizes D and is therefore merged, in present terms, in SpecD. However, in Northern Italian dialects D is lexicalized by the subject clitic, while in Standard Italian and similar languages, D is lexicalized by the finite verb, excluding the merger of DP in SpecD under ordinary economy considerations. What is more, the position of the postverbal subject remains to be defined entirely. We shall begin by considering standard theories of postverbal subjects in classical null-subject languages such as Standard Italian or Southern Italian dialects. According to the generally accepted theory, the unique argument of unaccusative verbs is an internal argument, similar to the object of transitive verbs. As proposed by Burzio (1986), therefore, its postverbal position can correspond simply to its thematic position, which we identify with the complement of V, as in (13): (13)
Viene Gianni. comes Gianni
Italian
By contrast, with unergative verbs, the unique argument has the properties of the external argument of transitive verbs. Once again, we can identify its postverbal position with its thematic position, if we take the latter to be SpecV, in accordance with the so-called VP-Internal Subject Hypothesis. This gives rise to structures of the type in (13'):
168
SUBJECTS, EXPLETIVES, AND THE EPP
(131)
Dorme Gianni. sleeps Gianni
Italian
Transitive verbs are more problematic if the postverbal subject essentially coincides with a thematic position. Indeed, transitive verbs display an SOV order, as in (14), while one would expect the VSO order if the subject is in SpecV and the object is in the complement of V: (14)
Scrisse una lettera tuo fratello. wrote a letter your brother
Italian
To explain this kind of data, it is necessary to assume that the object in (14) is not in its thematic position, but in some intermediate position between I and V, as illustrated in (15):
(15)
Note that though postverbal subject sentences are truth-functionally equivalent to their counterparts with preverbal subjects, there still is an interpretive difference between them. The preverbal subject generally corresponds to the Topic, or one of the Topics, of the sentence, while the postverbal subject is either the Focus or part of the Focus of the sentence. This is consistent with the idea that the sentential Focus is the element or elements bearing main stress in a sentence, which in turn corresponds to the most deeply embedded constituent, or constituents, in the sentence itself (Cinque 1993, Zubizarreta 1998). If so, VP-internal subjects are in Focus position, while VP-external subjects are Topics. Given these basic facts, it is perfectly possible to suppose that the realization of the subject in VP-internal position corresponds to the realization of a F(ocus) feature rather than an aspectual one.
PARAMETERS OF SUBJECT INFLECTION IN ITALIAN DIALECTS
169
If this Focus feature is associated with a functional category, then the postverbal subject in (13)-(15) should be in the Spec of such a category. Independent evidence for a low Focus projection hosting adverbs is indeed provided by Cinque (1999). As for the XP position in (15), our idea is that it can only be a Focus position once again. In effect, sentences of the type in (14) are felicitous answers to questions such as 'What happened?' that bear on the whole predicate. Thus the XP position in (15) is not to be assimilated to the target of object movement in proposals such as that of Johnson 1991. Though we agree with the latter that the object always surfaces in a position higher than its thematic position, we do not believe that this is the case because of the need to realize Agreement or Case or, in general, inflectional features. If the two DP positions in (15) are both inside the sentential Focus, there is no reason to exclude the possibility that the order of subject and object could be reversed. In effect, VSO orders are also found in languages like Italian, precisely with a double Focus interpretation, as for instance in (14'): (14')
Scrisse anche tuo fratello una lettera. wrote also your brother a letter
Italian
The discussion that precedes presupposes that on grounds of economy, the same Focus position cannot be realized by two distinct elements; in other words, multiple Specs are an impossibility (contra Chomsky 1995, 1998). Thus the object cannot be merged inside the Focus projection in addition to the subject, but rather must be generated in its own projection, precisely as indicated in (15). An analysis of the preverbal subject position of Italian is now also at hand. In particular we can assume that since the preverbal subject is topicalized, its position reflects the realization of a Topic feature. This and the previous analyses, though developed for Italian and more generally for classical null-subject languages, apply equally in Northern Italian dialects with subject clitics. In particular, nothing needs to be added to what has already been said about the postverbal position of the subject in Focus. As for the preverbal subject position, it is possible to argue that the subject clitic and the lexical subject never in fact appear in a head-Spec configuration, upholding Koopman's 1996 Generalized Doubly Filled Comp Filter (according to which either a head X or its Spec can be realized, but not both). Thus Manzini and Savoia 1998a and 1998b, to which we refer for a more detailed discussion, identify the position of the preverbal subject with aTop(ic) projection (i.e., SpecTop). Merger of the lexical subject in this position is unrelated to the realization of D; rather, it responds to the need for a Topic feature to be realized. This last analysis extends to subject clitic languages and to classical null-subject ones. Only preverbal subjects in languages like English are thus a means of satisfying the EPP; preverbal subjects in Italian and Italian dialects are entirely unconnected to this property.4 As for the topicalization of an object, the normal means of accomplishing it is Clitic Right or Left Dislocation, in the sense of Cinque 1990. Indeed note that for all that we have said so far, there is a fundamental asymmetry between subject and object, in that there is an inflectional projection D for the former, but no inflectional position for the latter. This account is sufficient for the data considered here, but it quickly becomes inadequate if we consider the full spectrum of data concerning the
170
SUBJECTS, EXPLETIVES, AND THE EPP
object; in particular the object, like the subject, can correspond to a clitic or it can be dropped. A theory of object inflections appears therefore to be required, parallel to the theory of subject inflections considered here but distinct from it so as to be able to account for various residual asymmetries between the two systems. Such a theory, which is discussed in Manzini and Savoia 1998c and Manzini and Savoia, forthcoming, is however outside the scope of the present article.
2. Agreement parameters with postverbal subjects 2.1. The problems In all of the examples considered so far, the preverbal subject agrees in Person and Number with the finite verb and in Person, Number, and Gender with the subject clitic, when the latter is present. Thus agreement with the preverbal subject is obligatory, independent of the null-subject, non-null-subject, or subject-clitic status of the language. Consider, for instance, agreement of the preverbal subject with both the finite verb and the subject clitic in Northern Italian dialects. Under the construal of A-movement in Chomsky 1995, this presents no particular difficulty. Quite simply, one can assume that on its way to its surface position in SpecD, the DP subject is attracted by the intermediate Spec! position, where it enters agreement with the I head. As for Chomsky 1998, the specialized rule Agree applies between the noninterpretable ())-feature set of the verb and the uninterpretable nominative Case feature of the DP; movement can be seen as a consequence of this operation. The question then arises whether these derivations can be replicated in a system in which it is the aspectual feature that is attracted by the base-generated DP. Before answering this question, it is worth considering the full range of agreement phenomena that characterize not only preverbal but also postverbal subjects. As is well known from Chomsky's 1995 discussion of English and French, in English postverbal subjects agree with the finite verb, as in (16), while in French the finite verb appears to agree with the expletive in the third-person singular rather than with the postverbal subject, as in (17) (the English examples also provide the translations of the French ones): (16) a. b.
There comes a boy. There come some boys.
(17) a.
II vient un enfant. it comes a child II vient des enfants. it comes of. the children
b.
French
Again the agreement parameter observed in (16)-(17) is entirely independent of the null-subject parameter as defined in (12). Indeed, in Northern Italian dialects we often find an expletive subject clitic determining agreement in the third-person singular even in the presence of a plural postverbal subject; this nonagreement in Number is of course reminiscent of French (17). Some relevant examples are provided in (18):
PARAMETERS OF SUBJECT INFLECTION IN ITALIAN DIALECTS
(18) a. b. a. b. a. b. a. b. a. b. a. b.
i 'fantf i 'vejian'dopo the children CIS come afterward 'dopo al 'veji i 'fajitj afterward CiS comes the children lir ma?'naei i 'drwomu do ID the children CiS sleep in there da ID u i 'drwom ir ma?'na3i in there CiS there sleeps the children kui ma'toti i Venu tjy 'tordi these children CiS come more late dop u 'verj kui ma'toti afterward CLS comes these children i pu'te e 'dormen the children CiS sleep de d 'la a 'dorma i pu'te in there CiS sleeps the children i pu'tirj i a b'vest al 'la:t the children ClS have drunk the milk al 'la:t a 1 a b'vest i pu'tirj the milk CiS it has drunk the children kwe ra'yattsi j'j enno ve'nuhi those boys CLS have come jj e vve'nuho de ra'yattsi CiS has come some boys
171 Casaccia
Castellazzo Bormida
Dego
Ciano d'Enza
Masi S. Giacomo
Dicomano
Nevertheless, a number of Northern Italian dialects display agreement in Number, Gender, and Person of the postverbal subject with the finite verb and with the subject clitic; this is typically true of the dialects of the Friuli, as illustrated in (19): (19) a.
b.
a. b.
i ka'nais ai duar u'i the children CLS.3PP sleep there 'ia a verj 'dopo she CLS.3PSF comes afterward u'i ai 'duar i ka'nais there CiS.Spp sleep the children 'dopo a verj 'ia afterward CLS.3PSF comes she i ka'nais a sorj vi'jiu:s the children CLS.3PP have come.MP a sorj vi'jiurs dei ka'nais CiS.Spp have come.MP some children
Vito d'Asio
Pinzano
Thus the same agreement parameter illustrated by (16) and (17) is reproduced for the clitic subject languages in (18) and (19). What is more, classical null-subject languages also display the same parametric variation. In these languages the postverbal subject typically agrees with the finite verb, as illustrated in (Ib) for
172
SUBJECTS, EXPLETIVES, AND THE EPP
Standard Italian and in (2b) for Southern Italian dialects. However, some dialects belonging to this type, notably in the area of the North-Central Marche, lack agreement between the finite verb and the postverbal subject, as illustrated in (20): (20) a.
b.
ki bur'dei Verjne 'do:p those children come afterward ki bur'dei 'dormne de 'la those children sleep in there 'do:p vjerj ki bur'dei afterward comes those children de 'la 'dorme ki bur'dei in there sleeps those children
Urbino
The relevant empirical findings in (2) and (18)-(20) are summarized in (21): (21) a. b. c. d.
+clitic,+agreement +clitic, -agreement -clitic, +agreement -clitic, -agreement
dialects of the Friuli typical Northern Italian dialects typical Southern Italian dialects dialects of the North-Central Marche
The fact that all four possibilities in (21) are actually realized, as well as the two possibilities in (16)-(17), suggests of course that there are two completely independent parameters, namely the null-subject parameter already systematized in (12) and a parameter concerning agreement with the postverbal subject. As far as the second parameter is concerned, there exist two types of languages. In the first type, corresponding to English, Standard Italian, or Northern Italian dialects of the Friuli, the postverbal subject agrees in all features with I (and with D); in the second type, the postverbal subject does not display such agreement. Chomsky (1995) argues for an account of (16)-(17) under which the agreement properties of the postverbal subject directly depend on the ^-features of the expletive. Indeed, there in English is not inherently specified for Number, Gender, and Person, while French il is specified as [third person, singular, masculine]. Thus, according to Chomsky 1995, in English the ^-features of the postverbal subject must delete its ^-features. The same operation, however, is not necessary, and hence not possible by economy considerations in French. As it turns out, this analysis cannot be straightforwardly translated into the framework of Chomsky 1998. Indeed, the presence on the postverbal subject of a nominative case feature means that the Agree operation involving it and the I position must take place in any case. This correctly predicts agreement as in English, but leaves us without an account for French. As we saw at the outset of this section, our theory of the null-subject parameter, and more generally of the raising of aspectual features to D(P), forces a different analysis of the agreement between a preverbal subject and I/D from both Chomsky 1995 and Chomsky 1998. Indeed, there is no successive-cyclic raising of the DP itself with all of its features through the agreement positions, but only a successivecyclic raising of the aspectual feature. At the same time the primitive operation Attract, which is meant to substitute for Agree, does not include any specific feature deletion mechanism. On the other hand, our theory also suggests an alternative
PARAMETERS OF SUBJECT INFLECTION IN ITALIAN DIALECTS
173
to the system for the raising of the associate in expletive constructions. As we shall argue in detail in what follows, there are empirical reasons to prefer our approach. 2.2. Agreement of the preverbal and postverbal subject Let us consider first agreement of D with the preverbal subject, in a system in which it is not the preverbal subject itself that raises but rather an aspectual feature. In fact, the solution to this problem presupposes the solution to an even more basic one—that of the agreement between I and D, which are here conceived as separate heads on the functional skeleton of the sentence. We have proposed above that the subject clitic is generated in the position where it surfaces, and that an aspectual feature is attracted by it, providing it with an argumental interpretation. Since the subject clitic itself never passes through the domain of I, it is not clear under this analysis how it comes to agree with I. The problem cannot simply be solved by saying that its position selects the I position, thus giving rise to a feature-checking configuration indirectly, since, as we have already mentioned, the space between I and D contains a number of projections, such as the clitic Neg typical of several Italian varieties. The solution that is effectively forced on us by the general form of our theory is that the aspectual feature itself is attracted by I. Agreement is essentially a reflex of the sharing of aspectual features by different positions. Thus the derivation of the simple example in (8) is to be further specified as in (22), where the OR role is attracted first to I, then to the subject clitic. Note that in accordance with what we shall suggest in section 2.3 below, the position of the subject clitic is identified in (22) with an N position rather than with D: (22)
[Np OR - ro [IP Oft - drwom [VP dfwom(Oft)]]] Castellazzo Bormida CiS sleeps
To be more precise, we can assume that I acts as an attractor of aspectual features in virtue of its nominal inflection features; the subject clitic will evidently attract it in virtue of its N and D features. The same mechanism also accounts for agreement between I and the preverbal subject in non-null-subject languages like English, on the assumption that the preverbal subject fills the SpecD position, as under the parameter in (12). In particular, derivations of the type in (10) are to be further specified as in (23), where the OR role passes through I on its way to D, whence the agreement between I and SpecD: (23)
[DP OR - John [IP Oft -1 [VP slept (Oft)]]]
The derivation of examples like (9) for Standard Italian proceeds in a similar manner; the only difference is that D is not independently realized; therefore the features of I supply the entire feature content of D: (24)
[IP OR - dorme [VP dormo (Oft)]] sleeps
Italian
We shall return at the end of this chapter (section 3) to the general question of the attraction exercised by any nominal feature on aspectual features, which is strictly bound up with the question of interpretability.
174
SUBJECTS,EXPLETIVES,AND THE EPP
The case of a preverbal lexical subject in subject clitic and null-subject languages differs only slightly from those already analyzed. We have already reached the conclusion that the preverbal lexical subject does not satisfy a D feature, but rather a Topic feature, associated with a Top position. Furthermore, the model that we have arrived at effectively bars a derivation under which the lexical subject is generated in thematic position and then moves to SpecTop position. Therefore the preverbal lexical subject is merged directly in SpecTop. Its nominal properties require it to be associated with an aspectual feature; in fact, its agreement with the verb and the subject clitic is just a reflex of their sharing of aspectual properties. Consider then the derivation of a sentence like (5) up to the D level, as in (25). The aspectual feature associated with V is attracted to the subject clitic via I, determining the agreement of these two projections. At this point the DP in SpecTop attracts an aspectual feature itself. Since the only aspectual feature is already associated with I and the subject clitic, this produces agreement between I, the subject clitic, and Top: (25)
[Topp OR- kulo dono [NP Oft - ro [IP Oft - drwom [Vp drwom (Oft)]]]] that woman CiS sleeps
In languages like Italian, which do not have an overt subject clitic, we have neverthless argued for the conclusion that a preverbal lexical subject appears in SpecTop position. Thus the derivation of the relevant sentences will follow exactly the same pattern, as in (25'). (25')
[Topp OR-Gianni [IP Oft - dorme [VP dormo (Oft)]]] Gianni sleeps
Italian
Interestingly, the same schema of derivation can be extended from (24) to all of the constructions that descriptively fall under the 'resumptive clitic' label, whereby the lexical argument and the clitic share one of the argumental slots associated with the verb. Note that this theory of necessity allows the same aspectual feature to be attracted by two or three different nominal elements, namely the verb, the clitic and the full DP. The question may legitimately be raised what rules out matching the same aspectual feature with any two DPs. In the general case, we can assume that the resulting sentence is interpretable, in Chomsky's 1995 terms, but only as gibberish; for having the same aspectual slot bound by two different DPs is like having a variable bound by two different quantifiers (Manzini and Roussou 1999). We can render this assumption more precise by reference to the idea (Manzini and Savoia 1998c) that between the already postulated D clitic and I, there is a full string of clitic positions corresponding roughly to the object clitics of Romance. Being associated with more than one clitic position (subject or object) is excluded for the same reasons as being associated with two scope positions. Agreement between nominal or nominally inflected positions that share the same aspectual features can be seen as the other face of the preceding question. Indeed, features such as N for the traditional third person, P for first and second person, or Num for plural (cf. section 2.3 below) correspond to properties crucially involved in fixing the reference of individuals and sets of individuals. The sharing
PARAMETERS OF SUBJECT INFLECTION IN ITALIAN DIALECTS
175
of these properties by elements associated by the derivation with the same aspectual feature can therefore be seen to fall out from interpretive necessity. If we go on now to expletive constructions, there is a straightforward reason to call into question the analysis proposed in Chomsky 1995 with regard to English and French. Consider, indeed, dialects such as (18), which have been described as having an expletive subject clitic which does not agree with the postverbal subject. Crucially, these dialects do not restrict postverbal subjects to indefinites, contrary to French. Thus they allow full pronouns to appear in postverbal position, provided they are focused; in this case a plural pronoun still does not trigger plural agreement on the verb and the subject clitic, as illustrated in (26). It is worth noticing that according to Zucchi 1996, in the dialect of Piacenza there is a contrast between postverbal definite subjects, which are possible without agreement, and postverbal pronoun subjects, which are not possible. This distribution of data is not necessarily replicated by the dialects below, even if it is true that agreement with the postverbal pronoun is at least possible in all dialects. As we shall see in section 2.4, this agreement option extends to all definites, and therefore does not single out pronouns (for further elaborations on this point, see Manzini and Savoia forthcoming): (26)
a Vi:n sol 'le/'lo/'lor CiS comes only she/he/they dop a vjerj 'luri afterward CiS comes they e js vy'nu '10 CiS has come they
Modena Ariano Polesine Mezzenile
Now, consider first- and second-person pronouns. By the discussion that precedes, there is no independent reason having to do with definiteness or with their pronoun status why they should be excluded from the expletive construction without agreement. Nevertheless, they are excluded from it. To be more precise, they can normally appear in postverbal position with a focused interpretation. In such case, however, both the verb and the subject clitic agree with them in all features, as exemplified in (27a); for the sake of comparison (27b) provides an example from the Friuli varieties which systematically present agreement: (27)
a.
b.
t al di 'te CiS it say you t vi 'ta CiS come you a 'parel 'mi CiS speak I te 'drwomi 'ti CiS sleep you ds ou 'fau 'me CLS it make I tu sos 'tu ko tu du'ars CiS are you that CiS sleep
Modena Alfonsine Viguzzolo Fontanigorda Mezzenile Vito d'Asio
176
SUBJECTS, EXPLETIVES, AND THE EPP
A comparison between the data in (18) and (26)-(27) leads us to the obvious conclusion that nonagreement between I, the subject clitic, and the postverbal subject is in fact restricted to Number and Gender; disagreement in Person is blocked in all dialects.5 As we have often stressed, agreement parameters are completely independent of the null-subject parameter. Thus we expect the same generalization to be true in classical null-subject languages and in classical non-null-subject ones. As anticipated in the discussion above, French does not allow us to check the prediction, since it restricts postverbal subjects to indefinites in any event.6 However, Italian dialects of the Northern Marche type in (20) abey by this generalization, as illustrated in (28): (28)
verj'g i come I si nut 'te have come you sen 'nuti 'no have come we
Urbino
The problem for Chomsky's 1995 theory is at this point evident. Under this theory, the raising of the associate is governed by the features of the expletive, with the result that there are languages in which the associate never raises. On the contrary, the systematic realization of agreement in Person leads us to the conclusion that even languages which do not realize Number and Gender agreement are languages which systematically raise the associate, since only thus can agreement in Person be explained. The main consequence of this observation is that the reasons that provoke the raising of the associate must work in all configurations where an expletive is doubled by a postverbal argument. Thus they cannot be dependent on the features of the expletive, which, as concluded by Chomsky (1995), are indeed subject to variation. Chomsky (1998) revises his previous approach in a number of ways. In the case of expletives like there, merger in Spec! satisfies the EPP feature. But there has only a noninterpretable person feature. This is deleted by the closest corresponding feature (set), namely that of I; conversely the full feature set of I cannot be deleted by the single Person feature of the expletive. Therefore being noninterpretable, they must be deleted by the associate, deleting in turn its noninterpretable nominative Case. By the same set of assumptions, expletives like English it, which possess the whole (J>-features set, are predicted to delete the <j)-feature set of the verb. In the case of English it, this schema of explanation works only because the associate is always sentential. Chomsky himself (1998:49, n. 90), however, notices the problematic status of French //. Though he does not elaborate on this point, the main problem would seem to be that if il deletes the agreement features of I, the latter cannot act as probe for the corresponding features of the associate. This correctly predicts that the associate does not agree with the verb, but it also leaves its status with respect to Case entirely open. Consider the motivation behind the raising of the associate in the grammar presented here. Quite simply, the expletive, in virtue of its nominal features, attracts an aspectual feature. But suppose that the only aspectual feature potentially avail-
PARAMETERS OF SUBJECT INFLECTION IN ITALIAN DIALECTS
177
able to D in a sentence is already associated with an argument in predicate-internal position. If so, the aspectual feature will be shared by the expletive with the predicate-internal argument. Agreement in Person ensues from this hypothesis. Nevertheless it remains to be explained why agreement in Number and Gender does not ensue. In what follows, we shall begin by looking in more detail at the features involved, namely Person, Number, and Gender. 2.3. (Lack of) agreement in Number (and Gender) of the postverbal subject The term (^features employed by Chomsky (1981, 1995) to refer to Number, Gender, and Person suggests that they behave like a coherent class. Though Chomsky (1998) allows for the Person feature to be represented in the absence of other features (as on the expletive there}, he maintains that the set behaves as a whole at least under the deletion operation, since the latter cannot apply selectively to a feature within it. Indeed, the presence of the sole Person feature on there is used to explain cases of full agreement (between the verb and the associate). The phenomena just reviewed concerning agreement with the associate suggest, however, a fundamental split between the Person feature, which always induces agreement, and the Number and Gender features, which do not. Let us then consider how the split between Person and Number is to be construed. To begin with, a careful study of the subject clitic system, of the type pioneered by Poletto (1993, 1997), reveals the existence of two separate positions for first- and second-person clitics and for third-person clitics. In particular, there is a considerable number of dialects in which the Negation, which is itself a clitic, appears after the third-person subject clitic, but before the first- or second-person clitic, as illustrated in (29), with some verb paradigms. In all cases the Neg clitic n precedes the second-person singular subject clitic (and the second-person plural subject clitic in the Firenze dialect), while it follows the third-person subject clitics; other subject clitics are not differentiated for (^-features or not realized: (29)
(e) (e)
urn Verjgo un tu v'vjeni e/la UITJ Vjene (e) un si Vjene (e) urn. vu vve'nihe e/le uirj Veqgano CiSCLS.Snot CLS.1/2 come(s) ne 'parlu ne te 'parli u/a ne 'parla ne par'lemmu ne par'lei i ne 'parlu CiS.3 not CiS.2 speak(s)
Firenze
Fontanigorda
178
SUBJECTS, EXPLETIVES, AND THE EPP a a
n 'dormi n at 'dormi al/la n(a) 'dorem a n dur'mema a n dur'mi i/li n 'dorem CLS CLS. 3 not CLS.2 sleep(s) a n 'dr0mu a n te 'dr0mi u n 'dr0ma a n dru'muma a n dru'mi i n 'dr0ma CLS CLS. 3 not CLS.2 sleep
'mia 'mia 'mia 'mia 'mia 'mia not 'meja 'meja 'meja 'meja 'meja 'meja not
Revere
S. Sebastiano Curone
These data fairly obviously motivate a split between a first- and second-person position and a third-person position for subject clitics. At the same time we are led to wonder whether first and second person vs. third person is the correct characterization of the split observed. An alternative view takes the third person to be absence of a positive value for the feature Person (cf. Benveniste 1966). Pushing this view further, we conclude that P(erson) is not a feature, but rather is a separate category which also has the property of attracting Asp features, and hence of being argumental. P contains whatever specifications individuate first and second person singular and plural. By the same test employed in (29) (i.e., the position of the Neg clitic), at least one additional position can be identified in the inflectional clitic string. Indeed, in a number of dialects, exemplified in (30), the Neg clitic precedes all of the differentiated clitics, both P and non-P, while it follows the undifferentiated one, as illustrated in particular by Sillano. Note that the pattern whereby the Neg clitic is found before all differentiated clitics is often also seen in varieties, such as Zoldo Alto, which do not realize the undifferentiated clitic: (30) no 'forme Zoldo Alto no te 'Sorme no -l/-la 'Sorme no Qor'mjorj no Sor'mi no -i/-le 'Sorme not CLS.2/3 sleep(s) (e) n(o) i 'dorma Sillano (e) non tu 'dorma (e) non le 'dorma (e) non sa 'dorma/dor'mjarj (e) non dur'middo (e) non le 'dormarj CLS not CLS. 1/2/3 sleep(s)
PARAMETERS OF SUBJECT INFLECTION IN ITALIAN DIALECTS
179
The data in (30) suggest that both the P position occupied by first- and second-person clitics, and the position occupied by third-person clitics are distinct from the position occupied by undifferentiated clitics. It is fairly natural to identify the position of the third-person clitic with N (i.e., with the categorial signature of non-P arguments). As for the highest position in the string, we note that when realized, it is compatible with both P and N specifications. We therefore tentatively identify this position with our original D category; indeed, we take it that the suggestion contained in its name—that it is a specification of definiteness/indefiniteness properties—is correct, and that as such it can combine with both P and N. The Neg test data also provide some evidence for a fourth position within the subject clitic string. Thus consider the Friulan paradigms in (31). At Vito, the Neg clitic precedes all differentiated forms, including i in the third person plural, while following the undifferentiated form a. At Montereale again the Neg clitic appears between the N form / and the D form a in the third person singular; however, in the third person plural it again follows i, which clearly is a differentiated form: (31)
no du'ar no te du'ars a no -1 du'ar no dur'mir) no dur'mi: i no du'ar CiS CLS.3 not CiS.2/3 sleep(s) no 'duar no tu 'duar a no 1 'duar no dur'min no dur'mi:s a no -i 'duar CiS not CiS.2/3 sleep(s)
Montereale
Vito d'Asio
This pattern can be taken to indicate that the i form is in a higher position than the / form at Montereale, though both are differentiated for (j>-features, as opposed to the undifferentiated form a. It is furthermore natural to identify the intermediate position between N and D occupied by i with the position Num postulated by Ritter (1991) or Bernstein (1991) for the structure of the DP. In short, our analysis of what are standardly considered the (^-features of D individuates at least three categories that are structurally ordered with respect to D itself, namely P and N corresponding to the feature Person, and Num corresponding to the feature Number. The schema in (32), which summarizes our conclusions, does not mention Gender. In fact, it is argued in recent studies (Ritter 1993, Di Domenico 1998) that even inside DP, Gender is always associated with one or the other of the categories already postulated; we will therefore disregard it from now on.
(32)
[D [Num [N [P
180
SUBJECTS, EXPLETIVES, AND THE EPP
Further support for the structural schema in (32) is provided by Manzini and Savoia, forthcoming; it is also worth noting that we agree with Poletto (1997) in individuating four subject clitic categories, though we disagree on their characterization. We are now in a position to consider sentences where the postverbal subject agrees in all features with the verb in I and with the subject clitic in D when present. Let us first consider subject clitic languages, as illustrated in (19) by dialects of the Friuli. It is true that they can simply be claimed to represent instances of 'clitic doubling' that do not fall under a theory of expletive constructions. But if we succeed in showing that the present theory can provide a uniform treatment for 'clitic doubling' and expletives, as well as for 'resumptive clitics,' then straightforward simplicity considerations favor it. We propose that in sentences of the type of (33) the OR feature of the lexical V becomes associated with the postverbal subject, which is closer to it than to any other D or DP. The derivation of the sentence then proceeds, eventually reaching the verbal and nominal inflections, which attract an aspectual feature in turn. Since the only available aspectual feature, namely OR, is already associated with the lexical argument i kanais 'the children,' it will be shared by it, by the verb in I, and by the subject clitic. In terms of the theory of agreement proposed here, the sharing of aspectual features in turn implies the sharing of inflectional features, which have referential import, and hence, in practice, agreement in all features (Person, Number, and Gender). Note that in the plural a subject clitic is realized in both D and Num, as in (33a) and (33b); on the contrary the masculine singular is realized only by a clitic in N, as in (33c), and the feminine singular by a clitic in D, as in (33d): (33) a. b. c. d.
[DP OR-a [^ump Oft-i... tip Oft- duar [vp Oft - i kanais Vito d'Asio CLS CiS.Spp.M sleep the children [DP OR-a [Nump Oft-s . . . [jp Oft-duar . . . [VP Oft las gardsonas . . . CiS CiS.Spp.F sleep the girls [NP OR- al. . . [IP Oft - duar . . . [vp Oft- il fantat. . . CiS sleeps the boy [DP OR-a . . . [IP Oft - duar . . . [VP Oft la gardsona . . . CLS sleeps the girl
Recall that, as already illustrated in relation to the table in (21), the parameters concerning agreement with the postverbal subject are entirely independent of the nullsubject parameter. Therefore, languages of the type of Italian have essentially the same derivation as dialects of the Friuli, even if no nominal inflection is overtly realized, as illustrated in (34) for the dialect of Modica (Sicily): (34) a. b.
[IP OR-rommi. . . [Vp Oft -u pittjiridjcju . . . sleeps the child [IP OR-ruammunu . . . [ V pOft -i pitt/iridjqM . . . sleep the children
Modica
We are now in a position to consider the parameter responsible for agreement or nonagreement in Number of the associate with the categories of verbal and nomi-
PARAMETERS OF SUBJECT INFLECTION IN ITALIAN DIALECTS
181
nal inflection. The same categories —D, Num, N, and P—that we have defined for the inflectional structure of the sentence constitute the functional skeleton of the DP as well. In the latter case, their logical, as well as morphological, relations have been carefully investigated. Thus standard semantic theories identify N with the predicative content of the DP; since N is nothing but a predicate, the referential properties of the DP reside in D. In effect, D is a quantifier and N represents its restriction. Thus we can assume that D attracts an N feature corresponding to its lexical restrictor. This latter feature could be seen as related to the argumental structure of the noun itself; in particular in the case of noneventive nouns it could be identified with the R role of Grimshaw 1995. Furthermore, assuming that the DP is embedded in a sentence, the D quantifier, while being restricted by the lexical content of N, also binds a variable within the structure of the VP; this we can take to correspond to the aspectual feature attracted by it. In short, within the DP, D necessarily attracts both Asp and N. Analogously we can assume that it is a minimal requirement on inflectional D as well that it should be associated with an aspectual feature, even if it is not associated with a lexical restrictor. Let us then consider the sentence in (35) from a subject clitic dialect in which the associate does not agree in Num with verbal and nominal inflections. In the dialect under examination, that of Premana (Lombardy), the expletive takes the form of an N clitic, exactly like the expletive of French. As in all other cases we have seen so far, the nominal properties of the finite verb and of the subject clitic attract an aspectual feature. Let us suppose, however, that the verb and subject clitic, precisely in virtue of their nominal properties, can attract in addition to the aspectual feature a role R as well. If we apply this hypothesis to the example under consideration, we obtain a derivation of the following type: (35)
[NP OR-al . . . [IP Oft dsyge . . . [vp Oft -R-i fi0i . . . CiS plays the children
Premana
In (35) the sharing of the aspectual feature leads not to the sharing of all features (Person, Number, Gender) with referential import, but only to the sharing of N/R features which correspond to the predicative content of the associate. The availability of this type of agreement in some languages, but not in others, can be expressed again as a lexicalization choice. It is not necessary, in other words, to assume that there is underlying variation in the derivation; thus the derivations in (33)-(34) can be refined along the same lines as (35) by introducing the R feature. The type of agreement seen in (33)-(34) lexicalizes the sharing of an aspectual feature through the sharing of all inflectional features that have referential import; it is in fact, as we shall say, a referential-type agreement. On the contrary, agreement with respect to the N feature only as in (35) can be characterized as a predicationaltype agreement; indeed, sharing of aspectual features is lexicalized through the sharing of the inflectional N feature corresponding to the predicative properties of a nominal expression. A derivation analogous to (35) characterizes languages without agreement in Num where the expletive is represented by an undifferentiated form in D, as illustrated in (36) with the Modena dialect (Emilia). Thus the languages that realize the expletive as an N form are only a subset of the languages without agreement; this
182
SUBJECTS, EXPLETIVES, AND THE EPP
represents an argument against any theory (including Chomsky 1995, 1998) that tries to tie the agreement properties of the expletive construction to the feature composition of the expletive itself. (36)
[DP OR-a . . . tip ©ft-zoga . . . [vp ©ft- R i putein . . . CiS plays the children
Modena
In (37), finally, we have indicated the derivation for those languages where no subject clitic is realized. In this case as well predicative agreement results in sharing of the N feature only: (37)
[IP OR - dorme . .. [Vp ©ft-R ki bur'dei. .. sleeps those children
Urbino
On these grounds, our model correctly predicts agreement in Person. Consider a sentence like (38a), from the Fontanigorda dialect (Liguria), in which we find a differentiated clitic realizing the inflectional P. As before, the verb and the subject clitic attract an aspectual feature, but contrary to what happens in (35)-(37), it is reasonable to assume that P forms do not have a predicative content. Rather, their interpretive content wholly reduces to referential features; in this respect they can be assimilated to D. The lack of a predicative, or N content, means that P pronoun associates can determine only referential agreement. Thus in (38a) the verb and the subject clitic attract an aspectual feature, which, being shared with the associate P pronoun, determines agreement in all features (in practice P) between the verb, the subject clitic, and the P pronoun itself. The same holds for cases like (38b-c) in which both a D and a P clitic are realized, or only a D clitic: (38) a. b. c.
[pp OR - te ... [IP ©ft - drwomi. . . [Vp ©ft - ti. . . Fontanigorda CiS sleep you [DP OR - a ... [Pp Oft - 1 . . . [IP Oft - dormi. . . [Vp Oft-ti. . .Revere CLS CiS sleep you [DP OR - a ... [IP Oft - durmem . . . [Vp Oft - nyalter . . . Revere CiS sleep we
A derivation analogous to that in (38) characterizes dialects in which clitic positions are not realized at all, as in (39): (39)
[IP OR-parlen . . . [yp ©ft-no . • • speak we
Urbino
Note that this theory is able to capture the fact that third-person pronouns can be treated as first- and second-person pronouns, triggering agreement in all features, or as lexical DPs, eventually triggering N-type agreement. In the latter case they are treated as having N-content; in the former they are treated as purely referential elements, without any predicative content. We have now analyzed the agreement data we presented at the outset. In particular, concentrating on subject clitic dialects, we have accounted for the systematic agreement of P associates with verbal and nominal inflections. We have furthermore formulated the parameter responsible for the possibility for a DP associate to agree or not with nominal and verbal inflections in Number. This parameter can be schematically formulated as in (40):
PARAMETERS OF SUBJECT INFLECTION IN ITALIAN DIALECTS
183
(40) If A and B share aspectual features (as a derivational property), they share (as a lexicalization property) a. all features with referential import b. the predicative N feature As we have argued more than once, the agreement parameter is entirely independent of the null-subject parameter. Thus classical null-subject languages present the same asymmetry with respect to agreement of the associate as subject clitic languages. We briefly mentioned in presenting the Friulan data that they can at least descriptively be characterized as cases of 'clitic doubling.' At the same time we stressed that obvious simplicity considerations favor a unitary treatment of such constructs and of expletive ones. Such a unitary treatment has in fact been provided in what precedes. What is more, we note that all cases descriptively labeled as 'clitic doubling' can now be theoretically characterized as expletive constructs with agreement, in which a clitic and a full DP argument share the same aspectual feature via attraction. What is more, the data in (18) and (21) clearly show that nonagreement in Num between the associate and the verbal and nominal inflections is constrained to the case in which the former is plural and the latter singular. Thus in no Italian dialect (or any other language that we know of) is it possible to have a singular associate with plural inflections; in other words, the nonagreeing expletive form of nominal and verbal inflections is always singular. The fact that this important generalization is usually left implicit in syntactic treatments presumably reflects the idea that some morphophonological default mechanism is responsible for this state of affairs. It should be noted, however, that the notion of 'default' has no clear independent role in current theories, not even phonological ones. Similarly, a purely syntactic solution is in principle preferable to having recourse to a completely separate component, morphology. Indeed, in the purely syntactic conception of features adopted here, singular is not the default value of the Num feature; on the contrary, singular and plural represent two completely independent specifications of the Num category. In particular, the plural corresponds to a realization of the Num category itself, while the singular simply corresponds to the case in which Num is not realized. Thus we explain why a nonagreeing expletive is always singular without having recourse to morphological default. Agreement with respect to the N feature simply excludes any realization of Num. We have already indicated above that due to limitations of space, a discussion of Gender is impossible here (see Manzini and Savoia forthcoming). Nevertheless, it is worth mentioning briefly that expletive clitics in languages without agreement in Num can still be differentiated in Gender; thus the expletive is masculine in (35) above, but feminine in (55) below. This clearly casts doubts on the idea that the nonagreeing expletive realizes default values, or at least on the explanatory value of such an idea, since if default is at stake, its value for Gender is not universal. On the contrary, our syntactic representations clearly distinguish between Num, which has its own autonomous realization, and Gender, which does not, as briefly mentioned in connection with (32); Gender therefore could simply be represented in N (cf. Di Domenico 1998).7
184
SUBJECTS, EXPLETIVES, AND THE EPP
These considerations lead us to clarify a further point. In languages of the type in (35)-(37), in which there is nonagreement in Num between the expletive and the associate, it is obvious that the nominal carries full Number (and Gender) specifications. This independence between the internal form of DPs and the agreement parameter is also as expected Going back now to (16)-(17), French appears to have all of the crucial properties of typical Northern Italian dialects from the point of view of the present theory, since raising of the associate fails to trigger agreement in Number with I and with the expletive. We shall return in the next section to the definiteness restrictions displayed not only by French but also by other languages. On the other hand, the agreement system of English does not have an exact parallel in any system examined so far. In particular, in English the postverbal lexical subject, which is generally taken to be indefinite, and hence by definition third person, agrees in Num features with the finite verb, hence with I; however, the expletive takes the invariable form there, which points to lack of agreement between the postverbal subject and the expletive. In fact, there are subject clitic systems that are are superficially similar to English. Thus in the Provencal dialect of Celle (Apulia) the associate agrees in all features with the verb in I, as illustrated in (41); however, the clitic subject takes an invariable form, not inflected for Number (or Gender), as is clear from the comparison with (42), where the lexical subject is preverbal. The alternation between o and aj in (41) appears to be governed by the alternation between verbs beginning with consonant and vowel, respectively; in any event, both forms are clearly distinct from what we take to be the agreeing D clitics in (42), which alternate between / and X": (41)
o d'doral srj'farj CLS sleeps the child o b'biA/fe 1 ejs CIS boils the water o Vinda ta ssa'rawa CiS comes your sister o dur'munda bs arj'farj Cl5 sleep the children D va'nunda bs arj'farj ClS come the children a'j ett9 vo'ni murj fi'awa CiS has come my child o 'sund9 va'ni mo fi'awa CLS have come my children
(42)
1 srj'farjrj i d'dora the child CiS sleeps '1 eja i b'bi/CXs the water CiS boils ta ssa'raws i Vindg da'poja your sister ClS comes afterward
Celle S.Vito
PARAMETERS OF SUBJECT INFLECTION IN ITALIAN DIALECTS
185
los srj'farj i dur'munda/i vs'nunda the children CiS sleep ICiS come murj fi'awa 'A etta va'ni a'iara my child CiS has come yesterday mo fi'awa i 'sundQ va'ni a'isre my children CiS have come yesterday It is worth noticing that while in English pronominal associates are independently excluded by the definiteness restriction, this is not the case in the dialect of Celle; thus, as expected, the olaj expletive appears with third-person pronominal associates, as illustrated in (43). Nevertheless the olaj expletive cannot appear with first- and second-person associates; in this case, the subject clitic is represented by the appropriate Person form, as again illustrated in (43): (43)
ma ta a'j n9 va o CIS
'seJ9 'dji 'sejs 'ti ette 'ija 'surj 'nus 'si 'vu 'sunds 'i:ss am/are/is PN
Celle S. Vito
Martin (1974) proposes an etymological derivation for Provencal expletives which assumes the Latin basis *ibi for / and the Latin basis *hoc for o, hence two locative forms. This interpretation effectively extends that already proposed by Ronjat (1937) for the la form of the expletive in Proven§al based on *illac, another locative. These reconstructions are debatable as such, but interesting precisely in that they seem to follow not an independently motivated etymological sequence, but rather a semantic interpretation a posteriori of the relevant forms. In present terms, there are good reasons to assume that olaj is in D, as is generally the case with undifferentiated subject clitics. Indeed, the differentiated subject clitic forms of the N/Num type are to be identified in this dialect with i and /. Since the verb agrees in all features with the associate, including Num, we know that the aspectual feature of the associate is attracted by I. The same feature can then be attracted by inflectional D as well. But if olaj themselves lack morphological specifications for Num (and Gender), we predict that there will be no overt reflex of agreement on the expletive. Let us then return briefly to English, with which we began the present discussion. The expletive of English, there, is actually a locative form rather than a D form. Though it must have at least a D feature, it lacks <j)-features. The outcome is not substantially different from having an undifferentiated subject clitic. Thus the aspectual feature of the associate can be attracted all the way to D, triggering agreement with I on the way, but with no overt morphological consequence on there itself.
186
SUBJECTS, EXPLETIVES, AND THE EPP
2.4. Definiteness
effects
In (19) above we presented data from typical Friulan dialects, in which the associate agrees in all features not just with the finite verb in I but also with the subject clitic. In some dialects of the same region, as well as in other Northern Italian dialects, the associate agrees with I and the subject clitic when it is definite. When it is indefinite, however, the associate may lack agreement in Number (and Gender) with I and the subject clitic, along the lines already illustrated in (18) for the core Northern Italian dialects. In (44), the examples in (a) show forms with postverbal subjects and agreement in all features, those in (b) provide a comparison with preverbal subjects, and those in (c) show the lack of agreement with a postverbal subject when the latter is indefinite. (44) a. b. c. a.
b.
c. a.
b. c.
i Verj i jio fi'oi CiS come the our children i jio fi'oi i Verj the our children CiS come a Verj ka'nais CiS comes children de 'le i 'droman i fi'0 in there CiS sleep the children i m erj kro'de:t i 'pja:t CiS to. me are fallen the dishes i fi'0 i 'droman de 'le the children CiS sleep there i 'pja:t i m erj kro'de:t the dishes CLS to. me are fallen a ka 'mia 1 e ve'jiy:t di ma'terj to house my CiS is come some children 'douppo i v'vermana di /i f fa afterward CiS come some/the children da 'la i d'dormsn i f fanti in there CiS sleep the children i ffanti i d'dormsno the children CiS sleep j e vvs'nuts di ffanti CiS is come some children
Montereale
Campodolcino
Gorafiglioano
In at least some of these dialects, nonagreement with the indefinite associate appears to be possible rather than necessary, as illustrated most clearly by the dialect of Gorfigliano (northern Tuscany). These examples pose a particular problem within a Minimalist grammar because of the optionality they seem to display. Given the general theoretical arguments that hold against optionality, it is most natural to construe the alternatives in (44) as indicating ambiguity. In particular it is possible that agreement relates to the specific use of the indefinite associate, and lack of agreement to its nonspecific use.
PARAMETERS OF SUBJECT INFLECTION IN ITALIAN DIALECTS
187
As before, we do not expect the parametric choice represented by (44) to depend on a particular value of the null-subject parameter. In other words, we predict that the same sensitivity to definiteness can be found in classical null-subject and nonnull-subject languages. The first prediction can be shown to be true. Indeed, we have seen in (2b) that in Southern Italian dialects, the postverbal subject typically agrees with the finite verb in all features. The same is true of Sardinian dialects. As exemplified in (45), however, Sardinian dialects typically do not display agreement with the postverbal subject when the latter is indefinite; the same is true of Central Calabrian dialects such as lacurso: (45)
ain'noi 'eni5i 'omminizi here comes [some] men 'drommini zos pit'tsinnozo sleep the children bi 'drommiti pit'tsinnozo here sleeps children Vinne 'tanti 'himmini came so.many women
Orroli Siniscola
lacurso
On the other hand, many Northern Italian dialects of the core type in (18), which normally lack agreement of the associate with the expletive, allow such agreement if the postverbal subject is definite. This is illustrated in (46), where the examples in (a) illustrate the nonagreeing option for postverbal subjects and the examples in (b) their agreeing option. A more accurate description of the phenomenon can be arrived at on the basis of examples like those for the dialect of Castellazzo Bormida. These involve disagreement and agreement, respectively, with an indefinite postverbal subject. As already mentioned, indefinites can have both a nonspecific and a specific reading; under the latter they form a natural class with definites, which are always specific. It is therefore natural to speculate that the true generalization pertaining to agreement with the postverbal subject in dialects such as (46) is that it is restricted to specific arguments. (46) a. b. a. b. a. b.
al Veji i ba'gai CL$ comes the children i 'vejian i ba'gai CiS come the children la a '?0ga i py'tlet there CiS plays the children la i '?0ga i py'tlet there CiS play the children a 'perla i 'o:m CiS speaks the men i 'perlsn i 'o:m CiS speak the men
Casaccia-Val Bregaglia
Revere
Modena
188
SUBJECTS, EXPLETIVES, AND THE EPP a. b. a. b. a.
b.
da'dr D u i 'drworn di 'omi there CiS CILoc sleeps some men da'dro i 'drwomu dir ma?'nai there CiS sleep some boys me 'tfamma i 'mie a'migi me calls the my friends i me 'tjarnmu '1 atri CLS me call the others a s e a'perts '4 ujo CiS.F itself is opened the door.M 4 e vs'nu i 'birj CiS is come the children i s e a'perta '4 ufo CiS.M itself is opened the door.M j em va'nu i 'birj CLS are come the children
Castellazzo Bormida
Fontanigorda
Ortonovo
Our interpretation of the data in (46) is supported by the observation that in contexts favoring a nonspecific reading of indefinites, such as existentials, agreement is indeed ruled out, as illustrated by the contrast between (47a) and (47b). Existential contexts can also be used to rule out the possibility that agreement with the postverbal definite subject, as in (47c), depends on it being a Topic: (47)
(ki 'g e ?) (who there is?) a. a 'g e di ra'ga:s CiS there is some boys b. * i 'g e:n di ra'ga:s CLS there are some boys c. i 'g e:rj ki ra'ga:s CLS there are those boys
Modena
As before, we expect the typology illustrated here with subject clitic dialects of the core Northern Italian type to be found in null-subject varieties as well. Indeed, at least some of the dialects of the northern Marche allow for both agreement and disagreement with the postverbal subject. In (48) the (a) examples illustrate agreement with the preverbal subject, and the (b) and (c) examples, disagreement and agreement, respectively, with the postverbal subject: (48) a.
i bor'dei 'verjgne/'dormne 'dop the children come /sleep afterward
el bor'dei 'dorme/'vjen 'dop b.
the child comes/sleeps afterward dla 'dorme i bor'dei there sleeps the children dop Vjen i bor'dei afterward comes the children
Monteguiduccio
PARAMETERS OF SUBJECT INFLECTION IN ITALIAN DIALECTS c.
189
dla 'dormne i bor'dei there sleep the children dop 'verjgne i bor'dei afterward come the children
As far as the definiteness effects considered so far are concerned, they provide an argument in favor of the idea that agreement may depend on the the properties of the attractee, in this case the associate, instead of on the properties of the attractor, in this case the nominal inflections of the sentence. Remember that according to the agreement parameter developed here, lack of agreement in Num between the associate and the verbal and nominal inflections reflects agreement with respect to the predicational N feature only; full agreement depends on the sharing of all features with referential import. Consider then the dialects in (44)-(45). Descriptively speaking, whether the associate is definite or indefinite, agreement can be of the referential type; however, only if the associate is indefinite can it also be of the predicational kind. It is natural to relate this difference to the different internal structure of definite and indefinite DPs. We can assume that D represents in any event the head of the DP; however, it is associated with different interpretive properties in the two cases. In terms of the present theory of features (as categories), this cannot simply depend on the presence in D of a two-valued feature ±definite, with -definite the default value. On the contrary, we can discriminate between definites and indefinites, assuming that D in all cases corresponds essentially to the variable that satisfies the R role of the Noun (as well as an aspectual role of the verb). Definites are then characterized by the presence of an individual variable whose value is typically assigned by a definite D; indefinites are characterized by the presence of a true variable, whose value is assigned by a quantification of some sort. With this much background we can now propose that in dialects of the type in (44)-(45), agreement in all features that have referential import (including Person, Number, and Gender) is restricted to cases in which the D head of the associate does in fact have individual denotation (i.e., to definites), as in (49a). On the contrary, when the D head of the associate does not correspond to a definite individual (or set of individuals), dialects of the type in (44)-(45) lexicalize agreement with respect to the N feature only, corresponding to the predicate content of the associate itself, as in (49b): (49) a. b.
[jp OR - drommini. .. [VP Qft-R-zos pittsinnozo.. . sleep the children [jp OR - drommiti.. . [Vp Oft-R-D pittsinnozo . . . sleeps children
Siniscola
Consider next the dialects in (46)-(48), which are characterized by a considerable amount of surface optionality, in that indefinite associates do not agree in Num but definite associates appear to be able to agree or not. The fact that N-type agreement (i.e., agreement with respect to the predicate N property) is possible with any kind of associate does not require any enrichment of the theory already given in section 2.3. It is simply a parametric choice of these dialects that the sharing of the aspectual feature can be lexicalized by agreement with respect to the predicative
190
SUBJECTS, EXPLETIVES, AND THE EPP
properties of the associate, hence by N-type agreement, as in (50a) and (50b). All that needs to be added is that when the D head of the associate corresponds to an individual variable, it is also possibile to lexicalize referential-type agreement (i.e., agreement with respect to all features), as in (50c): (50) a. b. c
-
[DP OR - a ... [IP Oft - perla . . . [VP Oft- R-di om . . . CiS speaks some men [DP OR - a ... [jp Oft - perla . . . [vp Qft-R- i om . . . CLS speaks the men
Modena
[NumP OR - i . . . [jP Oft - perten [VP Oft-R- i om . . . ClS speak the men
It is now time to draw into the picture non-null-subject languages of the classical type. English is generally described as requiring the associate to be indefinite in expletive constructions. The latter therefore are impossible if the associate is definite; or at least, as Chomsky (1995) points out, a definite associate is connected to a different interpretation than an indefinite one. Thus an indefinite associate gives rise to the typical existential reading in (5 la), while a definite associate gives rise to the list interpretation, as in (51b): (51) a. b.
There is somebody outside. There is John for a start.
The picture of English is further complicated by the fact that expletive constructions are restricted to a limited range of lexical verbs, which appears to be included within the unaccusative class. What is immediately relevant for present purposes is that, as predicted by our parametric theory, there are non-null-subject languages of the classical type which do not show either of these restrictions, for instance, German. On the other hand, French, which does not fall together with English under our version of the null-subject parameter, is standardly described as patterning with it in the restrictions imposed on the associate. Because the thematic/aspectual restrictions fall outside the scope of the present chapter, we shall leave the discussion of French and English open here. At this point, it remains for us to consider one last empirical argument put forward by Chomsky 1995. French does not allow for control of an adverbial sentence by a postverbal subject; such control is instead perfectly admissible in languages like Italian which present agreement with the postverbal subject, as exemplified by the contrast in (52); data of this type are provided by Cardinaletti 1997: (52) a.
Sono arrivati dei ragazzi senza avvisarmi. have arrived some boys without telling.me b. * II est arrive des gar§ons sans me le dire. it is arrived some boys without me it telling
Italian French
The data in (52) suggest that in languages where the associate determines agreement, as in Italian, it moves (or its features move) high enough in the tree to be in a position to c-command, and hence control into the adjunct (however control is to be construed); vice versa, this is not the case in languages where the associate does not agree. According to Cardinaletti 1997 this generalization holds for Northern
PARAMETERS OF SUBJECT INFLECTION IN ITALIAN DIALECTS
191
Italian dialects as well. Her data relating to Friulan-type dialects, in which the subject clitic agrees with the associate, indicate that control of an adjunct is possible. On the contrary, data for the Padova and Belluno dialects would show that in nonagreeing expletive constructions, control of the adjunct by the associate is not possible (Cardinaletti 1997:528). However, this type of data is not replicated by our survey; thus Northern Italian dialects which lack agreement of the associate systematically allow for control into the adverbial, as illustrated in (53). (53)
1 e 'jiu di ra'gass 'sensa 'dirmel Modena CiS has come some boys without telling.me 1 e v'nu i mi fi'ul 'genes 'di 'jiT :t Alfonsine ClS has come the my children without telling.me anything e 'jiy i me fi'0i 'Genoa 'direm 'jie:nt Revere CiS.has come the my children without telling.me anything e 'jiy di py'tlet a ?u'gar CiS.has come some children to play
What is more, the possibility of control of an adjunct sentence by the associate in a nonagreeing expletive construction also characterizes null-subject varieties, such as, for instance, the Sardinian ones illustrated in (54). Incidentally, these dialects normally present the same distribution of have and be auxiliaries as Standard Italian, with be for unaccusatives and have for the other verbal forms; nevertheless they switch to auxiliary have in nonagreeing expletive constructions with indefinites (Savoia and Manzini forthcoming and references quoted there). Since this latter choice of auxiliary is observed in (54) as well, we can confidently exclude the possibility that the postverbal subject is construed as topicalized or in any other way independently capable of explaining its ability to control the adverbial. (54)
dm a jk'niu pittrDk'kedjdjuzu 'sentsa de mmi 'nai 'nucjcja Orroli here has come children without of me tell anything a'9 enniu {Mttsok'kedjcjpzp '$entsa mi 44U 'narreSe Ardauli has come children without me it tell
Thus it appears that the constraint in (51) for French is to be studied in a different context from that suggested by Cardinaletti 1997 and Chomsky 1995. Not surprisingly, the same conclusion is implicit in the revision of the theory of agreement by Chomsky 1998. It is worth noticing briefly that though Cardinaletti 1997 concurs with Chomsky 1995 on the conclusion that the associate raises only in languages where it agrees with the verb in all features, she arrives at a different formulation of the relevant parameter. Remember that according to Chomsky 1995, raising of the associate in English, with consequent verb-associate agreement, depends on the fact that the expletive there lacks features; on the contrary, il has such features in French, so that the associate does not raise and does not determine agreement. Cardinaletti 1997 notices that this generalization is invalidated by German, among other languages, since German expletive es does have (J)-features but German expletive constructions have agreement of the verb with the associate. As an alternative Cardinaletti (1997:526) proposes that "those expletives that are unambiguously marked as nom-
192
SUBJECTS, EXPLETIVES, AND THE EPP
inative trigger agreement with the verb"; on the contrary, expletives not unambiguously marked for nominative require the raising of the associate, which thus determines agreement with the verb. In fact, French il is unambiguously nominative, whereas German es is both nominative and accusative. There are general reasons to doubt the theoretical status of the notion of abstract Case crucially involved in Cardinaletti's generalization (Manzini and Savoia 1998c and references quoted there); indeed, precisely because a form like es is morphologically not specified for case, its disambiguation can refer only to a notion of abstract Case. Nevertheless, it is interesting to consider whether there is any truth to the descriptive generalization itself. Our data concerning Northern Italian dialects undermine the empirical basis of this generalization. In particular, in some dialects the expletive takes a third-person singular form ambiguous between subject and object uses; however, it is still the expletive and not the associate that triggers agreement with the verb. This is true both in dialects where the expletive takes the masculine form, such as Casaccia (Alpine Lombardy), and in dialects where the expletive is feminine, such as Casola (northern Tuscany), as shown in (55). The (a) examples illustrate the form of the subject clitic in preverbal subject constructions, the (b) examples illustrate its form in expletive-associate constructions, while the (c) examples illustrate the form of the object clitic. (55) a.
b.
c. a.
b. c.
(la ni'nina) la 'dorm /la Verj (the little, girl) CiS sleeps/CiS comes (i ni'nirj) i 'dormana /i 'vejijiana (the children) CiS sleeps ICiS comes dop la Verj i ni'nirj afterward CiS comes the children nda kl as'tantsa la ga 'dorm i ni'nirj in that room CiS there sleeps the children i la 'laona CiS her wash (al ba'gai) al 'veji 'dopo (the child) CiS comes afterward (i ba'gai) i Vejian 'dopo the children CiS come afterward al 'veji i ba'gai CiS come the children al 'fa:-t? it do-you
Casola
Casaccia
3. Theoretical conclusions The preceding discussion makes it clear that the theoretical interest of morphosyntactic variation in Italian dialects goes far beyond the precise definition of the parameters involved. Indeed, while remaining within the broad conceptual framework of Minimalism, we have made several proposals concerning Universal Grammar. These are worth summarizing briefly. To begin with, our discussion of
PARAMETERS OF SUBJECT INFLECTION IN ITALIAN DIALECTS
193
the null-subject parameter in section 1 has led us to part from the theory of Chomsky 1995, 1998 in two major respects: first, there is a set of nominal inflections (D, etc.) in addition to the verbal inflection I. Second, the nominal inflections attract an aspectual feature which contributes an argumental interpretation to it according to the model of Manzini and Roussou 1999; as a consequence of this, there is no need for an empty category pro, responsible for this same argumental interpretation. The null-subject parameter itself reduces to whether the D/EPP feature is lexicalized by a D head (subject clitic languages), by a DP (non-null-subject languages), or by the verb (null-subject languages). In section 2 we introduced several innovations with respect to the theory of agreement. In our system, agreement of the verb with the preverbal subject is a byproduct of the attraction of an aspectual feature by several positions, typically I and D. We consider it highly significant that our theory forces us to an analysis of expletive-associate constructions alternative to Chomsky's (1995, 1998). As we have amply motivated in section 2, Chomsky's 1995 theory, based on the agreement properties of the expletive, faces at least one major empirical problem. For, even limiting ourselves to traditional expletive constructions, the disagreement of the verbal and nominal inflections with the associate never involves N vs. P features and is instead limited to Num (and Gender) features. This implies, in movement terms, that the associate always raises, since the match of N/P features can be guaranteed only by movement. In our system, the agreement of the associate is triggered not by the ^-features of I, but by the fact that nominal features in general attract aspectual features. By definition, in expletive contexts only an aspectual feature already associated with an argument is available. This aspectual feature being shared by the argument itself and by the inflectional positions determines their agreement in all features (Friulan dialects). The other value of the agreement parameter (typical Northern Italian dialects) depends on the possibility for agreement to involve just the predicational N properties of the associate. In general, for Chomsky 1995, 1998, (^-features are one of the motors of syntactic derivations. In the present system, on the contrary, agreement is not treated as a motor of the derivation but rather as a reflex of derivations that happen independently of it. In other words, agreement has a status analogous to that of Case, which does not drive derivations, but provides a morphological realization for their product. This result seems to us intuitively correct and conceptually desirable, providing indirect support for our overall analysis. The status of agreement in turn has wider consequences for the theory of grammar. According to Chomsky 1995,1998, noninterpretable features act as probes for interpretable counterparts; indeed, being noninterpretable, they must be deleted before the derivation reaches the LF interface. This system is crucially modeled on the deletion of the noninterpretable ^-features of I by the interpretable features of DP, which yields subject-verb agreement. If our alternative construal of agreement is correct, however, this piece of empirical motivation for the system no longer stands. As already emphasized, in our alternative model, nominal features attract aspectual features and agreement is just a reflex of the sharing of the same aspectual feature by several nominal positions. As pointed out in Manzini and Roussou 1999,
194
SUBJECTS, EXPLETIVES, AND THE EPP
we can maintain terminological consistency with Chomsky 1995 by assuming that every D is associated with a noninterpretable aspectual feature, which needs to be deleted by an interpretable aspectual feature originating with the lexical verb. Nevertheless, from an empirical point of view this appears to be entirely equivalent to saying that two interpretable features (i.e., D and OR/MEAS) must be matched in a local configuration. If the two theories are empirically equivalent, it seems to us that simplicity considerations favor the one with the fewer assumptions, hence in this case the direct matching of two interpretable features, which does not require the extra notational richness associated with a noninterpretable aspectual feature on D. The general conclusion that the grammar contains only interpretable features has in fact been argued for in the literature (Brody 1997). The last set of theoretical consequences that we wish to highlight concerns the theory of parameterization. While remaining within the bounds of the hypothesis that parameters concern lexical items or features, two broad models of parametrization have been developed within Principles and Parameters theory. The leading model in the Government and Binding framework has remained the one articulated by Rizzi (1982), precisely in connection with the null-subject parameter. In this model, which we may call macroparametric (Baker 1996), one parametric choice typically has consequences that involve a set of syntactic constructs. Within such a model it is possible to think that there is a limited number of parametric choices in Universal Grammar and that each parametric choice contributes a large chunk of the syntax specific to a given language. Such a model predicts a type of variation not unlike that envisaged outside the generative framework by typological studies. The alternative approach, which can be termed microparametric, predicts on the contrary that each parameter has strictly local consequences; these strictly local parameters vary independently of one another and combine freely, up to Universal Grammar, giving rise to fine surface parametrization. Empirical investigation into the morphosyntax of Italian dialects both concerning the syntax of the subject and other domains (Manzini and Savoia, forthcoming) crucially supports this microparametric view. Though the view appears to be dominant in the Minimalist conception of Chomsky 1995, it had earlier proponents as well. In particular, Manzini and Wexler (1987) stress that microparametrization is actually what one might expect on learnability grounds. As already pointed out, in a microparametric model, each parameter is strictly local and independent of other equally local parameters; this means that the parameter can be decided on inspection of strictly local evidence, and without reference to the setting of any other parameter. The characterization of the microparametric view can be sharpened beyond what may appear to be no more than common sense8 on the basis of the specific proposals advanced here on the null-subject parameter. Under the present view, not only is the structure underlying natural languages universal (cf. Cinque 1999), but so are the derivations resulting from the (universal) attraction properties of the various categories. The strictly local and independent character of the parameters derives from the fact that they effectively reduce to lexicalization choices.
PARAMETERS OF SUBJECT INFLECTION IN ITALIAN DIALECTS
195
Notes This work is a result of the collaboration of the two authors in all respects. Nevertheless, Leonardo Savoia takes responsibility for the first part of the chapter, through section 2.1, and Rita Manzini for the second part, from section 2.2 onward. Both authors take responsibility for the theoretical conclusions in section 3. 1. The proposal of a D category will be modified below in section 2.2. 2. Needless to say, once A-movement is abandoned in favor of feature attraction, the same conclusion is forced, if at all possible, for other types of movement. This result cannot be argued for within this chapter, but all we say is compatible with it. In particular, the head-movement notation employed throughout is compatible with a reading under which the head is merged in the position where it surfaces, say the I position for the finite verb, and the relevant feature of its lower 'copy,' in V for the verb, is attracted by it. In line with what we have seen so far, this latter feature could also be seen as related to the argumental structure of the predicate; in particular it could be identified with the event argument, e, of Higginbotham 1985. 3. A principled problem for all discussions of the null-subject parameter is implicit in our discussion of the Colfosco dialect, where we indicated that the first- and second- person plural forms are effectively null subject. Note that the same problem arises in a large number of Northern Italian dialects, which do not present the extra complications of the Colfosco one. Space limitations prevent us from delving deeper into this matter in the present chapter, except for noting that a lexical approach to parametrization yields a particularly flexible instrument for dealing with cases like this one. Indeed there is no special problem involved within the present framework in claiming that a dialect like Colfosco simply lacks lexicalization for the D position (or for the P position in terms of the more articulated structure developed in section 2.2) in correspondence of the first- and second-person plural. 4. Rizzi (1986) tests the nature of the position occupied by the preverbal lexical subject in null-subject languages by taking into consideration the behavior of quantified DPs, in particular negative ones. The basic idea is that negative quantifiers and the like are incompatible with Topichood because of their intrinsic properties; therefore, if they occur in the preverbal subject position of null-subject languages, the latter must be the subject position proper, rather than a Topic. In particular Rizzi (1986) points to a contrast between Standard Italian examples such as (i) and (ii), where in his judgment the focalization in (i) is well-formed, while the topicalization (Clitic Left Dislocation, or CLLD) in (ii) is not: (i)
Nessuno ho visto. Italian nobody I.have seen (ii) Nessuno 1'ho visto al cinema. nobody him.I.have seen at.the cinema In fact, it seems to us that by changing the pragmatic context (for instance, adding a separate focus), sentences of the type exemplified in (ii) become acceptable with CLLD, as in (iii); the same is true if the negative quantifier is given a specific reading, as in (iv): (iii)
Qui nessuno 1'ho mai visto vestirsi cosi. Here nobody him.I.have ever seen dress.himself like.this
Italian
(iv)
Nessuno (di questi) 1'ho mai visto al cinema. nobody of these him.I.have ever seen at.the cinema As for subject clitic languages, Poletto (1993) concludes that preverbal lexical subjects doubled by a subject clitic are indeed in Topic position, since the doubling by a subject clitic is excluded in the case of negative quantifiers, at least in the central Veneto dialects she considers. A look at our data reveals that overall, Northern Italian dialects allow for two possibilities. On the one hand, negative quantifiers and the like can be doubled by a differentiated subject clitic as in (v); on the other hand, they can be be immediately followed by the expletive (eventually a zero one) found in expletive-associate constructions as in (vi). The two possibilities actually alternate in the same dialect, as in (vii), showing that the choice is not parametric in nature: (v)
ni'sun al 'verj nobody CiS comes ne'gy.rj ai rn'majidje nobody ClS eat
Grizzo Premana
196
SUBJECTS, EXPLETIVES, AND THE EPP ni'Jeun i m 'tjema nobody CiS me call niq al 'majidsa nobody CiS eats ni'gynne al 'mendje nobody CiS eats
(vi)
de'guji a dor'mi nobody has slept ni'surj me'tjame nobody me calls nirj'Qyrj a 'majia nobody CiS eats ki ni'Jirj 'majidsa here nobody eats 'jiyrj e Vint nobody CiS comes
S.Mauro Pascoli Felizzano Villa di Chiavenna Andraz Chioggia Revere Fontanigorda Mezzenile
(vii)
ni'suq i 'mapa Stienta nobody CiS eat ni'surj a 'mapa nobody CiS eats Examples of the type in (vi) in fact suggest an analysis under which the quantified DP is focused rather than topicalized, binding a variable in the same position in which we would find a postverbal lexical subject in expletive-associate constructions. But the cases in (v) suggest a topicalization analysis which in the terms of the discussion of (iii)-(iv) turns out to be equally unproblematic. 5. An analogous generalization is reached by Rouveret (1994) on the basis of Welsh data. 6. The anonymous reviewer for this volume suggests that French does in fact have expletive-associate constructions with a definite associate, and that such constructions do allow for first- and secondperson pronouns, as in (i): (i)
II n'est venu que moi/toi French it not.is come that me/you 'Only I/you came' For the argument in the text to go through, this example will have to be dismissed in some way as irrelevant. The basic problem seems to us to be that the structure of such an example is far from clear; the presence of the complementizer que to introduce the pronoun even suggests that two sentences may be involved (cf. Donati 1998 on phrasal comparatives). If this latter idea is on the right track, then it is not clear on inspection what the focus position occupied by the pronoun is, given the presence of massive deletion. In short, further work is needed to establish (i) as a bona fide counterargument to our line of discussion. 7. It is one of the recurrent claims of the treatment proposed here that 'resumptive clitic,' 'clitic doubling,' and expletive constructions are just three different ways in which the same phenomenon of sharing of one aspectual feature by an inflectional D position and a DP argument can be realized. One potential problem is created by the apparent asymmetry between the case in which the DP argument precedes the D inflection and the opposite case; thus in all of the examples seen so far, a DP agrees in all features with an inflectional D element that follows it ('resumptive clitic'), while there can be or not be agreement in the reverse case ('clitic doubling' and 'expletive,' respectively). This asymmetry does not extend to all examples even of the range of languages considered here, as shown by the following French example pointed out to us by the anonymous reviewer: Des linguistes il en est venu beaucoup. French of. the linguists it of. them is come many 'Many linguists came' Therefore the pattern that seems to emerge from the body of the text will have to be treated as an artifact of the particular type of data considered. This conclusion is consistent with evidence that N-type
PARAMETERS OF SUBJECT INFLECTION IN ITALIAN DIALECTS
197
agreement (excluding Number) can characterize not only VSO languages (e.g., Welsh, cf. Rouveret 1994), but also SVO languages (e.g., the Belfast dialect of English described by Henry 1995). 8. In the words of the anonymous reviewer, "the common-sense—and Joosian—view that languages can vary along all sorts of unpredictably independent dimensions."
References Alexiadou, Artemis, & Elena Anagnostopoulou. 1998. Parametrizing AGR: Word order, Vmovement and EPP-checking. Natural Language & Linguistic Theory 16,491-539. Arad, Maya. 1998. VP-Structure and the Syntax-Lexicon Interface. Ph.D. dissertation, University College, London. Baker, Mark. 1996. The Polysynthesis Parameter. Oxford University Press, New York. Benveniste, Emile. 1966. Problemes de linguistique generate. Gallimard, Paris. Bernstein, Judy. 1991. DPs in French and Walloon: Evidence for parametric variation in nominal head movement. Probus 3,101-126. Borer, Hagit. 1986.1-subjects. Linguistic Inquiry 17, 375-416. Borer, Hagit. 1994. The projection of arguments. In Elena E. Benedicto & Jeffrey T. Runner (eds.), Functional Projections, 19-47. GLSA, University of Massachusetts, Amherst. Brandi, Luciana, & Patrizia Cordin. 1981. Dialetti e italiano: un confronto sul parametro soggetto nullo. Rivista di Grammatica Generativa 6, 33-87. Brandi, Luciana, & Patrizia Cordin. 1989. Two Italian dialects and the null subject parameter. In Osvaldo Jaeggli & Kenneth J. Safir (eds.), The Null Subject Parameter, 111-142. Kluwer, Dordrecht. Brody, Michael. 1997. Perfect chains. In Liliane Haegeman (ed.), Elements of Grammar: Handbook of Generative Syntax, 139-167. Kluwer, Dordrecht. Burzio, Luigi. 1986. Italian Syntax. Reidel, Dordrecht. Cardinaletti, Anna. 1997. Agreement and control in expletive constructions. Linguistic Inquiry 28, 521-533. Cardinaletti, Anna, & Ian Roberts. 1991. Levels of representation of agreement. Ms., University of Venice and University of Geneva. Chomsky, Noam. 1981. Lectures on Government and Binding. Foris, Dordrecht. Chomsky, Noam. 1982. Some Concepts and Consequences of the Theory of Government and Binding. MIT Press, Cambridge, Mass. Chomsky, Noam. 1986. Knowledge of Language. Praeger, New York. Chomsky, Noam. 1995. The Minimalist Program. MIT Press, Cambridge, Mass. Chomsky, Noam. 1998. Minimalist inquiries: The framework. MIT Occasional Papers in Linguistics 15, 1-56. Cinque, Guglielmo. 1990. Types of A'-Dependencies. MIT Press, Cambridge, Mass. Cinque, Guglielmo. 1993. A null theory of phrase and compound stress. Linguistic Inquiry 24, 239-298. Cinque, Guglielmo. 1999. Adverbs and Functional Heads: A Cross-Linguistic Perspective. Oxford University Press, New York. Di Domenico, Elisa. 1998. Per una teoria del genere grammaticale. UniPress, Padua. Donati, Caterina. 1998. Elementi di sintassi della comparazione. Ph.D. dissertation, University of Florence. Grimshaw, Jane. 1995. Argument Structure. MIT Press, Cambridge, Mass. Henry, Alison. 1995. Belfast English and Standard English: Dialect Variation and Parameter Setting. Oxford University Press, New York.
a
Subjects
Higginbotham, James. 1985. On semantics. Linguistic Inquiry 16, 547-593. Hornstein, Norbert. 1999. Movement and Control. Linguistic Inquiry 30, 69-96. Jaeggli, Osvaldo, & Kenneth J. Safir. 1989. The null subject parameter and parametric theory. In Osvaldo Jaeggli & Kenneth J. Safir (eds.), The Null Subject Parameter, 1-44. Kluwer, Dordrecht. Johnson, Kyle. 1991. Object positions. Natural Language & Linguistic Theory 4, 577-636. Kayne, Richard S. 1975. French Syntax. MIT Press, Cambridge, Mass. Koopman, Hilda. 1996. The Spec head configuration. Ms., UCLA. Manzini, M. Rita, & Anna Roussou. 1999. A minimalist theory of A-movement and control. UCL Working Papers in Linguistics 11, 403-440. Manzini, M. Rita, & Leonardo M. Savoia. 1997. Null subjects without pro. UCL Working Papers in Linguistics 9, 301-313. Manzini, M. Rita, & Leonardo M. Savoia. 1998a. Varieta romanze a Verbo Secondo. Studi in memoria di Giorgio Chiarini, 459—492. Alinea, Florence. Manzini, M. Rita, & Leonardo M. Savoia. 1998b. Negation parameters and their interaction in Italian dialects. Quaderni di lavoro dell'ASIS 2: 39-60, University of Padua. Manzini, M. Rita, & Leonardo M. Savoia. 1998c. Clitics and auxiliary choice in Italian dialects: Their relevance for the Person ergativity split. Recherches linguistiques a Vincennes 27, 115-138. Manzini, M. Rita, and Leonardo M. Savoia. (Forthcoming). / dialetti italiani. II Mulino, Bologna. Manzini, M. Rita, & Kenneth Wexler. 1987. Binding theory, parameters and learnability. Linguistic Inquiry 18, 413—444. Martin, Jean-Baptiste. 1974. Le pronom personnel de la 3e personne en francoprovencal central (formes et structures). Travaux de Linguistique et Litterature 12.1, 85-116. Nash, Le"a, & Alain Rouveret. 1997. Proxy categories in phrase structure theory. In Kiyomi Kusumoto (ed.), The Proceedings of the North East Linguistic Society 27, 287-304. GLSA, University of Massachusetts, Amherst. Platzack, Christen 1995. Null subjects, weak Agr and syntactic differences in Scandinavian. Working Papers in Scandinavian Syntax 53, 85-106. Poletto, Cecilia. 1993. La sintassi del soggetto nei dialetti italiani settentrionali. Unipress, Padua. Poletto, Cecilia. 1997. The internal structure of AgrS and subject clitics. Ms., University of Padua/CNR. Pollock, Jean-Yves. 1996. Language et cognition: Introduction au programme minimaliste de la grammaire generative. Presses Universitaires de France, Paris. Riemsdijk, Henk van, & Edwin Williams. 1986. Introduction to the Theory of Grammar. MIT Press, Cambridge, Mass. Ritter, Elizabeth. 1991. Two functional categories in Noun Phrases: Evidence from Modern Hebrew. In Susan Rothstein (ed.), Perspectives on Phrase Structure: Heads and Licensing, 37-62. Academic Press, New York. Ritter, Elizabeth. 1993. Where's gender? Linguistic Inquiry 24, 795-803. Rizzi, Luigi. 1982. Issues in Italian Syntax. Foris, Dordrecht. Rizzi, Luigi. 1986. On the status of subject clitics in Romance. In Osvaldo Jaeggli & C. Silva-Corvalan (eds.), Studies in Romance Linguistics, 391-419. Foris, Dordrecht. Ronjat, Jules. 1937. Grammaire historique des parlers provengaux modemes. Societe des langues romanes, Montepellier. Rouveret, Alain. 1994. Syntaxe du gallois. CNRS Editions, Paris.
PARAMETERS OF SUBJECT INFLECTION IN ITALIAN DIALECTS
199
Taraldsen, Knut Tarald. 1978. On the NIC, vacuous application, and the that-t filter. Ms., MIT. Distributed by Indiana University Linguistics Club. Vanelli, Laura. 1998. / dialetti italiani settentrionali net panorama romanzo. Bulzoni, Rome. Zubizarreta, Maria-Luisa. 1998. Prosody, Focus and Word Order. MIT Press, Cambridge, Mass. Zucchi, Alessandro. 1996. Subject clitics in Piacentino. Ms., Cornell University.
This page intentionally left blank
8 Subject Positions and the Placement of Adverbials Peter Svenonius
1. A sounding device for phrase structure The position of adverbs relative to other elements has been a critical diagnostic in the analysis of phrase structure. Earlier generative work on adverbs, such as Jackendoff 1972 and Ernst 1984, allowed for some flexibility in adverb placement, but noted certain regularities. Work on other aspects of clausal structure, such as Emonds 1976 and Platzack 1983, exploited the regularities, taking adverbs to mark phrasal boundaries, and used them to demonstrate the movement of other elements across them. Such work has actually been the primary motivation for functional heads in the clause. Morphological evidence is also used, but usually in a supporting role. In this section I briefly examine some of the basic arguments using adverbs to motivate models of phrase structure. My purpose here is not to restate the arguments that have been made, which are explicated clearly in such works as Webelhuth 1992, Holmberg and Platzack 1995, Vikner 1995, and references there. My purpose in this section is rather to point out the extent to which adverb placement has been used as a diagnostic for the movement or location of other elements. 1.1. Adverbs and heads Pollock (1989) proposed, based on the relative positions of verbs and adverbs in French and English, that the Infl node be split into at least two parts, Agr(eement) and T(ense), providing two different landing sites for the verb. Assuming that temporal and manner adverbials are adjoined to VP, below both Agr and T, but that negation is located between Agr and T, Pollock argued that Adv-V order was the result of no verb movement, Neg-V-Adv order was the result of short verb movement (to Agr), and V-Neg order was the result of long verb movement (to T). Belletti (1990) proposed a modification of Pollock's analysis which was motivated primarily by morphological considerations. She proposed that Agr (or AgrS) 201
202
SUBJECTS, EXPLETIVES, AND THE EPP
dominates T, rather than the other way around, since subject agreement morphemes typically appear farther from the stem than tense morphemes, in languages where they can be identified. The idea was that head movement of the verb to the functional positions was directly connected to the appearance of the inflectional morphemes.1 Despite various critiques, the Split-Infl hypothesis has been employed in a wide range of analyses. I assume some version of it for most of this chapter but in §4.5 I discuss an alternative which provides some advantages. 1.2. Adverbs and arguments Various work on the structure of Scandinavian, for example Platzack 1983, 1986a, 1986b, and Holmberg 1983,1986, has relied crucially on adverb placement to identify the VP boundary and thereby to demonstrate when verb movement takes place. For object shift in Scandinavian, and for scrambling in German, adverb placement has also proven useful in identifying the location of arguments, as in Webelhuth 1984, 1992, and Diesing 1992. This line has been extended to the location of subjects in Scandinavian in Jonas 1993 and Bobaljik and Jonas 1996, inter alia, where they argue that Icelandic, German, and one dialect of Faroese have two subject positions, while English, Mainland Scandinavian, and another dialect of Faroese have only one. This theory is pursued in Thrainsson 1997, Bobaljik 1995, and Bobaljik and Thrainsson 1997, where the connection between functional specifier positions and overt inflection is explored in detail. I discuss this issue more fully in §4.2 below. 1.3. Adverbs as primary motivation for clause structure A more recent development, pioneered by Alexiadou (1997) and Cinque (1999), has been to use adverbials themselves to motivate clause structure. In those approaches, arguments from inflectional morphology (cf. §1.1) are combined with arguments from the distribution of adverbs, which are taken to occupy specifiers of functional heads. Alexiadou identifies classes of adverbs with particular functional projections, both in terms of co-occurrence and in terms of word order. For example, Greek morphology motivates an aspectual projection for perfective versus imperfective features. These features interact with such adverbs as sinithos 'usually' and djo fores 'twice'; the former are possible with the imperfective form of a verb, but not with the perfective, while the latter show the reverse restriction. Alexiadou also shows that the relative order of adverbs matches the relative order of morphemes: suffixes indicating mood are farther from the stem than suffixes indicating aspect, and aspectual adverbs obligatorily appear lower in the tree than (i.e., to the right of) epistemic adverbs, which are sensitive to the mood of the sentence and arguably concern the same semantic domain as mood inflection. Alexiadou accounts for these facts by connecting the adverbs explicitly to the functional structure: adverbs occupy the specifier positions of the various functional heads. Thus epistemic adverbials, in the specifier of MoodP, precede aspectual adverbials, in the specifier of AspP; and morphological material motivates the same order of functional heads.
SUBJECT POSITIONS AND THE PLACEMENT OF ADVERBIALS
203
In some cases, Alexiadou adopts functional heads which have been motivated for other languages in part because they provide a location for adverbs in Greek. For example, Alexiadou notes that certain negative polarity adverbs, such as pja 'anymore,' appear below subject-oriented adverbs (and above aspectual adverbs), although the negation particle appears to be higher in the tree than AgrS, the head which Alexiadou argues to head the relevant functional projection for subject-oriented adverbs. There is in any case no inflectional indication of a negation head between AgrS and Asp; however, Zanuttini (1997) has argued on the basis of an extensive analysis of Romance dialects that there is a lower Neg projection, Neg2, below AgrS and above Asp. Alexiadou adopts a Neg2 projection for Greek as well, which then provides a site for the negative polarity elements. Cinque 1999 represents a very similar enterprise but with a wider and more cross-linguistic scope. Starting from the relative order of a wide selection of adverbials in Italian, and then introducing supporting evidence from the relative order of adverbials and the relative order of inflectional morphemes in other languages, he proposes over thirty separate functional heads. I take up specific aspects of Cinque's arguments below, including Nilsen's (1998) application of Cinque's model to Norwegian. 1.4. Specifiers and adjuncts A word or two needs to be said about the distinction between specifiers and adjuncts. The traditional assumption has long been that adverbs are adjuncts; this is assumed to be consistent with their optionality, their iterability, and restrictions on their extraction. Alexiadou 1997 and Cinque 1999 were noted above to argue, on the other hand, that adverbs occupy specifier positions. It is not clear that this raises any problems with respect to the properties of optionality, iterability, and difficulty of extraction. The optionality of a given adverb is not any more problematic on such a view than the optionality of any other filled specifier position (e.g., a possessor in a noun phrase); iterability is handled by the multiplicity of heads providing specifier positions; and restrictions on extraction can be stated in other ways (cf. for example Lasnik and Saito 1992). In fact, it is unclear that adjunction even necessarily exists on a model like Alexiadou's or Cinque's (cf. Hellan 1989, Hoekstra 1991 or Kayne 1994, where it is argued that there is no structural distinction between an adjunct and a specifier). Hoekstra (1991; cf. also Hellan 1989) specifically suggests that a specifier is, by definition, an adjunct which enters into an agreement relation with the head of the phrase it is adjoined to (cf. his p. 24). Uniqueness of specifiers, as Hellan (1989) argues, will generally be guaranteed by the nature of the relations that license them (e.g., feature checking). The iterability or noniterability of adjunct elements, on this view, will be determined by the nature of their licensing. An opposing view is provided by Kayne 1994, in which it is argued that more general constraints on phrase structure make impossible multiple adjunctions to the same phrase. Cinque's proposal in particular is fully consistent with Kayne's theory, while the one that I will propose here is not. Critical to this discussion is the segment-category distinction. A common assumption is that a specifier is dominated by a 'category' (i.e., a maximal projec-
204
SUBJECTS, EXPLETIVES, AND THE EPP
tion), while an adjunct is not; adjunction to a projection creates a two-segment configuration, and the adjunct is dominated only by one of the two segments (Chomsky 1986). If segments are really nondistinct from each other, as is sometimes assumed (cf. May 1985), and if there is multiple adjunction, then linear order of two elements adjoined to the same phrase will not reflect their scope. Thus, for May, the two LF structures in (1) for the sentence Exactly three girls kissed few boys are semantically equivalent (i.e., for May they would both be ambiguous). (1)
a. b.
[sexactly three girlsx [sfew boysy [s x kissed y]]] [sfew boysy [sexactly three girlsx [s x kissed y]]]
Similarly, if attributive adjectives in noun phrases involve multiple adjunction to NP, then the two structures in (2) should be equivalent, contrary to fact (cf. Svenonius 1993). (2)
a. b.
a [NP dangerous [NP dead [^p animal ]]] a [NP dead [^ dangerous [NP animal ]]]
The same observations carry over to adverbs. If adverbs are adjoined iteratively to the same node, say VP, then there are structures like those in (3). (3)
a. b.
Mockingbirds [Vp characteristically [VP often [Vp imitate woodpeckers ]]] Mockingbirds [yp often [yp characteristically [Vp imitate woodpeckers ]]]
Given that the two different word orders represented in the sentences in (3) do not mean the same thing, just as with the noun phrase examples in (2), one of the two assumptions made above is incorrect: either adjunction is not iterative, to the same node, or else segments count for determining c-command relations (i.e., scope is not determined under M-command). If multiple adjunction is eliminated, then there must be a very large number of maximal projections in the clause, as is demonstrated by Cinque 1996 (cf. §2 below). If, on the other hand, multiple adjunction is to be retained, then each segment must receive a separate interpretation, as in Heim 1982; see Lasnik and Saito 1992 for syntactic arguments that segments should count as categories, and Barbiers 1995 for arguments that they must count for calculation of c-command relations. I will return to this issue in §5 below. My working hypothesis in this chapter is that multiple adjunction is allowed. The strongest arguments against multiple adjunction (Kayne 1994) are theoretical in nature, and subject to empirical motivation; Cinque 1999 constitutes empirical support for that position, but to the extent that the analysis here is successful, it provides an argument against it.
2. Challenges to the 'Tight-fit' theory Ernst (1984), discussing Jackendoff's (1972) analysis of adverb placement and interpretation in English, refers to Jackendoff's proposal as a 'Tight-fit' theory, because of the restrictive way in which particular lexical items (i.e., adverbs) are linked to particular syntactic positions. Ernst points out that this leads to a great de-
SUBJECT POSITIONS AND THE PLACEMENT OF ADVERBIALS
205
gree of homonymity, since adverbs which can be used in different ways, for instance as 'speaker-oriented' or as 'manner' adverbials, must have distinct lexical entries. Ernst proposes instead a 'Loose-fit' theory, in which adverb meanings are sufficiently flexible that their interpretation can be determined in part by their position, leading to a reduction in lexical entries. Thus, for Ernst, one and the same adverb may be attached in multiple locations. In many ways the theories of Alexiadou and Cinque discussed in the previous section are reminiscent of the 'Tight-fit' position; there is, at any rate, a tight fit between adverb classes and syntactic positions. Despite the changes which have affected the standard theory since Jackendoff's work (and Ernst's), a number of problems still arise for any analysis which maintains that adverbs are consistently adjoined to certain nodes. Some are noted in Cinque 1999. Here I discuss some of the same problems as well as some additional ones. 2.1. Adverb placement versus adverb ordering The first problem is that adverbs show a fair amount of variability in placement, but are subject to strict ordering restrictions. For example, from (4b) it can be seen that already can appear before the finite auxiliary has (although this order is slightly marked), and from (4c) it is clear that probably can appear after has. (4)
a. b. c. d.
John has already left. John already has left. John has probably left. John probably has left.
However, although the two adverbs can co-occur in a single clause, as shown in (5a), they cannot appear in the order already-probably. (5)
a. John probably has already left b. ?? John already has probably left
Exactly the same facts can be demonstrated for Greek, as shown in (6)-(7) (cf. Alexiadou 1997:166). (6)
a. b.
(7)
a. b.
O Janis idi ehi figi. the John already has left O Janis ehi idi figi. the John has already left
Greek
O Janis pithanos efi figi. the John probably has left O Janis ehi pithanos figi. the John has probably left
Just as in English, the two adverbs can co-occur, but their relative order is strict. (8)
a.
O Janis pithanos ehi idi figi. the John probably has already left b. ?? O Janis idi ehi pithanos figi. the John already has probably left
Greek
206
SUBJECTS, EXPLETIVES, AND THE EPP
Alexiadou argues for examples like (la) that epistemic adverbs occupy SpecMoodP, and that Mood is higher in the tree than AgrS, the head which the finite verb occupies. For examples like (7b), she suggests that there is a second MoodP dominating the participle (see also Kayne 1993 and Collins and Thrainsson 1996 for analyses in which nonfinite verb forms are dominated by a variety of functional projections). However, if the distribution of adverbs is to be explained by linking them to functional projections, and if there are two sets of (some) functional projections in clauses with complex verbs, then additional assumptions are necessary to rule out examples like (8b). Similar facts can be demonstrated for Finnish (examples taken from Holmberg et al. 1993). (9)
a.
b. (10) a.
b.
Pekka ei olisi aina valittanut siita. Pekka not would always complained about.it 'Pekka wouldn't have always complained about it' Pekka ei aina olisi valittanut siita. Pekka not always would complained about.it
Finnish
Pekka ei ehka olisi valittanut siita. Pekka not perhaps would complained about.it 'Perhaps Pekka wouldn't have complained about it' Pekka ei olisi ehka valittanut siita. Pekka not would perhaps complained about.it
In (9)-(10) it is shown that aina 'always' and ehka 'perhaps' can appear both before and after the modal verb olisi. (lla) shows that the two adverbs can co-occur, and (lib) shows that their order is strict. (11) a.
Pekka ei ehka olisi aina valittanut siita. Finnish Pekka not perhaps would always complained about, it 'Perhaps Pekka wouldn't have always complained about it' b. * Pekka ei aina olisi ehka valittanut siita. Pekka not always would perhaps complained about.it
Holmberg et al. suggest there is LF-movement of adverbs; thus, in an example like (lOb), ehka 'perhaps' appears below the position in which it is interpreted and licensed, and must raise at LF. (lib) is blocked because of Relativized Minimality (Rizzi 1990), which prevents one adverb crossing another (at LF, as well as overtly). However, phrasal movement of adverbs is in general seriously undermotivated;2 allowing LF movement would tend to make false predictions with respect to the possible scopal interpretations of adverbs (a particularly clear statement of this can be found in Ladusaw 1988, for example). The problem is cast into sharper relief in Cinque's 1999 examples. He gives sets of Italian sentences to demonstrate that between the various adverb positions there are head positions which can serve as a landing site for the verb. For example, he gives the sentences in (12)-(13).3
SUBJECT POSITIONS AND THE PLACEMENT OF ADVERBIALS (12) a.
b. c. (13) a.
b. c.
207
Allora aveva forse saggiamente deciso di non presentarsi. Ita. then had perhaps wisely decided to not present. RFX 'Then he had perhaps wisely decided not to go' Allora forse aveva saggiamente deciso di non presentarsi. then perhaps had wisely decided to not present.RFX Allora forse saggiamente aveva deciso di non presentarsi. then perhaps wisely had decided to not present.RFX Non hanno rimesso di solito mica piu . . . tutto bene in ordine. not have put usually not anymore all well in order They haven't usually put everything well in order anymore' Non hanno di solito rimesso mica piu . . . tutto bene in ordine. not have usually put not anymore all well in order Non hanno di solito mica rimesso piu . . . tutto bene in ordine. not have usually not put anymore all well in order
As can be seen, the finite verb in (12) and the participle in (13) can appear in various positions with respect to the adverbs, but the ordering of adverbs is strict. For Cinque, this means that the verb optionally moves to one of several different functional heads. If movement is motivated by strong features, this suggests a large number of optionally strong features in Italian (quite a few, if one considers the whole range of adverbial positions). Below, I will propose an alternative account of this variation which makes no use of optionally strong features. 2.2. Subject positions A similar problem arises with subject positions. In many languages, the subject can appear in various positions with respect to adverbial elements (as has often been noted; fuller discussion and references are given in §4). For example, in Cinque's hierarchy of adverbs, an adverb like 'still' is quite low, but in Icelandic the subject can follow it, as shown in (14a).4 (14) a.
b.
Pess vegna 6'gra ennpa morg leikrit ahorfendum nutimans. Ice. this cause provoke still many plays audiences today's 'Because of this, many plays still provoke today's audiences' Pess vegna ogra morg leikrit ennb£ ahorfendum nutimans. this cause provoke many plays still audiences today's
Nilsen (1998) shows that the subject in Norwegian can precede or follow any adverb within IP (i.e., following an element in C) for the range of Cinque's adverbial classes that extends from cerlig talt 'honestly' down to snort 'soon' (a couple of levels below enda oifortsatt 'still'); a representative example is given below (from Nilsen 1997:17). (15) a.
. . . at Per tydeligvis ikke lenger bestandig vinner. Norwegian that Per evidently not anymore always wins '. . . that Per evidently doesn't always win anymore'
208
SUBJECTS, EXPLETIVES, AND THE EPP b. c. d. e.
. . . at tydeligvis Per ikke lenger bestandig vinner. that evidently Per not anymore always wins . . . at tydeligvis ikke Per lenger bestandig vinner. that evidently not Per anymore always wins . . . at tydeligvis ikke lenger Per bestandig vinner. that evidently not anymore Per always wins . . . at tydeligvis ikke lenger bestandig Per vinner. that evidently not anymore always Per wins
If, as Cinque argues, the various possible positions for the subject in the string of adverbials indicate different landing sites for the subject, then the Norwegian facts motivate twenty different subject landing sites (only fifteen in Italian; the subject cannot appear as low in Italian, according to Cinque). This is in addition to the thirty or so functional projections necessary to host the adverbs. Not only must there be twenty AgrPs (or TopicPs, FocusPs, etc.) in Norwegian, but each one must have an optionally strong feature, to motivate subject movement in those cases where it occurs. Cinque (1999, ch. 5) suggests that the different subject positions may reflect different interpretational possibilities (e.g., distributive versus collective interpretations, focus versus topic interpretations, etc., as in Beghelli and Stowell 1997, in which such interpretations are linked to specific functional positions). Indeed, there is some interaction between subject position (relative to adverbs) and interpretation. I discuss this more fully in §4. 2.3. Object shift The position of the object creates problems of a type very similar to those noted for subjects above. If object agreement morphemes are an indication, then object agreement is often low in the tree (e.g., it is placed below AgrS, Tense, and negation by Belletti 1990, below Tense and Mood by Julien 2000). A couple of suggestive examples showing object agreement inside tense and finiteness have been borrowed here from Julien 2000: (16) a.
b.
kaci -ma -ra -ygi raise-lsO-PAST-2sS 'you raised me' uni dal -oco -m -a -e he beat-caus-2sO-FiN-3sS 'he makes you beat'
Tarma Quechua
Santali
But object shift raises objects past not only aspectual adverbials, but also sentence adverbials. (17) a. b.
Han sendte det heldigvis ned. Hann sendi bad sem betur fer nidur. he sent it fortunately down 'He fortunately sent them down'
Norwegian Icelandic
SUBJECT POSITIONS AND THE PLACEMENT OF ADVERBIALS c. d.
Han sendte brevet heldigvis ned Hann sendi brefid sem betur fer niSur he sent the.letter fortunately down 'He fortunately sent the letter down' e. * Han har sendt {det/brevet} heldigvis ned f. * Hann hefur sent {baoYbrefi5} sem betur fer nidur he has sent it /the.letterfortunately down
209 Norwegian Icelandic
Norwegian Icelandic
(17a-b) show object shift of a pronoun across a sentential adverb meaning 'fortunately'; (17c-d) show the same thing for a full DP.5 Object shift of full DPs is restricted in Mainland Scandinavian, but not to the extent usually assumed, as argued in Nilsen 1997,1998.1 use a particle verb here to demonstrate that the adverbial is not postposed; (17e-f) establish the same fact in that they show that the object cannot precede a sentential adverbial when the environment for object shift is not satisfied. Since objects can also appear following these and other adverbials (cf. Nilsen 1997), it seems that a Cinquean analysis is forced to countenance a significant number of object landing positions. Nor is it enough to suggest that objects may avail themselves of the same landing sites as the subjects, since subject, indirect object, and direct object may all appear between the same pair of adverbs. Consider the following examples from Icelandic (cf. also Collins and Thrainsson 1996, and Nilsen 1997:27 for Norwegian). (18) a.
b. c.
M gafu vonandi alltaf einhverjir Sigridi hana tilbaka. then gave hopefully always somebody Sigrid it back 'Then somebody hopefully always gave it back to Sigrid' M gafu vonandi einhverjir Sigridi hana alltaf tilbaka. then gave hopefully somebody Sigrid it always back M gafu einhverjir Sigridi hana vonandi alltaf tilbaka. then gave somebody Sigrid it hopefully always back
Ice.
Since all three arguments can precede, intervene between, or follow the two adverbs, this seems to motivate three specifier positions, hence three heads, between every one of the head positions argued to exist for adverbial placement.6 Nilsen, taking Cinque's analysis of the adverbial positions to be correct, interprets the type of pattern in (18) as providing evidence against a movement to specifier account of subject and object positions, assuming that PF-reordering takes place. However, there are interpretive differences for different subject-adverb orderings (which I will detail in §4), differences which I locate in the LF component.
3. Introducing the right amount of flexibility into adverb placement The solution I propose to the problems noted above is straightforward: adverb attachment is driven by interpretation. This is the explicit assumption of earlier work (e.g., Jackendoff 1972, McConnell-Ginet 1982, Ernst 1984), and more or less drives the intuitions behind more recent work. However, it goes against the thrust
210
SUBJECTS, EXPLETIVES, AND THE EPP
of Alexiadou and Cinque, for whom there is a strict linking of adverb classes and functional categories, mediated in Cinque's theory by feature-checking. The Cinque and Alexiadou accounts are worked out in book length and are applied to a large number of specific cases. My account, in contrast, is less thoroughly developed and is applied only to one case, that of Germanic subject-adverb order. However, given that much of the received wisdom about adverb placement is based on exactly the kind of case that German subject-adverb order represents, my account, if successful, raises important questions for the functional head-based ones. As noted above in §2, the approach pursued here is closer to Ernst's 'Loose-Fit' theory than the other proposals. This introduces flexibility at three points. First, a node of a given category may not always receive the same interpretation. For example, an adverb might attach to an event-denoting node, and not to a node that denotes a state of affairs, but VP might variably denote an event or a state of affairs. Second, since interpretation is a matter not of Deep Structure category labels, the relevant factor in the Tight-fit theories, but of LF interpretations, any covert movement that affects interpretation can potentially affect adverb placement, including restructuring. Third, a node that contributes no semantic information that is relevant to the interpretation of a given adverb will be, in effect, invisible for that adverb. I discuss these various points in §3.1 immediately below. I make the additional assumption, as noted in § 1.4 above, that segments created by adjunction are nodes and receive their own interpretation. This assumption is logically independent from the assumption that adverb interpretation is sensitive to semantic interpretation only and not to syntactic category, but the two interact in a critical fashion. In §3.2 I show how these assumptions make sense of the head-adverb ordering facts discussed in §2.1 above, and in §3.3 I discuss adverb ordering restrictions. 3.1. Adverbs are attached according to semantic interpretation Consider examples like those in (19). (19) a. b. c. d.
Howard foolishly may have been trying to impress us. Howard may foolishly have been trying to impress us. Howard may have foolishly been trying to impress us. Howard may have been foolishly trying to impress us.
If the adverb in all of these sentences is the same, then these sentences indicate that such adverbs can be attached in a number of places. A head movement account is inadequate, since heads generally cannot cross other head positions. If all of the auxiliaries in (19a) occupy lower positions, and if they are all taken to have moved in (19d), then each one will have to have crossed at least two other head positions. Assume, then, that foolishly attaches to any of several maximal projections, perhaps under the conditions that it be c-commanded by the controlling subject, and that it c-command the node that denotes the event or action that the subject was foolish to perform. The aspectual, temporal, and modal information represented by the auxiliaries is simply not relevant to these conditions. (20) a.
Cari probably will have been finishing up by then.
SUBJECT POSITIONS AND THE PLACEMENT OF ADVERBIALS
211
b. Cari will probably have been finishing up by then. c. % Cari will have probably been finishing up by then. d. ?? Cari will have been probably finishing up by then. Here the semantics of the epistemic adverb dictate that the adverb scope over the tense, which is borne by the auxiliary will. If we assume that will is optionally in T or in Agr, then the adverb can occupy a single location (adjoined to TP) and scope over T. This is essentially the approach taken by Cinque. Assuming (20c-d) to be ungrammatical, as they are for many speakers,7 the correct results are achieved, but at the cost of an optionally strong head feature in Agr. Another approach is that taken by Holmberg et al. 1993, in which a definition of c-command is adopted which allows the adverb in (20b) to scope over the head, even though it follows it linearly: c-command is determined by the first maximal projection, and the adverb, being adjoined to TP, is not dominated by TP; therefore, the first maximal projection dominating probably is in fact AgrP, which also dominates will. However, as noted in §1.4 above, this kind of approach to c-command means that adjacent strings of adverbs mutually c-command each other; for example, already c-commands probably in an example like that in (21a) below, assuming that they are adjoined to the same maximal projection. (21) a. Howard will probably already have been finishing up by then, b. * Howard will already probably have been finishing up by then. If there is no asymmetric structural relation between the two adverbs in the examples in (21), then it is difficult to explain why (21b) is bad. In fact, it becomes generally difficult to explain why the linear order of adverbs correlates so well with their semantic scope. Instead, I suggest, will occupies the same node in (20a-b). Take that node to be, for example, Agr. The condition on probably is that it c-command tense, dictated by its interpretation. The interpretation of Agr is irrelevant to probably, so it can attach above or below that node. Note that it is not necessary that there be a functional T node that will has moved from; in fact, the Agr node can be the T node (i.e., it is Infl, a combination of Agreement and Tense and perhaps other features). What is necessary is that there be a node below probably in which the tense feature is interpreted. This is true as long as will has undergone head movement in (20b) from some lower position (it might be a lexical projection), and can be interpreted in that position (under reconstruction). If, on the other hand, reconstruction of heads is not permitted, then there must be a T node below the position of probably in (20b). 3.2. A floating strong head feature is unnecessary Given these assumptions, the examples in (12) and (13) in §2.1 above (repeated below as (22) and (23)) can be seen in a new light. Rather than having to assume a floating strong feature (i.e., one which can appear in any of several locations), for each of the two cases there can be a single strong feature in a functional head whose interpretation is not relevant to the adverbials, for example Agr; the verbal head obligatorily moves to Agr.
212
SUBJECTS, EXPLETIVES, AND THE EPP
(22) a.
b. c.
Allora aveva forse saggiamente deciso di non presentarsi. Ita. then had perhaps wisely decided to not present.RFX 'Then he had perhaps wisely decided not to go' Allora forse aveva saggiamente deciso di non presentarsi. then perhaps had wisely decided to not present.RFX Allora forse saggiamente aveva deciso di non presentarsi. then perhaps wisely had decided to not present.RFX
For the case in (22), say that the adverbs have to scope over T, but that Agr provides no information relevant to their interpretation. There is a strong feature in Agr, which attracts the verb. The verb always lands there, but the adverbs are optionally attached either to AgrP or to TP. Some may be attached to AgrP, others to TP. Alternatively, as noted above, there need not be any TP, but aveva can be taken instead to have moved from a lexical V position (as long as the tail of the chain it heads after movement contains a copy of the tense feature, or whatever feature is relevant for the adverbs). For the case in (23), assume another functional head, Agr, below T but above VP. The adverbs here must scope over VP, but once again, Agr is irrelevant. Agr contains a strong feature which attracts the participle. (23) a.
b. c.
Non hanno rimesso di solito mica piu . . . tutto bene in ordine. not have put usually not anymore all well in order 'They haven't usually put everything well in order anymore' Non hanno di solito rimesso mica piu . . . tutto bene in ordine. not have usually put not anymore all well in order Non hanno di solito mica rimesso piu . . . tutto bene in ordine. not have usually not put anymore all well in order
Again, this is not to say that Agr must be semantically vacuous, only that its contribution to interpretation is not relevant to this set of adverbials. The account sketched here reduces the number of functional heads necessary in the clause to two, whereas Cinque's and Alexiadou's require at least one per adverb. It also completely eliminates the need for a floating strong feature, and allows instead two consistently strong verbal features in the Italian clause. This proposal is not consistent with Kayne's claim that adjuncts are unique. In order for this model to be consistent with Kayne 1994, there must be one head for each adverb. However, this does not mean that Cinque's position must be adopted. Nothing in Kayne's theory requires that the heads be semantically coordinated with the phrasal expressions attached to them, given that interpretation occurs at LF, not at DS. The heads could even be semantically vacuous, in principle. Thus, a Kaynean version of the approach advocated here might involve free generation of vacuous Agr heads as needed to provide sites for adverb attachment. 3.3. Ordering restrictions Alexiadou and Cinque assume that ordering restrictions are determined by the obligatory sequence of functional heads. On my account, this explanation for ordering restrictions is not available, and I must appeal to semantic interactions
SUBJECT POSITIONS AND THE PLACEMENT OF ADVERBIALS
213
among the adverbs. Cinque notes (ch. 6) that in many cases, ordering restrictions can be motivated by semantic factors, for example the fact that temporally oriented adverbs precede aspectual ones. However, he suggests that there are several cases in which semantic considerations cannot explain the observed restrictions. He gives three examples. I will examine two of them briefly in turn. First, Cinque suggests (p. 135) that the relative ordering of 'evidentially' adverbials with epistemic ones, as seen in (24), is not semantically motivated. (24) a. Al evidently will probably give up. b. * Al probably will evidently give up. However, a semantic account seems eminently plausible. Take, for example, probably to be a function from propositions to propositions (probably too simple, but sufficient for present purposes). If evidently operates on a propositional object and returns something else, say a FACT in the sense of Vendler 1967, then probably will not be able to apply to its output. In support of this, it seems that it is difficult to embed a sentence with evidently under a higher clause that expresses the probability of the embedded clause. Cinque gives a sentence like the one in (25a), which is interpretable, though degraded. Note also that unlike (24a), evident in (25a) seems to assert evidentiality for arbitrary individuals, rather than for the speaker. Similar is (25b). (25) a. ? It is probable that it is evident that he is the guilty one. b. ? It is probable that Al evidently will give up. c. He discovered that Al evidently would give up. d. He probably discovered that Al evidently would give up. As indicated in (25c), the same clause is much better when embedded under a factive verb like discover, which takes fact complements (cf. Kiparsky and Kiparsky 1970). This higher clause can be modified by probably, as expected. Another example Cinque gives of an ordering which cannot be semantically motivated (p. 136) is the following: habitual adverbs like di solito 'usually' obligatorily precede terminative adverbs like piu 'no longer.' (26) a.
Dopo le 10, Gianni non beve di solito piu niente. after the 10 Gianni not drinks usually anymore anything 'After 10, Gianni doesn't usually drink anything anymore' b. * Dopo le 10, Gianni non beve piu di solito niente. after the 10 Gianni not drinks anymore usually anything
Ita.
In this case, I agree with Cinque that there is no obvious semantic reason that an adverb marking the termination of an event or event-type should not be able to scope over a habitual adverb. However, in this case, exactly that is quite possible in other languages, for example English. (27) a. b.
After 10, John usually no longer drinks anything, After 10, John no longer usually drinks anything.
The two examples show different interpretations in the expected way: (27a) means that it is usually the case that John no longer drinks (i.e., he drinks up until ten),
214
SUBJECTS, EXPLETIVES, AND THE EPP
while (27b) means that it is no longer the case that John usually drinks (i.e., he used to drink after ten). Thus, it seems that the restriction exhibited in the Italian example in (26) cannot be due to a universal ordering on the heads Habitual and Terminative, as Cinque suggests.8 3.5. Inflection Assuming that a similar story can be told for each pair of adverbs that are (more or less) strictly ordered, the motivation for the many functional heads proposed by Cinque is drastically reduced. But the beauty of the Alexiadou/Cinque approach is in the connection between adverb order and the ordering of inflectional affixes. For example, Cinque points out that 'epistemic mood' affixes appear outside past tense, as in the example in (28a) (from his p. 72, citing Bybee). (28) a.
Anti -ci re'an-aha -kon market-to go -PAST-PROBABLE 'I think he went to the market' b. He probably once was married. c. ?? He once probably was married.
Gam
As Cinque notes, this closely matches the required order for epistemic adverbials like probably and certain temporal adverbials like once, as indicated in (28b-c). I have suggested that the relative ordering is strictly determined by their interpretation; for instance, once may attach to an extensional object, while probably might attach to an intensional object. They both adjoin to TP nodes (or, say, IP; or VP, given certain assumptions about the interpretation of Tense; cf. §3.1 above), but if a lower node of TP is interpreted as intensional, then a higher one must be as well. Alternatively, (28c) is grammatical, but has the odd meaning 'At one time, it was the case that he probably was married' (the result of applying type-lowering to the intensional TP node under which probably is adjoined), which is odd because it asserts an indeterminacy about the past. On my account, then, why should the affixes on the verb in (28a) not be able to appear in the opposite order? Clearly, it is desirable, as on Cinque's account, that the ordering restrictions on adverbs and on morphemes follow from the same set of principles. This is true on my account, given a certain conception of the way the morphological component constructs complex words. Say, for example, that the tense affix is extensional in the sense that it attaches to a stem which projects a node with an extensional interpretation, and the epistemic affix is intensional in the same sense. This would mean that the inflected verb in the Garo example above projects an XP with an intensional interpretation, which would in turn mean that temporal adverbs cannot appear with that XP; however, the verb may also be associated with a trace in a V or T position, and temporal adverbs might attach to the VP or TP projected by that position. Thus the semantic-based account applies in exactly the same way to verbal affixes and phrasal adjuncts, but without a strict one-to-one matching of heads or adverbial phrases to functional categories and projections.
SUBJECT POSITIONS AND THE PLACEMENT OF ADVERBIALS
215
If the presence of a bound epistemic morpheme indicates the presence of an epistemic head in the syntax, then there will be a close correlation between the location of functional heads and the attachment site of adverbs, as in Cinque's theory; but in contrast to that theory, the fit will not be determined by feature checking, and will be as flexible as the semantics allow. I suggest that this is exactly the observed degree of flexibility, in contrast to the type of categorial rigidity ordinarily observed in morphological systems. The Mirror Principle (M. Baker 1985) maintains that morphological processes must occur in the same order as, for example, the grammatical function-changing processes which affect the lexical semantics of a stem (cf. note 1). One class of affixes that do not seem to be ordered by the Mirror Principle is the class of agreement affixes. Of course, the Mirror Principle makes clear (and, as far as I know, correct) predictions about the relative order of agreement affixes and grammatical function-changing morphemes. For instance, if a passive morpheme results in the promotion of a deep object to a surface subject, and a subject agreement morpheme shows agreement with the surface subject, then the passive morpheme must be closer to the stem than the subject agreement morpheme. But why is subject agreement so frequently farther from the stem than tense, as pointed out by Belletti? And why are there no 'deep' subject agreement morphemes? The distribution of agreement morphemes appears to be a much better candidate for an arbitrary syntactic fact, as opposed to a conceptually motivated fact. Even those accounts (Adger 1994) that ascribe a semantic contribution to Agr (specificity) don't make it clear why Agr morphology should be farther from the stem than tense or aspectual morphology. In §5,1 will propose (tentatively) a role for Agr that will make better sense of its high position in the clause. But my account will not derive this property from necessary conceptual considerations, so it retains a degree of syntactic peculiarity.
4. Subject positions in Germanic Before proceeding it is necessary to make clear what I am calling 'subject position' in Germanic, and to clear up some issues regarding the notion of 'topic.' In Germanic languages other than English, declarative main clauses are ordinarily V2, meaning that the finite verb appears in second position. Although subjects are common in first position (as in (29a) below), this position is not a 'subject position.' The exact characterization of the initial position in Germanic is not an easy matter. Traditionally it is described as a Theme position (in the Prague School sense), a Topic position (a closely related notion often characterized in terms of 'aboutness'; cf. Gundel 1974, Reinhart 1981), or as a position for 'background' information.9 However, none of these characterizations seems to capture initial elements like those in (29b-d), as has often been noted. (29) a.
Er kam deshalb nicht zur Arbeit. he came therefore not to work Therefore he didn't come to work'
German
216
SUBJECTS, EXPLETIVES, AND THE EPP b. c.
d.
Deshalb kam er nicht zur Arbeit. therefore came he not to work Ungliicklicherweise gibt es nur wenige Oasen. unfortunately gives it only few oases 'Unfortunately there are only few oases' Es gibt ungliicklicherweise nur wenige Oasen. it gives unfortunately only few oases
The traditional description of the initial element as a topic correctly captures the fact that it does not typically provide new information. However, more clearly needs to be said. In fact, new information is possible in first position, as in the Norwegian example below (from Faarlund 1992), in which the indefinite associate in an existential construction has been fronted to initial position (an element specifically excluded from topic- or themehood, e.g., in Firbas 1975). This example is legitimate as a response to a question like "How were the accommodations?," where 'features of accommodations' are salient, and therefore 'background' in some sense. (30)
Varmt og kaldt vann var detjo ikke warm and cold water was it after.all not 'After all, there wasn't hot and cold water'
Norwegian
If 'hot and cold water' were a topic, it could be expected either to have been commented on in the preceding discourse or to be commented on in the following discourse. This is not the case here. It seems that the initial position is typically used to indicate a change of emphasis or focus of attention in the discourse. Thus, when the initial element is argumental, it is better characterized as a SWITCH TOPIC or SHIFTED TOPIC, rather than as a CONTINUED TOPIC (cf., e.g., Aissen 1992, in which it is argued that this distinction is encoded in the grammar of Mayan, and King 1995 for Russian). For example, the following passage is licit in response to the question "Has the professor left?"") (31) a.
b.
c.
Ja, fortre veckor sen fick professorn besok Swedish yes for three weeks ago got the.professor visit av en herre som talade med stark brytning . . . of a gentleman as spoke with strong accent 'Yes, three weeks ago a man with a strong accent came to see the professor' De talade lange they spoke long 'They had a long talk' och sedan packade professorn och gav sig i vag and then packed the.professor and gave RFX in way med den frammande herrn . . . with the unknown gentleman 'Then the professor packed his luggage and left with the stranger'
Clearly, by the usual definitions of topic (cf., e.g., Givon 1990:900ff.), the professor is the topic of at least the first sentence. However, an expression referring to
SUBJECT POSITIONS AND THE PLACEMENT OF ADVERBIALS
217
the professor does not appear in initial position. This, I suggest, is because the speaker accepts the professor as having already been established as the topic by the preceding discourse, and he is therefore not a switch topic, but a continued topic. Thus the time adverbial 'three weeks ago' is initial in (3la), as it shifts attention to that time frame. Then the plural pronoun 'they' occupies initial position, as attention is now shifted to the two men. In the context established for the third sentence, then, either the professor or the two men could be treated as the continued topic, but the initial element is the temporal deictic element 'then,' shifting attention to a time immediately following the discussion referred to in the previous sentence. The expression 'the professor' appears in subject position, as a continued topic. I will not attempt to detail the nature of the initial position more fully, but will assume that the element in the sentence that is regarded by the speaker as most significantly redirecting the emphasis of the discourse is placed there. This is intended to include elements in contrastive focus, speaker-oriented adverbials, discourse connectives, scene-setting adverbials, and switch topics. Indefinites and elements expressing new information, including elements in (noncontrastive) focus, are ordinarily excluded. Unlike the situation with the switch-topic position in Mayan or Russian, the preverbal position in Germanic must be filled. This has the result that if there is no switch topic or other suitable element, a continued topic may appear in initial position (frequently the subject). The difference between a switch topic and a continued topic in initial position may be signaled by intonation (the continued topic is destressed; cf. Kristiansen 1996 for evidence from Danish). In the absence of any suitable element, an expletive may appear. Another difference between the initial position in Germanic and the switch topic position in Mayan or Russian is that if argumental, it must be coindexed with a gap in the IP (this could be modeled by saying that it is not a case position); this has the result that there are no 'multiple subject' constructions of the type known from Japanese and Russian. Consider the Russian sentence in (32a) (from Gundel 1974:185). (32)
a.
Ivan, ja ego ne ljublju. Ivan I him not like 'Ivan, I don't like him' b. * Ivan liker jeg han ikke. Ivan like I him not c. Ivan liker jeg ikke. Ivan like I not 'Ivan, I don't like' d. Ivan, han liker jeg ikke. Ivan him like I not 'Ivan, him I don't like'
Russian
Norwegian
Russian, like Japanese, allows DP shift topics that do not correspond to any gap in the sentence. As indicated in (32b), this is impossible in Norwegian. Ordinarily, an element in initial position corresponds to a gap in the sentence, as in (32c). Leftdislocation, as in (32d) is also possible, but then a coreferent pronoun appears in the initial position, and the left-dislocated element is outside the V2 clause
218
SUBJECTS, EXPLETIVES, AND THE EPP
(Icelandic has left-dislocation with a pronoun in situ; cf. Thrainsson 1979:59; but there, too, the left-dislocated element precedes what I have been calling the 'initial' position). Multiple subject constructions in languages like Japanese do not even require a coreferent pronoun to appear (cf., e.g., Kuno 1973 on Japanese sentences corresponding to "Fish, I like red snapper"; Gundel 1974:189 on Russian). Gundel (1974:69-71) claims that English left-dislocation has this property, but this seems to be true only when the 'dislocated' DP is supported by material such as as for). Thus, although simple SV or SVO sentences in English will typically be translated, out of context, into subject-initial sentences in other Germanic languages, the initial position is not primarily a subject position. Faarlund (1992) finds that in spoken Norwegian, 40% of sentences have nonsubject initial elements; similar statistics are demonstrated for German (Engel 1972). I follow the analysis which has developed out of the work of den Besten (1977), Platzack (1983), and so on, by which the landing site of the verb is C, essentially the same position as the location of the complementizer in a subordinate clause. I will refer to the complement of C as IP, at least when no more fine-grained level of detail is necessary.11 Inside IP there is what I call a 'subject position,' SpecIP. This, I will argue below, is a position for a potential continued topic like the internal topic of Aissen 1992 and King 1995.1 say 'potential' because if no other element occupies first position, the continued topic does; in other words, the subject position is occupied either by an overt element (generally the thematically highest argument in the sentence) or by the trace of a subject in the initial position. In (33a), the subject is in SpecCP, and its trace is assumed to occupy SpecIP, immediately after the verb. In (33b), SpecIP is occupied by the subject, han, immediately following the complementizer. In (33c), SpecIP is occupied by an expletive subject, immediately following the verb in C. (33) a.
b.
c.
Ole gikk inn i banken. Ole went in in the.bank 'Ole went into the bank' Jeg sa at han plutselig sprang ut igjen. / saw that he suddenly ran out again 'I saw that he suddenly ran back out' Etter han sprang det ei bevaepnet vakt. after him ran it a armed guard 'After him ran an armed guard'
Norwegian
In what follows, I will not distinguish between the subject position following a fronted verb and the subject position following a complementizer element, as there does not appear to be any structural distinction. On a split-infl analysis, it is possible that SpecIP corresponds to more than one position. I will discuss this possibility below.
SUBJECT POSITIONS AND THE PLACEMENT OF ADVERBIALS
219
4.1. German and the Mapping Hypothesis It has long been noted that the IP-internal syntax of German is sensitive to information structural considerations. Kratzer (1995) and Diesing (1992) have developed what is known as the Mapping Hypothesis, based on German data like that in (34) (from Kratzer, p. 153). The same basic facts obtain for Dutch (cf., for example, de Hoop 1992). (34) a.
b.
. . . weil sie immer Briefe aus Europa beantwortet German since she always letters from Europe answers '. . . since she is always engaged in answering letters from Europe' . . . weil sie Briefe aus Europa immer beantwortet since she letters from Europe always answers '. . . since she never leaves a letter from Europe unanswered'
According to the Mapping Hypothesis, there is a boundary, marked in (34) by the adverb immer 'always,' above which quantified elements receive a strong reading (in the sense of Milsark 1974), and below which they are caught under the scope of existential closure, receiving a weak or existential interpretation. Diesing identifies this boundary with the VP boundary, and states the Mapping Hypothesis as follows (1992:10): "Material from VP is mapped into the nuclear scope. Material from IP is mapped into a restrictive clause." However, it appears that simply being in the restriction or the nuclear scope of existential closure, or of syntactically present quantifiers, is inadequate to explain all the facts for German. Meinunger (1995:90) discusses the following example, where capitals indicate sentence accent. (35) a.
b.
. . . als er weider rauskam war auf einmal der HUND when he again out. came was of once the dog verschwunden. German disappeared '. . . when he came back out, all of a sudden the DOG had disappeared' . . . als er weider rauskam war der Hund auf einmal when he again out.came was the dog of once verSCHWUNden disappeared '. .. when he came back out, all of a sudden the dog had disapPEARed'
Here, the DP which appears variably above and below the VP-boundary is definite. The interpretation is in both cases Milsark-strong, but there is still an interpretational difference, which Meinunger argues has to do with topicality: 'the dog' is a topic in (35b), but not in (35a). This is consistent with Diesing's claim that DPs outside the VP boundary are 'presuppositional'; Meinunger proposes to account for this type of data by assuming that DPs to the left of boundary-marking adverbials occupy the specifiers of Agr phrases, and by linking Agr to topicality (using the model of topicality developed by Krifka). A similar approach is taken in Adger 1994, where Agr is linked to familiarity, in Heim's sense. Possibly, these ap-
220
SUBJECTS, EXPLETIVES, AND THE EPP
preaches obviate the need to locate existential closure at the VP level, since specific and generic indefinites may be topical or familiar in the relevant sense; see Cresti 1995 (and references there) on specific indefinites as 'topical.' 4.2. Scandinavian Bobaljik and Jonas (1996) take as a starting point Diesing's Mapping Hypothesis, expand it, and extend it to Scandinavian data. They argue that German and Icelandic have two subject positions, SpecAgrSP and SpecTP, but that Mainland Scandinavian and English have only one. Assuming that Diesing's boundary is at TP, rather than VP, they suggest that in sentences like (36b) (from their p. 196), where the subject follows an adverb, it occupies SpecTP. (36) a.
b.
I gaer klarudu bessar mys sennilega ostinn. yesterday finished these mice probably the. cheese 'Yesterday these mice probably finished the cheese' I gaer klarudu sennilega margar mys ostinn. yesterday finished probably many mice the. cheese 'Yesterday many mice probably finished the cheese'
Icelandic
Holmberg (1993) points out that adverb placement would seem to indicate two positions in Swedish as well, as in his example in (37) (from his p. 32), though he suggests that the lower position is reserved for subjects in focus (cf. his p. 38); hence (37b), without stress-marked focus, is marginal.12 (37) a.
Har nagon student mqjligen last boken? has any student possibly read the.book 'Has any student possibly read the book?' b. ?? Har mqjligen nagon student last boken? has possibly any student read the.book c. Har mqjligen NAGON STUDENT last boken? has possibly any student read the.book 'Has ANY STUDENT possibly read the book?'
Swedish
Many, though not all, speakers of Swedish concur; but the focus effect appears to be rather weak, in that the contrast needed to make (37c) good is not hard to achieve. In addition, this may vary with register; examples with nonfocused subjects are easy to find in texts, as in the examples in (38) (context makes it clear in each case that no contrast is intended), including older texts like the one whence (38c) is taken. (38) a.
b.
c.
Dar skall nu vajorna kalva. Swedish there shall now the.reindeer.does calve 'The reindeer does are going to calve there now' I Mars anfoll darpa Danmark Holstein-Gottorp. in March attacked thereon Denmark Holstein-Gottorp 'After that, in March, Denmark attacked Holstein-Gottorp' Har uppe aro namligen natterna mycket langa, kolden stra'ng . . . here up are namely the.nights very long the. cold severe 'Because up here the nights are very long, the cold severe . . .'
SUBJECT POSITIONS AND THE PLACEMENT OF ADVERBIALS
221
Norwegian has already been shown to allow adverb-subject order (in §2.2 above); an additional pair of examples is given here. (39) a.
b.
Sa provoserer Salomes mannshunger fortsatt dagens publikum. so provokes Salome's man.hunger still the.day's audience 'Then Salome's hunger for men still provokes today's audiences' Sa provoserer fortsatt Salomes mannshunger dagens publikum. so provokes still Salome's man.hunger the.day's audience
Some speakers report a sensitivity to focus, in the direction of the Swedish pattern. However, the general pattern for Norwegian is the one represented in (39); no intonationally signaled focus is necessary for adverb-subject order. On the other hand, subject-adverb order is strongly preferred with weak pronominal subjects, as indicated in (40) (this is true of Icelandic as well). (40) a.
Sa provoserer den fortsatt dagens publikum. so provokes it still the.day's audience 'Then it still provokes today's audiences' b. ?? Sa provoserer fortsatt den dagens publikum. so provokes still it the.day's audience c. Sa provoserer fortsatt DEN dagens publikum. so provokes still that the.day's audience 'Then THAT still provokes today's audiences'
Norwegian
In Danish, as Holmberg notes, subjects obligatorily precede adverbials in IP, regardless of focus. (41) a.
Naeste eftermiddag laa stenene endnu ur0rte. next afternoon lay the.stones still unmoved The next afternoon the stones still lay unmoved' b. * Naeste eftermiddag laa endnu stenene ur0rte. next afternoon lay still the.stones unmoved c. * Naeste eftermiddag laa endnu STENENE ur0rte. next afternoon lay still the.stones unmoved
Danish
However, adverb-subject order appears to be a recent casualty: Diderichsen (1946:191) gives (41b), without describing it as marked or archaic. In fact, in the text, when he introduces the examples in (42b-d), he uses the sentence in (42a). In that example, there is contrast with the preceding text, but the contrasted element is the adverbial expression ogsd i Hovedscetninger 'even in main clauses,' forming a focus set with 'subordinate clauses' (on his p. 189); the subject, Adverbialet 'the adverbial', is not in focus (in these examples I have preserved his capitalization conventions). (42) a.
b.
Dog kan ogsa i Hovedsaetninger Adverbialet staa foran Subjektet. . . yet can also in main.clauses the.adverbial stand before the.subject 'Yet the adverbial can precede the subject, even in main clauses . . .'
Saa maatte jo
/nok
/vist
Kongen b0je sig.
so must you.know/I.suppose/it.appears the.king bend rfx 'So of course the king had to yield'
222
SUBJECTS, EXPLETIVES, AND THE EPP c.
d.
Saa var pludselig D0ren sprunget op so was suddenly the.door burst open 'Then the door had suddenly burst open' I K0bstaederne har sasdvanligvis hvert Hus en lille Have in the.market.towns has usually every house a little yard 'In the market towns every house usually has a little yard'
However, none of these examples is fully acceptable to speakers I have consulted. It is worth noting at this point that English appears to pattern with Danish, if subject-aux inversion structures are taken as indicative. (43) a. Have any students {probably/possibly/already} read the book? b. * Have {probably/possibly/already} any students read the book? c. * Have {probably/possibly/already} ANY STUDENTS read the book? d. I think that {probably/possibly/already} some students have read the book. As indicated in (43b-c), adverb-subject order is impossible after a verb in C, with or without focus, as in Danish. However, English diverges from the pattern with respect to subordinate clauses, where adverbs quite freely precede the subject. The contrast, however, may be illusory. In general, English fairly freely allows main clause phenomena within embedded clauses (e.g., neg-inversion). Thus, (43d) may simply reflect the possibility of CP-recursion in English (however that should be analyzed). The similarity to Danish can be seen by comparing (43d) to (44). (44)
a.
* jeg tror at muligvis nogen student har laest bogen. I think that possibly some student has read the. book b. Jeg tror at muligvis har nogen student laest bogen. / think that possibly has some student read the. book 'I think that possibly some student has read the book'
Danish
Example (44a) is impossible in Danish (though it is perfect in Norwegian, and possible in Swedish with focus on 'some student'). Example (44b) is possible, as an instance of CP-recursion (cf. Vikner 1995), with the result that V2 is forced in the embedded clause. Taking English to be like Danish, I have shown three patterns: Icelandic and Norwegian, which allow both orders freely; Swedish, which has adverb-subject order only with focus; and Danish, which never allows adverb-subject order. Swedish and Danish appear to have recently been more like Norwegian and Icelandic. There is a difference, however, between Icelandic and Norwegian. The difference has to do with restrictions on what sorts of subjects may follow adverbial elements. I discuss this in the next section. 4.3. DP Readings As Bobaljik and Jonas (1996) note, definites are not generally licit following adverbs in Icelandic, as indicated in (45) below (cf. Sigurdsson 1990:50).
SUBJECT POSITIONS AND THE PLACEMENT OF ADVERBIALS
223
(45) a.
Nunahafa bofarnir liklega stolid smjorinu. Icelandic now have the.gangsters probably stolen the.butter 'Now probably the gangsters have stolen the butter' b. ?? Nuna hafa liklega bofarnir stolid smjorinu. now have probably the.gangsters stolen the.butter c. Nunahafa liklega BOFARNIR stolid smjorinu. now have probably the.gangsters stolen the.butter 'Now perhaps THE GANGSTERS have stolen the butter'
Example (45c) is acceptable with contrastive focus (indicated intonationally with an accent peak) on the subject, but not without. This is the same pattern as observed above in Swedish and Norwegian, except for the nature of the DP: in Swedish, focus is necessary to license even quantified subjects following adverbials, whereas quantified subjects are more generally licit in this configuration in Icelandic. In Norwegian, only pronouns require this additional stress. Bobaljik and Jonas take the Icelandic facts to support their analysis in terms of the Mapping Hypothesis, assuming that the low position (SpecTP, on their analysis) corresponds to the weak readings possible in Diesing's German sentences for indefinites below the boundary adverbs. However, there are some problems with this correlation, as Vangsnes (this volume) points out. I will return to these momentarily. Jonas (1993) shows that there are two different lects of Faroese: the one she calls 'Faroese+' is like Icelandic, in that adverb can precede quantified DPs, as in (47b), but not definites, as in (46b); and the one she calls 'Faroese-' is like Norwegian, in that all four of the sentences in (46)-(47) are acceptable. (46) a.
I dag tekur Jogvan kanska supan i skulan Faroese today takes John maybe soup in school Today John might take soup to school' b. * I dag tekur kanska Jogvan supan i skulan (* in Far+, ok in Far-) today takes maybe John soup in school
(47) a.
b.
I dag taka nakrar gentur kanska supan i skulan today take some girls maybe soup in school 'Today some girls might take soup to school' I dag taka kanska nakrar gentur supan i skulan today take maybe some girls soup in school
Faroese
Thus, for Jonas and for Bobaljik and Jonas, Icelandic, German, and Faroese+ have two subject positions (SpecAgrP and SpecTP), which can be distinguished syntactically according to the position of sentential adverbials, which attach to TP, above the lower subject position.13 The same two positions can be distinguished semantically, according to the interpretive possibilities: subjects in SpecTP can only be Milsark-weak, following Diesing (modulo the location of the boundary), while subjects in SpecAgrP can only be Milsark-strong. Mainland Scandinavian and English have only one subject position, SpecAgrP or SpecTP. According to Jonas 1993, to the extent that adverbs can appear preceding subjects in MS, this is because in those languages, adverbs can be attached to AgrP, unlike the situation in Icelandic, Faroese+, and German. The pattern for Icelandic is represented schematically in (48).
222 SUBJECTS, EXPLETIVES, AND THE EPP (48)
a. [AgrP bessar mys [TP sennilega [-jp [vp b. [AgrP [TP sennilega [TP margar mys [VP c. * [AgrP sennilega [AgrP pessar mys [Tp [vp probably these mice probably many mice
The structure in (48a) corresponds to (36a), (48b) to (36b). The order adverb-definite DP is impossible, because definite DPs cannot remain in SpecTP, and, as shown in (48c), adverbs cannot be adjoined to AgrP. The pattern for Norwegian, on this account, is demonstrated in (49). (49) a. [AgrP musene [TP sannsynligvis [TP [VP b. * [AgrP [TP sannsynligvis [jp mange mus [VP c. [AgrP sannsynligvis [AgrP musene [TP [VP probably the.mice probably many mice Here, no DP can remain in SpecTP, so (49b) is impossible; the subject-adverb order represented in (39a) above has the structure in (49a), like its Icelandic counterpart; but the adverb-subject order in (39b) has the structure in (49c). However, there are some problems with this account. First, it is unclear why adverbs should be allowed to attach to AgrP in Norwegian, but not in Danish, Icelandic, or German. Second, it is not clear that the two positions are adequate to explain the full range of word orders; recall from §2.2 and §2.3 above that there are many possibilities both for subject and for objects above, below, and between adverbials. Icelandic, for example, can have sentential adverbs both before and after a quantified subject (cf., e.g., (18) above); the second of the two, on this account, would have to be attached to some node lower than TP. Similarly, consider the Dutch examples in (50), from de Hoop 1992:182. (50) a.
. . . dat er waarschijnlijk enkele taalkundigen lui zijn Dutch that there probably some linguists lazy are '. . . that some linguists are probably lazy' b. . . . dat er enkele taalkundigen waarschijnlijk lui zijn that there some linguists probably lazy are Dutch, which has Transitive Expletive Constructions, should be like Icelandic, on Bobaljik and Jonas's account. In the examples here, there is an expletive after the complementizer, presumably in SpecAgrP. In (50a), then, the subject must be in SpecTP. However, in (50b), the subject also appears preceding the adverb, suggesting either that there is yet another subject position or else that adverbs can be attached lower than TP in Dutch as well. A third problem for the account sketched in (48)-(49) is that the interpretive possibilities for the different positions do not actually correlate with Milsarkian strength (nor, it seems, with quantificational restrictions). Note, for example, that the reading of the subject in (50a) above is the same as in (50b), that is, strong (partitive). In the same vein, Vangsnes (this volume; see also Vangsnes 1995) shows that strongly quantified elements, including universally quantified DPs, are possible in the lower position in Icelandic. In fact, the interpretive possibilities appear to vary significantly across languages, a problem for Diesing's account, and hence for Bobaljik and Jonas's.
SUBJECT POSITIONS AND THE PLACEMENT OF ADVERBIALS
225
Recall from §2.2 above that Nilsen (1997, 1998) applies a Cinque-style analysis to Norwegian adverb placement, finding something on the order of twenty different possible relative sites for the subject (e.g., above 'still' but below 'usually,' above 'usually' but below 'not,' above 'not' but below 'maybe,' and so on). He briefly addresses the issue of subject interpretation, suggesting that the boundary observed by Diesing may be located around the adverb kanskje 'maybe.' He gives the sentences in (51) (his 1997:23) to support this finding. (51) a.
b.
R0ykeforbudet br0t en student kanskje allerede i gar. Nor. the.smoking.ban broke a student maybe already yesterday 'A (specific) student might have violated the smoking ban as early as yesterday' R0ykeforbudet br0t kanskje en student allerede i gar. the.smoking.ban broke maybe a student already yesterday 'Some student (or other) may have violated the smoking ban as early as yesterday'
The preferred readings are as indicated in the glosses. Compare a much higher adverbial, sannsynligvis 'probably' (also from Nilsen 1997:23). (52) a.
b.
R0ykeforbudet br0t en student sannsynligvis allerede i gar. Nor. the.smoking.ban broke a student probably already yesterday 'A (specific) student probably violated the smoking ban as early as yesterday' R0ykeforbudet br0t sannsynligvis en student allerede i gar. the.smoking.ban broke probably a student already yesterday 'A student probably violated the smoking ban as early as yesterday' (AMBIGUOUS)
Here, according to Nilsen, the preferred reading is the strong one for the subject in (52a), whereas (52b) is ambiguous. This is expected if sannsynligvis marks a higher position than Diesing's boundary: a subject higher up must be outside the nuclear scope, while a subject lower down could be either outside it or inside it. Finally, consider one more pair from Nilsen 1997. (53) a.
b.
R0ykeforbudet br0t en student vanligvis uansett. Norwegian the.smoking.ban broke a student usually anyway 'A student usually violated the smoking ban anyway' (AMBIGUOUS) R0ykeforbudet br0t vanligvis en student uansett. the.smoking.ban broke usually a student anyway 'Usually some student or other violated the smoking ban anyway'
The adverb vanligvis 'usually' is lower down than kanskje. According to Nilsen, (52b) has only the weak reading, while (52a) is ambiguous, as expected if there are several possible positions above vanligvis, including positions inside and outside Diesing's scopal boundary. At first blush, these results strongly support a model in which Norwegian has several available subject positions. However, the results are not quite as neat as (51)-(53) suggest. For one thing, the readings given there are preferred readings,
226
SUBJECTS, EXPLETIVES, AND THE EPP
rather than being the only readings possible. The readings in (52a) and (53b) can easily be reversed, as shown in (54). (54) a.
b.
R0ykeforbudet br0t en eller annen student sannsynligvis Nor. the. smoking, ban broke a or other student probably allerede i gar. already yesterday 'Some student or other probably violated the smoking ban as early as yesterday' R0ykeforbudet br0t vanligvis en viss student uansett. the.smoking.ban broke usually a certain student anyway 'One particular student usually violated the smoking ban anyway'
If the strong readings for indefinites involve topicality, as argued in Cresti 1995, then what (51)-(53) show is that an indefinite preceding an adverbial has a tendency to be interpreted as topical (and that this tendency is stronger with certain adverbials than with others). This is overridden by using an expression like en eller annen student 'some student or other,' and is not critically associated with the relative position vis-a-vis adverbials. A similar independence of noun phrase interpretation from relative position can be demonstrated for Icelandic. (55) a.
b.
Pess vegna ogra ennba morg leikrit ahorfendum nutimans. this account provoke still many plays audiences today's 'For this reason, many plays still provoke today's audiences' M aetla margir malvisindamenn vonandi ad koma. then intend many linguists hopefully to come Then many linguists hopefully plan to come'
Recall from §2.2 that 'still' is a very low adverb in terms of relative order (following Cinque 1999 and Nilsen 1998; but cf. note 4). The quantified DP in (55a) is easily, even preferably, read as weak, but can also be taken to have a strong interpretation: 'Many plays are such that they still provoke today's audiences.' 'Hopefully,' on the other hand, is a very high adverb, relatively speaking, in that it must precede most other adverbs if they co-occur, including 'probably' and 'maybe.' Yet (55b), for at least some speakers, can be read with the subject taking narrow scope, that is, 'I hope that many linguists will come (but I don't care which ones).' On close scrutiny, it seems, the correlation between position and strong/weak readings breaks down completely (cf. Nilsen 1997 for some additional mismatches). However, there are still some clear tendencies to contend with, and the complex pattern of cross-linguistic variation demonstrated in §4.2. In §5,1 will provide an account for this variation that treats the scopal tendencies noted, and which correctly allows for the reversal of those tendencies in the presence of contributing factors. 4.4. Summary of the data In the previous two subsections, a complex pattern of variation was presented. Here I quickly summarize the pattern for the various languages, and recast the findings of §§4.2-4.3 in terms of topicality, specifically continued topichood.
SUBJECT POSITIONS AND THE PLACEMENT OF ADVERBIALS
227
German shows a sharp tendency for strong readings when DPs precede sentential adverbials; these strong readings manifest themselves variously as wide scope or presuppositional readings for quantified elements, and specific or generic readings for indefinites including bare plurals. DPs following sentential adverbials show a strong tendency for weak readings, including narrow scope for quantified elements and existential readings for indefinites. In addition, a topicality effect is observed for definites, as discussed by Meinunger; his example (35) from §4.1 is repeated as (56) below. (56) a.
b.
. . . als er weider rauskam war auf einmal der Hund verschwunden when he again out.came was of once the dog disappeared 'when he came back out, all of a sudden the dog had disappeared' . . . als er weider rauskam war der Hund auf einmal verschwunden when he again outcome was the dog of once disappeared 'when he came back out, all of a sudden the dog had disappeared'
As noted above, Meinunger argues that the difference in interpretation has to do with topicality: 'the dog' is topical in (56b) but not in (56a). According to Meinunger, the other interpretational effects noted by Diesing also follow from topicality. For German, objects also show this effect. In the other languages discussed here, objects (almost) never cross subjects within IP (this is also generally true of Dutch, which otherwise shows the same sort of interpretational effects as German). Norwegian was demonstrated above to freely allow subjects both above and below adverbials. Furthermore, Norwegian generally shows the same kinds of effects as German: subjects preceding sentential adverbials preferably show 'strong' readings, and subjects following sentential adverbials generally have a 'weak' reading. Norwegian also patterns with German when it comes to definites. (57) a. b.
Da nan kom tilbake ut when he came back out Da han kom tilbake ut, when he came back out
hadde plutselig hunden forsvunnet. had suddenly the.dog disappeared hadde hunden plutselig forsvunnet. had the.dog suddenly disappeared
The choice between the two appears to be a subjective one, based on the speaker's perception of the structure of the discourse. I argued above that initial position was appropriate for switch topics. Now I suggest that the IP-internal subject preceding a sentential adverbial is most naturally interpreted as a continued topic. Consider another example. (58) a.
b.
For eksempel kan vann plutselig trenge seg inn i treverket. Nor. for example can water suddenly force RFXin in the.woodwork 'For example, water could suddenly force its way into the woodwork' For eksempel kan plutselig vann trenge seg inn i treverket. for example can suddenly water force RFX in in the.woodwork 'For example, water could suddenly force its way into the woodwork'
228
SUBJECTS, EXPLETIVES, AND THE EPP
Both examples are perfectly acceptable, though indefinite subjects in actual discourse are rare (Kulbrandstad 1978 examined 200 pages of transcribed Norwegian speech, finding no weak indefinite subjects; Heltoft and Jakobsen 1996:200 observe for Danish that indefinite subjects are unusual and largely confined to fiction writing). The subject in (58a) cannot be interpreted as generic or specific, which is surprising on Diesing's account. Instead, the subject in (58a) is weak, but topical; (58a) is felicitous, for example, as the answer to the question "Could water damage the house?" (58b) is regarded as a slightly odd way to answer that question, because the subject is not topical. (58b) is, on the other hand, perfectly natural as an answer to the question "What could go wrong now?" (cf. Gundel 1974:60ff. on indefinite topics). The examples with indefinite and quantified subjects given in §4.3 succumb to a similar analysis. The question is not whether the subject is in the restriction of a generic or specific quantifier, but whether it is considered by the speaker to continue the topic of the discourse. Pronouns, of course, are most naturally seen in this way, so it is natural that unstressed pronouns should be odd following sentential adverbials. Swedish, it will be recalled, showed a sensitivity to focus. Here, the default is to place even nontopical DPs before sentential adverbials, unless they are clearly part of the focus of the sentence. Icelandic represented a different pattern. Quantified DPs behaved like Norwegian DPs, being apparently sensitive to the strong/weak parameter, but upon closer inspection, this broke down. I now suggest that the position for quantified DPs in Icelandic with respect to adverbials is also sensitive to topicality: weakly interpreted DPs generally make bad topics, and are therefore better following sentential adverbials. Quantified adverbials that are continued topics are preferably placed preceding sentential adverbials. As for definites, they followed the Swedish pattern, preceding adverbials unless contrastively focused. I have not undertaken a systematic investigation of Faroese. However, based on Jonas 1993, it appears that Faroese+ can be treated like Icelandic, as suggested there, while Faroese- might be like Norwegian. Finally, there is the Danish pattern, in which all DPs precede all sentential adverbials. English appears to pattern with Danish, at least with respect to subjectaux inversion structures, although in noninversion structures there is apparently quite free adjunction of adverbs before subjects. A brief historical note might be in place here. In Old Norwegian and Middle Norwegian legal documents, basing my observations on Christoffersen 1993 and M0rck 1999 respectively, full noun phrases almost universally follow sentential adverbials within IP, while pronominal subjects almost always precede.14 Some representative examples are given in (59), from Christoffersen. (59) a.
. . . skal iafnan sa fyr arf taka shall always that man inheritance take '. . . that man shall always receive the inheritance'
Old Norwegian
SUBJECT POSITIONS AND THE PLACEMENT OF ADVERBIALS b.
229
. . . oc take hann fyrst skulld sina upp sua sem domer doemde and takes he first debt his up such as judge deemed 'and first he settles his debt in the way that the judge determined'
At first, this looks like yet another pattern. However, examination of Modern Norwegian texts yields a similar pattern; full noun phrases are most commonly used when they are not anaphoric to the previous discourse and are therefore not continued topics. Thus it seems possible that Old Norwegian was not very different from Modern Norwegian. Old Icelandic sagas show a much higher incidence of subject-adverb order, with a variety of subject types (but pronominal subjects almost always precede adverbials). A few examples are given in (60); (60a-b) show that the same adverbial, jafnan 'always,' can appear both before and after a subject, and (60c-d) show the same for pa 'then.' (60c) shows that a definite subject can follow an adverb, in contrast to modern Icelandic (the context, in Bardarsaga Sncefellsdss, makes clear that contrast is not intended), and (60d) shows that a weak indefinite can precede it. (60) a.
b.
c.
d.
Hafa og foreldrar pinir jafnan i stormaelum stadid. Old Ice. have also parents your always in big.dealings stood 'Your parents have also always been involved in major dealings' . . . og spurdust jafnan ospektir nordan fra perm Ospaki. and was.heard always trouble north from them Ospak '. . . and from the north there was always news of trouble regarding Ospak and the people associated with him' Dundi p£ blodid um hann allan poured then the.blood over him all 'Then the blood poured all over him' Sau menn pa bloQrefjar nokkurar i utidyrum. saw people then blood.traces some in the.outer.door 'Then people noticed some blood traces in the doorway'
The pattern can be fairly closely duplicated in Modern Norwegian. Recall from (58) above that existentially interpreted indefinites can precede adverbials within IP if they are sufficiently 'topical.' Compare (60d): here 'people,' though indefinite, has a high degree of topicality. The previous sentences are 'The head of the household had disappeared. This, people thought was strange.' Thus it seems reasonable to assume that 'people' in (60d) is a continued topic. It is possible that the high incidence of full DP-adverb order represents a difference between Old Icelandic and Old Norwegian, but it seems more likely that it has to do with the nature of the texts, where the legal texts are dry and formulaic while the sagas employ a wider range of stylistic devices. Many of the examples are verbinitial (cf. Sigurdsson 1990 on VI in Icelandic), which may be significant. To conclude this section, there appear to be four major patterns: Norwegian and German, which can be analyzed fairly straightforwardly as sensitive to the notion of continued topic; Swedish, which is sensitive to focus; Icelandic, which is sensitive to definiteness; and Danish, which is rigidly subject-adverb.
230
SUBJECTS, EXPLETIVES, AND THE EPP
5. Analysis As a first pass, imagine an analysis based on Bobaljik and Jonas's analysis of Icelandic, utilizing the specifiers of Agr and T. First, assume that sentential adverbs attach to TP, recursively; semantic factors will determine relative ordering, as discussed in §3. Second, assume that subjects in Germanic generally move to SpecTP (this assumption will be revised below), for concreteness motivated by a strong D feature in T. Third, SpecAgrP is a position for a continued topic, as discussed in §4. This can be formalized by positing an optional strong +Topic feature in Agr (as in Meinunger 1995), matched by a +Topic feature placed on a continued topic. I will furthermore assume that the assignment of this +Topic feature to DPs can be grammaticized, in the sense that a given language may assign the +Topic feature automatically to DPs which have certain morphosyntactic characteristics, even if they are not continued topics in the discourse. In German and Norwegian, on this model, actual continued topics receive the +Topic feature, and they appear in SpecAgrP. In addition, weak pronouns are lexically marked +Topic, whether they are informationally continued topics or not. This ensures that subject pronouns always precede adverbs within IP (I return to object pronouns below). For Icelandic, definite D is marked +Topic, and definite DP subjects must therefore move to SpecAgrP. Quantified subjects remain in SpecTP unless they are in fact continued topics, just as in German and Norwegian. Nonsubjects must generally be prevented from moving to SpecAgrP. For the time being, assume that this is due to Relativized Minimality: a nonsubject cannot cross the subject in SpecTP (recall that strong D in T attracted the subject to that position). +Topic features on nonsubjects can be checked in Agr at LF; only one DP (the subject) must move overtly to satisfy the strong feature in Agr. Note that if the strong feature in Agr is optional, then it will appear only when needed to check +Topic on some DP. I return to the matter of objects in §5.3 below. In Swedish, nonfocused DPs are assigned +Topic, representing a greater degree of grammaticization than observed in Icelandic. Finally, Danish and English behave as though all subjects were marked +Topic, in that they all move to SpecAgrP. This is equivalent to there being a strong +D feature in Agr, and represents the complete grammaticization of the +Topic feature, which no longer serves to identify topics at all in those languages. The upshot is that what is being moved to SpecAgrP in the different languages is in fact a potential topic, where each language is sensitive to a different overt indicator of potential topicality. German and Norwegian, I contend, are the most discourse-configurational, in that full DPs which are not continued topics are not ordinarily forced to move to SpecAgrP. Icelandic is the next most discourse configurational, followed by Swedish. This account has the advantage that it makes relatively precise the distribution of the +Topic feature, compared with other accounts postulating such a feature. It also makes sense of the fact, observed in §3.5 above, that Agr is quite high in the clause (e.g., subject agreement inflection tends to be outside other verbal inflection); if Agr is linked to topicality, then it would be expected to be outside other functional projections which have to do with proposition-internal semantics, as topicality is in a sense meta-propositional.
SUBJECT POSITIONS AND THE PLACEMENT OF ADVERBIALS
231
The analysis does not sit well with the parameter that Bobaljik and Jonas argue for, by which Mainland Scandinavian does not have a SpecTP position at all. Their account is intended to unify a range of properties, including the possibility of (i) adverbs preceding subjects within IP, (ii) Object Shift for full DPs, (iii) Transitive Expletive Constructions,' and (iv) separate tense and agreement morphemes. These factors generally distinguish German and Icelandic on the one hand from Mainland Scandinavian and English on the other. However, as we have seen, Norwegian does have property (i). In addition, Nilsen (1997, 1998) has argued persuasively that Norwegian does in fact have property (ii) (as noted in §2 above). However, it does not have properties (iii-iv). Similarly, Dutch has (iii) but not (iv) (cf. Maling and Zaenen 1978). Such variation represents a compromise of the systematic correlation of features for which Bobaljik and Jonas's account was designed, strengthening the possibility that Mainland Scandinavian might have SpecTP after all. In subsection §5.11 refine the analysis, noting some remaining problems, and in §5.2 I propose a more dramatic refinement, one which does away with the Agr node altogether, while retaining the predictive power of the approach outlined here. In §5.3 I show how the account extends to object shift phenomena. 5.7. Refinements The analysis just sketched requires some immediate refinements. First, it has been necessary to stipulate that adverbs cannot adjoin to AgrP. Otherwise, adverb-subject orders would be possible in all cases, with all kinds of subjects. However, I believe that this problem has a well-motivated solution. I have argued elsewhere (Svenonius 1994) that IP is not entity-denoting; it is an open function. This can be modeled formally by saying that TP denotes a proposition, sort p (of type e), and that CP denotes an information unit, which is sort i, not sort p: an information unit is anchored to a context; C provides an anchoring, represented by w (for possible world). Agr is type
232
SUBJECTS, EXPLETIVES, AND THE EPP
Another potential problem regards the nature of the EPP in German and Icelandic. It is well known that certain expletive subjects do not appear within IP in those languages (cf. Platzack 1985, Vangsnes, this volume). Haider (1986, inter alia) argues that there is no EPP in German, and Bennis (1986) argues the same, at length, for Dutch. If these arguments are accepted, then it is a simple matter to revise the account above, by eliminating the strong D feature in T. Then subjects that are not attracted by one of the other strong features will raise to SpecTP only if they need Case, as on Dennis's account. Another refinement concerns the location of the subject with respect to adverbials. Recall from §2 that the subject can appear between adverbials, as demonstrated by Cinque and Nilsen. This means that it won't do to simply attach all adverbials to TP, below SpecAgrP and above SpecTP. But if they are allowed to attach to AgrP, then adverb-subject orders will be overgenerated. On the other hand, if they are allowed to attach to VP (or AgrOP, or some other lower node), then the conditions on adverb placement are substantially weakened (e.g., the account would predict that adverbs that scope over tense could not follow a subject which was in turn preceded by another adverb; this seems wrong), and subject-adverb orders will be overgenerated. All the examples with subjects preceding adverbials should easily admit nontopic readings, since the subject could be in SpecTP as well as in SpecAgrP. This is not the case; in (61a), repeated from (5la) above, and in (61b), the preferred reading of the subject is specific. (61) a.
b.
R0ykeforbudet br0t en student kanskje allerede i gar. Nor. the.smoking.ban broke a student maybe already yesterday 'A (specific) student might have violated the smoking ban as early as yesterday' Del hadde uheldigvis en student kanskje allerede br0t i gar. that had unfortunately a student maybe already broken yesterday 'A (specific) student unfortunately had maybe already violated it yesterday'
In fact, any example with adverbs both before and after the subject, such as (61b), indicates that adverbs must be attachable lower down than to TP, given the assumptions outlined above. However, in § 1.4 I adopted the position that there is no structural distinction between specifiers and adjuncts; this means that both the subject and the adverbs in both examples in (61) can be adjoined to TP; the adverb is merged first with TP, and then the subject is raised and adjoined, outside the adverb (this is basically equivalent to adjoining the adverb to T', in a model with a specifier-adjunct distinction). This is compatible with Chomsky's (1995: ch. 4) proposal of featural cyclicity (that a strong feature must be checked before the phrase in which it appears is embedded; cf. his p. 233), if adverbs are adjoined to projections of T, in contrast to the Alexiadou/Cinque model. This is not to say that no adverbs can be attached to lower nodes; in fact, in §3 above, I proposed exactly this, for certain adverbs; for example, if Tense is irrelevant to the interpretation of an adverb like stupidly, I suggested, then it can adjoin above or below Tense.
SUBJECT POSITIONS AND THE PLACEMENT OF ADVERBIALS
233
But the possibility of adjunction to TP below the position of the subject (the functional equivalent of adjunction to T) opens the way for an account without Agr, which I detail in §5.2. 5.2. Doing without Agr In this subsection I present an alternative treatment of subject-adverb order in Germanic that does not utilize two specifier positions at all. It requires some additional machinery, but avoids some of the difficulties discussed in §5.1, and simplifies others. The basic idea is that rather than reflecting two different subject positions, relative subject-adverb order simply shows the order in which the different elements have adjoined to IP. In other words, I do not make use of AgrP at all. This is consistent with the position of, for example, latridou 1990, C. L. Baker 1991, and Chomsky 1995: ch. 4, in which the motivation for Agr as an independent node in the clause is challenged. Recall that the evidence for the two positions, SpecTP and SpecAgrP, relies almost solely on adverbial data. But once adverbs are allowed to adjoin either before or after a subject has raised to a functional projection (the equivalent, in earlier versions of X'-theory, of allowing an adverb to attach to XP or to X1), the interpretation of that evidence changes substantially. I have now suggested that there is one derivation leading to adverb-subject order (i.e., one in which a subject raises to check features in T, and then an adverb is adjoined to TP), but two derivations leading to subject-adverb order: one in which the adverb is first adjoined to TP, and then the subject is raised to check features in T, adjoining outside that adverb, and another in which the subject raises to SpecAgrP. I am now suggesting that this second option can be dropped. The difference between a subject in SpecAgrP and a subject in SpecTP was taken above to be a question of whether that subject had a +Topic feature, including, for example, all definite subjects in Icelandic and all pronominal subjects in Norwegian, but also subjects in both of those languages that actually serve as continued topics. There was a +Topic feature in Agr which checked that +Topic feature, and was deleted (in accordance with the checking theory of Chomsky 1995). In addition, there was a strong D feature in T. Now I am assuming just one functional head, Infl. Thus the +Topic feature and the D feature must both be located there. Since D is strong, the +Topic feature will be checked whenever the subject is topical; thus the +Topic feature need not be strong. This eliminates the need to make the Topic feature in Agr optional. Adverb-subject order results when a subject raises to check D but does not check +Topic, because it is not topical. An adverb is merged afterward—note that the D feature is checked, but the +Topic feature is still unchecked at the point where the adverb attaches. Subject-adverb order, on the other hand, is the result of an adverb merging first with IP, with unchecked D and +Topic features, followed by raising of a topical subject, checking both of the features. What is clearly not allowed is adverb-subject order when the subject is topical in the relevant sense (i.e., pronominal in Norwegian, definite in Icelandic, unfocused in Swedish, at all in Danish). In this model, the illicit derivation would be
234
SUBJECTS, EXPLETIVES, AND THE EPP
one in which a topical subject adjoined to IP, checking both D and +Topic, and then an adverb merged with that IP. The generalization can be stated thus: (62)
An adverb may not attach to IP with a checked +Topic feature.
Since adverbs relatively freely follow the subject in all the languages discussed, adverb attachment to IP with an unchecked +Topic feature is fairly free. In Danish, it will be recalled, as in English, adverbs never precede the subject within IP. The simplest assumption to capture this pattern is that all DPs check +Topic in Danish, even focused ones. In other words, the +Topic feature and the categorial D feature of DPs are not distinct in Danish. The account can be cast in two slightly different ways. It can be thought of as a derivational account, in which case what matters for adverb attachment is whether the +Topic feature in Infl has been checked at the point in the derivation where the adverb is to be merged with IP. Subsequent checking of the +Topic feature in Infl, by a topical subject adjoined higher up, can then affect the entire IP, without consequence for the attached adverb. Alternatively, the account can be thought of representationally. In that case it is necessary to imagine that an IP node to which an adverb is attached still has a +Topic feature, even when the IP node higher up, to which the subject is adjoined, does not, due to checking and deletion. In other words, different nodes created by adjunction must have different featural specifications, as expected, given the discussion in §1.4 above. So far, (62) is a descriptive generalization, without any theoretical motivation. There is, however, a way to connect it with a larger framework of sentence interpretation. Recall from §5.1 above the rationalization for the failure of adverbs to attach to AgrP. The problem, I suggested, was that AgrP denotes an open function, and was therefore not the right type for adverbial modification. Specifically, I suggested that AgrP lacked 'anchoring' in order to become a full-fledged information unit. I propose that exactly the same account applies to IP with a checked +Topic feature. In other words, IP with the +Topic feature unchecked denotes a proposition, an intensional entity without the information-packaging structure of an information unit. An information unit contains additional information about topic and focus structure. This kind of information is organized at the CP level. C mediates anchoring to the context of the discourse, as I argued in Svenonius 1996. Above, I suggested that Agr was type
. This is most naturally understood in the context of Norwegian or German. In general, the elements there that check the +Topic feature are in fact topical. This means that when a topic moves to adjoin to IP, the IP node created to dominate that topic denotes an open function, ready for discourse anchoring (mediated, as noted, by C). This anchoring includes not only information about the topic, but also indicates whether the sentence is declarative, interrogative, and so on. When an expletive is adjoined to IP (to check the D feature), there is no topic in the sentence, but since pronouns are formally marked -t-Topic, the switch is set anyway (the possibility of expletives checking the feature suggests that it would be better labeled -Focus, as in Holmberg and Nikanne, this volume).
SUBJECT POSITIONS AND THE PLACEMENT OF ADVERBIALS
235
5.3. Accounting for object shift This account can treat object shift and scrambling phenomena in a parallel way, given two additional assumptions: Infl must be able to check +Topic features on more than one DP (cf. Chomsky 1995: ch. 4, Richards 1997), and something must ensure that subjects raise even when another DP (the object) has moved to the checking domain of Infl. I will assume that movement of two elements to the same checking domain involves crossing, rather than nesting, as detailed in Richards 1997: the subject raises first, because it is closer to the target (TP); then the object raises; but Shortest Move ensures that it targets the same node to which the subject is adjoined, thus attaching below the subject, resulting in subject-object order.16 In Norwegian and German, pronouns and full DP objects which are in fact topical are marked +Topic. Thus they move to adjoin to IP. This has the result that they precede sentential adverbials, since those adverbials cannot be attached afterward, in accordance with (62). However, the subject can also be attached to IP, but is attached first, resulting in subject-object-adverb order, as in (63). (63) a. b.
Heute lasen die Studenten es wahrscheinlich nicht. I dag leste studentene den sannsynligvis ikke. today read the students it probably not 'Today the students probably didn't read it'
German Norwegian
In German, it is also possible for an object to cross over a subject, appearing to its left; in the system adopted here, this means that a subject can move after an object has already raised (it is also possible, under more limited circumstances, in Swedish; cf. Holmberg 1986, Josefsson 1993). This is not possible in Norwegian, though I have no explanation as to why. It is also impossible in Dutch, so it is apparently not connected with OV order. A fact which apparently is connected with OV order is the fact that object shift cannot cross a verb, as indicated below. (64) a. b.
Heute haben die Studenten es wahrscheinlich nicht gelesen. I dag har studentene {*den} sannsynligvis ikke lest {den}. today have the students it probably not read it Today the students probably haven't read it'
I will not try to provide an account for this fact, either (cf. Holmberg 1997). What it shows is that the trigger for object shift cannot be a strong feature. This is consistent with the assumption made above, that +Topic is not a strong feature. In Icelandic and Mainland Scandinavian it is obligatory when possible with weak pronouns, and appears to be fully optional with full DPs. Arguably, with full DPs it is not strictly optional but linked with topicality. I will assume this to be the case, as it is with subjects. (65) a.
b.
Olabetalte {henne} heldigvis {*henne} tilbake. Ola paid her fortunately her back 'Ola fortunately paid her back'
Olabetalte {Kari} heldigvis {Kari} tilbake. Ola paid Kari fortunately Kari back 'Ola fortunately paid Kari back'
Norwegian
236
SUBJECTS, EXPLETIVES, AND THE EPP
Setting pronouns aside, the assumption is that when an object is informationally topical, which in Norwegian and German corresponds to being marked +Topic, it undergoes object shift if possible (e.g., if it does not have to cross a verb). This appears at first glance to be the opposite of the situation with Procrastinate, where a movement is postponed if it can be; here a movement is not postponed unless it must be. Such a situation can be handled by Economy in the sense of Chomsky 1995: ch. 2, if the derivation in which the object moves overtly is more economical that the one in which it does not; this is true if the overt movement makes unnecessary a separate covert movement, for example, to C. This is what I will assume.17 Swedish is different in that all (nonfocused) DPs are formally marked +Topic. However, the object shift facts are essentially the same as for Norwegian: pronouns obligatorily undergo object shift, and full DPs optio: ally do. (66) a.
b.
Lars gav {dem} lyckligtvis {*dem} till Ulla. Lars gave them fortunately them to Ulla 'Lars fortunately gave them to Ulla' Lars gav {pengarna} lyckligtvis {pengarna} till Ulla. Lars gave the. money fortunately the.money to Ulla 'Lars fortunately gave the money to Ulla'
Swedish
I assume exactly the same mechanisms as for Norwegian. If the object is informationally topical, then everything proceeds as in Norwegian. If, on the other hand, the object is not informationally topical, then it does not need to move to the checking domain of C at LF. Thus the overt movement of the object will violate Procrastinate, even if it is formally +Topic. The object may, on the other hand, move covertly to Infl (or its features may, following Chomsky 1995: ch. 4), to check its +Topic features, given the assumption made above that Infl may check +Topic features on more than one DP (e.g., the subject as well as the object). Similar assumptions apply for Icelandic, in which all definite DPs are formally marked +Topic, and for Danish, in which all DPs check the +Topic feature. This is only a rough sketch, and additional issues need to be addressed. As the account now stands, (62) will ensure that no sentential adverbs precede an informationally topical object that has been permitted to escape VP. However, as seen in §2.3 above, object shift sometimes leaves an object between sentential adverbs (even with light pronouns, in Swedish; cf. Holmberg and Platzack 1995:153, note 11). This might represent movement to check some feature other than +Topic, or it might mean that (62) is sensitive to whether the element having checked the +Topic feature is Nominative. I will not speculate further on this matter here.
6. Conclusion I have examined the evidence for multiple subject positions in Germanic; as this evidence was primarily wrapped up in questions of adverb placement, my route took me through various difficult issues regarding adverbials, which I realize I have not done justice to. However, I think that I have established that the assumption made in a great deal of the foundational literature—that adverbs can be used to
SUBJECT POSITIONS AND THE PLACEMENT OF ADVERBIALS
237
identify specific clausal boundaries—is too optimistic, and that a certain degree of variability must be countenanced. However, there is also a great degree of rigidity, something which I suggested could be handled by a proper semantic treatment. I then examined the nature of the different subject positions, and suggested that a properly conceived notion of topicality could account for most of the differences in interpretation noted in the literature, essentially in agreement with Adger 1994 and Meinunger 1995. For sentential adverbials, I have allowed an important part of the variability of adverb placement to play itself out in the order of adjunction: adverbs can, I suggested, adjoin before or after the adjunction of a feature-checking element such as a subject. This leads ultimately to such a reduction in the structural role of Agr that its elimination can be considered.
Notes Thanks to Tom Ernst and an anonymous reviewer for Oxford University Press for useful comments on an earlier draft. Thanks also to David Adger, Anders Holmberg, Porbjorg Hr6arsd6ttir, 0ystein Nilsen, and Tarald Taraldsen for helpful discussion; to the participants in the Troms0 Workshop on Subjects, Expletives, and the EPP; and to the many people who have provided me with native speaker intuitions. 1. This is consistent with M. Baker's 1985 Mirror Principle. The Mirror Principle states that morphological derivations must directly reflect syntactic derivations (and vice versa); e.g., if an 'applicative' process first promotes an oblique argument to direct object status, and then a passive process promotes that direct object to subject status, then the morphological processes must also apply in that order. If head movement of the verb to combine with a T or Agr head is classified as a 'syntactic derivation' in this sense, then Belletti's ordering is in fact forced by the Mirror Principle. 2. Incorporation of adverbs is well-attested (cf., for example, Rivero 1992, Alexiadou 1997). However, incorporation is arguably motivated by morphological considerations, such as whether a morpheme is or is not a bound one (M. Baker 1988, Rizzi and Roberts 1989), absent in phrasal movement. 3. Example (12) is Cinque's (17), p. 49. Example (13) is based on his (1), p. 45. Cinque's original sentence (la) is: [i] Da allora, non hanno rimesso di solito mica piu sempre Italian since then, not have put usually not any.longer always completamente tutto bene in ordine. completely everything well in order 'Since then, they haven't usually not any longer always put everything well into order' Cinque also gives five other possible placements for rimesso, one after each of the five adverbial expressions following it in [i]. 4. Tom Ernst cautions (p.c.) that 'still' can also appear relatively high; but the fact is that subjects can follow preverbal adverbs in Icelandic quite generally. 5. The fact that sem beturfer (literally 'as better goes') is phrasal, while heldigvis is a single word, is irrelevant; the same pattern obtains for Icelandic single-word adverbs like sennilega 'probably' and for Norwegian phrasal adverbs like cerlig talt 'honestly' (literally, 'honestly said'). 6. Nor do the arguments move as a unit: adverbs may intervene between the subject and indirect object, or between the indirect object and direct object, as long as elements following adverbs are not pronominal (since object shift is obligatory when possible for pronouns). The relative order subject-indirect object-direct object is fairly strict. 7. Jackendoff (1972:67) gives an example like (20c)"?" (and on p. 76 a slightly different one is "*"); on p. 81 it is ruled out by Jackendoff's formal rules, and the relative acceptability of some adverbs after Modal + have is noted as an anomaly; Ernst (1984:255) marks such a sentence with "(?)" and notes that it is "not fully acceptable to many speakers." I will take it to be ungrammatical here.
238
SUBJECTS, EXPLETIVES, AND THE EPP
8. An anonymous reviewer observes that the obligatory Italian ordering also holds of the corresponding French adverbs d'habitude and plus. Thus the restriction on Italian is not entirely parochial, either. 9. The literature on this subject is vast. See, for example, Abraham 1992 or 1995: chs. 13-14, for some discussion of German, with references; Rognvaldsson 1982 for discussion and references for Icelandic; Faarlund 1992 for Norwegian. 10. Uttered by the butler in the Swedish translation of Herge's (1953) Les Aventures de Tmtin: Objectif Lune, p. 1 (translated by Karin and Allan B. Janzon, Bonnier Carlsen Bokfdrlag, Stockholm). 11.1 will not discuss the alternative view that subject-initial sentences are IPs, while nonsubject initial sentences are CPs. See, e.g., Zwart 1993 or Vikner 1995. 12. Cf. E. Kiss 1996 for arguments for two subject positions in English. 13. The two positions can also be distinguished by the possibility of an expletive: the expletive appears in the higher subject position, so it cannot co-occur with contentful subject appearing there. See Vangsnes (this volume) for detailed discussion of the so-called Transitive Expletive Construction. 14. Thanks to Endre M0rck for discussion of Middle Norwegian. 15. This is not to say that there is no independent evidence for the head positions Agr and T; cf., e.g., Manzini and Savoia, this volume. However, as I noted in §3.5, releasing adverbs from the very strict confines of feature-checking in a spec-head configuration introduces a degree of flexibility in their placement just when the semantics of the head does not completely determine the applicability of the adverb, as is arguably the case with a head like Agr. Thus I would argue that even if the more radical stance taken here is rejected, that of eliminating Agr altogether, the account proposed for adverb-subject order is superior to one which connects Germanic subject-adverb order to subject movement to SpecAgrP. 16. If movement involved nesting, then subject-object order would reflect a derivation in which the object moved first, then the subject. But in the model to be developed here, it would then be difficult to explain EPP effects (because a shifted object would check the strong D feature that I am assuming forces subject raising). 17. One possibility, roughly sketched, is that the object must move covertly to the checking domain of C, because it is topical, and C mediates information structure. If the verb moves overtly to C, then a shifted object can pass its +Topic feature up to C as a free rider. This favors overt object shift just in case the verb moves to C. If the verb does not move to C, then additional movement will be necessary for the object regardless, rendering the Procrastinate violation gainless and therefore impossible. However, this account faces various problems, including some of the ones noted in Holmberg 1997 for competing accounts.
References Abraham, Werner. 1992. Wortstellung im Deutschen—theoretische Rechtfertigung, empirische Begrundung. In Ludger Hoffman (ed.), Deutsche Syntax. Ansichten und Aussichten, 484-522. Walter de Gruyter, Berlin. Abraham, Werner. 1995. Deutsche Syntax im Sprachenvergleich. Grundlegung einer typologischen Syntax des Deutschen. Gunter Narr Verlag, Tubingen. Adger, David. 1994. Functional Heads and Interpretation. Ph.D. dissertation, University of Edinburgh. Aissen, Judith. 1992. Topic and focus in Mayan. Language 68,43-80. Alexiadou, Artemis. 1997. Adverb Placement. John Benjamins, Amsterdam. Alexiadou, Artemis, & Elena Anagnostopoulou. 1998. Parametrizing AGR: Word order, Vmovement, and EPP-checking. Natural Language & Linguistic Theory 16, 491-539. Baker, C. L. 1991. The syntax of English not: The limits of core grammar. Linguistic Inquiry 22,387-430. Baker, Mark. 1985. The Mirror Principle and morphosyntactic explanation. Linguistic Inquiry 16, 373^16.
SUBJECT POSITIONS AND THE PLACEMENT OF ADVERBIALS
239
Baker, Mark C. 1988. Incorporation. University of Chicago Press, Chicago. Barbiers, Sjef. 1995. The Syntax of Interpretation. Ph.D. dissertation, Rijksuniversiteit te Leiden. MIL Dissertations 14. Holland Academic Graphics, The Hague. Beghelli, Filippo, & Tim Stowell. 1997. Distributivity and negation: The syntax of each and every. In Anna Szabolcsi (ed.), Ways of Scope Taking, 71-107. Kluwer, Dordrecht. Belletti, Adriana. 1990. Generalized Verb Movement. Rosenberg & Sellier, Turin. Bennis, Hans. 1986. Gaps and Dummies. Foris, Dordrecht. Besten, Hans den. 1977. On the interaction of root transformations and lexical deletive rules. Ms., University of Amsterdam. [Published 1983 in Werner Abraham (ed.), On the Formal Syntax of the Westgermania, 47-131. John Benjamins, Amsterdam.] Bobaljik, Jonathan. 1995. Morphosyntax: The Syntax of Verbal Inflection. Ph.D. dissertation, MIT. Bobaljik, Jonathan, & Dianne Jonas. 1996. Subject positions and the roles of TP. Linguistic Inquiry 27, 195-236. Bobaljik, Jonathan, & Hoskuldur Thrainsson. 1997. Two heads aren't always better than one. Ms., Harvard and McGill Universities and University of Iceland. Chomsky, Noam. 1986. Barriers. MIT Press, Cambridge, Mass. Chomsky, Noam. 1995. The Minimalist Program. MIT Press, Cambridge, Mass. Christensen, Kirsti Koch. 1991. AGR, adjunction, and the structure of Scandinavian existential sentences. Lingua 84, 137-158. Christoffersen, Marit. 1993. Setning og sammenheng: Syntaktiske studier i Magnus Lagab0ters landslov. Ph.D. dissertation, Agder College. ADH-serien 65, Agder distriktsh0gskole, Kristiansand. Cinque, Guglielmo. 1999. Adverbs and the Universal Hierarchy of Functional Projections. Oxford University Press, New York. Collins, Chris, & Hoskuldur Thramsson. 1996. VP-internal structure and Object Shift in Icelandic. Linguistic Inquiry 27, 391-444. Cresti, Diana. 1995. Indefinite Topics. Ph.D. dissertation, MIT. Diderichsen, Paul. 1946. Elementcer dansk grammatik. Ejnar Munksgaard, Copenhagen. Diesing, Molly. 1992. Indefinites. MIT Press, Cambridge, Mass. Emonds, Joseph. 1976. A Transformational Approach to English Syntax: Root, Structurepreserving, and Local Transformations. Academic Press, New York. Engel, Ulrich. 1972. Regeln zur "Satzgliedfolge." Zur Stellung der Elemente im einfachen Satz. Linguistische Studien 1, 17-75. Ernst, Thomas. 1984. Towards an Integrated Theory of Adverb Positions in English. Ph.D. dissertation, Indiana University. Indiana University Linguistics Club, Bloomington, Indiana. Faarlund, Jan Terje. 1992. Norsk syntaks ifunksjoneltperspektiv. Universitetsforlaget, Oslo. Firbas, Jan. 1975. On the thematic and non-thematic section of the sentence. In Hakan Ringbom (ed.), Style and Text: Studies Presented to Nils Erik Enkvist, 314-334. Skriptor, Stockholm. Givon, Talmy. 1990. Syntax: A Functional-Typological Introduction. John Benjamins, Amsterdam. Gundel, Jeanette. 1974. The Role of Topic and Comment in Linguistic Theory. Ph.D. dissertation, University of Texas at Austin. [Published 1988. Garland, New York. Page numbers here refer to the Garland version.]
240
SUBJECTS, EXPLETIVES, AND THE EPP
Haider, Hubert. 1986. Configurationality in disguise: Word-order and the V-2 property. In Werner Abraham & S. de Meij (eds.), Topic, Focus, and Configurationality, 39-64. John Benjamins, Amsterdam. Heim, Irene. 1982. The Semantics of Definite and Indefinite Noun Phrases. Ph.D. dissertation, University of Massachusetts, Amherst. Hellan, Lars. 1989. A two-level X-bar system. Working Papers in Scandinavian Syntax 44, 61-74. Heltoft, Lars, and Lisbeth Falster Jakobsen. 1996. Danish passives and subject positions as a mood system: A content analysis. In Elisabeth Engberg-Pedersen et al. (eds.), Content, Expression, and Structure: Studies in Danish Functional Grammar, 199-234. John Benjamins, Amsterdam. Hoekstra, Eric. 1991. Licensing Conditions on Phrase Structure. Ph.D. dissertation, Rijksuniversiteit Groningen. Grodil, Groningen. Holmberg, Anders. 1983. The finite sentence in Swedish and English. Working Papers in Scandinavian Syntax 3. Holmberg, Anders. 1986. Word Order and Syntactic Features. Ph.D. dissertation, University of Stockholm. Holmberg, Anders. 1993. Two subject positions in IP in Mainland Scandinavian. Working Papers in Scandinavian Syntax 52, 29-41. Holmberg, Anders. 1997. The true nature of Holmberg's Generalization. In Kiyomi Kusumoto (ed.), Proceedings of the North East Linguistic Society 27, 203-217. GLSA, University of Massachusetts, Amherst. Holmberg, Anders, Urpo Nikanne, Irmeli Oraviita, Hannu Reime, & Trond Trosterud. 1993. The structure of INFL and the finite clause in Finnish. In Anders Holmberg & Urpo Nikanne (eds.), Case and Other Functional Categories in Finnish Syntax, 177-206. Mouton de Gruyter, Berlin. Holmberg, Anders, & Christer Platzack. 1995. The Role of Inflection in Scandinavian Syntax. Oxford University Press, New York. Hoop, Helen de. 1992. Case Configuration and Noun Phrase Interpretation. Ph.D. dissertation, Rijksuniversiteit Groningen. Grodil, Groningen. latridou, Sabine. 1990. About Agr(P). Linguistic Inquiry 21, 551-576. Jackendoff, Ray. 1972. Semantic Interpretation in Generative Grammar. MIT Press, Cambridge, Mass. Jonas, Dianne. 1994. The TP parameter in Scandinavian syntax. In Cecilia Hedlund & Anders Holmberg (eds.), Proceedings of the XFVth Scandinavian Conference of Linguistics and the VIHth Conference of Nordic and General Linguistics, Special Session on Scandinavian Syntax, 33-59. Gothenburg Papers in Theoretical Linguistics, Gothenburg University. Josefsson, Gunlog. 1993. Scandinavian pronouns and object shift. Working Papers in Scandinavian Syntax 52, 1-28. Julien, Marit. 2000. Syntactic Heads and Word Formation: A Study of Verbal Inflection. Ph.D. dissertation, University of Troms0. Kayne, Richard S. 1993. Toward a modular theory of auxiliary selection. Studia Linguistica 47, 3-31. Kayne, Richard S. 1994. The Antisymmetry of Syntax. MIT Press, Cambridge, Mass. King, Tracy. 1995. Configuring Topic and Focus in Russian. CSLI, Stanford, Calif. Kiparsky, Paul, & Carol Kiparsky. 1970. Fact. In Manfred Bierwisch & Karl Erich Heidolph (eds.), Progress in Linguistics, 143-173. Mouton, The Hague.
SUBJECT POSITIONS AND THE PLACEMENT OF ADVERBIALS
241
Kiss, Katalin E. 1996. Two subject positions in English. The Linguistic Review 13,119-142. Kratzer, Angelika. 1995. Stage-level and individual-level predicates. In Gregory Carlson and Francis Pelletier (eds.), The Generic Book, 125-175. University of Chicago Press, Chicago. Kristiansen, Erik. 1996. Topic continuity and prosody: An experimental study in Danish. In Elisabeth Engberg-Pedersen et al. (eds.), Content, Expression, and Structure: Studies in Danish Functional Grammar, 261-282. John Benjamins, Amsterdam. Kulbrandstad, Lars Anders. 1978. Presenteringskonstruksjonen i moderne norsk—struktur, funksjon og bruk. M.A. thesis, University of Oslo. Kuno, Susumu. 1973. The Structure of the Japanese Language. MIT Press, Cambridge, Mass. Ladusaw, William. 1988. Adverbs, negation, and QR. In Linguistic Society of Korea (ed.), Linguistics in the Morning Calm 2: Selected Papers from SICOL-1986, 481-488. Hanshin Publishing Company, Seoul. Lasnik, Howard, & Mamoru Saito. 1992. Move Alpha: Conditions on Its Application and Output. MIT Press, Cambridge, Mass. Maling, Joan, and Annie Zaenen. 1978. The non-universality of a surface filter. Linguistic Inquiry 9, 475-497. May, Robert. 1985. Logical Form: Its Structure and Derivation. MIT Press, Cambridge, Mass. McConnell-Ginet, Sally. 1982. Adverbs and logical form. Language 58,144-184. Meinunger, Andre. 1995. Discourse Dependent DP (De-) Placement. Ph.D. dissertation, University of Potsdam. Milsark, Gary L. 1974. Existential Sentences in English. Ph.D. dissertation, MIT. [Published 1979. Garland, New York.] M0rck, Endre. 1999. Midtfeltet i mellomnorske heilsetninger. Norsk Lingvistisk Tidsskrift 17, 183-200. Nilsen, 0ystein. 1997. Adverbs and A-shift. Working Papers in Scandinavian Syntax 59, 1-31. Nilsen, 0ystein. 1998. The Syntax of Circumstantial Adverbials. M.A. thesis, University of Troms0. Platzack, Christen 1983. Germanic word order and the COMP/INFL parameter. Working Papers in Scandinavian Syntax 2. Platzack, Christer. 1985. The Scandinavian languages and the null subject parameter. Working Papers in Scandinavian Syntax 20. Platzack, Christer. 1986a. COMP, INFL, and Germanic word order. In Lars Hellan & Kirsti Koch Christensen (eds.), Topics in Scandinavian Syntax, 185-234. Reidel, Dordrecht. Platzack, Christer. 1986b. The position of the finite verb in Swedish. In Hubert Haider & Martin Prinzhorn (eds.), Verb Second Phenomena in Germanic Languages, 27-47. Foris, Dordrecht. Pollock, Jean-Yves. 1989. Verb movement, Universal Grammar, and the structure of IP. Linguistic Inquiry 20, 365-^24. Reinhart, Tanya. 1981. Pragmatics and linguistics: An analysis of sentence topics. Philosophica 27, 1-38. Richards, Norvin. 1997. AgrVP. Paper presented at the 13th Comparative Germanic Syntax Workshop, Cornell University, Ithaca, NY. Rivero, Maria-Luisa. 1990. The location of nonactive voice in Albanian and Modern Greek. Linguistic Inquiry 21, 135-146.
242
SUBJECTS, EXPLETIVES, AND THE EPP
Rivero, Maria-Luisa. 1992. Adverb incorporation and the syntax of adverbs in Modern Greek. Linguistics and Philosophy 15, 289-331. Rizzi, Luigi. 1990. Relativized Minimality. MIT Press, Cambridge, Mass. Rizzi, Luigi, & Ian Roberts. 1989. Complex inversion in French. Probus 1, 1-30. Rognvaldsson, Eirikur. 1982. Urn ordarod og fxrslur ffslensku. Master's thesis, University of Iceland. [Published 1990 as vol. 2 of Malfraedirannsoknir. Malvisindastofnum Haskola Islands, Reykjavik.] Sigur5sson, Halld6r Armann. 1990. VI declaratives and verb raising in Icelandic. In Joan Maling and Annie Zaenen (eds.), Modern Icelandic Syntax, 41-69. Academic Press, New York. Svenonius, Peter. 1993. The structural location of the attributive adjective. In Erin Duncan, Eric Potsdam, & Philip Spaelti (eds.), Proceedings of the Twelfth Annual West Coast Conference on Formal Linguistics, 438-454. CSLI, Stanford, Calif. Svenonius, Peter. 1994. Dependent Nexus: Subordinate Predication Structures in English and the Scandinavian Languages. Ph.D. dissertation, University of California, Santa Cruz. Svenonius, Peter. 1996. Predication and functional heads. In Jose" Camacho, Lina Choueiri, & Maki Watanabe (eds.), The Proceedings of the Fourteenth West Coast Conference on Formal Linguistics, 493-507. CSLI, Stanford, Calif. Swart, Henriette de. 1991. Adverbs of Quantification: A Generalized Quantifier Approach. Ph.D. dissertation, Rijksuniversiteit Groningen. Grodil, Groningen. Thramsson, Hoskuldur. 1979. On Complementation in Icelandic. Ph.D. dissertation, Harvard University. [Published 1979. Garland Press, New York.] Thrainsson, Hoskuldur. 1996. On the (non-)universality of functional categories. In Werner Abraham et al. (eds.), Minimal Ideas: Syntactic Studies in the Minimalist Framework, 253-281. John Benjamis, Amsterdam. Vangsnes, 0ystein Alexander. 1995. Referentiality and argument positions in Icelandic. Working Papers in Scandinavian Syntax 55, 89-109. Vendler, Zeno. 1967. Linguistics in Philosophy. Cornell University Press, Ithaca, NY. Vikner, Sten. 1995. Verb Movement and Expletive Subjects in the Germanic Languages. Oxford University Press, New York. Webelhuth, Gert. 1984. German is configurational. Linguistic Review 4, 203-246. Webelhuth, Gert. 1992. Principles and Parameters of Syntactic Saturation. Oxford University Press, New York. Zanuttini, Raffaella. 1997. Negation and Clausal Structure. Oxford University Press, New York. Zwart, Jan-Wouter. 1993. Dutch Syntax: A Minimalist Approach. Ph.D. dissertation, University of Groningen. Grodil, Groningen.
Index A vs. A-bar 38, 77, 87, 126 Adjunction 203-204 Agr 59-60, 67 n. 14 and case 38 as a projection 73, 201 as target of head movement 211-212 connection to morphology 202, 215 in C-T dependency 131, 136 interpretation of 59-60, 219, 230 topic feature in 72, 230 Agreement 8, 57-58, 86-87, 99 Agree 59-60, 162 and the Mirror Principle 215 as a definition of specifier 203 as sharing of aspectual features 173, 181, 193 inC30 in I 34 object 208 postverbal subject 170, 180, 184 referential vs. predicative 181, 183, 193 subject 73, 86-87,170, 173 with the associate 8, 57, 175 with the expletive 7, 170, 172, 176 Attract 162
subject and 86, 119, 120 Celtic: See Irish Circular reference: See referentiality, self Clitic as functional category 160, 178-179 doubling 180 landing site 37 negation 177 subject 33, 160, 165, 177 Complex inversion 32-33
Bare noun phrase: See Indefinites Binding as a diagnostic for subjecthood 4, 87, 102 n. 21, 134 Breton 143-144 Burzio's Generalization 6
ECP: See Empty Category Principle Empty Category Principle (ECP) 38-39 EPP: See Extended Projection Principle Ergative languages 4, 5 Expletive 5-9 and agreement 7, 32, 35, 95 and case 53, 55-57, 90-92, 191 and control 190 and Definiteness Effect 53, 54-55 as a predicate 8 extraposition 6-7, 76, 111-113 in French 12, 31-32, 34-35, 151 n. 23, 176, 191, 196 n. 6 in German 128, 145, 191 in Hungarian 111-113 in object position 6 in V2 47-^8, 128-129
Case and binding 89 Case/agreement chain 7 checking 38, 56-57, 66, n. 12 inherent 55 of expletives 8, 46, 62, 64, 90-92, 117, 192 partitive 55, 90-92 strong and weak 13, 55-57 structural 55
Danish 19, 129, 145, 146, 221, 222, 230 Definite DPs internal structure of 189 Definiteness effect 7, 46-54, 54-57, 63, 65 n. 4, 66 n. 12 full vs. third person agreement 186, 189 on low subject 222-223, 224 Dependency 130 conditions on 132 recoverability of 132 Discourse configurationality: See topic prominence Dutch 224
244
INDEX
inVallader31,35 it vs. there 1,35,118 null 8, 32, 37, 161 semantic content of 8, 62 topic 9, 95 weather 5-6, 76 with no associate 7 Extended Projection Principle (EPP) absent in Irish 11 and agreement 10, 77, 80-81, 133 and anchoring 61 and case 20 and deixis 61-62 and predication 20 andT132 and topicality 83, 115, 116 as a strong D feature 20, 58, 126, 136 in nominals 127 in non-finite clauses 134-135 in small clauses 20-21 in V2 clauses 128-129 in VSO languages 129, 142-144 interaction with focus 79-80, 83, 97 satisfied by adverbials 81 satisfied by head movement 11 satisfied by V to I 58, 77, 164 universality of 9-11, 77, 107-108 Faroese 223 Feature interpretable vs. non-interpretable 39, 79, 193 lexical vs. agreement 60 movement 34, 39, 161-162 strength 22, 34, 58, 80 uniquely expressed in a chain 33 Finnish 206 Floated quantifier 88, 102 n. 21 Focus licensing low subject 109, 220-222 [±Foc] feature 79-80, 111 French 78 Functional category and markedness 141 and morphology 214-215 motivation for 74, 131, 201-204
German 126, 128, 137-139, 145, 146, 215-216, 219, 223, 227, 230 Gothic 147 Greek 202-203, 205 Head movement long 143-145 of finite verb: See V to I of participles in Italian 207, 212 ofTtoAgr 127 Hebrew 76 Icelandic 96, 146, 207, 208, 209, 220, 223, 226, 230 Identification 30, 59-64, 125 Impersonal passives 7 Indefinites as subject: See under subject bare 50-51,108 internal structure of 189 strong readings of 226, 227, 228 Individual-level: See predicate Irish 11,131 Old Irish 143 Italian 16, 206-207 Japanese 217-218 Lexicalization parameterized 165,194 Long Head Movement (LHM): See head movement Mapping Hypothesis 52-53, 79, 219-220, 223 Markedness 141 Middle Norwegian 228 Mirror Principle 215 Multiple specifiers: See Multiple Subject Construction Multiple Subject Construction (MSC) and multiple specifiers 98-100 in Dutch 224 in Finnish 72, 90-98 in Icelandic 43-44 in Japanese and Russian 217-218 with strong DPs 48^9
INDEX Norwegian 45, 59, 207-208, 209, 216, 217, 218,221,225,227,230 Null subject: See pro drop Object Shift 79, 208-209, 235 Old Icelandic 229 Old Norwegian 228 Parameter 194 Partitive Hypothesis 55, 91-92 Predicate individual-level 49-50, 113 stage-level 50, 113 Predication 116, 119 EPP and: See Extended Projection Principle (EPP) Pro drop 10-11, 17,61,62 and expletives 71, 75, 94 in Chinese 10-11 in Finnish 19, 75-76 parameter 74-75,77,133,157,166-167 without pro 193 Projection Principle 9 Quasi-argument 6, 10, 20, 75, 119 Reconstruction 36, 87, 89 Referentiality of arguments 133 self: See circular reference Russian 78, 217 Segment-category distinction 203-204, 210 Specifier-adjunct distinction 203-204 Stage-level: See predicate Strong quantifier in associate position 13,46-54 in topicless sentence 115-117 subject 110 Stylistic Fronting 81 Stylistic Inversion 12, 35-38 Subject: See clitic as topic 4, 110, 169 characteristics 118-119 indefinite 108,109, 116-118 multiple positions 37, 44-46, 56-57, 117
245
postverbal 35-38, 46^8, 167-168 strongly quantified: See strong quantifier thematic 3, 60, 120 VP-internal31, 113, 116-118 Swedish 216, 220, 230 Taraldsen's Generalization 10 Tense 129 as a projection 211 Thai-trace effect 38-39 Thematic role and role of pro 161 as aspectual feature 162 Topic 215-218 A vs. A-bar 87 absence of 113, 116 and EPP: See Extended Projection Principle and topicality and subject 73, 84 as a feature 138, 230 as an A-bar position 87 as topic 115 continued 216-217, 218, 226-229, 230 null operator 139-140 prominence 4, 71, 78, 113, 122, 230 switch 216-218 topical indefinites: See indefinites, strong readings of Transitive Expletive Construction (TEC): See Multiple Subject Construction Vtol and agreement 59, 80 in English 211 in Italian 206-207, 211-212 V2 8, 9, 137-147 and complementizers 145 initial position in 140-141, 215-217 origin of 146-147 with empty first position (VI) 138-141 Vallader 30-31,34-35 Welsh 142 Yiddish 146