The Unaccusativity Puzzle: Explorations of the Syntax-Lexicon Interface (Oxford Studies in Theoretical Linguistics, 5)

The Unaccusativity Puzzle

Oxford Studies in Theoretical Linguistics General editors David Adger, University of York; Hagit Borer, University of Southern California

Advisory editors Stephen Anderson, Yale University; Gennaro Chierchia, University of Milan; Rose-Marie Dechaine, University of British Columbia; Elan Dresher, University of Toronto; James Higginbotham, University of Southern California; Pat Keating, University of California, Los Angeles; Ruth Kempson, School of Oriental and Mrican Studies, University of London; James McCloskey, University of California, Santa Cruz; Gillian Ramchand, University of Oxford; Maria-Luisa Zubizarreta, University of Southern California This series provides a forum for cutting-edge work in theoretical linguistics. Its focus is on the interfaces between the subcomponents of grammar and between grammar and other components of the mind. PUBLISHED

r. The Syntax of Silence Sluicing, Islands, and the Theory of Ellipsis by Jason Merchant 2.

Questions and Answers in Embedded Contexts by Utpal Lahiri 3. Phonetics, Phonology, and Cognition

edited by Jacques Durand and Bernard Laks 4. The Syntax-Pragmatics Interface Concept Formation and Verbal Underspecification in Dynamic Syntax by Lutz Marten 5. The Unaccusativity Puzzle Explorations of the Syntax-Lexicon Interface edited by Artemis Alexiadou, Elena Anagnostopoulou, and Martin Everaert IN PREPARATION

Tense, Mood, and Aspect

edited by Alessandra Giorgi, James Higginbotham, and Fabio Pianesi The Ecology of English Noun-Noun Compounding by Ray Jackendoff The Oxford Handbook of Linguistic Interfaces edited by Gillian Ramchand and Charles Reiss

[published in association with the series]

The Unaccusativity Puzzle Explorations ofthe Syntax-Lexicon Interface

edited by ARTEMIS ALEXIADOU ELENA ANAGNOSTOPOULOU

and MARTIN EVERAERT

OXFORD UNIVERSITY PRESS

OXFORD UNIVERSITY PRESS

Great Clarendon Street, Oxford oX26DP Oxford University Press is a department of the University of Oxford. It furthers the University's objective of excellence in research, scholarship, and education by publishing worldwide in Oxford New York Auckland Bangkok Buenos Aires Cape Town Chennai Dar es Salaam Delhi Hong Kong Istanbul Karachi Kolkata KuaiaLumput Madrid Melbourne Mexico City Mumbai Nairobi Sao Paulo Shanghai Taipei Tokyo Toronto Oxford is a registered trade mark of Oxford University Press in the UK and in certain other countries Published in the United States by Oxford University Press Inc., New York

© Editorial matter and organization Alexiadou, Anagnostopoulou, and Everaert © The several contributors and in this collection, Artemis Alexiadou, Elena Anagnostopoulou, and Martin Everaert, 2004 The moral rights of the authors have been asserted Database right Oxford University Press (maker) First published 2004 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, without the prior permission in writing of Oxford University Press, or as expressly permitted by law, or under terms agreed with the appropriate reprographics rights organizations. Enquiries concerning reproduction outside the scope of the above should be sent to the Rights Department, Oxford University Press, at the address above You must not circulate this book in any other binding or cover and you must impose the same condition on any acquirer A catalogue record for this title is available from the British Library Libraty of Congress Cataloging in Publication Data The unaccusativity puzzle: explorations of the syntax-lexicon interface / edited by Artemis Alexiadou, Elena Anagnostopoulou, and Martin Everaert. p. em. - (Oxford studies in theoretical linguistics ; 5) Includes bibliographical references and index. 1. Grammar, Comparative and general-Verb 2. Grammar, Comparative and general-Syntax 3. Lexicology. 1. Alexiadou, Artemis. II. Anagnostopoulou, Elena. III. Everaert, Martin. Iv. Series. P381.U532003

415'.6-dc22

2003060967

0199257647 (hbk) ISBN 0199257655 (pbk)

ISBN

13579108642 Typeset in Adobe Garamond by Peter Kahrel Ltd., Lancaster Printed in Great Britain on acid-free paper by Biddies Ltd., www.biddles.co.uk

Contents

General Preface

vii

Abbreviations

viii

Contributors

xi

Introduction

1

ARTEMIS ALEXIADOU, ELENA ANAGNOSTOPOULOU, AND MARTIN EVERAERT 1.

A Semantics for Unaccusatives and its Syntactic Consequences

22

GENNARO CHIERCHIA 2.

Unaccusativity as Telicity Checking

60

ANGELIEK VAN HOUT

3. Unergative Adjectives and Psych Verbs HANS BENNIS

4. Voice Morphology in the Causative-Inchoative Alternation: Evidence for a Non-Unified Structural Analysis of Un accusatives

114

ARTEMIS ALEXIADOU AND ELENA ANAGNOSTOPOULOU

5. Unaccusative Syntax and Verbal Alternations

137

DAVID EMBICK

6. Against an Unaccusative Analysis of Reflexives

159

TANYA REINHART AND TAL SILONI

7. Unaccusatives and Anticausatives in German

181

MARKUS STEINBACH

8. Syntactic Unaccusativity in Russian MAAIKE SCHOORLEMMER

20 7

Contents

vi

9. Gradience at the Lexicon-Syntax Interface: Evidence from Auxiliary Selection and Implications for Unaccusativity

243

ANTONELLA SORACE 10.

Unaccusativity in Saramaccan: The Syntax of Resultatives TONJES VEENSTRA

11.

The Grammar Machine

288

HAGIT BORER 12.

Acquiring Unaccusativity: A Cross-Linguistic Look

332

JANET RANDALL, ANGELIEK VAN HOUT, rURGEN WEISSENBORN, AND HARALD BAAYEN

References

355

Index

371

Oxford Studies in Theoretical Linguistics

General Preface

This volume has its origin in a workshop on unaccusativity organized by the Research Centre for General Linguistics (ZAS, Berlin) and the Netherlands Graduate School in Linguistics (LOT) hosted in Berlin in May 1998. However, the volume is independently structured and includes papers that were not presented during that event, such as the contributions by Chierchia, Reinhart, and Siloni and Sorace. The volume explores unaccusativity from different angles, and investigates various aspects of the phenomenon, such as syntactic versus semantic approaches, the mechanisms driving the projection of arguments, the role of (semi-)functional heads in determining verb class membership, the status of unaccusativity diagnostics, the nature of the special morphology associated with unaccusative predicates, and the importance of unaccusativity for (second) language acquisition research. The editors would like to thank David Adger and Hagit Borer for including the volume in their series, the contributors for their co-operation, two external reviewers for their insightful comments, and ZAS and LOT for the financial support that made the workshop possible. Artemis Alexiadou, Elena Anagnostopoulou, Martin Everaert

Abbreviations

ABS

ACC

Act AGR

AS ASH ASP AUX

d. CS DAT DEF DET

D1STR DO

ECM EPP F

GCC GEN

IEPS 1NSTR 1NTRANS 10

1PF

LOC M

MID N

Nact

absolutive case accusative case active agreement Argument Structure Auxiliary Selection Hierarchy aspect auxiliary elitic Conceptual Structure dative case default (agreement) determiner distributive (interpretation) direct object exceptional case-marking Extended Projection Principle feminine General Condition on A-chains genitive case Inferrable Eventual Position or State instrumental case intransitive indirect object imperfective locative masculine middle neuter non-active

Abbreviations NEG NOM

OE OM PASS PASSPART PAST

PF

PL

PPP PRT

PVC REFL SG

SU SUB}

SVC TNS TRANS

TRS UH UTAH

negation nominative case object experiencer object marker passive passive participle past tense perfective plural present-participle phrase partitive case perception-verb construction reflexive singular subject subject serial-verb construction tense transitive transitive reflexive sentence Unaccusative Hypothesis Universal Theta-Assignment Hypothesis

ix

Contributors

Artemis Alexiadou, University of Stuttgart Elena Anagnostopoulou, University of Crete Harald Baayen, Max Planck Institute for Psycholinguistics Hans Bennis, Meertens Instituut, Royal Netherlands Academy of Arts and Sciences Hagit Borer, University of Southern California Gennaro Chierchia, University of Milan David Embick, University of Pennsylvania Martin Everaert, Utrechts Instituut voor Lingulstiek OTS Janet Randall, Northeastern University Tanya Reinhart, Utrechts Instituut voor Lingulstiek OTS Maaike Schoorlemmer, Utrechts Instituut voor Lingulstiek OTS Tal Siloni, University of Tel-Aviv Antonella Sorace, University of Edinburgh Markus Steinbach, University of Mainz Angeliek van Hout, University of Groningen Tonjes Veenstra, John F. Kennedy Institute for North American Studies, FU Berlin Jiirgen Weissenborn, University of Potsdam

Introduction Artemis Alexiadou) Elena Anagnostopoulou) and Martin Everaert

This volume presents a collection of chapters of recent generative research into unaccusativity, which explore this key phenomenon from different angles. The volume has its origin in a workshop on unaccusativity, organized by the Research Center for General Linguistics (ZAS, Berlin) and the Netherlands Graduate School in Linguistics (LOT), hosted in Berlin in May 1998.' However, the book is independently structured-the contributors here are not the same as those participating in that event! In Part I of our introduction we introduce the theoretical background and the main issues in the unaccusativity research. In Part II we offer summaries of the chapters.

Part 1. Theoretical background: Issues in the study of the un accusative hypothesis 1.

THE PHENOMENON

Unaccusativity has been tal<en as a starting point in the study of the complex properties of the specification of verbs and verb classes.) The Unaccusative Hypothesis, as first , We would like to acknowledge these institutions for financially supporting that event. Alexiadou's research was partially supported by the DFG grant AL 5541r-I. , The chapter by Gennaro Chierchia deserves special mention. It has been circulated as a manuscript (from Cornell University) for many years, and has been very influential. It is published here for the first time in its original form. The author has written a postscript ro this chapter. J Observe that different terms are employed for the same phenomenon. We refer to the phenomenon by the term unaccusativity as originally introduced by Perlmutter (1978). Burzio (1981, 1986) uses the term ergative verb. Another term sometimes found in the literature is split intransitivity.

Artemis Alexiadou, Elena Anagnostopoulou, and Martin Everaert

2

formulated by Perlmutter (1978), and later adopted by Burzio (1981), was a syntactic hypothesis that claimed that there are two classes of intransitive verbs, unaccusative and unergative verbs, each associated with a different underlying syntactic configuration. 4 In Relational Grammar this was expressed as a distinction between verbs taking a final subject originating as an initial direct object (unaccusatives) and verbs taking a final subject that was also an initial subject (unergatives). From a Government-andBinding perspective (see Chomsky 1981 and subsequent work), an unergative verb takes a theta-marked deep-structure subject and no object, whereas an unaccusative verb takes a theta-marked deep-structure object:

(1)

a. b.

NP [yp V] [ypVNP]

unergative John sings unaccusative John came

Different classes of verbs have been analysed as unaccusative or as having an unaccusative alternate. We discuss some of these here. Rosen (1981, 1984) and Burzio (1981, 1986) argue that reflexive verbs such as the Italian verbs in (2) are unaccusative: (2)

a.

b.

Giovanni si vergognara. Giovanni himself ashames 'Giovanni is ashamed.' Gianni se e fotografato. Gianni himself is photographed 'Gianni has photographed himselE'

In Burzio (1986) ample evidence is provided that the surface-structure subjects in (2) are derived from a deep-structure object position: (3)

a.

b.

[s e [vp si vergognara Giovanni] [s Giovannij [yp si vergognara eJ

Chierchia, Embick, and Reinhart and Siloni (all in this volume) discuss this class of verbs (cE also Steinbach's chapter). Many transitive verbs allow an intransitive variant whose subject corresponds to the direct object of the transitive verb (see Partee 1965). This is illustrated in (4) and (5) with examples from French and Dutch (see, among others, Hoekstra 1984 and Everaert 1986 for Dutch; for French, Ruwet 1972 and Zribi-Hertz 1987; for Italian, Burzio 1986). The inchoative/anticausative variant ((4b), (5b» is generally analysed as unaccusative: (4)

a.

Jean brisera Ie verre. Jean will break the glass

4 Observe that the unaccusative-unergative distinction can only be formulated in a theory which distinguishes between subject and object, defined phrase-structurally as the grammatical functions which are the dedicated positions of proto-agent and proto-patient roles. If one does not believe in this dichotomy, then there is no reason ro believe that the class of (monadic) predicates is distinguished into a class of unaccusatives and a class of unergatives. It could just as well be a tripartite or quadrate distinction.

Introduction

(5)

b.

Le verre se brisera. the glass REFL will break 'The glass will break.'

a.

Hij verspreidde het gerucht. he spread the rumour Het gerucht verspreidde zich. the rumour spread itself 'The rumour spread.'

b.

3

As (4) and (5) illustrate, in many languages a reflexive clitic is added to the unaccusative variant of the verb (see the chapters by Alexiadou and Anagnostopoulou, Embick, Reinhart and Siloni, and Steinbach, in this volume). This is, however, not always the case, as is discussed for French (see (6)) in Ruwet (1972), for Dutch in Everaert (1986) and for German in Haider (1985): (6)

a. b.

Lennemi a couIe Ie bateau. the enemy has sunk the boat Le bateau a coule. the boat has sunk

Another class frequently investigated in the unaccusativity literature includes motion verbs, illustrated by the Dutch examples in (7)-(8). Such verbs are unergative (see (7)) which, when accompanied by a directional prepositional phrase (8), show all the characteristics of unaccusative verbs, such as the choice of auxiliary BE ((7b) vs (8b)), and the lack of impersonal passives ((7c) vs (8c)) (see Hoekstra and Mulder 1990; Narasimhan, Di-Tomaso, and Verspoor 1996; Rappaport Hovav and Levin 2000, among others): (7)

a. b. c.

(8)

a. b. c.

Hij loopt. 'He walks.' Hij heeftl*is gelopen. Er wordt gelopen. Hij loopt naar huis. , 'He walks home. Hij is/*heeft naar huis gelopen. ?*Er wordt naar huis gelopen.

Verbs of this type figure prominently in the chapters by Randall et al., van Hout, and Veenstra. Although the unaccusative-unergative dichotomy was introduced and discussed for one-place predicates, unaccusativity is not limited to such verbs. Two-place predicates such as experiencer-object verbs of the piacere class, taking a theme and an experiencer argument (see (9a) and (9b) for Italian and Dutch, respectively) and two-place double object verbs taking a theme and a goal/source argument (see ((lOa)

Artemis Alexiadou, Elena Anagnostopoulou, and Martin Everaert

4

and (rob» for English and Dutch, respectively) have also been argued to have derived subjects, thus qualifying as two-place unaccusatlves (see den Besten 1982; Belletti and Rizzi 1988; GrewendorfI989; Fanselow 1992, among many others). (9)

a. b.

(10)

a.

b.

Questo piace a Gianni. this appeals to Gianni Dat boek bevalt mij. that book pleases me The ring passed to Mary. De teugels ontglipten hem. the reins slipped him 'The reins slipped from his hands.'

A second class of object experiencer verbs, ofted called the preoccupare ('worry') class (including, for example, interest, attract,jrighten, disgust, excite), poses additional complications.

(n)

Questo preoccupa Gianni. 'This worries Gianni.'

Despite their thematic similarity to the verbs of the so-called piacere class, preoccupare verbs share relevant properties with transitive/causative verbs and there is debate on whether they should be classified as unaccusative or not (Belletti and Rizzi 1988; Grimshaw 1990; Pesetsky 1995, for example). This issue is addressed in Bennis's chapter. The unaccusative hypothesis is not limited to verbs but extends to other categories. Cinque (1990) has argued that adjectives can be divided into unaccusative and unergative as well. Italian (12) and (13) exemplifY unaccusative and unergative adjectives, respectively (anaphora and ne-cliticization diagnose the difference between the two classes of adjectives; see section 2 for some discussion).

(12)

a.

b. (13)

II PropriOi destino non era noto a nessunoi. his own destiny was not well-known to anybody Ne so no note solo aIcune (delle sue poesie). of-them are well-known only some (of his poems)

a. *1 Proprii amici non sono riconoscenti a nessunoi. his own friends are grateful to nobody b. Ne sono buoni pochi (dei suoi articoli). of-them are good few (of his articles)

This distinction is discussed in Bennis's contribution.

Introduction

2.

5

DETERMINING THE UNACCUSATIVEUNERGATIVE DISTINCTION

2.I.

Some diagnostics

A number of phenomena have been taken to be sensitive to unaccusativity. 1hese include (see Burzio 1986; Donna Jo Napoli 1988; Levin and Rappaport Hovav 1995, among others):

(a) auxiliary selection In most Romance and Germanic languages (English and Spanish being the exceptions) unaccusative verbs, such as French a1'1'ive1' (arrive), select BE, while unergatives, French rougir ('become red'), for example, select HAVE (see Haider and Rindler-Schjerve 1987; Perlmutter 1989; Cocchi 1994; Ackema 2000; Reuland 2000; Chierchia, in this volume; Randall, van Hout, Weisssenborn, and Baayen, in this volume; Sorace, in this volume). (14)

a.

b.

Marie est arrivee en retard. 'Marie arrived late.' Marie a rougi de honte. 'Marie became red with shame.'

(b) possibility to appear in resultative constructions A resultative phrase denotes the state achieved by the referent of the NP it is predicated of as a result of the action denoted by the verb. Resultative phrases may be predicated only of the object of a transitive verb, never of the subject. Intransitive verbs, then, divide into two groups: resultative phrases can appear with unaccusatives, but not with unergatives. 1his is exemplified in (15) (van Voorst 1985; Tsujimura 1994; Levin and Rappaport Hovav 1995, among others; see also Rappaport Hovav and Levin 200I):

(15)

a. She licked the peanut butter clean. b. *Dora shouted hoarse. c.

1he bottle broke open.

(c) prenominal perfect/passive participles Participles of transitive verbs can occur as attributive predicates of the nouns corresponding to their direct objects, as illustrated in (16a). Unergative verbs cannot be converted to such adjectival forms, as shown in (16b), but this is possible with unaccusative ones as shown in (16c) (see Williams 1981a; Hoekstra 1984; Levin and Rappaport Hovav 1986; Grewendorf 1989; Grimshaw 1990; Zaenen 1993, among many others; see Pesetsky 1995 for critical discussion of the validity of the diagnostic for English):

6 (16)

Artemis Alexiadou, Elena Anagnostopoulou, and Martin Everaert

der gekiH~te Student the kissed student b. *der gearbeitete Student the worked student c. der eingeschlafene Student the fallen asleep student

a.

(d) ne-cliticization In languages such as Italian, direct objects share the property that eliticization of a partitive phrase by the elitic ne is possible only with direct objects (17a). Certain intransitive verbs, namely unaccusatives, permit ne-eliticization (17b), while othersunergatives-do not (18) (see Belletti and Rizzi 1981; Lonzi 1985; Burzio 1986, among others): (17)

a.

b. (18)

Giovanni ne ha insultati due. John of them has insulted two Ne arrivano molti. of them arrive many *ne telefonano molti of them telephone many

For other languages similar tests, based on extraction, have been proposed: en-extraction in French (Legendre 1989) and the wat-voor/was-fUr split in Dutch/German (den Besten 1982). . (e) impersonalpassives Unaccusative verbs cannot be passivized. In contrast, unergative intransitive verbs allow the impersonal passive (Perlmutter 1978; Grewendorf 1989; Zaenen 1993): (19)

a.

Er werd hier door de jongelui veel gedanst. it was here by the young people a lot danced b. *Er werd door de kinderen in Amsterdam gebleven. it was by the children in Amsterdam remained

Under the assumption that the crucial characteristic of the passive is the absorption of the subject 8-role, it is clear that no such absorption is possible in the case of unaccusative verbs (19b)-while it is possible with unergatives (19a). It is well known that some of the tests listed above apply only to certain languages, or groups of languages. In the literature, several authors have attempted to provide lists of possible diagnostics for particular languages. To mention a few examples: for French, Legendre (1989), Ruwet (1991); for German, Fanselow (1985), Grewendorf (1989); for Dutch, Hoekstra (1984); for Russian, Neidle (1989), Pesetsky (1982); for Spanish, Torrego (1989); for Greek, Alexiadou and Anagnostopoulou (1997). To illustrate this, we list some examples of unaccusativity diagnostics that are less well known and widespread cross-linguistically.

Introduction

7

German allows so-called split phrases, i.e. phrases where the head and its satellites are separated: (20)

a.

b.

Er hat immer dreckige KIeider an. KIeideri hat er immer dreckige ei an. 'AI; for clothes, he always wears dirty ones.'

Grewendorf (1989) argues that split phrases are not allowed in subject position of transitives (2Ia) and unergatives (2Ib), but are allowed for unaccusatives (2IC, d) (21)

a. *Studenten haben fleiEige

das Seminar besucht. students have hard-working the seminar visited b. *Studenten haben fleiGige telefoniert. students have hard-working called c. Fehler sind dem Hans vermeidbare unterlaufen. mistakes are Hans avoidable occurred d. Widerspruche sind dem Richter mehrere aufgefallen. inconsistencies are the judge many stricken

It has been argued that in Russian, distributive po-phrases are limited to nonoblique VP-internal NPs, a claim which qualifies this construction as a syntactic diagnostic for unaccusativity. The distribution ofpo-phrases is illustrated in (22) (Pesetsky 1982; Schoorlemmer, in this volume); (22a) is an example of a transitive verb, (22b) of an unaccusative, and (22C) of an unergative: (22)

a.

Ja dal kazhdomu mal' chiku po jabloku. I gave every boy po apple.DAT 'I gave every boy a (different) apple.' b. Po jabloku upalo s kazhdogo dereva. po apple.DAT fell from every tree 'A (different) apple fell from every tree.' c. *V kazhdoj komnate smejalos' po devushke. in every room laughed po girl.DAT 'A (different) girl laughed in every room.'

For Georgian, an ergative language, Harris (1981) discusses several diagnostics for unaccusativity. One diagnostic is case marking. In one specific tense-aspect category, the II series, the case marking can be summarized as follows: Transitive Active intransitive (unergative) Inactive intransitive (unaccusative)

Subject -ma -ma

Direct Object -i

-i

In (23), transitives are verbs with final subjects and objects (to use the terminology of Relational Grammar). Harris argues that active intransitives-that is, unergativeshave an~argument that is initial and final subject, while inactive intransitives are verbs

Artemis Alexiadou, Elena Anagnostopoulou, and Martin Everaert

8

with an initial object and a final subject-that is, unaccusatives. This difference is reflected on the case marking of the single argument. Suppletion of verb roots, sensitive to the number feature of an argument, is another phenomenon where the unaccusative-unergative distinction in Georgian seems to surface. The verb 'kill' is an example: mOfvla is used for 'kill' with a singular object, daxoca with a plural object (24a, b); in the corresponding unaccusative constructions (24c, d), the same suppletion is found:

(24)

a.

b.

c.

d.

Mgel-i movlfali. wolf-NOM ISU:3DO.kill.u 'I killed the wolf' Mgl-eb-i davxoce. wolf-NOM ISU:3Do.kill.u 'I killed the wolves.' Mgel-i mol}Vda. wolf-NOM 3su.ldll.u 'The wolf died.' Mgl-eb-i daixoca. wolf-NOM 3SU.kill.u 'The wolves died.'

To summarize, unaccusativity diagnostics are not necessarily cross-linguistically valid. However, whatever the precise diagnostics are for a particular language, in essence they all rely on: (i) the absence of an external thematic role (impersonal passives)-as is the case in Relational Grammar, and the Principles-and-Parameters theory-or the absence of a thematic role for the subject function which is of an 'object type'-as is the case in Lexical Functional Grammar. (ii) a movement relation between object and subject position (ne-cliticization)-for those frameworks that allow for derivations. With respect to the latter issue, the notions deep and sUfface unaccusativity were introduced in the literature (Levin and Rappaport Hovav 1995; also Bresnan and Zaenen 1990). Diagnostics of surface unaccusativity apply only when the argument of an unaccusative verb remains in its deep-structure position, that is, the object position, such as ne-cliticization in Italian or split phrases in German. Diagnostics for deep unaccusativity are all those diagnostics where the surface position of the argument is irrelevant, such as impersonal passives, adjectival participles, auxiliary selection, and participation in the causative-inchoative alternation.

2.2.

Unaccusativity mismatches

A well-known puzzle surrounding unaccusativity is that unaccusativity diagnostics do not uniformly pick up the same class of verbs, both within and across languages. This

Introduction

9

leads to what has been referred to by the term 'unaccusativity mismatches' (see van Voorst 1985, L. Levin 1986, Eisenberg 1989, McClure 1990, Dowty 1991, Kathol 1991 for German, Zaenen 1993 for Dutch). Observe, for instance, the Dutch examples in (25).

(25)

a.

b. c.

de gevallen/*gewerkte/ /*gebleven/*gebloede jongen the fallen/ worked/ stayed/ bled boy De jongen is gevallen/*gewerkt/ /gebleven/gebloed. the boy is fallen/ worked/ stayed/ bled Er wordt *gevallen/gewerkt/ /*gebleven/*gebloed. there is fallen/ worked/ stayed/ bled

Examples (25a-c) clearly show that vallen ('fall') is an unaccusative verb and werken ('work') an un ergative: only vallen is allowed as a prenominal perfect participle; vallen takes BE, werken takes HAVE; only werken allows impersonal passivization. On the other hand, such verbs as blijven ('stay') and bloeden ('bleed') show mixed behaviour; they cannot form prenominal perfect participles, and should then be unergative. But they take BE and do not allow impersonal passivization, behaving on a par with unaccusatives. Cross-linguistically, the same indeterminacy characterizes several verb classes. For instance, in the domain of psychological predicates there seems to be no consistency across languages as to the behaviour of the Experiencer Object (EO) class with respect to the various unaccusativity diagnostics. For example, in Italian and Dutch, pleasetype predicates take auxiliary BE. On the other hand, in German, predicates such as gefollen 'please' select auxiliary HAVE and do not form attributive participles and impersonal passives (see Grewendorf 1989). Moreover, in French the selection of etre 'be' as perfect auxiliary is restricted to a narrow set of un accusative verbs (see Sorace, in this volume), and there are languages such as English, Spanish, and Greek, in which all intransitive predicates uniformly select auxiliary HAVE. Note, finally, that we can only truly speak of unaccusativity mismatches if we are certain that the diagnostics in question are clear-cut. & a matter of fact, for many diagnostics, perhaps all, this is not the case. Thus, in languages such as Spanish and Greek, unaccusative verbs allow bare plurals in post-verbal position (26b), while unergative verbs disallow them (26a) (see Torrego 1989 for Spanish, Alexiadou 1996 for Greek, also Alexiadou and Anagnostopoulou 1997).

(26)

a. *Epezan pedhia. were playing children 'Children were playing.' b. Irthan pedhia. came children 'Children came.'

However, bare plurals in post-verbal position become acceptable with unergative verbs when a locative adverbial phrase is added to the sentence in both Spanish and Greek, as illustrated in (27) for Greek:

10

(27)

Artemis Alexiadou, Elena Anagnostopoulou, and Martin Everaert Edo pezun pedja. here plaY.3PL children 'There are children playing here.'

Moreover, not all unaccusative predicates accept bare plurals, and judgements of native speakers vary as to the degree of grammaticality of such sentences. For instance, verbs of change of state related to adjectives or nouns are not entirely grammatical in Greek:

(28)

??Pagosan potamia. froze rivers 'Rivers froze.'

In Greek, bare plurals seem to identifY (it) verbs of appearance or existence, (b) break verbs-that is, verbs of change of state that are not related to adjectives or nouns (and, hence, they do not really qualifY as an unaccusativity diagnostic for this language).

3.

THE LEXICON-SYNTAX INTERFACE AND UNACCUSATIVITY

3.1. Levels of Representation Concentrating on the lexical semantics of a verb and the syntactic structures it can occur in, we can discern at least three different levels of representation for the present discussion: (i) a lexical-semantic representation, (ii) a lexical-syntactic representation, and (iii) a syntactic structure representation. The lexical-semantic representation of a predicate, often called lexical conceptual structure (LCS), is the 'deep' semantic description, which is probably unique for any particular predicate, or class of predicates. LCS decomposes the meaning of a verb into structures containing variables and meta-predicates (such as CAUSE, BE, etc.). Such a semantic description is mapped onto a more syntax-like representation, often called predicate-argument structure or argument structure (AS). It is widely assumed that the unaccusative-unergative distinction is encoded at this level. AS specifies how many arguments a verb requires and to which syntactic argument positions these are linked, for instance by making a distinction between external and internal theta roles (Williams I98Ia). It has been proposed to specifY further internal thematic roles (Marantz I984) or case properties of predicates (Belletti and Rizzi I988). The argument-structure representation is not unique for individual predicates or classes of predicates. Two different predicates, walk and sleep, for example, will probably have the same argument structure. This level of representation is then mapped onto a syntactic representation. Although essentially different, lexical conceptual structure and argument structure are part of the lexical representation of a predicate and thus part of the lexicon, which is distinguished from syntax. In other words, lexical semantic properties are directly reflected in argument structure, and the mapping from argument structure to syntax is, in most cases, trivial. Crucially;

Introduction

II

this entails that there is no direct relation between syntax and the lexical semantics of predicates, the LCS, but only between syntax and AS; this is sketched in (29): Lexical conceptual structure

Argument structure

Syntactic structure

Lexicon

..

Syntax

By and large, (29) represents the position on lexical representations tal<:en by such authors as Carrier and Randall (1993), Grimshaw (1990), Levin and Rappaport Hovav (1986,1988), Tenny (1987), and Zubizarreta (1987). In Jackendoff's work, and in the work of Van Valin (1990), for instance, LCS is defined in such a way that it incorporates all relevant aspects of AS. As such, lexical semantic notions are, in principle, accessible to syntax. This is also the position argued for in Levin and Rappaport Hovav (1995), who formulate mapping rules that directly relate syntax and lexical semantics to each other.

3.2. The Universal Alignment Hypothesis The debate on the unaccusative-unergative distinction can be summarized in terms of the following questions: (i) Is the class of unaccusative predicates semantically defined? This can be further decomposed into several questions. For example, is unaccusativity defined at LCS? Is there a separate level of AS and LCS? Are there semantic features guiding the distinction? (li) Is the unaccusative-unergative distinction syntactically encoded? (iii) What is the role of AS? These issues were first addressed in Relational Grammar. In Perlmutter and Postal's (1984) Universal Alignment Hypothesis, the answer to these questions was clearly formulated (without reference to the notions given in (29)):

(30)

There exist principles of universal grammar which predict the initial relation [= syntactic encoding], borne by each nominal in a given clause from the meaning of the clause.

In other words, unaccusativity can indeed be semantically defined and syntactically encoded. This is also the view adopted by Levin and Rappaport Hovav (1995). They extensively argue that the syntactic properties of intransitive verbs are semantically determined. They do so by isolating the semantic factors that decide whether a verb is unaccusative or unergative through a thorough examination of the behaviour of verbs that belong to a range of semantic classes in a number of syntactic constructions.

Artemis Alexiadou, Elena Anagnostopoulou, and Martin Everaert

12

In all approaches that take unaccusativity to be reflected in the syntax, an explanation for the correspondence between semantic properties and syntactic behaviour has been provided in terms of the postulation of certain linking principles, between, on the one hand, LCS and AS, and, on the other hand, between AS and syntax. The latter force the relevant argument of an unaccusative verb to be realized in direct object position, while requiring that of the unergative to be realized in subject position. Such proposals are based on principles like the Uniformity of Theta-Assignment Hypothesis (Baker 1988) or the Aspectual Intelface Hypothesis (Tenny 1992, 1994). Burzio (1986), a proponent of the syntactic approach, gives a positive answer to question OJ) above, and does not address question (i). Baker (1988) and Pesetsky (1995) follow the tradition of Relational Grammar and answer both questions positively. In non-derivational approaches-such as the ones advocated by Van Valin (1990), Dowty (1991), and Bresnan (20OI)-the answer has also been that there is a direct relation. That is, the syntactic behaviour of verbs in unaccusativity tests is a reflection of their semantics. On these views, though, the two classes of intransitive verbs can be differentiated on semantic grounds. In turn, the semantic characterization obviates the need to attribute different syntactic representations to the verb. Unaccusativity is not necessarily syntactically encoded.

3.3. Determining factors of unaccusativity Several authors have argued that (the lack of) agentivity is the crucial notion underlying unaccusativity (see Perlmutter 1978, Pinker 1989, Pustejovsky 1995, among others): the single argument of unergatives tends to be agentive, while the single argument of unaccusatives tends to be a theme or a patient or an undergoer. Perlmutter (1983) lists verb meanings that tend to belong to each class: 5

(31)

a.

b.

Generally unergative predicates i. Predicates describing willed or volitional acts e.g. work, play, speak, talk, smile, grimace, wink, walk, box, knock, bang, laugh, dance; manner-of-speaking verbs, e.g. whisper, shout, bellow; predicates describing sounds made by animals e.g. bark, quack, roar. ii. Certain involuntary bodily processes e.g. cough, sneeze, burp, sleep. Generally unaccusative predicates: i. Predicates expressed by adjectives in English; predicates describing size, shapes, weights, colours, smells. ii. Predicates whose initial nuclear term is semantically a patient, for example, burn,fall, drop, sink,jloat, tremble, shake, melt,jreeze, evap-

orate, solidifY, crystallize, dim, redden, darken. 5 Pesetsky (1995) correctly points out that the Universal Alignment Bypothesis does not commit one to the position that a specific thematic role is always linked to a specific grammatical configuration (which could be seen as the idea behind the Universal Theta-Assignment Hypothesis). However, Perlmutter's listing of verb meanings suggests that that position is what he has, in fact, in mind.

Introduction

13

iii. Predicates of existing or happening, such as exist, happen, occur, take

place. iv. Involuntary emission of stimuli, for example shine, glow, clink, pop,

smell, sting. v. Aspectual predicates: begin, start, stop, cease. According to other views (see Tenny 1987 and others), the crucial semantic factor is telicity: unaccusative verbs tend to be telic, while unergative ones tend to be atelic (see Borer, van Hout, in this volume). Based on the existence of unaccusativity mismatches (see section 2.2), Rosen (1984) and others have argued that meaning alone is not predictive of class membership, as there is no single semantic property common to all unaccusative verbs selected by all diagnostics in various languages. For instance, Rosen observes that there exists a class of verb meanings which may go either way depending on the language-for example, die is unergative in Choctaw but unaccusative in Italian. Levin and Rappaport Hovav (1995) argue that the semantic notions of activity and change of state, as well as the notion of internal and external causation, are to be facets of meaning that figure prominently in the determination of a verb's status with respect to the Unaccusativity Hypothesis. On this view, there is a correlation between the syntactic diagnostics and the semantic properties of intransitive verbs. Given that states of affairs can be conceptualized in different ways, one would expect that certain verbs, if they appear in unaccusative syntactic environments, impose accomplishment-achievement/external causation readings; when they occur in unergative syntactic environments they impose activity/internal causation readings (see also B. Levin 1993). So, on this view, there is a continuum between the two classes, and the mixed behaviour of certain predicates depends on the fact that they are compatible with both types of interpretation.

3+

The syntax: of unaccusativity

The developments in syntactic theorizing have resulted in a situation where the picture of the syntax of unaccusativity drawn in Burzio can no longer be maintained. Once VP-shells are introduced (Larson 1988), thematic roles are no longer forced to occur in unique positions, and the subject-object asymmetry is no longer expressed as a specifier-complement asymmetry. Moreover, with the introduction of elaborated functional structures (Ouhalla 1988; Pollock 1989) and the VP-internal subject hypothesis (Kitagawa 1986; Koopman and Sportiche 1991) the nature of A-movement has changed. In the early 1980s the single argument of unaccusatives was generated as a D-structure object (VP-internally) and moved to an A(rgument)-position in IP-the position where the subject of unergatives was generated. Nowadays, in both unergatives and unaccusatives the single argument is generated in the VP domain and moves to the functional domain, blurring the distinction between the Merger and the S(urface)-syntax mapping of the two. Since the specifier-complement distinction is

14

Artemis Alexiadou, Elena Anagnostopoulou, and Martin Everaert

also lost, a way to express structurally the difference between subjects of unergatives and subjects of unaccusatives is through the postulation of an asymmetry in terms of semi-functional heads, v/Voice, for instance (Kratzer 1994; Chomsky 1995). According to some theories in the 'light-v' framework, the subject of an unergative is introduced by a semi-functional head v, while the unaccusative argument belongs to the lexical verb, as depicted in (32). vP

(Unaccusative)

vP

(Unergative)

~

I

NP

v'

~ VP ~

v

V

b.

NP

v'

~ VP

v

I

V

This type of representation reopens the way for a syntax that directly reflects lexical decomposition (Dowty 1979; Marantz 1993; Kratzer 1994; von Stechow 1995). Such analyses are adopted in the chapters by, among others, Alexiadou and Anagnostopoulou, Bennis, and Embick. Note that assuming an approach along the lines of (32) has the result that some of the diagnostics, e.g. ne-cliticization-which crucially relied on the notion of government and the specifier-complement distinction-need to be reformulated. However, as pointed out in section 2, most of the other diagnostics were either poorly understood as to how they relate to unaccusativity (auxiliary selection, for example) or more easily described in terms oflexical primitives (the theme-externalization rule in adjectival passives formulated in Williams 1981a; the expression of agent/animacy in impersonal passives and -er nominals). Syntactic approaches to unaccusativity do not assume a clear-cut distinction between syntax and lexical semantics. In these approaches, semantics is not a primitive deriving unaccusativity; rather, verbal meaning is expressed via certain syntactic configurations. Such theories can be further divided into two sub-types. For one type of approach there is no D-structure representation distinct from a lexical semantic representation. For example, Dowty (1979), following the generative semantics literature, develops a lexical-decomposition analysis of verbs, according to which their semantics and syntax are indistinguishable. Hale and Keyser's (1993, 1998) configurational theta-theory (also assumed in Chomsky 1995), claims that unaccusativity depends on the way in which arguments combine with predicators that belong to different categories (A, V, P), thus experiencing different relations and having different combinatorial options. A different view, which also crucially relies on syntactic structure, is pursued in Borer (1994 and in this volume; see also van Hout 1996, in this volume). According to Borer, no hierarchical or thematic information is associated with arguments in verb entries. Arguments receive their interpretation by being in particular aspectual specifiers, which enforce subject-of-result or subject-of-process readings. This accounts for the different aspectual properties associated with each class. For un accusative and

Introduction unergative verbs the representation ofVP is identical; the verb is associated with a single unordered argument. It is often assumed that the semantics-based approach to unaccusativity is incompatible with the syntax-based approach. The former assumes that semantic features are primitives and that the syntax depends on a set of linking rules. The approach based on syntax claims that syntactic configurations are the primitives from which meaning can be deduced. There are reasons to believe that both approaches have their limitations (see So race's contribution for discussion and criticism of approaches relying on linking rules as well as purely syntactic ones). For example, a semantics-based approach fails to capture regularities and parametric differences in the lexicon-syntax mapping that crucially depend on the syntactic structure introduced by prepositions, verbs, adjectives, and nouns (see Hale and Keyser 1993, and, especially, 1997). On the other hand, a configurational theta theory has little to say on issues such as alternating vs. non-alternating unaccusatives, a difference that, as Levin and Rappaport Hovav (1995) convincingly argued, depends on the semantic notion of 'internal' vs. 'external' causation. Levin and Rappaport Hovav build on Chierchia's (in this volume) and Reinhart's (2000) view of alternating unaccusatives as being basically dyadically formed on the basis ofS-role reduction. The question that arises is whether a combination of the two approaches is feasible. What are the semantic features that pre-exist, how much is derivable from structure, and how do the two combine?

3.5. Acquisition The study of unaccusativity is one area where acquisition research has had a profound influence on syntactic research. It. is evident that unaccusative verbs pose a learning problem. In many contexts they are identical to unergative verbs in their surface syntax. Acquisition research has shown that children are sensitive to the distinction between unergatives and unaccusatives (van Hout et al. 1993; Snyder et al. 1995; van Hour 1996; Verrips 1998; Babyonyshev et al. 200I, among many others). The question that arises is how children distinguish between these two types of intransitive predicates, and what property guides them in this distinction. Influential theories of language acquisition crucially rely on the lexical semantics of verbs to help learners bootstrap their way into the preliminary projection of the syntax of argument structure (semantic bootstrapping; see in particular Grimshaw 1981; Pinker 1984, 1989, and subsequent literature). According to another view, early knowledge of the syntax of argument structure helps the child acquire the meaning of the verbs associated with that structure (syntactic bootstrapping; see Gleitman 1990). The acquisition of the unergative-unaccusative distinction is often explained in terms of semantic bootstrapping, thus arguing for a semantic or a linking-based approach to unaccusativity. More recently, a number of papers on second-language acquisition and the unaccusative-unergative distinction have appeared (Balcom 1997; Yuan 1999; Hirakawa 200I; Sorace and Somura 200I, and others). The relevant research questions revolve around the well-known distinction between universal developmental paths

16

Artemis Alexiadou, Elena Anagnostopoulou, and Martin Everaert

and transfer: can adult L2 learners acquire properties of the L2 lexicon that are not deducible from the input and that cannot be transferred from L1? Research on the acquisition of lexical alternations in a second language has shown that the problems encountered by L21earners in the initial stages, and the overgeneralizations they make, are consistent with those found in L1 acquisition, regardless of the native language. These issues are briefly addressed in Sorace's contribution.

Part II. The chapters A common view on the distinction between unaccusatives on the one hand and passives and middles on the other hand holds that the absence of distinct verbal morphology in unaccusatives correlates with the absence of an external thematic role in the syntax of unaccusatives, in contrast to passives and middles, in which such a correlation does not hold. Several chapters address the accuracy of the above statement. In what follows we offer a summary of the contributions which either directly address the issues discussed in Part I or explore the implications of the presence of special morphology for the structural representation of unaccusatives and the status of the external thematic role in them. Gennaro Chierchia proposes the adoption of an (independently motivated) theory of properties as the semantic algebra in terms of which truth conditions are recursively specified. The main characteristic of such a theory is that properties are taken to be primitives, linked to their arguments via predication. The latter is viewed as a mapping from properties into propositional functions. Thus, in a sense, predicates come in two forms-as properties and as propositional functions (with predication connecting them). Chierchia indicates how to use such a theory in compositionally interpreting fragments ofLF-that is, in mapping LF into 1£ The main assumption is that clausal structures come about via predication, which Chierchia takes to be associated with INFL-the Predication Principle. This principle, which plays a key role in his approach, is 'configurational' in the sense that it makes crucial use of syntactic configurations (much like the projection principle). One of the consequences of the Predication Principle is that verbs that are associated with propositional functions must have certain characteristics: they must take their argument within the VP and must be associated with an expletive subject. These are central features of unaccusatives. Chierchia further argues that unaccusatives should be regarded as instantiating a special type of reflexivization (see Reinhart and Siloni, Steinbach, and Embick, all in this volume). Such a form of reflexivization is also involved in the 'externalization' of the internal argument of unaccusatives via NP movement. This hypothesis accounts for (a) the unstable character of their valence; (b) the association of unaccusatives with reflexive morphology; (c) the control properties of da Sf phrases; and (d) the aspectual properties of unaccusatives. 6 Finally, Chierchia's semantics leads to a reformulation of 6 In his postscript, Chierchia addresses the question of whether an operation such as REFL can be regarded as the very operation that derives unaccusatives.

Introduction the aux-selection rule ofItaiian as a purely semantic domain condition on the meaning of essere 'be' that exploits the notion of subject-affected property. Such a reformulation may pave the way for a genuinely nondisjunctive account of the factors triggering essere selection (see also Sorace, in this volume). For Angeliek van Hout telicity is the defining property of unaccusativity. She argues that unaccusativity reflects the mapping configuration in which the verb's single argument moves through the specifier position of AgrO on its way to the specifier position of AgrS. Movement through AgrO is triggered when the predicate is telic, because, as two-argument verbs also show, AgrOP is the locus of telicity checking. Examining first telic-atelic alternations with two-argument verbs in Dutch, van Hout demonstrates that, in order to receive a telic interpretation, a direct object needs to be projected; an oblique object or an intransitive one do not trigger telicity. Compare the telic in vijfminuten een botherham eten 'eat a sandwich in five minutes' versus the atelic urenlang van de taart eten 'eat of the cake for hours' and urenlang eten 'eat for hours'. Particle verbs and prefixed variants thereof do not show this flexibility; they are obligatorily transitive (that is, they require a direct object). Compare een boterham opeten 'eat up a sandwich' versus *van de taart opeten 'eat up from the cake' and *Ik at op 'I ate up'. These verbs are inherently telic. The correlation between telicity and transitivity is explained in a lexicon-syntax mapping system based on semantic feature checking. Assuming that the event structure of a predicate must be identified (Grimshaw 1990), the telicity feature is checked in AgrOP (Borer 1994; van Hout 1996). The argument in Spec, AgrO is associated with the event participant that ends in a final state. Applying this mapping rule to single-argument verbs, telic single-argument predicates are predicted to be unaccusative-that is, the argument moves through AgrOP to check telicity before moving to subject position. And indeed they are unaccusative. On the basis of two unaccusative diagnostics in Dutch-auxiliary selection and prenominal perfect participles-the unergative-unaccusative split lines up exactly along the atelic-telic dimension. The same holds for flexible single-argument verbs-for example, atelic and unergative lopen 'walk' versus telic and unaccusative naar huis lopen 'walk home' and verb/particle verb pairs such as atelic branden 'burn' and telic verb randen 'burn up'. According to van Hour, there is no effect of any additional unaccusative determinant in Dutch. She concludes with a comparison of her analysis with other theories of unaccusativity, including aspectual ones. Hans Bennis starts from Chomsky's (1995) proposal concerning the representation of arguments in terms of a light v introducing the external argument. This allows a rather straightforward implementation ofBurzio's generalization. Assuming that v--not Vassigns accusative Case and introduces the External Argument (see also Kratzer 1994; cf. Hale and Keyser 1993), v-less verbal structures are unaccusative. Bennis argues that this account is only partially correct, and that all three possible syntactic configurations are found-namely, a v-less structure; v embedding a VP but not introducing an external argument; and a v embedding a VP and introducing an external argument. To substantiate this point, Bennis turns to the domain of adjectival projections (a-AP), where all three configurations are attested. Apart from adjectives with an external argument

18

Artemis Alexiadou, Elena Anagnostopoulou, and Martin Everaert

he argues that there are two further types of unaccusative adjectives: complex and simplex adjectives. The former include an a without an external argument, while the latter lack an a altogether. He furthermore extends this typology to the verbal domain. He argues that the three classes of psych predicates identified by Belletti and Rizzi (1988) correspond to the three possible configurations. Psychological verbs of the preoccupare class show the predicted properties of complex unaccusatives (they include a light v), while verbs of the piacere fall under simplex unaccusativity (they lack a light v) (cf. Alexiadou and Anagnostopoulou's contribution). Artemis Alexiadou and Elena Anagnostopoulou argue for a decomposition approach to predicates along the lines of Kratzer (1994) and von Stechow (1995), on the basis of the distribution of voice morphology associated with detransitivization in the causative-inchoative alternation. In Alexiadou and Anagnostopoulou's approach, unaccusativity is not linked to the presence vs. absence of v/Voice-contra the approaches mentioned in section 3.4, cf. (32). The investigation of (a) the consistency vs. gaps in the distribution of detransitivizing morphology in Greek and (b) the semantic and syntactic properties de-transitivizing marking correlates with, lead to a more refined structural representation of unaccusativity than is usually assumed. More specifically, Alexiadou and Anagnostopoulou argue that anticausatives do not have a unified structure (a related view is expressed in Borer 1991; cf. Bennis, in this volume) and that the patterns of voice morphology associated with the causative-anticausative alternation in Greek can be accounted for if there are at least three structures underlying the formation of anticausatives. Anticausatives are formed on the basis of an intransitive v BECOME/RESULT which embeds an Adjectivep, a Voicep, or a possessive construction. In his contribution, David Embick focuses primarily on morpho-syntactic properties of unaccusatives, passives, and reflexives. In particular, the nature of the morphological syncretism, labelled u-syncretism by Embick, is examined within the framework of Distributed Morphology. Embick advances the idea that the source of this pattern is a particular syntactic property, namely, the lack of an external argument (contra Reinhart and Siloni, in this volume). The syntax generates passives, reflexives, and unaccusatives, each fully specified for distinct features. U-syncretism results from the realization ofv in a particular structural environment, as represented in (33), where -X refers to the feature associated with non-active morphology. (33)

v H v- XI_ no external argument

By underspecifying -X, an answer is provided to the common morphological pattern of constructions with distinct syntactic and semantic properties. Tanya Reinhart and Tali Siloni offer a critical survey of the literature on the relation between reflexives and unaccusatives, and take a stand against un accusative analyses of reflexives (contra Embick, in this volume). Such analyses derive reflexives via a process of dethematizing the external argument, thus forming unaccusative-like entries (Marantz 1984; Grimshaw 1990; Pesetsky 1995, among others). Initially, Reinhart and Siloni question the empirical basis of such approaches, and ask if these approaches

Introduction

19

advance a lexical derivation (Marantz 1984; Grimshaw 1990) or a syntactic one (Pesetsky 1995). Reinhart and Siloni also supply evidence from Romance and Semitic languages to show that the grammatical subject of reflexives is the external argument, unlike the subject of unaccusatives. Following proposals by Chierchia (in this volume) and Reinhart (1996), they argue that both reflexives and unaccusatives are derived from the corresponding transitive entries by the same type of reduction operation, which operates on the internal argument to form reflexives and on the external argument to produce unaccusatives. Finally, they attribute the differences between French-type and Hebrew-type reflexives (for example, with regard to productivity) to the different component of the grammar (lexicon vs. syntax) in which the reflexive reduction operation takes place. Markus Steinbach focuses on the derivation of anticausatives in German, especially of reflexive anticausatives. He claims that such constructions are syntactically transitive reflexive sentences (TRS) and that they are multiply ambiguous in that, apart from the anticausative interpretation, TRSs can also yield a reflexive, a middle, and an inherent-reflexive reading. The main claim is that the four different types ofTRSs, do not differ syntactically, but only in their semantic/thematic interpretation. In constructions which include a reflexive pronoun, this is either interpreted as the second (internal) argument of the verb (reflexive) or it indicates a valency reduction operation. This operation either takes the form of saturation of the first or external argument (in middle constructions) or its reduction (in anticausatives and inherent reflexives; see Chierchia this volume). The (thematic) ambiguity ofTRSs is derived at the interface between syntax and semantics. The analysis is based on a modified version of a binding theory, relying on notions developed in Reinhart and Reuland (1993) and Pollard and Sag (1994) and on the (independently motivated) distinction between structural (nominative and accusative) and oblique (dative) case in German. On the basis of Russian, Maaike Schoorfemmer argues that unaccusativity must be syntactically encoded. Specifically, she discusses two syntactic configurations that are limited to transitive and unaccusative verbs, thus providing diagnostics for syntactic unaccusativity in Russian. The first diagnostic is the distributive po-phrase 'one X each', which can occur only in the object position of a transitive verb and occurs with unaccusative, but not with unergative verbs. The second diagnostic is telic aspectuality, which Schoorlemmer assumes to arise from the presence of an internal argument with particular quantificational properties. Since this type of telicity is also found with unaccusative verbs, the natural assumption is that the single argument of such verbs is in fact syntactically internal. The analysis of telic aspectuality is extended to derive imperfectivity, which emerges as a further syntactic unaccusativity diagnostic. The only way to generalize over properties of unaccusative verbs, as they show up, is to assume that they are syntactically encoded as properties of internal arguments. Antoneffa So race reviews some of the arguments for the Auxiliary Selection Hierarchy (ASH) and suggests some implications for theories of split intransitivity. The ASH indicates that within each of the two classes of intransitive verbs, some verbs require a given auxiliary more categorically than others. As Sorace discusses, not only

20

Artemis Alexiadou, Elena Anagnostopoulou, and Martin Everaert

auxiliary selection, but a number of syntactic manifestations of the unaccusativeunergative distinction in Italian and other languages are lexically constrained and tend to be acquired in a gradient fashion, starting with some verbs and gradually spreading to other verbs. The crucial components identified by the ASH are telicity and agentivity, which define a structured hierarchy of verb types based on universal aspectual relations. The extremes of the hierarchy consist of maximally distinct 'core' verbs, which display consistent unaccusative-unergative behaviour. In contrast, peripheral verb types closer to the centre are susceptible to syntactic variation. The mapping of the core verbs onto unaccusative or unergative syntax is largely invariant across languages, whereas intermediate mappings may vary cross-linguistically because different languages may have different cut-off points along the hierarchy. The ASH points to an analysis of split intransitivity in terms of gradient lexical-semantic effects on the syntactic realization of arguments, and thus firmly places the solution to the puzzle at the lexicon-syntax interface (contra Borer, in this volume, and van Hout, in this volume). She further investigates the predictions made for theories of first and second-language acquisition. Tonjes Veenstra explores the syntax of the resultative construction which has been viewed as an unaccusativity diagnostic. Focusing on Saramaccan, in which resultatives are expressed by means of serial-verb constructions, he argues for a unified analysis of the syntax of resultative constructions found in languages without serialization (such as the Germanic languages) and in languages with serialization. The analysis relies on a Larsonian VP-shell configuration in which the second VP is an adjunct to the first VP (see Larson 1990; Veenstra 1996a). Major parts of the discussion concentrate on the internal make-up of the second VP, addressing the following questions: Is the shared argument part of the second VP or not? Is it a full-blown VP? Is it a passivized VP? Is it an unaccusative VP? Veenstra argues that the shared argument is not part of the second VP, that it belongs to the first one. Moreover, he shows that the second VP is a full nonpassivized, non-unaccusative VP. He also defends the position that there is an aspectual projection inside the second VP in serial-verb constructions. Moreover, it is not the case that only transitive verbs can participate in resultative serial-verb constructions. The problem oflearnability of unaccusativity is addressed by Hagit Borer. She talces an argument-structure approach: syntactic as well as interpretational properties of arguments are independent of information stored in lexical entries (van Hout 1996, Borer 1994 among others). In Borer's model, both internal and external arguments are assigned their interpretation in the specifiers of functional, aktionsart, categories, and the verb associated with those structures is interpreted in accordance with its general conceptual meaning as augmented by these independently licensed arguments. In such a model, language learners could be assumed to pass through a developmental stage at which their knowledge of the structure and the interpretation of arguments is complete, yet their understanding of the lexical verb may remain flawed. Borer argues that precisely such a stage exists in the acquisition of Hebrew. Children go through a stage during which their argument structure is syntactically well-formed, but the verb associated with this argument structure is inappropriate from the adult perspective.

Introduction

21

Errors illustrating this developmental stage consist of the following types: 0) valency errors with attested root pairs; (ii) valency errors with unattested root pairs; and (iii) morpho-phonological innovations based on existing roots. These errors are explained if one assumes that the child has the computational means to construct the syntactico-semantic argument structure-so, understanding the meaning of the sentences in question-but has an incomplete knowledge of the way in which that syntactic structure determines the particular morpho-phonology of the verbal form. Rather, given an exposure to a particular root in some environment, the child forms a preliminary concept associated with that root. She then proceeds to randomize across the available, well-formed, morpho-phonological templates, having failed to acquire a full lexicalized knowledge of the relationship between a particular, often idiosyncratic, morphophonological instantiation of a particular root and the argument frame within which it may be used. Janet Randall, Angeliek van Hout, Jiirgen Weissenborn, and Harald Baayen examine unaccusatives in two closely related languages, Dutch and German, and investigate how they are acquired. A close look at the adult languages shows that Dutch and German distinguish the two classes in different places. On the assumption that a given unaccusative verb means the same thing in the two languages, the difference must lie in the rules that link the semantics of a verb to either an unergative or an unaccusative syntax. German appears to base its semantics-syntax linking on 'locomotion', while Dutch seems to employ 'endpoint'. If this is the case, then linking rules cannot be universal, but must be learned for each language. An experiment was designed to test the relevance of three factors that have been proposed to operate in determining intransitive verb class: (1) whether the participant is animate or not (Ernie vs. a block); (2) whether the verb is telic, implying that an endpoint is reached (Ernie disappeared vs. Ernie ran); and (3) for verbs that have an endpoint, whether this is encoded on the verb itself or added by a PP in which there is an additional syntactic clue that the verb encodes an endpoint (Ernie disappeared in the barn vs. Ernie ran into the barn). Coincidentally, verbs with endpoint PPs were locomotion verbs. Subjects participated in a sentence-completion task helping puppets 'learn' novel verbs for novel actions that were acted out by puppets in a series of videotaped scenes designed to test the three factors. The auxiliary verb that subjects chose in the sentence-completion task was used as the mark of verb class. The results showed that semantic factors appear to determine unaccusativity. For all subjects, verbs with clear endpoints tend to be classified as unaccusative. The presence of an animate actor also plays a significant role in signalling unergativity in all but the youngest Dutch subjects. When these two factors compete, however, endpoint appears to be more important, and the verb is considered unaccusative. German children behave like Dutch children initially, relying on endpoint rather than locomotion, although this is not the relevant distinction in their target language, and not the one the German adults used. Whether endpoint was carried in the PP or in the verb itself also made a significant difference, presumably because when it appears in a Pp, the endpoint is more readily detectable. The results provide ' evidence for two semantic universals in determining unaccusativity.

I A Semantics for Unaccusatives and its Syntactic Consequences Gennaro Chierchia

1.1.

INTRODUCTION

Since its original formulation, due in its current form to Perlmutter (1978), the unaccusative hypothesis has played a major role in linguistic theory, as it raises fundamental questions on the nature of grammatical relations and on the syntax-semantics map. The central idea at the basis of the Unaccusative Hypothesis is that the surface subject of a class of intransitives really is, at some level, an object. In what follows, I will outline an explicit (truth-conditional) semantics for unaccusatives and explore its consequences for syntax. I will try to show that the semantics represented below takes us some steps forward in our understanding of unaccusativity. My inquiry will be limited largely to Italian and the question of whether what I propose extends to other languages will have to be left for further research. At the same time, the advantage of focusing on Italian is that such a language offers an extremely rich and well-charted territory for the study of unaccusativity, as attested by the extensive and insightful work of Rosen (1981), Burzio (1986), and others. This paper was originally written in 1989 and it has been circulated as a manuscript (from Cornell University) for many years. It is published here for the first time in its original form. I have added a postscript to this chapter, especially in reaction to Reinhart (1996).

A Semantics for Unaccusatives The chapter is organized as follows. I will take the moves from Burzio's proposal concerning the syntactic structures associated with unaccusatives and try to provide a semantics for such structures. I will argue that this semantics sheds light on phenomena previously unaccounted for (pertaining to the interaction of unaccusatives and reflexives and to aspectual properties of unaccusatives) and leads, furthermore, to a new formulation of the aux-selection process in Italian, superior to the options currently available.

I.2.

BACKGROUND

Syntactic preliminaries

I.2.I.

Within the Principles-and-Parameters framework, the Un accusative Hypothesis is implemented roughly as follows. A verb comes with a list of8-roles (its arguments), one of which is generally realized outside of the maximal projection of the verb (cf. Williams 1983). So, for example, in (I) you see a sketch of the type of information that a verb entry contains.

a.

b. c.

piangere: eI (agent) distruggere: eI, 82 (agent, theme) dare: eI, 82, 83 (agent, theme, goal)

'cry' 'destroy' 'give'

In (I), for each verb it is indicated how many 8-roles it assigns and what their (rough) semantic content is. The external argument is in italics. Some verbs, such as seem, lack an external argument. This feature is generally associated by principles of grammar with the possibility for such verbs to have expletive subjects and to undergo raising. Un accusatives are taken to have an argument structure like the one of seem, in that they lack an external argument. This is exemplified in (2).

a. b. c.

venire: 81 (theme) affondare: (eI), 82 (agent, theme) mangiare: eI, (82) (agent, theme)

'come' 'sink' , , eat

Venire in (2a) takes its only argument internally, in contrast with the unergative piangere in (ra). In (2b, c), where parentheses indicate 'optional' arguments,' one can see two widespread types of transitive-intransitive alternants. In affondare, the external 8-role is 'optional', thus the intransitive member of the pair will be unaccusative. In mangiare, the internal 8-role is optional, thus the intransitive member of the pair will be unergative. From these assumptions, it follows that the underlying structure associated with these verbs is as shown in (3).

, Note the quotes. The notion of 'optionality' rdevant here will be discussed shortly.

Gennaro Chierchia

24

a.

b.

IP

~

NP

IP

~

I'

~

~

I Gianni

'Gianni eats.'

I'

NP

I

VP

mangia

VP

~

I

e

V

NP

I

I

affonda Gianni 'Gianni sinks.'

The NP Gianni occupies the canonical subject position in (3a) and the object position in (3b). Structures such as (3b), just like raising structures, are associated with either the occurrence of a phonologically null expletive in subject position or with movement into such position of the post-verbal NP. This is basically the line tal<:en in Burzio'swork One question that immediately arises in this connection concerns the nature of the alternations in (2b, c). The notion of 'optional argument' is poorly understood at present. More specifically, the projection principle requires that the a-structure of a predicate be projected at every level of representation. If we were to allow 'optional arguments' freely, the predictive power of the principle would be seriously undermined. It follows, then, that if we want the projection principle to have bite, it is preferable to assume that the members of each pair in the alternations in (2b, c) are related to one another by means of some operation on a-roles. Either one derives the transitive member of the pair from the intransitive one by an operation that adds a role, or one derives the intransitive from the transitive by an operation that removes a a-role. The question becomes what the exact nature of these operations is. Various hypotheses on the syntactic side of this questions have been formulated (see, for example, Burzio 1986: ch. 1, or Fagan 1988) but, as far as I know, the issue of their truth-conditional import has never been explicitly addressed. The latter issue is, however, central to any attempt to gain proper understanding of the semantics of the structures in question. In order to address such an issue (and, more generally, the issue of the semantics of unaccusatives), we will have to mal<:e some general assumptions concerning how truth conditions are associated with syntactic structures. The assumptions we will make are quite minimal, perhaps simplistic. But they will do for our present purposes. I will adopt the customary view that L(ogical) F(orm) is the level of syntax that feeds into truth-conditional interpretation. Let us assume, for explicitness, that truthconditional interpretation takes the form of a map from LF onto I(ogical) f(orm): a logic for which truth and entailment are known quantities. The map from LF onto If is, as far as possible, compositional: one derives the meaning of a complex structure by assigning first a meaning (= If) to its constituents. These meanings are then composed using an extremely restricted range of semantic operations, such as function application, abstraction, function composition, and perhaps some type shifting principles.

A Semantics for Unaccusatives

25

The next issue, then, is: which logic will be best suited for this task? Clearly, the choices one makes in this connection will channel truth conditions in certain ways rather than others, and will have, thus, far-reaching empirical consequences, just like the choice of primitives in syntax does. So the scope of this question exceeds by far what we can address here. But we must at least consider those aspects of the question that will be of direct relevance to our goal. To this task the following section is devoted.

1.2.2.

Semantic preliminaries: property theory

I suggest that we adopt an intensional logic where properties and propositions are taken as primitives. The logic in question is going to be a simple first-order language in which individual variables and constants are of three sorts or types: the type of basic entities, the type of properties, and the type of propositions. This means that the domain in terms of which our logic is interpreted is divided into at least three subdomains: a domain of basic entities (people, chairs, etc.) a domain of properties (running, loving Mary, etc.), and a domain of propositions (john run,fohn love Mary, etc.). Properties and propositions are regarded just as a special sort of individuals. The logical behaviour of properties and propositions is spelled out in terms of a set of axioms, whose formulation need not concern us here. 2 If you like, you can think of properties as actual regularities occurring in the world. For example, running is what is common to every situation where running goes on. Similarly, you can think of propositions as types of eventualities. For example, John run is what is common to all the situations where John runs. (I suggest this not because it is of much importance here, but simply because attaching some intuitive content to these notions might help in grasping how the system works.) What I just said is summarized in the diagram in (4). I use e (entities) for the type of individuals, p for the type of propositions, IT for the type of properties, and u (urelements) for the type of ordinary individuals.

u

p 7t

The main characteristic of properties is that they can be predicated of other entities. In order to do so, given that properties are individuals, we need a relation of some kind to connect them to their arguments. Such a relation will be a predication relation. The result of predicating a property of an individual via the predication relation is a proposition. Let us exemplify.

2

See e.g. Chierchia and Turner (1988).

26

Gennaro Chierchia

(5)

If r is a property and u an individual (of any sort) and U is the predication relation, then Ur(u) is the proposition that u has property r.3

Now, what we just did can be seen in a slightly different perspective. Let (a, b) be the type of functions from as into ~s. Thus, in particular, if e is the type of individuals and p the type of propositions, (e, p) would be the type of functions from entities into propositions (propositional functions). We can view predication, i.e. 'u, ('up') as mapping properties (entities of type 1t) onto propositional functions. I.e. we can say that if r is a property, ur is a propositional function, which means that ur applied to an individual yields a proposition. All this is schematized in (6).

(6)

a.

b. c. d.

(a, b) = the type of functions from as into ~s (e, p) = function from entities into propositions (propositional functions) u ('up'): 1t ~ (e, p); uris of type (e, p) Example: urunQ) =John runs

So, for example, we can express the fact that an individual say John, has a certain property, say running, as in (6d). To summarize, predication can be viewed as creating (for each property) a map from individuals onto propositions. That is, we can regard it as associating with each property a propositional function as indicated in (6). The predication map will have an inverse-that is, a function that turns propositional functions (back) into properties. This map can be viewed as a way of 'nominalizing' propositional functions. I use for it the symbol in (7a) and assume that nominalization will satisfY the condition in (7b). (7)

a. b.

('down'): (e, p) ~ 1t nu r = r

n

Accordingly, the diagram in (4) can be completed as shown in (8).

(8)

3 The rationale behind the notation is the following. Predication is a special relation analogous to set membership. In a pure relational notation, where all relations are prefixed, we might write:

(i)

U(run, j)

'running and John stand in the imtantiate relation' or 'John instantiates running'

Now, if relations are thought of as function valued functions (:.. la Schoenfinkel-Montague-Curry), then 'U' becomes a function from properties into a function from individuals into a proposition (of type

(ll,(e,p»): (ii)

U(run) (j)

By omitting the first set of parentheses and writing ,urun' for 'U(run)' we obtain what we have in the text. This notation is meant to underscore the analogy between predication and verb inflection (which typically attaches to the verb stem).

A Semantics for Unaccusatives

27

The diagram in (8) illustrates the way in which the semantic domains we are going to need to interpret English are classified. In Chierchia (1985) and Chierchia and Turner (1988), I have tried to show that a system of this sort plays an important role in characterizing the notion of 'finite clause'. I will argue here that it also can playa key role in characterizing the semantics of unaccusatives. We have discussed so far properties and propositions. What about relations? Well, a simple (though not the only) way of representing them in the present system is as indicated in (9). We can represent a two-place relation as a function from entities into properties. Something like John loves Mary will, thus, be derived as shown in (9a). Similarly for three-place relations. (9)

a.

Two-place relations: (e, n). Example: u [love(m)](j) = John loves Mary

/-----

u [love(m)] =loves Mary, type (e, p)

j =~hn, type: e

I

love(m) = love Mary, type n

~

love, type: (e, n)

b.

m = Mary, type: e

Three-place relations: (e, (e, U

n». Example:

[give(s)(w&p)](j) = John gives War and Peace to Sue

/

U

[give(m) (w&p)] = give W&P to Sue, type (e, p)

-

j = John, type: e

I

give(s) (w&p) = give W&P to Sue, type n

~

give(s), type: (e, n)

w&p =War and Peace, type: e

~

give, type: (e,(e,

n»

s =Sue, type: e

Looking at (9b) for example, if one feeds an argument, say Sue, into give, one obtains give Sue, a function from entities into properties-that is, a two-place relation. If one further feeds into that another argument, say ~r and Peace, one obtains a property, the property of giving War and Peace to Sue. Predication can apply to this, turning it into a propositional function, which then applies to John, yieldingJohn gives ~r and Peace to Sue. This illustrates how an n-place relation can be represented as a (curried) n-l property-valued function. I will still refer to property-valued functions as 'n-place relations', as it is less cumbersome and mnemonically more useful. By the same token, given a relation like love in (9a) or give in (9b), I will refer to the argument that is fed last (j in (9a-b)) as the last argument of the relation, to the one fed in immediately before it, as the second to the last (m in (9a), w&p in (9b)), and so on.

Gennaro Chierchia The system that I have just sketched has, I believe, several advantages over more standard systems, such as Montague's. For one thing, it embodies a more intensional view of properties and propositions. By treating them as basic, we can leave open their identity conditions, which means that two logically equivalent propositions will not necessarily be identical. In Montague's system, in which propositions are identified with sets of worlds, logically equivalent propositions turn out to be the same object, which leads to well-known problems in connection with propositional attitudes. But quite independently of this feature, the following aspect of the system just sketched will playa crucial role for our purposes. Predicates playa double role. On the one hand, they are just individuals, perhaps aspects of the world (entities of type n). On the other hand, predication turns them into unsaturated, argument-taking functions (of type <e, p»). This double role will enable us to give semantic content to the notions of internal vs. external predication, which seems to be at the heart of unaccusativity. We are now in a position to see how a property theory of the type just sketched can be used in interpreting LF in general, and unaccusative structures in particular.

I.3. 1.3.1.

INTERPRETING LF

Relation changing operations

Let us address now the question of the semantics of operations on 8-roles. As pointed out above, in the Principles-and-Parameters framework, relation-changing processes involve operations on 8-roles. For example, detransitivization-the alternation in (2C)-can be cast as an operation that lexically saturates the object 8-role (see, for example, Rizzi 1986a); passive is generally taken to involve an operation that saturates or absorbs the subject 8-role. And so on. What one would like to determine is which operations on meanings such operations on 8-roles are associated with. This question has been studied fairly extensively in the semantic literature (see Dowty 1982 or Keenan and Faltz 1985, for example). The basic approach emerging from this line of research is that there are a limited number of ways in which the logical structure of a relation or property can be modified. So operations on 8-roles can be taken to correspond to algebraic operations on relations. It is perhaps appropriate in this connection to present some of the standard operations discussed in the literature, adapting them to the present semantic system. In each instance, I consider only the simplest possible case. Generalizing them to more complex cases presents no problem, as far as I can see. Let us begin with passive. One can interpret passive as a map from relations onto properties, which fills in the slot that corresponds to the subject (i.e. the last slot) with a variable and closes it off existentially. Such an operation is spelled out in (IO). The logical type of the operation is given in (IOi), its truth-conditional import in (IOii).

A Semantics for Unaccusatives (10)

29

Passive (P) i. Logical type ofp: (e, rc) ~ rc ii. Content: U[P(see)] (x) H 3y U [see(x)] (y) iii. Closest Montague-style equivalent: A x3y [see (x) (y)]

In (lOiii) I provide for comparison the closest Montague-style version of passivization. An operation of this sort appears to provide a simple way of giving truth-conditional content to the claim that passive morphology absorbs or saturates the subject 8-role. Notice, in particular, that the suppressed argument is still semantically there, which makes it possible for it to manifest itself in certain cases. For example, as often noted, such an argument can control infinitival adjuncts, as in (n). (n)

The boat was sunk [PRO to collect insurance].

The implicit agent appears to be the understood controller of the infinitival clause in (n). This is possible because the agent, even if not syntactically projected, is present in semantic structure (in the form of an existentially quantified variable) and thus accessible for this kind of control (but not, say, for controlling reflexives). A structurally similar operation is the one arguably involved in what used to be called 'indefinite object deletion-that is, alternations of the kind given in (2C). This is illustrated in what follows.

(I2)

Detransitivization (D) i. Logical type ofD: (e, IT) ~ IT ii. Content. U[D(eat)](x) H 3y U [eat(y)](x) iii. Closest Montague-style equivalent: A x3y [see (y) (x)]

The operation D in (I2) corresponds to an operation which saturates the object 8-role. Note that the only difference between (I2) and (10) is the slot where the existentially quantified variable is inserted. In (n) it is the last slot (the one corresponding to the subject), in (12) it is the second to the last (the one corresponding to the object). So (12) is an object-affecting operation, while (II) is a subject-affecting one. A further important operation is reflexivization:

(13)

Reflexivization (R) i. Logical type ofR: (e, IT) ~ IT ii. Content: U[R(wash)](x) H U [wash(x)](x) iii. Closest Montague-style equivalent: Ax [wash (x) (x)]

This operation identifies the two arguments of a relation, thereby reducing it to a property. In this sense, it is both a subject- and an object-affecting operation. In the lexicon this operation is presumably involved in the interpretation of forms of reflexivization such as self-prefixing in English, or the alternation between transitive and intransitive shave. In the syntax, reflexive clities are obvious candidates for applications of such an operation. There are many aspects of the behaviour of reflexive clitics that mesh well with the idea that they are semantically associated with R. For example,

30

Gennaro Chierchia

reflexive elitics do not display NP behaviour; they are thus unable to conjoin with other NPs, which makes sense in view of the fact that R is not of the same type as NPs. Furthermore, non-arguments (such as VP-adjuncts) are typically unable to eliticize. This follows on the present view from the fact that adjuncts are properijt-modifiers (in our framework, functions of type (n, n» while reflexive elitics need relations to operate on. Thus the hypothesis that reflexive elitics are semantically associated with the relation reducer R seems well motivated. By the same token, R by itself cannot be what is involved in the interpretation of non-eli tic reflexives, as English himselfor Italian se stesso. Nonelitic reflexives are fullfledged NPs-for example, they can be conjoined with other NPs as in (14).

(14)

John likes only himself and Bill.

It is elear that one cannot interpret (14) by means of an argument-reducing operation such as the one in (13). What has been proposed in this connection is that such nonelitic reflexives as himselfare associated with two things: an ordinary pronoun meaning and an operator that encodes an instruction to bind the pronoun meaning to a suitable antecedent (subject to a domain condition, such as principle A of the Binding Theory). An explicit proposal along these lines can be found, for example, in Bach and Partee (1980) (cf. (56) for an illustration). As nonelitic reflexives are associated with an ordinary pronoun meaning, nothing will prevent them from conjoining with other NPs. Nor will they be prevented from occurring in adjunct position, to the extent that they can find a local antecedent in the relevant domain. To summarize, we have discussed a number of semantic operations on relations that can be viewed as the semantic counterparts of operations on 8-roles. The central question that now arises is: which of these operations, if any, is involved in unaccusative alternations such as the one between transitive and intransitive sink? Before we can address this question, however, we need to be more explicit on how these operations are used in interpreting LF. I.3.2.

The map from LF onto If

The only substantive assumption that I will be making on the LF-lf map is the following. Just as elauses arise by syntactically predicating a VP of a subject, the semantic value of elauses, namely propositions, arise by semantically predicating a property of an individual. I will execute this by assuming that the predicator ('u') is associated with the inflectional head 1°. In other words, 'U' is what 1° 'means'. I will call this the predication principle. This principle is very much in the same spirit as the projection principle: it requires that the map from syntax onto meaning is, in some sense, transparent. Perhaps, the predication principle could subsume the extended projection principle, along lines similar to those explored in Rothstein (1983). The predication relation takes two arguments: an individual and a property. If argument structure must be syntactically projected and if predication is realized as JD, it follows that [NP, IP] must be syntactically projected.

A Semantics for Unaccusatives

31

Be that as it may, let me exemplifY how one might envisage, on the basis just sketched, the iF-If map. Consider a simple structure such as the one in (15):

(15)

IP, U[love(m)](j)

~

Np, j

If, U [love(m)]

~

VP, love(m)

I, U

~

V, love

John

Np,m

I

I

Mary

love

In (15), I indicate next to each node its If representation. As pointed out above, I assume that in general the iF-If map is compositional, that is, it proceeds bottom-up and the translation of the mother is determined by the translation of its daughters and a restricted number of semantic operations (such as function application or composition).4 In (15) we use function application all the way through. Now, there are several ways in which this interpretive procedure is simplistic. For one thing, it might be the case that the interpretive procedure has to access information contained at other levels (such as D-structure or S-structure). Moreover, it has been argued (Higginbotham 1985 and Parsons 1988, for instance) that verbs have an extra 'Davidsonian' argument slot for events. These are issues I ignore here only for simplicity. The crucial factor, from the present perspective, is the predication principle, that is, the claim that a proposition comes about only by using the predication relation to connect a property (an entity of a certain sort) with an argument, no matter how this is executed. With these provisos in mind, let us move on to consider how passive works.

(16)

Ip, U [P(hit)](j)

~

NP, j

If, U [P(hit)]

~

VP, hit

I, uop

~

Johnj

V, hit

I

was hit

NP, Id (= Aaa)

I

ei

In (16) passive morphology is interpreted as the composition uop (= AR U[P(R)]) of the predication operation 'u, with the passivization operation P; syntactically passive morphology saturates the subject 8-role (see, for example, Baker 1988); its semantic counterpart P closes off existentially the argument slot corresponding to such role. NP 4

We follow on this score the Montagovian tradition.

Gennaro Chierchia

32

traces (unlikeA'-bound ones) are taken to be semantically inert, and interpreted as the identity map. So the meaning of the VP is obtained by applying the identity map associated with the trace to the semantic value of the V. This has the effect that the meaning of the V is passed up. Then P and predication apply to it and the result applies, in turn, to the subject. Again, I am leaving out many details that, though important for other purposes, are irrelevant to the task at hand.

1.3.3. Expletives and predication An interesting problem for the predication principle arises in connection with verbs like seem. There are expressions that are syntactically VPs and yet semantically they are clearly propositions, like seems that John was here. The problem is illustrated in (17): (17)

IP

~

NP

I', U [seem(p)] = UNDEFINED

~

UVP, seem(p) [type of seem: (p,p)]

I,U e

~

V, seem

I

seem

Np, p

I

that p

The issue here is that seem appears to be taldng a proposition as its (only) argument. In other words, seem is of type (p, p) and tal<:es its propositional argument internally to its maximal projection. Once the propositional argument of seem is provided, we get a proposition. And propositions are the wrong sort of objects to be predicated of anything: they are complete structures. Thus the predication operator is undefined for them. What ways out do we have? Two types of route are open to us: either we give up the idea that clausal structures universally involve predication in the semantic sense-that is, we give up the predication principle-or, if the predication principle is right, we must somehow turn seems that p into a property-that is, something that can take an argument. Let us pursue this strategy, to see what it involves. Imagine defining a function of the following sort: (18)

Expletivization (E) i. Logical type ofE: (p, n) ii. Semantic content: U[E(seem(p))] (x) =

seem(p), ifx=..L { UNDEFINED, ifx .. ..L.

Here is the idea. Expletivization applies to a proposition and turns it into a property. E(seem(p)) is a property that predicated of an arbitrarily chosen funny object (which I notate as '.1') yields the proposition seem(p). If E(seem(p)) is predicated of anything else but the funny object, the result is undefined. The idea is not new. It is just a variant

A Semantics for Unaccusatives

33

of a standard Montaguesque strategy for dealing with expletives. Notice, incidentally, that E, as defined, is a subject-affecting operation: it adds a sort of truth-conditionally inert subject to a proposition (Le. a o-place relation). In this sense, it falls together with passive and reflexivization, which are subject-affecting too. By means ofE, we can rescue the structure in (17) as follows:

(19)

IP, U [E(seem(p))](.l) = seem(p)

~

NP, .1

If, U [E(seem(p))]

~

VP, seem(p)

I,U it

~

V, seem

I

seem

NP, p

I

that p

E is to be thought of as a type-shifting operation that is called upon when simple function application will not work (as is generally the case for type shifting).5 In structures such as (19), at the VP level we are stuck with an object that predication cannot apply to. Hence we resort to expletivization that turns seem(p) into a property. This property can only apply to an argument that does not affect truth conditions (namely, an expletive). This amounts to pushing the Predication Principle as far as possible. If the principle is right, then positing something like expletivization is the only way to cope with the existence of verbs like seem. If we do not want to resort to expletivization, we will have to resort to a different interpretive procedure for 'normal' cases (such as (15)-(16)) vs. cases like seem. As I do not find this particularly desirable, I will stick to the Predication Principle. There is a further role that expletivization might play. We have assumed that passive absorbs the 8-role of the subject-that is, it existentially quantifies over the subject slot-and externalizes the argument corresponding to the object. Imagine, now, an operation that, instead, simply saturates with an existential quantifier the subject slot, without externalizing anything. Such an operation would have a semantics of the following kind:

(20)

I-Passive L Logical type ofP]: (e, rr) --? (e, p) ii. Semantic content: P](help) (x) dy U [help (x) ](y) (i.e. 'x is helped')

Truth-conditionally, P and p] are just alike. The only difference between them is one of logical type: P yields properties as output, p] yields propositional functions. Since it does not call for externalization of an internal argument, nothing prevents 5 Cf. Partee and Rooth (1983) for a general discussion of type-shifting principles. My use of E follows closely the spirit of their proposal.

Gennaro Chierchia

34

generalizing it to intransives. That is, one can assume that e in (20i) is optional. In the case of intransitives, PI would work as follows: (21)

a. b.

PI (dance) =::ly U [dance] (y) (I.e. 'it is danced') Generalized type ofpI: «(e,) n) -7 «(e,) p)

In plain words, Pr is impersonal passive, an operation that, unlike regular passive, fails to externalize the internal argument if there is one. But this means that whenever we have an impersonal passive we will wind up with something propositional at the VP level, as in the case of seem. This will force us to resort to expletivization, which will create the external expletive argument that predication needs. In (22) I illustrate the whole process with a German example. (22)

a.

Es wird ihm geholfen it is him (dat.) helped

b.

IP, U [E(Pr (help (x) ))](..1) =::ly U [help(x)](y)

~

Np, ..1

I', U [E(Pr (help(x)))]

~

VP, help (x)

I, U

~

es

Np, x wird

I

ihm

V, help

I

geholfen

In (22b), helfen is passivized via PI' This means that when we apply the verb to its argument, after impersonal passivization, we get a proposition at the VP level. Consequently, just as with seem, we must resort to expletivization to avoid ill-formedness. Thus, assuming expletivization enables us to capture the difference between personal and impersonal passive in a simple and semantically explicit way. We have essentially two variants of the same operation, which differ as to whether a property or a propositional function is created. If the latter, our theory of predication will require resorting to expletivization. Several aspects of the syntax of impersonal passives remain of course open (pertaining, for instance, to case theory). But however they are settled, it is conceivable that the relevant semantics should go along the lines proposed here.

1.3+ Expletivization and unaccusatives It should be clear by now what our basic hypothesis concerning the semantics of unaccusatives will be. As they are like seem (and, perhaps, impersonal passives) in lacking an external argument, their logical type must be as shown in (23) and expletivization will be involved in their interpretation. , . , a. arrivare, 8r, (e, p) arnve b. camminare, 8r, n 'walk'

A Semantics for Unaccusatives (24)

35

IP, U [E(arrive(g))](.l) = arrive(g)

~

Np,.l

I', U [E(arrive(g))]

~

VP, arrive(g)

1, U

~

V, arrive

e

I

Np, g

I

arriva

Gianni

Unaccusatives differ from unergatives just as propositional functions differ from properties. It then follows from the Predication Principle that unaccusatives are prevented from taking their argument externally. Ifwe try to generate a structure such as (25), in which the argument of arrivare is directly generate externally, the derivation will not go through:

(25)

IP

~

NP, g

I',

U

[arrive] =UNDEFINED

~

I, U Gianni

VP, arrive (type: (e,p»)

I

arriva

When we try to apply predication to arrivarewe get something ill formed, for 'U' is only defined for properties. And this structure cannot be rescued by any of the known strategies. This forces arrivare to take its argument internally; which in turn forces one to resort to expletivization to avoid getting stuck with a proposition at the wrong level. Let me try to summarize so far. I have sketched a way to interpret syntactic structures of the kind familiar from work in the Principles-and-Parameters framework using a property theory that crucially distinguishes properties from propositional functions and that views predication as a map from the former onto the latter. The central aspect of the syntax-semantics map that I have proposed is that a clause relates a predicate to a subject via a predication relation, not only in the syntax but also in the semantics (the Predication Principle). This entails that VPs must be associated with the kind of things that can be predicated, namely properties. But it does not exclude that certain verbs might be associated with propositional functions. Given, however, the Predication Principle, these verbs turn out to be, in a sense, 'defective'. They are predicted to have the following features: their arguments must be fed in internally and they will have an expletive subject (via expletivization). So why are there verbs like seem? Why are there unaccusatives? Or, equivalently, why may verbs lack an external argument, while, at the same time, the subject position must always be projected? It seems to me that our theory of properties and its encoding in syntactic structure (the Predication Principle) offers the beginning of an answer

Gennaro Chierchia to these questions. On the proposed approach, predicates exist in two related forms: as properties and as propositional functions, with predication relating the two. This leads one to expect verbs to be associated with properties (and, in the case of trans itives, with property-valued functions). But to rule out verbs directly associated with propositional functions we would have to pass a law. Such verbs must have unaccusative characteristics. These assumptions (if one buys property theory) are rather minimal. This makes me hope that extensions and refinements of the LF-lf map will leave them unaffected. There are a good deal of questions which the above considerations leave unsolved and which must be answered for the present line of inquiry to be viable. For example, we still have to say something on the nature of unaccusative transitive-intransitive alternations. Furthermore, in unaccusative structures, the internal argument can be externalized via NP-movement. How are the resulting structures to be interpreted? I believe that the answer to these two questions is, perhaps surprisingly, related. To this issue the following section is devoted.

1.4. 1.4.1.

UNACCUSATIVES AS REFLEXIVES

On the nature of unaccusative alternations

fu pointed out, it is typical of unaccusatives to enter alternations of the type in (26).

(26)

a.

b.

Gianni ha affondato la barca. the boat.' 'Gianni sank La barca eaffondata. 'The boat sank.'

The question that arises here is how the transitive and intransitive members of the pair in (26) are related to one another. The projection principle requires that members of transitive-intransitive pairs of this sort be related by some admissible operation on Sroles. 1£ however, we start by talting the intransitive member of the pair as basic and derive the other via, say, causativization, it would still remain to be explained why the argument of intransitive ajfondare should be an internal one. On the other hand, if we take the transitive member of the pair as basic, we might have a better chance at explaining why the argument left behind is internal. Imagine, for instance, that intransitive ajfondare is derived from the transitive via an operation such as passive. We have seen that passive has the option ofleaving an internal argument behind (c£ (20)). If this independently available option is selected, the unaccusative behaviour of the intransitive member of the pair in (26) would follow. While a story along these lines might be on the right track, there are various considerations that militate against something like passive as the right choice for unaccusatives. The main empirical reason, as far as I know, lies in well-known contrasts like the one in (27):

A Semantics for Unaccusatives (27)

37

a. The boat was sunk [PRO to collect insurance] b. *The boat sank [PRO to collect insurance]

If unaccusatives involved the same operation as passivization, the contrast in (27) would remain a mystery. We would be left without a clue as to why the implicit argument of passives is capable of controlling the adjunct clause in (27a) but not in (27b). If passive does not work here, where else can we look? Should we make up some other way of saturating the subject a-role of transitive affimdare? Well, before doing that, it might not be a bad idea taking a closer look at the operations already available. The point is that, as often observed, in alternations such as those in (26), the transitive member of the pair tends to be causative. This entails that its meaning will have to be specified in terms of a CAUSE predicate. Let us assume, then, following, for example, Dowty (1979), that its meaning is specified as in (28a). The relation in (28a) holds between x and y just in case some action ~ of y causes a(x) to be the case, where a is some suitable, possibly abstract, predicate (like, say, being under water). In (28b) I give for comparison the equivalent of (28a) in the standard Montagovian notation.

(28)

a. b.

AxnAy3~[CAUSE(U~ (y), u a (x)] Ax Ay3P [CAUSE(P(y), a (x)]

What I would like to suggest is that the meaning of unccusative affondare is a reflexive form of the causative (28a) (which I will abbreviate as C(a». Its representation is as in (29a):

(29)

a.

b. c.

affondarelv = R(affondareTV) (= R(C(a), where R is as in (13» La barca eaffondata. 'The boat sank.' U [R(affondareTV)] (the boat) = a property of the boat causes the boat to sink

So I am essentially proposing to interpret the boat sank as something like the boat sank itself. But this is not quite right, of course. What we actually want is to interpret the boat sank as: some property of the boat (or some state the boat is in) causes it to go down. That is, with unaccusatives the causing factor must be understood, not as an action, but statively. This, it seems to me, captures exactly how the boat sank and the boat sank itself differ in meaning. When we say 'the boat sinks itself' we are imputing to the boat the capacity of performing an action; we are anthropomorphizing it. But for the boat to sink, it suffices that the boat has or comes to have a property that causes its sinking. The reflexive character I am attributing to unaccusatives is of a special, static nature. There are many conceivable ways of implementing this. We can define a reflexive operator that has this stativity requirement built in. Or we can assume that the CAUSE operator (implicitly present in the structure of sink) is neutral as to what type of property is involved (i.e. whether it is an action or a state). It is then the (in)animacy of the subject that triggers an implicature as to whether the causing factor involves an action

Gennaro Chierchia or a state. I will follow the latter strategy here, as it is simpler. But very little would change if this strategy turns out to be inadequate. Reflexivization as we have defined it above identifies subject and object and externalizes the remaining argument. Thus, we still have to explain how come the argument left behind is an internal one. The answer to this question is virtually forced upon us: the kind of reflexivization involved in deriving unaccusatives must 'leave behind' an internal argument. That is to say, it must have the following characteristics: (30)

'Internal' reflexivization: Rr a. Logical type ofRr: (e, rr) ~ (e, p) b. Content: Rr(sink) (x) = U[sink(x)] (x)

Rr is to R just what P r is to P. Truth conditionally, Rr and R are exactly alike, the difference residing in the logical type of their outputs. In fact, if passive is allowed to leave internal arguments behind (in our terms, to have propositional functions as output) there should be no reason why this option should be banned to other argumentmanipulating operations. The upshot of this proposal, then, is that intransitive members of an unaccusative alternation are related to their transitive counterpart via an operation of reflexivization that has these two features: (a) the causing factor is understood statively, and (b) the reflixivization operation is an 'internalizing' one. In (31) I provide a sample derivation: IP, U [E(Rr(sink) (the boat»] (..1) = Rr(sink) (the boat)

(31)

~

Np, ..1

If, U[E(Rr(sink) (the boat»]

~

VP, Rr(sink) (the boat)

I, U e

~

Y,Rr (sink)

I

eaffondata

NP, the boat

I

la barca

As you can see, the only difference with what we had previously (c£ (24» is that unaccusative sink has the structure Rr(sink), where sink is a two-place causative relation. This hypothesis also provides an answer to what happens when the internal argument of unaccusative ajfondare is externalized via NP movement. In this case, we have to assume that the externalizing operation of reflexivization R is used (as opposed to Rr). In (32) is an example. We can now think of the relation between transitive and intransitive sink as follows. Intransitive sink is derived from the transitive one via reflexivization, with the possibility of freely choosing between Rand R r. IfRr is chosen, unaccusative sink will have to take its argument internally. IfR is chosen, the argument of sink must be externalized via NP movement. Thus NP movement is interpreted very differently in unaccusative structures from the way it is interpreted in passives.

A Semantics for Unaccusatives (32)

39

IP, U [R(sink)] (the boat)

~

NP i

I', U[R(sink)]

~

VP, R(sink)

I, U la barca

I~ V,R(sink)

I

eaffondata

Np, Id

I

ei

An immediate consequence of the hypothesis that unaccusatives (of this kind) are a special case of reflexives is that the contrast in (27), repeated here as (33), can now be derived. Consider:

(33)

a. b.

*La barca eaffondata per [PRO far incassare l'assicurazione a Giuseppe] to make Giuseppe collect insurance.' 'The boat sunk Gianni eannegato per [PRO far incassare l' assicurazione a Giuseppe] 'Gianni drowned to make Giuseppe collect insurance.'

In (33a) the only available controller is the boat (both cause and causee, on our proposal); but the boat as such cannot collect or malce anyone collect insurance, whence the deviance of (33a) (and of (27a)). This predicts that if the subject of an unaccusative can perform the relevant action, the sentence should become grammatical. This indeed seems to be so, as (33b) shows. The structure of (33b) is identical to the one of (33a), but the subject Gianni in (33b) can make Giuseppe collect insurance and the sentence is grammatical. The next question to ask concerns unaccusatives that lack a transitive counterpart. Does it make any sense to try to extend to them the view that unaccusatives are reflexives? And if not, how is NP movement interpreted in such cases? There are two logical possibilities here, it seems to me. One is to assume simply that unaccusatives that lack a transitive counterpart are just lexically marked as lacking an external argument (in our terms, as being of type <e, p»). For verbs like seem, whose internal argument is propositional, this seems very plausible. But for verbs whose internal arguments is non-propositional, another option is available. One can assume that these verbs, too, involve reflexivization of a covert, abstract relation. In other words, unaccusatives (seem & Co. aside) would uniformly have the structure R( C(a)), where R is (externalizing or internalizing) reflexivization, C is a causative operator, and a is a (abstract) one-place predicate. C(a) sometimes surfaces as an actual transitive verb and sometimes does not. Before we try to assess the merits and faults of this proposal, let me illustrate it a bit further by means of a more concrete example. Take the verb crescere (,grow'). In standard Italian, it is only intransitive (see 34a). There is of course a twoplace relation, meaning roughly 'cause to become bigger', which constitutes the interpretation of the English transitive grow. That relation, one can assume, underlies the representation of unaccusative crescere in Italian.

Gennaro Chierchia The two hypotheses we are contrasting are the following: Hypothesis I: unaccusatives lacking a transitive counterpart are reflexivizations of abstract causative relations (apart from such cases as seem). Hypothesis 2: unaccusatives lacking a transitive counterpart are simply listed in the lexicon as lacking an external argument. Perhaps Hypothesis 2 might strike one as being formally simpler. This is only apparent, however, at least within the present set of assumptions. Earlier we gave an arguably simple account of how NP movement in unaccusatives of the sink type should be interpreted. Obviously, under Hypothesis 2 such an account would not extend to unaccusatives lacking a transitive counterpart, while under Hypothesis I it would. That is to say, under Hypothesis I you can maintain that the If of (Hb), for instance, is as in (Hc): (34)

a.

*Gianni ha cresciuto pomodori e lattuga tutta la vita. 'Gianni grew tomatoes and lettuce all his life.' b. I pomodori sono creSClUtl ei . 'The tomatoes have grown.' c. U[R(C(a»](the tomatoes), where R is the 'externalizing' reflexivization and C( a) is whatever relation is associated with transitive grow in English.

The If in (34c) is the same as the one in (32). Per contra, on Hypothesis 2 we would need two different ways of interpreting NP movement in unaccusative structures: one for unaccusatives with a transitive counterpart and one for those without one. On the basis of this discussion, it would seem that, after all, Hypothesis I is preferable on grounds of formal simplicity. And it turns out to be preferable on empirical grounds as well. In the next section I consider four arguments that seem to favour it.

1.4.2. Empirical evidence 1.4.2.1.

Unstable valence

As C. Rosen pointed out to me, unaccusatives tend to be unstable in their valence. They often oscillate from transitive to intransitive and vice versa, both diachronically and across dialects. For example, while in standard Italian crescere is intransitive (c£ (3¥», there are dialects in which transitive uses (such as (35a» do occur.

(35)

a.

b.

I figli, Gianni li ha cresciuti bene. his sons, Gianni them (el.) has raised well R(C(a»

And we find transitive uses even of such verbs as morire ('die'), ribellare ('uprise'), suicidarsi ('commit suicide'). In contrast, unergative verbs such as sudare ('sweat') and piangere ('cry') seem not to undergo these shifts in valence. Oscillations of this type in unaccusatives are just what one would expect, if their

A Semantics for Unaccusatives

41

representation is as in (35b). What happens to an unaccusative lacking a transitive counterpart when it is transitivized is simply that the transitive C(a) embedded in its lexical representation surfaces as an autonomous lexical entry. If, on the other hand, these unaccusatives were simply listed in the lexicon as lacking an internal argument, there would be no reason for them to undergo shifts in valence any more easily than their unergative cousins. 1.4.2.2.

Reflexive morphology

The second argument is based on the fact that a significant class of unaccusatives is marked by reflexive morphology: (36)

a.

b.

rompere-rompersi 'break' aprire-aprirsi 'open' irritare-irritarsi 'irritate' etc. inginocchiarsi 'knee' scontrarsi 'collide' arrabbiarsi 'get angry' etc.

In (36a) we see examples of transitive-intransitive alternations of the unaccusative type where the intransitive member of the pair is obligatorily marked by a reflexive morpheme (the reflexive clitic si). In (36b) we see examples of unaccusatives without a transitive counterpart-these are also obligatorily marked by a reflexive. Reflexive marking of this kind, which in every respect behaves like a regular reflexive clitic, is sometimes called 'inherent' and is generally viewed as a pure marking of unaccusativity (see, for example, Burzio 1986: ch. 3). The question that arises now is the following: why should unaccusatives be marked by reflexive morphology, as opposed to any other one? As far as I know, this question has not yet been answered satisfactorily. 6 Structurally, unaccusatives are parallel to passives, not to reflexives. The present approach clearly offers a basis for a principled account of such a phenomenon. It has been observed occasionally that unaccusatives (and/or middles) and reflexives appear to have some common semantic traits. But this semantic similarity has never been actually spelled out. The present theory provides a substantive hypothesis concerning what exactly unaccusatives have in common with reflexives and how they differ from them. Unaccusatives are viewed as reflexivizations of causative transitive forms. But they are reflexivizations of a special sort. The causing factor is not perceived as an action performed by the subject, but as a property or state of it. Thus, what is special about unaccusatives marked by si as opposed other unaccusatives is simply that the former wear their meaning on their sleeves. All unaccusatives have an implicit reflexive operator in their If. In some cases, the reflexive operator is 6 Within Relational Grammar a 'retro-advancement' from 2 to I is postulated, which creates the same structural configuration associated with reflexives (cf. Rosen 1981: 71 ff.)

Gennaro Chierchia lexically incorporated into the meaning of the verb without any morphological reflex. In others, the reflexive operator actually shows up in its usual form: a clitic, a piece of verb morphology.

1.4.2.3. 1he distribution ofda se phrases A consequence of the view that unaccusatives are reflexives is that the subject of unaccusatives should systematically be associated, not only with the semantic role associated with the object (say, theme), but also with the one associated with the subject. In John kills himself, John is both the theme and the agent of the killing. Similarly in the boat sunk, the boat is claimed to be both the theme and the cause of the sinldng (not the agent, of course, since it is inanimate and hence can perform no action). In other words, the subject of unaccusatives should be associated with the entailments that characterize themes as well as with the entailments that characterize causes. If we can find grammatical constructions sensitive to these roles, we should be able to test this consequence, which, as far as I can tell, is peculiar to the present hypothesis. I think that there is indeed, in Italian, a structure sensitive to the 'cause' role, which enables us to test our claim. The structure in question has to do with the idiomatic anaphor da Sf, similar to the English by itself, exemplified in (37). (37)

Gianni mi ha picchiato da se. Gianni hit me by himself

Da Sf is a modifier, occurs in adjunct position, and its antecedent must be construed as the sole cause of the event under consideration. That is what da Sf contributes to meaning. Sentence (37) says that Gianni is the only agent of the hitting. The da Sf anaphor appears to have the following properties. It disallows long-distance antecedents (38a) and is subject oriented (38b).

(38)

a. *Gianni sostiene che io sia venuto da se. Gianni claims that I came by himself 'Gianni claims that my arrival was his responsibility.' b. *10 ho picchiato Gianni da se. I hit Gianni by himself

The facts in (38a) suggest that da Sf is subject to principle A of the binding theory. This would account also for (38b), since da Sf is an adjunct and the only C-commanding local antecedent would be the subject. The following facts further support the claim that the antecedent of da Sf must be an agent or cause:

(39)

a. *Gianni conosce illatino da se. Gianni knows Latin

by himself

b. *Gianni ha sudato da se. c.

Gianni sweat by himself A: Tu hai fatto sudare Gianni. 'You made Gianni sweat.'

A Semantics for Unaccusatives

43

B: No, ha sudato da se. 'No, (he) sweat by himself.' Example (39a) shows that with statives da se is bad, as their subject is non-agentive. With verbs denoting unvoluntary physical functions such as sudare ('sweat'), da se is also bad (see (39b)). It improves somewhat only to the extent that their subjects can be understood as causing the physical change. The dialogue in (39c) provides a context in which this happens. Given all this, we expect da se to be impossible in passives, for the only possible controller (the subject) is dearly non-agentive. This expectation is indeed borne out: *La porta estata aperta da se the door was opened by itself

(40)

Sentence (40) is completely ungrammatical, whether the agent is expressed or not. A similar point can be made with so called impersonal si: (41)

*Questo libro si legge da se this book one (d.) reads by itself

Ifwe try to construe si in (41) as impersonal (as on the reading indicated by the gloss), the sentence is ungrammatical. The only reading that (41) has is the anthropomorphic one where the book reads itself; in such a case, however, si is being interpreted as a reflexive and the sentence is, consequently, agentive. Consider now the case of unaccusatives. According to our theory, the subjects of unaccusatives are systematically associated with the causer role, for they are reflexivization of causatives. Hence, we expect that they should be acceptable antecedents for da se phrases. And indeed they are: . (42)

La porta si eaperta da se. 'The door opened by itself' b. La barca eaffondata da se 'The boat sunk by itself' c. *La barca estata affondata da se. 'The boat was sunk by itself'

a.

Sentences such as (42b) and (42C) provide a nice minimal pair involving a passive and an unaccusative. These contrasts are very sharp. This test provides us with a way to distinguish our hypothesis from a conceivable alternative. One could argue that in unaccusatives the cause/agent role is simply absent. That is to say, we could assert that in the unaccusative member of the transitive-intransitive pair the subject a-role is just not there. A daim along these lines has been put forth in connection with English middles (c£, for example, Keyser and Roeper 1984). This hypothesis could perhaps provide an alternative account for the control facts considered above in (27)-but it would leave the behaviour of da se phrases in the dark. There would be no reason why the subjects of un accusatives are legitimate antecedents of phrases requiring precisely

Gennaro Chierchia

44

an agent/cause role, while the subjects of passives are not. On the other hand, under the present theory the distribution of da se phrases falls into place. 1.4.2.4. The aspect ofunaccusatives

The claim that unaccusatives are reflexivizations of causative structures has a further consequence concerning their aspect. Causatives involve semantically the bringing about of a certain state. There can be causative states; but if a causative is non-stative, it will have to be telic (for the coming about of the relevant state provides a culmination for the event). It follows that non-stative unaccusatives cannot be activities, but have to be accomplishments or achievements. This expectation appears to be borne out, as noted by several researchers (c£ especially Van Valin 1987 and Levin and Rappaport Hovav I988). The aspectual class of unaccusatives is telic in the overwhelming majority of cases. This can be illustrated with particular cogency by considering verbs like run, as the following argument shows, adapted from Van Valin's work. Run is generallyatelic (i.e. an activity or a process). In Italian, corr'ere ('run') has both an unergative and an unaccusative variant, as attested by the fact that it can talce both essere (the aux selected by unaccusatives) and avere (the aux selected by unergatives). Yet these two variants of run have different aspectual properties. The canonical test to distinguish telic from atelic verbs is modification by flr- and in-phraees, as illustrated in (43), talcen from Dowty (1979): (43)

a.

b. d. e.

John pushed a cart for an hour ??John drew a circle for an hour ??John pushed a cart in a minute John drew a circle in a minute

For-phrases are odd with telic events and fine with atelic ones. In-phrases behave the opposite way: they are odd with atelic events, fine with telic ones. Many different accounts of these facts have been pursued. One of the most complete to this date remains that of Dowty's. He provides an aspectual calculus where, essentially, telic events are analysed in terms of a cause operator, which is then provided with an interval semantics in whose terms the distribution of various adverbials can be derived. The point of relevance to us is that unaccusatives seem to pattern systematically with telic events with respect to tests such as those in (43): (44)

a. b.

??Gianni ecorso in giardino per un'ora. Gianni AUX run in the garden for an hour Gianni ha corso in giardino per un'ora. Gianni AUX run in the garden for an hour

The choice of aux in (44b) shows that we are dealing with the unaccusative variant of the verb run. Modification by a for-phrase is odd, which suggests that we are dealing with a telic event in this case. This contrasts with what happens if the choice of aux

A Semantics for Unaccusatives

45

favours towards the unergative interpretation. The behaviour of in-phrases reinforces this pattern. (45)

a. b.

Gianni e corso in giardino in un minuto. Gianni AUX run in the garden in a minute ??Gianni ha corso in giardino in un minuto. Gianni AUX run in the garden in a minute

What is interesting is that this happens with a verb whose basic meaning does not lend itself naturally to a telic interpretation. Yet to the extent that such a verb is classified as unaccusative, a telic interpretation is superimposed. That is, correre is interpreted as 'run to point x', which bounds the activity and gives it a culmination that makes its aspectual class shift. And this behaviour appears to be completely general. The correlation between unaccusativity and telicity is highly systematic and calls for a principled account. The present line of inquiry appears to provide a basis for it. If the representation of unaccusative correre is of the form R(C(a))-say, cause oneself to bring about a change in location in a certain way-then we would expect it to be classified as telic, since (non-stative) causatives have a built-in culmination. If unaccusatives were simply listed in the lexicon as lacking an external argument, nothing would seem to follow concerning their aspect. Their distribution across the various aspectual classes would be expected to be random, contrary to fact.

I.4.3. Intermediate summary On the basis of the foregoing discussion, I conclude that the hypothesis that unaccusatives are generally reflexivizations of causatives (that is, that they have the logical form in (46a)) is well supported. For such proposition-taking verbs as seem, this option is not available (propositions do not undergo reflexivization). Hence such verbs simply have to be listed in the lexicon as having the type shown in (46b). Verbs of this subclass are expected not to have the properties associated with the structure in (46a). They should thus not undergo shifts in valence, nor should they ever be associated with reflexive morphology. Moreover, they are not expected to be telic.

(46)

a.

b.

R(C(a)) seem: (e, p)

This analysis provides us with a semantics for unaccusatives which is not only truthconditionally adequate and explicit. It also accounts for transitive-intransitive alternation in a way that restores the predictive power of the projection principle and bypasses the dubious notion of optional argument. And it explains certain properties of unaccusatives previously either unnoticed (like their distribution in connection with da se phrases) or unaccounted for (like the presence of reflexive morphology and their aspectual class). Further consequences concern the much-discussed process of aux-selection in Italian, to which I turn next.

46

Gennaro Chierchia

1.5.

AUX SELECTION

In Italian, the aux esse/I! is selected over avere in the following cases: (47)

a. b. c. d.

passives reflexive clitics unaccusatives impersonal si

The most sweeping attempt to account for this distribution within the Principlesand-Parameters framework is Burzio (1986). Burzio's account can be schematically summarized as follows: (48)

a.

NP a V ~

b.

essere is selected iff a binding* relationship obtains between the subject

~

and a nominal contiguous to the verb. The asterisk on the term binding indicates that not all types of binding will do. Binding of such non-clitic reflexives as Sf stesso does not trigger essere selection. Burzio's approach does remarkably well in trying to reduce the seemingly unrelated contexts in (47) to a coherent whole. Still, one cannot fail to notice that an awkward disjunction remains. Essere is selected if and only if the subject binds the object position or if it binds a subject clitic, where the latter side of the disjunction accommodates the auxselection properties of impersonal si constructions. The question we would like to ask here is whether having an explicit semantics tal{es us any further in our understanding of what it is that mal{es the contexts in (47) a natural class. The following considerations suggest that this might indeed be so. I. 5.I.

Subject-affectedness

Consider the semantic operations associated with the structures in (47) (leaving aside for the moment impersonal si). Passive requires P, reflexive clitic require R, unaccusatives require Rand/or E. What do these operations have in common? The answer seems clear: they are all subject-affecting operations. As discussed in section 1.3.1, the operations just mentioned affect a specific projection ('slot') of a relation, namely the one that corresponds to the subject (in our notation, the last one). P existentially quantifies over the subject slot. R identifies the subject with the object slot. E adds a truth-conditionally inert subject slot. So the generalization that suggests itself is that essere is defined exclusively for subject-affected properties-that is, properties in the range of subject-affecting operations. Let us spell out this hypothesis a bit further. As auxs are verb modifiers, they can be viewed semantically as functions from properties into properties. The function/

A Semantics for Unaccusatives

47

argument structure of, say, sentences such as (49a, c) can be specified as in (49b, d) respectively. (49)

a.

b. c.

d.

Gianni estato picchiato. Gianni AUX-PAST beaten-PAssPART 'Gianni was beaten.' U[essere(P(beat) )](g) Gianni ha sudato. Gianni AUX-PAST sweated U[avere(sweat)](g)

Since auxs are property modifiers, they can be sensitive to the semantic make-up of what they modifY. For example, in Spanish there are two forms of the copula, ser and estar; ser seems to select the class of properties that are somehow 'inherent' or 'essential', while estar selects the complement of such class. The choice of essere vs. avere in Italian, on the other hand, is sensitive to a different factor: subject affectedness. More explicitly, the range of subject-affecting operations will be the set: {rEn: ::Jg[f(g) =rn, where n is the set of properties, g any n-place relation and f a subject-affecting operation. This set constitutes the domain of essere; its complement (namely, n - {rEn: ::Jg[f(g) = r]}) the domain of avere. All this can be summarized as follows: (50)

a. b.

Esse/'{! is a function defined only on subject affected properties. A subject affected property is a property in the range of a subject-affecting operation ({rEn: ::Jg[f(g) = r]}).

If this is correct, it must be the case that impersonal si is also semantically associated with a subject-affecting operation. This hypothesis has a great deal of intuitive plausibility, for, as it has been observed several times, impersonal si sentences are very close in meaning to passives. And if indeed it turns out to be correct that impersonal si corresponds semantically to a subject-affecting operation, then (50) would provide us with a maximally simple, truly non-disjunctive, characterization of the environments in which essere is selected, and would thus, arguably, constitute an improvement over (48) above. I believe that within the limits of the present work, it will be impossible to establish conclusively the thesis I have just outlined: the syntax and semantics of si are one of the most complex aspects of Italian grammar, and there is no way that I can do justice to it. However, I would like to offer some considerations that I think lend strong plausibility to the claim that the semantics of si is formally on a par with the semantics of passive.

1.5.2.

Impersonal si

As noted by Cinque (1988), among others, salient features of the semantics of impersonal si are the following: si is restricted to ranging over a (plural) group of humans,

Gennaro Chierchia and is sensitive to discourse factors, such as the generic or episodic character of the sentence, the presence of adverbs of quantification, and the like. This latter aspect of the meaning of si is illustrated by the following examples.

(SI)

a.

b. c.

d. e.

f g.

Si e bevuto molto ieri sera. si AUX.PAST drunk a lot last night 'People/we have drunk a lot last night.' 3xp[drink a lot (xp)] Se si invita qualcuno, 10 si deve intrattenere if si invites someone, him (d.) si must entertain 'If you invites someone, you have to entertain them.' VXp Vy[invite(xp'Y) -7 entertain(xp,y)] Talvolta, se si invita qualcuno, 10 si deve intrattenere. 'Sometimes, if one invites someone, one has to entertain them.' In Italia, si beve molto vino. 'In Italy, one drinks a lot of wine.' VXp [xp in Italy -7 xp drinks a lot of wine]

In (SI) we have an episodic sentence and si is interpreted as having quasi-existential force, as indicated in (SIb), where xp is a variable ranging over groups of people? In (SIC) we have a conditional, and si acquires quasi-universal force. These examples show, furthermore, that si licenses what has come to be known as donkey anaphora. In fact, as is generally the case with donkey-type anaphora, the universal quantificational force in such examples as (SIC) is obtained as a default only when no explicit adverb of quantification is present. If such an adverb is overtly present, it typically determines the quantificational force of the indefinites within its scope. Thus, for example in (SIC), si has again a quasi-existential force under the scope of talvolta ('sometimes'). So for (SIC) to be true it suffices to entertain some guests, not all guests. Finally, in (SI/) we have a generic sentence (which can be analysed as a conditional of a special sort-see Krifka 1988, for example) and, consequently, si acquires again quasi universal force. These facts strongly suggest that si should be regarded as indefinite in the sense of Discourse Representation Theory.8 This theory treats indefinites as variables (whose value has got to be 'new' in the relevant context). Their quantificational force is then determined in systematic ways by the presence of overt or implicit unselective quantifiers. Within the present framework, a first step towards a discourse-representationtheoretic treatment of si that does justice to the previous observations could be that in (S2).

7 There is also an 'empathy' effect with the speaker-that is, the speaker adopts the point of view of the subject. For relevant discussion, see Sells (1987) and Cinque (1988). 8 See Kamp (1981) and Heim (1982). An alternative formulation of Discourse Representation Theory that maintains the basic insight of Kamp and Heim but treats indefinites as existentially quantified NPs (using a 'dynamic' logic), can be found in Groenendijk and Stokhof (1987) and Chierchia (1988).

A Semantics for Unaccusatives (52)

a.

b. c.

49

Logical type of si: (n, p) Truth-conditional content: sip (r) = Ur(xp) [xp a variable ranging over a group of people] Example: slp(lavora) = u[lavora(xp)]

In (52a) the semantic operation associated with si (denoted as sip) is defined as a map from properties onto propositions. Such a map plugs a distinguished variable (restricted to groups of humans) into the subject slot of a property. The quantificational force of this variable is then determined along the lines of Kamp's and Heim's theory. It is easy to see that the operation in (52) corresponds closely to passive-see in particular our definition of impersonal passive PI in (21). The main differences between sip and PI appear to be that sip but no PI restricts the variable introduced in the subject slot to groups of humans and that sip, but not PI' does not close existentially the variable it introduces. One cannot fail to notice in this connection that the second difference might just reflect an inadequacy of our semantics for passive as originally formulated in (IO) and (21). The point is that the implicit agent argument of passives does interact with quantificational adverbs and conditionals, just as other indefinites do:

(53)

a.

b.

If this medicine is taken ip. the morning, it must be tal<:en on an empty stomach. Vx Vy[x tal<:es y in the morning ~ x takes y on an empty stomach]

In (53a) we see that in the antecedent of a conditional the implicit agent of a passive gets quasi-universal force, much like si and other indefinites. This suggests that indefinites-si and the implicit arguments of 'agentless' passive-should be treated uniformly. If indefinites are treated as variables, then both si and regular passive should simply introduce a variable into subject position. This modification of our semantics, if ultimately warranted, does not affect, as far as I can tell, any of the points made here. How does this proposed semantics mesh with syntax? Consider the LF for impersonal si constructions proposed in Cinque (1988), reproduced in (54). (54)

IP

~

I'

NP

~

pro

I

VP

~

lavora

Agr (no features)

si [+arg]

Cinque, building on previous work on the topic, argues that in such structures as (54),

si is an [NP,IP] clitic that receives the subject a-role, while the subject position is occupied by an expletive pro. On the basis of our semantics for si given in (52), the interpretation of (54) proceeds as follows:

Gennaro Chierchia

50

IP, U[E(sip (work»] (1-) =sip (work) =Uwork(xp)

(55)

~

NP,1-

I' U[E(sip (work»]

~ I

~ pro

Agr,U (no features)

VP, work lavora

si, sip [+arg]

The interpretation of (54) is completely parallel to the one of impersonal passive. Let us consider this in some detail. The interpretation proceeds bottom up, as usual. In Infl we have U and sip. There are two options here, depending on the order of application of these operations. Either we get Usip(work) or, on the opposite order, sip~work). In either case, the result is ill formed, as types do not match. Hence, we must resort to type shifting. Our operator E, in particular, enables us to get the well formed U[E(sip(work»], a property that looks for an expletive subject. And such property finds the expletive it needs in subject position. The resulting logical form then reduces to Uwork(xp), where the free variable will be quantificationally elosed on the basis of the principles of Discourse Representation Theory. So, assuming that si is associated semantically with sip, the route from structures like (54) into their lfs follows from general principles. Things are of course more complicated than what I have been able to say here might lead the reader to believe. 9 But perhaps the above considerations provide enough details to lend some preliminary support to the thesis that the interpretation of impersonal sentences in Italian must involve a passive-like subject-affecting operation.

1.5.3. Further consequences On the basis of our discussion, it would appear that the suggestion put forth at the beginning of section I. 5 is not just wishful thinking. The aux-selection process in Italian appears to be amenable to a maximally simple semantic characterization. Here I would like to point to two further interesting consequences of our formulation of aux selection as a semantic constraint on the meanings of essere and avere. In introducing our treatment of reflexivization in section 1.5, we pointed out that non-elitic reflexives (such as English himselfand Italian se stesso) could not be assimilated to an argument reducing operation such as our R. They must be associated with a binding operator that links the reflexive to its antecedent. In (56) I illustrate the differences between clitic and non-elitic reflexives, using Bach and Partee's (1980) approach to the latter.

9 In particular, Cinque (1988) argues that there is also a non-argumental si that occurs with a non-pleonastic pro. I believe that the semantics sketched here is compatible also with this aspect of Cinque's proposal, but I will not pursue the issue here.

A Semantics for Unaccusatives (56) a.

51

IP, !..xu [love (x) ] (x) (g) = U [love(g)] (g)

~

If,!..xu [love(x)](x)

NP, g

~

VP, OJ (love(x)

I,u

~

I

ama

b.

NP, Xi'storeO~O

V, love

Gianni j

= n!..xu [love(x)] (x)

I'

1

se stesso j

IP, U [R(love)](g)

~

NP, g

If, U [R(love)]

~

VP, love

I,uR Gianni j

~

V, love

I

Np, >taa

I

Leaving irrelevant details aside, Sf stesso in (56a) is associated with an ordinary pronoun-meaning and an operator in store. This operator is then assigned scope (perhaps at ymax, perhaps in infl) and links the reflexive to its antecedent (subject to principle A). I leave it open here how this storage operation is to be syntactically realized. In keeping with the tenets of Property Theory, R (love) and !..xU [love (x) ] (x), the interpretations of si ama and ama Sf stesso, respectively, while truth-conditionally equivalent, are distinct properties. And given the syntax and semantics of non-elitic reflexives, there is no obvious way in which VPs containing them can be regarded as subject affected properties in our sense. The fact that they do not trigger essere selection is a consequence of this. Burzio tries to account for the different behaviour of elitic vs. non-elitic reflexives by arguing that essere selection and the licensing of elitics are S-structure phenomena, while the binding theory (responsible for the distribution of non-elitic reflexives) applies at LF, a controversial hypothesis. It is also significant that a similar difficulty (an unwanted binding theoretic relation accidentally meeting the description for essere selection) arises in an untelated area, namely the pro-drop parameter (as Burzio 1986: ch. 2 himself points out). It is generally assumed that the pro-drop parameter has to do with the 'richness' ofinfl or agr. A way of spelling this out (the one adopted by Burzio) is to regard agr as a pronominal element, essentially a [NP, IP] elitic. As such, it will be coindexed with the subject position so as to license an empty category in that position W

The proposal and notation are tal,en from Bach and Partee (1980).

Gennaro Chierchia

52

and to make the recovery of its content possible. So, the structure of a sentence such as (57a) will be as in (57 b). (57)

a.

b.

Telefona. pro phones [NP eJ agr j telefona

But (57b) appears to be just the configuration that should trigger essere selection. Now, proposals as to the exact formulation of the pro-drop parameter vary a great deal. But they all share the assumption that pro-drop phenomena of the type found in Italian are based on a syntactic relation between agr and the subject position, a relation that appears to be structurally similar to the one between the subject elitic si and the subject position. Burzio addresses this issue briefly (1986: 93) and suggests that the two relations in question should be viewed as 'complementary'. In the present approach, this issue does not arise at all. Pro is a pronominal element and its semantics will be analogous to the one of its overt counterpart he. It could not be associated with an operation on relations, in the way impersonal si constructions can (and, ifI am right, must). Accordingly, structures such as (57b) are irrelevant to aux selection. The point that emerges is that by stating essere selection as a condition on the domain of essere and avere (the first being a function defined for the set of subject affected properties, the second for its complement) we can derive which binding relations are going to affect such process. In particular, there is no danger that 'wrong' types of binding (like binding of se stesso or of pro) may get in the way.

1.6.

SUMMARY AND (TENTATIVE) CONCLUSIONS

Unaccusativity phenomena are at the centre of an intricate node ofissues in the grammar of Italian and in Universal Grammar. My proposal is tentative and incomplete. It remains to be seen, for example, whether my formulation of the essere selection rule can be maintained once causatives and 'reanalysis' constructions are talcen into consideration. It also remains to be seen, for example, how the present semantics interacts with the definiteness effect associated with unaccusatives (cf. Belletti 1988). And, of course, it remains to be seen how the present line does in connection with unaccusativity in other languages. Let me try to summarize briefly what I have tried to do. I have advocated the adoption of an (independently motivated) theory of properties as the semantic algebra in terms of which truth conditions are recursively specified. The main characteristic of such a theory is that properties are taken as primitives and linked to their arguments via predication, where the latter is viewed as a map from properties onto propositional functions. Thus, in a sense, predicates come in two forms-as properties and as propositional functions (with predication connecting them). I then indicated how to use such a theory in interpreting compositionally fragments ofLF, that is, in mapping LF

A Semantics for Unaccusatives

53

onto 1£ The main assumption here is that clausal structures come about via predication, which I have taken to be associated with infl (the predication principle). This principle, which plays a key role in our approach, is 'configurational' in the sense that it makes crucial use of syntactic configurations (much as the projection principle). One of the consequences of the predication principle is that verbs that are associated with propositional functions must have certain characteristics: they will have to take their argument within the VP and will have to be associated with an expletive subject. These are central features of unaccusatives. I have furthermore argued that unaccusatives should be regarded as a special type of reflexivization. Such a form of reflexivization is also involved in the 'externalization' of the internal argument of un accusatives via NP movement. I have argued that this hypothesis accounts for (a) the unstable character of their valence, (b) the association of unaccusatives with reflexive morphology, (c) the control properties of da Sf phrases, and (d) the aspectual properties of unaccusatives. Finally, our semantics has led to a reformulation of the aux selection rule of Italian as a purely semantic domain condition on the meaning of essere that exploits the notion of subject affected property. Such a reformulation may pave the way to a genuinely non-disjunctive account of the factors triggering essere selection. While a great deal of this picture might turn out to be modified in the light of future work, one thing does seem to emerge from it. What we assume concerning the semantics (= If) of unaccusatives is bound to mal{e a significant difference in our understanding of the phenomenon. While syntactic research has convincingly shown that a simple-minded semantic account of unaccusativity will not take us far, it also seems to be the case that a purely syntactic account of the phenomenon will not take us all the way. Neither the syntax nor the semantics of unaccusativity can be reduced to an epiphenomenon of the other.

POSTSCRIPT The main idea put forth in this chapter is that unaccusativity is a form of reflexivizadon. Is such a hypothesis still viable? The debate on this question remains open. Here I would like to add a few observations on it, most of them prompted by Reinhart (r996), which insightfully reformulates and develops such a line of investigation. As is well known, reflexivization can be viewed as an operation that reduces the arguments of a relation by identifYing them:

(r)

REFL(K)

=

AU K(U,U)"

Such an operation turns a relation into a property (and can of course be generalized to n-place relations, but we will not get into that here). It is useful, I think, to visualn

REFL

corresponds to R in the chapter.

Gennaro Chierchia

54

ize what REFL actually does in a concrete example. Let D = {a, b, c, d} be our domain of discourse. Assume that the extension ofK (in a relevant situation) is as in (2a). Then the result of applying REFL to it will be as in (2b): (2)

a.

(a, b) (b,c)

K= (c,d) (a, a) b.

(d,d) REFL(K) = {a, d}

Under standard assumptions about relations, properties, etc., there is at most one operation like REFL. REFL is entirely defined by what it does: it applies to sets of ordered pairs and turns them into sets by selecting those individuals forming pairs constituted by identical objects. There are no two ways of doing that. (There can be, of course, two different symbols or structures associated with the same operation.)I2 I have argued in 1989 (this chapter), following much other work, that in the languages of the world, REFL is the semantic counterpart of at least some forms of reflexive marking. For example, there are reasons to believe that REFL is involved in the semantics of the reflexive clitic si in Romance. The question is, can REFL also be the very operation that derives unaccusatives? It would be great if it were so, for it would reduce the essence of a complex phenomenon to a simple and rather well-understood operation on argument structure. I believe, however, that the answer to such question has to be 'no'. Let me motivate my scepticism. Take your prototypical verbs that have unaccusative alternates, say, break or sink. The following are clearly contradictory.

(3)

a.

b.

John broke the door, but the door did not break. The explosion sank the boat, but the boat did not sink.

What these elementary examples show is that ifKTR is a two-place relation (transitive) and KIN is its unaccusative alternate (intransitive), anything that occupies the second slot ofKTR (the one corresponding to the object) has to be in the extension of KIN' That is to say, (4)

Main fact about unaccusative alternations: For all u, u', ifKTR (u,u'), then KIN(u')

But REFL does not and cannot yield that. Quite the opposite. Given what REFL means, for any K, REFL(K) has to be a proper (possibly empty) subset of the individuals related by K, unless, of course, K is already a reflexive relation. I) A formal proof of this fact can n This is true also if possible worlds are added. It is not necessarily true in a framework where properties and relations are taken as primitives (c£, e.g., the Property Theory in the chapter). In such a setup, two functions, relations, etc. that do the same thing can be distinct functions (as functions, relations, etc. are not identified with their set theoretic graphs). 13 The land of reflexivity that K would have to satisfY to get the generalization in (4) via an application

A Semantics for Unaccusatives

55

be readily provided, but just look at (2), and you immediately see why. REFL merely collects in a set all the pairs of the form (u,u) in the extension of the relation; all pairs of the form (u, u'), with u;
(5)

a. The explosion sank the boat. b. The gust of wind opened the door. c. An earthquake moved these rocks.

Cases like these have the form K(e,x), where e is an eventuality and x an ordinary individual. Notice one thing. The eventuality e that surfaces as subject must clearly affect or involve the object x. The explosion must be sufficiently near the boat, the gust of wind must invest the door, etc. In e-role talk, there must be some kind of thematic relation that links the event surfacing as subject to the object. To make things' more explicit, let us couch the relevant relation, namely sink, in a Davidsonian format, where every verb has an implicit argument ranging over eventualities. In this format, something like the explosion sank the boat would have a logical form of roughly the following type: (6)

::Ie' sank(e,x,e') [where e=the explosion]

This is to be read as follows. There is a sinking e' of the boat x brought about by the explosion e. This can of course be further analysed by factoring out explicitly the relevant e-roles, for example along the lines proposed in Parsons (1990). The important point is that there are two eventualities at play here-the one associated with the subject (the explosion) and the one brought about by it (the actual sinking). It is inconceivable that the eventuality expressed by the subject bears no significant relation to

of REFL is the following: (i)

\fx[3yK(y,x)

---7

K(x,x)]

At first sight, relations such as sink do not satisfY this constraint. From John sinks the boat it does not ~eem to follow that the boat sinks itself(which is, in fact, a deviant sentence). I do believe that the meaning of the boat sinks is in some sense close to that of the boat sinks itself. But the problem facing any approach along these lines is of course to spell out in exactly what sense this is so,

Gennaro Chierchia the one brought about by it. These two eventualities must share a 8-role-Iet us call this '8-sharing'. Clearly, 8-sharing is perfectly general, and applies to every relation that takes events as arguments. Let us make this explicit: (7)

8-sharing For every eventuality e, e' and every individual x, ifK(e, x, e') holds, then there is a thematic relation 8 such that 8(e, x).

Not much to quarrel over so far. Now a further observation. Intuitively, 8-sharing is a form of hidden reflexivization, in the sense that one individual plays a double role in two distinct events. What (7) esentially says is that x is, say, the theme of e' and at the same time the theme (or some other 8-role) of e. It is this form of reflexivization that I think is at the basis of unaccusative alternations. Unaccusativity brings this hidden form of reflexivization to the surface. Here is one way of making this intuition explicit. Consider any relation K that can take as subject argument an eventuality. It is straightforward to turn it into a reflexive relation K" as in (8). (8)

a.

b.

Kr= AxAy Ae' 3d 8 [K(e, y, e') 1\ 8 (e, x) I\x=y] sinkr =AxAy Ae' 3d 8 [sink(e, y, e') 1\8 (e,x) I\x=y]

The operation [ Y (call it 'reflexive closure') simply pulls out the argument that the subject event shares with the event being brought about. Because of8-sharing (which is simply a factual observation) such an operation is guaranteed to be well defined. Reflexive closure, in effect, puts a copy of the object argument into the first argument slot of the relation, thereby making it reflexive. Now take a particular output of such operation, say sinkr, and suppose there is a word that lexicalizes it. What properties would this word have? How could it enter in a syntactic derivation? The most directly relevant observation in this connection is that it would not make sense to apply sinkr to two distinct arguments, for the result would be necessarily false (by definition of [ Y). The arguments of sinkr must be identical for the resulting sentence to make sense. Projecting sinkr into a usable syntactic structure must guarantee identity of arguments. The computational system, as presently conceived, gives us two and only two ways of doing so. One way is to ensure that the two arguments are syntactically identical copies of one another: (9)

The boat sinkr the boat

When this configuration is created, that is to say, when two identical copies are around (in the correct configuration), one of them does not get to be pronounced (Le. it is a trace). Another conceivable way to ensure identity of arguments is to resort to REFL (Le. 'real' reflexivization): (ro)

The boat REFL(sinkr)

RELF gets rid of (saturates, discharges, or whatever you want to call it) the inner argument of a relation by forcing it to be identical with the outer argument. This second strategy is tantamount to the composition of REFL with [ y. If REFL is associated with

A Semantics for Unaccusatives

57

special morphology, we would expect such morphology to show up whenever this second option is selected. These two strategies appear to be exactly the two main syntactic realizations of unaccusative verbs, both within and across languages; the first one involves movement (i.e. copying); the second, reflexive morphology. Thus, across languages, the mapping from sinkr onto its admissible syntactic manifestations is in some sense both trivial and forced upon us by the very nature of []r, which in turns rests on a-sharing. I find it tempting to conclude that something like is indeed what is behind unaccusativity, the piece needed to integrate and understand the role that REFL seems to play in it. One immediate consequence of this view is that verbs that cannot take eventualities as overt subjects (for example, verbs that impose animacy requirements on their subject) obviously cannot undergo reflexive closure. Thus verbs like notice, love, or psych verbs of the fear class will be unable to undergo reflexive closure and hence will not have unaccusative alternates. This simply follows from the fact that one cannot say things like the explosion notices/fears Bill and consequently a-sharing does not apply. On the other hand, psych verbs of the piacere ('please') or scare type in principle can undergo reflexive closure, for they patently have the relevant property of allowing eventualities as subjects (cf., for example, the statue scaredJohn vs. seeing the statue

[r

scaredJohn). Does this proposal derive the main fact about unaccusative transitive intransitive alternation, namely (4)? Yes, on one assumption. The assumption one needs is the following. If an ordinary individual x sinks y, then there must be an eventuality e of which x is the agent that stands in the sink-relation to y. Let me be explicit about it: (n)

Assumption sink (x, y, e') ---+ 3e [Ag(e,x) /\ sink (e, y, e')]

This ensures that if x sinks y, there is an eventuality that causes the sinking (typically, what x does) out of which we can extract the argument that such eventuality shares with the actual sinking. For my money, assumption (n) is likely to be correct. As a concrete illustration of this idea, we might tal<:e Parsons's (1990) proposal on the semantics of sink, namely:

(12)

a. b.

John sank the boat 3e 3e'[Ag(e,John) /\ cAusE(e, e') /\ Th(e', the boat) /\ SINK(e')]

The assumption in (n) is a logical consequence of the semantic representation for sink at the basis of example (12).'4 We may of course disagree on how to derive (n) (via lexical decomposition or not, via projection in the syntax of the relevant semantic heads or not, etc.). We may also disagree on how to explain the less than perfect synonymy between pairs like (13a, b).

'4

I ignore tense here. Moreover, I assume that the verb sink is defined in the obvious way, namely:

sink = AeAYAe' [1h(e', y) 1\ CAUSE(e, e') ASINK(e')]

58

(13)

Gennaro Chierchia

a. John broke the window. b. John's blow broke the window.

These matters are important and substantive. But in the end, someone's breaking something does necessarily correlate with an event that brings the breaking about. And this should suffice for our observations about 8-sharing to apply: the causing event must share a 8-role with the event that is being caused. I have mentioned several times that the net effect of [Y is maldng a relation reflexive. It might be useful to actually visualize how: (14)

a.

(a, b, e,) (e,b, e,) K= (c,d,e,) (e',d, e2 ) (e",f,ej )

b.

(b, b,e,) Kr = (d,d,e,) (f,f,e)

Think ofK as a relation where the first member of the triplet corresponds to the subject, the second to the object, and the third is the Davidsonian argument. Imagine a situation where {a, b, c, d, f} are ordinary individuals, ie, e', eft, e" e v ej } are eventualities and the extension of K is as in (14a). Thus there are three K-eventualities (e" e2 , ej ), three eventualities that cause them (e, e', eft) and a and c are the agents of e and e' respectively, while eft does not have an agent. Applying [ ]r to K in such a situation yields (14b). Essentially, the subject argument is deleted and the object argument gets copied in its place. This is done by pulling out of the causing arguments (namely, e, e', eft) what it shares with the event en which is being caused. The operation is both easy to grasp and formally simple. Rule (14b) shows how an unaccusative relation would actually look according to the present hypothesis. The properties of unaccusatives appear to be readily derivable if we assume that they look as in (14b). Notice, for instance, that the main fact about unaccusative alternations (those in (4)) comes out correctly. Notice, moreover, that the subject argument of the unaccusative relation Kr is identical to its object argument (an inherent reflexivity of sorts) and that such an argument is non-agentive. Plus, we have already remarked on why something that looks like (14b) would have to be structurally realized either exploiting NP movement or reflexive morphology. Summarizing, such transitive relations as sink, break, etc. admit of eventualities as subject. Moreover, if ordinary individuals can be taken as subjects of such relations, then there is an eventuality of which they are agents that causes the sinking, breaking, etc. (assumption (II)). In either case, we have an eventuality that brings about the sinking, breaking, etc. The causing eventuality will share an argument (typically the theme) with the event being caused (8-sharing). It is in this form of8-sharing that reflexivity is hidden. All we have to do is pull out the argument that the causing eventuality shares with the eventuality that is being brought about. One way of doing so is via the operation [ Yof reflexive closure. The effect of such an operation is focusing on the portion of the causing event which specifically involves the object ('focusing

A Semantics for Unaccusatives

59

on the inner cause', to borrow Levin and Rappaport Hovav's term) and thus factoring out external causing agents. Pure REFL by itself cannot suffice to explain unaccusative alternations. It fails to derive what we called the main property of such alternation (see (4». REFL plus [ Y can, with minimal and straightforward assumptions on the syntax/semantics map. To put it in slightly different words, Universal Grammar surely has a number of devices designed to alter the argument structure of relations. The suggestion I am maldng is that among such devices, there are two forms of reflexivization: plain vanilla REFL and reflexive closure [Y (or, if you wish, the composition of REFL with []'). The former is potentially total, in the sense that it can in principle apply to any relation. The latter is inherently partial. It can only apply to relations that allow event arguments as subjects. It exploits 8-sharing and its communicative purpose is the elimination from the argument structure of the verb of the external cause. 15 The complexity of the relevant phenomena surely requires more elaborate mapping hypotheses than those we could entertain here. '6 My guess, for what it is worth, is that such necessary elaborations can and must maintain a variant of the present idea (inherent reflexivity) as their semantic core. At any rate, I keep feeling that trying to be as explicit as possible as to the semantic import of unaccusativity is a fruitful heuristic. '5 [1' corresponds to R] in the chapter. Both novelty and continuity with regards to the general line adopted there should be fairly clear. Reflexive Closure, as developed in this postscript, is independent of property theory and of any particular hypothesis on the syntax semantics map. It also mal(es explicit the relation to the event based approaches that have developed since. Finally, it factors out differently pure reflexivity from what is special about unaccusatives formation. ,6 One relevant area is that of Aktionsart. On this issue, I do not share Reinhart's (1996) pessimism as to its relevance to unaccusativity. Unaccusatives tend to be either telic or (more rarely) stative. Atelic verbs that display unaccusative properties are of very special types: inchoatives such as cool and motion verbs like roll. The former are iterations of (telic) achievements; the latter when in unaccusative forms tend to behave in a telic way.

2 Unaccusativity as Telicity Checking Angeliek van Hout

2.1.

INTRODUCTION

Which lexical-semantic properties determine unaccusativity-the division of intransitive verbs into two syntactically distinct classes-and how does the lexicon-syntax mapping system account for this? I argue that a single factor plays a role in Dutch: telicity. No effects are attested for agentivity or notions such as internal versus external control. I propose a configurational theory of unaccusativitywhich is driven by feature checking of this aspectual feature, and has no use for thematic roles. The role of telicity in the lexicon-syntax interface of two-argument verbs predicts the syntax of one-argument verbs: telic one-argument verbs are unaccusative, whereas atelic one-argument verbs are unergative.' I argue that two-argument verbs need to

The write-up of this research was supported by grant 300-75-025 from NWO (Netherlands Organization for Scientific Research) which is hereby acknowledged. I thank the organizers and participants of the ZAS workshop for their questions and comments, and the editors and Alice ter Meulen for remarks and comments on an earlier version. I I use the terms 'one-argument verb' and 'two-argument verb' when I refer to a verb's lexical property of taking one or two arguments, and reserve the terms 'transitive' and 'intransitive' for the syntactic verb frames in which verbs appear in sentences. I will not be using any additional lexical features such as mark-

Unaccusativity as Telicity Checking

61

project a direct object in order to establish a telic interpretation. I model this generalization at the lexicon-syntax interface where the aspectual feature of telicity must be syntactically checked in object position; specifically, AgrOP is the locus of telicity checking. A natural consequence of this approach is that telic one-argument verbs must also check their telicity feature in AgrOp, which, I argue next, yields unaccusativity. On this view, unaccusativity reflects a particular syntactic movement configuration for one-argument verbs in which the single argument moves through the object position (specifier position of Agr, 0) on its way to the subject position (specifier position ofAgr,S), where it needs to satisfY the Extended Projection Principle (EPP). Movement through object position is triggered when the predicate is telic. This constitutes the syntax of unaccusativity. Unaccusativity, then, is a semantically determined syntactic reality. The single argument of an atelic predicate, on the other hand, moves straight to the specifier ofAgrS to satisfY the EPP. This latter movement constitutes unergative syntax. The unergative and unaccusative configurations are illustrated in (1).

(1)

a.

b.

Un ergative syntax

Unaccusative syntax

AgrSP

AgrSP

~

~

Spec DPi

AgrS'

~ AgrS [+EPP]

VP

D

Spec DPi

AgrS'

~ AgrS [+EPP]

AgrOP

~ Spec

AgrO'

~

AgrO [+telic]

VP

D

Data from Dutch one-argument predicates support the prediction that telic oneargument verbs are unaccusative. The unaccusative diagnostics in Dutch yield two verb classes that split along aspectuallines. Telic one-argument predicates select auxiliary zijn ('be') and their perfect participle forms can be used as prenominal modifiers; atelic one-argument verbs, on the other hand, selecthebben ('have') and their participles cannot be used prenominally. The claim that there is a role for telicity in defining unaccusativity is not new. In the 1990S it became increasingly clear that telicity is involved in defining or co-defining unaccusativity (Grimshaw 1990; Borer 1994; Tenny 1994; Levin and Rappaport Hovav 1995). Here I provide further data and arguments for its crucial role and show that other lexical-semantic properties are not involved. Offering a more comprehensive theory of the lexicon-syntax interface, including both one- and two-argument verbs, I derive a more principled explanation of the role of telicity for unaccusativity. ing a verb's arguments lexically as internal or external arguments, nor will I use thematic-role labels, except occasionally ill order ro point to one of the two arguments of a two-argument verb.

Angeliek van Hout In section 2.2, argument alternations with two-argument verbs lead to the generalization that telicity requires the projection of a direct object. I formulate an aspectual mapping theory that implements this generalization. Working with two unaccusative diagnostics-auxiliary selection and participial modification-section 2.3 then shows how this generalization makes the right predictions for one-argument verbs. This leads to the configurational definition of unaccusativity illustrated in (lb). In section 2·4 I discuss how this analysis compares to other theories of unaccusativity, including other aspectual ones.

2.2.

TWO-ARGUMENT VERBS AND TELICITY

Tenny (1994) points out a strong connection between telicity and direct objects, show2 ing that aspect is involved in the lexicon-syntax interface. In section 2.2.1 I examine so-called variable-behaviour verbs in Dutch: verbs which occur in multiple verb frames. A comparison of these verbs with their particle and prefixed variants shows clearly that telicity requires transitivity. I summarize the analysis I proposed in van Hout (1996, 2000), which connects this generalization to object Case and telicity checking in AgrOP in section 2.2.2. The data in this article are drawn from Dutch, but they hold more generally for Germanic languages.

2.2.1.

Direct objects and telicity

The aspectual effects of linking variable-behaviour verbs onto different verb frames show that only verbs in a transitive frame can yield telicity. An intransitive frame without a direct object or a verb in what I will call a conative frame with an oblique object does not give telicity.3 Consider the variants of schrijven ('write')-intransitive in (2a) , conative in (2b), and transitive in (2C)-and see how the temporal modifiers bring out telicity and atelicity. (2)

a.

b.

Elena heeft jarenlang/*binnen een jaar geschreven. Elena has years-long/within a year written 'Elena has been writing for years/*within a year.' Elena heeft jarenlangl*binnen een jaar aan haar proefschrift geschreven. Elena has years-long/within a year at her dissertation written 'Elena has been writing (away) at her dissertation for years/*within a year.'

2 I use 'aspect' to refer to lexical aspect (or Aktionsart, situation-type aspect), not grammatical (or: viewpoint) aspect. Telicity is determined at the VP level, not simply by a verb itself, so I mostly use the term 'telic predicate', including the verb, its object and other aspectually relevant elements in the VP, such as directional PPs and particles, rather than 'telic verb'. J The term 'conative frame' is typically used for predicates like shoot at the bird, which have an attempt reading (cf. Latin conare ('attempt')). I extend its use here to all verb frames in which the verb's second argument appears, not as direct object, but as oblique object-that is, in a PP in Dutch.

Unaccusativity as Telicity Checking

c.

Elena heeft haar proefschrift *jarenlang/binnen een jaar geschreven. Elena has her dissertation years-long/within a year written 'Elena has written her dissertation *for years/within a year.'

Although schrijven ('write') can appear in multiple frames, one of its particle variants, afichrijven ('finish writing') can only appear in a transitive frame; it requires an argument to be projected as direct object, see the examples in (3). The same goes for other particle verbs derived from schrijven: opschrijven ('write up'), inschrijven ('register'), voorschrijven ('prescribe'), uitschrijven ('take out of registration'). (3)

a. *Elena heeft jarenlang afgeschreven. Elena has years-long off-written b. *Elena heeft aan haar proefschrift afgeschreven. Elena has at her dissertation written c. Elena heeft *jarenlang/binnen een jaar haar proefschrift afgeschreven. Elena has years-long/within a year her dissertation off-written 'Elena finished writing her dissertation *for years/within a year.'

This pattern is not particular to schrijven. Many two-argument verbs and their particle verb and prefixed variants exhibit the same atelic-telic effects in the transitive-intransitive-conative alternation. These involve so-called verbs of destruction and creation. More examples include the following. (4)

eten 'eat' roken 'smoke'

opeten 'eat up' oproken 'finish smoking', uitroken 'smoke out', beroken '(blacken with) smoke' lezen'read' uitlezen 'finish reading', overlezen 'read over', doorlezen 'read over', herlezen 'reread' zingen 'sing' nazingen 'copy by singing', bezingen 'celebrate (in song)' spelen 'play' uitspelen 'finish playing', overspelen 'play again' graven 'dig' opgraven 'dig up', uitgraven 'dig out', afgraven 'dig off', begraven 'bury' tekenen 'draw' uittekenen 'work out by drawing', natekenen 'copy by drawing', intekenen 'register' verven 'paint' opverven 'paint up', oververven 'repaint', afoerven, 'finish painting' bakken 'fry' uitbakken 'fry (till all fat is out)', aanbakken 'fry (a little)' knippen 'cut' verknippen 'cut wrongly', uitknippen 'cut out', inknippen 'cut in' snijden 'cut' insnijden 'cut in', afinijden 'cut off' slurpen 'slurp' opslurpen 'slurp up' bouwen 'build' verbouwen 'renovate', ombouwen 'rebuild (into something else)', uitbouwen 'extend by building' scheuren 'tear' verscheuren 'tear up' , aficheuren 'tear off', uitscheuren 'tear out' smelten 'melt' versmelten 'melt (away)', omsmelten 'melt (into something else)'

Angeliek van Hout The morphologically simplex verbs correlate with different aspectual semantics across different verb frames: telic readings with a transitive frame and atelic with intransitive and conative frames. These morphologically simplex verbs themselves are neither telic or atelic. Rather, the verbs acquire their aspectual value by virtue of the verb frame they occur in. In other words, it is at the VP level (verb plus its object), not the lexical verb, that (a)telicity is determined. Particle and prefixed variants, however, do not show variable projection behaviour, nor do they have different aspectual options. All particle and prefixed variants listed above are inherently telic. Their meanings differ slightly from the simplex variants in that the particle or prefix specifies a notion of culmination added on to the verb's original meaning. Different particles spell out different kinds of culmination moments, as indicated by the glosses. Importantly, these inherently telic verbs have no option but to appear in a transitive frame; their verb frame is fixed. Notice that it matters which particle variant one chooses. Another variant of schrijven, doorschrijven ('continue writing') , differs from any of the particle verbs just listed. It cannot project its argument as a direct object, (5c), only as oblique object, (5b), or it must leave it implicit, (5a).

(5)

a.

Elena heeft jarenlang/*binnen een jaar doorgeschreven. Elena has years-long/within a year through-written 'Elena has been writing on for years/*within a year.' b. Elena heeft jarenlang/*binnen een jaar aan haar proefschrift Elena has years-long/within a year at her dissertation doorgeschreven. through-written 'Elena has been writing on at her dissertation for years/*within a year.' c. *Elena heeft haar proefschrift doorgeschreven. Elena has her dissertation through-written

The particle door is actually fairly productive; it can take any activity-denoting verb, yielding the meaning of 'Von' or 'continue V-ing'. Note that this meaning is inherently atelic; doorschrijven is an atelic predicate without a culmination moment. Such atelic particle verbs are flexible, appearing in an intransitive or conative frame, but do not allow their second argument to appear as direct object. This flexible or inflexible projection pattern can be summarized by the generalization in (6). Schoorlemmer (in this volume) reaches exactly the same conclusion for Russian.

(6)

Generalization over atelic-telic alternations with two-argument verbs In order to form a telic predicate, the verb projects an argument in direct object position.

2.2.2.

Telicity checking and object case

In the 1990S many researchers have noted the association between telicity and direct

Unaccusativity as Telicity Checking objects; equally many theories have been proposed to implement this association. 4 The proposal that I have developed in van Hout (1996, 2000) presents a particular implementation of the idea that the semantic notion of telicity also figures as a syntactic entity. This theory is summarized below. Given that verb-frame alternations are expressions of different event types (telic or atelic), mapping is sensitive to a verb's event type, which must thus be talcen as a syntactically relevant property. I propose that telicity is a semantically meaningful, morpho-syntactic feature that may be introduced into a computation. In a Minimalist framework (Chomsky 1995), the correlation between telicity and direct objects is captured by what I call event-feature checking. Telicity is checked in direct object position, that is, in the functional layer above VP-that is, AgrOP. A telic feature is checked in AgrOP by way of Specifier-Head agreement, attracting an object argument to its specifier position. Telicity checking correlates with a particular kind of object case, namely, strong Case-explained below-which also talces place in AgrOP. Given that other predicative elements than the lexical verb determine the event type of a verbal predicate, including prefixes and particles, the lexicon-syntax mapping is designed to look at aspectual properties of the whole VP in which the verb appears, instead of projecting up from purely lexical properties of the verb alone as in the tradition of standard lexicon-syntax theories. The data in section 2.2.1 have shown that, in order to get a telic reading, there must be an object. However, not just any transitive verb frame gives telicity. There is an additional factor: the semantics of the object. 5 Verkuyl (1972) observed that only objects that denote a specific quantity of stuff induce telicity; direct objects that denote homogeneous stuff, such as mass terms or bare plurals, do not. In Krifka's (1989) terminology, which I will follow here, non-homogeneous, telicity inducing objects are called 'quantized' and homogeneous ones 'non-quantized'. Consider the various objects of eten ('eat') in (7). (7)

a.

b.

Claartje heeft *urenlang/ in IO minuten een spekulaasje/ +twee Claartje has *hours-long/in IO minutes a ginger-cookie/two spekulaasjes gegeten. ginger-cookies eaten 'Claartje ate a ginger cookie/two ginger cookies *for hours/in IO minutes.' Claartje heeft urenlang/ *in IO minuten spekulaas/ spekulaasjes Claartje has hours-long/*in IO minutes gingerbread/ginger-cookies gegeten. eaten 'Claartje has eaten gingerbread/ginger-cookies for hours/*in IO minutes.'

4 These include van Voorst (1988), Van Valin (1990), Dowty (1991), Pustejovsky (1991), Levin and Rappaport Hovav (1992a, 1995), McClure (1993), Borer (1994), den Dikken (1994), Tenny (1994), Travis (1994), Schmitt (1995), Ramchand (1997), and Arad (1998b). , There are in fact additional requirements. The verb must not be stative and it must have the lexical property of incrementality. The latter property differentiates verbs of destruction and creation from verb like push and carry, which do not give telicity unless there is a directional PP or particle to express a path.

66

Angeliek van Hout

Even though both alternants are transitive, only (7a) is telic because the objects are quantized, een speculaasje ('a ginger cookie') and twee speculaasjes ('two ginger cookies'). The mass-term object speculaas ('ginger bread') and the bare plural speculaasjes ('ginger cookies') in (7b), on the other hand, are non-quantized and the predicate is atelic. See Verkuyl (1972, 1993) and Krifka (1989, 1992) for two formal analyses of this mapping between objects and events. There is in fact a three-way interaction between a predicate's event-type features (telic versus atelic), particular argument positions in the syntactic configuration (direct object versus oblique object), and the semantics of the noun phrase in object position (quantized versus non-quantized). All three properties interact. If there is no direct object, as in intransitive or conative verb frames, there can be no telicity. If there is an object in a transitive frame, but it is non-quantized, as with mass terms and bare plurals, there is no telicity either. A theory of telicity needs to contain all three kinds of properties as its ingredients. In order to incorporate the effect of the semantics of the object noun phrase in the syntax-semantics interface, I use de Hoop's (1992) Case theory and van Geenhoven's (1996) refinement of it. De Hoop claims that an object's morpho-syntactic case is associated with the semantic interpretation of the object noun phrase. She introduces the syntactic notions of strong and weak Case and relates them to strong and weak readings of noun phrases, respectively. Following de Hoop's suggestion that the strong-weak Case distinction is also associated with telicity, I argue that telicity is correlated with an object with strong Case. I propose that telicity in the syntax-semantics interface is indirectly established via object Case. The objects in (7a) have a strong reading, therefore they carry strong Case. 6 They yield telicity since strong Case gives a quantized reading. The mass term and the bare plural objects in (7b), on the other hand, have wealc readings/They carry weak Case and are semantically incorporated, with the effect that they yield atelicity-see van Hout (2000) for further details. Hence, object Case is indirectly associated with telicity through the association of the strong-weak object Case distinction-a syntactic distinction-with different noun-phrase interpretations. Different readings are associated with (non)-quantization of the noun phrase which on their turn determine (a)telicity. The strong-weak Case distinction has a syntactic consequence as well, as the two cases are associated with different object positions. Basically, weak objects remain in VP whereas strong objects move out to the functional layer. AgrOP determines the checking domain for strong Case checking. Strong Case thus attracts the object to the Specifier of AgrOP. Weak Case, on the other hand, is checked inside the VP in the object's base position. Hence, quantized objects may raise to AgrOP to check strong 6 A strong reading involves either a generalized quantifier reading of type «e, t), t) or a referential reading of type e. 7 Adjusting de Hoop's Case theory, van Geenhoven (I996) argues that objects with weak Case are predicative indefinites of type (e, t). These predicates are absorbed by an incorporating verb of type «e, t), (e, t» in a process of Semantic Incorporation and as such they introduce a restriction on the individual(s) that the verb applies to. Their existential interpretation derives from the lexical semantics of the incorporating verb, yielding existential quantification of the direct object.

Unaccusativity as Telicity Checking Case, but mass term and bare plural objects must stay in the VP. The latter cannot raise to AgrOP to check strong Case because their meanings are semantically incompatible with strong readings, which would be triggered by the strong Case that is assigned in that position. In sum, telidty is taken as a morpho-syntactic feature. It is an interpretable feature and its checklng domain is AgrOP. Telldty attracts an object to the spedfier of AgrOP. This is stated in (8).

(8)

Telicity checking A telic event type feature is checked via Spedfier-Head agreement in AgrOP. It triggers movement of a noun phrase to the Spedfier of AgrOP. The predicate's event-type properties must be compatible with the tellc feature.

In this analysis the telidty-checklng requirement explains the generalization in (6) at the end of the previous section. The assumption that telldty checking involves AgrOP derives the generalization that a telic predicate requires the presence of an object. A derivation without an object, that is, an intransitive or a conative frame, wlll not converge. Since AgrOP is also the locus of strong Case assignment, telidty checking correlates with strong Case assignment (at least, in the case of two-argument verbs; see below for one-argument verbs). Quantized NPs check strong Case upon raising to the Spedfier of AgrOP. They also check telidty there. Non-quantized NPs such as mass terms and bare plurals, on the other hand, remain in the VP, where they get weal{ Case and thus do not check telldty. This yields the following three types of derivation for two-argument verbs. First, a telic predicate has a quantized object with strong Case in the spedfier of AgrOP; (9) lllustrates the crudal part of the derivation of (7a). Second, an atelic predicate has a non-quantized object with weak Case in VP; (10) lllustrates the derivation of (7b).8,9Third, an atelic predicate may have a quantized object as an obHque object; (12) lllustrates the derivation of a conative verb frame such as (II).

(9)

Telic two-argument predicate in transitive frame with AgrOP AgrSP

~

AgrS'

Spec

DPi

~

AgrS

Spec

DPj

AgrOP

~

AgrO'

~

[strong Case] AgrO [Helic]

VP

D

Objects with weak Case may be NPs rather than DPs. I will not explore this here. Atelic transitive predicates with verbs from the push class can have a quantized object and still be atelic (e.g., push a cart/two carts for hours). These involve a derivation with the object in AgrOP, but without a telicity feature. Whereas verbs of creation and destruction are compatible with a telicity feature, push-verbs are not, unless they appear with a directional PP or particle. See n. 5. 8 9

68 (10)

Angeliek van Hout Atelic two-argument predicate in transitive frame with no AgrOP AgrSP

~

Spec DPj

AgrS'

~ AgrS

VP

~

V'

~

V

NPj

[weak Case] (II)

(12)

Claartje heeft van het spekulaasje gegeten. Claartje has from the ginger-cookie eaten 'Claartje ate from the ginger-cookie.' Atelic two-argument predicate in conative frame with no AgrOP (object inside PP) AgrSP

~

Spec DP j

AgrS'

~ AgrS

VP

~

V'

~

V

PP

~

P

DPj [oblique Case]

This theory of telicity at the lexicon-syntax interface deals with two-argument verbs and their objects. What does it involve for telic one-argument verbs?

2.3.

ONE-ARGUMENT VERBS AND TELICITY

The telicity-checking theory as defined for two-argument verbs predicts unaccusative syntax precisely for telic one-argument predicates, which must also have their telicity checked in AgrOP (section 2.3.1). I apply two unaccusative diagnostics to Dutch oneargument verbs (section 2.3.2) and show that these diagnostics support this prediction (se_ction 2.3.3). I then investigate the role of agentivity for the classification of intransitive verbs with the same diagnostics and find that it is not involved (section 2.3.4).

Unaccusativity as Telicity Checking I will argue that the analysis offers part of an explanation of Burzio's (1986) generalization (section 2.3.5).

2.3-1. Extending telidty checking to one-argument verbs Applying the same telicity-checking requirement in the case of one-argument verbs, a natural consequence is that the single argument of telic one-argument verbs must move to AgrOP to check telicity. It will then move on to subject position to fulfil the Extended Projection Principle (EPP).1O In van Hout (1996, 2000) I argue that this kind of movement configuration constitutes unaccusativity; (13) is repeated from above."

(13)

Telic one-argument predicates with unaccusative syntax AgrSP

~

Spec DPi

AgrS'

~ AgrS

AgrOP

~

Spec

AgrO'

ti

~ AgrO [+telic]

VP

D ti

The mapping theory predicts that telic one-argument verbs are unaccusative because they need to check their telicity in AgrOP; atelic verbs, on the other hand, are unergative. In this section I will examine the role of telicity for two unaccusative diagnostics in Dutch, namely, auxiliary selection and prenominal participial modification. I will focus the discussion in particular on variable-behaviour verbs that are sometimes unergative and at other times unaccusative. They form minimal pairs and show most clearly which semantic factors are involved in unaccusativity. First, I brieRy discuss the unaccusative diagnostics in Dutch.

2.3.2. Unaccusativity diagnostics in Dutch Perlmutter (1978) discovered that the class of one-argument verbs divides in two types-unergatives and unaccusatives. He observed that in a number of construcW For lack of a better theory I use the EPP from Principles and Parameters terminology to refer to the effect that single arguments must move on after having checked telicity in AgrOP, because sentences need subjects. II Borer (1994) proposes basically the same configuration on the basis of a different line of reasoning. See the comparison of unaccusativity theories in section 2.3.4.

Angeliek van Hout dons, the subject of an unaccusative verb behaves similar to an object of a transitive verb and different from a subject of a transitive or unergative verb; subjects of unergatives behave just as transitive subjects. Following this up in the generative framework, Burzio (1986) concluded that subjects of unaccusatives are underlyingly objects, just like subjects of passives. I focus here on two constructions that show the split: auxiliary selection and participial modification. Auxiliary selection has been taken by many as a valid unaccusative diagnostic since Perlmutter's (1978) original discussion (see Grimshaw 1987, for instance). It is found in a number of languages, including Dutch, German, Italian, and Basque. Levin and Rappaport Hovav (1995) point out that it is unclear why auxiliary selection should be sensitive to argument structure, movement of an underlying object to subject position, or, in the present framework, movement of the single argument via AgrOP. Yet, in languages that have two different perfect auxiliaries for their intransitives, such as Dutch, unergatives such as lachen ('laugh') in (14a) and transitives such bakken ('bake') in (14c) choose HAVE, whereas unaccusatives such as vallen ('fall') in (14b) select BE.

(14)

a.

b.

c.

Susan heeft gelachen. 'Susan has laughed.' De vaas is gevallen. the vase is fallen 'The vase has fallen.' Inge heeft een appeltaart gebald{en. Inge has an apple-pie baked 'Inge has baked an apple pie.'

Levin and Rappaport Hovav (1995) take auxiliary selection to be a diagnostic of what they call 'deep unaccusativity', which means that the diagnostic is sensitive to the D-structure status of the single argument. In Italian, for example, unaccusative verbs take essere ('be') independent of whether the subject appears pre- or post-verballythat is, in S-structure subject or object position. Everaert (1995) shows that auxiliary selection in Dutch turns out to be sensitive to semantic as well as syntactic properties. The best generalization for Dutch seems that it is a sufficient, but not necessary, condition that the verb be intransitive and telic to select zijn ('be'). Granting that there is a definite need for a theory that explains why unaccusatives take BE as their auxiliary and unergatives HAVE (and not the other way around, for example), I will here work on the premise thatit is a valid diagnostic adding the following argument. If it turns out that in addition to auxiliary selection, another diagnostic in the same language classifies the intransitives along the same lines, the validity of auxiliary selection as an un accusative diagnostic is supported. Below I will explore the role of telicity for auxiliary selection and compare it in each case with another purported diagnostic: prenominal participle modification. The two diagnostics indeed work together in each case and divide the intransitive predicates in the same two groups. Perlmutter found that prenominal participial modification is another unaccusativity diagnostic in Dutch. The pattern shows that objects of transitives and subjects

Unaccusativity as Telicity Checking of unaccusatives behave alike and are distinguished from subjects of unergatives and transitives. When used as a prenominal modifier, a perfect participle of a transitive verb can be used to modifY the object (see 15a), but not the subject (15b). A participle of an intransitive verb can be used to modifY the subject of an unaccusative (as in 15c) but not an un ergative (see 15d). Participial prenominal modification is therefore a deep unaccusativity diagnostic, because it is sensitive to the underlying position of the single argument.

(15)

a.

de gebakken appeltaart 'the baked apple pie' b. *de gebald{en kok 'the baked cook' c. de gevallen vaas , 'the fallen vase d. *het gelachen meisje 'the laughed girl'

Zaenen (1993) surveys another alleged unaccusative diagnostic in Dutch-impersonal passivization-and finds that it cuts diametrically across the division drawn by auxiliary selection and participial modification. Impersonal passivization is sensitive to controllability, whereas auxiliary selection and participial modification are sensitive to telicity. Zaenen concludes that impersonal passivization is not a reliable test for unaccusativity; only the latter two are good diagnostics. In particular, prenominal modification shows clearly how subjects of unaccusatives pattern with objects of transitives, and subjects of unergatives with subjects of transitives. I will not discuss here why auxiliary selection and participial modification are sensitive to unaccusativity, but simply act on the premise that they are good unaccusative diagnostics. Future work will have to explain how exactly these constructions are sensitive to the underlying object position of the verb's single argument or its movement from object to subject position via AgrOP.

2.3.3. Unaccusativity and telicity Variable-behaviour verbs have played an important role in analyses of unaccusativity (Hoekstra 1984; Hoekstra and Mulder 1990; Levin and Rappaport Hovav 1992a, 1995; Borer 1994; Arad 1998b). Verbs that sometimes show unergative behaviour and sometimes unaccusative, depending on context, are interesting because they can shed light on which semantic factors contribute to unaccusativity. I will now show that auxiliary selection and participial modification yield the same division of one-argument predicates into two classes. Next I show that a predicate's telicity, as predicted, is the sole semantic factor that defines the class of unaccusatives; there is no role for mapping rules defined on agentivity or immediate cause. I will examine the behaviour of these flexible verbs in the context of additional PPs, particles, or prefixes. A first illustration comes from verbs of manner of motion such as lopen ('walk').

Angeliek van Hout Basically, lopen is an atelic predicate; it combines with a durative time adverbial (I6a). When it talces a directional PP the predicate becomes telic; naar huis lopen ('walk home') is non-homogeneous and combines with a frame adverbial, (I6b). Auxiliary selection is sensitive to telicity: the former takes hebben ('have'), the latter, zijn ('be'). This pattern is firmly established in the unaccusativity literature.

(16)

a.

b.

John heeft urenlang gelopen. John has hours-long walked 'John walked for hours.' John is in vijf minuten naar huis gelopen. John is in five minutes to home walked 'John walked home in five minutes.'

Applying participial modification to these two variants of lopen shows that atelic lopen cannot be used a prenominal modifier, while telic naar huis gelopen ('to home walked') can.

(17)

a. *de gelopen jongen

b.

the walked boy de naar huis gelopen jongen the to home walked boy 'the boy who walked home'

So, both diagnostics indicate that lopen, when it appears in an atelic predicate, is unergative, but in a telic predicate it is unaccusative. Similar unergative-unaccusative alternations based on the telic-atelic distinction, are found with other kinds of verbs. Compare how the diagnostics work for atelic slapen ('sleep') in (I8a-a') and its telic particle variant inslapen ('fall asleep') (literally, 'insleep') in (I8b-b'). Atelic slapen selects hebben. Its participle cannot be a prenominal modifier of the single argument, whereas telic inslapen selects zijn, and its participle is a fine modifier.

(18)

a.

Winnie heeft dagenlang geslapen. Winnie has days-long slept 'Winnie slept for days.' a'. *de geslapen jongen the slept boy b. Winnie was binnen vijf minuten ingeslapen. Winnie was within five minutes in-slept 'Winnie fell asleep within five minutes.' b'. de ingeslapen jongen the in-slept boy 'the boy fallen asleep'

Another illustration comes from verb pairs with a prefixed variant. One example is the pair branden-verbranden ('burn-burn (away)'). The prefixed verb is the telic vari-

Unaccusativity as Telicity Checking

73

ant in this pair while the simplex verb is atelic. Again the diagnpstics work together and classify branden in (19a-a') as unergative and verbranden in (19b-b') as unaccusative.

(19)

a.

Het houtblok heeft urenlang gebrand. the log has hours-long burned 'The log burned for hours.' a'. *het gebrande houtblok the burned log b. Het houtblok is in een uur verbrand. the log is in an hour VER-burned 'The log burned (away) in an hour.' b'. het verbrande houtblok the VER-burned log 'the burned log'

Some verbs, drogen ('dry') for example, can freely take on a telic or atelic interpretation as brought out by the context. With the help of the durative temporal modifier urenlang ('for hours') in (20a) versus the frame adverbial in een uur ('in an hour') in (2ob), drogen can be established as an atelic or telic predicate, respectively. The auxiliary varies along the lines of the telic-atelic distinction.

(20)

a.

b.

De was heeft urenlang gedroogd, maar bleef toch vochtig. the laundry has hours-long dried, but remained still damp 'The laundry dried for hours, but still remained damp.' De was is in een uur gedroogd. the laundry is in an hour dried 'The laundry dried in an hour.'

Looking at the past particle gedroogd ('dried'), it can modify the single argument in (21a), but only under the telic interpretation-that is, with the entailment that the laundry is dry; (21b) is not felicitous. (21)

a.

de gedroogde was 'the dried laundry' b. Ide gedroogde was die nog steeds vochtig is the dried laundry that still damp is

Drogen has a telic particle variant opdrogen ('dry up') in (22a-a') and a telic prefixed variant verdrogen ('dry out') in (22b-b'). Both classify as unaccusative under the two diagnostics. (22)

a.

Het water is in een uur opgedroogd. the water is in an hour up-dried 'The water dried up in an hour.' a'. het opgedroogde water the up-dried water 'the dried-up water'

Angeliek van Hout

74 b.

Het grasveld is in een week verdroogd. is in a week VER-dried the lawn 'The lawn dried out within a week.' b'. het verdroogde grasveld the VER-burned lawn 'the dried-out lawn' Many other pairs of one-argument verbs and their particle and prefixed variants show an atelic-telic effect and behave in similar ways as the examples in (18)-(22) under the two unaccusative diagnostics. Here are some more examples. (23)

drogen 'dry' reizen 'travel' lopen 'walle'

opdrogen 'dry up', verdrogen 'dry out', uitdrogen 'dry out', indrogen 'dry in' aJreizen 'set off to travel' ajlopen 'expire', inlopen 'warm up for walking', uitlopen 'sprout'

draaien 'turn' , , zwermen swarm bloeien 'flower' gloeien 'glow' groeien 'grow' stromen 'stream' dwalen'roam' koken 'boil' hongeren 'starve' branden 'burn'

rotten 'rot' slijten 'wear' dampen 'steam'

omdraaien 'turn around', uitdraaien 'turn out', indraaien 'turn in', afdraaien 'turn off' uitzwermen 'swarm out', opzwermen 'swarm up' opbloeien 'revive' opgloeien 'glow up', uitgloeien 'glow out' opgroeien 'grow up', vergroeien 'grow to become deformed' wegstromen 'stream away' afdwalen stray', verdwalen 'getlost' inkoken 'boil down', overkoken 'boil over' verhongeren 'starve (to death)' verbranden 'burn away', ontbranden 'ignite', opbranden 'burn up', uitbranden 'burn out', inbranden 'burn in' verrotten 'rot (away), verslijten 'wear out', uitslijten 'wear out', inslijten 'wear in' verdampen 'evaporate', uitdampen 'evaporate'

2.3.4. Unaccusativity and agentivity Having established that the unergative-unaccusative alternation is sensitive to telicity, I will now examine if another semantic factor, agentivity, is also relevant to unaccusativity. In the early theories unaccusatives were defined as verbs whose single argument role is a Theme, whereas unergatives are verbs with an Agent role (see, for example, Burzio 1986; Chomsky 1986; Pinker 1989; Grimshaw 1990). In a lexicon-syntax mapping that obeys a principle such as Baker's (1988) Universal Theta-Assignment Hypothesis, this lexical-thematic difference leads to unaccusativity for one-argument verbs, because an Agent always maps onto subject position, hence linking as unergative, and a Theme onto object position, linking as unaccusative. Many unaccusativity theories of the 1990S have since pointed to the role of telicity (see section 2.4 for further discussion

Unaccusativity as Telicity Checking

75

and comparison), but the role of the Agent versus Theme distinction (or some version of it) still lingers on in most analyses. Does it playa role in Dutch unaccusatives? An unaccusativity theory that incorporates a somewhat different notion than agentivity is proposed by Levin and Rappaport Hovav (1992a, 1995). Along with a linking rule based on telicity, they formulate two further linking rules. These only take effect when the telicity rule does not apply because it takes precedence; so the extra rules apply to atelic predicates. 12 The so-called Immediate Cause Linking Rule links the argument of a verb denoting the immediate cause of an eventuality as the external argument, as unergative. The definition of 'immediate cause' is based on the notions of internal and external causation. If an eventuality can only be controlled by the person or thing engaged in it, this participant is an internal causer and has immediate control. For example, the running of a person initiates from their own actions, the flowering of a rose is given in the natural design of roses, and the ticking of a clock comes about by the clock's internal make-up. Such arguments are mapped as external arguments. If neither the Telicity nor Immediate Cause Linking Rule apply, that is, in the case of atelic verbs with a single argument that is not an immediate cause, a Default Linking Rule maps the single argument as the internal argument, as unaccusative. Levin and Rappaport Hovav analyse the class of roll-verbs as a case in point (roll, bounce, slide, float). Their single argument is not an internal cause of the moving action; instead, they must be put into action by some external cause. These verbs map as unaccusatives. The Dutch verbs rollen ('roll') and lopen ('walk') belong to the class of manner-ofmotion verbs, which are ambiguous between an externally and an internally caused version, depending on the agentivity of the participant. When a person rolls down the hill or across the room, this participant causes his own action, which is thus internally caused. But when a marble is set in motion and starts rolling around, it is externally caused. The Immediate Cause Linking Rule predicts that a predicate denoting an atelic and internally caused event is unergative, whereas an atelic predicate that is externally caused comes out unaccusative by the Default rule. Their telic variants, independent of internal or external control of the single participant, are unaccusative because of the Telicity linking rule which is ordered higher than the Immediate Cause Linking Rule. The data in Dutch do not support this. Instead, they show that only telicity plays a role in unaccusativity. Examine first the set of examples with rollen. In (24) it takes an internal cause participant, a dog, and in (25), a participant that has no direct control, a marble. The purpose clause in (24a) brings out the agentive character of the participant (no such r, In fact, Levin and Rappaport Hovav argue that the proper characterization is not telicity, but involves the notion 'directed change', which subsumes telicity, but also includes two classes of atelic unaccusative verbs: degree achievement verbs such as cool, widen, harden and atelic verbs of inherently directed motion such as rise, descend, and foIl. Verbs of these two classes tal<e the typical atelicity-indicating temporal modifiers such as for hours-but still select the BE auxiliary. The prenominal-participle test also works fine, but only on a telic, culminating, interpretation of these verbs. A verharde weg ('paved road', lit. 'hardened road') is hard, and not somewhat hard. So, one could argue that these verbs are ambiguously telic or atelic and on an atelic interpretation they may still denote quantized events. This issue needs to be further explored.

Angeliek van Hout

purpose clause would be possible in (25a) because the marble does not have control over its rolling). The a-examples give atelic predicates and the b-examples, telic ones, as demonstrated by the temporal adverbials. Both diagnostics-auxiliary selection and participial modification-are applied.

a.

(24)

Leika, de hond, heeft urenlang gerold (om haar energie kwijt te Leika the dog has hours-long rolled (to her energy rid to ralcen). get) 'Leika, the dog, rolled for hours (to get rid of her energy).' a'. *de gerolde hond the rolled dog b. Leika is in vijf minuten de berg afgerold. Leika is in five minutes the hill off-rolled 'Leika rolled down the hill in five minutes.' b'. de van de berg af gerolde hond the from the hill off rolled dog 'the dog that rolled down the hill'

(25)

a.

De knikker heeft minutenlang over tafel gerold. the marble has minutes-long over table rolled 'The marble rolled around on the table for minutes.' a'. #de gerolde knikker the rolled marble b. De knikker is binnen een minuut de tafel afgerold. the marble is within a minute the table off-rolled 'The marble rolled off the table within one minute.' b' . de van tafel afgerolde knikker the off table off-rolled marble 'the marble that rolled off the table'

The examples show that the unergative-unaccusative distinction is set along the lines of telicity only; there is no role for internal or external control. In particular, the cases of atelic internal versus external control are crucial here. Compare (24a-a') versus (25a-a'): both are unergative as determined by auxiliary selection and participial modification, whereas the Immediate Control Linking Rule (in conjunction with the Default Linldng Rule) predicts that the external cause variant should be unaccusative. The next examples with lopen ('walk') make the exact same point. In (26) John is in internal control of his walking, while the water in (27) is not (at least, not clearly). Again, unaccusativity is not sensitive to this distinction. It only regards telicity. (26)

a.

John heeft de hele nacht gelopen. John has the whole night walked 'John walked all night.'

Unaeeusativity as Telicity Checking

77

a'. *de gelopen jongen the walked boy John is in five minuten naar de bus gelopen. John is in five minutes to the bus walked 'John walked to the bus within five minutes.' b'. de naar de bus gelopen jongen the to the bus walked boy 'the boy who walked to the bus'

b.

(27)

a.

Het water heeft urenlang langs de muren gelopen. the water has hours-long along the walls walked 'The water ran along the walls for hours.' a'. *het gelopen water water the run b. Het water is binnen een minuut de zee ingelopen. the water is within a minute the sea in-walked 'The water ran into the sea within a minute.' b'. het in zee gelopen water water the into sea run 'the water that ran into the sea'

To summarize this section, I have examined the role of telicity for flexible behaviour one-argument verbs and applied the two unaccusative diagnostics Dutch offers, auxiliary selection and prenominal participial modification. I have also looked at the role of another semantic factor, agentivity, or rather, internal and external contro1, and found that it does not playa role in the mapping of Dutch intransitives. Telicity alone determines the unergative-unaccusative behaviour of flexible intransitives. I now reformulate the mapping generalization from section 2.2.1, and extend it over one- and two-argument verbs: (28)

Generalization over atelie-telic alternations In order to form a telic predicate, the verb projects an argument in directobject position. This can be either an overt or an underlying object.

2.3.5. Telicity checking, case, and Burzio's generalization What is the relationship between unaccusatives and object Case? Un accusatives-in the aspectual approach outlined earlier-move first to AgrOP to check telicity and move on to AgrSP to conform to the EPP. This means that they do not receive strong object Case. In fact, the strong-weak Case distinction does not apply to them because no argument remains in VP or AgrO P. It only applies to two-argument verbs and has the effect that when the object of a telic predicate moves to AgrOP to check telicity it will receive strong Case there. It also has the effect that telic two-argument predicates cannot take mass term or bare plural objects, because these are not semantically

Angeliek van Hout compatible with the strong readings that strong Case triggers. For one-argument verbs these effects simply do not show, because their single argument receives subject Case, which is not associated with particular readings. Thus, the telic two-argument verb opdrinken ('drink up') is ungrammatical with a bare plural object in (29a), while the telic one-argument predicate van de berg af rotlen ('roll off the hill') can take a bare plural subject in (29b) (to the extent that these kinds of subject are possible in Dutch; an expletive er ('there') subject would be preferred).

(29)

a. *Hij heeft biertjes opgedronken.

b.

he has beers up-drunk Honden zijn van de berg af gerold. dogs are of the hill off rolled 'Dogs rolled down the hill.'

However, there is an association between unaccusativity and case assignment in a different way. Looking at passive and unaccusative verbs in Italian and comparing them to transitives and unergatives, Burzio (1986) offers the generalization in (30). (30)

Burzios generalization Only a verb with an external argument is able to assign accusative Case.

This generalization describes the fact that the internal argument of passives and unaccusatives cannot stay in its object position, because there is no accusative Case for it, since these verbs (by his definition) do not have an external argument. So the internal argument must move to subject position where it can get nominative Case. The telicity-checking theory developed above may offer the starting point of an explanation of this generalization. Assuming that specifier positions can only be occupied once during a derivation, there cannot be another argument that ends up as object in the specifier of AgrOP in the derivation of an unaccusative verb-after all, this position is already occupied by the telicity-checking argument. Hence, a telic one-argument verb (on my definition of the class of unaccusatives) does not allow any other argument in the Specifier of AgrOP after it has moved to the Specifier of AgrSP. In other words, it cannot 'assign accusative Case'. So the telicity-checking approach to mapping derives as one oflts consequences exactly a situation that Burzio's generalization describes and may thus explain half of Burzio's generalization-the half that states the implication that a verb without an external argument (here, a verb whose single argument moves through AgrOP to AgrSP) cannot assign accusative Case. In the telicity-checking theory this is derived by the interaction of three independent sets of conditions: the telicity mapping condition, conditions on structural object Case assignment, and conditions on the filling of specifier positions. The theory developed so far says nothing about the other half of Burzio's generalization-the implication that a verb with an external argument can assign accusative Case. Several chapters in this volume discuss Burzio's generalization and seek to derive it from an approach to unaccusativity that links it to the presence or absence of a vP projection, the projection that introduces external arguments. In particular, Bennis

Unaccusativity as Telicity Checking

79

derives Burzio's generalization by locating the objective Case feature in v and not V, so that a derivation without a vP-that is, without an external argument-cannot assign object case.

2.4.

COMPARING ASPECTUAL AND OTHER THEORIES OF UNACCUSATIVITY

The configurational view on unaccusativity developed above derives Perlmutter's (1978) original generalization that the single argument of unaccusatives shares properties with direct objects of two-argument verbs. In my analysis, subjects of telic oneargument verbs, as well as objects of telic two-argument verbs, move through AgrO .13 The configurational approach to the lexicon-syntax mapping that I advocate takes unaccusativity out of the lexicon by claiming it is not a particular lexical specification that determines the split in two intransitive verb classes. I will compare my approach to other approaches of unaccusativity. The aspectual view on unaccusativity argues against traditional theories that define unaccusatives in terms of the thematic properties of a verb's arguments (Chomsky 1986; Baker 1988; Pinker 1989; Grimshaw 1990; Reinhart r996). Such thematic accounts typically assume the Universal Theta-Assignment Hypothesis: an Agent is mapped onto syntax as an external argument-that is, onto subject position. A Theme, on the other hand, is mapped as an internal argument and originates in object position, from where it must move up to subject position for reasons ofEPP and Case checking. On the aspectual account, thematic labels do not play any role in the linking system whatsoever. The data in (24)-(27) show that the single argument's thematic status cannot distinguish unergatives from unaccusatives. The present account thus leads to the abolishment of thematic roles as driving forces in the lexicon-syntax interface in the domain of one-argument verbs. Instead, it supports mapping theories that do without thematic labels. The role of the EPP remains the same. My approach furthermore does not assume any non-thematic, lexical specification of internal and external argumenthood of a verb's arguments, as originally proposed by Williams (198ra) and followed by many, including several chapters in this volume (Chierchia, Reinhart and Siloni, Steinbach). Variable-behaviour verbs are a case against fixed lexical specification. Depending on context or the presence of particles or prefixes, one-argument verbs sometimes appear as unergative, sometimes as unaccusative. Mapping on the basis of fixed lexical specification would lead to a duplication of the lexical entries for all these verbs, as they would have to be listed twice (once with an external argument, once with an internal one). This would lead to redundancy.

'3 I have not shown how movement via AgrOP derives the behaviour of unaccusarives in the two diagnostic constructions, and, thus, why subjects of unaccusatives and objects of transitives behave similarly. This is left for future research.

80

Angeliek van Hout

The aspectual approach to the lexicon-syntax interface was initially inspired by Tenny's (1994) groundbreaking work on the interaction of event structure and the role of the various arguments (subject, direct object, indirect object). Tenny describes the aspectual function of a direct object in terms of the argument bearing a so-called Measure role. An indirect object (such as a PP) may get a Delimiter role. On this approach, verbs are lexically specified to assign aspectual roles. They are listed with their aspectual theta grid, which is similar to how thematic theta grids are lexically specified in traditional approaches: for each verb one must lexically specifY which aspectual roles it assigns. Again, variable-behaviour verbs are a problem for this approach. All data presented earlier, with two- and one-argument verbs, suggest that a verb is not lexically tied to a particular theta grid, whether it is a thematic one or an aspectual one 01' one marking its argument as internal or external. Verbs can appear in multiple frames; crucially, many can appear in telic and atelic predicates. Other elements in the predicate, such as directional PPs and particles, co-determine telicity. Unless one posits massive lexical ambiguity-all these verbs can be telic as well as atelic and thus have two different aspectual role grids-this variable behaviour cannot be accounted for and an important generalization would be missed. The account outlined here supports Levin and Rappaport Hovav's (1992a, 1995) work on unaccusativity in that it posits a central role for telicity.I4 It differs from their account, in that here, telicity is the only determining factor in unaccusativity, whereas Levin and Rappaport Hovav list several different linking rules that may classifY an argument as a direct internal one. In addition to their 'telicity' rule, the Directed Change Linking Rule, they formulate the Existence Linking Rule, which says that the single argument of a verb whose existence is asserted or denied is mapped as a direct internal argument, and the Immediate Cause Linking Rule, which says that immediate-cause arguments become external arguments (see section 2.3.4). Furthermore, there is a Default linking rule that simply says that any argument that does not fall under the scope of any of the linking rules is a direct internal argument. The Existence Linking Rule takes care of verbs such as appear, disappear, stay, remain; the Immediate Cause Linking Rule classifies verbs of emission such as shimmer and glow as unergatives, since their single argument is in immediate control of the event (there is no external causer for something to glow or shimmer). The Default Linking Rule then classifies non-agentive verbs of manner of motion such as roll and bounce as unaccusative, because their single argument is not an immediate cause. Neither do these verbs express directed change or existence, so the argument would not be classified unaccusative by these two linking rules. I found, however, that the single argument's internal or external control properties do not playa role in Dutch. The data in (24)-(27) show that it is not a factor in unaccusativity, so there is no support for the Immediate Cause Linking rule. Moreover, on Levin and Rappaport Hovav's account it remains rather arbitrary why it is precisely telic one-argument verbs that become unaccusative and atelic ones unergative, rather '4

See n.

12.

Unaccusativity as Telicity Checking

81

than the other way around (atelic verbs are unaccusative and telic ones unergative). In the present proposal, the importance of the object position for telicity is generalized across one- and two-argument verbs and their aspectual mapping behaviour. The generalization that telic one-argument predicates are unaccusative now comes as no surprise. The theory outlined here thus supports only one of Levin and Rappaprt Hovav's linking rules. My theory is very similar to Borer's (1994) theory of argument linking. Borer proposes a system in which argument projection is not based on the properties of lexical entries as such (like theta grids, lexical-conceptual structures, or argument structures), but on the properties of functional projections-in particular, their aspectual properties. The order of arguments in syntax is constituted as a result of a set of aspectual constraints on the realization of specifiers. These specifiers, in turn, are associated with particular interpretations. In particular, Originator is linked with subject position and Event Measure with the object position of telic predicates. Borer thus takes a fully configurational approach to argument linking, rejecting lexical-entry driven approaches. Her model leads her to conclude that the initial projection of arguments onto positions inside the VP does not matter. What matters is their movements in the functional layer above VP. Borer concludes that the arguments within the VP are hierarchically unordered. There is no lexical distinction between external and internal arguments, nor is there a VP-internal syntactic distinction between subjects and objects. There is no lexical specification that drives the ultimate positions of the arguments either. This is a huge step away from traditional linking approaches which all distinguish initial subject and object positions. My approach supports Borer's approach for one-argument verbs-in fact, it is almost a notational variant. However, on the basis of arguments from two-argument verbs I believe that there is a need for distinguishing subject and object positions for two-argument verbs inside the VP after all. In section 2.2 I discussed the role of the semantics of the direct object, showing that only quantized objects yield telicity. I argued that these move out to AgrOP. On the other hand, non-quantized objects, such as mass terms and bare plurals, do not yield telicity and in my proposal they do not need to move out of the VP, getting their weak Case assigned in the object position inside VP (see derivation (10)). For these cases-atelic transitives-Borer creates another functional projection outside VP-namely, FP-which does not yield any aspectual role for the argument that goes there. It is unclear what triggers movement to FP, except for the theory-internal argument that all the arguments must move out of the VP. Until there is further evidence that objects with weak Case move out of the VP, I maintain that at least a position for objects must be distinguished in the VP, rather than having unordered argument sets in the VP. The sole factor in determining unaccusativity in my theory (and also in Borer's) is the telicity of a predicate, which triggers movement to a certain projection in the functional domain. As for atelic predicates, this theory predicts unergative mapping for one-argument verbs. Stative verbs are atelic and are thus not expected to map as unaccusatives. There are, however, two classes of statives that are potential counter-

Angeliek van Hout

82

examples. Stative one-argument verbs such as blijven ('stay') and zijn ('be') select auxiliary zijn. 1hey do not give good results with prenominal modifiers, however, so they are not obviously unaccusative. Sorace (2000, this volume) argues that such verbs fall in the middle of a gradient hierarchy of auxiliary selection, and considers them as not prototypically unaccusative. 1he other class are stative verbs of the piacere class of psych verbs, which are argued to be unaccusative, at least in Italian (see Bennis, in this volume, for instance). Since these verbs are atelic, unaccusativity is not predicted in the aspectual approach. It may be, then, that the present theory is incomplete, in the sense that it covers only half of unaccusativity, and must be extended with a theory that can tal<:e stative psych unaccusatives into account. 1he property that seems to be crucial for these is that they lack an external argument. Several chapters in this volume working with this notion of unaccusativity explore ways of how the make-up of the vP layer plays a role (presence or absence of vp, vP with or without an external argument, etc.). See in particular Bennis, Embick, and Alexiadou and Anagnostopoulou. It remains to be seen if the two approaches-the aspectual one and the lack-of-external-argument one-can meet in the middle, or if they turn out to be incompatible.

2.5.

CONCLUSIONS

Unaccusativity reflects a mapping configuration in which the verb's single argument moves through the specifier position of AgrOP on its way to the specifier position of AgrSP. Movement through AgrOP is triggered when the predicate that contains the verb is telic, because AgrOP is the locus for telicity checking. 1his was first demonstrated for two-argument verbs, then extended to one-argument verbs. 1here are several lines of further research that have not been explored here. First of all, I have used auxiliary selection and participial modification as Dutch unaccusativity diagnostics, but I have not analysed why they can diagnose it. 1he structure proposed for unaccusatives should be employed to find an analysis. Furthermore, different languages show different unaccusative diagnostics. It is a challenge to see if these can be linked to the configuration presented here. A further step is thus to develop the theory for other languages and see if it holds cross-linguistically. Levin and Rappaport Hovav (1995) distinguish several subclasses of unaccusatives. In the present framework there is only one class: telic one-argument verbs. It must be seen whether this is in fact a proper cross-linguistic generalization. 1his raises the issue of the universality of the lexicon-syntax mapping system-do languages differ parametrically? Alternatively, the variation across Dutch, English, and other languages may turn out to be an artefact resulting from looking at the wrong unaccusative diagnostics. Learnability considerations may help determine this issue (c£ van Hout, Randall, and Weissenborn 1992, 1993). Given that unaccusatives at the surface look just like unergatives with one single argurrient-that is, they do not wear

Unaccusativity as Telicity Checking their underlying objecthood on their sleeves-children must rely on overt diagnostics in order to classify one-argument verbs. The learnability of unaccusativity therefore requires some form of a syntactic bootstrapping procedure (van Hout 1996; Borer, in this volume). This may lead to a conclusion in favour of parametric variation of unaccusativity (Randall, van Hout, Weissenborn, and Baayen 1994 and in this volume). A final issue is the effect of the Extended Projection Principle (EPP), which I have used in my analysis as a feature to trigger the single argument of an unaccusative to move to AgrSP. It is unclear to me how this principle must be viewed in the minimalist framework. The real question is: why do sentences need to have their subject position filled? The answer to this question will also explain the second movement of the single argument in an un accusative configuration.

3 Unergative Adjectives and Psych Verbs Hans Bennis

3. I.

INTRODUCTION

The idea that the verb phrase is a layered constituent has been around in the literature for some time-see Larson (1988) and Hale and Keyser (1993), for instance. In the Minimalist Program, as sketched in Chomsky (1995,2001), shell theory plays an important role as well. Here I will argue that a particular interpretation of shell theory may provide insight into long-standing problems such as the nature of syntactic ergativity and the structure of psychological verbs. Let us consider a VP-structure with a V-type shell (the vP) on top of it. Given binary branching, this structure is able to accommodate three argument phrases, namely, the complement ofY, the specifier of Y, and the specifier of v. In Chomsky's words: The internal arguments occupy the positions of specifier and complement ofY. Accordingly, the external argument cannot be lower than [Spec, vJ. If it is [Spec, v], as I will assume, then the v--VP configuration can be taken to express the causative or agentive role of the external argument. It would be natural to extend the same reasoning to transitive verb constructions

I would like to thank SjefBarbiers, Marcel den Dikken, and the editors of this volume for their comments on an earlier version of this chapter. It has been presented in Berlin (1998) and Leiden (1998). I thank the audiences there for their remarks.

Unergative Adjectives and Psych Ve>rbs generally, assigning them a double-VP structure as in (U5) [=(I)-HBJ, the agent role being understood as the interpretation assigned to the v-VP configuration. A V-object construction is therefore maximal, not V'. [... J If intransitive (unergative) verbs are hidden transitives, as Hale and Keyser suggest, then only unaccusatives lacldng agents would be simple VP structures. (Chomsky 1995: 315-16)

vP

(I)

~

Spec

v'

~

v

VP

~

Spec

V'

~

V

Compl

Chomsky's position has several interesting implications. First of all, it is argued that the nature of thematic roles is essentially configurational-that is, the presence of small v expresses the interpretative concept of agentivity or causativity, which in turn requires the specifier of v to be interpreted as a position for the argument bearing the thematic role ofAgent or Cause (c£ Kratzer 1996). In this perception the lexicon does not have to specify the presence or the nature of the external argument. An indication whether v should be added to the V-projection is sufficient. Since the meaning of a lexical verb might be decomposed into an activity or causativity part and a core verbal part, we may eliminate the external argument from the lexical representation altogether. To some extent this is the approach adopted by Hale and Keyser, although the actual execution of this idea differs substantially. A similar consequence is that unaccusativity is structurally expressed; not by the lack of an underlying external argument, but rather by the absence of the v-projection. Again, this might be the result of the lack of agentivity/causativity in the meaning of the verb, in which case unaccusative verbs do not have to be listed in the lexicon as such.

3.2.

BURZIO'S GENERALIZATION

This perspective on unaccusativity-unaccusative projections are VPs, unergative projections are vPs-gives rise to a change in perspective on the presence of arguments. As will be clear, the presence of an external argument depends on the presence of a small v, and vice versa; or in Chomsky's words: 'The external role is a property of the v-VP configuration, and a specifier bearing this role is therefore a necessary part of the configuration; a transitive [or, rather, unergative-HB] verb assigns an external role by definition' (Chomsky 1995: 316).

Hans Bennis

86

This proposal relates the presence of vP to the interpretive property of agentivity/ causativity. Absence of a small v gives rise to unaccusativity. However, there are nonagentive, unergative constructions, in particular stative ones. If we want to capture the unergative-unaccusative distinction in structural terms-vP vs. VP-we should not assimilate the v-projection with one particular meaning type. Small v can be either dynamic or stative (cf. Kratzer I994). In a sentence such as (2a) the external argument is interpreted as an Agent as the consequence of v being dynamic; in (2b) the external argument is interpreted as Possessor,' due to the fact that v is stative, whereas the structure in (2C) is unaccusative given the lack of a v-projection. (2)

a. John greets the audience. b. John knows the audience. c.

John dies.

This brings us to the generalization in (3). If v is dynamic: [Spec, v] is Agent If v is stative: [Spec, v] is Possessor

Qohn greets the audience) (John knows the audience)

From this perspective we may easily capture Burzio's generalization-lack of external argument: lack of accusative Case-in structural terms. 2 In order to do so we need to give the v-projection a role, not only in the presence of an external argument, but also in the assignment of accusative Case. The ingredients for the latter role of the small v are also present in the Minimalist Program. After a decade of Agr-projections, Chomsky tries to eliminate Agr as a separate functional projection (Chomsky I995: sect·4·ro). The obvious problem for a theory in which formal relations or checking relations are relations between a head and its specifier is the question how to assign accusative Case to the object. The main reason for the introduction of AgrO was to create a Spec position in which it is possible to assign Case to the object. In order to get rid ofAgrO, Chomsky argues that a phrase may have multiple specifiers. When we adjoin the object to XP, a second specifier position is created that may be involved in Case assignment. If we talce v to be a head able to assign accusative Case, and if an object can be assigned Case by being adjoined to vP, we no longer need AgrO for Case assignment. We are now in a position to account for Burzio's generalization in structural terms. If it is indeed v, and not V, that is able to assign accusative Case, the absence of v has two consequences: there is no external argument and no accusative Case. This leads to the conclusion that the two implications below are both valid, since the presence of an external argument and the availability of accusative Case crucially depend on the presence of v:

(a) if there is no external argument, no (structural) accusative Case can be assigned;

I 2

The thematic role Possessor indicates that the argument possesses the state denoted by the VP. Other attempts to derive Butzio's generalization can be found in Reuland (2000).

Unergative Adjectives and Psych Verbs (b) if no (structural) accusative Case is available, no external argument can be generated. One related question should be raised in connection with this multiple-specifier approach. If the external argument is generated by Merge in the specifier postion of v, and the internal argument can be raised/adjoined to vP (the outer spec position) in order to check its objective Case feature, why cannot the external argument, which is also present in [Spec, v] (the inner spec position), check the objective Case feature of v? If it could, we incorrectly predict internal arguments to become subjects (raising to [Spec,T]) and external arguments to show up as objects. Of course, this is a general question to the multiple specifier approach. Chomsky argues that 'Subj inserted by Merge in [Spec, v] is not in the checking domain of v, because it does not head a nontrivial chain' (Chomsky 1995: 352).3 Chomsky's approach crucially separates thetapositions from Case-positions, or rather theta-checking from Case-checking, in terms of his Chain Condition: 'an argument is a non-trivial chain CH = (a, t), where a has raised for feature checking and tis in a a-position' (Chomsky 1995: 312). Let us assume that the complementarity of theta-checking and Case-checldng is a general property of the theory. It then follows that only raised arguments can check the Case feature of v. This gives us the results we want: [Spec, v] is a position in which both theta-checking and Case-checldng may apply, but only with respect to different arguments. The argument that checks its theta-feature in [Spec, v] gets interpreted as the agent, the causer or the possessor. This argument has to move to a higher projection in order to check its Case feature. The argument that checks its Case feature in [Spec, v] is the argument with objective Case; this argument must have been moved from a thetaposition lower in the structure in order to establish a non-trivial chain. Adopting the multiple specifier approach, we provide a simple explanation of Burzio's generalization. This part of the theory now contains three ingredients:

a. an external argument is present iff a v-projection is added to VP; b. accusative Case can be checked in [Spec, v] only; c. only internal arguments may check the Case-assigning feature that is intrinsic to v. Together these ingredients mal<:e Burzio's generalization follow from the theory and we can get rid of this unidentified theoretical object. However, as I will demonstrate in the following sections, this perspective is too simple. I will extensively show that Burzio's generalization is partially wrong. I will argue against the idea that the presence of a small v is necessarily connected to the presence of an external argument (point a). To be more precise, only one of the following implications is valid (AI). Implication A2 appears to be incorrect.

3 Chomsky extensively argues that adjunction of the subject to its own projection, creating an outer spec position, is not an available option either. This implies that the subject has to move out to check its Case feature.

88

Hans Bennis

AI if an external argument is present, v has been generated; A2 if v is generated, an external argument has to be present. If A2 is not correct, the theory predicts the occurrence of vP without external argument. This approach thus gives us three distinct configurations: i. VP without a v-projection; ii. a v-projection without an external argument; iii. a v-projection with an external argument. The external argument can be absent due to the absence of vP (i) or to the absence ofe-assigning properties intrinsic to v (ii). 'We thus distinguish two different types of ergativity: simplex ergativity (i) and complex ergativity (ii).4 We will see that (i) and (ii) differ in various ways. In order to substantiate this proposal empirically, I will first concentrate on adjectival constructions in Dutch. It will be shown that three different adjectival construction types have to be distinguished structurally. In order to do so, I will introduce the concept of small a. The three different construction types correspond to (i)-(iii) above: AP, aP without EA, and aP with EA. Mter having established the three-way system for adjectival phrases, I concentrate on present participles. These participles are somewhere in between verbs and adjectives. Particularly interesting are present participles of psychological verbs. Especially the occurrence of the experiencer argument is significant. If the present-participle construction is internally verbal the experiencer argument may be realized, whereas the occurrence of an experiencer leads to un grammaticality in adjectival present-participle constructions. From this we conclude that the experiencer is structurally Case-marked, which in turn leads to the conclusion that psychological verb constructions of the relevant type (the preoccupare class; cf. Belletti and Rizzi 1988) are to be analysed as vP. The next step in the argumention involves the argument that the relevant class of psychological verbs is ergative in the thematic sense-that is, no external argument has been generated. If correct, we have shown the existence of verbal constructions of type (ii). It now follows that the three different types of psychological verb correspond to the three different construction types distinguished earlier: the temere class is argued to be a regular transitive construction of type (iii), the preoccupare class is of type (ii), and the piacere class behaves like the simplex ergatives of type (i). The final conclusion of this chapter is that Burzio's generalization should not have any theoretical status. It is partially wrong, and, insofar as it is correct, the consequences follow from the theory without additional stipulation. Moreover, I argue that unaccusativity is theoretically a meaningless concept. The crucial properties with

4 The idea that unaccusatives can be either simplex or complex is also proposed, in a different context, in Alexiadou and Anagnostopoulou (in this volume). They appeal to the relative complexity of structure in order to accommodate the systematic presence or absence of morphology associated with detransitivization in different classes of unaccusatives in Greek.

Unergative Adjectives and Psych Verbs respect to the presence and absence of arguments and their configurational properties are indirectly related to Case theory only. The most important distinction that is introduced here is the distinction between simplex ergativity and complex ergativity.5,6

3.3.

SIMPLE ERGATIVE ADJECTIVES

Cinque (1989, 1990) argued that the class of adjectives should be divided into ergative and non-ergative adjectives. He discusses a number of criteria which provide substance to a distinction along these lines for Italian and German. Below I discuss three phenomena that show that a division into ergative and unergative adjectives is motivated for Dutch as well (for a more extensive motivation of this distinction in Dutch, see Bennis 2000). It is evident that unaccusativity is irrelevant here, given that adjectives do not assign structural Case. In order to develop a division into ergative and non-ergative adjectives, we will introduce other criteria to show that (44) and (4b) differ with respect to ergativity (4a is ergative, 4b, unergative).

(4)

a.

b.

Deze mensen zijn mij bekend. these people are me known 'These people are known to me.' Deze mensen zijn mij trouw. these people are me loyal 'These people are loyal to me.'

3.3.1. Inversion In Dutch, inversion of subject and object is highly restricted. If we do not take into account wh-movement-that is, movement to [Spec, C) as in question formation, topicalization, and relativization-the subject generally precedes the (in)direct object. However, there is one class of exceptions: when the subject is an underlying direct object, as in the case of passives and ergatives, the indirect object may precede the subject. The topic of argument inversion in Dutch has been amply discussed in the 5 I use the term 'ergativity' (in the spirit of Burzio 1986) for what other people in this volume use the term 'unaccusativity'. The reason is that an important part of the discussion focuses on adjectives. Since adjectives do not assign structural accusative Case in general, the term 'unaccusative adjective' seems tautological, in contrast to the term 'ergative adjective'. The crucial point is that the ergative-unergative distinction refers to a thematic distinction, rather than a Case distinction. 6 Some chapters in this volume, notably those by Borer and van Hout, take unaccusativity to depend on telicity; and do not address the presence or absence of the external argument as a crucial property of ergative constructions. These proposals agree with the account presented here that unaccusativity is mainly a syntactic phenomenon that is not determined by properties of the verb in the lexicon. Other chapters, notably those by Alexiadou and Anagnostopoulou and Embick, take the external argument property to be central and do not address the notion of telicity. It is not so clear to me how telicity could be taken to be the relevant notion in the case of the stative unaccusatives that are presented here.

Hans Bennis literature (den Besten I982; Everaert 1982; Hoekstra I984; Bennis I986, for instance). In (5a) we see that the order IO-SUBJ leads to ungrammaticality in ditransitive sentences. In (5b, c) we observe that the order IO-SUBJ is possible in passive and ergative clauses'?

(5)

a. b. c.

dat die jongen hem/* hem die jongen beloofde naar huis te gaan that that bOYDO himro/himIO that bOYDo promised to go home dat die jongen hem/hem die jongen werd voorgesteld that that bOYDo himro/himro that bOYDo was introduced dat die jongen hem/hem die jongen opviel that that bOYDo himro/himro that bOYDO struck

Given that adjectives in Dutch may occur with indirect objects, we have the opportunity to check whether a superficial subject is or is not derived by movement from object position. If inversion is available, the subject has been moved, just as in (5b, c). It is indeed the case that the class of adjectives that have indirect objects can be divided into adjectives that allow inversion, and adjectives that do not. This can be observed in (6) and (7).

(6)

a. b.

dat dat probleem hem/hem dat probleem duidelijk is that that problemDo himro/himro that problemDo clear is dat die jongen hem/hem die jongen bekend is that that bOYDo himro/himIO that bOYDo well known is dat die jongen hem/*hem die jongen trouw/gehoorzaam is that that bOYDo himro/himro that bOYDo loyal/obedient is

(7)

We thus observe that the class of adjectives can be divided into adjectives that pattern with ergative verbs (duidelijk 'clear', bekend'well known'), and adjectives that behave as unergatives (trouUJ 'loyal', gehoorzaam 'obedient').

3.3.2. As clauses The second criterium, which is also applicable in German and Italian, concerns the occurrence of adjectives in so-called as clauses (c£ Stowell 1987; Cinque I989). All has been demonstrated by Stowell, the gap in as clauses always corresponds to an underlying sentential object. Simple examples of this can be found in (8). In (8a) the gap is in object position, whereas in (8b) the gap corresponds to the subject/object of a passive and in (8c), to the subject/object of an ergative verb. The non-occurrence of subject gaps is demonstrated in (9).

(8)

7

well.

a.

Zoals ik zei e, houdt Jan van slald<:en. as I said, John likes snails

In examples (5)-(7) the indirect object is a pronoun. However, inversion is possible with full DPs as

Unergative Adjectives and Psych Verbs b. c. (9)

91

Zoals e door iedereen beweerd wordt, houdt Jan van slakken. as by everybody is said, John likes snails Zoals e mij opgevallen is, houdt Jan van slakken. as me struck is, John likes snails

a. *Zoals e zijn onschuld bewijst, had Jan geen slakken gegeten. as proves his innocence, John had not eaten snails

b. ?*Zoals e mij van zijn gelijk overtuigt, lust Jan geen slald<:en. as me of his right convinces, John does not like snails We encounter the same division in the case of adjectives. Some adjectives appear in the as-construction, whereas other adjectives do not. This difference can be explained if we assume that the adjectives in (10) may select a sentential object, contrary to the adjectives in (II).

(10)

a. b.

(II)

Zoals e mij duidelijk werd, houdt Jan niet van slakken. as me clear became, John does not like snails Zoals e bekend is, houdt Jan niet van slakken. as is well known, John does not like snails

a. *Zoals e grappig is, houdt Jan van slald<:en. as is funny, John likes snails

b. *Zoals e verrassend is, houdt Jan van slald<:en. as is surprising, John likes snails Ifit is the correct conclusion that the empty element in (10) is in object position-that is, the thematic position for the sentential argument is the object position of the adjective-it follows that the surface subject of the same two adjectives in (12) is moved from object to subject position, thus giving rise to an ergative analysis.

(12)

a. b.

Jans afkeer van slakken is duidelijk. Johns revulsion at snails is clear Jans afkeer van slakken is mij bekend. Johns revulsion at snails is me well known

This test indicates that adjectives of the type duidelijk ('clear'), bekend ('well-known), zeker ('certain'), and waarschijnlijk ('probable') are ergative in their thematic organization.

3.3.3. Complementizer selection A third criterion to distinguish ergative adjectives from unergative ones is the possibility to select the complementizer of a related argument clause. It is welllmown that the form of the complementizer of a finite sentential complement (that or whether) depends on properties of the matrix verb. This selection property only shows up when the sentential argument is the object of the verb. Sentential subjects or adjuncts do not show this phenomenon. In the case of clauses selected by an adjectival head, we

Hans Bennis observe that variability in complementizer selection shows up with adjectives such as duidelijk ('clear') and bekend ('well known), but not with adjectives such as acceptabel (,acceptable') and plezierig ('pleasant'). A negative context generally supports the selection of the complementizer of ('whether') in complement clauses. We observe that on-prefixation cooccurs with ofselection in the case of adjectives of the duidelijk class only.

(13)

a. b.

Het is on-duidelijk/on-bekend dadofJan naar huis gegaan is. it is unclear/unknown that/whether John is gone home Het is on-acceptabellon-plezierig dat/*ofJan naar huis gegaan is. it is unacceptable/unpleasant that/whether John is gone home

The examples in (13) again show that the (clausal) argument of the adjective in (13a) is the thematic object of the adjective, whereas the clausal argument in (13b) is not an object.

3.3.4.

The a-shell

The arguments given above suffice to show that the ergative-unergative distinction is motivated in the adjectival domain. The question now arises how we can account for this distinction in structural terms. In section 3-I it was shown that in minimalist terms verbal ergativity may be structurally expressed by the absence of a v-shell. The occurrence of an external argument forces the presence of a v-shell in unergatives. The same type of reasoning applies to adjectival projections. Ergative adjectives such as dUidelijk and bekend can be analysed as bare APs. The indirect object is generated in [Spec,A], whereas the object/subject of the adjectives is generated as the complement of A. Unergative adjectives such as trouw and gehoorzaam show up with an external argument. The indirect object is again generated in [Spec,A]. To create a structural difference between ergative and unergative projection, we need to create another layer of structure. If we adopt structural parallellism between A- and V-projections, the consequence is that unergative adjectives require an adjectival shell on top of the bare AP. Small a in adjectival projections is fully symmetrical to small v in verbal projections. The two structures are presented in (14).

(14)

a.

b.

AP

~

Spec

it

mij

~

aP ~

a' deze ~ mensen a AP Spec

Compl

A

dezemensen

bekend

~

Spec

mij

it

I

A

trouw

Unergative Adjectives and Psych Verbs

93

In both (14a) and (14b) the indirect object mij appears to be inherently Case-marked. In (14a) the object of bekend moves to a higher projection (TP) in order to receive Case. In (14b) the external argument deze mensen moves to TP for Case reasons as well. Although the superficial strings of these constructions are similar, as in (4), we have seen that there are good arguments to suppose that the underlying structures are substantially different. Additional corroboration for a difference in underlying structure concerns the semantic relation between the adjective and the DP deze mensen in (4). In (4b!I4b) the DP deze mensen receives an interpretation that is similar to the 'possessive' interpretation of the subject of stative verbs, c£ (2b/3). This interpretation is not available in (4a!I¥). In line with the view that there is a direct relation between syntactic structure and thematic organization, we assign small a an interpretation that is comparable to the interpretation of (stative) small v. The a-shell in (I4b) differs in two respects from the v-shell in unergative verbal projections. First, small a is not able to assign Case. This implies that adjunction to aP does not create an outer specifier position in which Case features can be checked. An internal argument has to be moved to a position higher than [Spec,a] in order to receive Case. A second difference between a and v is that the aspectual interpretation of the small head appears to be constant in the case of adjectives. Adjectival projections are stative, whereas verbal projections are sometimes stative, sometimes dynamic.

3.4.

COMPLEX ERGATIVE ADJECTIVES

Within the class of unergative adjectives, a systematic division into two groups is called for. I will demonstrate that constructions of the type in (15a) are different from the type in (15b), and provide three arguments to show that a structural distinction between them is warranted. For more argumentation on this issue I refer to Bennis (2000).

(15)

a.

b.

Jan is trouw/gehoorzaam. John is loyal/obedient Dat is trouw/gehoorzaam. that is loyal/obedient

3.4.1. Indirect objects The construction of the type in (15a)-type A-may involve the realization of an indirect object with adjectives such as trouw ('loyal') and gehoorzaam ('obedient'). Interestingly, the same indirect object is not permitted if the adjectival subject does not refer to the person who possesses the adjectival property (jan in (16a)), but rather to the theme that expresses the adjectival property (dat in (16b) )-type B.

Hans Bennis

94 (16)

a.

b.

Jan is mij trouw/gehoorzaam. John is me loyall obedient Dat is (*mij) trouw/gehoorzaam. that is (me) loyall obedient

[type A] [type B]

There is no interpretive reason to expect this difference between types A and B. Nor is it dear how the absence of the indirect object could be related to Case reasons. Why would inherent, lexically determined Case be absent in (16b) , where it is present with the same adjectives (trouw, gehoorzaam) in (16a)? The difference between types A and B thus appears to be determined by structural considerations.

3.4.2. Prepositional objects A similar difference between type A-the subject is the possessor of the adjectival property-and type B-the subject is the nominal phrase that expresses the adjectival property-we find with respect to the occurrence of prepositional complements. In the case of the adjectives trots ('proud') and nerveus ('nervous'), the object can be expressed in a PP headed by the preposition op and voor, respectively, but only if the subject is the possessor. Again, there is no interpretive reason to expect the difference in grammaticality between (17a, ISa) and (17b, 19b) to occur.

(17)

a.

b. (IS)

a. b.

Jan is trots (op zijn overwinning). John is proud (of his victory). Jans houding is trots (*op zijn overwinning). John's attitude is proud (*ofhis victory). Jan is nerveus (voor het examen). John is nervous (about the exam). Jans gedrag is nerveus (*voor het examen). John's behaviour is nervous (*about the exam).

A difference which is similar to the difference between (17a, ISa) and (17b, 19b) has also been discussed by Higgins (1973) and Pesetsky (1995) for English adjectives. They observe that not all adjectives show this pattern, as is demonstrated in (19) and (20). (19)

(20)

a.

b.

John is clear (to his students). John is reminiscent (of his father).

a. b.

John's words are clear (to his students). John's behaviour is reminiscent (of his father).

The same is true for the Dutch equivalents of (19) and (20), as is shown in (21). (21)

a. b.

Jan is duidelijk (voor zijn studenten). John is clear (for his students) Jans houding is duidelijk (voor zijn studenten). John's attitude is clear (for his students)

Unergative Adjectives and Psych Verbs

95

Crucially, the adjectives in (19)-(21) are ergative (c£ section 3.3). We may thus hypothesize that the non-occurrence of prepositional objects is restricted to unergative adjectives which do not have a possessor subject.

3+3. Implicit arguments Another criterion that divides the class of unergative adjectives into two groups is the presence of an implicit argument. In the construction of type B-the construction which lacks a possessor subject-the possessor argument can be added in a PP headed by van ('of'). 8 Of course this option is not available to the construction of type A, since a possessor is already present in the structure. This is shown in (22).

(22)

a.

b.

Jan is aardig/vervelend (*van Henk). John is nice/annoying of Henk Dat is aardig/vervelend (van Henk). that is nice/annoying of Henk

If we take the van-phrase in (22b) to be parallel to the door-phrase ('by') in verbal constructions, the pair in (22) is reminiscent of the active-passive alternation. The external argument of the adjective in (22a) is realized in an optional PP in (22b).9Thematically both the subject in (22a) and the object of the preposition van in (22b) refer to the possessor of the adjectival property. An argument that corroborates a passive-like analysis is the fact that the possessor is implicitly present in (22b) when the optional van-PP happens to be absent. Just as in passive clauses, this implicit argument can be syntactically activated in controlling the subject of infinitives. This is shown in (23) and (25). (23)

a.

b.

Het is aardig (van Jan) om PRO iets tegen mij te zeggen. it is nice ofJohn for PRO something to me to say Het is vervelend (van Jan) om PRO dat aan mij over te laten. it is annoying ofJohn for PRO that to me to leave

In (23) we find embedded subject clauses of the infinitival type. The relevant point here is that the PRO-subject of the infinitive is controlled by the possessor argu8 The presentation simplifies the actual situation considerably (cf. Bennis 2000). The class of unergative adjectives can be divided in at least two different groups: adjectives denoting mental properties (c£ Stowell 1991), such as the ones discussed here (am¥iig, trOt/lV, etc.), and adjectives denoting psychological states such as moeilijk (,difficult') and blteressam ('interesting'). In the B-type construction, the latter group of adjectives selects a possessor (or experiencer) in a PP that is headed by voor ('for'), instead of van ('of).

0)

a. Jan is moeilijk!interessant. John is difficult/interesting

b. Dat is moeilijklinteressant voor Jan. that is difficult/interesting for John Here I will leave out of consideration this distinction within the class of unergative adjectives. 9 The fact that a sentence such as *Jall is dat aardig/ vervelend is ungrammatical, is not due to thematic considerations, but rather to the fact that no structutal accusative is available for the thematic object of the adjective.

Hans Bennis ment of the adjective. Even when no lexical possessor is available, the sentence must be understood in such a way that the person who is interpreted as the subject of the infinitival clause possesses the relevant adjectival property. This can be made explicit by replacing the adjectives in (23) by adjectives that do not have an implicit possessor, such as the ergative adjectives discussed in section 3.3. The ungrammaticality of (24a) and the contrast between (24a) and (24b) are easily explained if we assume that in this kind of infinitivals, PRO should be controlled. The lack of an implicit possessor argument forces the subject clause to be finite. (24)

a. *Het is duidelijklwaarschijnlijk om PRO iets tegen mij te zeggen. it is clear/probable for PRO something to me to say b. Het is duidelijk/waarschijnlijk dat hij iets tegen mij zal zeggen. it is clear/probable that he that he something to me will say

A similar argument can be made with respect to the much debated 'easy-to-please' construction in (25). (25)

a.

b.

Oat is aardig (van Jan) om PRO e tegen mij te zeggen. that is nice (ofJohn) for PRO e to me to say Oat is vervelend (van Jan) om PRO e aan mij over te laten. that is annoying ofJohn for PRO e to me to leave

I take the 'easy-to-please' -construction of the type in (25) to be the result of wh-movement of an empty operator to the [Spec,C] of the infinitival clause. This empty operator is coindexed with the matrix subject (dat in (25)). The relevant point here is the fact that PRO in the subject position of the infinitival clause is controlled by the implicit argument of the adjective. Even if the van-phrase is lacking, the sentence must be interpreted in such a way that the PRO-subject of the infinitival clause is identified as the implicit possessor of the adjectival property.

3.4+ An analysis of type B Let us try to make sense of the differences between type A and type B unergative adjectives in a structural way. A list of the differences between these two constructions is set out in Table 3.1. The thematic organization appears to indicate that the relation TABLE 3. I. Differences between types A and B

Thematic interpretation of subject Occurrence of indirect object Occurrence ofPP-object Implicit possessor Vtln-PP

Type A

TypeB

Possessor Yes Yes

Theme

No No

Yes Yes

No No

Unergative Adjectives and Psych Verbs

97

between types A and B is parallel to the relation between active and passive verbs. Let us assume this to be the case. Until now I have not said anything about passives. However, the line of argumentation taken here forces us to tal<:e a particular stand on this issue. I argued above that in verbal constructions the presence of an agent or agentivity is determined by the presence of small v. Given that passives-in contrast to ergative verbal constructions-show agentivity and given that they arguably contain an implicit agent, passive structures have to be vP. In the spirit of various proposals in the literature, we may argue that passive morphology is generated in v. The passive morpheme is then taken to host the implicit external agent (cf. Jaeggli 1986; Baker, Johnson, and Roberts 1989). The difference between passive and ergative structures can thus be traced to the presence or absence of small v. The similarity between the two resides in the fact that the theme argument shows up as the surface subject. In order to account for this we seem to be forced to assume that Burzio's generalization is reinvoked at the v-level: if v does not assign a thematic role to the inner spec position, it does not assign structural objective Case to the outer spec position either. However, if we integrate Bal<:er, Johnson, and Roberts's (1989) passive analysis into our account, the correlation between the absence of an explicit agent and the unavailability of Case can be made to follow. Baker et al. argue that the implicit argument (or rather the passive -en morpheme) should receive Case. If that is correct, we may assume that the passive morpheme in v checks its objective Case.IO In this way objective Case can no longer be checked by an object DP in the outer Spec of v. The passive part of Burzio's generalization can thus be made to follow from an analysis that is basically similar to the analysis given by Baker, Johnson, and Roberts (1989). We now have identified two verbal constructions in which the absence of an external argument (inner spec of v) prevents the assignment of objective Case (outer spec of v). This leads to the configurations in (26) for passive and ergative verbal projections. (26)

b.

a. Passive

Ergative

vP

VP

~

~

Spec

v'

e

~

V'

Spec

v

VP

[+pass]

~ Spec V'

~

V

DP

~ V DP

m The minimalist theory has identified two structural configurations in which formal relations can be established. In addition to the Head-Spec configuration discussed earlier, checking also takes place if two elements cooccur in a head position. The head-head configuration is the one relevant here. The passive morpheme in v is able to check its Case feature with the Case feature in v.

Hans Bennis

Coming back to the adjectival constructions under consideration here, we can express the difference between type A and type B unergative adjectives in a straightforward way. The structural difference between type A and type B concerns the availability of an external argument in the specifier position of the aP. Just as in the verbal domain there are two different types of ergativity in the adjectival domain. Simplex ergativity in the case of adjectives such as duidelijk ('clear') and bekend ('well-lmown') (27b), and complex ergativity in the case of unergative adjectives that have been stripped of their external argument (27a). (27)

a.

b.

Complex ergative (c£ 22b)

Simplex ergative (c£ 4a)

aP

AP

~

~

Spec e

a' ~

~

A duidelijk

AP

a

A'

Spec

~

DP

it

Spec

~

A leuk

DP

In both cases the DP-the theme which expresses the adjectival property-has to be moved to a position in which it can check its Case feature. No structural Case is available within the projections (27a) and (27b). The DP thus has to move to the external subject position [Spec,T] in order to reach a position in which the Case feature might be checked. n Although the thematic organization, the presence of an implicit argument, and the optional presence of a van-phrase in type B follow from this passive-like analysis, it does not yet explain why indirect objects and prepositional objects do not occur in type B constructions. This question appears to be complicated by the fact that indirect objects and prepositional objects do occur in configurationally similar passive constructions, as in (28). (28)

a. b.

Dat boek werd mij overhandigd. that book was me given Dat werd tegen mij gezegd. that was to me said

II Of course the landing site of the DP that is on the run to find a Case-checking environment may also involve the specifier position of vP in the case of small-clause complements. This is what appears to happen in sentences such as in (i).

(i)

Ik vind dat leuk/duidelijk. I find that nicel clear

Unergative Adjectives and Psych Verbs

99

Similarly, as we have seen in section 3.3, simplex ergative adjectives may occur in combination with indirect objects. I will argue that the answer to this apparent puzzle lies in (relativized) minimality. Let us assume that movement of an object always targets the outer spec position of the XIx-projection. 12 In simplex ergatives the target position is [Spec,V] or [Spec,A]; in complex ergatives and unergatives, [Spec, v] or [Spec,a]. It now follows that in ergative constructions object movement is not obstructed by minimality or locality considerations, since the adjoined object will always be the local binder of its trace. The presence of an indirect object in [Spec,V/A] makes no difference since both specifiers are equidistant with respect to the object position. We thus expect ergative constructions to be able to appear with indirect objects. The situation differs if it concerns movement from the VP/AP-domain to the specifier position of vPlaP. It has been argued (c£ Chomsky 1995) that in verbal constructions, movement from VP is contingent on head-movement ofY. We may take movement to small v to be a first step in the journey from V to T. If we assume that in verbal constructions the verbal domain is enlarged by movement of V to v, the presence of an indirect object in [Spec,V] has no effect whatsoever on the movement of the object to [Spec, v]. No minimality problem arises. We thus expect objects to target [Spec, v], independent of the presence of another DP in [Spec,V] (the indirect object) orland [Spec, v] (the external argument). In adjectival constructions of the type aP the situation is different. There is no obvious reason to assume that A has to move to the functional domain. If there is no head movement, movement of the object out ofAP may give rise to minimality violations. This situation occurs if [Spec,A] is filled with an argument. By moving the object to [Spec,a] the intervening argument in [Spec,A] is closer to the trace of the object than the moved object in [Spec,a], thus ensuing a violation of the Minimal Link Condition (Chomsky 1995).'3 We thus are able to explain the fact that internal arguments do not show up in adjectival constructions of type B. The at first sight surprising asymmetry between complex ergative adjectives and simplex ergative adjectives, and between complex ergative adjectives and complex ergative verbs (passives), now follows from locality considerations. I4This corroborates the distinction between simplex and complex ergative constructions. Let us summarize the major results of this discussion of adjectival constructions: • simplex ergative adjectival constructions (adjectives such as duidelijk 'clear') are bare APs; n This would follow if adjunction is restricted to minimal (heads) and maximal projections only. lfVP/ AP is taken to be a segment of vP/aP, object movement in non-ergative configurations should target vP/aP directly. For discussion of this issue, see Chomsky (1995: ch. 4). 13 A crucial aspect in this analysis is that there is no movement from A to a. This might be problematic in that attributive adjectives show some inflection in Dutch. Another problem concerns the possibility of A-to-Q raising, as proposed in Corver 1997. I will have to assume that no movement is involved in these cases. I will leave this problem for further research. '4 This is also true for the English contrast given in (17)-(20), which has been discussed by Higgins (1973) and Pesetsky (1995).

Hans Bennis

100

• other adjectival constructions have an a-shell on top of AP;

• aP might be ergative (complex ergatives) or unergative; • due to minimality, indirect objects and prepositional objects are barred from complex ergative, adjectival constructions; • passive verbal constructions are structurally similar to complex ergative, adjectival constructions; • ergative verbal constructions pattern with simplex ergative, adjectival constructions. It is clear that until now the notion of ergativity has largely been separated from the issue of Case checking. The ergative-unergative distinction has been argued to be relevant in contexts in which Case is irrelevant, as in adjectival phrases where no structural Case is available in general. In the following sections I will return to verbal constructions to see to what extent ergativity and Case are interrelated.

3.5.

PRESENT PARTICIPLES

An interesting bridge between adjectival and verbal constructions is presented by the issue of present participles (cf. Bennis and Wehrmann 1990). At least in Dutch, present participles appear to be exclusively adjectival in their distribution. They occur in attributive position within DP (29a), they appear in adjunct position within the middle field (29 b) , and some of them show up in predicative position, such as the copula construction in (30a) and the small-clause construction in (30b). These positions are generally unavailable for verbal projections. (29)

a.

b. (30)

a.

b.

een ontroerende/pratende machine a touching/talking machine De machine stond ontroerend/pratend op mijn bureau. the machine stood touching/talking on my desk Die machine is ontroerend/*pratend. that machine is touchingltalking Ik vind die machine ontroerend/*pratend. I find that machine touching/talking

Although externally adjectival, there are convincing arguments to claim that the internal mal{e-up of present-participle phrases (PPPs) can be verbal. The PPP may contain direct objects with structural Case and small-clause complements. These possibilities point at a regular verbal construction, since direct objects and small clauses are not found within the class oCreal' adjectives. Some examples are given in (31) and (32).

(31)

a.

een [mij veel overlast bezorgende] machine a [melo much troubleDo causing] machine

Unergative Adjectives and Psych Verbs

(32)

101

b.

De machine stond [mij veel overlast bezorgend] op mijn bureau. the machine stood [melo much troubleDo causing] on my desk

a.

de [[mij gelukkig] makende] machine the [[me happy]sc making] machine De machine stond [[mij gelulddg] makend] op mijn bureau. the machine stood [[me happy]sc making] on my desk

b.

In the domain of PPPs there is a very clear distinction between present participles of dynamic verbs, and present participles of stative verbs. PPPs with dynamic verbs only occur in attributive and adjunct position, as in (29), (31), and (32), but not in a predicative position, as in (30)' This is not the case for stative verbs, among which the object-experiencer verbs are most prominent. PPPs of these verbs show up in predicative position as well, as in (30). Interestingly, the internally verbal status of these stative PPPs is much less clear. This is especially true for stative PPPs in predicative position. I will provide three arguments to show that stative PPPs might be analysed as adjectival internally: adjectival stress, intensification, and prefixation. In Dutch, particle verbs have stress on the particle. This is true both for dynamic and stative particle verbs. Dynamic PPPs also show this stress pattern. However, as sCion as we find stative PPPs in predicative position, stress shifts to the verbal part of the present participle, corresponding to the pattern of complex adjectives. This is demonstrated in (33).

(33)

a.

b. c.

d.

Jan zal mij uit-zwaaien/*uit-zwaiien. John will me out-wave Die baan zal mij uit-putten/*uit-putten. that job will me ex-haust het mij uit-zwaaiende/*uit-zwaaiende publiek the me out-waving audience Die baan is *uit-puttend/uit-puttend. that job is ex-hausting

[+dyn] [+stat] [+dyn PPP] [+stat PPP]

In contrast to (33a-c) the PPP in (33d) appears to be adjectival internally. The same point can be made with respect to intensification (c£ Brekke 1988; Borer 1990, for English).15 In Dutch there are various intensifiers. The interesting point is that heel ('very') can only be added to adjectival projections, whereas erg ('very') is possible both " Brekke (1988) explicitly argues that only PPPs of experiencer verbs are internally adjectival, whereas other PPPs are internally verbal. This appears to be similar to the situation in Dutch, although the evidence in favour of a categorial distinction in English is much wealcer than the evidence available for Dutch. The major argument produced by Brekke is the fact that intensification by very separates PPPs of experiencer verbs ('very interesting') from PPPs of other verbs ('*very jumping'). If it is correct that intensification by very is a test for true adjectival status, only PPPs of experiencer verbs qualifY as adjectival projections. Borer argues against the qualifYing status of very intensification. She concludes that all PPPs are adjectival. For Dutch the evidence that at least dynamic PPPs are internally verbal is overwhelming, as we have seen above. The fact that intensification in Dutch seems to pattern with English, thus appears to support the distinction that is defended in Brekke.

Hans Bennis

102

in verbal and adjectival contexts. The contrast between (34a) and (34b) follows if we take the PPP verrassend ('surprising') to be adjectival.

(34)

a. b.

(35)

a. b.

Zijn optreden is erglheelverrassend. his performance is very surprising Hij verrast mij ergl*heel. he surprises me very een ergl*heel snurkende man a very snoring man Deze man snurkt ergl*heel. this man snores very

It appears that no semantic or lexical restriction on intensification is relevant. The non-occurrence of heel in (35a) strongly suggests that dynamic PPPs are verbal, whereas the occurrence of heel in (344) points at an adjectival internal structure. A third argument for the dual nature ofPPPs is the fact that adjectival affixation is possible with stative PPPs only, as is clear from on-prefixation and comparative formation. (3 6)

a. b.

(37)

een on-opwindende man an unexciting man een angstaanjagend-ere situatie a frightening-er situation

a. *een on-slapende man an unsleeping man

b. *een werkend-ere student a working-er student These three argument suffice to show that PPPs of dynamic verbs are internally verbal, whereas PPPs of stative verbs can be adjectival. However, this qualification requires some further refinement. It seems to be the case that dynamic PPPs are always verbal, whereas the categorial make-up of stative PPPs is dependent on the structural position of the PPP. In adjunct position (attributive or adverbial) the stative PPP can be either adjectival or verbal, whereas an adjectival internal structure appears to be required for PPPs in complement position (predicative). This is demonstrated in (38) and (39). (38)

a. b.

(39)

,6

a.

de het antwoord wetende/alwetende geleerde the the answer knowing/all-knowing scientist Die geleerde is *het antwoord wetend/alwetend. that scientist is the answer knowingl all-knowing het telkens weer ?6pwindende/opwindende verhaalI6 the again-and-again exciting story

If both possibilities are available in principle, there is a preference for adjectival stress. However,

Unergative Adjectives and Psych Verbs

b. c.

d.

!O3

Dat verhaal is telkens weer *6pwindend/opwindend. that story is again-and-again exciting het heel *6pwindende/opwindende verhaal the very exciting story het *on-6pwindende/on-opwindende verhaal the unexciting story

In (38) it is shown that stative PPPs may have direct objects in adjunct position only. In (39) we observe that stative PPPs may have verbal or adjectival stress in attributive position, whereas adjectival stress is required in the case ofPPPs in complement position. Moreover, adjectival stress is required as soon as adjectival modification shows up (intensification with heel or on-prefixation). From this we conclude that the following empirical generalizations hold: • PPP in adjunct position can be verbal; • PPP can be adjectival iff V is stative; • PPP must be adjectival if: a. PPP is in complement position; b. PPP shows adjectival modification (adj. morphology/intensification). The relevant point to be made here concerns the appearance of the experiencer argument in the case of PPPs of psychological verbs. It appears to be the case that the experiencer may occur in PPP constructions only if the internal structure of the PPP is verbal. If the PPP is adjectival the object experiencer (OE) must be left out. This is demonstrated in (40). (40)

a.

b. c.

d.

e.

de mij 6pwindende gebeurtenis the me exciting happening Die gebeurtenis is (*mij) opwindend. that happening is me exciting de (*mij) opwindende gebeurtenis de (*mij) heel opwindende gebeurtenis the me very exciting happening de (*mij) onopwindende gebeurtenis the me unexciting happening

The particle verb op-winden ('to excite') is a psychological verb of the object experiencer (OE) type. In (40a) we observe that the OE (mij) may appear in a PPP construction. Above it was argued that PPPs of stative verbs in attributive position are structurally ambigious between an adjectival and a verbal internal structure. In (40b) it is demonstrated that the OE (mij) does not appear when the PPP is a predicate in complement position. The ungrammaticality of the OE in (40b) can be related to the informants agree in that verbal stress is not impossible in these cases. The unacceptability of verbal stress in (39b-d) is clear and contrasts with the reduced acceptability of verbal stress in (39a). Similarly, the unacceptability of adjectival stress is absolute in (4oa, c).

Hans Bennis categorial nature of the PPP. As demonstrated above, the PPP in complement position is adjectival. The conclusion that the presence of the object experiencer depends on a categorial distinction, is supported by the data in (4oc-e) as well. The OE (mij) becomes impossible in attributive position when the PPP has adjectival properties, such as adjectival stress (4oc), intensification with heel (40d), or adjectival prefixation (40e). On the other hand, when the PPP is arguably verbal, as in (4oa), the OE (mij) is possible. If we assume that the object experiencer must be Case-marked structurally, we can provide a simple explanation of the contrast between (40a) and (40b-e). Adjectival projections do not feature Case-checking heads; if the OE must be structurally Case-marked, we expect the OE to occur in verbal projections only. We thus find ourselves in the position that we have strong arguments to claim that the presence of an object experiencer in PPPs is dependent on the presence of structural Case. At this point I disagree with an approach of the type adopted in Belletti and Rizzi (I988), in which the object experiencer is assumed to have inherent Case. The correlation between the categorial status (verbal vs. adjectival) and the appearance of an object experiencer cannot be explained easily if we assume the object experiencer to have inherent Case, since, as argued above (c£ (4)), inherent Case can be present both in Dutch in verbal and adjectival constructions. In section 3.2 I argued that structural objective Case depends on the presence of small v. Combining these results, we are now led to the conclusion that an object experiencer PPP may contain a vP and that the OE must be moved to the outer Spec position of this v in order to check its objective Case feature. The major result of this section is the insight that the object experiencer ofPPPs of psych verbs should receive structural objective Case. In the next sections I deal with regular OE-verbs. The most important issue concerns the argument structure of psych verbs and the relationship between the presence of an external argument and the availability of objective Case.

3.6.

COMPLEX ERGATIVE VERBS

In the literature the argument structure of psychological verbs and the related projection of arguments into the syntactic structure has been the cause of much debate (see, for example, Belletti and Rizzi I988; Pesetsky I995; Grimshaw I990; Arad I998a). Most authors agree that an unaccusative perspective on object experiencer verbs, as proposed by Belletti and Rizzi, is incorrect (Pesetsky, Grimshaw, Arad). In this section I provide arguments to claim that psychological verbs are ergative, though not unaccusative. They are the verbal counterparts of the complex ergative adjectives discussed in section 3+ The proposal defended here is in between the unaccusative analysis of Belletti and Rizzi (I988) and the unergative analysis of the others. Let us first compare standard psych-verb constructions with their 'real' causative counterparts. In (4I) a number of typical psych-verb constructions are listed. The

Unergative Adjectives and Psych Verbs experiencer is present as an object, and the theme-the object which causes the emotion-appears as the subject. Related to these constructions are the examples in (42), in which the subject is the possessor of the object that causes the emotion (the Causer). In this case, the object that causes the emotion can be present in an optional PP. (4I)

Dat gedrag amuseert/ontroert/verbaast/interesseert/ ... mij. that behaviour amuses/moves/astonishes/interestsl ... me

(42)

Jan amuseert/ontroert/verbaast/interesseert/ ... mij met dat gedrag. John amuses/moves/astonisheslinterestsl ... me with that behaviour

The sentences in (42) appear to be normal transitive clauses with an external argument, generated/merged in [Spec, v]. The experiencer is a direct object in all relevant respects. Two arguments in favour of the direct-object status of the experiencer in (42) concern the availability of passive variants, and the occurrence in participle constructions. Passive variants of (42) are presented in (43).

(43)

a. b.

Ik werd door Henk ge'interesseerd voor taalkunde. I was by Henk interested for linguistics Ik werd door Jan ontroerd met een serenade. I was by John moved with a serenade

Given the analysis of passivization along the lines of Balcer, Johnson, and Roberts (I989), as discussed above, it follows that the direct object cannot check its Case by movement to the outer Spec of v, since the passive morpheme checks the objective Case feature in v. The object thus has to move to [Spec,T] in order to pass the Case requirement for DPs. It thus follows that in causative constructions of the type in (42) the experiencer is structurally Case marked, either by adjoining to vP in Active clauses (42), or by adjoining to TP in passives (43). A similar argument can be derived from the distribution of past participles. If the participle of a verb is used in attributive position within DP, the head noun of the DP corresponds to the (underlying) direct object. This is demonstrated in (44) for a normal bitransitive verb. In (45), we observe that causative psych verbs behave similarly.

(44)

De jongen overhandigt de voorzitter een cadeau. the boy gives the chairman a present b. *de overhandigde jongen (the given boy) c. *de overhandigde voorzitter (the given chairman) d. het overhandigde cadeau (the given present)

(45)

a.

a.

De jongen amuseert het publiek met een redevoering. the boy amuses the audience with a speech b. *de geamuseerde jongen (the amused boy)'? c. het geamuseerde publiek (the amused audience)

'7 This phrase is ungrammatical in the intended interpretation in which de jongen ('the boy) is the causer of the emotion.

Hans Bennis

106

d. *de geamuseerde redevoering

(the amused lecture)

If causative psych verb constructions of the type in (42) are regular transitive constructions indeed, it appears to be the case that the psych verb constructions in (41) are the ergative counterparts of these transitives. However, not in the normal sense of ergativization. The experiencer object of the unergative clauses in (42) still appears as the direct object of the ergative clauses in (41). What happens to become subject in the 'ergative' variant is the object that causes the emotion (the Theme); the phrase that shows up in a prepositional phrase in the non-ergative construction. There are at least four arguments for claiming that (41) is ergative: (i) the internal argument of the unergative construction shows up as the subject of the ergative construction (c£ (41) vs (42»; (ii) the external argument of the unergative construction can be added in an optional PP; (iii) the two arguments of the ergative construction behave as internal arguments with respect to inversion (section 3.3-1); (iv) the subject of the ergative construction behaves as an object with respect the occurrence in as clauses (section 3.3.2). The arguments (ii)-(iv) are illustrated below. As in the complex adjectival construction discussed in section 3.4, the external argument of the unergative construction can be added to the ergative construction in a van-phrase. Examples are provided in (46). Dat verbaast/irriteert/ ... mij van hem. that astonisheslirritates/ ... me of him

(46)

In section 3.3.1 it was shown that the inversion of arguments in Dutch is restricted to cases in which the surface subject is an underlying direct object. The fact that the construction in (41) readily allows this kind of inversion constitutes a strong argument in favour of a derivational approach. Instances of inversion are given in (47b). (47)

a.

b.

dat die voorstelling mij amuseert/behaagt/irriteert/ .. . that that performance me amuses/pleases/irritates/ .. . dat mij die voorstelling amuseert/behaagt/irriteertl .. . that me that performance amuses/pleases/irritates/ .. .

A final argument concerns the ~ppearance of psych verbs in as clauses. In the examples in (48) the experiencer is lexically realized, and we encounter a gap in subject position. As argued above, the gap in as clauses corresponds to a sentential underlying object (see also Stowell 1987). If that is correct, it supports an ergative analysis of the construction in (41). (48)

a.

b.

Zoals mij telkens weer verbaast, houdt Jan van slakken. as me again-and-again surprises, loves John snails Zoals mij altijd irriteert, wast Jan zijn handen niet voor het eten. as me always irritates, washes John his hands not before dinner

Unergative Adjectives and Psych Verbs

!O7

The conclusion from this section thus far is that the subject of an EO psych-verb construction as (4I) is derived by movement from object position. Apparently this contradicts the conclusion we have reached in the preceding section, where it was argued that the experiencer is assigned objective Case. If the subject of a psych construction of the type in (4I) is an object, the construction does not have an external argument. If Burzio's generalization is correct, the lack of an external argument implies the absence of structural objective Case. Given that both structural claims-EO psych verbs have no EA and the EO receives structural, objective Case-have been extensively motivated, we have to conclude that Burzio's generalization is empirically wrong. Let us try to provide a structural analysis of EO psych verbs. Given that the EO bears structural objective Case, the verb phrase should contain a v-layer. Given that the EO construction should receive an ergative analysis, no external argument is generated in [Spec, v]. The two arguments in an EO construction should be analysed as internal arguments-that is, arguments within VP. The question that remains is how these two arguments are projected: which of the two is generated in the specifier position of V and which one in complement position. There are various arguments suggesting that the experiencer is in [Spec,V] and the Theme (the object that causes the emotion) in complement position. For instance, NP-inversion in Dutch shows up only ifboth orders are structurally motivated. In those cases in which the 'thematic order' differs from the 'Case order', as in passives ofbitransitive verbs or ergative verbs with an indirect object, NP inversion can be observed. If that is the correct generalization, it follows that the occurrence ofNP-inversion in EO constructions, as in (47), leads us to analyse the experiencer as being generated in [Spec,V]: the nominative-accusative order in (47a) is the order that corresponds to 'Case order', whereas the opposite order in (47b) should correspond to the underlying order. Let us therefore assume that EO is generated in [Spec,V]. This results in the underlying structure in (49).

Spec

v'

o

~ v

VP

~

Spec EO

V'

~ V

Compl THEME

Both arguments in (49) have to be moved in order to check their Case features. Given the presence of v--which automatically implies the interpretive aspect of agentivity/ causativity-and the absence of a Case-absorbing passive morpheme, accusative Case is available. If it concerns a finite clause, nominative Case is available as well. EO will be adjoined to vP in order to check accusative Case. Given that V moves to v and v to T, the Theme can be moved to [Spec,T] without violating locality or minimality

Hans Bennis

108

constraints. The important point is that the exceptional behaviour of EO psych verb constructions can be accounted for without stipulation by introducing the concept of complex ergativity. Before drawing a conclusion based on the distinction between simplex and complex ergativity, I will demonstrate how the variety of psych verb constructions can be accounted for in a framework as sketched above.

3.7.

PSYCH VERBS

In the literature, a distinction has been made between three different types of psych verb construction. The first distinction is whether the thematic experiencer shows up as the subject or the object of the construction. This difference is illustrated by the standard examples in (50). The relevant groups of psych verbs are also known as the temere (Italian, 'fear') class of psych verbs (subject experiencer) and the preoccupare (Italian, 'preoccupy') class (object experiencer). (50)

a. b.

John fears snakes. Snakes frighten John.

The second division is more subtle. The class of verbs with object experiencers is often divided into 'unergative' OE psych verbs (thepreoccupareclass; OE1) and 'ergative' OE psych verbs, the piacere (Italian, 'please') class of psychological verbs (OE2). In this section I will indicate how the three groups of psych verbs are structurally distinguished within the framework developed here. Let us first concentrate on subject-experiencer verbs. AB far as I can see, there is no reason to consider subject psych verbs to be a separate class of verbs. They belong to the class of stative verbs with an external argument (Le. vP, where v is [+statl-cf. section 3.2). The similarity between the two can be illustrated by the possibility to change these verbal constructions into possessive constructions which are construed by using the possessive verb hebben ('have') and changing the verb into a bare nominal, as in (52). .

(51)

a. b.

(52)

a. b.

Jan kent mijn antwoord. John lmows my answer Jan vreest/bewondert/waardeert mijn antwoord. John fears/admires/appreciates my answer Jan heeft kennis van mijn antwoord. John has knowledge of my answer Jan heeft vrees/bewondering/waardering voor mijn antwoord. John has fear/admiration/appreciation for my answer

Subject experiencer verb constructions are regular transitive constructions in which the head of the v-layer is interpreted as stative. There appears to be no structural difference between the constructions which are traditionally labelled as being 'stative' and

Unergative Adjectives and Psych Verbs those which are called 'subject experiencer verb constructions'. 18 More interesting is the group of object experiencer verbs. In section 3.6 the case of complex ergative psych verbs was discussed at some length. The question now arises how we can account for the distinction between the two groups of object experiencer verbs (OEI vs OEz). Before jumping to the rather obvious conclusion, I will first illustrate that a division into two groups of object experiencer verbs is indeed motivated in Dutch on empirical grounds. There are at least four criteria to substantiate this division: (i) (ii) (iii) (iv)

auxiliary selection; passivization; past participle constructions; present participles constructions.

Well-known is the fact that OE-verbs fall in two groups with respect to the selection of the auxiliary. In general, ergative verbs select the auxiliary zijn ('be'), whereas unergative verbs select the auxiliary hebben (cf. Hoekstra 1999). Some OE-verbs (OEz) select zijn (53a), others (OEI) select hebben (53b).

(53)

a.

b.

Dat is mij bevallen/tegengevallen/overkomen. that is me pleased/disappointed/happened Dat heeft mij geamuseerd/geirriteerd/opgewonden. that has me amused/irritated/excited

This indicates that OEz-verbs such as bevallen are ergative, contrary to the OEI-verbs that have been discussed in section 3.6. OEz-constructions (53b) have passive variants, whereas no passives are available for OEI-verbs of the type bevallen. '9 This is shown in (54).

(54)

a. *Ik word daardoor bevallen/overlcomen.

b.

I am thereby pleased/happened lk word daardoor opgewonden/geirriteerd. I am thereby excited/irritated

A third difference between the two classes ofOE-verbs concerns the behaviour of past participles in attributive position. Whereas the head of the relevant DP corresponds to the object ofOEI-verbs (56), it corresponds to the subject ofOEz-verbs (55).>°

18 One major difference between 'know' and 'fear' is the fact that passivization is easily available for verbs of the subject experiencer type, whereas it appears to be impossible for 'know'. Structurally, the occurrence of passivization is expected, since these verbs are regular unergative verbs. The ungrammaticality of passivization in the case of 'know' verbs remains to be explained. 19 It appears to be the case that the participle in (54b) is adjectival: the passive auxiliary worden 'become' can be replaced by the verb raken 'get' and the participle allows adjectival modification with heel (cf. sect. 5). However, it does not seem to diminish the relevance of the contrast in (54). 20 The ungrammaticality of (55c) and (5Gb) concerns the intended interpretation only. (55c) can be interpreted in such a way that the director is the theme argument, and not the experiencer.

Hans Bennis

no (55)

a.

(56)

a.

Dat gedrag bevalt de directeur. that behaviour pleases the director b. het bevallen gedrag the pleased behaviour c. *de bevallen directeur the pleased director Dat gedrag amuseert de directeur op. that behaviour amuses the director b. *het geamuseerde gedrag the amused behaviour c. de geamuseerde directeur the amused director

Finally, the two classes of OE-verbs behave differently with respect to their occurrence as present participles. Whereas OEl-verbs are ambiguous between an adjectival and a verbal internal structure, as has been discussed in section 3.5, no such ambiguity arises with OE2-verbs of the type bevallen. These latter verbs are exclusively verbal internally, similar to non-stative verbs. The non-occurrence of OE2-verbs with adjectival intensification (57a) and in complement position (58a) contrasts with the behaviour of OEl-verbs. 21 (57)

a.

b. (58)

een *heel/erg tegenvallende voorstelling a very/very disappointing performance een heel/erg opwindende voorstelling a very/very exciting performance

a. *Die voorstelling is bevallend. that performance is pleasing b. Die voorstelling is opwindend. that performance is exciting

Having established that a distinction between OEl-verbs and OE2-verbs is motivated, the question arises as to how this distinction can be accounted for structurally. As indicated above, the solution is quite obvious. OEl-verbs are the complex ergative verbs discussed in section 3.6 above. This allows us to analyse OE2-verbs as simplex ergative verbs. Let us investigate how the differences between OEl and OE2-verbs may follow from a structural difference in terms of simplex vs. complex ergativity. The selection of the auxiliary (hebben vs zijn) has always been a theoretical problem. It provided a test to distinguish ergative from unergative verbs, but is was unclear how " The verb apva/len ('to strike') does not behave as expected. It is one of the OE2-verbs with respect to auxiliary selection (zijn), absence of passivization, and past-participle behaviour. However, with respect to present participles it behaves like OEl-verbs: it allows adjectival modification, adjectival prefixation, adjectival stress, and it occurs in complement position. I have no idea why this verb shows this atypical behaviour.

Unergative Adjectives and Psych Verbs

III

it should be accounted for. On the basis of the discussion here, we may formulate a provisional analysis: hebben is selected if the complement is vp, otherwise the auxiliary is zijn. 22 The occurrence of passives was argued to be dependent on the presence of the v-layer (section 3+4). Given that the v-layer is absent in simplex ergatives, no passivization of OE2-verbs is expected. The difference with respect to past-participle interpretation may follow as well. In OE2-verbs, the experiencer is an indirect object with inherent Case. Given that the head noun in an attributive participle projection corresponds to the underlying direct object, it is to be expected that the head noun corresponds to the surface subject in simplex ergative OE2-constructions. For OEIconstructions the matter is more complicated. In these complex ergatives, both DPs may be considered direct objects: the theme is a direct object thematically, whereas the experiencer is a direct object with respect to Case. Apparently, structural accusative Case is the relevant criterion in selecting the argument that is targeted in participle constructions. This fits in with the general approach to this phenomenon in which it is assumed that attributive participles are passive participles. Finally, the difference in behaviour with respect to present participle formation may also be related to the structural difference between simplex and complex ergativity. The crucial point is that the lack of a v-layer in simplex ergatives makes it impossible to initiate a categorial change from v to a. Such a categorial change is only possible if v is stative, just like adjectives. The fact that only unergative stative verbs and complex ergative verbs allow such a categorial transformation, supports the theory as developed here. We thus conclude that the set of psychological verbs can be divided into unergative stative verbs (subject experiencer verbs), complex ergative verbs (OEI-verbs) and simplex ergative verbs (OE2-verbs). There appears to be no clear reason to argue that psychological verbs are to be distinguished theoretically. It is also unclear whether the notion 'experiencer' has any theoretical significance. What has to distinguished in the lexicon is whether a particular verb has a v-layer or not. Preferably this distinction is related to an aspectual distinction. A second division concerns the question whether a particular v-projecting verb may occur without external argument (unergatives vs complex ergatives). Again it might be the case that the availability of this option is semantically determined. The fact that typically psychological verbs (OEI) appear as complex ergatives might be the result of the absence of the external argument. In that case the meaning of the OEI construction might be the consequence of the absence of the external argument, combined with the presence of the causativity in v.

12 The problem with such a formulation of auxiliary selection is the fact that the auxiliary is zijn in the case of passives. In sect. 3+4 it was argued that passives are vPs. It might be possible to maintain the proposed aux-selection procedure if we analyse passive participles as adjectival externally (i.e. aP). This would also provide an answer to the question why passive participles appear in attributive position, contrary to active participles. For reasons of space, I will not elaborate on this issue here. Another way to approach this issue is to claim that hebben is selected if the v is able to assign Case to [Spec, vJ. If not, i.e. in the case of simplex ergatives and passives, the auxiliary is Zijll.

Hans Bennis

Il2

3.8.

CONCLUSION

Just as in adjectival constructions, it appears to be the case that the concept of verbal ergativity is not monolithic structurally. We have in fact encountered three different configurations in which the external argument is absent:

(a) there is no v-layer on top ofVP; consequences: no external argument, no accusative Case; (b) there is a v-layer on top ofvp, and v contains a passive morpheme; consequences: the passive morpheme absorbs the external argument and accusative Case; (c) there is a v-layer on top of VP and no external argument is generated; consequences: accusative Case is available. These three configurations correspond to (simplex) ergative constructions, passive constructions, and experiencer object constructions, respectively. Theoretically, it does not require much to obtain this three-way distinction. The theoretical assumptions required to reach this result are: • • • •

the verb phrase is/can be a layered constituent; the external argument is the specifier of v; structural accusative Case is checked in vP; the passive morpheme is generated in v; it absorbs the thematic role and the Case features of v; • v mayor may not have a phrase generated in its specifier position.

As to the concept of 'unaccusativity', I conclude that it lacks any theoretical relevance. The availability of accusative Case is an automatic consequence of the presence of v. The accusative-Case feature can be checked in the outer Spec position of v (structural accusative Case) or in the head position (passive morpheme). Absence of the v-layer implies the absence of accusative Case. The concept of'ergativity' is not a p'rimitive of the theory either. It is interpreted as the empirical correlate of structures in which no external argument has been generated. This can be due to the lexically or semantically determined absence of the v-layer ('simplex ergativity') or to the absence of an argument in the specifier position of v ('complex ergativity'). The theoretical axiom that is known as 'Burzio's generalization' is now deprived of any theoretical significance as well. It is indeed true that the presence of an external argument correlates with the availability of accusative Case. If an external argument is present, it indicates the presence of a top-layer or an x-layer; if this layer happens to be of the type vP, accusative Case is available. However, it is not the case that the availability of accusative Case is indicative for the presence of an external argument. We have seen that EO psych-verb constructions are a case in point: accusative Case is available although an external argument is absent. As claimed in section 3.2, Burzio's generalization is partially wrong-insofar it is correct, it follows from the theory without additional stipulations.

Unergative Adjectives and Psych Verbs

II3

Finally, the approach to unaccusativity adopted here makes no use of the semantic concept of telicity, as many others have proposed. I have defended a purely syntactic analysis of unaccusativity which does not directly involve the syntax-semantics interface. Empirically, this analysis is strongly supported by the occurrence of stative unaccusatives. If object-experiencer verbs are indeed ergative verbs, the semantic approach that makes crucial use of concepts such as telicity and agentivity is seriously weakened.

4 Voice Morphology iri the Causative-Inchoative Alternation: Evidence for a Non-Unified Structural Analysis of Un accusatives Artemis Alexiadou and Elena Anagnostopoulou

4.r.

INTRODUCTION

The Unaccusativity Hypothesis, as formulated by Perlmutter (1978) and later adapted by Burzio (1981) and others, claims that the class of intransitive verbs is not homogeneous, but consists of two subclasses-unaccusative verbs and unergative verbs. From Earlier versions of this chapter were presented at the Workshop on Unaccusativity in Berlin, May I998; at the Poster Session at NELS 29, Oct. I998, at the University of Delaware; at the CUNY Linguistics Colloquium, October I998; at the Linguistics Colloquium of the University of Princeton, November 2000; and at the applicative festival in MIT, January 200I. We thank these audiences for their comments. We also thank Len Babby, Maggie Browning, David Embick, Martin Everaert, Christiane Fellbaum, Hans Kamp, Arnim von Stechow, Hubert Truckenbrodt, Edwin Williams, and two anonymous reviewers for their suggestions. Alexiadou's research was partially supported by the DFG grant AL 5541r-I and a J. Seeger fellowship from the University of Princeton, which are hereby acknowledged. A much condensed presentation of parts of the material discussed here is included in Alexiadou and Anagnostopoulou (I999). Apart from being more elaborate, the analysis presented here is also different, as we no longer take transitivity to be a primitive.

Voice Morphology

II5

a Principles-and-Parameters perspective (PP) (Chomsky 1981 and subsequent work), each class is associated with a different underlying syntactic configuration: an unergative verb takes a deep-structure subject and no object, whereas an unaccusative verb takes a deep-structure object and no subject (cf. I):

a. NP [yp V] b. [yp VNP]

unergative John dances unaccusative John returned

Early syntactic research on unaccusativity concentrated on the view that a D-structure object is the complement of V (strictly), as depicted in (I), thus expressing the distinction between the two classes in terms of the specifier-complement asymmetry. More recent research, though, has abandoned this representation due to the impact of the structures proposed in Larson (1988) according to which, internal arguments are not uniformly represented as complements of V strictly.' In Hale and Keyser's (1993, 1998) configurational theta theory, too, which takes thematic roles as defined in terms of the relations 'specifier' and 'complement' of specific categories, theme arguments are also represented as either specifiers or complements of heads. A different line of research, which also drops the specifier!complement distinction, relates argument interpretation to the positions arguments occupy in the functional domain, in particular their placement with respect to Aspectual heads, rather than relying on lexical information to link arguments with particular X-bar positions in the VP domain (see, for example, Borer 1994 and in this volume and van Hout 1996 and in this volume). According to this proposal, arguments are interpreted as eventoriginators when they are specifiers of high Asp and as event-measurers when they are specifiers oflow Asp. The general idea that the various verb classes are distinguished syntactically through differences in functional structure is also pursued in the semantic proposals of, for instance, Kratzer (1994, 2000) and von Stechow (1995). In these proposals the external argument is not an argument of the verb, but rather, it is introduced by a separate head Voice. In Kratzer, Voice adds an agent! causer when combining with action predicates and an experiencer! possessor when combining with stative predicates. In such views it is natural to propose that the presence or absence of VoiceIv is the determining factor for the classification of predicates into the unaccusative or unergative class (see also Chomsky 1995, Collins 1997, and section 4.3 for more discussion). While for Borer the difference between the two classes is the result of the presence of specific aspectual heads and how the argument moves to or through those heads, for an approach embedded within Kratzer's assumptions it is the presence vs. absence of v that would make predicates behave one way or another. We will argue for an approach along the lines of Kratzer and von Stechow, but crucially we will not link unaccusativity to presence vs. absence of Voice. Our contribution to this discussion comes from an examination of the distribution of voice r See also Marantz (1993) for a more radical proposal along the same lines, according to which internal arguments are always specifiers.

Artemis Alexiadou and Elena Anagnostopoulou

n6

morphology associated with detransitivization in transitivity alternations, and in particular, the causative-anticausative alternation. The investigation of (a) the consistency vs. gaps in the distribution of this morphology and (b) the semantic/syntactic properties such a marking correlates with, leads us to a picture of the structural representation of unaccusativity which is more refined than the one sketched above in that we do not take anticausatives to have a unified structure (a related view is expressed in Borer 1991 and Bennis, in this volume). More specifically, we argue that the patterns of voice morphology can be accounted for if there are at least three structures involved in the formation of anticausatives in Greek. We propose that anticausatives are formed on the basis of an intransitive v BECOME/RESULT which embeds either an Adjectivep, or a VoiceP or a possessive construction. The chapter is structured as follows: in section 4.2 we present the factual background for our discussion, namely the distribution of 'special' morphology on the intransitive variants of alternating verbs. In section 4.3 we present our theoretical assumptions, and an explanation that has been proposed in the literature for the distribution of this special morphology. In section +4 we turn to a discussion of the Greek verb classes and consider the morphological patterns these exhibit. In sections 4.5 and 4.6 we discuss the regularities that emerge from the distribution of 'special' morphology in Greek and offer our explanation for the differences among the various patterns.

4. 2 .

DE T RA N SIT I V I Z I N G M 0 R PH 0 LOG YIN

CHANGE-OF-STATE UNACCUSATIVES, PASSIVES, AND REFLEXIVES

As is welllmown, across languages several verbs, depending on their semantic type, enter in transitivity alternations; see the examples in (2). The intransitive/inchoative or anticausative counterpart of the alternation qualifies as an unaccusative predicate. (2)

a.

b.

The window broke Mary broke the window

A pattern found in a number of unrelated languages involves the presence of morphology associated with de-transitivization of the intransitive member of the alternation. This morphology is shared by reflexive and/or passive predicates and has the form of a pronoun, a clitic, or verbal inflection (Partee 1965; Reis 1973; Marantz 1984; Everaert 1986; Cinque 1988; Chierchia 1989; Klaiman 199I; Reinhart I997; von Stechow 1995; Alsina 1996, among many others). In (3) we illustrate this pattern on the basis of Greek. Greek has two sets of forms for Active and Non-active inflection across two tenses and two aspects.

Voice Morphology (3)

a. Active forms of grafo ('write') Imperfective PN 1SG 2SG

3SG 1PL 2PL

3PL

b.

Non-past graf-o graf-is graf-i graf-ume graf-ete graf-un

Past e-graf-a e-graf-es e-graf-e graf-ame graf-ate graf-ane

II7

Perfective Non-past grap-s-o grap-s-is grap-s-i grap-s-ume grap-s-ete grap-s-un

Past e-grap-s-a e-grap-s-es e-grap-s-e grap-s-ame grap-s-ate e-grap-s-an

Non-active forms of grafo

Imperfective PN 1SG 2SG

3SG 1PL 2PL

3PL

Non-past grafo-me grafe-se grafe-te grafo-maste grafe-ste grafo-nde

Perfective Past graf-o-muna graf-o-suna graf-o-tan graf-o-mastan graf-o-sastan graf-o-ndan

Non-past graf-t-o graf-t-is graf-t-i graf-t-ume graf-t-ite graf-t-un

Past graf-tik-a graf-tik-es graf-tik-e graf-tik-ame graf-tik-ate graf-tik-an

As is signalled by (4b), Non-active morphology occurs on the anticausative members of the causative-anticausative alternation: (4)

a.

b.

0 Janis

ekapse ti supa. the John.NoM burnt.Act the SOUp.ACC 'John burnt the soup.' I supa kegete. the SOUp.NOM burns.Nact 'The soup is burning.'

Other Non-active morphology environments in Greek are passives, inherent reflexives, and reflexive constructions involving a transitive verb prefixed with the marker afto ('self') (Rivero 1992; Embick 1998; Anagnostopoulou and Everaert 1999):2

(5)

a.

To vivlio diavastike ktes. (Passive) the book.NOM read.Nact yesterday 'The book was read yesterday.'

2 There is a fourth case involving the Non-active form, namely, deponent verbs (Mackridge I985: Embick 1998), which are not part of an alternation. Deponents are transitive verbs which lack active forms altogether. Syntactically, deponent verbs behave exactly like all other transitive verbs: the deponent verb takes a nominative subject and an accusative object:

(i)

metahirizome to leksiko. (Deponent) use-Nact the dictionary '1 use the dictionary.'

Artemis Alexiadou and Elena Anagnostopoulou

u8

TABLE 4.1. Syncretism in reflexives, anticausatives, and passives

Tigre Motu 'O'odham Modern Greek Kanud Margi Uigur Wowora Udmurt Nimboran Danish (-5)

b.

c.

Reflexive

Anticausative

Passive

x (x) x x x x

x x x x x x x

x x x x x x x x x x x

x x

x x x

I Maria htenizete. (Inherent reflexive) the Mary.NoM combs.Nact 'Mary combs herself.' I Maria afto-katastrefete. (Self-reflexives) the Mary.NoM self-destroys.Nact 'Mary destroys hersel£'

Tahle 4.1 (from Haspelmath 1990: 36) illustrates the systematicity of this syncretism for a number of typologically unrelated languages.

4.3.

DETRANSITIVIZING MORPHOLOGY IN THE ABSENCE OF EXTERNAL ARGUMENTS

What is the common property of anticausatives, passives, and reflexives which detransitivising morphology is sensitive to? One influential proposal suggested in the literature is that the property in question is the absence of an external argument (see Marantz 1984, for instance). Various formulations of this have been developed (see the discussion in Embick, in this volume). We concentrate here on a recent proposal for Greek which is embedded within the set of assumptions put forth in Kratzer (1994) and Chomsky (1995). As mentioned in section 4.1, Kratzer (1994) and Chomsky (1995), among others, have argued that a functional category Voice (or Event Phrase or little v Phrase) determines the transitive vs. intransitive and the eventive vs. stative nature of a predicate. In (6) v introduces the external argument (Williams 1981a).

Voice Morphology (6)

vP

~

v'

DP

~ VP ~

v

VO

Comp

A recent trend in the literature is to assume that v is always present in transitive and unaccusative constructions and may have the following properties (see, for example, Harley 1995; Collins 1997; Marantz 1997; Embick 1998 and in this volume; Arad 1999; Travis 1999; Alexiadou 2001). (7)

a. b. c.

d. e.

v is the locus of agentivity, i.e. of features relevant to the licensing and interpretation of external arguments. v bears Case features for the object (Burzio's Generalization results from a and b). v bears features related to eventivity. v bears features related to the licensing of a manner component (manner adverbs). v comes in two types: one that introduces an external argument, and one that does not.

In this approach, the difference between transitives, passives, reflexives, and unaccusatives is analysed in terms of differences in the feature specification of v and the presence or absence of an external argument. Concerning voice morphology in Greek, Embick (1998) argues that the presence of Non-active morphology is sensitive to the absence of an external argument: 3 (8)

V -7 V-VOC[NonAct]/_ No external DP argument

In particular, Embick proposes that Non-active morphology is the morphological reflex of the absence of an external argument, where external argument is interpreted literally-that is, when v has no specifier, then v bears Non-active morphology. According to Embick, voice morphology does not correspond to any syntactically Active element (for example, it is not a ditic-like element absorbing a a-role, as suggested in, for instance, Bal{er, Johnson, and Roberts 1989). Nact is a morphological feature which is assigned in specific syntactic configurations, and reflects properties l The formulation in (8) differs from the one adopted in Embick (this volume), where use of the term u-syncretism is made and the common morphological properties of passives, reflexives, and anticausatives are treated in terms of underspecification. (i) illustrates Embick's representation (in this volume):

(i)

v H v-Xl_no external argument

By underspecif)ring -X, Embick offers an explanation for the common morphological properties of the three classes.

Artemis Alexiadou and Elena Anagnostopoulou

120

of the syntactic environment, rather than itself affecting an alternation. 'The feature [nact] is assigned post-syntactically to the verb when it (or the v-V complex) is not in a local relationship with an external argument. On this view, voice features are properties of morphology and thus, all occurrences of [nact] belong to the morphology component. 4 However, the actual distribution of voice morphology in anticausatives strongly suggests that this analysis is only partially correct. As we show in the following sections, (8) accounts for some but not all the patterns to be discussed (see also 'Theophanopoulou-Kontou 2000). We point out that the structure in (6), with the external DP argument missing, is not sufficient to account for the properties of unaccusative predicates and needs to be refined. To account for the distribution of voice morphology in Greek we propose that anticausatives are built on the basis of an operator denoting result (see Kratzer 1994, 2000)5 which we will call here BECOME/RESULT. 'This BECOME/ RESULT should be considered closely related to, but not identical to, Dowty's (1979) and Van Valin's (1990) BECOME operator which yields achievements, according to the predicate decompositions in (9):

(9)

a. STATE: predicate' (x) or (x, y) b. ACHIEVEMENT: BECOME predicate' (x) or (x, y) c.

d.

ACTIVITY: (±agentive): DO (x))[predicate' (x) or (x, y)] ACCOMPLISHMENT: CAUSE 'V where is normally an activity predicate and 'V an achievement predicate

For the typology in (9), accomplishments are possible only in the presence of a CAUSE operator which embeds a BECOME operator. 'This entails that anticausatives with an accomplishment reading-such as many deadjectival verbs-must necessarily be considered to be underlyingly causative, as suggested by, for example, Chierchia (in this volume). However, we start from the assumption that deadjectival anticausatives are basic, and causatives are derived from the addition of a higher causative layer (see also Hale and Keyser 1993). 'Thus the aspectual properties of an accomplishment must be compatible with the BECOME operator we postulate, which should not be limited to achievement readings. While we take V[BECOME/RESULT]-that is, a v type that does not project a specifier-to be related to all the patterns we discuss, we propose that voice morphology is never associated with this head and is always associated with a separated head Voice hosting agentivity and manner features (Hale and Keyser 1993; Levin and Rappaport 4 The difference between deponent verbs (see n. 2) on the one hand, and passives, reflexives, and anticausatives, on the other hand, is that the former are specified for this feature inherently, while the latter are assigned this feature in a particular syntactic configuration. 5 Kratzer discusses at great length two types of 'result' states, namely, Parsons' (1996) target states and result states. She argues that different types of participial constructions belong to the former or the latter class as can be diagnosed by a series of criteria. Our use of the term RESULT is deliberately left vague, since we have not investigated in detail how change-of-state verbs would fit in Kratzer's typology. The suggestion that the operator BECOME and Kratzer's RESULT operator can be collapsed and can be employed in the analysis of change-of-state anticausatives is due to Arnim von Stechow (personal communication).

Voice Morphology

121

Hovav 1995; Alexiadou 2001). Since certain classes of anticausatives systematically lack 'special' morphology, we must conclude that detransistivizing morphology is never associated with the 'specifier-less' v [BECOME/RESULT] (contra Embick) but rather with Voice embedded under it in (lOb). We furthermore propose that there are three configurations that are involved in the formation of unaccusatives as shown in (10); VoiceP is included in one of them. (10)

a.

b.

vP

~

~

v

BECOME

vP

AP ~

RESULT

VoiceP

I

Voice'

~

Voice

c.

HAVE

VP

a wrinkle

As shown in the structures in

(10), BECOME selects either an AP or a possessive construction-that is, it combines with a state which it brings into existence, as suggested by Dowty (1979). On the other hand, RESULT selects a verbal phrase-it combines with an event and brings about its target/result state, as suggested by Kratzer (2000). We further argue that in Greek, voice can carry manner features without necessarily carrying agentivity features, and that 'detransitivizing' voice morphology is sensitive to manner rather than agentivity (see Zombolou, in progress).6 Ifwe are right, then manner is the property unifiying passives, unaccusatives, and reflexives in Greek.

4.4.

THE DISTRIBUTION OF DETRANSITIVIZING

MORPHOLOGY IN GREEK ANTICAUSATIVES

4.4.1. The general picture

As in many other languages, Non-active morphology in Greek is always used with passives/reflexives, but only sometimes with anticausatives. For example, the anticausative in (lIb) surfaces with Active morphology. 6 Potentially, this property differentiates a language such as Greek with syncretism in passives, unaccusatives, and reflexives from a language such as English, where passives are distinguished from rhe other rwo categories. One could speculate that English voice necessarily combines agentivity with manner, while Greek Voice can have manner without having agentivity.

Artemis Alexiadou and Elena Anagnostopoulou

122

(II)

a.

b.

0 Janis adiase ti sakula. the John.NoM emptied the bag.Acc 'John emptied the bag.' 1 sakula adiase. the bag. NOM emptied.Act 'The bag emptied.'

Gaps of this type are never found in passives and reflexives, an asymmetry that is unexpected by the condition (8) and should be explained. A further complication concerning the distribution of voice morphology in anticausatives is that there are many cases in which an intransitive alternating verb can surface with both Active and Non-active forms, as illustrated in (12). (12)

a. b. c.

0 sismos gremise to ktirio. the earthqual<:e.NoM demolished the building.Acc To ktirio gremise apo mono tu. the building coliapsed.Act by itself To ktirio gremistil<:e apo mono tu the building coliapsed.Nact by itself

Given the morphological similarity between passives and anticausatives when the latter have Non-active morphology, a test which can successfully distinguish them is the type of optional prepositional phrase denoting the agent/causer licensed with them (see Alexiadou and Anagnostopoulou 1999 and Mavromanolaki 2002 for a recent discussion and further references). As the contrast in the examples in (13) illustrates, the preposition apo (,from') followed by a DP denoting the agent (corresponding to the English 'by' -phrase) is licensed with passives but not with anticausatives.

(13)

a.

To vivlio diavastike apo ton Petro. the book NOM read.Nact by the Peter 'The book was read by Peter.' b. *To bukali adiase apo ton Petro. the bottle. NOM emptied.Act by the Peter 'The bag emptied.' c. *1 supa kaike apo to Jani. the soup burnt.Nact by the John

(Passive)

(Anticausative)

(Anti causative)

On the other hand, apo followed by the anaphoric element mono to which a possessive elitic attaches (an expression corresponding to Italian da Sf (Chierchia, in this volume) or English by itself(Levin and Rappaport Hovav 1995)) is not permitted with passives, but is permitted with anticausatives (it is also not allowed with unergatives; see Levin and Rappaport Hovav 1995): (14)

a. *To vivlio

diavastike apo mono tu. the book NOM read.Nact by itself

(Passive)

Voice Morphology b. c.

To bukali adiase apo mono tu. the bottle. NOM emptied.Act by itself To pani skistike apo mono tu. the cloth tore.Nact by itself

123

(Anticausative) (Anticausative)

By this test, both Active (I4b) and NonActive (I4c) qualify as anticausatives, since both can co-occur with 'by itself'. And yet, only (I4c) is described by (8). In the following sections, we argue that the distribution of Act/Nact voice morphology shows certain regular patterns which lead to generalizations calling for an explanation. In our discussion we distinguish between two groups. The former group refers to patterns, which either straightforwardly fall under (8) or represent a double classification of predicates in the unaccusative and unergative class, again conforming with (8). The latter group refers to patterns that are not described by (8) and suggest that the structural analysis of these constructions is not as uniform as has often been assumed.

4.4.2. The cases that fit the condition [-external argument] In previous work (Alexiadou and Anagnostopoulou 1999), we have identified four classes of alternating verbs, depending on the type of morphology that occurs on the intransitive variant. In the the first class of verb alternations, Active morphology appears with the transitive verb, while Non-active morphology appears with the intransitive and the passive. This is summarized in Table 4.2 and is illustrated in (15).

(15)

c£

0 giatros giatrepse to Gianni. the doctor. NOM healed the Giannis.Acc 'The doctor healed John.' giatreftike apo monos tu. b. 0 Giannis the Giannis.NOM healed.Nact by himself giatreftike apo to giatro. c. ?O Giannis the Giannis.NOM healed.Nact by the doctor

a.

(Causative)

(Anticausative) (Passive)

The Non-active form qualifies as inchoative in (I5b) (the presence of a PP equivalent to the English 'by itself' is all right), and as passive in (I5c) (the presence of a PP introducing an agent is also all right)'? Other Greek verbs belonging to this class are TABLE 4.2. Class I Transitive

Intransitive

Active form

Non-active form

7 Note though that the agentive phrase in (I5c) is somewhat deviant. It is certainly better with the verb therapevome ('be healed'). For reasons that are not clear to us, this is often the case in Greek, where the passive is not as productive as in other languages.

Artemis Alexiadou and Elena Anagnostopoulou

124

kommatiazo ('tear'), miono ('decrease'), eksaJanizo ('diminish'), veltiono ('improve'), diplasiazo ('double'), singentrono (,collect/gather'), dhiadhidho ('announce/spread a rumour'), vithizo ('sink'), and sindeo ('connect'). Evidently, these verbs can be accommodated under (8). In the second class, Active morphology generally appears with the transitive variant, while Non-active morphology is used with the intransitive and with the passive-that is, they have the morphological pattern found with Class-I verbs. This is summarized in Table 4.3 and is illustrated in (16). TABLE 4.3. Class lIa

(16)

a.

b.

Transitive

Intransitive

Active

Non-active

0 Janis ekapse to vivlio/ti supa. the John.NoM burnt.Act the bookAcclthe SOUp.ACC 'John burnt the booklthe soup.' To vivlio/i supa bike. the bookNoM/the soup. NOM burnt.Nact 'The book/the soup burnt.'

However, in a limited set of cases (with a restricted class of arguments), the pattern illustrated in Table 4.4 emerges-namely, a pattern in which the intransitive verb surfaces with Active morphology (17). In this respect, Class II differs from Class 1. TABLE 4.4. Class lIb Intransitive Active

(I?)

fotia kei/*kegete. the fire.NoM burns.Act/*Nact 'The fire burns.'

Other verbs that behave similarly are Jotizo (,lighten), and psihreno ('cool') (verbs of light-emission and weather verbs). The first subcase of this class, (I6b), falls under (8). As for the few cases where Active morphology appears (17), they qualifY as unergatives by at least three criteria: (i) They cannot undergo the causative alternation:

(18)

*0 Janis ekapse ti fotia. the John.NoM burnt.Act the fire.Acc 'John burnt the fire.'

Voice Morphology

125

(ii) They cannot form adjectival passives: (19)

a. *1 fotia ine kameni.

vs.

b.

the fire is burnt To vivlio ine kameno. the book is burnt

(iii) The Active variant surfaces whenever the single argument, in (17) the fire, has an inherent property that causes the eventuality expressed by the verb to happen. This can be identified with Levin and Rappaport Hovav's (1995) notion of 'internal causation', underlying the syntax of unergatives. Crucially, when the verb surfaces with the Active, the single argument does not undergo a change of state (unlike, for example, the soup or the book). We conclude that Class-lIb intransitives are unergatives. As pointed out in section 4.3, unergative verbs include a v-marked [+external argument]. The absence of special morphology is expected, and thus this Class supports (8) rather than contradicting it, despite appearances to the contrary. Let us now turn to the second group of verb alternations.

4.4.3. The cases that do not fit the condition [-external argument] The third class of alternating verbs exhibits Active morphology both on the transitive and the intransitive variants of alternating verbs. This is summarized in Table 4- 5 and illustrated in (20). TABLE 4- 5. Class III TranistivelAnticausative Active Form

(20)

a.

b.

cf.

c.

0 Janis adiase ti sakula. the John.NoM emptied.Act the bag.Ace 'John emptied the bag.' I sakula adiase. the bag.NoM emptied.Act 'The bag emptied' 1 sakula adiastike apo to Jani. the bag.NoM emptied.Nact from the John

(Causative)

(Anticausative)

(Passive)

Many Greek verbs behave like this, including asprizo ('whiten'), kokinizo ('redden'), mavrizo ('blacken') (all change of colour verbs), katharizo ('clean'), stroggilevo ('round'), klino ('close'), anigo ('open'),plateno ('widen'), stegnono (,dry), stenevo ('tighten'), skureno ('darken'), stenevo ('narrow'), kathistero ('delay'), alazo ('change'), and ksepagono ('defreeze'). These verbs do not fall under (8). The fourth class of verb alternations consists of two subcases again. When the single

Artemis Alexiadou and Elena Anagnostopoulou

126

argument is inanimate both forms are possible, as summarized in Table 4.6 and illustrated in (21) (a case of apparent 'optionality' to which we return). TABLE 4.6. Class IVa

(21)

a.

b.

Transitive

Intransitive

Active

Non-active/Active [-animate]

0 Janis lerose to trapezomandilo. the John.NoM dirtied.Act the tablecloth.Acc 'John dirtied the tablecloth.' To trapezomandilo lerosellerothike apo mono tu. the tablecloth. NOM dirtied.Act/Nact by itself 'The tablecloth got dirty.'

On the other hand, with animate arguments only the Non-active form is possible, as summarized in Table 4.7 and illustrated in (22). TABLE 4.7. Class IVb

(22)

a.

b.

Transitive

Intransitive

Active

Non-active [+animate]

0 Janis lerose ti Maria. the John.NoM dirtied.Act the Mary.Acc 'John dirtied the Mary.' I Maria lerothike/*lerose. the Mary.NoM dirtied.Nact

Other verbs that behave similarly are zarono ('wrinkle'), tsalakono ('crumple'), zesteno (,heat'), skizo ('tear'), erimono ('desert'), madao ('pluck'), skorpizo, ('scatter'), gremizo ('demolishlcollapse') stravono ('bend'), ksekubono ('unbutton'), dialio ('dissolve'), ksediplono ('unfold'), and tentono ('stretch'). We will call these verbs the ('wrinklel crumple') class. 8 In (2Ib) both forms qualifY as anticausatives, as is evidenced by the fact that 'by itself' is possible with the Active and the Non-active. The form with Non-active falls under (8), but the form with Active does not. The fact that Non-active morphology is obligatory when the single argument is animate (22b), shows that the morphology

8 We call them this because we first noticed the pattern with tsalakollo 'crumple'. Interestingly, English does not seem to have unaccusative verbs of this type. It seems to us that this cross-linguistic difference follows from our proposal; see the comment on English in n. 6.

VrJice Morphology

127

is sensitive to the feature [+animateJ. These cases can be analysed as being reflexive predicates, which in Greek always have Non-active morphology. In what follows we will not be concerned with the properties of these predicates (see Embick, Reinhart and Siloni, and Steinbach, all in this volume, for some discussion on reflexive predicates); we will concentrate on Class IVa. Table +8 summarizes the patterns presented so far. Class III and the Active found with class IVa cannot be explained via (8). In the following sections we concentrate on exactly these cases. TABLE 4.8. Summary

Class I Class IIa Class IIb Class III Class IVa Class IVb

4.5.

Transitive Intransitive

Described by (8)

Active Active Active Active Active Active

Yes Yes Yes (unergative) No Yes (Nact)/no (Act) Yes (reflexive)

Non-active Non-active Active Active Active/Non-active [-animate] Non-active [+animate]

REGULARITIES IN THE DISTRIBUTION OF THE ACTIVE/NON-ACTIVE FORM

The question that naturally arises at this point is whether there are regularities behind the distribution of Active morphology in the cases that do not fall under (8). The answer to this question is positive. As we will show, the distribution of the Active form in Class III is systematic. Moreover, in the case of Class IV, where both forms are possible (the wrinlde/crumple class), the difference in form relates to a difference in meaning. Starting with Class III, it is striking that the verbs which systematically lack Nonactive morphology are those that belong to the class of deadjectival verbs (as first pointed out in Alexiadou and Anagnostopoulou 1999). In fact, we have not found any deadjectival verbs that show Non-active morphology; we have found change-ofstate verbs that show Active morphology and are not deadjectival, but not the reverse. A closer look into the morphological make-up of deadjectival verbs falling under Class III shows that they consist of a (usually adjectival) stem, to which affixes such as iz-, -ev, -en, and -on attach, followed by the verbal inflection. The corresponding adjectives consist of the same stem to which the adjectival endings attach:

(23)

Class III: Deadjectival Verbs

verb

Adjective

aspr-iz-o 'whiten' kokin-iz-o 'redden'

aspr-os/i/o 'white' kokin-os/i/o 'red'

Artemis Alexiadou and Elena Anagnostopoulou

128

mavr-iz-o 'blacken' kitrin-iz-o 'yellow' prasin-iz-o 'green' kathar-iz-o 'clean' stroggil-ev-o 'round' plat-en-o 'widen' stegn-on-o 'dry' sten-ev-o 'tighten' skur-en-o 'darken'

mavr-os/i/o black' kitrin-os/i/o 'yellow' prasin-os/i/o 'green' kathar-os/i/o 'clean' stroggil-os/i/o 'round' plat-is/iali 'wide' stegn-os/i/o 'dry' sten-os/i/o 'tight' skur-os/i/o'dark'

Following and modifying Alexiadou (2001), we propose that -iz, iaz-, -ev, -en, and -on are overt reflexes of BECOME/RESULT. 9 On this view, the structural representation of the verbs listed in (23) contains a v, which embeds an adjectival stem. IO If the only overt reflexes of the v present in the structure are the verbal suffixes listed above, then we have a representation compatible with the absence of Non-active voice morphology. Turning to the verbs of Class IV, which display optionality in their form, again we observe that the optionality is only apparent. It is important to note that there are meaning differences between the two forms (Active vs. Non-active), which are repeated again in (24)'" (24)

a.

b.

0 Janis lerose to trapezomandilo. the John.NoM dirtied.Act the tablecloth.Acc 'John dirtied the tablecloth.' To trapezomandilo lerose/lerothike. the tablecloth dirtied.ActiNact 'The tablecloth got dirty.'

While the Active form denotes a partial change, the Non-active form may denote a change of state that takes place completely.I2This can be seen on the degree to which 9 Our first intuition was to propose that these affixes are instantiations of a causative v. Given our analysis, though, these elements are instances of a change-of-state v. A plausible candidate for a causative v is the element -poio, as in megalo-poio ('I exaggerate a situation'), which can never surface without carrying voice morphology in the anticausative variant:

(i)

I katastasi *megalopoiise/ megalopoiithike. the situation exaggerated.Act/exaggerated.Nact

W Note that the label adjectival stem could be seen as a toot not specified for a lexical category (see Embick this volume for a discussion on roots and the role of the lexicon). n That such a difference might be expected was first pointed out to us by Bill McLure (personal communication). " Potentially, the active-Non-active distinction in (24b), which we describe as the difference between partial and total change, can be stated in terms of the aspectual distinction between an achievement and accomplishment (Dowty 1979). Tests such as compatibility with the imperfective/progressive show that the active form qualifies as an achievement, while the Non-active form qualifies as an accomplishment or, rather, what Dowty has called a degree achievement. In terms of aspectual properties, then, it seems that the Non-active form behaves similarly to deadjectival verbs surfacing with active morphology which also qualifY either as accomplishments or as degree achievements (see Mavromanolald 2002 for discussion). Since we have not carefully looked into the aspectual properties of change-of-state verbs we prefer not to state the distinction in such terms.

vnice Morphology

129

the object is affected by the event. If it is completely affected, the Non-active form must be used; if it is only incompletely affected (a change of state only of part of the object), then the Active form can be used. This difference is explicated in (25)-(27). In the first conjunct, which states that the change of state is partial, the Active form is comfortably used. In fact, the Active entails and not simply implies 'incomplete change'. This is shown in the second conjunct, where complete change is asserted by the adverb 'completely'. Here only the Non-active form is well-formed. The Active is ruled out as a contradiction. I)

(25)

To ktirio gremise se ena simio alIa den gremistike/*gremise the building collapsed.Act in one spot but NEG collapsed.Nact/*Act entelos. completely

(26)

To trapezomantilo lerose se ena simio alIa den lerothike/*lerose the tablecloth dirtied.Act in one spot but NEG dirtied.Nact/*Active entelos. completely

(27)

To pukamiso tsalakose s' ena simio alIa den tsalalmthike/*tsalalmse the shirt wrinlded.Act in one spot but NEG wrinkled.Nact/*Act entelos. completely

The reader should not be misled into thinking that the Non-active form exclusively denotes complete change. The same examples show that the Non-active form is also compatible with partial change. The Non-active is well-formed in the first conjunct which asserts partial change:

(25') To letirio gremistike se ena simio alIa den gremistike/*gremise the building collapsed.Nact in one spot but NEG collapsed.Nact/*Act entelos. completely (26') To trapezomantilo lerothike se ena simio alIa den lerothike/*lerose the tablecloth dirtied.Nact in one spot but NEG dirtied.Nact/*Active entelos. completely (27')

To pukamiso tsalakothike s'ena simio alIa den tsalalmthike/*tsalakose the shirt wrinkled.Nact in one spot but NEG wrinkled.Nact/*Act entelos. completely

I) Of course, in order for the meaning of the whole sentence not to be a contradiction, negation needs to be introduced in one of the conjuncts; in (25)-(27) the negation happens to be on the second conjunct, but it could also be on the first.

Artemis Alexiadou and Elena Anagnostopoulou Thus, while the Active asserts incomplete change, the Non-active is compatible with both total and partial change. On the basis of the discussion in this section, we conclude that there are generalizations to be made for the distribution of Active and Non-active morphology in Greek anticausatives, namely:

(i) Active Morphology systematically occurs on (a) deadjectival verbs and (b) wrinlde/crumple-verbs asserting partial change. (ii) Non-active morphology occurs on wrinkle/crumple-verbs that may denote complete change. In the next section, we propose that these generalizations can be captured through the postulation of distinct structures for each case.

4.6.

STRUCTURES OF ANTICAUSATIVES AND

DISTRIBUTION OF (NON- )ACTIVE MORPHOLOGY

4.6.1. Deadjectival verbs Starting from deadjectival verbs, we analyse (28a) as resulting from (28b), as already anticipated in section 4.5 (see the discussion of examples (23)). According to this analysis, deadjectival verbs are built upon a BECOME v which embeds an adjective. The presence of BECOME derives the change-of-state interpretation, as proposed by Dowty (1979), who builds on the generative semantics literature (Lakoff 1965; McCawley 1968). Thus, the structure (28b) receives the interpretation in (29). Assuming that predicate adjectives assign an external argument (Higginbotham 1985 among others), v is embedding a small-clause complement (28b). (28)

a.

To pukamiso aspr-is-e the shirt whitened

b.

vP

~ v

iz

AP

~

DP

At

I

to pukamiso

A

I

aspr(29)

whiten: become white '4

'4 It has been argued that the degree-achievement interpretation of 'whiten' results from an ambiguity between 'become white' and 'become whiter'; see, for example, Abusch (1986) and Borer (1991).

Voice Morphology

13 1

Causatives of deadjectival verbs are formed when (28b) is embedded under a causative v:

(30)

vP

~

DP

v'

~

vP

CAUSE

~ BECOME

AP

4.6.2. Non-active wrinkle/crumple verbs Turning to Non-active wrinlde/crumple verbs, we propose that these also have a BECOME/RESULT component, similar to the deadjectival verbs discussed in the previous section. Unlike deadjectival verbs, however, N-act wrinkle/crumple verbs are formed on the basis of RESULT v which takes as complement a Voice phrase lacking agentive features but containing manner features: '5

(31)

a.

Deadjectival verbs

b.

Nact-wrinkle/crumple

vP

vP

~ BECOME

AP

~ RESULT

VoiceP

~

Voice

VP

The Non-active morphology is associated with Voice. What the two classes have in common is that they both lack agentivity features. The distribution of adverbials provides evidence that this analysis is correct. Manner adverbs are well-formed with verbs with the structure in (3Ib), while they are ruled out with verbs having the structure in (3Ia), which lacks Voice and, by hypothesis, manner: (32)

a.

to pani skistike prosektika the cloth tore.Nact carefully b. *to pukamiso asprise prosektika the shirt whitened.Act carefully

Agentive features are not present in either of the structures in (31), as is evidenced by the fact that the presence of agent oriented adverbs such as deliberately is impossible; see (33). 15 Kratzer (1993, 2000) algues on the basis of German adjectival passives that in German RESULT cannot combine with Voice. However, this is apparently open to parametric variation. In Greek, crucially, RESULT does combine with Voice in adjectival passives, as argued in Anagnostopoulou (2001).

Artemis Alexiadou and Elena Anagnostopoulou

132

(33)

a. *To pani skistike epitides/ me prothesi. the cloth tore.Nact deliberately/on purpose b. *To pani asp rise epitides/ me prothesi. the cloth whitened.Nact deliberately/on purpose

Moreover, note that the presence of an agentive 'by' -phrase is impossible in (32a)as shown in (34)-which provides evidence for a separation of manner adverbs and agentive features: (34)

*To pani skistike prosektika apo ti Maria. the cloth tore.Nact carefully from the Mary

The existence in Greek of a Voice head with manner features independently of agentive features is found in another empirical domain as well. As argued in detail in Alexiadou (2001), Greek event nominalizations have exactly the same internal malceup-they contain a voice head with manner features but they lack agentive features. Once again, adverbs discriminate the two types of features:

(35)

a.

I katastrofi ton arhion prosektika. the destruction the documents.GEN carefully b. *1 katastrofi ton arhion epitides the destruction the documents.GEN deliberately

In conclusion, the following structures express the similarities and differences between transitives, passives, and unaccusatives with these verbs:

(36)

a.

b.

Transitive voiceP

VoiceP

~

DP

/~

Voice'

~

Voice +manner, +agent c.

VP

Unaccusative vP

~ RESULT

Passive

VoiceP

I

Voice'

~

Voice +manner, -agent

VP

Voice +manner, +agent

VP

voice Morphology

133

Note that in this analysis, anticausatives are more complex than causatives and passives. On the other hand, we saw in (28) and (30) that in the case of deadjectival verbs, the causative is more complex than the anticausative. I6 This concludes the structural analysis of Non-active wrinkle verbs, which differ from deadjectival verbs in including a verbal layer. In the next section we turn to the Active Wrinkle verbs.

4.6.3. Active wrinkle-verbs Recall the major difference between Active and Non-active wrinkle verbs, namely that the Active form asserts partial change: (37)

To pukamiso tsalakose s'ena simio alIa den tsalakothike/*tsalakose the shirt wrinkled.Act in one spot but NEG wrinkled.NactiAct endelos. completely

If we were to represent Active wrinkle-verbs as having an adjectival component (as assumed in Dowty 1979 for all anticausatives, despite the lack of evidence for many of them), we would not be able to accommodate the incomplete change entailments they bear. To capture the partial change interpretation of Active-wrinlde verbs, we propose that they have a part-whole or possessive substructure as a component in their representation, as illustrated in (38). The shirt has a wrinlde somewhere.!There is a wrinlde somewhere in the shirt.

(38)

But this cannot be the whole story, as they are not stative predicates. Unlike statives, for example, they have a habitual interpretation in the simple present:

a.

(39)

b.

0 Giannis agapai ti Maria (#sixna/efkola). the Giannis loves the Mary (#often/easily) To pukamiso tsalakoni sixna/efkola. the shirt wrinldes.Act often/ easily

Moreover, they are incompatible with the progressive behaving like achievements: (40)

To pukamiso *tsalakoni [no progressive interpretation]/ the shirt wrinldes.Act tsalalwnete [progressive interpretation] wrinkles.Nact

16 Given our proposal in section 4.6.1. that CAUSE may embed BECOME/RESULT (cf. (30)), and given our claim that BECOME/RESULT may embed VoiceP, we predict that the combination CAUSE + BECOME/ RESULT + Voice is possible. Potentially, this structure is instantiated in Greek through the periphrastic causative construction in (il, under a monoclausal analysis of such structures (Cinque 1999; Wurmbrandt 1998).

(i)

ta nea ekanan to Jani na stenahorithi. the news made the John-Acc SUB] worry-Nact

Artemis Alexiadou and Elena Artagnostopoulou

134

The two components in the meaning of Active wrinkle are represented in (41):

(41)

vP

~

XP

BECOME

~

the shirt

~ HAVE

a wrinkle

Here v BECOME embeds a possessive construction. As with deadjectival verbs there is no source for voice morphology in (41) and thus no source for manner modification. '7 (42)

a.

To hirio gremistike methodika. the building demolished.Nact meticulously b. *To hirio gremise methodika. the building demolished.Act meticulously

In (41) a nominal is involved in the formation of the construction. And indeed nouns are systematically related to the Active-wrinlde verbs, e.g. leroma ('a dirty spot'), tsalakoma ('a wrinke'), gremisma ('demolition), etc. '8 Note, finally, that the two components, BECOME and HAVE, are expressed in the possessive 'get' in ?he shirt got a wrinkle, for instance. If it is correct that the 'maximalaffectedness' reading in English is expressed in the get passive (Klaiman 1991), we could relate the alternation of Active and Non-active in Greek Class IV to the alternation between ?he shirt got a wrinkle and ?he shirt got wrinkled in English.

4.6.4. Summary We proposed that anticausatives generally have the structure in (43): (43)

vP

~ BECOME/

XP

RESULT

17 These two groups-the one with active and the one with Non-active morphology-behave differently with respect to the tests diagnosing internal argumenthood (see the Appendix). They give worse results with the active form, though the contrast is not very sharp. This is expected if the two classes do not receive a uniform structural analysis, and if active wrinkle verbs do not actually have internal arguments. 18 The nouns involved belong to the malmo group. As Kolliakou (r995) and Alexiadou (20or) noted, these nouns have different aspectual properties from other derived nominals. In principle the structure of these nominals could be further decomposed so that in the representation in (4r) we could further analyse 'a wrinkle' according to Alexiadou (2oor)-but we will not go into that here.

Voice Morphology

135

Depending on the type of XP embedded under BECOME/RESULT, there are at least three types of anticausatives in Greek: 0) BECOME + predicate (deadjectival); (ii) RESULT + VoiceP (wrinlde-Non-active); and (iii) BECOME + possessive construction (wrinkle-active). Non-active voice morphology is associated with Voice embedded under RESULT, when this is present. Thus, Non-active morphology is linked to Voice yielding 0) passives; (ii) reflexives; and (iii) adjectival passives. It is sensitive to two properties unifYing these three constructions, namely, absence of a specifier (Embick 1998, in this volume) and manner (see also Zombolou, in progress).

ApPENDIX Interestingly, the Active and Non-active forms of 'wrinlde' verbs show a distinct behaviour in tests diagnosing internal argumenthood of the single argument, though the effect is not always very strong (Alexiadou and Anagnostopoulou 1997):

Possessor sub-extraction (1)

a.

(2)

a. ?*Tinos gremise to letirio? whose collapsed the building.NoM b. ?Tinos gremistilee to letirio? whose collapsed.Nact the building. NOM

Tinos diavases to vivlio? whose read.2sG the book ace 'Whose boole did you read?' b. Tinos efige to pedi? whose left the child 'Whose child left?' c. *Tinos etrekse to pedi? the child whose run

Bare plurals in post-verbalposition (3)

a.

(4)

a. *Gremisan letiria. collapsed buildings b. ?Gremistikan letiria. collapsed.Nact buildings

I Maria tiganise psaria. Mary.NoM fried fish.pL b. Irthan pedia. came children c. *Gelasan pedia. laughed children

Possessor clitics. Borer and Grodzinslcy (1986) argue that possessive datives can be used as an unaccusativity diagnostic because they can only be construed with a DP in a properly governed position (Le. head-complement relation). The same test can be used for Greek.

Artemis Alexiadou and Elena Anagnostopoulou

136 (5)

a.

(6)

a. *Mu lerose i bluza. my dirtied the shirt. NOM b. ?Mu lerothike i bluza. my dirtied.Nact the shirt. NOM 'My shirt got dirtied.'

Mu skotose to pedi. my killed.3SG the child.Acc 'He killed my child.' b. Mu efige to pedi. my left the child. NOM 'My child left.' c. *Mu etrekse to pedi. my ran the child.NoM

This is reflected in the structures proposed.

5 Unaccusative Syntax and Verbal Alternations David Embick

5.r.

INTRODUCTION

Syntactic and lexico-semantic properties are both implicated in the study of unaccusativity. Accordingly, a persistent question in the discussion of unaccusativity as a theoretical notion has been whether or not a reduction is possible in either direction. Typically, the question has been whether the verbs that behave syntactically as unaccusatives do so as a result of their lexical semantics. In broader terms, this is part of the larger question of whether or not the syntactic behaviour of a verb is (uniquely) determinable from its semantic properties. The association of arguments of a predicate to syntactic positions is referred to as the question of 'Linking' in theories of the interface between syntax and the lexicon; Pesetsky (1995) and Levin and Rappaport Hovav (1995) provide overviews of this and related issues. Similar questions apply to verbs that enter into verbal alternations, concerning whether the verbs that do and those that do not enter certain alternations are distinguishable by semantic criteria. The focus in this chapter is not on these types of lexico-semantic question, but is instead For helpful discussion I would like to thank Artemis Alexiadou, Elena Anagnostopoulou, Karlos Arregi, Rajesh Bhatt, Rolf Noyer, and Alexander Williams. I would also like to thank the audience at the ZAS/ LOT Workshop on Unaccusativity for comments.

David Embick on the syntactic structures and syntactico-semantic features that are involved in unaccusative syntax. 'Unaccusative syntax' is understood here in a structural sense as referring to cases in which an external argument is not projected. In the sense I intend, unaccusative syntax is found both in unaccusatives in the standard sense, as well as in passives, which are syntactically intransitive in lacking an external argument, but nevertheless agentive. The structural factor uniting these contexts, the absence of an external argument, underlies a number of cross-linguistically common syncretisms-that is, cases of identical morphological realization in distinct syntactico-semantic contexts. Syncretisms of this type, in which disparate syntactic constructions show 'the same' or similar morphology, are crucial to the understanding of the manner in which syntax and morphology relate to each other and to other parts of the grammar. Much of the chapter is devoted to showing the role that unaccusative syntax plays in defining such syncretisms as those mentioned. In section 5.2, I will discuss the importance of the unaccusative analysis of reflexives in the analysis of patterns such as that alluded to above. This aspect of the chapter will also involve a revision to that analysis for certain systems. Second, in section 5.3 I will discuss the nature of the morphological syncretism that centres on unaccusative syntax, and show that it arises by morphology being sensitive to the absence of an external argument. In section 5.4 I discuss alternatives to the analysis presented in sections 5.2 and 5.3. One alternative considered is that the absence of an external argument is encoded in a syntactico-semantic feature. The second alternative is based on a recent critique of the unaccusative analysis of reflexives, which holds that reflexives have, in effect, the syntax of unergatives. Finally, section 5.5 concludes. Some background is required before the discussion proceeds. One set of assumptions I will mal{e concerns the relationship between syntax, morphology, and the lexicon, and stems from work in Distributed Morphology (Halle and Marantz 1993,1994, and related work). Particularly relevant is the architecture provided by this theory, in which morphology interprets the output of the syntactic derivation by (among other things) adding phonological content to positions in a hierarchical structure:

(I)

The Grammar

Syntactic derivation

.. (Spell Out)

IMO'PhOI"~ PF

LF

The syntax manipulates terminal nodes containing abstract features, at least for functional heads. There is a difference in the theory between functional heads on the one hand and the members of the open-class vocabulary on the other. The latter are

Unaccusative Syntax and Verbal Alternations

139

referred to as 'Roots' (in the notation ",-,ROOT), and are category-neutral. The abstract features present on the functional heads in the syntax are spelled out with phonological content in the process of Vocabulary Insertion, which tal<:es place in morphology. Individual Vocabulary Items consist of a phonological exponent and the features that it is associated with. Thus, for example, the English plural /-z/ is associated with the following Vocabulary Item: -z H [PL]. Vocabulary Items are inserted such that the item with the greatest subset of the features on a node will win out over its competitors. As a result of this, exponents, the morpho phonological objects that are inserted into these nodes, may be underspecified with respect to the morphosyntactic context in which they appear. This type of underspecification will figure prominently in the analysis presented in section 5.4. This grammatical architecture forces a particular approach to the study of verbal alternations. There is no extra-syntactic lexicon in which word-formation of any type, or in particular the derivation of one verbal class from another, can take place. I What there is to say about verbal alternations is essentially syntactic, and consists in identifYing the structures and features underlying particular alternations. In this way there are clear connections with the Hale and Keyser approach to argument structure (Hale and Keyser 1993 and related work), at least to the extent that the structures proposed in that framework are actually part of the syntax and not some other component. Every theoretical framework has to list certain types of unpredictable information, whether the special meaning found with kick the bucket, or the basic sound-meaning connections found in a Root such as ",-,DOG. In the framework assumed here, there is a further component of the grammar, the Encyclopedia, in which special meanings of the type found with idioms, light-verb constructions, and, for that matter, simple Roots are listed. Certain aspects of what is sometimes called lexical semantics are therefore stored in this list. Among other things, the fact that certain verbs enter transitivity alternations (?he vase broke and John broke the vase) while others do not (?he books arrived and *John arrived the books) implicates Encyclopedic knowledge-that is, is related to the semantic differences between the two roots. 2

5.2.

REFLEXIVIZATION AND

UNACCUSATIVE SYNTAX The idea that unaccusatives and passives do not have external arguments is a familiar one. Here I will review some further assumptions concerning (I) the licensing of external arguments, and (2) the unaccusative analysis of reflexives. Based ultimately on arguments that the external argument is not an argument of the verb per se (c£ • This position will be articulated in greater detail in sect. 5.2. 2 See also Borer (in this volume) and van Hout (in this volume) for related perspectives.

David Embick Marantz I9 84) , Kratzer (I994, I996) proposes that such arguments are the specifiers of a functional head that takes the VP as an argument. Although the connections here are somewhat tenuous, the head in question is sometimes identified with the v ('small v') employed in Chomsky (2000) and subsequent work (cf. also Hale and Keyser I993); I will assume this identification for convenience here. 3 The structure of a transitive vP is as follows:

(2)

vP

Transitive

~ op v ~ v

--ip

I

~ --iop

[~~J

On this implementation, [AG] is a semantic feature with the properties proposed by Kratzer. Thus, the v head has interpretable (and uninterpretable) feature content. Returning to the connection between passives, reflexives, and unaccusatives, I will assume the structure in (3) for the vP in a passive. Like a transitive, this v contains the feature [AG], which is responsible for the agentive interpretation of the passive; the external argument op and the Case feature are absent, however (see below).4 (3)

vP

Passive

~

v

--ip

I~ [AG]

--iop

The structure of unaccusatives is similar; the primary difference is that these will lack the feature [AG], and will therefore not be interpreted agentively. Reflexives differ in terms of how exactly they manifest unaccusative syntax. As noted above, a pattern found in a number of untelated languages involves the appearance of identical morl Naturally the possibility always exists that there are several heads in the VP broadly construed, v, Voice, etc. 4 In the passive, the idea is that the Agent argument, which is licensed by the feature [AG] on v, does not appear; Kratzer gives [AG] the following semantics:

(i)

AG* = Ax,Ae,[Agent(x)(e)]

A further conjunction rule is needed to identifY the event argument here with that of the VP. Kratzer's discussion is centred on the case of actives, in which a DP is projected in the specifier of the head bearing [AG] and is taken by that head as an argument. It seems plausible that the agentive interpretation, when no DP argument is present, is derivative of a process akin to existential closure, although I cannot pursue this here.

Unaccusative Syntax and Verbal Alternations phological marking in the intransitive members of transitivity alternations, certain types of reflexive, and passives.5 Many attempts to capture the similarities between these morphosyntactic contexts have been based on Marantz's (1984) proposal that each of these environments involves the absence of an external argument; ultimately this is derivative of the feature [-logical subject] in that framework. 6 .fu implemented in a series of subsequent analyses (Kayne 1988; Pesetsky 1995; McGinnis 1997, among others) focusing on sE-ditics in Romance, the derivation of reflexives involves the diticization of an anaphoric external argument, with subsequent raising of the object to a position from which it binds the anaphor. That is, the ditie is generated in the specifier of v, as the initial configuration in (4) shows. Reflexive I

Reflexive II

vP

~

CL

vP

~

DP

v

~ v

(

v~ ~~

CL

AG [

1~

t

Case

The external argument then diticizes onto v, satisfYing the Case feature of v such that the internal argument will check Nominative Case. Note that the internal argument is required to raise, in order to bind the diticized anaphor. This requirement makes reflexives of this type unlike unaccusatives, in which there is no such requirement (c£ section 5.5.2).7 There is a further question concerning the nature of the operation that combines CL with the verb; the arrow in (5) is therefore somewhat figurative, indicating that the ditic and the verb will get together at some point.

) These three do not exhaust the full range of configurations that appear in such systems. The primary motivation for the unaccusative analysis of reflexives in Marantz (1984) was syntactic, however. Nevertheless, the connection between this analysis and the analysis of the syncretism is clear. 7 The mechanics of the raising of the internal argument present difficulties within current Minimalist assumptions about movement, particularly if the v in such structures as those in (4) and (5) define strong phases-the most natural assumption, if this is in fact a type of transitive. If the internal argument remains in situ, then this v could not be a phase. The argument would be phase-internal, and could not raise to T subsequently in the derivation, in that it would not appear at the phase boundary. What is needed is for the internal argument to move to [Spec, vl in order for subsequent movement to be possible. But movement of the object to adjoin to vP cannot be for Case reasons if the Case feature of v is checked by the cliticization of the external argument to this head. 6

David Embick The account sketched above is referred to as the 'unaccusative analysis of reflexives', for obvious reasons. Two points about this analysis figure prominently in the discussion to come. The first is about what unifies reflexives and unaccusatives according to this treatment. The two are similar only to the extent that the full (Le. non-elitic) DP originates as an internal argument. In other respects they differ: the reflexive is agentive, while the unaccusative is not; and the reflexive is syntactically transitive, in that in the initial stages of the derivation it has an external argument. So the unaccusative analysis of reflexives holds that reflexives and unaccusatives have some properties in common; not that they are identical. The second point about the unaccusative analysis of reflexives concerns whether or not the elitidzed external argument approach is always appropriate for systems in which reflexives and unaccusatives show some common properties. As I will show below, reflexives in some languages can show unaccusative syntax without showing the exact derivation outlined in (4)-(5). With the idea behind the unaccusative analysis of reflexives at hand, I will now examine the nature of the morphological syncretism often found with passives, reflexives, and unaccusatives. For convenience, I will refer to this as the 'u-syncretism'. The analysis I present is based on the idea that a particular structural property is at the heart of this morphological pattern. In particular, unaccusative syntax, represented abstractly in (6), is the relevant factor:

(6)

Unaccusative syntax

vP

~

v

VP

~

... Verb ... DP Although passives, reflexives, and unaccusatives differ in a number of ways in terms of feature content and other aspects of their derivation, they all have a common subpart, represented in (6). In some languages, morphological realization is sensitive only to this structural property. When this situation occurs, the u-syncretism results.

5. 3 .

ILL US T RAT ION

Modern Greek shows a version of u-syncretism, in which Non-active (Nact in glosses) morphology appears in a number of syntactic environments. The analysis of Nonactive voice in Modern Greek illustrates three primary points about the nature of u-syncretism: (7)

Points from the discussion of Modern Greek Non-active voice appears in the context of unaccusative syntax; and this type of syntactic configuration is found with passives, unaccusatives, and certain reflexives.

I.

Unaccusative Syntax and Verbal Alternations

143

The formation of certain reflexives in Greek shows a type of unaccusative syntax, but this derivation does not involve a cliticized external argument; rather, an adverbial element is prefixed to the verb. 3. Reflexive interpretation depends on verb class-some verbs, the body-action type (wash, etc.), receive reflexive interpretation with the signal of unaccusative syntax alone. Other verbs require something in addition. 2.

Modern Greek verbs inflect for two voices morphologically-Active and Non-active. Non-active voice appears in three basic types of alternation. 8 To begin with, passives (8a) and the intransitive versions of some verbs appearing in transitivity alternations (8c) show this form: 9

a.

(8)

b.

c.

0 Yanis katastrafike. the Yanis destroy.Nact.3sG 'Yanis was destroyed.' tsalcizo breakAct 'break-Transitive' tsalcizome breakNact 'break-Intransitive'

Not all verbs that enter transitivity alternations show Non-active voice in the intransitive form. Such patterns are quite common cross-linguistically, a point which warrants an excursus. The default hypothesis given the theory of syntax-morphology interactions assumed here is that the morphological difference should directly correlate with a structural or featural difference in the verb classes showing morphological differences (Alexiadou and Anagnostopoulou, in this volume, develop the latter type of approach). An analysis of this type is found in Hale and Keyser's (1998) discussion of two different types of alternating verb, one basically transitive, the other, basically intransitive (c£ also Alexiadou and Anagnostopoulou 1999). To tal(e a pertinent example, Hale and Keyser propose the structures in (9) and (10) for derivations based on 'adjectival' Roots, along the lines of the English verb redden (Hale and Keyser's V appears as v). (9)

Intransitive vP

~

DP

v

~

v

"ROOT

In addition, Non-active voice appears invariably with a small class of verbs, the deponents; some such verbs show Non-active form in spite of being in what appears to be transitive, active, syntax (see Embick 8

1998). 9

See Alexiadou and Anagnostopoulou (in this volume) for the transitivity alternations.

David Embick

144

(10)

Transitive

vP

~

v

DP

~

v

vP

~

DP

v

~ v

"ROOT

There is an important question at this point as to whether something like (9) is unaccusative or not. In the structural sense in which I am employing the term here, it mayor may not be. On the one hand, the v head has an argument in its specifier. On the other hand, the DP is a sister ofwhat appears to bea type of complex predicate, and so in that sense might not be truly external. In yet another sense, however, ifwe define 'unaccusative' as 'non-agentive intransitive', it would count as unaccusative, with a non-agentive v. A further possibility involves reference to the features on v, with modifications to the Hale and Keyser analysis; see Alexiadou and Anagnostopoulou (in this volume). Whether or not one adopts Hale and Keyser's structures, it is a distinct possibility that some of the difficulty in identifying clear patterns of unaccusative behaviour stems from the existence of the two distinct notions of unaccusativity discussed in this paragraph. Different phenomena and different languages might be sensitive to either the 'No External Argument' or the 'No Agentive v' properties. Differences in (among other things) morphological realization in transitivity alternations could be captured in these terms, in accordance with the strongest hypothesis about the interface. If this is unworkable, other options, such as contextu<Jl allomorphy, are available. Returning to Modern Greek, two types of reflexive also appear with Non-active voice. First, certain verbs are interpreted as reflexive when they appear with the Nonactive voice:IO (II)

I Maria xtenizete kathe mera. the. NOM Maria.NOM comb.NacqsG every day 'Maria combs herself every day.'

In addition to the type illustrated in (II), there is also a second type, with the verb prefixed with afto- 'self' and Non-active voice:

(12)

0 Yanis afto-katastrafike. the Yani self-destroy.Nact.3SG 'Yani destroyed himself'

IO In addition to verbs dealing with various body actions, such as 'wash', there are other verbs capable of receiving this interpretation as well. See Manney (2000) for examples.

Unaccusative Syntax and Verbal Alternations

145

What is of note in this type is the connection between afto- and the Non-active morphology: both are required for a reflexive interpretation. If the former is absent only a passive interpretation is possible, and if the latter is absent the sentence is ungrammatical. l l These two types of reflexive contrast with reflexives formed with an actual anaphor, in which the Active form appears:

(13)

VIepo ton eafto mu. see-Act.lsG the self my 'I see myself.'

In the case of reflexives with afto-, it seems clear that this element effects reflexivization. The question that remains concerns how exactly this talces place. Given what we have seen above in the discussion of Romance sE-clitics, one possibility is that aftois the realization of an external argument that has cliticized onto the verb. A second possibility is that afto- is in effect adverbial. Modern Greek shows what is classified as 'adverb incorporation' more generally, with the adverbial elements being realized prefixally on the verb." Whether or not the adverb has actually incorporated, or is present in a type of compounding (which I consider more likely), the point is that if this option is correct, the derivation is unlike what happens in the Romance case. The afto- prefix appears in nominalizations like afto-katastrofi ('self-destruction'), and this is a type of nominalization in which an Agent is not licensed (cf. Alexiadou 2001). If the afto- prefix realized a clitic external argument, i.e. an Agent, it could not appear in nominalizations of this type. The fact that such nominalizations are possible shows that afto- is not the external argument. Consider now the following structure, in which afto is shown attached to the verbal Root:

(12)

Afto-reflexive

v

I

AG

DP

afto In terms of the presence of a v that licenses agentive interpretation, the structure is like that of a passive. However, the main verb in this case has the interpretation 'selfV'. There is a single argument, the D P complement of the Root. In this structure, the

n Putative exceptions show afto- with active morphology; however, these do not receive a reflexive interpretation. See Rivero (1992) and Embick (1998). U Cf. Rivero (1992) for adverb incorporation, although Rivero does not propose this type of derivation for afto- prefixed verbs. Rivero's proposal is that afto is an incorporated direct object. The reason for the systematic appearance of Non-active voice in these cases would not be attributable to the absence of an external argument if this were the case, and her proposal will not be considered further.

David Embick object is the object of a self-V action; that is,fohn was self-destroyed. In this sense, the sole DP argument comes to be interpreted as agentive in a derivative fashion-that is, without having originated as the specifier of vP. With the body-action type verbs like the example in (II), something further must be said. The most consistent would be that the reflexive interpretation is possible because of the Encyclopedic semantics of the verb-that is, the self-related component is part of the verb's meaning. It is tempting to form an analogy with such English examples as John got dressed, where the interpretation can be agentive in spite of the apparently passive syntax, something which does not happen for other verbs (c£ John got arrested). The case of afto-prefixation indicates a great deal about the nature of u-syncretism. The actual reflexivization that occurs is brought about through afto-. Non-active voice appears in addition. It is clear from what we have seen above that Non-active morphology does not actively reflexivize anything. For one, it is simply not required to do so in the afto- case. It is, moreover, clearly not reflexivizing the verb in the unaccusatives and the passives we have seen above. Foreshadowing the discussion of the next section, the generalization is instead that the Non-active voice simply appears when there is no external argument; that is, it signals unaccusative syntax. How precisely it does so will be addressed in detail in the next section. In the rest of this section I illustrate the points raised in the discussion of Greek in two additional languages. Fula (West Atlantic, Niger-Congo) has a three-way voice system, with endings for what are labelled active, middle, and passive voice. I ) The forms referred to as middle occur in the intransitive variants of certain verbs which are transitive in Active form, and indicate an action that occurs without reference to an external agent (Arnott 1970: 25 6):'4

(15)

a.

b.

'O-besd-ii sheede. 3SG-increase-AcT price 'He increased the price.' Sheede besd-ake. price increase-MID 'The price increased.'

Within a class of verbs pertaining to bodily action, the middle form has a reflexive interpretation; thus for the verb wash, the Active, Middle, and Passive are as follows: (16)

a.

b.

'O-loot-ii biyiko. 3sG-wash-AcT child 'She washed her child.' 'O-loot-ake. 3SG-wash-MID

I) In the following discussion I draw on the work of Arnott (1956, 1970), as well as the discussion of Klaiman (1991, 1992). 14 It also appears in certain statives. According to Arnott 'the emphasis is on the state and not the means'.

Unaccusative Syntax and Verbal Alternatiom

c.

147

'She washed herself' 'O-loot-aama. 3SG-wash-PAsS 'She was washed.'

Outside of this class of verbs, it is not possible to form reflexives by simply applying middle-voice endings to transitive verbs; instead, the reflexive suffix +/ -it- appears on the verb, along with the middle voice:

(17)

a.

b.

wara kill.ACT 'kill' war-t-o kill- REFL-MID 'commit suicide'

Contrasting verbs of this type with the type illustrated in (16), Arnott (1970: 342) notes that these refer to 'actions which it is unusual for a Fulani to perform on himself.' As examples Arnott gives fimmba ('shave'), moora ('dress hair'), ndaara ('look at'), ta'ya ('cut') and others. '5 The reflexive interpretation is possible for these verbs only when they have both -t/ -it and the middle voice. '6 The pattern is thus as follows. For a specific set of Roots, a reflexive interpretation is possible with the middle-voice morphology, which signals unaccusative syntax, alone; this is parallel to the wash-type verbs in Greek above, subject to differences in the identity of the actual verbs. With other verbs, a specific reflexive affix is required. The middlevoice morphology appears in these reflexives as well, signalling unaccusative syntax. In Fula, it is quite possible that the Romance-type analysis, with a didcized external argument, is the appropriate one for the reflexive affix. What is important for our purposes is that it is the reflexive affix that is actually responsible for reflexivization. The middlevoice morphology merely functions to signal unaccusative syntax, which is also present in such reflexives. The connection to Greelc reflexivization above is thus quite dear. Further evidence for an adverbial analysis of certain reflexivizing elements, as in Greek afto- above, can be found as well. I illustrate this in Tolkapaya (a dialect of Yavapai, a Yuman language). The discussion here is based on the treatment of Munro (1996).'7 Tolkapaya shows a signal -v which is sensitive to the absence of an external 15 With shave-type verbs it is possible to form middle-voice forms without the reflexive suffix -tl-it; in such cases, however, the interpretation is not reflexive (Arnott labels it 'causative reflexive'):

(i)

a. Mi-femmbii-mo. 1SG-Shave.AcT-3SG 'I shaved him.' (active) b. Mi-femmb-ake. 1SG-shave-MID 'I got myself shaved.'

16 According to Arnott (1956: 134), the reciprocal suffix appears only with active endings. I will not investigate the reasons behind this difference here. 17 As are the transcriptions. I would like to thank Pamela Munro for clarifYing a number of points con-

David Embick

argument. Some verbs are interpreted reflexively or as stative passives when suffixed with -v-, glossed here as 'MINUS' (for 'minus an argument') in accordance with Munro's practice:'8

(18)

Hamany-che chthUl-v-i. child-NOM wash-MINUS-ABS 'The child is washed/The child washes himsel£'

In addition to the interpretations given above, the same form with -v can also be interpreted as 'washable'. Munro classifies the standard interpretations of verbs with -vas reflexive, passive, and middle-that is, configurations in which no external argument is projected. The example above shows a reflexive interpretation with the verb wash. With other verbs, a reflexive interpretation is not possible with the -v suffix alone, but is found if the element yeem, glossed here as 'self', is present. When yeem appears without -v the interpretation is adverbial, meaning 'all alone', 'all by oneself'. This type of verbdependent difference is illustrated in the following two verbs, for 'cut hair' and 'cut', from Munro (1996). The former is interpreted reflexively with -v- alone, whereas the latter has a passive (stative) interpretation with the -v-suffix:

(19)

chiir-i 'cut (someone's) hair' chiir-v-i 'cut one's hair'

(20)

chkyat-i 'to cut' chkyat-v-i 'to be cut' yeem chkyat-v-i 'cut oneself'

Munro notes that in general the adverbial yeem forces the reflexive interpretation; otherwise it mayor may not be available, depending on the verb involved, apparently.19 In this section I have examined three languages in which Reflexivization interacts with or involves the absence of an external argument, and in which a morphological syncretism between reflexives and other syntactic configurations results. The casestudies emphasize the point that in such systems, verbs with reflexive interpretation do not allow for a uniform syntactic analysis cross-linguistically, and in some cases differ structurally within the same language. Even languages which show the familiar syncretism of reflexive with unaccusative and Passive show different properties in the reflexive system when it is examined in greater detail. 20 In particular, while the analysis with cerning the Tolkapaya facts. The analysis I present is in some sense an instantiation of Munto's idea that Tolkapaya -v is involved in valency reduction. 18 In other cases, the range of interpretations of -v- suffixed verbs thus varies somewhat idiosyncratically. Munro notes such cases as s~hlok-i ('to take by the handle'), s~hlok-v-i ('to have a curved handle'), in which the semantic relationship is rather indirect. 19 Munro notes that yeem may be optionally present with verbs such as wash in (16). 20 It should be pointed out that there is no necessary connection between this type of reflexivization and the type of voice marking we have been examining. In addition, there are languages that mark mor-

Unaccusative Syntax and Verbal Alternations

149

a cliticized external argument (cf. (4) and (5)) seems to hold for the Romance type of reflexivization and its unaccusative properties, a further type has been identified here in which there is unaccusative syntax and an 'adverbial' type of reflexivization. In addition, there is a question concerning the nature of the 'special' class of verbs that lend themselves to reflexive interpretation with the morphological marker of unaccusativity alone. This could be because of the special Encyclopedic semantics of these verbs, an implementation of something which is often referred to as 'inherent reflexivity'. Up to this point, the analysis has concentrated on showing the relevance of the absence of an external argument to the study of certain verbal alternations. I will now turn to the question of how morphological realization relates to the external argument property.

5·4·

IMPLEMENTATION

The question raised by the u-syncretism is as follows: how is it that passives, reflexives, and unaccusatives can have the same morphology, while at the same time being semantically and syntactically distinct? As we have seen, these configurations have a subptoperty in common, in that they all lack external arguments. The analysis I propose makes the morphological patterns a reflex of this common structural property. It holds that for the purposes of morphological realization, the absence of an external argument plays a defining role, so that in effect other differences between passives, reflexives, and unaccusatives are ignored by the morphology. Some background is required in order to understand the nature of this approach to syncretism. To begin with, morphological elements may be underspecified with respect to the syntactic environment in which they appear. 'Underspecification' in this sense refers to the properties of phonological exponents with respect to syntacticosemantic environments in which these exponents will be inserted. To take a simple example, consider the suffixes found on Romanian adjectives (the discussion here is based on Noyer 1998):

(21)

Romanian adjectival endings Singular Plural Masculine -0 -i Neuter -0 -e -e Feminine -a

phologically the absence of external arguments, while also marldng reflexives with active voice. Creek, a Muskogean language, shows a suffix -k V; which appears in the intransitive members of transitivi ty alternations and certain passive-like forms (cE Martin 1991; Hardy 1994). Once again these are environments in which external arguments are absent, as in Martin's analysis. Reflexives, however, are formed with a reflexive prefix, and show active morphology-that is, no -k V suffix. Cases of this type are of interest from the perspective of the approaches of Marantz (1984) and McGinnis (1997), which hold that all 'elitic' reflexive systems involve unaccusative syntax.

David Embick

15 0

While the exponents -Ii and -i appear in only a single environment, each of the other two suffixes, -0 and -e, appear in more than one position. The theory assumes that the morphosyntactic positions in which Vocabulary Insertion is taking place are fully specified. So, concentrating on the plural for exposition, there are three distinct feature bundles to consider prior to insertion:

(22)

Feature bundles

a.

[+Masc 1 +PI

b.

[+Neut 1 +PI

c.

[+Fem +PI

1

Consider now the vocabulary items in (23).

(23)

[+masc +pl] H -i [+pl] H-e

These rules are ordered by the familiar principle according to which the most specific rule will take precedence over one that is less specified. Applied to the three feature bundles in (22), the desired results are captured. For (22a), the conditions for the insertion of -i are met; in each of (22b, c), the most specific rule that can apply inserts -e. In this example, -e is underspecified with respect to the morphosyntactic environment it is inserted into, so it appears in both feminine and neuter plurals. However, it is the same -e that appears in each case. The underspecification of phonology with respect to morphosyntax or syntax/semantics in this manner allows for the pervasive patterns of syncretism found in natural language to be captured systematically, a point which has figured prominently in the critique ofLexicalist approaches to morphology. Returning to the u-syncretism, the syntax generates passives, reflexives, and unaccusatives, each fully specified for the distinct features that make these distinct verbal constructions. The functional head v is overtly realized in a number oflanguages. The most direct statement of the u-syncretism thus involves saying that this pattern results from the realization of v in a particular structural environment. Using -X to refer to the feature or signal associated with Non-active type morphology, the representation of this is as follows: (24)

v H v-X/_No external argument

The spell-out here is stated somewhat abstractly, in terms of the symbol -X. This is necessary in order to capture further properties of morphology that determine how the morphological sensitivity to the absence of an external argument is manifested. In Modern Greek, forinstance, the -X here would be an abstract feature [NonAct], which underlies the realization of entire sets of Non-active agreement endings; see Embick

Unaccusative Syntax and Verbal Alternations (1998) for discussion. 2I In Tolkapaya, however, there is a single exponent, -v, associated with the absence of an external argument. In that language, then, (24) will result in the spell-out of this actual exponent. By underspeci£}ring -x, so that it is sensitive to only the absence of an external argument, we have an answer to why passives, reflexives, and unaccusatives can have the same morphological realization while still being syntactico-semantically distinct. Syntactico-semantically they are distinct; but they have a common structural property which is directly relevant to morphological realization. The underspecification of morphology with respect to syntax in this way is one of the defining features of Distributed Morphology, and realizational approaches to morphology more generally. Lexicalist theories are incapable of capturing syncretism without appealing to accidental homophony, a type of analysis that clearly misses the relevant generalizations. 22 While the u-syncretism is centred on a syntactico-semantic object and its features, there is no syntactico-semantic feature [-external argument]; rather, the immediate environment of v determines morphological realization. 23 The apparent effects of a [-external argument] feature are reduced to a syntactic head subject to contextual allomorphy or feature assignment in morphology; and the head and these processes are each independently required. In the next section I examine alternatives to the analysis I have provided. Before doing so, I would like to stress a few points concerning the approach I have adopted here. I am not assuming that all languages show some version of the u-syncretism in the way I have discussed it here. Neither am I assuming that all languages show some aspects of unaccusative syntax in the derivation of reflexives. The focus has instead been on the nature of the u-syncretism in the languages that show it, as a means of exploring properties of the syntax-morphology interface. Further cross-linguistic questions, concerning the fact that notional reflexives seem to be associated with a number of distinct syntactic configurations, will be touched on in the conclusion.

5.5.

ALTERNATIVES

There are two possible alternatives to the analysis presented above. One alternative is that there is actually a feature [-external argument] present in the syntax; this is addressed in section 5.5.1. The second alternative is that the unaccusative analysis of reflexives, which figures crucially in the analysis above, is incorrect. This position has been taken by Reinhart (1997) and Reinhart and Siloni (in this volume) and is the topic of section 5.5.2.

21 It might also be the case that it is a v head with certain features that is subject to (22); see Alexiadou and Anagnostopoulou (in this volume). 22 For the general point about lexicalist problems with syncretism there is an ample literature; Noyer (1997, 1998) provides an overview. See Embick (1998, 2000) for further discussion of syncretism in voice morphology. 2) This position is also explored in Lidz (1998).

David Embick

5.5.1. A [-external argument] feature in syntax? In determining whether there are grounds for positing a [-external argument] feature in the syntax, two questions must be asked: (1) is a [-external argument] feature necessary for syntactic reasons? and (2) is a [-external argument] feature necessary for semantic reasons? The distinction here follows one which is based on the division of feature-types by Chomsky (1995). The answer to the first question is that the presence or absence of external arguments is, in syntactic terms, derivative of the workings of other features. There is thus no syntactic reason to posit such a feature. If the second question were answered affirmatively, the implication would be that a [-external argument] feature is required for interpretive properties found in each of the cases found in typical u-syncretisms. There is no reason to posit such a feature, as what needs to be said about the interpretation of passives, reflexives, and unaccusatives is derivative of other factors. In the rest of this subsection I expand each of these points. In recent syntactic discussions the status of external arguments revolves around relationships between Case and other properties of the clause. Specifically, v without an external argument does not have a Case feature; this is an element ofBurzio's Generalization-no external 8-role, no accusative case. But these effects are derivative. Consider the following cases, on the assumption that an internal argument is present (derived from Chomsky 2000):24 with Case on v and no external argument, the Case feature ofT will not be checked; with no Case, but with an external argument present, the internal argument will not have Case. Something has to be said about unergative verbs as well; the simplest is that the biconditional External Argument H Case does not hold. In any case, there is no reason to posit a feature [-external argument] on syntactic grounds, given that any effects this feature might have are subsumed by other derivational mechanisms. An additional possibility is that the no external argument property is encoded as an interpretable feature. This would be in addition to such interpretable features as [AG], and perhaps features relating to eventualities and to properties such as 'Inchoativity', associated with v heads. In light of the common syncretisms analysed earlier, this type of treatment would necessarily hold that there is a common interpretive component to passives, reflexives, unaccusatives, and so on. The problem lies in specifYing what this feature would be; or, what it would be required for in the first place. There do not seem to be any obvious interpretive connections between passives, unaccusatives, and reflexives beyond their structural similarities. For example, it seems highly unlikely, given an analysis of the passive which has an Agentive v, no external argument, and so forth, that another interpretable feature is required to indicate that there is no external argument. Given this, there is no reason to posit such a feature, unless its existence is absolutely forced; see Embick (2000) for additional discussion.

'4

With Case and External Argument, we have a normal transitive; without either, an unaccusative.

Unaccusative Syntax and Verbal Alternations

153

5.5.2. A (partially) lexical alternative I now turn to a discussion of a Lexical treatment of the reflexive-unaccusative syncretism, found in Reinhart (1997, 2000) and Reinhart and Siloni (in this volume). Before addressing the specific details of these proposals, I will first provide a background discussion of how the lexicon is viewed from the perspective of Distributed Morphology. The interest in doing so is not merely historical. The motivation behind the architecture assumed in Distributed Morphology stems from a series of arguments showing that a generative lexicon of the familiar type is not justified. In assessing a particular lexicalist account, the first question to be asked is whether or not a mode of composition separate from that provided by the syntax is necessary. The lexicon, as typically defined, serves (at least) two roles. First, it is the module of the grammar where a certain type of combination of elements takes place-specifically, the lexicon is the place where words are stored and derived. This is the generative aspect of the lexicon: it generates words, while the syntax generates structures in which words are combined into phrases, and so forth. The second role played by the lexicon is as a component in which arbitrary or unpredictable information is simply listed; this is the lexicon in the Bloomfieldian sense. In justif}ring the existence of such a module, the focus must be on the former aspect, and not the latter. Any theory can (and in fact must) simply encode or list unpredictable information somewhere. The real questions concern the generative aspect of the lexicon, and are questions about modularity-whether the processes deriving words are distinct from those that assemble words into phrases. A clear example of the division between lexical and syntactic is the discussion of 'adjectival' and 'verbal' passives in Wasow (1977). After identif}ring a number of differences between the two proposed categories, Wasow states his solution in modular terms: Adjectival passives are created in the lexicon, verbal passives in the syntax. The modes of composition for the two differ, and stem from the modular distinction between lexicon and syntax. In a more articulated form, the arguments for modularity cover both the phonological and interpretive domains. In phonology, Lexical Phonological operations were associated with a set of characteristic properties not shared by complex forms created in the syntax (c£ Kiparslcy 1982). On the semantic side, lexical entities were taken to have a distinct property as well: the possibility of 'special' or idiosyncratic meanings. In contrast, objects created in the syntax could not have such meanings. As Marantz (1997) discusses in detail, the modularity arguments for a generative lexicon break down on both the phonological and semantic branches of the grammar. On the phonological side, the breakdown of the modularity argument is found in cases in which objects that have to be assembled syntactically nevertheless show 'lexical' phonology (c£ Hayes 1990 for a statement of the problem from a lexicalist perspective, and Embick 1995 for a case study from a non-Iexicalist point of view). Semantically, 'special meanings' are not found with simplex lexical items alone, but must be associated with objects created in the syntax. In each case, the breakdown of the modularity arguments points to the same conclusion: there is no generative

David Embiek

154

lexicon. 25 In the present case, the burden on a lexicalist approach is to show that the operations required to capture a given alternation simply cannot be syntactic, and that they justifY the existence of a generative system separate from the syntax. Reinhart (1997, 2000) proposes that the similarities between reflexives and unaccusatives are the result of a lexical process, Reduction, which alters the argument structure of basic transitive verbs; the general approach relates directly to that presented in Chierchia (in this volume). The statement of reduction as a general process is as in (25).

en

The R represents a relation-in the derivation of reflexives, a SELF-function. The in the output of the process indicates that either the first or the second of the two original roles may be reduced. In order to derive unaccusatives, the external argument is removed by reduction. In the derivation of reflexives, the claim is that it is the internal argument that is reduced. On this view, unaccusatives have unaccusative syntax, while reflexives do not. The morphological similarity between the two is supposed to stem from the fact that both reflexives and unaccusatives have undergone the same operation. Of course, the operation in the two cases is the same only to the extent that an argument has been removed; if we were to focus on which argument is removed, it would look as if there are two operations at play. The proposal is thus not able to capture the morphological connection between reflexives and unaccusatives straightforwardly; this problem will only multiply as further environments that often pattern in the same way morphologically, such as passives, are taken under consideration. The discussion of these proposals falls into two components. The first is architectural, and addresses the question of whether the behaviour of reflexives and unaccusatives provides evidence for a generative lexicon. The second part addresses an implication of Reinhart's view, which is that reflexives of the relevant type must be unergative. Reinhart and Siloni (in this volume) provide a set of arguments that attempt to support this claim by showing that reflexives and unaccusatives differ with respect to certain diagnostics. The first objection against these proposals is the most general, and the most telling. The assumption that underlies Reinhart's project is that operations relating to reflexives and unaccusatives are lexical-that is, that they take place in a Lexicon in the sense familiar from Lexical Phonology and Morphology and related theories. The proposals are thus only meaningful to the extent that the existence of a lexicon with operations of the relevant type can be justified. In the present context, Reinhart's Reduction rule can only be justified if it can be shown that it requires a type of operation that is statable only in lexical terms. However, Reinhart provides the relevant argument on this point. Noting the formation of reflexives from small clauses (Jean se eroit intelligent), Reinhart argues that the reduction that creates this type of reflexives cannot be lexical, but must be syntactic because it involves elements outside of the verb's lexical argument structure. She then notes that 'the reduction operations ... obey precisely 25

See also Borer (in this volume) for this position.

Unaccusative Syntax and Verbal Alternations

155

the same constraints when they apply in the lexicon and in the syntax' (2000: 15). With no independent reason to suspect that a generative lexicon of the relevant type is necessary, this is precisely the type of anti-modularity argument that makes the lexical treatment unjustifiable. Reinhart acknowledges that a syntactic operation is required in order to handle some of the phenomena that her lexical treatment is intended to handle, and that there is no principled difference between the operation in the two domains. In light of the discussion of the lexicon above, there is therefore no reason to think that unaccusatives and reflexives need to be derived via a lexical operation. That is, there is no lexicon of the relevant type, and therefore no reduction in the lexicon. We are left, then, with the question ofwhether there is evidence for a syntactic operation of Reduction, as discussed in Reinhart and Siloni (in this volume)!6The nature of this operation, when it is syntactic, is as follows. In the derivation of French reflexives, se appears as a sort of intransitivizer-its function is to absorb Case. In the syntax the verb has two a-roles, but only one full DP is licensed, by virtue of the aforementioned Case absorption. The application of LF-Reduction identifies the two a-roles, such that they are both assigned to the single DP' 2? A consequence of the Reduction view is that reflexives are unergatives. Reinhart and Siloni argue that reflexives and unaccusatives do not always pattern together, and that the unergative analysis is therefore correct. The arguments share a common property. The underlying assumption in each case is that in the unaccusative analysis of reflexives, reflexives and unaccusatives are exactly the same. However, this is not what the unaccusative analysis of reflexives asserts. Rather, this analysis holds that (certain) reflexives share a particular property with unaccusatives. Other differences, such as agentivity, and the presence of a diticized external argument in some languages, differentiate the two. Many of the facts that Reinhart and Siloni present are thus compatible with either analysis. 28 I will illustrate with an argument based on facts from Italian, which, in addition to exemplifying the point just mentioned, connects with the discussion of the previous section. The context in question involves reduced relatives in Italian. The pattern with non-reflexives is familiar, with unaccusatives and passives being grammatical in the reduced relative, but unergatives and transitives ungrammatical: Reduced-relative pattern {;uomo 'the man' [arrivato a Milano] [arrestato dalla polizia] *[telefonato a suo nonno]

'arrived in Milan' 'arrested by the police' 'telephoned to his grandfather'

26 Reinhart and Siloni suggest that Reduction applies in the lexicon for Hebrew, but in the syntax (more precisely at LF) for languages such as French. The basis for this claim lies in a difference in productivityonly certain verbs are reflexivizable by Reduction in Hebrew, while in French it is generally available. This type of reasoning is a basic tenet of the lexical-syntactic distinction. 27 There is a question here about how the morphological identity between reflexives and unaccusatives could be captured in such a system; but I will continue with the question of whether the syntactic analysis it implies is justified. 28 Although in the case of Hebrew the arguments offered by Reinhart and Siloni are more straightforward.

David Embick The contrast that Reinhart and Siloni base their argument on is in (27); unaccusatives with si are possible in reduced relatives, while reflexives with si are not: (27)

a.

II bicchiere [rottosi ieri] apperteneva a mio nonno. the glass broken-sl yesterday belonged to my grandfather 'The glass broken yesterday belonged to my grandfather.' b. *Luomo [lavatosi ieri] e mio nonno. the-man washed-sl yesterday is my grandfather 'The man washed yesterday is my grandfather.'

On a syntactic level, the restriction on reduced relatives may be stated as follows: no external arguments in the reduced relative. On the analysis defended here, reflexives are transitive in the sense that they do involve the projection of an external argument, the ditic. Thus the reflexive is crucially unlike unaccusatives and passives, in which no external argument is present at any stage of the derivation. The difference in (26) follows from the fact that 'Active' v, with an external argument, cannot be the complement of perfect aspect in a reduced relative. 29 From this perspective, then, the reduced-relative facts can be captured by either the unaccusative or the unergative analysis. A further question is whether this pattern of selection constitutes evidence for the syntactic visibility of a [-external argument] feature. The difference between the reflexives on the one hand and the passives and unaccusatives on the other, is that the former have an external argument (realized as si) at a particular stage in the derivation, whereas the passives and unaccusatives never do. This, of course, crucially assumes a treatment of passives that involves no 'suppression' of a merged argument; see Embick (1997). If it is assumed that the head immediately above the vp, presumably the aspectual head forming the relative, is sensitive to the presence or absence of a specifier in its complement, then there is no need to posit a [-external argument] feature based on this case. Or the restriction could be derivative of other syntactic principles. A discussion of these issues is found in Iatridou, Pancheva, and Anagnostopoulou (2001), although it is not dear at this point what the relevant factors are and how they are to be captured. However, whatever form this syntactic property tal<:es, it is not the case that this pattern counter-exemplifies the morphological treatment of the u-syncretism advanced in section 5.3. As we have seen, the implementation of effects like those related to Burzio's generalization does not refer directly to [-external argument] features; there is therefore no reason to think that whatever factors the reduced-relative generalization reduces to will require such features. 3D '9 This is on the assumption that reduced relatives are not derived from a clausal structure with whizdeletion; see Bhatt (1999) for further discussions of reduced-relative structures. )0 There is a further implication if the reduced relative pattern is dealt with in the terms suggested above. If reflexives are excluded from the reduced relative environment because they have an external argument at some stage in the derivation, then the diagnostics for the 'unaccusativity' of reflexives of this type must be sensitive to post-syntactic factors. Thus, for instance, the fact that be is selected rather than have as the auxiliary for reflexives is a morphological fact, however the conditions are to be stated. Given the full range of factors that seem to playa role in determining auxiliary selection, this conclusion does not seem

Unaccusative Syntax and Verbal Alternations

I57

Two primary points emerge from the discussion of Reduction-based treatments of un accusatives and reflexives. First, Reinhart's Reduction-based account explicitly acknowledges that the operation must be syntactic in certain cases, and thus cannot provide evidence that a Lexical derivation is required. Second, the Reduction account is based on the claim that the unaccusative analysis of reflexives is inadequate, and that the reflexives in question are really unergative. While Reinhart and Siloni (in this volume) identifY a number of questions about the unaccusative analysis of reflexives as applied to Romance, it seems clear that these points do not impact directly on the analysis of the u-syncretism presented in the last section.

5.6.

CONCLUSIONS

The concerns of this chapter have been primarily morpho syntactic, and have been directed at the nature of the morphological pattern that unites a number of distinct constructions, the u-syncretism. Following a strong hypothesis about the syntaxmorphology interface, the arguments presented above show that the u-syncretism is centred on the absence of an external argument. Further arguments showed that this pattern arises when morphology is underspecified with respect to syntax, and treats passives, reflexives, and unaccusatives in the same way because each of these configurations lacks an external argument. This underspecification-based approach, which holds that features or morphological exponents that appear in the u-syncretism are morphological, was shown to be superior to alternatives based on syntactico-semantic features and Lexical operations. As is always the case, a number of questions remain, one of which I will now address. In the discussion of reflexives throughout the chapter, we have seen many syntactic structures associated with reflexive interpretations. One question that such patterns raise is whether or not one should expect a more uniform syntactic encoding of this sort of semantic phenomenon. For instance, much of the motivation behind the Reduction-based approach developed by Reinhart stems from the desire to offer a uniform semantic operation for all types of notional reflexivization. At a morphosyntactic level, it appears that there are different ways of deriving reflexive interpretations, both within and across languages. In addition to what we have seen earlier in the various case studies, consider English examples of the type John hit himself, Mary washed, Susan got dressed, and Fred self-destructed, each of which is notionally reflexive, while showing apparently distinct syntactic configurations. A further assumption is that Reduction as an operation drives syntactic considerations. In view of the morphosyntactic heterogeneity of the structures that are interpreted reflexively, it seems that this

implausible; cf. also the distinction between 'deep' and 'surface' unaccusativity from Levin and Rappaport Hovav (I996). Reinhart and Siloni (this volume) identifY problems with the use of auxiliaty selection as an unaccusativity diagnostic, however, suggesting that further factors must be considered.

David Embick assumption should be reversed. Assuming that an operation like Reduction is needed for the semantics of reflexivization, one could say that Reduction is triggered by certain configurations generated by the syntax. Recall the idea that Reduction is required at LF for languages like French, suggesting that it would be possible to develop a uniformly LF-based notion of this operation. The idea that there is a uniform semantic basis for reflexive interpretation could then be maintained. This of course assumes that an operation like Reduction is required for the semantics of reflexivization-which is another topic altogether.

6 Against the Unaccusative Analysis of Reflexives Tanya Reinhart and Tal Siloni

6.1.

INTRODUCTION

Over the last two decades the unaccusative approach to reflexive verbs has become rather popular, particularly in works dealing with Romance reflexives. Under the unaccusative approach, the subject of reflexive verbs is an underlying object just like the subject of unaccusative verbs. Although the starting point of most of these studies is Romance reflexives, one would not expect reflexive predicates to have a fundamentally different argument structure cross-linguistically. We contend that the arguments that led linguists to the unaccusative approach can all be handled by a version of the more traditional view that tal{es reflexive verbs to be unergative predicates. Moreover, we show that when reflexives are submitted to syntactic tests of unaccusativity, they systematically fail the tests in a variety oflanguages. More specifically, their subject does not pattern with internal arguments. We believe that the morphological similarity often attested between reflexives and un accusatives is not due to a common argument structure, but to the basic operation at the heart of their derivation. For helpful comments and suggestions, we would like to thank Marc-Ariel Friedemann, Martin Everaert, and the audience at the Round Table on Argument Structure and Reflexives (Paris, I999).

Tanya Reinhart and Tal Siloni

160

Across languages, reflexive verbs have different morphological instantiations. While in French (and Romance in general) reflexive verbs are formed by reflexive elitics (Ia), in English they are morphologically identical to their transitive alternate (Ib). And while in Hebrew reflexives mostly appear in the so called hitpa'el verbal template (Ie), in Dutch, they appear with the reflexive element zieh (Id): (I)

a. b. e.

d.

Max se lave. Max SE washes Max washes. Max mitraxec. Max washes Max wast zich. Max washes ZICH

Furthermore, while in English, Hebrew, and Dutch the set of reflexives is a elosed set, in Romance languages the phenomenon is productive.' Nonetheless, we argue that across languages, reflexivization is essentially the same phenomenon. We adopt the null hypothesis regularly assumed in works on argument structure that the different thematic instantiations of a verbal concept are derived from the same underlying thematic structure. Unlike approaches that decrease the role of the lexicon from an operative component to a list of items (for example, Borer, in this volume; Embick, in this volume; Marantz 1997), we assume that the lexicon is a computational component, where derivational operations can apply. We argue that reflexives are derived from their transitive alternate by an operation that identifies the external ann internal argument and reduces the latter. Further, we attribute the somewhat different nature of reflexive verbs in Hebrew, Dutch, and English vs. Romance to the distinct component of grammar in which the operation applies: lexicon vs. syntax. In contrast with recent proposals that insert the external argument syntactically through vP (Chomsky 1995) or VoiceP (Kratzer 1996), we believe that predicates bear lexical information with regard to their external argument. Hence, the latter can be involved in lexical operations. The chapter is organized as follows. In section 6.2 we reject the possibility that reflexive elitics are object elitics. Section 6.3 discusses the operation of reduction, which, we believe, is the operation that derives reflexive verbs. In section 6.4, we examine and discard the arguments advanced by proponents of the unaccusative analysis in favour of their approach. Section 6.5, in turn, provides cross-linguistic evidence that the subject of reflexive verbs is not an internal argument. In the last section we show how the distinctions between reflexive verbs in Hebrew, Dutch, and English vs. Romance can be straightforwardly accounted for if reflexives can be derived either in the lexicon or in syntax. , 'The fact that zich can only appear with a limited set of verbs is what led Reinhart and Reuland (1993) to the claim that zich is licensed in local contexts only when the predicate is reflexive (what they call intrinsically reflexive).

Against the Unaccusative Analysis

6.2.

r6r

AGAINST AN OBJECT-CLITIC ANALYSIS

Roughly, when we say 'reflexive verbs' we mean verbs denoting an action that the Agent argument applies to itself (or a state of mind the Experiencer argument has with regard to itself). Reflexive verbs appear in what is often called reflexive morphology; but cross-linguistically this morphology is also shared by other types of predicate, such as unaccusatives, middles, and passives. In Romance languages, reflexive verbs are formed by means of elitics labelled 'reflexive elitics', which are similar to pronominal object-elities (compare (2a) with (2b)). A natural assumption is therefore that reflexive elitics, on a par with object elitics, are associated with the object position, as schematized in (2C):

(2)

a.

b.

Jean se lave. Jean SE(REFLcl) washes 'Jean washes.' Jean lei lave t i. Jean himcl washes 'Jean washes him.'

An object elitic analysis of reflexives: c.

Jean sei lave t i. Jean SE(REFLcl) washes

However, by now there is considerable evidence against an object-elitic analysis of Romance reflexives (2C). As already shown by Kayne (1975), reflexives do not pattern with transitive verbs. The reflexive eli tic, then, cannot simply be the object elitic of a transitive entry. Various arguments lead to that conelusion. Consider first the context of expletive insertion in French, illustrated in (3a). While transitive verbs are disallowed in this environment (see 3b), reflexive verbs do occur there, as shown in (3c)! If reflexives were transitive entries, we would expect them to be completely impossible in the post-verbal position of expletive constructions, just like transitive verbs: (3)

a.

11 est arrive trois fllles. there is arrived three girls b. *11 lesi a denonces ti trois mille hommes ce mois-ci. there themcl has denounced three thousand men this month-here c. 11 s' est denonce trois mille hommes ce mois-ci. there SE is denounced three thousand men this month-here 'Three thousand men dep.ounced themselves this month'

2 Judgements may vary among speakers. According to Kayne (1975), example (3c) is entirely grammatical. Some speakers judge it as marginal. Importantly; speakers agree that there is a clear difference in grammaticaliry between transitives (3b) and reflexives (3C).

Tanya Reinhart and Tal Siloni

162

Additionally, Kayne (1975) has observed that French causative constructions, too, treat transitives and intransitives differently and that reflexives pattern with intransitives. When the verb embedded under the causative verb jail'e ('make') is a transitive verb, its subject must be introduced by the preposition a ('to'); see (¥). When the lower verb is intransitive, its subject cannot be introduced by a, as in (4b).3 As is elear from (4 C) , when the direct object of the embedded verb is a pronominal elitic, the verb patterns with transitive entries. But when the lower verb is reflexive, its subject surfaces without the preposition (4d), just as the subject ofintransitive verbs. Notice that the different positioning of pronominal elitics and reflexive elides in the causatives of (4) suggests in itself that they deserve a different syntactic treatment. . (4)

laver Max a Paul. I will+malce wash Max to Paul 'I will malce Paul wash Max.' b. Je ferai courir Paul. I will+make run Paul 'I will malce Paul run.' c. Je Ie ferai laver a Paul. I himcl will+malce wash to Paul. 'I will make Paul wash him.' d. Je ferai se laver Paul. I will+make SE wash Paul 'I will make Paul wash himself'

a. Je ferai

Further, as is well known, transitive verbs use the auxiliary avoil' ('have') to form complex tenses. Reflexives, in contrast, employ hre ('be'). As is the case when the auxiliary etre is used, agreement is always obligatory on the past participle of the reflexive verb (5b). Past-participle agreement with direct-object elities is, in contrast, optional at least in certain French dialects; see (5a). As noted by Sportiche (1998), if reflexive elities were simply object elities, this difference would be unexpected.

(5)

a.

b.

l

(i)

Marie les a decrit(es). Marie themcl has described 'Marie described them' Marie s'est decrit*(e). Marie SE is described 'Marie described herself'

The subject ofintransitives is an accusative argument; when it is eliticized, the accusative elitic is used: Je Ie ferai courir. I himd will+make run 'I will make him run.'

Not only verbs taking an accusative complement disallow an accusative subject but also transitive verbs taking a sentential complement (ii): (ii)

Jean a fait penser *(11.) Marie qu'dle etait belle. Jean has made think to Marie that she was beautiful 'Jean made Mary think that she was beautiful.'

Against the Unaccusative Analysis The same can be shown in Icelandic, for example, where Case is morphologically marked. The argument is suggested by Marantz (1984) on the basis of data from Andrews (1982). Sentence (6a) is an Exceptional Case Marking construction. The accusative subject of the subordinate elause is the simplex anaphor sig. As expected, the predicative adjective sterkan ('strong') must bear accusative Case in agreement with its subject sig. Example (6b) is a paraphrase of (6a), containing a reflexive verb, which is formed by the reflexive suffix -st. If -stwere a reflexive object elitic, one would expect its predicative adjective to also appear in accusative Case. However, as shown in (6b), the adjective must surface in nominative in agreement with hann, the subject of the reflexive verb. The reflexive suffix, then, is not the elitic version of the anaphor sig, and is not associated with the object position. (6)

a.

b.

Hann telur sig vera sterkan. (Icelandic) he. NOM believes himsel£Acc to+be strong.Acc Hann tel-st vera sterkur. (Icelandic) he. NOM believes-REFL to+be strong. NOM

Indeed, the object-elitic analysis of reflexive verbs is not promising, and two alternative trends of research have consequently been developed. On the one hand, it has been argued that reflexive verbs are the output of a lexical operation of absorption or reduction, which applies to a transitive entry, targeting its internal argument and producing an intransitive verb. Among proponents of this lexical approach are Grimshaw (1982), Wehrli (1986), and Chierchia (this volume). Under this view, the reflexive elitic is associated with the internal a-role, but in the lexicon, not in syntax. On the other hand, it has been proposed that the reflexive elitic is, in fact, associated with the external a-role, and the reflexive verb is therefore an unaccusative verb, as its subject is the internal argument. Among the defenders of the unaccusative approach, some argue in favour oflexical absorption of the external argument (Bouchard 1984; Grimshaw 1990; Marantz 1984), while others believe that the external argument is present in syntax via se (Kayne 1988; Pesetsky 1995; Sportiche 1998, among others). In sections 4 and 5 we argue against the unaccusative analysis of reflexives in either its lexical or syntactic version. But first, in the next section, we adopt the proposal that reflexives are products of reduction of the internal argument. In the last section we further refine the proposal, suggesting that reduction may have a syntactic as well as a lexical mode of application.

6.3.

REDUCTION

If the reflexive elitic is not the internal argument in syntax, it is only natural to explore the possibility that it is nonetheless associated with the internal a-role-in the lexicon, not in syntax. Indeed, Grimshaw (1982) suggests that the reflexive elitic is a marker oflexical reflexivization, which is a lexical operation binding the internal argument

Tanya Reinhart and Tal Siloni by its external co-argument, rendering the former syntactically inaccessible. Similarly, according to Wehrli (1986), reflexive se absorbs the internal argument, which is consequently unavailable to syntactic processes. The lexical option is elaborated in Chierchia (in this volume). Following him, we assume that an operation labelled 'reduction' can operate in the lexicon on transitive entries to produce reflexive verbs. Reduction applies to a two-place relation (predicate), identifies the two arguments, and reduces the relation to a property. Reflexive reduction in the lexicon turns a transitive entry such as wash in (7a) into an intransitive entry whose single 8-role is the external 8-role (8,), as schematized in (7b). Reduction requires identification of two arguments, thereby capturing the interpretation of reflexive sentences. Thus the output is always interpreted as schematized in (7c). For a more precise analysis of the semantics see Chierchia (in this volume). In section 6.6 we will propose that reduction can apply not only in the lexicon, but also in the syntax ofLE This will derive the distinctions between reflexives in Hebrew (and English and Dutch) and Romance, respectively. (7)

a. wash (81'8,) b. Reduction: R(wash) (8,) c. R(wash)(x) H Ax [x wash x]

6.p. A constraint on reduction reduction applies under identification of two 8-roles, a two-place relation is required. In Romance, the argument to be reduced in reflexivization can be either an accusative or a dative argument, as illustrated in (8a) and (8b), respectively:

.fu;

(8)

a. Jean s'est introduit a Paul.

b.

Jean SE is introduced to Paul 'Jean introduced himself to Paul.' Jean s' est achete une voiture. Jean SE is bought a car 'Jean bought a car for himself.'

However, whether the reduced argument is accusative or dative, identification must take place with the external argument when the verb is a three-place predicate. Thus, while it is possible for an anaphor in situ to be bound by an internal co-argument, as illustrated in French (9a) and Hebrew (9b), reduction involving two internal coarguments is entirely inconceivable; see (9C):4 (9)

a.

?Sur cette photo Jean n'a montre les enfants qu'a eux-memes. on this picture Jean not has shown the boys but to themselves

4 French nonclitic anaphors, like other nonclitic pronominals, show restrictions that need not interest us here (stress on the anaphor often improves the example). The important point here is the difference in grammaticality between (9a) and (9c). An example equivalent to (9c) cannot be constructed in Hebrew.

Against the Unaeeusative Analysis

b.

Dan her' a Ie-dina' et ' acma ba-tmuna. Dan showed to-Dina.Acc herself in+the-picture e. *Jean Si' est montre l'enfanti • Jean SE is shown the boy And for the same reason, predicates that lack an external a-role are incompatible with reflexive clitics, as already observed by Burzio (1981, 1986), among others. Sentence (IOa) contains a raising predicate with an embedded small-clause complement and a dative argument. The predicate cannot occur in a reflexive form (see lOb), although the reflexive interpretation is possible when a nonclitic anaphor is used, as shown in (lOe):

(IO)

a.

Jean leur semble intelligent. Jean to+themcl seems intelligent b. *Jean se semble intelligent. Jean SE seems intelligent e. Jean ne semble intelligent qu'a lui-meme. Jean not seems intelligent but to himself

Likewise, reduction cannot apply when the external a-role is not free, not available because it is subject to another operation. This explains Kayne's (1975) original observation that reflexive clitics are incompatible with passivization, as illustrated below by Italian examples (from Rizzi 1986b). Again, while pronominal objects of passives may either surface in situ (see lIa) or cliticize (lIb), reflexive sentences can be expressed by means of a nonclitic anaphor (as in lIe), but not by means of a reflexive verb (ud), as the subject in question is a derived subject, not the external argument, which has been 'demoted' by passivization: (II)

Gianni estato affidato a lui. Gianni was entrusted to him b. Gianni gli estato affidato. Gianni to+himcl was entrusted e. Gianni estato affidato a se stesso. Gianni was entrusted to himself d. *Gianni si estato affidato. Gianni SI was entrusted

a.

Reduction is thus a constrained operation. We propose it must obey the following constraint, to which we return in sections 6.4 and 6.6:

(12)

Reduction can only apply to a pair of free a-roles, one of which is external.

6.3.2. Reduction of the external argument It is known that across languages the so-called reflexive morphology tends also to

Tanya Reinhart and Tal Siloni

166

appear on unaccusative verbs. In Romance, an unaccusative can occur with the same elitic that appears on reflexive verbs, as illustrated by French examples:

(13)

a.

b.

La branche s' est cassee. the branch SE is broken Jean s' est evanoui. Jean SE is fainted

And in Hebrew, where there are several verbal forms an unaccusative verb can take, many occur in the same form as reflexive verbs, in the hitpa'el verbal template. Chierchia (this volume) argues that the fact that reflexives and unaccusatives can share the same morphological form can be explained if unaccusatives are also derived from a two-place verb by some sort of reduction. That is, reflexive morphology tends to appear when a derivational operation (reduction) takes place. The actual reduction operation which Chierchia proposes for unaccusatives is rather different from reflexive reduction. Retaining his insight, Reinhart (2Oora) assumes two reduction operations, depending on whether the internal argument is reduced or the external one. Internal role reduction, Reflexivization, creates reflexive predicates, whose interpretation is schematized in (144). External role reduction, which forms unaccusative predicates, eliminates the external role altogether (and is therefore labelled expletivization), as schematized in (I4b). (14)

a. b.

Internal Reduction: Reflexivization: RR (V) (x) H Ax [Vex, x)] External Reduction: Expletivization: RE (V)(x) H Ax [Vex)]

Technical details aside (see cited references for extensive discussion), the basic idea is that se casser in (13a) is derived from its transitive alternate casser by reduction that targets the external argument. When there is no transitive alternate, as in the case of s'evanouir (13b), reduction applies to an abstract transitive alternate, which is frozen, not accessible to use by the active lexicon. Thus, when reduction applies to the internal role, the external role is syntactically realized, and a reflexive verb is obtained. When the external role is reduced, the internal argument is syntactically realized, resulting in an unaccusative entry.5 Under this view, then, reflexives and unaccusatives can, and often do, bear the same morphology because both are the outputs of a derivational operation (reduction). They do not, however, share the same argument structure. Reflexives, unlike

5 According to Reinhart (20OIa), expletivization reduces only a Cause a-role. Note that Wehrli (1986) has already suggested, without elaborating, that unaccusative se occurs when lexical absorption of the external argument tal<es place. Likewise, Borer and Grodzinsky (1986) have proposed that in Hebrew the hitpa'e! forms are derived either by elimination of the external argument-which gives rise to unaccusatives-or by elimination of the internal argument-which forms reflexives. Finally, a word on terminology. Forms that do not have a transitive alternate (s'tvanotlir in (13b), for example) are often called by traditional grammars 'intrinsic reflexives'. Reinhart and Reuland (1993) use the term differently (see n. 1). As the term is ambiguous, we do not use it here. It will become clear in section 6.6 that what Reinahrt and Reuland call 'intrinsic reflexives' are lexical reflexives in our terms, as they are formed in the lexicon.

Against the Unaccusative Analysis unaccusatives, are unergative entries; their subject is the external argument. This view is in contradiction with the unaccusative approach to reflexives, which talces reflexive verbs to be unaccusative predicates to the extent that their subject is an internal argument. The unaccusative analysis of reflexives has become very popular in the last two decades starting with Marantz (1984) (see Bouchard 1984; Kayne 1988; Grimshaw 1990; Pesetsky 1995; Sportiche 1998; Embick, in this volume). Under the unaccusative approach, the fact that reflexives and unaccusatives can share the same morphology simply follows from the fact that reflexives are unaccusatives as their subject is an internal argument. Interestingly, however, what initiated this line of research was not so much the form shared by reflexives and unaccusatives, but other arguments, which are examined below.

6.4.

ON THE UNACCUSATIVE ANALYSIS

According to the unaccusative approach, the subject of reflexives is an underlying object which has to raise to subject position for Case reasons, because the reflexive morphology absorbs its Case. The approach has two major variants: lexical and syntactic. While under the former the external argument is absorbed in the lexicon (Bouchard 1984; Marantz 1984; Grimshaw 1990), according to the latter, the external argument is present in syntax via the reflexive elitic se (Kayne 1988; Pesetsky 1995; Sportiche 1998). As the arguments that led linguists to the unaccusative approach are basically the same whether the approach is lexical or syntactic, we abstract away from this difference in our discussion. Recall first that the starting point of most studies defending the unaccusative path is Romance reflexive elitics. However, the analysis should hold across languages, or else the hypothesis is weakened as we end up elaiming that reflexive verbs are not the same phenomenon cross-linguistically. It is worth noting at the outset that among our sample languages here, Dutch reflexives are hardly analysable as unaccusatives. Dutch zich is not a elitic and it would be difficult to explain how it gets associated with the external 8-role. According to Reinhart (2000), it is a residue of the internal argument affected under reflexive reduction.

6+1. Morphological evidence Although proponents of the unaccusative analysis do not heavily rely on the fact that unaccusatives and reflexives can share the same verbal form, the morphological angle does playa role in their argumentation. Marantz (1984) mentions that in Albanian, for example, reflexives and passives share the same form (see 15), arguing that the morphology in question appears when the subject is an underlying object. The same argument can be made in French, where in addition to unaccusatives (see (13) above), also middles, (16a), and a limited set of passives (see 16b) can be formed by means of

Tanya Reinhart and Tal Siloni

168

'reflexive' morphology. 1he same is true for Hebrew, as well as other languages (see Embick, in this volume, for an overview of a few languages). (15)

Agimi lahet. (Albanian) Agim wash.3SG i. 'Agim washes himself.' ii. 'Agim is washed.'

(16)

a.

b.

Ces tomates se vendent bien. well these tomatoes SE sell 'These tomatoes sell well.' Le crime s'est commis ce matin. the crime SE is committed this morning 'The crime was committed this morning.'

(Zribi-Hertz 1982)

Moreover, in French (and Italian) reflexives and unaccusatives select etre ('be') and not avoir ('have') as their auxiliary (see 17), just like passives and middles. 6 The choice of etre, according to the unaccusativity defenders (Grimshaw 1990; Pesetsky 1995, for instance), signals that the subject position hosts a derived subject. Sportiche (1998) mal{es the same point on the basis of the behaviour of participle agreement. Assuming that in sentences with the auxiliary etre participles always agree with the underlying object, he concludes that the subject of reflexives must be a derived subject, as it triggers agreement on the participle (again, just like in unaccusatives, passives, and middles): (17)

a.

b.

La voiture est passe-e. the car is passed-AGR Marie s' est lave-e. Jean SE is washed-AGR

The morphological arguments, however, are not very strong. The fact that different diatheses of a verb may appear in the same morphological guise does not mean that their derivations are of the same nature, nor that they share the same type of grammatical subject. The behaviour of participle agreement is a consequence of the choice of auxiliary. And auxiliary selection, in turn, is an intricate matter, which is not yet well understood, although it has received much attention in the literature (for example, Hoekstra 1984; Ackema 1995; Everaert 1996; Friedemann and Siloni 1997; Reinhart 2000). Note, incidentally, that neither reflexives nor unaccusatives consistently choose BE cross-linguistically. At any rate, clearly, the simplest procedure to determine whether reflexives in French and Italian use BE due to their unaccusative character or due to different factors is first to submit their subject to syntactic tests that discriminate between external and internal arguments. Section 6.5 is devoted entirely to such

6 More precisely, in French, unaccusatives tend to choose the auxiliary etre, but there seem to be instances of unaccusatives with avoir.

Against the Unaccusative Analysis tests. AI> will become elear in the course of the section, there is robust evidence that the subject of reflexive verbs systematically patterns with the subject of unergatives; it is the external argument, unlike the subject of unaccusatives. Yet, defenders of the unaccusative analysis did not base their argumentation solely on morphological grounds.

6.4.2. Syntactic evidence A popular syntactic argument in favour of the unaccusative analysis is the incompatibility of reflexive elities with verbs lacking an external argument (Bouchard 19S4; Kayne 19S5; Grimshaw 1990; Pesetsky 1995; Sportiche 1995). Neither raising predicates nor verbs in passive can take a reflexive elitic, as shown in (lOb) and (lId) above. This follows particularly well from the view that the reflexive elitic absorbs or bears the external 8-role. If the latter is not available, due to the nature of the predicate or due to passivization, the reflexive elitic cannot appear. However, the same generalization easily follows also from the constraint imposed on reduction (see 12): reduction can only apply to a pair of free 8-roles one of which is external. AI> discussed in section 6.p, with both raising predicates and passives, there is no free external 8-role for reduction to be possible. Furthermore, the empirical coverage of the constraint in (12) is larger than that of the account offered by the unaccusative approach. Dutch presents a case where the latter fails while the former works. Consider the expletive constructions in (IS), where the external argument of the predicate is demoted (and implicit) just as in regular passives. They can accommodate transitive verbs (see ISa) as well as unergatives (Isb) (the so-called impersonal passives). But they do not allow reflexives (see ISC) nor unaccusatives (as in ISd).

(IS)

a.

Er werd een kind gewassen. there was a child washed b. Er werd gedanst. there was danced c. *Er werd zich gewassen. there was ZICH washed d. *Er werd gegroeid. there was grown

Reflexives and unaccusatives, then, seem to eluster together. However, as mentioned in the beginning of the section, Dutch zich can hardly be argued to be associated with the external argument and zich gewassen to be an unaccusative entry. Thus, while under either approach (ISd) is impossible because passivization is incompatible with unaccusative entries as there is no external argument available, (ISC) remains unexplained under the unaccusativityview. The reason is that the construction'disallows the reflexive verb, although zich gewassen cannot be argued to be an unaccusative predicate, which does not have an external argument. Under the reduction view, in contrast, the impossibility of (ISC) as well as (ISd) and (9c), (lOb), and (nd) is due to the same reason:

Tanya Reinhart and Tal Siloni

170

their derivation involves reduction, which is impossible due to the constraint in (12), which limits reduction to predicates with external and internal free 8-roles? Finally, the strongest argument against deriving reflexives through a lexical operation targeting the internal argument is offered by Marantz (1984). His argument is based on the Icelandic example in (6b) above, but the same can be illustrated in French. Consider the Exceptional Case Marking construction in (19a) and its reflexive equivalent in (19b). The matrix predicate considere does not take a DP as its internal argument. Max in (I9a), to which considere assigns accusative Case, is the subject of the small clause, receiving its 8-role from the adjective intelligent. As it is not an argument of considere, a lexical operation on the argument structure of the verb cannot affect it. Marantz concludes that reflexivization absorbs the external argument, and that the subject of reflexives is therefore the underlying object. Example (19b), according to him, would be analysed as schematized in (20): jean, the subject, is the internal argument.

(19)

a. b.

Jean considere Max intelligent. Jean considers Max intelligent Jean se considere intelligent. Jean SE considers intelligent

Marantz's analysis of (19b): (20)

Jean j se considere [tj intelligent] Jean SE considers intelligent

But, in fact, reflexivization into ECM complements poses a problem to any lexical analysis, whether it reduces the external or internal argument. Reflexivization entails linking two arguments, identifYing them in our terms. If it takes place in the lexicon, only two co-arguments (arguments of the same predicate) can be involved. In 7 Clearly, the operation forming passives is incompatible with reduction. Two remarks are nonetheless in order. As noted by Martin Everaert (personal communication), if zich is analysed as an anaphoric argument in situ and is further assumed to be a definite expression, examples such as (18c) can be ruled out on a par with (i), which is impossible because the nominal expression is definite (and compare this with (18a)). As mentioned earlier, we do not believe zich is a reflexive argument in situ, and will not examine this possibility here. Second, plaire ('please') and deplaire ('displease') in (ii), which are experiencer verbs with unaccusative syntax (Belletti and Rizzi 1988; Pesetsky 1995; Reinhart 20mb), do allow reflexive clitics (Arad 1998a; Landau 20m). Experiencer verbs of the ptaire-class are not very common, and not all members allow it; see (iii). We will not investigate this exception here, but note that it is completely unexpected under the unaccusative approach, which associates se with the external a-role. The constraint in (12), in contrast, can be refined to allow it.

(i) *Er werd het kind gewassen. there was the child washed

(ii)

Jean se plalt/ deplalt sur cette photo. Jean SE pleases/displeases on this picture 'Jean pleases/displeases himself on this picture.'

(iii) *Jean se manque. Jean SE misses (experiencer verb)

Against the Unaccusative Analysis ECM constructions, the two relevant arguments are not co-arguments; in (I9b) the two arguments that the operation applies to are not arguments of the same predicate. Hence, they cannot be linked to each other in the lexicon. To link the two arguments, lexical analyses it la Marantz-which absorb the external argument in the lexiconhave to impose a syntactic condition on an element (the external argument) which is no more available in syntax as it was absorbed in the lexicon; such a condition is ad hoc and implausible. This may be what led other linguists to prefer the syntactic version of the unaccusative analysis, under which the two arguments in question are present in syntax, and can therefore be in syntactic binding relations. In the subsequent section, however, we provide decisive evidence that the subject of reflexive verbs is an external argument, unlike the subject of unaccusatives. This evidence, we believe, refutes any unaccusative approach to reflexives. Note, in addition, that reflexivization into ECM complements is not a phenomenon that holds across languages. We do not find anything of the sort in Hebrew (2Ia) or English (22a). An anaphor pronoun in situ must be used in these languages to obtain the relevant interpretation (21b, 22b): (21)

a. *Dan mitxasev 'intiligenti. Dan considers.REFL intelligent b. Dan maxsiv 'et 'acmo 'intiligenti. Dan considers.Acc himself intelligent

(22)

a. *Dan considers intelligent. b. Dan considers himself intelligent.

Mter presenting cross-linguistic evidence against the unaccusative analysis, we propose our solution to the ECM puzzle, arguing that the different behaviour of French-type and Hebrew-type reflexive verbs follows from the distinct component of grammar in which the reduction operation applies: syntax vs. lexcion.

6.5.

AGAINST THE UNACCUSATIVE ANALYSIS

So far, then, there does not seem to be evidence in favour of the unaccusative analysis of reflexives. Moreover, as mentioned above, the unaccusative hypothesis is hardly extendable to Dutch reflexives. The next obvious move is to check whether reflexives in languages other than Dutch have the characteristics of unaccusatives (which was not really done in the literature cited above). As will become clear below, reflexives systematically fail tests of unaccusativity. More specifically, their subject does not pattern with internal arguments. This is true for a variety of languages, such as Hebrew, English, Russian, and the Romance family itself, which was the starting point for most studies defending the unaccusative approach. It is welllmown that the French quantitative elitic en can eliticize only out of the object position. It can thus serve as a test to discriminate between the internal and

Tanya Reinhart and Tal Siloni

172

external argument in a post-verbal position. Sentence (23a) contains an unaccusative verb; en cliticization is possible; see (244). Examples (23b-c) constitute a minimal pair: (23b) is a reflexive verb, and (23c) is an unaccusative with 'reflexive' morphology. While the latter allows en cliticization (see 24c), the former disallows it; see (24b).8 This is straightforward if the subject of reflexives is an external argument, unlike the subject of unaccusatives.

(23)

a.

b.

c.

(24)

11 est arrive trois fllles hier soir. there is arrived three girls yesterday evening 'There arrived three girls yesterday evening.' (?)Il s'est lave beaucoup de touristes dans ces douches publiques, there SE is washed many tourists in these public showers recemment. recently 'Many tourists washed in these public showers recently.' 11 s' est casse beaucoup de verres dans ce lave-vaisselle. glasses in this dishwasher there SE is broken many 'Many dishes broke in this dishwasher.'

a.

11 en est arrive trois hier soir. there of+themcl is arrived three yesterday evening b. *11 s'en est lave beaucoup dans ces douches publiques, in these public showers there SE of+themcl is washed many recemment. recently c. 11 s' en est casse beaucoup dans ce lave-vaisselle. in this dishwasher there SE of+themcl is broken many

According to Guglielmo Cinque (personal communication, cited by Grimshaw (1990: 184, n. 3», the same pattern holds in Italian, as illustrated in (25).9 (25)

a.

Ne sono arrivati tre. of+themcl are arrived three b. *Se ne sono vestltl tre. SI of+themcl are dressed three

Reduced relatives supply another syntactic context to distinguish between external and internal arguments. Reduced relatives with the so-called past participle do not allow predicates with an external argument (see Siloni 1995,1997); hence, the contrast between (26a-b) , whose reduced relative contains an unaccusative and a passive, respectively, and (26c), where the predicate in the reduced relative is an unergative verb. 8 Recall that certain speakers already find (23b) somewhat marginal. Nonetheless, for all speakers, (24b) is completely impossible, whether they judge (23b) as marginal or entirely acceptable. 9 Italian speakers seem divided on (25b), some categorically niling it out, others accepting it. All the spealters accept (25a).

Against the Unaccusative Analysis

173

Consider now the examples in (27). The reduced relative in (27b), which contains a reflexive predicate, is impossible. But, when the same morphology is used to form an unaccusativeverb, the reduced relative is acceptable-see (27a). Once again, reflexives do not pattern with unaccusatives. IO I.:uomo arrivato a Ginevra e una spia. the man arrived in Geneva is a spy b. I.:uomo arrestato dalla polizia e una spia. the man arrested by the police is a spy c. *I.:uomo telefonato a suo nonno e una spia. the man telephoned to his grandfather is a spy

(26)

a.

(27)

a.

II bicchiere rottosi len apparteneva a mio nonno. the glass broken-sl yesterday belonged to my grandfather b. *I.:uomo lavatosi len e mio nonno. the man washed-sl yesterday is my grandfather

Within the Semitic family, reflexives do not pattern with unaccusatives either. In Hebrew, the subject of reflexives is decisively an external argument; it fails diagnostics of internal arguments just like the subject of unergatives and unlike the subject of unaccusatives. As observed by Shlonsky (1987), among others, there are two types of post-verbal subject in Hebrew. One type appears in triggered inversion also labelled stylistic inversion, which is licensed by an XP immediately preceding the verb [XP V S]. These postverbal subjects do not concern us here. Another type of post-verbal subjects, which do not require a pre-verbal trigger, is found with unaccusatives (see 28a) and passives, shown in (28b); these post-verbal subjects are internal arguments. External arguments do not allow simple inversion [V S] (see 28c). As shown below, reflexives cannot appear in simple inversion (see 28d) just like unergatives, while unaccusatives with identical morphology (in the hitpa'el verbal template) do allow it (see 28e), on a par with other predicates whose subject is an internal argument. II

(28)

a.

Nisbar masehu. broke something

10 Judgements are due to Guglielmo Cinque and Alessandra Lukinovich. In French the test is not applicable, as participial relatives of this land disallow elitics altogether (unlike relatives with the so-called present participle; see Siloni I995, I997 for discussion). Arguing in favour of the unaccusative analysis of reflexives, Embick (in this volume) responds that reduced relatives indeed disallow reflexive predicates because they project an external argument, but the latter is not their subject but rather the elitic si. While this reasoning may supply an alternative explanation to the ungrammaticality of (27b), it is absolutely inapplicable to all the other arguments we present in this section, as they rely on tests that directly identifY internal arguments. II Additional factors, such as focus, affect the choice of post-verbal subjects and may therefore malre certain examples less acceptable than others. Arguably, stylistic inversion involves V-raising out ofIP (Shlonsky and Doron I992), while in simple inversion, the subject stays in its VP-internal position and SpecIP is filled by a null expletive. Hence, by and large, simple inversion is a trait of pro-drop languages. If a null expletive is not selected in the numeration, the subject has to raise to SpecIP to check the EPP feature.

Tanya Reinhart and Tal Siloni

174

b.

Ne' ecru slosa xayalim ba-hafgana. were+arrested three soldiers in+the-demonstration c. *Rakdu slosa yeladim ba-mesiba. danced three boys in+the-party d. *Hitlabsu salos dugmaniyot ba-knisa. dressed three models in+the-entrance e. Hit' alfu slosa xayalim ba-hafgana. fainted three soldiers in+the-demonstration The same holds in embedded contexts. Raising predicates disallow [5 V] order in the embedded infinitival clause (for reasons which need not interest us here). Unaccusative verbs allow [V 5] order and can therefore realize their subject in the subordinate clause (see (29a-b)). Reflexives such as (29d), in contrast, just like unergatives such as (29c), disallow simple inversion [V 5] and hence cannot realize their subject in the embedded clause (note that the unaccusative in (29b) and the reflexive (29d) share the same morphology).

(29)

a.

Crixim le-hagi' a 'asara talmidim. need to-arrive ten pupils b. Crixim Ie-hit' asef 'asara talmidim. need to-gather ten pupils , c. *Crixim li-rkod asara talmidim. need to-dance ten pupils d. *Crixim le-hitlabd 'asara talmidim. need to-dress ten pupils

Modification by possessive datives can also be used to detect internal arguments in Hebrew. & noted by Borer and Grodzinsky (1986), possessive datives can only modifY internal arguments. Hence, they can serve as possessors to subjects of unaccusatives, as in (30a-b), but not to subjects of unergatives (see 30c). & expected, reflexives such as (30d) behave justlike unergatives (again, note that the predicates in (30b) and (30d) share the same morphology).I2,'3

I l Another test that seems to point to the same direction is the possibility to use kol as a negative polarity item. Doran and Mittwoch (1987) note that this is only possible with internal arguments. As expected, reflexives fail the test just like unergatives. However, the diagnostics should be taken with some caution; not only this use of kol belongs to formal language, but in addition it is better with [-animate] nouns, which makes the test more difficult for reflexives whose subject is obligatorily an Agent in Hebrew (we use triggered inversion contexts below as the negative polarity reading is easier to obtain with post-verbal subjects). We will not discuss this construction here.

(i)

'etmol 10 higi'a leol faksl?talmid. yesterday not arrived any fax/pupil

10 'avad leol maxsev/po'el. yesterday not worked any computer/worker

(ii) "etmol (iii)

*' etmol

10 hitraxec leol yeled. yesterday not washed any boy

Against the Unaccusative Analysis (30)

a.

Sney sfarim naflu Ie-dan. two books' fell to-Dan b. Ha-simla hitkamta Ie-dina. the-dress wrinkled to-Dina c. *Ha-kelev saxav Ie-dina the-dog lay to-Dina d. *Ha-xatul hitgared Ie-dina. the-cat scratch.REFL to-Dina

In Russian, genitive of negation provides a test of unaccusativity: internal arguments can bear genitive Case when their predicate is negated (Pesetsky 1982). Unaccusatives can appear in the same form as reflexives in Russian. Importantly, however, while the former pass the test of negation (when they are negated their subject can appear in genitive) (see 31a), the latter fail it (see 31b), as do unergatives (see 31C):14

(31)

a.

Ne objavilos' studentov. NEG showed up students.GEN b. *Ne pomylos' studentov. NEG washed students.GEN c. *Ne tancevalo studentov. NEG danced students.GEN

Finally, even in English it seems that there is evidence that the subjects of reflexives are external arguments. Agent nominals-also known as -er nominals-can be derived only from predicates with an external argument (as their name suggests); hence the contrast between (32a) and (32b). As expected, reflexives pattern with unergatives: they can give rise to agent nominals (32c):15 (32)

a.

She runs so fast because she is an experienced runner.

b. *She moves so gracefully because she is an experienced mover. c.

She dresses slowly because she is an elegant dresser.

I) In literary Hebrew the verbal form nitpa'el can sometimes be used instead of hitpa'el (that is, the Mishnaic instead of the Biblical form, which is the current form in Modern Hebrew). Although the use is limited to literary language, it seems (to us and to the informants we have consulted) that nitpa'el cannot be used as a reflexive 0), but can occur as an unaccusative (ii). Again, this is unexpected, if reflexives and unaccusatives belong to the same class:

(i)

hu hitlabeS/*nitlabeS; histarek/*nistarek; hitraxecl*nitraxec. he dressed combed washed

(ii)

hu hitkaveclnitkavec; hitkamet/nitkamet; hem hit'asfu/nit'asfu. he shrunk wrinlded they gathered

'4 Thanks to Irena Botwinik-Rotem, Nora Goldshlach, and Lea Nash for the Russian data. Judgements vary among speakers as genitive of negation is not equally productive for all Russian speakers. See Pesetslry (1982) and.Schooriemmer (this volume) for the use of the preposition po in its distributive meaning ('one x each') as an additional test of unaccusativiy. 15 This does not mean, of course, that all reflexives and unergatives can form Agent nominals.

Tanya Reinhart and Tal Siloni

17 6

We believe it is now evident that unaccusatives and reflexives do not belong to the same syntactic class; the difference between them is as substantial as that between unaccusatives and unergatives. This leaves us with the ECM puzzle. How come in Romance languages reflexivization can apply to two arguments which are not 8-coarguments (see 19b)? Or, in other words, if reflexivization is essentially the same phenomenon across languages, why do languages differ with regard to the possibility to reflexivize into ECM complements (see 19b vs. 2M and 22a)?

6.6.

REFLEXIVE REDUCTION: LEXICAL AND SYNTACTIC APPLICATION

In French (and Italian), reflexivization is possible into ECM complements, as illustrated in (19b), repeated in (33a). The same is impossible in Hebrew (and English), as shown in (21a), repeated in (33b).

(33)

a.

Jean se considere intelligent. Jean SE considers intelligent b. *Dan mitxasev 'intiligenti. Dan considers.REFL intelligent

Moreover, while in French, reflexivization is a productive operation, in Hebrew it is limited to a closed class of elements. Similarly, while in French the argument reflexivization targets can be a dative element (in addition to accusative), as illustrated in (8b) or in (34), in Hebrew, reflexivization of datives seems not to be attested. (34)

Jean s' est envoye une lettre. a letter Jean SE is sent 'Jean sent a letter to himself'

In Reinhart and Siloni (1999), we argue that the occurrence of reflexive verbs in ECM constructions in French-type languages, but not in Hebrew-type languages, correlates with their productivity in the former and non-productivity in the latter. Despite the distinctions, we argue, reflexive verbs are one single phenomenon across languages. Reflexive entries are always derived from their transitive alternate by the operation of reduction targeting the internal argument, along lines proposed by Chierchia (in this volume). However, according to our proposal, reduction has two modes of application: a lexical mode and a syntactic mode. In Hebrew-type languages, reflexives are products of lexical application, while in French-type languages they are the output of syntactic application. We summarize the proposal below. The distinctions between Hebrew-type and French-type reflexives are derived from the distinct component of grammar in which reduction applies. While in Hebrew reflexives are derived by reduction in the lexicon, in French their derivation involves

Against the Unaccusative Analysis

177

syntactic reduction. Crucially, in the two types of language, the reflexive morphology (se or the verbal template hitpa'e/) has the effect of eliminating a Case feature of the verb. However, in Hebrew the choice of reflexive morphology requires reduction in the lexicon, while in French se is a Case absorber allowing reduction to talce place in the syntax ofLE In Hebrew, reflexive reduction implies both thematic and Case reduction in the lexicon. A transitive entry (35a) which undergoes reflexive reduction forms an unergative entry (35b), which does not bear an accusative Case assigning feature. The relevant numeration therefore contains in addition to a verb reduced to assign a single 6-role (6,), one realizable DP that will check the EPP (Extended Projection Principle) feature on, as schematized in (35c): (35)

Lexical reflexive reduction: a. V (6,,6 2 ) b. Reduction: R(V)(6,) (R(V)(x)) H Ax [V(x,x)] Case reduction c. Numeration: {... IEPP ... R(V(6,)), DP}

In French, the reflexive morphology also absorbs a Case feature of the verb, but there is no thematic reduction in the lexicon. Se is a Case absorber which is not contingent upon lexical thematic reduction, as schematized in (36b).'6This results in a discrepancy between the number of6-roles and the number of Case features available to assign. The relevant numeration thus includes a verb bearing two 6-roles, but only one realizable DP, because a Case-assigning feature of the verb was eliminated (see (36b)). If two DPs are selected, the derivation will crash due to lack of Case. Reduction applies at LF-see (36c). Under LF-reduction, the two 6-roles of the predicate are identified to the extent that both are assigned to the same DP, as it is the only realizable DP. This thematic discharging obtains at LF an interpretation equivalent to that of lexical reduction; see (35b). The two 6-roles are available in syntax but checked against one single DP: (36)

LF reflexive reduction: a. V (6,,62 ) b. Numeration: {... IEPP ... SE V(6,,6J, {DP}} SE: Case absorption c. LF:[DP(8,,8 2 )SEV+I ...]

The mechanism of LF-reduction is incongruous with the traditional formulation of the 6-criterion, as two 6-roles are assigned to one argument. However, we assume with Chomsky (1995) and Reinhart and Reuland (1993), among others, that the criterion is not indispensable, as its consequences, it seems, can be derived by independent 16 For the sake of simplicity, the base predicate is a two-place predicate; modulo the additional argument, the same holds for a three-place predicate. The absorbed Case may be either accusative or dative.

178

Tanya Reinhart and Tal Siloni

modules of the theory (see cited references). Note that the mechanism requires the insertion of SE. It cannot apply to transitive entries as the derivation will crash due to unchecked Case-assigning features of the verb. Hebrew does not have a sE-type Case absorber that allows the predicate to postpone thematic reduction until LF. Its reflexive morphology requires reduction to apply in the lexicon. As is often the case with outputs oflexical operations, the set of reflexives is a closed set. We leave open the question of what defines the set. Note nonetheless that across languages it is roughly the same set of predicates (the so-called grooming verbs, for instance). No instance of dative-reflexivization is attested in Hebrew. This could be a basic trait of lexical reflexivization or a side effect of the definition of the set. I7 Further, as in Hebrew the operation is obligatorily lexical, it can only operate within the same 8-grid. Hence, there are no reflexive ECM predicates (see (33b», as their formation involve two 8-roles which belong to two distinct 8-grids. In French, reduction is a syntactic operation. It is productive just as other syntactic procedures. Accordingly, the group of reflexives is not a closed class, and the operation can target accusative as well as dative arguments (see (34». Further, reduction in French applies in the syntax of LF. At that level, we assume, the predicates that compose the ECM structure can form a complex predicate (along lines proposed by Reinhart and Reuland (1993) and references cited there). In (33a), for example, the verb considere and the adjective intelligent allow the formation of the complex predicate intelligent-considere, as schematized in (37). Consequently the 8-roles of intelligent and considere become 8-roles of the same predicate-the complex predicate formed at LF-and can therefore be subject to LF-reduction. That is, they can be assigned to the same DP (see (36c»,Jean in (37): (37)

LF: [Jean se [intelligenti-considere]j tJ Jean SE intelligent-considers

Reflexivization of ECM predicates is thus limited to languages with LF-reduction. Only in these languages can reduction in ECM environments respect the constraint in (I2)-repeated here in (38)-with the emphasis that the pair of8-roles in question must indeed be a pair of co-8-roles: (38)

Reduction can only apply to a pair of free co-8-roles, one of which is external.

Dutch seems to pose a problem to our proposal, as it exhibits a closed class of reflexives on the one hand (see 39a-c), and reflexives in ECM constructions on the other hand-see (40). The former is a characteristic oflexical reduction, while the latter is typical of syntactic reduction. This is unexpected under our account:

17 Reciprocal verbs in Hebrew are also lexical products, but they can be derived from a transitive entry by reduction of the dative argument; still, the operation also eliminates their accusative Case (see Siloni 2001). If this is the case, then lexical reduction of dative arguments is not in principle impossible. This also suggests that lexical reduction operations (or more generally, valency reducing operations) entail accusative Case reduction.

Against the Unaccusative Analysis (39)

(40)

I79

a.

Max wast zich. Max washes ZICH b. *Max haat zich. Max hates ZICH c. *Max hoorde zich. Max heard ZICH Maxhoorde [zich zingen]. Max heard ZICH sing

However, as shown by Reinhart and Reuland (I993), zich ('him/herself') is licensed in two distinct syntactic environments: with reflexive predicates and with nonreflexive predicates. When the predicate is reflexive, zich is licensed in a local context. When the predicate is not reflexive, zich cannot be bound by a co-argument; it is, then, what is often called a long-distance anaphor. Following Everaert (I986), Reinhart and Reuland (I993) argue that predicates such as wast in (39a) are inserted as reflexive predicates. 1hey constitute a closed class of elements, as shown by the impossibility of (39b--c). In our terms, they are the products of lexical reduction. 1he predicate hoorde does not belong to the class of predicates that allow lexical reduction. Hence, zich is ruled out in (39c). Still, it can appear in (40), although the predicate is not reflexive because it is the subject of the ECM complement, not a 8-argument of the matrix verb. Hence, it is not syntactically bound by a co-argument but by an argumenr of a distinct predicate, the embedding predicate hoorde. Sentence (40), which superficially seems to be a structure analogous to the French example (33a) is, in fact, an instance of syntactic binding, not the output of LF-reduction as is its French counterpart. Dutch simply uses the same element (zich) in both contexts: when lexical reduction tal<:es place and in the context of non-local syntactic binding!8

6.7.

SUMMARY

We have argued that the unaccusative analysis of reflexive verbs must be discarded, as the subject of reflexives does not pattern with internal arguments. Reflexives are unergative entries whose subject is an external argument, unlike the subject of unaccusatives. We adopt the view that reflexive verbs are derived from their transitive alternate by a reduction operation that reduces the internal argument provided that it is identified with its external co-argument. Further, we suggest that reduction has two ,8 Note that we assume with Reinhart and Reuland (1993) that complex predicate formation at LF is an optional operation. The formation of a complex predicate in (40) would have the annoying result of making the two relevant arguments co-arguments, in contradiction with what the relevant binding condition requires. We do not address here the long-standing mystery of why zich cannot appear in the direct object position of a small clause, although it can be embedded under a PP.

180

Tanya Reinhart and Tal Siloni

modes of application: a lexical mode and a syntactic one. This is what makes reflexives show somewhat different characteristics in Hebrew-type languages vs. French type-languages. In our view, then, reflexivization is essentially the same phenomenon cross-linguistically. The distinctions between the two types of reflexive-in Hebrew and French, for instance-follow from the different component of grammar in which reduction applies.

7 Unaccusatives and Anticausatives in German Markus Steinbach

7.r.

INTRODUCTION

Although German does not belong to the class of split-ergative languages, it has, like many other Indo-European languages, two types of one-place predicate which differ in several respects. Perlmutter (1978) calls them unergative and unaccusative (cf. also Burzio 1986; Zaenen 1988; Levin and Rappaport Hovav 1995). These differences, as discussed in Grewendorf (1989) and Fagan (1992), for example, concern, among other things, auxiliary selection, participles used as attributes, -er-nominalization, VPtopicalization, impersonal passives, and stress assignment. 1 The first three differences between unergative and unaccusative verbs are illustrated by the following examples. Consider auxiliary selection first. Unergative verbs, like most two-place verbs, form the present perfect with the auxiliary haben ('have'). Unaccusative verbs use the auxiliary sein ('be'). This is illustrated in (I). We would like to thank Artemis Alexiadou, Hans-Martin Gartner, Juliane Mock, Christian Richard, Wolfgang Sternefeld, RalfVogel, and the organizers and the audience of the Workshop on Unaccusativity in Berlin. I For a more detailed discussion of the distinction between unergative and unaccusative verbs in German, we refer the reader to Haider (1985, 1993); Wunderlich (1985); Grewendorf (1989); Ehrich (1992); Fagan (1992); Eisenberg (1999); Primus (1999); Wanner (1999); Abraham (2000).

Markus Steinbach

I82

(I)

a.

Clara hat geschlafen. (Unergative) Clara has slept b. Clara ist errotet. (Unaccusative) Clara is blushed 'Clara has blushed.'

Second, (2C) and (2d) illustrate that a noun modified by the present participle of a two-place verb such as lesen ('read') is linked to the first argument of this verb, which is realized as the subject in the corresponding active sentence. By contrast, a noun modified by the past participle gelesene ('read') is linked to the second argument, which is realized as the direct object. Examples (2a) and (2b) show that only the unergative verb schlafen ('sleep') patterns with two-place verbs in this respect. Again, unaccusatives differ from unergatives. The noun corresponding to the sole argument of the unaccusative verb en';Hen ('blush', lit. 'redden') can be modified, not only by the present participle, but also by the past participle. (2)

a.

b. c. d.

die schlafende/*geschlafene Clara (Unergative) the sleeping/slept Clara die errotende/errotete Clara (Unaccusative) the blushing/blushed Clara die lesende/*gelesene Clara (Le. Clara reads the book) the reading/*read Clara das *lesende/gelesene Buch (Le. Clara reads the book) the *reading/read book

Thc third difference concerns -er-nominalization, which has a variety of interpretations in German (c£ e.g. Meibauer I995). The crucial difference between unergatives and unaccusatives can be observed with -er-nominals that refer to persons (Nomina agentis). Only the -er-nominals in (3a), which are derived from unergative verbs, can refer to the person performing the action described by the verb. This interpretation is impossible for the -er-nominals in (3b) derived from unaccusative verbs. (3)

a.

Schlafer, sleeper b. *Erroter, *blusher

Sprecher, Denker (Unergative) speaker thinker *Ankommer, *Erwacher (Unaccusative) *arriver *awaker

Unaccusative verbs differ from unergative verbs at least in their semantics and selectional properties. Typical unaccusative verbs select a semantic argument which has proto-patient properties as opposed to typical unergative verbs, which select a semantic argument with proto-agent properties (c£ Dowty I99I). Following Grimshaw (1990), we assume two different lexical entries and call the sole semantic argument of the unergative verb in (4a) external semantic argument and the one of the unaccusative verb in (4b) internal semantic argument! On the syntactic side, matters are less 2

Note that the picture might be more complex. According to Dowty (1991: 607), one-place verbs might

Unaccusatives and Anticausatives in German clear. An additional syntactic distinction must be syntactically motivated. It is still an open question whether unaccusative verbs in German also differ from their unergative counterparts in their syntactic representation, as, for example, argued in Grewendorf (1989).3 For unaccusatives two possible D-structure representations are at hand. Either they share the syntactic representation (4c) with unergatives or the sole argument of un accusatives, unlike that of unergatives, is linked to the complement position ofVo (vp,Spec is either empty or does not project), as is illustrated in (4d). Thus, the basic question is whether in German subjects of unaccusative verbs are base-generated in object position or whether the subjects of unergative and unaccusative verbs are basegenerated in the same position-VP,Spec, for instance. This is illustrated in (4C).4

a.

b. c.

d.

Schlafen (x) (Unergative) Erroten «x» (Unaccusative) [yp NP [Va]] [yp [NP Va]]

The diagnostics for unaccusativity in German yield no clear evidence for the claim that unaccusativity in German must also be encoded in syntax (cE, for example, the controversial discussion in Grewendorf 1989 and Fagan 1992). A purely syntactic explanation of the differences between unergative and unaccusative verbs is not very convincing, since in principle these differences can be traced back to the lexical distinction between unergative and unaccusative verbs. In addition, the diagnostics for unaccusativity do not always yield a clear classification of one-place verbs and in most cases additional intervening factors must be considered (such as aspectuality in the context of auxiliary selection and modifying participles, or information structuring in the context ofVP-topicalization).5 differ in the degree of unaccusativity and unergativity. 'The "cut" between unergative and unaccusative arguments is indeterminate, vatying, I argue, according to the same parameters.' There might be a continuum from typical unaccusative verbs (selecting an argument with many proto-patient ptoperties) to typical unergative verbs (selecting an argument with many proto-agent properties). 3 The same problem arises for all languages that distinguish between unergative and unaccusative oneplace predicates (cf. the discussion in Dowty 1991). " 4 In a more elaborate framework such as that developed in Hale and Keyser (1993), for example, unaccusatives have less 'VP-structure' than unergatives. Unaccusatives only project the simple basic VP in (ii). The internal argument is base-generated in VP,Spec and moves to IP,Spec to receive or check nominative case. In contrast with unaccusatives, unergatives contain an additional (functional) vP. The external argument of unergatives is either base-generated in IP,Spec or in vP,Spec, as (0 illustrates (see also Radford 1997, Grewendorf 2002, and Alexiadou and Anagnostopoulou, in this volume). For a different kind of syntactic implementation of unaccusativity, which is compatible with the analysis of unaccusatives proposed here, see van Hout, in this volume.

(i) (i0

Lp NP VO [yp VOl] [vp NP [V0]]

5 In German, the tests do not provide a uniform picture. Take, for instance, the -er-nominal Diufer ('walker'). Laufir can refer to a walking person, although it is derived from the unaccusative verb taufin ('walk'), which selects the auxiliary sein ('be'). Another example is impersonal passive, which is predicted to be ungrammatical with unaccusative verbs. Nevertheless, whereas some unaccusative verbs can be passivized, some unergatives cannot (cf. also Nerbonne 1982). In addition, VP-topicalization is not completely

Markus Steinbach Here I do not re-address the validity of the respective diagnostics. Instead, I will assume that German distinguishes unergative from unaccusative one-place predicates only in the lexicon, as (4./l) and (4b) illustrate. I will mainly investigate a closely related question, namely, the derivation of anticausatives, which might shed some new light on the syntax of unaccusatives. German not only has two different types of one-place verbs, but also two different kinds of anticausatives. As opposed to the syntactically simpler form in (5a), the more complex form in (5b) obligatorily selects a reflexive pronoun in object position (SICH stands for the reflexive pronoun).

(5)

a.

b.

Der Stock bricht. (Non-reflexive anticausative) the stick breaks 'The stick breaks.' Der Stock biegt sich. (Reflexive anticausative) the stick bends SICH 'The stick bends.'

Although neither verb in (5) assigns an external 8-role, only the first one is unaccusative. In contrast with the non-reflexive anticausative in (5a), the reflexive anticausative in (5b) is syntactically transitive with a reflexive pronoun in the position of the accusative object. In the next section I argue that reflexive anticausatives are normal transitive reflexive sentences (TRS) in German. Reflexive anticausatives are basically two-place causative verbs, which permit reduction of the first semantic argument. In addition to the anticausative interpretation in (5b), TRSs also receive a reflexive, a middle, and an inherent reflexive interpretation. The analysis of TRSs (including reflexive anticausatives) sheds new light on the syntactic representation of non-reflexive anticausatives and unaccusatives in general. The syntax of anticausatives thus provides new insight into the syntax of unaccusatives. Non-reflexive anticausatives such as (5a), as we will see, are the connecting link between such TRSs as (5b) and unaccusatives such as (Ib). This chapter is organized as follows. The next section provides a brief overview of the basic properties of anticausatives and embeds reflexive anticausatives in the wider context of other TRSs. It is shown that TRSs are systematically ambiguous and that the different interpretations of a TRS do not differ in syntax. Finally, the main shortcomings of syntactic and lexical explanations of (some) TRSs are briefly discussed.

excluded for unergatives (cf. Haider 1985 and 1990 and Fragan 1992) and auxiliary selection depends at least partly on the aspectual interpretation, which is closely related to the internal argument, that is, the proto-patient (c( Dowty 1991). (i)

Peter hat getanzt/ist in den Saal getanzt Peter has danced/is into the hall danced

Another problem for these diagnostics is that many of the relevant data cannot clearly be judged (un-)grammatical. Note, finally, that Jacobs (1993) develops a theory of stress assignment which refers to the semantic properties of arguments and makes use of the notion of proto-patient. See also Fagan (1992), who argues that the diagnostics for unaccusativity in German are not reliable, and the discussion in Dowty (1991).

Unaccusatives andAnticausatives in German Section 7.3 argues for a different and, until now, new way of deriving TRSs. The semantic ambiguity ofTRSs is derived at the interface between syntax and semantics. The accusative reflexive pronoun is systematically ambiguous between an argument and a non-argument interpretation. The latter indicates valency reduction and mediates the linking of the subject to the internal semantic argument of the verb. This approach offers a uniform analysis of the ambiguity ofTRS. In section 7.4 we turn to unaccusatives again. It follows from the analysis of reflexive anticausatives that a syntactic derivation is not available for non-reflexive anticausatives and unaccusatives in general. Otherwise, the differences between non-reflexive and reflexive anticausatives and the systematic ambiguity ofTRSs illustrated in the next section remain unexplained. Therefore, our analysis of anticausatives supplies an argument against a syntactic derivation of unaccusatives. Unaccusatives and unergatives share the same syntactic representation given in (4c). The last section briefly summarizes the main findings.

7.2.

REFLEXIVE ANTICAUSATIVES AND OTHER

TRSs

Recall that German, like many other Indo-European languages, has two different types of anticausative-reflexive and non-reflexive (c£ also Alexiadou and Anagnostopoulou, in this volume). Non-reflexive anticausatives such as brechen ('break') in (6b) equal unaccusative verbs such as erriHen in (I b) . The sole argument of the nonreflexive anticausative is an internal argument, which is linked to the accusative object in the causative counterpart in (6a). Non-reflexive anticausatives must not have a reflexive pronoun in the position of the accusative object. In contrast, reflexive anticausatives such as biegen ('bend') are ungrammatical without a reflexive pronoun in object position, as (7b) illustrates.

(6)

a. b. c.

(7)

a. b. c.

Hans-Martin bricht den Stock Hans-Martin breaks the stick Der Stock, bricht (*sich,) the stick breaks S1CH Further examples: rollen ('roll'), fliegen (,fly), trocknen ('dry), zerbrechen ('smash'), zerknittern ('crumple'), abbrechen ('break off), einfrieren (,freeze'), auftauen ('thaw'). Hans-Martin biegt den Stock Hans-Martin bends the stick Der Stock, biegt *(sich,) the stick bends SICH Further examples: offnen ('open'), schlie6en ('close'), flillen ('fill'), leeren ('empty'), verdunkeln ('darken'), erhellen ('light up'), drehen

Markus Steinbach

IS6

('turn), aufwarmen ('warm up'), aufldaren ('solve'), falten ('fold'), glatten ('smooth'), vergro~ern ('enlarge'), verldeinern ('reduce') In (6) and (7) the syntactic subject of the anticausative corresponds to the accusative object of the causative counterpart. Hence, both the subject of the anticausative and the object of the causative are linked to the internal semantic argument of the verb. The resulting interpretation of both kinds of anticausative is a one-place predicate selecting only an internal argument. However, this interpretation relates to different syntactic and lexical representations. At first sight, the non-reflexive anticausative in (6a) is expected to be the optimal realization of a one-place unaccusative verb. In contrast with non-reflexive anticausatives, their reflexive counterparts are syntactically transitive with a reflexive pronoun in the position of the accusative object, although the resulting interpretation contains only one semantic argument. Interestingly, the anticausative interpretation is not the only possible interpretation for transitive reflexive sentences. In addition to the anticausative interpretation, TRSs such as (5b) can also yield a reflexive, a middle, and an inherent-reflexive reading. These four interpretations ofTRSs are illustrated in (S). (S)

a.

b.

c.

d.

Ralf rasiert sich. (Reflexive interpretation) Ralf shaves SICH 'Ralfis shaving.' Dieser Roman liest sich gut. (Middle interpretation) this novel reads SICH well 'This novel reads well.' Die Tiir offnet sich. (Anticausative interpretation) the door opens SICH 'The door opens.' Ralf erkaltet sich. (Inherent-reflexive interpretation) Ralf catches.a.cold SICH 'Ralf is catching a cold.'

The ambiguity ofTRSs in German is not accidental. The reflexive pronoun in (S) can, but need not, be linked to a semantic argument. It is ambiguous between an argument interpretation in (Sa) and a non-argument interpretation in (Sb-d). In the latter case it indicates valency reduction. The reflexive pronoun in (Sa), which is linked to a semantic argument, is called argument-reflexive. The reflexive pronouns in (Sb-c), which induce valency reduction, are called non-argument-reflexives. 6 This ambiguity of (weak) reflexive pronouns can be found in many Indo-European languages.? What 6 The argument-reflexive is also called 'echt reflexiv', 'reflexive sich', and 'anaphorical sich'; the nonargument reflexive, 'unecht reflexiv', 'medial sich', 'lexical sich', 'non-thematic reflexive', and 'Fiigungs-sich' (cf. Haider 1985 and Abraham 1995). Fagan (1992) calls the argument-reflexive 'referential' and the nonargument-reflexive 'nonreferential'. 7 In contrast with most Indo-European languages, German does not distinguish weak from strong reflexive pronouns. German has only one form (mich 'me, myself', dich 'you, yourself', sich 'himself, herself, itself, themselves', UllS 'us, ourselves', and euch 'you (pl.), yourSelves'), which corresponds to both weak

Unaccusatives and Anticausatives in German every analysis of reflexive anticausatives has to take into consideration is the systematic ambiguity ofTRSs. This section and the next section focus exclusively on the syntax and semantics ofTRSs (especially on the ambiguity of the reflexive pronoun). 8

7.2.1. The thematic ambiguity ofTRSs The ambiguity ofTRSs results from two different steps of interpretation. First, the reflexive pronoun in TRSs can, but need not, be linked to an argument of the verb. This ambiguity is illustrated in (9). (9)

Argument interpretation vs. valency reduction TRS argument reflexive ~ Reflexive interpretation ~ non-argument reflexive ~ Valency reduction

The schema in (9) shows that the reflexive pronoun can be linked to the second (or internal) argument of the verb, resulting in the reflexive interpretation in (8a). The structure in (10) is a simplified semantic representation of sentence (8a). The reflexive pronoun is linked to the second semantic argument and it is coreferent with the first argument of the verb-r is an individual type-variable and stands for Ralf. R stands for the two-place predicate rasieren ('shave').

R (r r,) (reflexive interpretation =8a) " Alternatively, the reflexive pronoun is not linked to an argument variable of the verb, as can be seen in (8b-d). In this case it indicates valency reduction. If the reflexive pronoun is not linked to a semantic argument itself, it mediates the linking of the subject to the second semantic argument. The first semantic argument of the verb, which is not linked to syntax, can either be saturated or reduced. Argument saturation yields the middle interpretation and argument reduction the anticausative and inherent reflexive interpretations. 9 (10)

(II)

Argument saturation vs. argument reduction Argument saturation ~ Middle interpretation . Anticausative interpretation Valency reduction ~ Argument reduction ~ Inherent reflexive interpr.

and strong forms in other languages. The German reflexive pronoun can be focused and coordinated, for example (a typical property of strong forms) and it must be used with body-part reflexives, in middle constructions, and inherent-reflexive verbs (a typical property of weak forms); see Steinbach (2002b) for a detailed discussion. See also n. IS. 8 A long-standing problem is the adverbial modification that middle constructions usually require. Iwata (1999) and Steinbach (200M, b) argue that the 'adverbial effect' is not crucial for the grammaticality of middle constructions because it 'is not so much a matter of syntactic subcategorization requirements as one of informativeness' (Iwata 1999: 528). The 'adverbial effect' can be derived from conditions on assertions to be pragmatically licensed. 9 Argument reduction and argument saturation can be defined in the foIIowingway-cf. also Chierchia (in this volume) for operations on semantic arguments ('Op' stands for operaror; see n. 10): Saturation: (AY P (x,y»)(a) --7 (AY Op(x) P (x,y»)(a) Reduction: (AY P (x,y»)(a) --7 (AY P (y»)(a)

188

Markus Steinbach

In contrast with sentence (8a), which contains an argument-reflexive, sentences (8b), (8c), and (8d) contain a non-argument-reflexive. Saturation of the first (or external) argument results in the middle interpretation in (12a). The first argument is implicitly present and bound by a generic operator (which also binds the situation variable in middle constructions; we return to this later). Saturation does not change the semantic adicity of the predicate. IO Reduction yields the anticausative and inherent-reflexive interpretation in (12b) and (12C), but now the first argument is completely removed from the semantic representation. The outputs in (12b) and (12C) are one-place predicates. (The variable x in (12a) is the implicit argument of the middle construction bound by the generic operator GEN; 0 stands for the reduced first argument in anticausatives and inherent reflexives.)"

(12)

a.

b. c.

GEN(x) L (x, r) (Saturation: middle interpretation = 8b) o (0,t) (Reduction: anticausative interpretation = 8c) E (0,r) (Reduction: inh. refl. interpretation = 8d)

In the next section, I argue that argument saturation is less restricted than argument reduction; middle constructions are therefore much more productive than anticausatives. In German, middle constructions are grammatical with nearly allldnds of verb (c£ Fagan 1992 and Steinbach 2002b).I2 Unlike English, German also has impersonal middle constructions, which are derived from one-place verbs. Unergatives such as flirten (,flirt') in (13a) and un accusatives such as sterben ('die') in (13b) can both undergo middle formation. Impersonal middle constructions are also transitive and reflexive and they require an impersonal subject-the third-person neuter personal pronoun es. In this respect they differ from impersonal passives in German, which must not have an impersonal subject. '3 W The free argument variable in middle constructions must be bound by some operator. Implicit arguments can either be bound by the existential quantifier or by the generic operator. The latter triggers a 'global' generic interpretation, which covers all human beings in the universe of discourse, c( below. As opposed to languages like Russian or Modern Greek, German shows a division of labour with respect to the interpretation of the implicit first argument: whereas passives are responsible for the existential interpretation, the generic interpretation is associated with middle constructions. The 'modal' interpretation and the adverbial modification, which are typical for middle constructions, can be derived from genericity (cf. Steinbach 2002b) .. It Like anticausatives, inherent-reflexive verbs are derived from an underlying two-place predicate. However, they differ from the former in allowing only the anticausative interpretation. Their first argument must be reduced. I do not want to discuss here whether this restriction on inherent reflexives can be derived from their semantics or whether inherent reflexives are listed in the lexicon as idiomatic expressions. 12 Individual-level predicates such as abstammen ('be descended from) or heifen ('be called') and zeroplace predicates are exceptions. The former cannot undergo middle formation because they do not provide an event variable that can be bound by the generic operator in middle constructions (c( below). Zero-place predicates are excluded because they do not select an argument variable at all. I) In impersonal passives the third person neuter pronoun es is only grammatical in sentence-initial position of matrix clauses. In contrast with the pronoun in impersonal passives, the pronoun in impersonal middle constructions is a genuine (impersonal or pleonastic) subject that can also occur in the middle field.

(i)

a.

. .. wei! (*es) gestern viel gelacht wurde. (Impersonal passive) ... because it yesterday much laughed was ' ... because yesterday people were laughing a lot.'

Unaccusatives and Anticausatives in German a.

(13)

b.

Mit blauen Augen flirtet es sich leichter. (Unergative) with blue eyes flirts it SICH more. easily 'With blue eyes, flirting is easier.' Gesundheitsstudie: In welchem Bezirk stirbt es sich am friihesten. study.on.health: in which district dies it SICH at. the earliest 'Study on health: in which district do people die youngest.' (Unaccusative)

Another important restriction concerns dative reflexive pronouns, which cannot indicate valency reduction (cf. also Fagan 1992 and Abraham 1995). The subject of a middle construction or anticausative cannot correspond to the dative object of the 'active' or causative counterpart. Dative reflexive pronouns are always interpreted as semantic arguments of the verb-the non-argument reading is not available. Therefore, sentence (14a) can only yield a reflexive interpretation. Nevertheless, verbs selecting a dative object can undergo middle formation in principle. In the impersonal middle construction in (14b), the dative object preserves its case and it is again the accusative reflexive pronoun that indicates valency reduction. I4

a.

(14)

b.

hilft sich doch leicht. Ein Obdachloser a homeless.person.NoM helps SICH.DAT PARTICLE easily 'A homeless person helps himself easily.' (Reflexive interpretation) *'You can help a homeless person easily.' (*Middle interpretation)

. .. wei!

*(es) sich in diesem Bett gut schliift. ... because it SICH in this bed well sleeps ' ... because you sleep well in this bed.'

(Impersonal middle)

'4 Maling (2001) argues that dative case is not to blame. She defines the relevant restrictions in terms of thematic roles: goal arguments cannot be promoted to subject. In German, marked (or oblique) case forms such as dative crucially differ from structural case forms (cf.n. 29). The difference between structutal and oblique case-forms in middle formation is nicely illustrated by the minimal pair in 0) and (H). The meaning of the German verbs trejfen (,meet') and begegne/l ('meet') is almost identical, but they do not assign the same case to their objects. Trejfen assigns accusative and begegnen dative case. Not surprisingly, only the accusative object of trejfen can be promoted to subject in middle constructions. Hence, 0) is ambiguous between a middle and a reflexive interpretation. The corresponding sentence in (ii), containing the dative verb begegnen, only receives a reflexive interpretation. A middle interpretation is only possible in the impersonal middle construction in (Hi). Further arguments for an analysis based on the distinction between structural and oblique case forms can be found in Steinbach (2002b).

0)

Nette Menschen treffen sich hier liberal!. nice people.NoM meet SICH here everywhere 'You can meet nice people allover this place.' (Middle interpretation) 'Nice people meet all over the place.' (Reflexive interpretation)

(H)

Nette Menschen begegnen sich hier liberal!. SICH.*ACe/DAT here everywhere nice people.NoM meet 'Nice people meet allover the place.' (Only reflexive interpretation)

(Hi)

Netten Menschen begegnet es sich hier liberal! nice.DAT people.DAT meets it. NOM SICH here everywhere 'You can meet nice people everywhere.' (Middle interpretation)

Markus Steinbach

b.

leicht. Einem Obdachlosen sich doch hilft es a.DAT homeless.person.DAT helps it. NOM SICH PARTICLE easily 'You can help a homeless person easily.' (Middle interpretation)

Note that German draws a clear morphological distinction between accusative and dative reflexive pronouns in first and second person singular.I5 Like the third-person reflexive pronoun, the first- and second-person reflexive pronouns are also ambiguous, but once again only in the position of the accusative object. Example (15a) is a middle construction, example (15b), an anticausative construction. (15)

a.

b.

Du verkaufst dich gut- ich meine, dein Buch verkauft sich gut REFL.2SG well I mean your book sells SICH well you sell 'You sell well-I mean, your book sells well.' Ich beruhigemich schon wieder. I calm. down REFL.ISG again 'I will calm down again.'

In German, only accusative reflexive pronouns are ambiguous between an argument and a non-argument interpretation. Moreover, the ambiguity of the accusative reflexive pronoun is restricted to the position of the accusative object. Hence, accusative and dative reflexive pronouns in prepositional phrases are also unambiguous, like the dative reflexive pronoun in (14a). They are always linked to a semantic argument of the verb, thus leading to a reflexive interpretation. (16)

a. b.

Ralf legte das Buch neb en sich. Ralf put the book next. to SICH.ACC Ralf erzahlte den Kindem von sich. Ralf told the children about SICH.DAT

The observations made so far are summarized in (17). Following related phenomena in other languages, I call the reflexive pronoun in the position of the accusative object a 'middle marker' (cf. Kemmer 1993). Unlike other Indo-European languages, German '5

(i)

This is illustrated by the paradigm in (i). Accusative and dative reflexive pronouns in German

Accusative Dative

1sg

2sg

3sg

1pl

2pl

Wi

mich mir

dich dir

sich sich

uns uns

euch euch

sich sich

The first- and second-person singular are clearly morphologically marked for accusative and dative case. The assumption that reflexive pronouns in German are not assigned case, as proposed, for example, in Muller (1993), cannot be maintained. For further discussion of Muller's analysis, see Vogel and Steinbach (1998) and Steinbach (2002b). Note that dative reflexive pronouns developed very late whereas non-argument interpretations of the accusative reflexive pronoun can already be found in early stages of German. Unlike accusative reflexive pronouns, it is not until the beginning of the eighteenth century that the modern usage of dative reflexive pronouns seems to have been established. At earlier stages (as well as in some dialects of Modern German) no morphological distinction between personal and reflexive dative pronouns can be found. This observation is in line with the analysis developed here.

Unaccusatives and Anticausatives in German has no morphosyntactic middle marker (in the strict sense of the word), but a free morpheme in a specific syntactic position. Nevertheless, the accusative reflexive pronoun in object position shows the same ambiguity as middle markers in other languages. I6

(17)

Middle marker and reflexive marker in German Middle marker Inherent reflexive interpretation Anticausative interpretation Middle interpretation Reflexive interpretation

Reflexive interpretation

ACC-RP

DAT-RP;P+ACC/DAT-RP

Reflexive marker

The schema in (ll) captures the fact that there is a strong correlation between reflexive pronouns and middle markers. However, reflexivity cannot be reduced to middle voice because not every reflexive marker is also a middle marker. The middle marker in German has two essential properties: it must be a reflexive pronoun and it must occupya specific syntactic position, the position of the accusative object.

7.2.2. The syntax ofTRSs We showed that TRSs have four different readings. As will become clear in this section, these interpretations do not differ in syntax. Both the argument- and the nonargument-reflexives obey the same restrictions on word order in the middle field, which is restricted by (the interaction of) various constraints (which we cannot discuss in great detail here). For present purposes it is only important to mention that these restrictions apply to argument- and non-argument-reflexives in exactly the same way. This can be seen in the following examples for main and subordinate clauses.I?

(r8)

a. Die Tur offnet sich schnell. (Non-argument-reflexive) b. *Die Tur offnet schnell sich the door opens quicldy srCH 'The door opens quicldy.'

,6 In Indo-European languages, middle markers have quite different morphosyntactic properties. Possible middle markers are verbal inflection, a verbal affix, a verbal elitic, or a weak reflexive pronoun. Typical interpretations for a middle marker are, among others, passive, middle, anticausative, detransitive, and reflexive. We come back to this issue in the next section. For Greek, see Alexiadou and Anagnostopoulou, in this volume; for Italian, Chierchia, in this volume. '7 One constraint regulates the ordering of nominative and accusative: nominative precedes accusative in the unmarked order. A second constraint states that pronominal elements tend to be right-adjacent to Co, the so-called Wackernagel position. Further constraints refer to information structure, thematic roles, and animacy. A more detailed discussion of the relevant constraints can be found in Lenerz (I977); Reis (I987); Uszkoreit (I987); Fagan (I992); Vogel and Steinbach (I998); Milller (I999); Gartner and Steinbach (2000).

Markus Steinbach c. Ralf rasiert sich schnell. (Argument-reflexive) d. *Ralf rasiert schnell sich Ralf shaves quieldy SICH 'Ralf shaves quicldy.' (19)

a. b. c.

d. (20)

· .. wei! sich die Tiir schnell offnet. (Non-argument-reflexive) · .. wei! die Tiir sich schnell offnet. because the door SICH quicldyopens · .. wei! sich Ralf schnell rasiert. (Argument-reflexive) · .. wei! Ralf sich schnell rasiert. because Ralf SICH quieldy shaves

a. · .. wei! sie sich schnell offnet. (Non-argument-reflexive) sich sie schnell offnet. b. *... wei! because SICH it quicldyopens (it = the door) · .. wei! er sich schnell rasiert. (Argument-reflexive) sich er schnell rasiert. d. *... wei! because SICH he quieldy shaves c.

These examples illustrate that the reflexive pronoun is neither a elitic nor incorporated into the verb, regardless of its interpretation. Besides, Gartner and Steinbach (1997, 2000) show in detail that German has no syntactic (or special) elitics. Note finally that both argument- and non-argument-reflexives differ from elements that appear preferably adjacent to the verb, such as Karten in Karten spielen ('play cards') or davon in davonschleichen ('sneak off). (21)

a. . .. wei! Heidi sich oft davongeschlichen hat. b. *... wei! Heidi sich davon oft geschlichen hat. because Heidi SICH off often sneaked has ' ... because Heidi often sneaked off.' c. *... wei! die Tiir oft sich offnet. d. . .. wei! die Tiir sich oft offnet. because the door SICH often opens

In contrast to such verb-object combinations, reflexive pronouns need not be adjacent to the main verb. So far there is no empirical evidence to tell the non-argumentreflexive from the argument-reflexive in syntax. However, there are some differences between argument- and non-argument-reflexives. Only the former can be coordinated, fronted, questioned, bear narrow focus, or occur in the scope of a focus-sensitive operator. These differences are illustrated in (22) by means of focus. (22)

a.

Ralf rasiert [SICHh Ralf shaves SICH b. *Die Tiir offnet [SICHh the door opens SICH

Unaccusatives andAnticausatives in German

I93

Although non-argument-reflexives are not completely identical to argument-reflexives, examples such as (22) do not provide a basis for a special syntactic status of nonargument-reflexives. These differences are clearly due to the different semantics of argument- and non-argument-reflexives and can be best explained by independently needed theories of focus, fronting, and coordination, as has been argued by Fagan (I992) and Steinbach (2002b). Consider again example (22). Recent theories offocus divide the semantic representation of a sentence into two parts (c£ Stechow 1991, for instance). The first part corresponds to the focus, the second part to the background. The background results from replacing the focus of the clause by a variable. The semantic background (or, according to Rooth 1992, the focus-semantic value of a sentence) is a set of alternatives to the ordinary semantic value of that sentence. In the next section I argue that the reflexive pronoun is not linked to a semantic argument variable in middle constructions and anticausatives. k a consequence, no focus-background structure can be generated for sentence (22b). Hence, a sentence with narrow focus on a non-argument-reflexive is ungrammatical. The argument-reflexive, on the other hand, is linked to the second semantic argument. SimplifYing somewhat, the focus-semantic value of the reflexive pronoun in (22a) is a set of individuals {Ral£ Hans, Martin, etc.} and the focus-semantic value of the whole sentence is the set of propositions given in (23). (23)

1/ [s Ralf rasiert sichp] W= {R (r (x» I x E E}, E the set of individuals = the set of propositions of the form 'Ralf is shaving x'

This strategy can also be applied to the other points in question. Focus-sensitive operators bind the focus of a sentence or constituent. The focus-sensitive operator only, for example, presupposes the ordinary semantic value of the sentence or constituent and asserts that no other alternative from the set of alternatives is true. Again, the nonargument-reflexive cannot be bound by a focus-sensitive operator because no focussemantic value can be generated. Questions are yet another example. In semantics, questions can be analysed as sets of possible answers (Karttunen 1977). The meaning of a question corresponds to the focus-semantic value of its answer. According to Biiring (1997), a sentence S can be uttered as an answer to a question Q (given a common ground CG) if the focus-semantic value of S is identical to the meaning of the question Q. kking for a non-argument-reflexive pronoun is again impossible because it does not have a focus-semantic value. Let us complete this discussion with a final remark on coordination. IS ,8 Fronting is a little bit more complex. In German, a constituent can occupy the sentence-initial position if it is (a) the first argument in the unmarked word order, (b) the focus, (c) the topic, or (d) linked to the text or discourse. Note that accusative objects are always marked in sentence-initial position. Condition (b) also excludes non-argument-reflexives, because they cannot be focused. Neither can they be linked to the discourse since they must be bound sentence-internally. Hence, condition (d) does not apply. Reflexive pronouns can only be sentence-internal topics (S-topics). Biiring (1997) analyses S-topics parallel to focus: they induce alternatives similar to the focus. This means that the ungrammaticality of fronting non-argument-reflexives can also be explained on the basis of the impossibility of focusing them. For a more detailed discussion see Steinbach (2002b).

Markus Steinbach

194

(24)

a.

Ralf rasiert sich und seinen Onkel. uncle Ralf shaves SICH and his 'Ralf is shaving himself and his uncle.' b. *Die Tilr offnet sich und das Fenster. the door opens SICH and the window

Coordination requires syntactic and semantic identity of the conjuncts (c£ e.g. the zeugma-effect). Hence, a reflexive pronoun that is coordinated with another NP (in our example seinen Onkelor das Fenster) must also be interpreted as a semantic argument of the verb. Otherwise the second conjunct cannot be linked to a semantic argument and does not receive a thematic interpretation. '9 To sum up, the different meanings ofTRSs do not correspond to different syntactic representations. In syntax there is only one kind of accusative reflexive pronoun that is semantically ambiguous between an argument and a non-argument reading. This semantic ambiguity is responsible for the fact that only the argument-reflexive can be questioned, focused, fronted, and coordinated.

7.2.3. Syntactic and lexical approaches Nearly all approaches derive TRSs in the lexicon and/or in the syntax. While most linguists agree that anticausatives should be derived in the lexicon, it is less clear whether middle constructions are to be derived in the lexicon by a lexical rule of middle formation or in the syntax by means of A-movement. Both types of analysis somehow manipulate the selectional properties of the underlying verb. This section shows that neither of these analyses offers a uniform and conclusive derivation of middle constructions and, more generally, of the ambiguity ofTRS. Recall from the previous section that argument- and non-argument-reflexives do not differ in syntax. Nevertheless, syntactic analyses of middle constructions as proposed in Haider (1987) and Schacht! (1991) assumes two different syntactic representations. The argument-reflexive is analysed as a complement of the verb (see 25a) whereas the non-argument-reflexive is base-generated as a VP-adjunct that 'absorbs' the external theta role (see 25b).20

19 The same holds true in sentence (i), which does not involve coordination of two NPs but, rather, of two sentences, one of which contains a gap in the position of the verb. Again, the antecedent and the deleted element must receive the same interpretation. However, the verb liffilen ('open') is interpreted differently in each conjunct. In the first conjunct, it is a two-place predicate (,Hans opens the window), and in the second, a one-place anticausative predicate ('the door opens'). Sentence (i) is thus correctly excluded.

(i)

*Hans offnet das Fenster und die Tilr sich. Hans opens the window and the doorslcH

W For syntactic analysis of English and Dutch middle constructions see Hale and Keyser (1993), Hoekstra and Roberts (1993), and Stroik (1992, 1999). These approaches are criticized in Zribi-Hertz (1993),Ackema and Schoorlemmer (1995), and Steinbach (2002b).

Unaccusatives andAnticausatives in German (25)

a.

b. c.

d.

195

Ralf rasiert sich. b Ralf b rasiert,] [yp sich t,JJ Die Tiir offnet sich. b Die Tiir [co offnet,] [yp sich [yp t,JJJ

Apart from the fact that this distinction is not empirically motivated, it is confronted with further problems. 2I Haider and Schacht! assume that the external theta role cannot only be assigned to VP,Spec (or IP,Spec) but also to a syntactic adjunct in A'position, which conflicts with the theta criterion. In addition to this, they have to stipulate that German has two different ways of assigning accusative case: either to an A-element in the complement position of the verb or to an PI. -element adjoined to VP. But this stipulation is only necessary to derive non-argument-reflexives. Besides, it is not sufficient to derive impersonal middles because one-place predicates cannot assign accusative case at all. Note that they must also distinguish two kinds of binding relation, which complicates binding theory. 22 According to Haider's analysis of non-argument-reflexives and auxiliary selection, we would expect unaccusative verbs either to be ungrammatical in middle constructions or to select the auxiliary sein ('be'). However, unaccusative verbs can undergo middle formation and that they 'select' the auxiliary haben ('have') in middle constructions (c£ section 7.2.1). Another problem concerns anticausatives. In anticausatives, the first argument is not implicitly present. Hence, no external e-role needs to be 'absorbed' in syntax by the non-argument-reflexive adjoined to VP. Anticausatives must be derived in the lexicon by an additional rule of argument deletion (c£ (26) below). Therefore, these syntactic accounts cannot offer a uniform analysis of the anticausative, the middle, and the reflexive interpretation ofTRSs-that is, the ambiguity of the accusative reflexive pronoun. It remains an open question why the middle marker is an accusative reflexive pronoun. 2 ) For a more detailed discussion of (further) shortcomings of syntactic analyses, see Fagan (1992), Zribi-Hertz (1993), Ackema and Schoorlemmer (1994), and Steinbach (2002b). Fagan (1992) and Bierwisch (1996, 1997)'argue for an alternative lexical analysis of middle constructions in German.>4 According to Bierwisch, the following principle accounts for the reflexive pronoun in middle constructions and reflexive anticausatives.

2> Note that this distinction conflicts with the observations made in the previous subsection. Moreover it does not help to explain the differences berween argument- and non-argument-reflexives mentioned above. These specific properties of non-argument-reflexives cannot be reduced ro A'-properties since adjuncts can in general be focused, fronted, questioned, or coordinated. " Both kinds of reflexive pronoun mUst be bound in syntax, but argument-reflexives are also semantically bound. '3 We already mentioned that syntactic analyses assume that the external role can be assigned to the subject position or alternatively to a VP-adjunct. It is, however, unclear why only reflexive pronouns have the property of 'absorbing' the external theta role in VP-adjoined position. According to Haider and Schacht!, in principle every VP-adjunct can receive accusative case and the external a-role. '4 For lexical analyses of English and Dutch middles see Ackema and Schoorlemmer (1994, 1995).

Markus Steinbach

196

(26)

Whenever an argument position Ax is directly dominated by a non-genuine argument position AY, Ax is assigned the feature [+REFL].

A non-genuine argument position does not bind a variable in the semantic form (SF) of the verb. Consider the following example, which is a simplified lexical entry of the reflexive anticausative verb offoen Copen'). (27)

Ojfnen: Ax Ay [(y CAUSE)

[BECOME [OPEN

x]]]

In the anticausative reading the argument position Ax does not bind a variable in SF (this is indicated by the parentheses). According to principle (26), the object position receives the feature [+REFL]. In this analysis the reflexive pronoun results from a valency reduction operation that only affects the SF of a verb. 2 ' However, this treatment of non-argument-reflexives cannot explain the occurrence of reflexive pronouns in impersonal middle constructions, because the underlying lexical entry has only one argument position, as can be seen in (28). Principle (26) cannot apply to such examples. (28)

lachen Claugh'): Ax

[LAUGH

x]

Lexical analyses are therefore forced to introduce an additional non-genuine argument position to feed principle (26). Otherwise, personal and impersonal middle constructions cannot be treated alike. That is why Bierwisch (1997) assumes that middle constructions have at least two lexical entries of their own; these templates are necessary to derive the core cases. Template (29a) is for personal middle constructions, (29b) for impersonal middle constructions derived from one-place predicates CN' stands for the generic implicit argument). (29)

Middle templates MT,: two-place verbs: [+M] AVAYAZ [VNy] MT2 : one-place verbs: [+M] AVAyAz [VN]

a. b.

With these two templates, both personal and impersonal middle constructions can be derived from underlying one- and two-place verbs. Insertion of the two-place verb lesen ('read') in MT, yields, for example, the following 'middle verb'. (30)

lesen [+v, -N, +M]: Ay Az [N READ y]

Nevertheless, an additional stipulation (namely, principle (31)) is necessary to prevent two-place predicates from being inserted in the 'impersonal' template MT" which would result in an ill-formed lexical entry!6

Note that iiffnen still subcategorizes two syntactic argument positions. The output is given in (i) and the corresponding sentence in (ii). (i) Ab Ay Az [b V NJ (ii) Ralf zeigt sich sich im Spiegel. Ralf shows SICH SICH in. the mirror 'Ralf shows himself to himself in the mirror: 25

26

Unaccusatives and Anticausatives in German (3I)

197

N must occupy the highest argument position of V

Another problem is that dative objects are expected to undergo middle formation because principle (26) is not sensitive to case. However, we mentioned in section 7.2 that dative reflexive pronouns cannot yield a non-argument interpretation. The same problem arises with three-place predicates such as the one in (32). Even if this lexical analysis assumes an additional middle template MT3 for three-place predicates, it cannot derive the correct lexical representation. According to principle (26), the dative object Ay dominated by the non-genuine argument position receives the feature [+REFL] in (32a). But sentence (32b) shows that it is again the accusative reflexive pronoun that indicates valency reduction. (32)

a.

b.

Ax-ACC Ay-DAT Az-NOM [CAUSE N [BECOME [pass (y,x)]]] weil sich sliGer Hustensaft kleinen Kindem because S1CH sweet coughing.syrup.NOM small children.DAT besser einfloGt better pour.into

I complete the discussion with two basic remarks. Fiist, the template analysis is forced to assume at least two different lexical entries in order to derive personal and impersonal middle constructions parallel to anticausatives. Second, like syntactic analyses, it does not offer a uniform analysis ofTRSs. The same holds true for Fagan's (I992) analysis.>7 Both lexical accounts cannot explain why (only) reflexive pronouns in the position of the accusative object are ambiguous between an argument and a non-argument interpretation. Even if it is assumed that principle (26) derives the distribution of non-argument-reflexives, it does not explain why in most Indo-European languages an argument position dominated by another non-genuine argument position must be linked to a reflexive pronoun (i.e. receives the feature [+REFLJ). Note finally that there is no morphological evidence for a lexical or syntactic derivation of anticausatives and middle constructions. In German, verbs in TRSs are always morphologically active. For a discussion offurther shortcomings oflexical analyses, see Steinbach (2002b). In conclusion, neither a syntactic nor a lexical approach offers a conclusive and uniform derivation of the ambiguity ofTRSs. The next section outlines an alternative post-syntactic analysis ofTRSs. This analysis is based on two assumptions, which are both independently motivated. The thematic ambiguity ofTRSs is derived from a single underlying syntactic representation at the interface between syntax and semantics.

In semantics (i) is a two-place verb with an implicit object that corresponds to such examples as Ra(ftrinkt ('Ralfis drinking/drinks'). In syntax, (i) projects two reflexive pronouns because of principle (26). But the corresponding German sentence (ii), with a dative and an accusative reflexive pronoun, does not yield interpretation (i). 27 Alrhough Fagan (I992: I75) argues that the reflexive pronoun indicates some change in argument structure, she does not offer an explanation for this but simply assumes that the lexical entry of the middle construction contains a reflexive pronoun.

Markus Steinbach

7.3.

THE AMBIGUITY OF TRANSITIVE REFLEXIVE SENTENCES AND THE [±R] -DISTINCTION

Our analysis is based on a modified version of the binding theories of Reinhart and Reuland (1993) and Pollard and Sag (1994), which incorporates non-argument-reflexives in binding theory. The basic theoretical concepts are briefly introduced in section 7.3-1. In section 7.3.2, this theory is applied to TRSs and it is illustrated how this analysis accounts for the thematic ambiguity ofTRSs.

7.3.1. Basic concepts I follow Pollard a~d Sag (1994) in assuming that the binding principles should be defined relative to semantic predicates. The relevant parts of their binding principles are given in (33). (33)

Principle A: a locally o-commanded anaphor [i.e. reflexive pronoun] must be locally o-bound O-binding: Y locally o-binds Z just in case Y and Z are coindexed and Y locally o-commands Z ...

Pollard and Sag define o-command on the basis of the relative obliqueness of co-arguments!8 In German, the relevant notion seems to be case. Nominative subjects are less oblique than accusative objects, which are less oblique than dative objects. 29 Hence, a dative object can be bound by an accusative object and a nominative subject whereas an accusative object can only be bound by a nominative subject. The examples in (34) illustrate that accusative objects asymmetrically bind dative objects, which asymmetrically bind more oblique objects (c£, for example, Grewendorf 1988 and Vogel and Steinbach 1998). (34)

a.

Man hat die Gaster einanderr vorgestellt. one has the guests.AcC each other.DAT introduced 'The guests were introduced to each other.' b. *Man hat den GastenI einanderI vorgestellt. one has the guests.DAT each other.Acc introduced

,8 A synsem object Y is less oblique than a synsem object Z iffY precedes Z on the SUBCAT list of the same lexical head. An alternative definition could refer either to the order of semantic arguments on the predicate argument structures or to grammatical relations or case. The latter one would state that an argument linked to the nominative (subject) is less oblique than an argument linked to the accusative object, both of which are less oblique than an argument linked to the dative object. '9 In contrast with Pollard and Sag, I do not think that the obliqueness of arguments has to be fixed separately for each lexical entry. Instead, I assume the general obliqueness hierarchy in (i) for German, which manifests itself in various phenomena (cf. Vogel and Steinbach 1995,1998; Steinbach 2002b; and n. 30).

(i)

Obliqueness hierarchy for German Nominative (subject) > accusative object> dative object> other oblique objects

Unaccusatives and Anticausatives in German

199

. .. weil (es) den beiden Henkernr voreinanderr graute ... because (it) the two executioners.DAT of. each other dreaded ' ... because the two executioners dreaded each other' d. *... weil (es) vor den beiden Henkernr einander r graute ... because (it) of the two executioners each other.DAT dreaded

c.

In addition to Pollard and Sag's principle A, the concept of maximal A-chains is necessary to derive the ambiguity ofTRSs. The definition of maximal A-chains is given in (35) (c£ Reinhart and Reuland 1993 and Fox 1993).

(35)

a.

b.

MaximalA-chain A maximalA-chain is any sequence of coindexation of syntactic arguments that satisfies antecedent government General Condition on A-chains (GCC) A maximal A-chain (a" ... , an) contains exactly one link-ar-that is both [+ RJ and case-marked

According to (35), a well-formed maximal A-chain contains only one [+RJ-expression, which must be its head. Hence, every [+ RJ-expression in the position of a syntactic argument must head its own maximal A-chain. What do 'syntactic argument' and '[+RJ' mean? Two final assumptions, which are both independently motivated, are necessary to derive the ambiguity ofTRSs. 1.

2.

German distinguishes between structural and oblique case. Nominative and accusative are structural, dative is oblique. Only NPs that are assigned (or check) structural case are syntactic arguments. 3D Compared to the other pronominal elements, the cD-features of (weak) reflexive pronouns are maximally underspecified. Therefore, (weak) reflexive pronouns are not specified for the feature [RJ. Depending on the syntactic context, they can either be [+ RJ or [-R]. All other pronominal and nominal expressions are specified as [+ R],l'

30 In German, dative case differs from nominative and accusative in many respects. Dative is morphologically marked whereas nominative and accusative are unmarked. This difference manifests itself in nominalizations, free relatives, uninflectable indefinites, CP -complements, and a certain kind of idiom. Furthermore, German has so-called free and multiple datives but no systematic multiple occurrences of accusatives or nominatives. Case movement-that is, movement into a case position-is an exclusive property of accusative and nominative. Moreover, dative objects cannot undergo middle formation and tough-movement, and the word-order constraints on dative objects are different from those on accusative objects and nominative subjects. In addition, dative objects, unlike accusative objects, cannot be dropped in sentence-initial position. Finally, Bader, Bayer, Hop£, and Meng (1996) find a difference with respect to processing: the parser prefers assignment of structural (accusative) case (to a case-ambiguous NP) over oblique (dative) case. For further differences and discussion see Vogel and Steinbach (1998) and Steinbach (2002b). 3' Reflexive pronouns in German differ from personal pronouns at least in two dimensions: they are not specified for gender and they lack nominative case. Reflexive pronouns are always less specified than personal pronouns. The third-person personal pronoun ibr, for example, is specified as ([ +sing -plur], [-1p -2P +3P], [-masc +fem -neuter], [-nom -ace +datll and the corresponding reflexive pronoun sicb is specified as ([0], [-rp -2P +3P], [0], [-nom +acc +datll.

Markus Steinbach

200

The first distinction is language-specific. I do not assume that all languages distinguish between structural and oblique case in the same way German does. The second distinction seems to be more universal. In various languages, (wealc) reflexive pronouns are ambiguous between an argument and a non-argument interpretation. The fact that (weak) reflexive pronouns are maximally underspecified and hence not specified for the [Rl-feature seems to be responsible for their ambiguity.3 2 In German, a reflexive pronoun that is [-Rl and assigned structural case must not head a maximal A-chain. The GCC requires that syntactic arguments specified as [-Rl are antecedent-governed by another syntactic argument specified as [+ Rl. On the other hand, if a reflexive pronoun is [+ Rl it must head its own maximal A-chain. The GCC, in combination with the assumption that reflexive pronouns are not specified for [Rl, accounts for the ambiguity ofTRSs. This is illustrated in the next section.

7.3.2. A-chains and the interpretation of reflexive pronouns Sentence (36) is a simple example of a transitive sentence without a reflexive pronoun. (36)

liest einen Roman. Suse Suse.NOM reads a novel.Acc 'Suse is reading a novel.'

It contains two syntactic arguments (Suse and einen Roman), which are both specified as [+ Rl. Hence, each of these two syntactic arguments must head its own maximal Achain. The relevant part of the structure is given in (37). CP and verb movement are omitted for the salce of simplicity. (37)

a. b.

[AgrSP Suse r [AgrOP einen Roman 2 [yp tr [V't 2 liestllll chain I = Suser-tr; chain 2 = einen Roman 2- t2

This syntactic representation yields the following interpretation. Chain 2, which consists of the accusative object and the trace in complement position, is linked to the second (internal) argument of the verb (cf. (38b)); chain I (the subject and its trace in VP-Spec) is linked to the first (external) argument (c£ (38c)).

(38)

a. b. c.

L (x,y) Ay L (x,y) (r) ~ L (x,r) Ax L (x,r) (s) ~ L (s r)

A sentence with two syntactic arguments that are both [+ Rl necessarily contains two maximal A-chains, each linked to a semantic argument. The same holds true for a J' English seems to be an exception. Note, however, that English also distinguishes weak from strong forms. The strong form is the reflexive pronoun x-self, the weak form is morphologically empty in English. Like weal, reflexive forms in other Indo-European languages, it triggers a reflexive interpretation only with certain kinds of verb (verbs of grooming, for instance) and it must be used in middle constructions and anticausatives (c£ Steinbach 2oo2a).

Unaccusatives andAnticausatives in German

201

reflexive pronoun specified as [+Rl Recall that reflexive pronouns are not inherently specified for [R]. Depending on context, they can either be [+ R] or [-R]. Consider first the [+ R] -reflexive in (8a), which is repeated in (39). (39)

Ralf rasiert sich. Ralf.NOM shaves SICH

The reflexive pronoun must head a maximal A-chain because it is specified as [+ R]. Thus, (39) contains two maximal A-chains. Again, both chains are linked to a semantic aJ;gument of the verb. (4 0 )

a. b. c. d, e.

[AgrSP Ral~ [AgrOP sich-[+R]2 [yp t, [v' t2 rasiert]]]] chain I = Ral~-t,; chain 2 = sich-[+R]2-t2 R(x,y) Ay R (x,y) ~ [SICH] ) ~ R (x, [SICH]) Ax R (x, [SICH]) (r) ~ R (r,[SICH])

Recall that the binding principles are restricted to co-arguments along the lines of Pollard and Sag (1994). A locally o-commanded reflexive pronoun must be locally 0bound, i.e, coindexed with a less oblique co-argument. In (41) the accusative reflexive pronoun must be coindexed with the subject. (41)

R (p,[SICH]) ~ R (r" [SICH],)

The final semantic representation in (42) results from A-abstraction on the antecedent. Following Reinhart (1983), all arguments that are coindexed with the antecedent are converted into variables bound by the A-operator.

Now consider [-R] -reflexives. If a reflexive pronoun receives structural case and is specified as [-R], it must be antecedent-governed by another syntactic argument that is specified as [+R]. Otherwise it fails to meet the GCC in (35b). In (43) the non-argument-reflexive is included in a complex maximal A-chain headed by the syntactic subject (a [+Rl-expression). The following anticausative construction illustrates the interpretation of non-argument-reflexives.

(43)

a.

b. c.

Die Tiir, (= [+RJ) offnetsich, (= [-R]) the door. NOM opens SICH [AgrSP die Tiir-[+RJ,-NoM [AgrOP sich-[-R],-Acc [yp t, t, offnet]]] chain = die Tiir,-sich-[ -R],-t,-t,

Sentence (43a) contains only one chain that can be linked to an argument position. This chain consists of the syntactic subject and the reflexive pronoun and it is interpreted in its VP-internal base position (the complement ofY, which corresponds to the second argument) just like the object chain (chain 2) in the previous two examples. Hence, the syntactic subject is linked to the second (internal) argument of the verb. Recall that in anticausatives the first (external) argument is deleted by the semantic rule

Markus Steinbach

202

of argument reduction (cf. n. 9). The final semantic representation is given in (44c). (44)

b.

0 (x,y) Ay 0 (x,y) (t)

c.

0 (0,t)

a.

Reduction of the first argument is more restrictive than argument saturation. While the semantic rule of argument saturation can be applied to almost all verbs, argument reduction changes the basic meaning of the verb and can thus only be applied to certain verbs. Reduction of the first argument is only possible if the verbs does not entail volitional involvement in the event or state for this argument-that is, if the first argument is not specified as [+mental state involved] (cf. Dowty 1991; Reinhart 1996; Steinbach 2002b; Chierchia, in this volume). The thematic interpretation of middle constructions can be derived in a similar way. Note that in middle constructions the first argument is saturated and not reduced. In middle constructions the free argument variable is bound by a generic operator, as can be seen in (45). The generic operator also binds the situation variable s. Middle constructions are thus characterizing or habitual sentences (cf. Krifka et al. 1995 and Cohen 1996). Simplifying somewhat, the generic operator can be analysed as the phonologically empty counterpart of the frequency adverb usually. Thus, sentence (45) would be true if a situation in which someone reads the book Biene Maja is very likely to be a situation in which this book is easy to read for this person. (45)

a.

b.

Biene Maja liest sich leicht. Biene Maja reads SICH easily 'Biene Maja reads easily.' GENs,x,y [y = biene-maja] [read(s, x,y) & easy(s)]

In impersonal middle constructions, the complex A-chain (es-sich) is not linked to an argument. The impersonal subject es, which is a quasi-argument only fulfilling a specific syntactic function, is not interpreted in semantics. Hence, no semantic argument of the verb is linked to syntax and the semantic argument of the one-place predicate is again bound by the generic operator. Reduction of the sole argument of one-place verbs is impossible because this would yield a zero-place predicate. The derivation of the thematic ambiguity ofTRSs is summarized in (46). A reflexive pronoun can either head its own maximal A-chain (line 3) or it is part of a more complex maximal A-chain headed by a [+R]-expression in subject position (line 2). The former links the [+R]-reflexive to the second argument, similar to other nonreflexive accusative objects. The latter links the subject to the second argument. In German, this ambiguity is restricted to reflexive pronouns in the position of the accusative object. Only accusative objects can be antecedent-governed by another [+ R]-expression, the nominative subject. By contrast, dative objects are not syntactic arguments because dative case is oblique in German. Hence, they are correctly excluded from A-chain formation, which is a necessary condition on the non-argument interpretation of reflexive pronouns. So far, our analysis derives the observations

Unaccusatives andAnticausatives in German (46)

20 3

Simple and complex chains

Simple chain Complex chain Two chains Two chains

Syntax

Semantics

[+R-NP]-[-R-TRACE] [+R-NP]-[- R-SICH]-[-R-TRACE]-[-R-TRACE] [+R-NP]-[-R-TRACE]; [+R-SICH]-[-R-TRACE] [+R-NP]-[-R-TRACE]; [+R-NP]-[-R-TRACE]

One argument One argument Two arguments Two arguments

made in section 7.2 and offers a uniform analysis of the thematic ambiguity ofTRSs. This is summarized in (47). (47)

Only a reflexive pronoun in the position of the accusative object is a middle marker in German.

7.4.

BACK TO UNACCUSATIVES

This analysis of TRSs predicts that unaccusatives and unergatives share the same underlying VP-structure. Recall that a maximal A-chain is interpreted in its base (or tail) position. In a simple transitive sentence the subject chain (chain I in (37) and (40)) is linked to the first argument of the verb via VP, Spec and the object chain (chain 2 in (37) and (40)) is linked to the second argument via the complement position of VO. The following two linking principles for German account for this. (48)

a.

VP, Spec is linked to the first argument of the verb.

b.

The complement ofVo is linked to the second argument of the verb.

Middle constructions and anticausatives are also subject to the linking principles in (48). According to (48b) the complex maximal A-chain in (43) is interpreted in its base position, the complement position ofVo. Hence, in middle constructions and anticausatives, the syntactic subject is linked to the second argument of the verb because of the non-argument-reflexive. The linking conditions in (48) are not sensitive to the distinction between internal and external arguments, which is repeated in (49). (49)

a. b.

Unergative predicates: P (x) Unaccusative predicates: P «x»

Unergative and unaccusative verbs select only one argument. According to (48), the subject chain is always linked via VP, Spec to the first (external or internal) argument of the verb. The principles in (48) contradict a theory that reflects the lexical distinction between unaccusatives and unergatives in their syntactic representation. If we assume the structures given in (50), the linking conditions in (48) cannot apply to

Markus Steinbach

20 4

unaccusatives because they predict that the complement position of yo must be linked to a (non-existent) second argument.

(50)

a.

b.

Unergative: [yp NP VOl Unaccusative: [yp [yo NP YOll

The analysis ofTRSs proposed here, then, predicts that un accusatives and unergatives share the same VP-structure, namely (50a). The linking principles in (48) contradict a theory that reflects the lexical distinction between unergatives and unaccusatives in syntax. Structure (50a) is in line with our derivation of non-argument-reflexives. The sole syntactic argument, the nominative NP, must always be linked to the sole semantic argument of the one-place predicate, regardless of whether this semantic argument is external or internal. Matters are a little more complicated if we choose structure (50b) for unaccusatives, which requires a slight modification of the linking principles in (48). According to this modification, the specifier position would be linked to the external semantic argument, and the complement position of yo to the internal semantic argument of the verb. (48 ')

Linking principles for a syntactic representation of unaccusativity: a. Spec ofVP is linked to the external argument of the verb. b. The complement of yo is linked to the internal argument of the verb.

As a result, unaccusative predicates are expected to have two grammatical syntactic realizations. The internal argument can be linked either to an NP in the complement position or to a complex A-chain. Unaccusatives should therefore be syntactically intransitive and transitive reflexive. This is illustrated in (51) and (52).

(51)

a. *Clara errotet sich. b. c.

Clara blushed S1CH [AgrSP Clara-[+Rlr-NoM [AgrOP sich-[-Rl,-ACC [yp t, [Vo t, errotetllll chain = Clara,-sich-[-Rl,-t,-t,

The structure in (5Ib) contains a complex A-chain. Like subjects in middle constructions and anticausatives, the subject in (5I) can in principle be interpreted in the complement position. The unaccusative interpretation of (51) should therefore be grammatical-but it is not. If we accepted structure (50b) and the modified linking principles, we would expect sentence (51) to yield the same interpretation as the unaccusative in (52). (52)

a.

b. c.

Clara errotet. Clara blushed [AgrSP Clara-[ +Rl,-NOM [yp [yo t, errotetllll chain = Clara,-t,

The same applies to impersonal middle constructions, which can be derived from unergative and unaccusative verbs. Recall that the complex maximal A-chain containing the reflexive pronoun and the impersonal subject es are not linked to a non-exist-

Unaccusatives and Anticausatives in German

20 5

ent second argument. However, this does not affect the (implicit) first argument of the verb, which might be internal or external. The implicit first argument is always bound by the generic operator in middle constructions. In conclusion, argument linking in German is only sensitive to the order of the arguments. Further properties of semantic arguments, such as the internal-external distinction, are irrelevant for the linking of arguments. This may be seen as indirect evidence that German does not distinguish between unaccusatives and unergatives in syntax. All intransitive sentences share the same underlying VP-structure in (50a). This VP contains only one Np, which is always linked to the first semantic argument of the one-place verb. This argument also applies to theories that assume layered VPstructures (c£ n. 4). It follows that in German, all subjects of intransitive verbs are base-generated in the specifier of the same VP (or vP). Hence, the analysis proposed by Hale and Keyser (1993) for English cannot be transferred to German. Let us conclude this section with a final remark on the non-reflexive anti causative, which is repeated in (53).

(53)

a. b.

Hans-Martin bricht den Stock. Hans-Martin breaks the stick DerStock bricht (*sich). the stick breaks (*SICH)

Recall that reflexive anticausatives are basically two-place causative verbs. The anticausative interpretation is derived from an underlying two-place predicate by a semantic rule of argument reduction. Hence, the anticausative is derived from a basic causative form. Non-reflexive anticausatives, on the other hand, are basically oneplace unaccusative verbs, which only select an internal argument. In this case the unaccusative one-place verb (53b) is the underlying form and the causative verb in (53a) the derived form (c£ also Wunderlich 1993). The causative interpretation results from adding an external argument. Sentence (53b) is not derived from (53a) via argument reduction; rather, (53a) derives from (53b) via argument addition. Thus, nonreflexive anticausatives are basically unaccusative one-place predicates that allow causativization. This difference in the lexical representation of these two kinds of anticausatives accounts for the syntactic differences between reflexive and non-reflexive anticausatives-namely, the presence or absence of the reflexive pronoun.

7.5.

CONCLUSION

In German, anticausatives are either unaccusative or transitive reflexive. Here I focused on the latter, which are discussed in the context of other interpretations of transitive reflexive sentences. In addition to the anti causative interpretation, TRSs also yield a reflexive, a middle, and an inherent-reflexive interpretation. This semantic ambiguity ofTRSs is systematic and can be found in many other Indo-European

206

Markus Steinbach

languages. It is therefore worthwhile to offer a uniform analysis ofTRSs. I proposed a post-syntactic analysis that is based on the distinction between structural and oblique case and [±Rl-expressions. The ambiguity ofTRSs follows from these two assumptions and the General Condition on A-chains. A reflexive pronoun in the position of the accusative object can either be interpreted as a [+Rl-expression linked to the second semantic argument of the verb (argument interpretation), or it is interpreted as a [-Rl-expression that indicates valency reduction (non-argument interpretation). In the second case the syntactic subject is linked to the second semantic argument of the verb. Hence, the complement position ofVo is always linked to the second argument of the verb whereas the specifier of the VP is linked to the first argument of the verb. These linking principles are only sensitive to the distinction between first and second argument-not to the distinction between internal and external argument. All intransitive sentences thus have the same underlying structure. The syntactic subject is always linked to the sole (and consequently, the first) semantic argument of the verb. The analysis ofTRS proposed in this chapter therefore provides an indirect argument against a syntactic distinction between unaccusatives and unergatives in German. Unaccusatives differ from unergatives only in their lexical representation.

8 Syntactic Unaccusativity in Russian Maaike Schoorlemmer

8.1.

INTRODUCTION

Since their discovery in I978 (Perlmutter I978), unaccusative verbs have fascinated linguists working in the fields of syntax, lexical semantics, and the interaction of the two. Over the years, many syntactic diagnostics have been found that distinguished between two classes of intransitive verb. The subjects of one class pattern with the subjects of transitive verbs (unergative verbs); the subjects of the other class pattern with the objects of transitive verbs (unaccusative verbs). The standard explanation of this distinction has been that unaccusative verbs project their grammatical subjects like passive verbs do: in object position. There have also been indications that many of the verbs identified as unaccusatives share semantic properties that might directly determine their behaviour in the diagnostics. It might not then be necessary to assume an outright syntactic, configurational difference between the two classes. One form of synthesis between these two points of view is offered in Levin and Rappaport Hovav (I995), who argue that there is a syntactic pendant to unaccusativity which is learnable on the basis of the semantic distinctions identified. The chapter I would like to thank the audience of the Berlin workshop for useful comments. I also thank Olga Boril<, Ljudmila Geist, Angeliek van Hout, Volodja Kiiminov, Svetlana Poljakova, and Mario van de Visser for judgements, comments, and discussion. All remaining errors are my own.

Maaike Schoorlemmer

208

is broadly intended to support this conclusion on the basis of material from Russian. I will discuss two syntactic configurations in Russian that are restricted to transitive and unaccusative verbs; these constructions show evidence of syntactic unaccusativity, that is, of a syntactic encoding of whatever lexical feature is the basis of the distinction between unaccusative and unergative verbs. Levin and Rappaport Hovav argue that different languages may adopt different semantic properties as the basis for this syntactic distinction. The chapter provides good diagnostics for identifYing un accusative verbs as well as some discussion of the semantic factors likely to be operative in those unaccusative verbs identified by these diagnostics. A further possible approach to the question of the nature of the difference between unaccusative and and unergative verbs is the hypothesis that there is not really any distinction between the syntactic and semantic properties of verbs, in the sense that a particular semantic type of argument will always appear in a particular syntactic configuration. I will refer to this as the 'direct mapping approach'. The diagnostics identified here will be used to define the set of un accusative verbs in Russian, which will then allow us to evaluate this latter view on the projection of arguments. Under the direct mapping approach, semantic factors relevant to the unaccusative-unergative distinction are predicted to have other syntactic effects as well. I will argue that this is not the case in some of the factors identified, which is evidence for approach to the projection of internal and external arguments in terms of Linking Rules, as proposed by Levin and Rappaport Hovav. The chapter is organized as follows. Sections 8.2 and 8.3 are devoted to the first diagnostic context: distributive po-phrases. In section 8.2 I introduce a contrastive distribution of two different kinds of distributive expression in Russian and argue that the relevant contrast cannot be due to a specific semantic property of the relevant verbs. A provisional syntactic explanation will be provided in section 8.3, based on the cross-linguistically observed generalization that PPs-in contrast with DPs-must occur as internal arguments. The second diagnostic context involves telic aspectuality and Russian aspectual morphology. The relevance of telic aspectuality and the role of the internal argument for the Russian aspectual system is laid out in section 8+ In section 8.5, I extend its application by focusing on the derivation of imperfective forms of verbs as a diagnostic for the presence of an internal argument, and therefore of an unaccusative verb in the case ofintransitivity. Section 8.6 argues for a way of using this test to identifY underived unaccusative verbs. Section 8.7 addresses the issue ofidentifYing specific linking rules that are active in Russian. The conclusions are summarized in section 8.8.

ail

8.2.

DISTRIBUTIVE PO

The distribution of a particular kind ofPP-namely, po-phrases without numeralscan be used as a test for syntactic unaccusativity. I will argue that the distributional

Syntactic Unaccusativity in Russian

20 9

facts are a reflection of a more general restriction on the distribution of argument PPs in Russian and other languages, which is that they must be internal arguments.' In Russian, direct objects of transitive verbs can be replaced by a PP meaning 'one X each'. It turns out that some, but not all, intransitive verbs allow their subjects to be replaced by the same Pp, displaying the parallel behaviour of objects of transitive verbs and subjects of some intransitive verbs that is the classic format for unaccusativity diagnostics. This test for unaccusativity was first identified by Pesetsky (1982), and refined in Borik (1995).

8.2.1. Bare po-phrases as internal argument PPs Russian has argument PPs headed by po with a distributive meaning (glossed as DIST in the examples).

(I)

a.

b.

My polucili knigu. we received bookAcc My poluCHi po knige. we received DIST bookDAT 'We received a book each.'

Such po-phrases usually, but not necessarily, cooccur with a kazdyj ('each, every') phrase. (2)

a. b.

Kazdyj rebenok poluCH po knige. every child received DIST book.DAT Deti polucili po knige. children received.pL DIST bookDAT 'The children received a book each.'

The crucial property of distributive po-phrases is that they can replace the direct object of a transitive verb, but not its subject. Replacement of such subjects is illustrated in the following examples.

(3)

a. *V kazdoj komnate po rebenku poluCH(-o) knigu. in each room DIST child.DAT received book 'In each room a child received a book' b. *V kaZdoj kvartire po marne razdala(-o) (detjam) jabloki. in each flat DIST mother.DAT gave children.DAT apples 'In each flat a mother gave apples (to children).'

The reason for the ungrammaticality might be that in the event of a PP subject the , Another bona fide syntactic test for unaccusativity in Russian is genitive of negation (Babby 1980). However, for this test to work, the verb needs to be unaccusative and an expression of existence (see Levin and Rappaport Hovav 1995: 153 on the effects of expression of existence in English). As a result of this, the test identifies only a small subset of the total number of unaccusative verbs, which include suscestvovat' ('exist'), prisutstvovat' ('be present'), and ostat'sja ('stay, remain').

Maaike Schoorlemmer

210

verb lacks proper inflection. We will see below (cf. (5» that Russian default agreement is perfectly capable of dealing with distributive PP subjects. I conclude that there is a true subject-object asymmetry in the distribution of distributive po-phrases. Intransitive verbs split into two classes with respect to the occurrence of distributive po-phrases. The subjects of the first class behave like the objects of transitive verbs, and allow replacement by a distributive po-phrase. Examples of such verbs with DP subjects are provided in (4).

(4)

a.

b.

c.

d.

V gOrSke ros cvetocek. in pot grew.M flower 'There was a flower growing in the pot.' V kastrjule kipela voda. in pot boiled.F water 'There was water boiling in the pot.' V konjusne stojala losad'. in stable stood-F horse 'There was a horse in the stable.' Iz karmana toreala pereatka. from pocket stuck-out.F glove 'There was a glove sticking out of the pocket.'

The subjects of the examples in (4) can be replaced by distributive po-phrases, as in (5). The verb then gets default agreement, glossed as DEF.

(5)

a.

b.

c.

d.

V kazdom gOrSkeIV gOrSkax roslo po cvetocku. pot/in pots grew.DEF DIST flower.DAT in each 'In each pot grew a flower.' V kazdoj kastrjuleIV kastrjuljax kipelo po litru vody. in each pot/in pots boiled.DEF DIST litre.DAT of-water 'A litre of water was boiling in each pot.' V kaZdoj konjusnelV konjusnjax stojalo po losadi. in each stable/in stables stood.DEF DIST horse.DAT 'There was a horse in each stable.' Iz kazdogo karmana tOrCalo po pereatke. from each pocket stuck-out.DEF DIST glove.DAT 'A glove was sticking out of each pocket.'

The subjects of the second class behave like the subjects of transitive verbs and do not allow replacement by a distributive po-phrase. Examples of such verbs with DP subjects are provided in (6), and their ungrammatical counterparts with distributive subjects in (7).

(6)

a.

V kvartire smejalsja mal' Cik. in flat laughed boy 'A boy was laughing in the flat.'

Syntactic Unaccusativity in Russian

b.

c.

(7)

2II

Na ploseadke begala sobacka. on playground ran-around dog 'A dog was running around on the playground.' Na sosednej ulice fume! gruzovik on neighbouring street made-noise truck 'A truck made noise nearby.'

a. *V kazdoj kvartire/v kvartirax smtjalos'

po mal'dku. in each flat/in flats laughed.DEF DIST boy.DAT 'A boy was laughing in each flat.' b. *Na kazdoj ploseadke/na ploscadkax bega!o po sobacke. on each playground/on playgrounds ran-around.DEF DIST dog.DAT 'A dog is running around on each playground.' c. *KaZdyj den' na sosednej ulice fumelo po gruzoviku. each minute on neighbouring street made-noise.DEF DIST truckDAT 'A truck made noise in a neighbouring street every minute.'

I will refer to the class of verbs that allows a distributive po-phrase subject as unaccusatives, and to those that do not as unergatives. The two questions to be answered now are, What is the difference between these two groups of verbs? and How does that difference account for their behaviour with respect to distributive po-phrases? Before getting down to possible semantic or syntactic distinctions between these predicates, I want to rule out two potential reasons for the ungrammaticality of the examples in (7). First of all, as mentioned before, there is no general ban on agreement between finite verb and a po-phrase subject, as illustrated in (5). It follows that the inability of the verb to agree with a po-phrase subject cannot be the reason for the ungrammaticality of the examples in (7). Secondly, one might conjecture that the ungrammaticality of these examples is due to the fact that they express quantification in addition to distribution. After all, the interpretation of the po-phrase subject or object is 'one X each', which could be interpreted as a quantified expression. Under this scenario, however, we predict the verbs in (7) to resist other quantificational subjects as well, which is a false prediction, as illustrated in (8).

(8)

a.

b.

c.

V kvartire smejalis' dva mal'cika. in flat laughed.pL twO.NOM boYS.GEN 'Two boys laughed/were laughing in the flat.' Na ploscadke begajut tri sobacki. on playground run-around.pL three. NOM dogS.GEN 'Three dogs are ninning around on the playground.' Na sosednej ulice sumjat dva gruzovika. on neighbouring street make-noise.pL twO.NOM trucks.GEN 'Two trucks are noisy in a neighbouring street.'

The conjecture must therefore be rejected that the ungrammaticality of po-phrase

Maaike Schoorlemmer

212

subject with unergative verbs is due to the inability of these verbs to take quantified subjects. 2

8.2.2. Numeral po-phrases-no semantic explanation It is impossible to formulate an explanation for the distribution of distributive pophrases that relies entirely on the semantics of the relevant distributional constituents and/or verbs. Assuming that Levin and Rappaport Hovav are right, a semantic factor can probably be identified that distinguishes the unaccusative and unergative verbs identified in the previous subsection. Suppose that there is a correlation between the occurrence of distributive po-phrases and this semantic factor. In order to show that semantics is all there is to unaccusativity, this property would also need to account for the specific distributional phenomenon observed. I will show that the behaviour of intransitive verbs with respect to distributive po-phrases has to do with the syntactic structure of this kind of constituent, and not with its (semantic) distributive properties. The conclusion must therefore be that the explanation relies on the syntactic configuration projected by unaccusative verbs as opposed to unergative ones. Observe that if we want to argue that unaccusativity is strictly a semantic phenomenon, establishing a correlation between this semantic factor and triggering unaccusativity in acquisition is not enough. We also need to explain how this factor affects the distribution of distributive po-phrases. An intuitive first version of this might read as follows: 'Something in the semantics of the verb disallows a distributive argument'. When we look at distributivity of objects in English, a likely candidate for this 'something' is stativity. The examples below contain telic and atelic eventive as well as stative predicates (in (9) and (10), respectively), and distributivity is possible only in the eventive (telic or atelic) ones.

(9)

a. John read a newspaper every day. b. John tortured a prisoner every day. c. *John resembled his boss every day.

(10) a.

The children received an apple each. The children tortured a spider each. c. *The children resembled a parent each.

b.

Example (9c) is ungrammatical even in the unlikely event of John having a different boss every day. Distributivity of this type, which semantically is very similar to distribution using po-phrases in Russian, requires the expression of an event in English, which therefore excludes stative and/or individual level predicates. The following hypothesis might then be proposed to account for the distribution of po-phrases in Russian: (II)

A distributive po-phrase argument requires an eventive verb 2

I will return to the combination of quantification and distributivity in section 8.3.

Syntactic Unaccusativity in Russian

21 3

By this hypothesis, eventivity is the crucial property distinguishing verbs whose subjects behave like transitive subjects or transitive objects. This hypothesis faces two problems. First of all, the class of unaccusative verbs as identified by this diagnostic does not differ from the class of unergative verbs as far as eventivity is concerned. Verbs from both classes can be used in a context like 'when I entered .. .' which forces a progressive interpretation on the imperfective verb.

(12)

Kogda ja vosla, when I entered, b. .., v kastrjule kipela voda in pot boiled.F water ' ... there was water boiling in the pot' c. . .. v konjusne stojala losad' in stable stood-F horse ' ... there was a horse in the stable'

(13)

a.

.., v kvartire smejalsja mal' Cik

b.

in flat laughed boy ' ... a boy was laughing in the flat' .., na sosednej ulice fumel gruzovik on neighbouring street made-noise truck ' ... in a neighbouring street a truck was making noise'

The second problem is more general. Suppose we succeed in finding a semantic property that unifies unaccusative verbs as opposed to unergative ones and the presence of which can be related directly to the inability of unergative verbs to take distributive pophrases. We are then immediately faced with the problem that a different kind of distributive po-phrase exists-namely, the numeral po-phrase-which is allowed as the subject of either type of intransitive predicate. Numeral po-phrases are distributive like the po-phrases discussed so far, which I will henceforth refer to as 'bare po-phrases'. However, numeral po-phrases differ from bare po-phrases by including a numeral and by occurring as subjects of unergative verbs (the relevance for unaccusativity of the presence of a numeral is due to Borik 1995). This is illustrated in (14). (14)

a.

b.

V kaZdoj kvartire smejalos' po dva mal' Cika in each flat laughed-DEF DIST two-NOMIACC boys.GEN 'Two boys laughed/were laughing in each flat.' Na kaZdoj ploscadke begaet po tri sobacki: on each playground run-around-DEF DIST three.NoMlAcc dogS.GEN 'Three dogs run around/are running around on each playground.'

Maaike Schoorlemmer

214

c.

Kazdyj den' na sosednej ulice sumit po odnomu every day in neighbouring street makes-noise-DEF DIST [one gruzoviku. truck].DAT 'Every day, one truck makes noise in a neighbouring street.'

With intransitives, the two types of po-phrase both trigger default agreement on the verb. We have been looking for a way in which the distinction between two types of verb with respect to distributive bare po-phrases could be made to work if it were solely based on semantic properties of the verbs in question, as stated in hypothesis (n). Given that the verbs under discussion do occur with numeral po-phrases that are equally distributive, it is impossible to maintain that the ability to allow distributive subjects as such causes the ungrammaticality of the examples in (7). For this, they would have to be semantically classified not to allow bare po subjects, although they do allow subjects with numeral po, which is also semantically distributive and, like bare po, may involve distribution of singular elements. Not only has the ability to allow distributive subjects been found to be an unlikely candidate for a semantic factor triggering unaccusative-like behaviour, it is also unlike any other semantic factors that have been proposed to force unaccusative behaviour. The type of semantic factor that is found to correlate with unaccusativity across languages has something to do with 8-roles, aspectual properties, change of state, etc. None of these factors can be made responsible for the limited distribution of distributive po-phrases with either transitive verbs or intransitive verbs. I conclude that the semantics of the verbs in question cannot explain the distribution of distributive pophrases.

8.3.

EXPLANATION IN SYNTACTIC TERMS

Recall the standard analysis of the distinction between unaccusative and unergative verbs in terms of argument configuration. Semantics cannot explain the occurrence of distributive po-phrases, and I now want to argue that this syntactic property-argument configuration-can. Of course, we should meet the same standards as those set for the semantic explanation: we need to indicate not only how to encode the distinction, but also how the encoding explains the observed behaviour. The point of departure is a lexical feature on unergative verbs that induces a syntactic projection of the subjects as external arguments, and/or a lexical feature on unaccusative ones that induces syntactic projection of the subject as an internal argument. I will assume that this distinction takes the form of the projection ofVP (unaccusative verbs) vs. the projection of both VP and vP (unergative verbs) (Chomsky 1995). Secondly, there must be a syntactic distinction between (at least) the two types ofpo-

Syntactic Unaccusativity in Russian

21 5

phrase, one that can be made relevant to the hierarchical position of the argument. In this section, I will offer an approach in terms of the PP-DP distinction. But before getting down to my specific proposal, let me first eliminate from the discussion some factors that might seem relevant to it. Numeral po-phrases must occur in the scope of a quantifier such as kazdyj ('each, every') when they are the subject of an unergative verb. Consider the examples in (15) containing numeral po (and compare these with (14)).

(15)

a.

b.

c.

V kaZdoj kvartire/*V kvartirax smejalos' po odnomu mal' ciku. in each flatlin flats laughed DIST [one boy].NOM 'A boy was laughing in each flat.' Na kaZdoj ploscadke/??Na ploscadkax begaet po dve on each playground/on playgrounds runs-around DIST [one sobacki. dOgJ.NOM 'A dog is running around on each playground.' Kazduju minutu/*vse vremja na sosednej ulice sumit po each minute/all time on neighbouring street makes-noise DIST tri gruzovika. [three trucks].NOM 'A truck makes noise in a neighbouring street every minute.'

It turns out that in those contexts where bare po-phrases are impossible, numeral pophrases require the presence of a quantifier. It is impossible to tell whether this is a true difference between the two kinds of po-phrase, because the relevant context to check it (bare po-phrases as subjects of unergative verbs) is not available (see the examples in (7)). Po seems to show different case-marking properties in the two types of po-phrase. In ordinary PPs with this preposition, in bare po-phrases and in numeral po-phrases with the numeral odin ('1'), it assigns dative case. This is illustrated in (16), (2), and (14c), respectively. (16)

a.

b. c.

d. e.

My Sli po reke. we walked along river. OAT otpravit' posylku po pocte send package by post. OAT Poezda Sli po raspisaniju. trains ran by timetable. OAT Rodstvenniki po mated vse vysokie. relatives by mother. OAT (are) all tall Ja ne mogla najti specialista po sovremennoj lingvistiki. I not could find specialist in modern linguistics. OAT

In numeral po-phrases, when the numeral is higher than odin, the numeral occurs in the nominative or accusative (or paucal; see Franks 1995) case, as in (1¥, b)-in other

Maaike Schoorlemmer

216

words, it loses the ability to assign dative case. 3 Since bare po does not cooccur with numerals, it is unclear whether it really differs from numeral po in this respect. 4

8.p. A further distinction There is a further distinction between numeral and bare po-phrases to suggest that there is a true syntactic distinction between the two types of constituent; numeral pophrases can be relativized, bare po-phrases cannot. (17)

a.

Po dva jabloka, kotorye my razdali, ne roskos'. DIST two apples that we distributed (are) no luxury b. *Propalo po jabloku, kotoroe my razdali. we distributed disappeared DIST apple.DAT that

The contexts are chosen so as to be compatible with the distributional restrictions on either type of constituent. It is clear from these examples that distributivity could not be responsible for this distinction. My proposal is that category could be. It works as follows. It has been argued by a number of authors in different frameworks that across languages, complement PPs must be internal arguments. Explanations for this have been provided in terms of case-marking (Pesetsky 1982), the inability of verbs to select their external arguments (Marantz 1984), and the inability of the preposition to incorporate with the verb from an external argument position (Neeleman 1997).5 The observed distinction in distribution between bare a numeral po-phrases could be explained straightforwardly if it could be shown that the following state of affairs holds: , Borovikoff (2001) treats numeral po-phrases with the numeral I as bare po-phrases, sole!y on the basis of the case facts. This, however, does not match the distributional behaviour of these po-phrases, where it is the presence of the numeral that matters, not its cardinality or case marking properties (Borik 1995). 4 Crucially, I am not assuming that bare po-phrases contain an empty numeral 1. If they did, there would be no way of establishing a syntactic difference between the two constructions, and therefore no way to account for the different syntactic behaviour. 5 In some languages, including Dutch, PPs can replace any DP argument. Some examples are in (i) and (ii). (i)

Over Groningen is korter. 'Via Groningen is the shortest way.'

(ii)

Op het dalr vergt we! vee! voorbereiding, maar het is leuker dan in de tuin. on the roof requires a lot of preparation, but it is more fun than in the garden

Whether or not a PP can be used in this way in this way seems to be independent of the hierarchical status of the constituent (it can even be the external argument of a transitive verb). Rather, it seems to be a language-particular property, which is also illustrated by the fact that even a sentence such as tinder the stairs is a good hiding place is ungrammatical in Russian with the PP as the grammatical subject. This is illustrated in the examples in (iii)-(iv), where the subject status of the PP can be read off the verbal agreement. (iii) *Do dnja rozdenija bylo ploxoj den' dlja ekzamena. was-N bad daY-M for exam before birthday (iv) *Cerez Novgorod bylo samaja korotkaja doroga. shortest waY-F via Novgorod was-N the

Syntactic Unaccusativity in Russian (18)

a.

b.

217

Numeralpo-phrases are DPs Bare po-phrases are PPs

A distinction along these lines also immediately explains the difference in relativization. There is further cross-linguistic evidence for the DP status of numeral po-phrases as well as other constituents in which a preposition modifies a numeral. Examples of such constituents are up to thirty people, around thirty children. In relativization in English, for example, preposition stranding is impossible in such cases, although it is generally possible with argument PPs. (19)

a. b.

(20)

a. *The thirty children who/that I saw up to. b. *The thirty children which I saw around.

The three factors (that) this argument relies on. The many people he stood up for.

Like numeral po-phrases, other constituents in which a preposition modifies a numeral, in both Russian and English, behave like DPs in a number ofways: they can be external arguments, they allow relativization, and they do not allow preposition stranding.

(21)

a. b.

C.

(22)

a.

b. C.

V etom zdanie rabotaet s desjatok celovek. in this building works around ten.ACC people from Okolo pjati celovek podalo zajavlenie ob uxode. around five.GEN people gave statement about resignation 'Around five people handed in their resignations.' Do 20i celovek igralo v karty na kazdoj lekcii. up-to twenty.GEN people played cards at every lecture Propalo te samye do trex jablok, kotorye zdes' ldalo. disappeared the very up-to three apples that here lay Propalo s djuZinu jablok, kotorye zdes' ldali. disappeared around dozen apples that here lay Po dva jabloka, kotorye my razdali, ne roskos'. DIST two apples that we distributed (are) no luxury

Similar facts can be provided for Dutch and English. I take these facts as evidence for a DP status of the relevant constituents, which is why I will refer to them as quasi-PPs. Having established that quasi-PPs, including numeral po-phrases, are DPs, at least one part of the puzzle-the DP-like distribution of numeral po-phrases-can be explained. However, we have the contrast with bare po-phrases to account for.

8.3.2. The categorial status of bare po-phrases The remainder of this story would be straightforward if bare po-phrases showed clear signs of being PPs. In that case, their distribution would be a mere reflection of their

Maaike Schoorlemmer

2I8

categorial status. However, the PP status of bare po-phrases is not obvious. Apart from the clear PP-like behaviour under relativization illustrated in (17), other phenomena point in the direction of a DP status of bare po-phrases. So, whereas the DP complement of an argument PP can relativize, this is impossible with the nominal constituents included in either numeral or bare po-phrases.

(23)

a. b.

(24)

a. *Propalo tri jabloka, do kotoryx zdes' ldalo. disappeared three apples up-to which here lay b. *Propala djuzina jablok, s kotorye zdes' ldali. disappeared dozen apples, around which lay here c. *Dva jabloka, po kotorym my razdali, ne roskos'. two apples, DIST which we distributed, (are) no luxury

(25)

The three factors on which this argument relies. The many people for which/whom he stood up.

*Jabloko, po kotoromu/kotoroe my razdali, ne roskos' apple DIST which.DATlAcc we distributed (is) no luxury

If we interpret the facts in (24) as evidence for a DP status of quasi-PPs, then the facts in (25) should be taken as evidence for a DP status also of bare po-phrases. My conclusion is that there is a true distinction between the syntactic properties of numeral and bare po-phrases. It is possible that the correct characterization of this distinction is as in (18). In that case, the observed differences between them are straightforwardly accounted for, but the occasional DP-like behaviour of bare po-phrases needs to be explained. This is the absence of relativization as well as the ability of bare po-phrases to occur as PP grammatical subjects of unaccusative or passive verbs (see n. 5). I have nothing further to say about the relativization facts, but the occurrence as a grammatical subject of bare po-phrases might have something to do with the absence of a thematic contribution by the preposition, which allows a bare po-phrase PP to get theta-marked as opposed to ordinary PPs. Alternatively, both types of constituent are DPs, and there is some other distinction between them that causes the observed differences. A candidate for such a distinction might be the hierarchical position of the actual preposition. In a numeral po-phrase, the distributive marker very obviously only has semantic scope over the numeral, whereas in bare po-phrases its hierarchical position is similar to that in a pp, taking scope over the entire NP. This is illustrated in (26a) for numeral po-phrases and (26b) for bare po-phrases. 6 (26)

a.

[DP [NumP

b.

[DP P

bp P [Q]] Numo [NP]J]

[NP]]

6 More work needs to be done on these structures if only to account for the case properties of such constituents. I leave this matter open for now.

Syntactic Unaccusativity in Russian

2I9

These are speculative structural proposals, but what is crucial is that the constituent containing the preposition also contains the NP constituent in the case of a bare pophrase, whereas the preposition is contained in a specifier of the projection line of the noun phrase in the case of a numeral po-phrase. The actual mechanism restricting the occurrence of PPs to internal argument position might then be sensitive to the relatively high structural position of the preposition in the bare po-phrase as opposed to the numeral po-phrase. Under both approaches, the crucial point will be what restricts the distribution to internal argument position of PP complements cross-linguistically as well as bare po-phrases in Russian. Observe that, if we succeed in unifying the distribution of properties ofPP complements and bare po-phrases, neither Marantz's nor Neeleman's explanation for the distribution of PPs can be correct. The inability of selecting an external argument (Marantz) or the requirement that the selected preposition incorporate (Neeleman) cannot be crucial in determining the distribution of PP complements, since a po-phrase is never headed by a selected preposition. What PP complements and bare po-phrases do have in common is a configuration in which the complement of the verb categorially is a pp, but nevertheless it is the DP complement of the preposition that is the semantic complement of the verb. This state of affairs seems to require an approach along the lines of Pesetsky's original proposal which rests on a potential discrepancy between c-selection and s-selection. Then, if bare po-phrases turn out to be PPs, the crucial factor must be category-related, as follows: a PP complement must be an internal argument whenever the preposition is not the sole contributor of thematic information to the DP contained in it. If bare pophrases do turn out to be DPs, the crucial factor must be defined in purely structural terms: any constituent must be an internal argument in which a preposition dominates the lexical projection of the argument and does not contribute thematic content. In both cases, reference must be made to the non-thematic nature of the distributive marker. I will leave an explanation of these generalizations for further research, as well as the final verdict on the categorial status of bare po-phrases.

8.3.3. Concluding distributivity It is very hard to find two unaccusative diagnostics in any language that pick out exactly the same set of verbs. We could jump on this fact and declare that they just all use a slightly different semantic criterion and none of them necessarily have anything to do with syntactic unaccusativity. However, apart from making some correlations extremely hard to explain, I think this approach misses the generalizations that do exist. My hypothesis is therefore that, to the extent that it can be shown that a diagnostic requires a syntactic explanation, it is possible that it imposes additional restrictions on the verbs or subjects that may particpiate in the relevant context. This implies that a diagnostic may identifY either all unaccusative verbs or a subset-if it is based on a syntactic principle it will not be too permissive. I will illustrate the effect of this phenomenon on the diagnostic context of bare po-phrases.

Maaike Schoorlemmer

220

The verbs in (27a) have so far been identified as allowing bare po-phrase subjects; those in (27b) show similar behaviour?

(27)

a.

b. c.

rasti ('grow), kipet' ('boil'), stojat' (,stand'), upast' (,fall'), toreat' ('stick out'); soxnut' ('dry'), drognut' ('shiver/freeze'), zamerznut' ('freeze'), sidet' ('sit'), bolet' ('be ill'), ldat' ('lie'); spotknut'sja (,trip'), obaldet' ('be/grow astounded'), ustat' ('be/grow tired').

The verbs in (27c) cannot be used with a bare po-phrase subject, despite the fact that they show up as unaccusative by other diagnostics and do not seem to differ in lexical semantics from the other ones in any significant way. It looks as though an additional factor limits the occurrence of bare po-phrases with intransitive verbs. As it is unclear what this factor is, let me just rule out the unique influence of a few potential candidates (see section 8.7 for some more discussion of the relevant factors): • it cannot be aspect, since both (pa, b) and (27c) contain perfective verbs; • it cannot be the verb's necessarily taking an animate subject, since this applies to all verbs in (27C) but also to most in (27b); • it cannot be the verb's necessarily taking a human subject, since this applies to obaldet', but not ustat' or spotknut'sja (all in (27c)); • it cannot be something in the verb's meaning that implies a lack of control in the action; this applies to the verbs in (27c), but also most of those in (27a) and (27b). For the time being, I conclude that the class of unaccusative verbs comprises at least those that allow bare po-phrase subjects. To conclude this discussion, let me summarize what I have done so far. The aim of the chapter is to provide additional evidence for Levin and Rappaport Hovav's thesis that unaccusativity is a syntactically expressed phenomenon, acquired on the basis of semantic regularities in (most of) the verbs involved. So far, we found that the distribution of bare po-phrases is impossible to explain by appealing exclusively to differences in lexical semantics distinguishing unaccusative and unergative verbs. I proposed that it is the lexical category or syntactic structure of the constituent in which the bare po-phrases occurs (as opposed to numeral po-phrases and other semiPPs) that does account for it.

7 Some speakers allow many verbs in this construction whose participation in the unaccusative class cannot possibly be confirmed independently, for example, kurit' ('smoke'), igrat' ('play'), risovat' ('draw'), pit' ('drink'), vjazat' ('knit'). My hypothesis is that these speakers do not really have distributive po at all. Whenever it looks as if they are using it, they use numeral po, in which the numeral odnomulodnoj ('one') has been elided. Being able to use optionally transitive verbs with po should therefore disqulifJ speakers from the distributive po-test.

Syntactic Unaccusativity in Russian

8.4.

221

TELIC ASPECTUALITY

In this second part of the chapter I tackle yet another syntax-related phenomenon and assess its value as an unaccusativity diagnostic for Russian: aspectuality and its effect on grammatical aspect in Russian. Again, the point will be that there is a systematic distinction between internal and external arguments that cuts across the event-participation of these arguments. I will show that it is possible to distinguish between lexically triggered perfective aspect and perfective aspect derived on the basis of compositionally derived telicity. In the latter case, internal arguments playa crucial role. If intransitive verbs show the relevant behaviour they must be unaccusative verbs whose subject is a syntactically internal argument. There is extensive argumentation in the literature that many instantiations of telic aspectuality rely on the presence in the sentence of an internal argument with particular quantificational properties. The idea is based on the observation that a sentence with a transitive verb can be telic only if at least the following conditions are fulfilled: (a) it contains an eventive verb; (b) there is a direct object present; and (c) the object is neither generic, nor a bare plural (+SQA in Verkuyl's 1972 terms). Following Verkuyl (1972, 1993) and Krifka (1989), I propose that aspectuality is a compositionally derived property, and that the quantificational properties of an internal argument partly determine its value. The two possible values are [telic] and [durative] (or [atelicl). For those verbs that can compositionally derive telic aspectuality, perfective grammatical aspect is derived in the absence of such interfering factors as progressive, habituality, etc. (see Schoorlemmer 1995). Observe that any exclusively lexical marking of aspectual properties misses out on the generalization that an object is required in most cases. This requirement has been found to extend to intransitive verbs. If unaccusative verbs have syntactic internal arguments, these are predicted to participate in the calculation of aspectuality, like objects of transitive verbs. This prediction is the basis of the reasoning in the next few sections. Here I will first argue for the existence in Russian of verbs that show perfective aspect despite the absence of compositional telicity or even a direct object that could participate in the derivation of it. I then go on to show the differences between intransitive verbs of this kind and perfective intransitive verbs whose perfective aspect does rely on the presence of a direct object that participates in the derivation of compositional telicity. Finally, I will argue that this latter class should be treated as unaccusatives on the basis of their syntactic properties-that is, the presence of a direct object that participates in the computation of compositional telicity. By introducing the distinction between lexical and compositionally derived aspect, this section will lay the groundwork for the two subsequent ones. Next, in section 8.5, I will argue that this approach allows us to use a specific type of aspectual morphology, which is restricted to compositionally derived aspect, as a diagnostic for unaccusativity. In section 8.6

Maaike Schoorlemmer

222

I show that under specific conditions, the underived counterparts of prefixed unaccusative verbs may be assumed to be unaccusatives as well.

8.4.1. Russian semelfactives as lexical perfectives Russian verbs differ as to whether or not they have a lexically marked value for perfective aspect. Many languages have a class of verbs that show perfective-like properties (in Russian, perfective aspect) without the aid of an internal argument. Among these are semelfactive verbs: verbs that express one instance of a potentially repetitive action. I claim that semelfactives are one of a number of verb classes in Russian whose lexical semantics directly trigger perfective aspect. I will concentrate on semelfactives because their special lexical semantics occurs in English too. Some English semelfactive verbs are illustrated in (28). (28)

a. b. c.

d.

The light flashed in an hour/for hours. I blinked in an hour/for hours. I coughed in an hour/for hours. He shrieked in an hour/for hours.

In English, nothing in the verb's morphology indicates whether it is intended in its semelfactive or repetitive reading. In their most natural interpretations, modification with in an hour forces the semelfactive reading, modification with for hours forces the repetitive reading. 8 In Russian, the two readings directly correlate with the verb's grammatical aspect.

(29)

a. b. c.

Ja morgnula. I blinked.pF Ja kaSljanula. I coughed.pF Onkrilmul zaminutu. he shouted.pF in minute

a'. b~

c'.

cas ami. Jamorgala I blinked.IpF for.hours casami. Ja kaSljala I coughed.IPF for. hours Onkriear casami. he shouted.IPF for. hours

Perfective aspect triggered by semelfactivity is usually marked with the suffix -nu-; such verbs lack 'perfectivizing' prefixes. There are good reasons to assume that this type of perfectivity is independent of the presence or properties of an internal argument. First of all, in many sentences with semelfactives, the test-context of ' in an hour/minute' gives very awkward results, from which I conclude that telicity is not involved in deriving perfective aspect with these verbs. The two logical possibilities that are open to us here are: • telicity has sources and diagnostics other than the ones discussed so far; • perfective aspect has sources and diagnostics other than telicity.

8 However, modification with in an hOllr is also compatible with the repetitive readings-see (29a), for instance. In this case, the interpretation is that the light started to flash (repetitively) after an hour.

Syntactic Unaccusativity in Russian

223

If we follow the first option, we conflate telicity and perfective aspect, which makes it impossible to pursue any further discussion of the two phenomena in isolation. In following the second option, my assumption is that telicity is a last-resort mechanism to provide a grammatical aspect value for a predicate-in the absence of lexical aspect, a telicity value is calculated which then determines grammatical aspect (see Schoorlemmer 1995). This approach is supported by the data in (30). To the extent that semelfactiveverbs are (optionally) transitive, there is no evidence that the internal argument plays any role in deriving temporal boundedness. (30)

a.

Vasja tolknul dver'. Vasja pushed door 'Vasja gave a push into (the) door.' Sobaka maxnula xvostom. dog wagged tail.1NsTR 'The dog wagged its tail once.' Vasja pnul masinu nogoj Vasja kicked car leg.1NsTR 'Vasja kicked (the car).'

b.

c.

Many of these verbs express movement with a body-part, which must be expressed in the instrumental case. Verbs such as torture, which cannot derive telic aspectuality, also take such instrumental objects. The presence of the objects here must therefore be assumed to have no aspectual consequences: the verbs are perfective or imperfective 'of their own accord'. I conclude that semelfactive verbs are lexically marked as such in Russian and other languages. In Russian, this marking directly triggers perfective aspect. I will refer to this as lexical aspect. With such verbs, no compositionality is involved in deriving the clause's aspectual properties. To summarize, Russian is assumed to derive perfective aspect on the basis of either of two properties (see Schoorlemmer 1995 for details and discussion): • directly, on the basis of a lexical aspect features such as semelfactivity ('once only); • compositionally, on the basis of telic aspectuality (which in turn depends on argument properties). I will refer to these different mechanisms as 'lexical aspect' and 'compositional aspect' respectively. 9 I want to mention some other lexical aspect triggers in Russian. Some classes of verbs have lexically marked imperfective aspect. Examples are members of the already mentioned 'torture' class, as well as pre- and suffixed verbs with such meanings as 'to V slightly now and again'. Apart from semelfactivity, there are other semantic features

9

In this discussion, I abstract away from the effects of progressive and habituality.

Maaike Schoorlemmer

224

that function as lexical triggers for perfective aspect, such as 'to V for awhile', 'to V for a long time', 'to begin/end V-ing' (marked with specific prefixes, see (36) and (37)). When I discuss specific properties of lexical aspect as exemplified by semelfactives, it should be borne in mind that the same story applies to these cases as well.

8.4.2. Telic on the basis of argument properties We have seen that there is at least one semantic class of verbs that can derive perfective aspect purely on the basis of its lexical properties. This, however, is not the entire story of perfective aspect in Russian. Many verbs do not have such lexical properties, and can only derive perfective grammatical aspect on the basis of compositional aspect. The relevant class of verbs includes both transitives and intransitives. Telic compositional aspect by definition requires the presence of an internal argument (see introduction to this section). Therefore, an intransitive verb showing perfective aspect on the basis of compositional aspect-that is, lacking lexical aspect-must have an internal argument subject; it must be unaccusative.IO I will now show that there is such a class of verbs, and that it lacks the properties attributed to lexical aspect above. First of all, compositionally telic intransitive perfective verbs are perfectly happy in the telicity diagnostic context 'in an hour/week'. This is illustrated in (31).

(31)

a.

b.

c.

Za sutki, stariki umerli ot bolezni. in day, old. guy died. PF of illness 'In a day, the old guys died of the illness.' Za minutu strel'by vse okna razbilis.' in minute o£shooting, all windows broke.pF 'Within minutes of the shooting all windows broke.' Za polgoda zasuxi ves' skot sdox. in half. year of. drought all cattle died.pF 'In six months of drought, all cattle had died.'

Secondly, as illustrated in (32), such intransitives do not show the semelfactive/ repetitive distinction with different aspectual forms. (32)

ti.

b.

c.

Starik umiral casami. old. guy died.IPF for.hours 'The old man was dying for hours.' Okna razbivalis' casami. windows broke for. hours 'For hours, windows were breaking.' Skot sdyxal nedeljami. cattle died for.weeks 'Cattle were dying for weeks.'

10

It also follows that all perfective unergatives are lexically perfective.

Syntactic Unaccusativity in Russian

225

This means that the imperfective does not necessarily imply repetition (as do the imperfective counterparts of semelfactive verbs), bur, for example, focuses on the duration of the event or describes repetition in terms of a large number of subjects. The perfective does not necessarily have the meaning of 'a single instance of a potentially repetitive action'. As in the case of ordinary transitive verbs, the aspectual forms are used strictly grammatically, with no lexical strings attached. Also, like most transitive verbs, these verbs show perfective prefixation and imperfective suffixation (cf. the examples in (41); see section 8.5 for more details). My claim is that these perfective verbs are perfective on the basis of compositional aspect, which in turn derives from the presence of a (+SQA) internal argument. II That means that if they can be perfective they must derive telic aspectuality on the basis of an internal argument. This claim makes some predictions regarding the co occurrence of aspect morphology and compositional telicity that I will elaborate on in section 8.5. In the remainder of this section, I present the argument against a purely semantic approach to telic aspectuality as a perfective-triggering mechanism.

8-4-3. A semantic approach to telic aspectuality My claim has been that there are cases of perfective aspect that do not rely directly on the properties of the relevant verbs, but that arise as a result of the presence of compositional telicity, which in turn relies on the quantificational properties of internal arguments. I will now briefly review the possibilities of accounting for the distribution and behaviour of perfective aspect in Russian purely on the basis oflexical properties of verbs. What we need in order to reformulate the need for an internal argument in semantic terms is a way of malcing this argument required input into a compositional mechanism computing the telicity value. Crucially, this mechanism should then not rely on syntactic position. Assuming that the computation of compositional aspect is either a syntactic or a LF mechanism, this alternative marking of the relevant argument must be present in syntax at the latest. Also, it cannot be just a lexical marking such as 'human' or 'property to which the visual cortex is sensitive', because those do not generally participate in either syntax or LF computations (although they might do in lexical or pragmatic mechanisms-see section 8.7). This leaves only a vety limited set of possibilities. Playing the devil's advocate, I will pursue the line of a lexical marking of the kind that syntax could be sensitive to (such as 'gender' or 'person').I2 Let me first concentrate on those transitive and intransitive verbs showing compositional telicity that seem to involve some form of change of state, a semantic component responsible for one of Levin and Rappaport Hovav's linking rules (Levin and

II Elsewhere, I have linked rhis property to the ability of the relevant verbs to derive complex event nominals (Grimshaw 1990), and to occur in the passive when transitive. Perfective verbs wirh lexical aspect show neirher of rhese properties (Schoorlemmer 1995,1997,1998). " And that language acquisition is probably sensitive to as well, see Levin and Rappaport Hovav 1995.

226

Maaike Schoorlemmer

Rappaport Hovav 1995: 146). In order to make the 'change-of-state' property of a predicate visible to syntax or LF, we would need a lexical marking either on the verb or on the argument. Notice that such a marking is necessary for both transitive and intransitive verbs. In the case of transitive verbs it is obvious that marking the verb is not good enough, because it would not allow us to choose between the two arguments. Let me therefore explore the possibility of lexical marking on the argument that actually expresses the change of state. In the case of build a house this seems pretty straightforward: the house is being built, and when it changes from being 'unfinished' to 'finished' the event ends-the object expresses the end-point. However, there are many cases where it is much harder to define which argument actually expresses the change of state. Take read a book, for example. The book does not change from being read-if anything changes it is the reader's brain. Or, alternatively, stage a play, where the play is not what changes, it is the performance, or the audience. The solution that has been proposed for this is to rephrase the problem in terms of 'measuring out', following Tenny's (1987, 1994) 'measure role'. The argument carrying the measure role is the one that measures out the change of state, for example in size (a book being read), in incrementality (a house being built), or in degree of readiness (a chicken being boiled). The event is delimited by the measure argument reaching the full measure. Translating 'measure' into 'change of state', we can say that in the case of build a house the argument that gets more and more finished is the one that expresses the change. In the case of read a book the book can be taken as the measurement, the factor that delimits the event of reading. But the problem of stage a play remains, and a new one presents itself with verbs of motion. In verbs such as carry, the subject and object move together, making it hard to decide which expresses the 'change of state'. In addition to this problem, we find that the 'measure' is a point in some implicit or explicit direction (a 'path') distinct from the argument that changes position. Crucially, however, the latter is still relevant fOl telicity:

(33)

a. The children went home in an hour. b. *Children went home in an hour.

The moved argument must not be generic in a telicity-inducing context, although in both cases there is a clearly defined local end-point. Now, if the relevant argument must be the one expressing the 'measure' in order to make the right predictions for clauses like read a book, then how do we know that with verbs of motion the argument relevant for telicity is the one that changes position? And even if this problem is solved, some cases remain whose potential telicity seems impossible to derive from the type of event-participant in a predictive way. I conclude that this type of semantic label on arguments cannot be the factor determining which argument participates in compositionally deriving telicity. Observe that these objections carryover to more recent approaches to the connection between lexical semantics and syntactic projection, by which arguments of verbs project internally or externally on the basis of the verb's lexical semantics (van Hout 1996 and in this

Syntactic Unaccusativity in Russian

227

volume), or even project independently of any projecting predicate, a predicate which is then interpreted as a modifier of a pre-existing configuration of arguments and corresponding aspectual properties (Borer, this volume). Alternatively, under a view based on linking rules, specific lexical information may be invoked to determine the internal or external projection of a specific class of intransitive verbs' subjects. In language acquisition, once the notion of internal and external projection of such arguments has become firmly established, this view allows room for exceptions to linking rules to occur-cases where the child establishes the argument structure of the verb entirely on the basis of its syntactic behaviour, separately from lexical semantics. Also, under this view we might expect a component of lexical semantics to playa role in the determination of the verb's syntactic properties that does not otherwise playa role in syntax at all. This prediction will be the topic of section 8.7. .

8.4.4. Telicity as an unaccusativity diagnostic I propose the following scenario. The semantic labels discussed above participate in a calculus that determines whether arguments project as internal or external arguments using linking rules. For example, all 'incremental themes' and 'measurements' project internally. Since internal arguments are a necessary condition for deriving telic aspectuality, the result of this calculus determines whether a clause may be telic or not. This predicts a crucial distinction between unergative and unaccusative verbs-only the latter derive perfective aspect on the basis of compositional aspect. This distinction can be very easily illustrated on the basis of English examples.

(34)

(35)

a. *John fell for hours.

b.

The baby slept for hours.

a.

John fell in an hour. The baby slept in an hour.

b.

Only unergative verbs can occur in atelic clauses modified by 'for hours'. Both types of verb occur in clauses modified by 'in an hour', but with different readings: the event expressed in the unaccusative verb has reached its completion, the one in the unergative verb has only started. In Russian, the difference between unergative and unaccusative verbs shows up in a different way as well. Compositional aspect, which requires the presence of an internal argument, is possible only with transitive and unaccusative verbs. In such cases, pairs of verb forms exist of either aspect which are totally synonymous; see (41), for example. Alternatively, unergative verbs can be perfective only on the basis of specific lexical properties that are absent in their imperfective counterparts, leading to pairs of perfective and imperfective verbs that are not totally synonymous. Synonymous pairs of verbs will be discussed in great detail in the next section; some non-synonymous pairs (other than semelfactive-repetitive ones) are provided in (36).

Maaike Schoorlemmer

228

(36)

a. b. c.

Imperfective

Perfective

smejat'sja 'laugh' smejat'sja 'laugh' igrat' 'play'

rassmejat'sja 'burst out laughing' otsmejat'sja 'stop laughing' poigrat' 'play for a while'

In all cases, the perfective variant of such non-synonymous pairs differs from the imperfective one by having been supplemented with some component of 'boundary': a beginning in (36a), an end in (36b), and a limited timeframe including a beginning and an end in (36c). In many cases, the distinction between synonymous and non-synonymous pairs is very hard to draw, particularly in the case of change-of-state verbs. Fortunately, as I will argue in the next section, compositional aspect has a morphological reRex in both transitive and intransitive verbs that is not found in lexical aspect. As a result, we can actually identifY unaccusative intransitive verbs on the basis of their morphological makeup.

8.5.

DERIVED IMPERFECTIVE AS AN

UNACCUSATIVITY DIAGNOSTIC The difference between lexical and compositional aspect is at least partially reRected in the morphological relationship between the perfective and imperfective forms of the verballexeme. I will show that suffixation to derive the imperfective variant of a prefixed perfective verbal form only occurs with compositional aspect. Again, the reasoning will be that, if this phenomenon is encountered in intransitive verbs, it means that their subjects are internal arguments-that is, they are unaccusative verbs. I3

8.5.1. Aspectual morphology The system of aspectual morphology in Russian is quite complex, which is one reason why finding a morphological clue for unaccusativity is not straightforward. I begin by looking at the morphology of perfective verbs.

'3 Some Russian intransitive verbs that follow this pattern take oblique dative complements instead of direct objects. Examples are vozrazit'/ vozraiat' ('object to'), pomoC'lpomagat' ('help'), and izmmit'/ izmenjat' ('betray). We might treat these cases in either of two ways: the dative object functions like a direct object for the derivation of compositional telicity, or the predicates are unaccusatives of the piacere class (Belletti and Rizzi 1988). In the latter case, it is the grammatical subject that participates in the derivation of compositional telicity, being an underlying object. I leave this matter open for now.

Syntactic Unaccusativity in Russian

229

As mentioned before, lexical semelfactivity is marked on the verb with the suffix -nu-, which also serves as the perfectivity marker with these verbs (see the examples in (45)). All other perfective verbs (lexical or syntactic) are derived with prefixes (except a very small number of underived perfective verbs). Examples of perfective verbs with lexical aspect are in (37), without it, in (38) and (39).

a.

b.

po-spat' pro-spat'

c.

za-revet'

a.

na-pisat'

b.

po-stroit' s-suCit'

c. d. a.

s-varit'

pere-pisat'

b.

vy-pisat'

c. d.

do-citat' za-cerknut'

'sleep for a while' dolgo 'sleep for a long time' 'start-to-bawl' 'write' 'build' 'twist, spin' , coole'

, , copy , subscribe to' , read to end' , cross out'

Prefixation itself therefore tells us nothing about the aspectual source of a perfective verb: it could be lexical or compositional. Prefixation of the kind viewed in (37) is analogous to what was found in (36) above: the addition of the affix results both in a semantic change and lexical perfective aspect. In the cases in (38), the prefix only signals perfective aspect, which results in a prefixed variant of the underived imperfective form that is synonymous with it. (4°)

a. b. c. d.

Peifective

Impeifective

na-pisat'

pisat' stroit' suCit' varit'

po-stroit' s-suCit' s-varit'

'write' 'build' 'twist, spin' , coole'

The perfective variants of these pairs, being derived without the addition of lexical semantics to their underived counterparts, must therefore be assumed to constitute compositional aspect. Morphologically, however, these cases are identical to lexical perfective aspect marked by prefixes. This is not the case in the third type of aspectual derivation. We now find prefixed perfective verbs that differ semantically from any existing underived counterparts, a situation similar to the one found in pairs of forms in which the perfective is derived by lexical aspect. There is an important difference, however, which is that in these cases, an imperfective form is derived from the prefixed perfective using suffixation, a so-called 'secondary imperfective'. This is true of the verbs in (39), as is illustrated in (41) .'4 14 Possible suffixes are: -yva-I-iva-, -va-, and -a-, which derive a uniform conjugation type for all secondary imperfective verbs.

Maaike Schoorlemmer

230

Perfective

(41)

a. b. c. d.

perepisat' vypisat' docitat' zacerknut'

Imperfective perepis-yva-t' vypis-yva-t' docit-yva-t' zacerk-iva-t'

,

,

copy 'subscribe to' 'read to end' 'cross out'

The secondary imperfective verb form has a very high degree of morphological regularity and semantic predictability: it is used only for deriving synonymous imperfective forms of prefixed perfective verbs. Being otherwise synonymous, these verbs therefore represent perfective and imperfective forms of verballexemes. We have seen that having a secondary imperfective counterpart is a signal that the verb derives perfective aspect using compositional telicity, which by our earlier definitions means that the verb projects an internal argument. As a result, any intransitive perfective verb with a secondary imperfective counterpart must be an unaccusative. Given the total synonymity of both verbal forms I also assume total identity of argument structure, which means that the secondary imperfective counterpart of an unaccusative perfective verb form is itself also unaccusative. Let me review the reasoning set up so far. If a verb participates in secondary imperfectivization it shows syntactically derived, compositional aspect. If a verb participates in compositional aspect it requires an internal argument with certain quantificational properties in order to derive perfective aspect. Therefore, if such a verb has a secondary imperfective form it must have an internal argument when perfective. Since there is no difference between the perfective and imperfective form except their aspect (just as there is no difference between sleeps and slept, apart from tense), this means that the verb takes an internal argument in general. If intransitive, this verb must then be considered an unaccusative verb. Phrased in this way, telic aspectuality in combination with secondary imperfectivization functions as an unaccusativity diagnostic. This diagnostic again identifies only unaccusatives, but not necessarily all. Verbs like those in (38) do not have secondary imperfectives. Semantically and syntactically, they behave like verbs with secondary imperfectives (as in (39)), but the trouble is that you cannot tell just by looking at the morphology. Morphologically, they pattern with semelfactives (and other lexical perfectives) in having underived, non-prefixed verbs as their imperfective 'counterparts'. We therefore have to allow for verbs that do not show secondary imperfective but that do participate in compositional aspect on the basis of an internal argument. Sometimes it is hard to decide whether a prefix introduces a lexical aspect-trigger or not, which is why I have talcen pains to establish a way of looldng at the morphology to tell us whether we are dealing with compositional aspect or not. It turns out that there are cases where the imperfective morphology tells us very clearly that we are dealing with compositional aspect. In those cases, whenever the verb is intransitive, we know it must be unaccusative.

Syntactic Unaccusativity in Russian

23 1

8.5.2. Identifying unaccusative verbs in Russian In the remainder of this section, I present the aspectual forms of a large number of prefixed intransitive verbs in Russian to show that they are unaccusatives. I will present them as members of morphological and semantic classes. Further subdivisions have to do with the semantics of the underived verbs underlying them, and will be elaborated on in section 8.6.2. In all cases, the forms are given in the order perfective-imperfective.

8.5.2.1. Intransitive change-ofstate verbs • Change of state ending in -nu/0This is a different -nu than the one encountered in semelfactives. Unlike semelfactive -nu-, it is always unstressed and is truncated in the masculine past tense form.

a.

b. c. d. e.

f

izdoxnut' vysoxnut' umolknut' uvjanut' promoknut' zamerznut'

'die' 'dry' 'fall silent, stop (noise)' 'wither, wilt, droop' 'get wet, soak' 'freeze'

izdyxat' vysyxat' umolkat' uvjadat' promokat' zamerzat'

vyzjabyvat' 'grow-cold'

vyzjabnut'

• Change of state ending in -et' (with e-conjugation)

(44)

(45)

a.

b. c. d.

obaldet' onemet' uspet' sozret'

obaldevat' onemevat' uspevat' sozrevat'

'be dumbfounded' 'grow dumb' 'manage, be on time' 'ripen, mature'

a. b.

zabolet' zapret'

zabolevat' zaprevat'

'fall ill' 'rot'

• Other intransitive change-of-state verbs. The verbs in (47) ending in -et' have a different conjugation than the verbs in (44), as illustrated by the different imperfective form. (46)

a.

b. c. d. (47)

a.

b. c. d.

sovpast' vyzdorovet' vyrasti sgnit' vskipet' sgoret' zaxvorat' zacvesti

sovpadat' vyzdoravlivat' vyrastat' sgnivat' vskipat' sgorat' zaxvaryvat' zacvetat'

'coincide' , recover , , grow up , rot'

,

'get to a boil' 'burn down' 'fall ill' 'flower, come into full bloom'

Maaike Schoorlemmer

23 2

(48)

a. b. c. d. e.

ustat' umeret' (u)past' opozdat' zastrjat'

ustavat' umirat' padat' opazdyvat' zastrevat'

'grow tired' 'die' 'fall' 'be/run late' 'get stuck'

8.5. 2 •2 . Verbs ofmotion Underived imperfective verbs of motion in Russian come in two types, so-called directed and undirected verbs of motion. [5

a. b.

xodit' idti 'walk (around, on several occasions)' 'walk (in one direction)' plavat' plyt' 'swim/float (in one direction)' 'swim/float (around, on several occasions)'

The directed verb idti means 'to walk uninteruptedly (in a particular direction)', undirected xodit'means 'to walk with interruptions and/or in different directions' .16 A simple illustration of the differences is given in the examples in (50). (50)

a.

b. c.

Vanja sel v skolu. Vanja walked.DIR toward school Vanja xodil po parku. Vanja walked. UNDIR around park Vanja xodil v skolu. Vanja walked.uNDIR toward school 'Vanja was a schoolboy, used to go to school.'

In (50a) a unidirectional path that is traversed in one occasion is described. The event described in the other two examples does not fulfil either of these criteria, and therefore the undirected verb is used: in (50b) because it is not unidirectional, in (50c) because it describes multiple events. Despite these differences, the verbs express the same manner of motion and the same (imperfective) aspect. Both types derive prefixed perfective forms, but they differ in semantics. Prefixed undirected verbs of motion retain the specific undirected lexical semantics (51). Prefixed verbs derived from directed verbs of motion cannot generally be said to be directed or undirected (52).

a. b.

xodit' 'walk around' plavat' 'swim/float around'

poxodit' 'walk around for a while' zaplavat' 'start swimming/floating around'

15 Although the discussion of verbs of motion in this subsection applies to both transitive and intransitive verbs, I will provide only intransitive examples. 16 The traditional definition of the distinction in terms of direction only is refined in PiftOn (1997) on the basis of Polish facts.

Syntactic Unaccusativity in Russian (52)

a.

b.

idti 'walle' plyt' , . , sWIm

233

vyjti 'go', 'go out' doplyt' 'arrive swimminglfloating'

Prefixed directed verbs of motion like those in (52) derive secondary imperfective verbs, as opposed to prefixed undirected verbs of motion. Observe the contrast in (53) (undirected verb stem) and (54) (directed verb stem). *poxazivat'J7 'walk around for a while' *zaplavyvat' 'start swimming around'

(53)

a. b.

poxodit' zaplavat'

(54)

a. b.

vyexat' vyezZat' vyplyt' vyplyvat' vylezt' vylezat'

c.

'drive, travel out' swim, float, sail out' 'climb, clamber out'

,

This contrast indicates a distinction in aspectual mechanism: prefixed directed verbs of motion show compositional aspect, whereas prefixed undirected verbs of motion show lexical aspect. The intransitive verbs among them can therefore be concluded to be unaccusative. Before concluding this section, let me briefly elaborate on the morphology of secondary imperfectivization in verbs of motion. The morphological derivation of secondary imperfective forms of verbs of motion very often involves stem suppletion, which means that instead of attaching the imperfectivizing affix to the prefixed (directed) verb, it is attached to the stem of the prefixed undirected counterpart; see (55a-c).J8 Sometimes the stem of the undirected counterpart simply replaces the directed stem (55d-e).

(55)

a. b. c. d. e.

vyletet' ypolzti ybdat' vyjti ybresti

vyletat' vypolzat' vybegat' vyxodit' vybrodit'

'fly out' , crawl out' , run out' 'walk out, leave' 'wander out'

Such pairs of aspectual forms do not behave in any way differently from ordinary pairs of forms expressing compositional aspect. I take this to mean that these irregular secondary imperfective forms do not constitute special forms of undirected verbs of motion, but secondary imperfective forms of prefixed directed verbs of motion. I therefore conclude that prefixed directed verbs of motion are unaccusative when intransitive.

17 This form does exist, but it differs semantically from the prefixed perfective in adding both habituality and diminutive semantics: 'to tend to walk around very slowly, in very small stretches'. It cannot be considered a secondary imperfective counterpart to any prefixed verb. 18 Attachment of the imperfectivizing affix to the undirected stem is reflected in a stressed final vowel, which in the case of (55b-c) contrasts to the stress on the underived undirected verb.

Maaike Schoorlemmer

234

In this section, I have argued for secondary imperfectivization as a morphological unaccusativity diagnostic that relies on the participation of internal arguments in the derivation of telic aspectualitywlth many verbs. This diagnostic has allowed me to identify intransitive verbs as unaccusatives that largely correspond to semantic classes of verbs displaying the same behaviour in other languages. These include change-of-state verbs, change-of-position verbs, and directed verbs of motion. I will return to the issue of semantic classes ofverbs showing unaccusative behaviour in the following sections. The secondary imperfective diagnostic can identifY only derived unaccusative verbs. In the next section, I address the question of the argument structure of underived counterparts of prefixed unaccusative verbs.

8.6.

ARGUMENT STRUCTURE AND DERIVATION

We have looked at prefixed intransitive verbs and identified certain groups of them as unaccusatives. However, there are many underived intransitive verbs for which this method does not apply. I will now argue that we can actually maintain the null hypothesis, that barring major semantic shifts the underived counterpart of a prefixed unaccusative is also unaccusative. I show that the derivation of a prefixed unaccusative verb from a non-unaccusative always gives rise to the addition of the reflexive-marker -sja. It follows that the null hypothesis can be maintained in those cases where the prefixed unaccusative verb does not carry the reflexive marker.

8.6.1. Underived verbs Two morphological procedures derive verbs from other verbs in Russian: prefixation and the addition to the verb of the reflexive marker -sja. Examples with prefixation are given in (39) and (56) (the prefixed verbs show compositional aspect), examples with -sja arein (57). (56)

a.

b.

stojat' (*ocered') stand (queue) wait your turn in a queue exat' (*goroda) travel (cities)

c.

dumat' (*plan) think (plan)

a.

stroit' build myt' wash

b.

stroit'sja be/get-built myt'sja wash-refl

vystojat'/vystaivat' ocered' out-stand .PF/rPF queue ob' exat' lob' ezzat' goroda around-drive.pF/rpF cities travel from one city to the next obdumat' I obdumyvat' .PF/rPF plan through-think plan

Syntactic Unaccusativity in Russian c.

vstretit' meet (someone)

235

vstretit'sja meet (reciprocal)

Both adding -sja and prefixation may be vacuous with respect to the verb's argument structure,1 9 but usually they are not. Observe that the two procedures operate in opposite directions: if anything, prefixation adds an argument, adding -sja in a sense reduces the verb's argument structure by turning a transitive into a passive, an anticausative, a reflexive, or removing the syntactic activity of the internal argument (see Gerritsen 1990). More generally, I will talk of reduction in argument structure in the case of any of the following changes in argument structure: reduction in number of syntactic or semantic arguments, demotion of their status with respect to a thematic hierarchy, change from external to internal syntactic status, or loss of syntactic activity. Let us tal{e a look at what happens when transitive verbs are input to prefixation. There are two options: they are still transitive after prefixation (compare (39) above), or they are not, in which case they are also affixed with -sja. Examples are given in

(58).

(58)

a.

citat' read

b.

govorit' talk, speak

c.

pit' drink

nacitat'sja much.read.sja 'get fed up reading' razgovorit'sja apart. talksja 'get carried away talking' dopit'sja do beloj gorjacki until.drinksja until delirium 'drink oneself into a delirium'

With all these verbs, the reflexive marker signals the derived verb's inability to express the internal argument of the related underived verb as a direct object. This generalization is absolute to the extent that there are no cases of underived transitive verbs turning intransitive by just adding a prefix. This can be formulated as follows:

(59)

a.

b.

Whenever prefixation derives an intransitive verb from a transitive one, -sja is added in addition to the prefix. v/rans ~ prefix_V/ntrans_*(_sja)

I now propose to extend to all verbs in Russian the generalization that reduction of the argument structure by prefixation leads to the addition of -sja on the derivate.

(60)

Whenever, as a result of prefixation, a verb's argument structure is reduced, -sja is added to the verb.

Observe that by this hypothesis, we predict that reduction in argument structure of intransitive verbs also leads to the addition of -sja. The derivation of an un accusative '9 Prefixation without argument structure effects is found in lexical perfectives such as those in (36); in cases like gl'ozit'lgrozit'sja ('threaten') the reflexive marker has no effect on the argument structure.

Maaike Schoorlemmer verb from an unergative one (where the argument structure is reduced from projecting a single external argument to a single internal argument) should therefore have the same effect. This hypothesis is expressed in (61). (61)

a. b.

Whenever an unaccusative verb is derived from an unergative one by prefixation, -sja is added in addition to the prefix. Viunerg => prefix_Vtnacc_*(_sja)

An example of this kind of reduction, which never occurs without the addition of the reflexive marker, is provided in (62). vrat' lie

zavrat'sja/zavirat'sja become entangled in lies, become a compulsive liar

Here, the prefixed derivate shows secondary imperfectivization in evidence of an un accusative status. There is no evidence for an unaccusative status of the underived imperfective verb. Recall that I am testing the following hypothesis: whenever we have an underived counterpart to an unaccusative prefixed verb, the underived (and semantically closely related) verb is also unaccusative. In order to test this, we want to be able to identify the following situation: the prefixed verb is unaccusative, but its underived counterpart is not. Observe that in such a case prefixation would be accompanied by a reduction in the verb's argument structure in the sense described in (60) and (61): the external argument 'turns' internal as a result of deriving the prefixed verb. My hypothesis is that argument structure reduction never occurs without the addition of -sja, whether the underived verb is transitive or intransitive. It follows that a situation that violates the null hypothesis would invariably result in the presence of -sja on the prefixed unaccusative. It also follows that, in the absence of -sja on the prefixed unaccusative, we are justified in concluding that the underived verb is also unaccusative. In some cases, it is not clear exactly what kind of 'reduction in arguments' is marked by -sja-sometimes it may not be a clear-cut reduction in number or kind. It is therefore a theoretical possibility that adding a prefix and -sja to an intransitive underived verb leaves the unergative or unaccusative status of the verb intact. Some potential examples of this are given in (63), where the prefixed intransitive verb does not have a secondary imperfective.

(63)

a. b.

begat' izbegat'sja run (undirected) exhaust oneself running vyt' razvyt'sja howl get carried away howling

For this reason, prefixed verbs that do show -sja are not included in the diagnostic. All we can say is that if -sja is absent and the prefixed verb is unaccusative, then so is the unprefixed verb.

Syntactic Unaccusativity in Russian

237

8.6.2. Identifying underived unaccusative verbs We now know that if a prefixed unaccusative verb without -sja has an unprefixed morphological base, the latter is also unaccusative. This allows us to add a large number of verbs to the inventory of unaccusatives provided in previous sections. We will now go over the groups of prefixed verbs identified as unaccusatives to see what their unaccusative root verbs look like and what their meanings are. 8.6.2.I. Root verbs denoting a change ofstate

• Some underived root verbs of change-of-state verbs in -nu/0- denote a change of state (see (42)). (64)

a. b. c. d. e.

f

izdoxnut' die (cattle) vysoxnut' dry umolknut' fall-silent uvjanut' wilt promoknut' get wet, soale zamerznut' freeze, cool

izdyxat'

doxnut' id.

vysyxat'

soxnut' id.

umalkivat'

?molknut' id.

uvjadat'

vjanut' id.

promokat'

moknut' soak

zamerzat'

merznut' freeze, be cold

• The same is true for root verbs of most change-of-state verbs ending in -e- (see (44))·20

(65)

b. c. d.

onemet' grow dumb uspet' manage, be on time sozret' ripen, mature

onemevat'

nemet' id.

uspevat'

spet'

sozrevat'

zret'

ripen id.

• The following other intransitive change-of-state verbs have change-of-state roots: (66)

a. b.

c.

vyzdorovet' recover vyrasti grow up sgnit' rot

vyzdoravlivat'

zdorovet' grow stronger

vyrastat'

rasti grow

sgnivat' id.

gnit' id.

20 The underived root verb of some verbs does not exist-for example, obaldet'lobaldevat', 'be dumbfounded': *baldet'.

Maaike Schoorlemmer 8.6.2.2. Root verbs denoting a state or situation

Root verbs of prefixed change-of-state verbs may also denote states or situations. Observe that under the argumentation presented in this section they must still be assumed to be unaccusatives (cf. (43), (45), and (47), respectively). vyzjabnut' freeze

(67) (68)

a. b.

(69)

a. b. c. d.

zabolet' fall ill zapret' rot

vyzjabyvat' zabolevat'

zaprevat'

vskipet' get to a boil sgoret' burn zaxvorat' fall ill zacvesti come into bloom

zjabnut' freeze, be cold

bold be ill pret' id. vskipat' sgorat' zaxvarivat' zacvetat'

kipet' boil goret' burn, be ablaze xvorar be ill cvesti Rower, be blooming

8.6.2·3· Root verbs denoting directed motion In section 8.5.2 I argued that prefixed verbs of motion with secondary imperfective counterparts are unaccusatives, whether the secondary imperfective form is regular or irregularly derived from the undirected stem. By the reasoning followed in this section, this means that the underived counterparts of the (perfective) prefixed verbs are unaccusatives. This in turn me;ll1s that, as in many other languages, directed verbs of motion are unaccusative verbs in Russian. The complete set of Russian underived intransitive directed verbs of motion is given here. (70)

a.

b. c. d. e.

f g.

vyjti walk out, leave vyexat' drive, travel out vyplyt' swim, Roat, sail out vybresti wander out vyletet' fly out vypolzti crawl out vybdat' run out, leave running

vyxodit' vyezzat' vyplyvat' vybrodit' vyletat' vypolzat' vybegat'

idti walk, go exat' drive, travel plyt' swim, float bresti wander letet' fly polzti crawl bezat' run

Syntactic Unaccusativity in Russian h.

vylezt' climb, clamber out

vylezat'

239

lezt' climb, clamber

In this section I have argued that whenever prefixation reduces the argument structure of the underived verb in any sense it is accompanied by the addition of the reflexive marker. As a result, deriving an unaccusative verb from a non-unaccusative will always lead to such an addition. The absence of the reflexive marker on a perfective intransitive verb form with a secondary imperfective counterpart therefore is evidence for unaccusativity of the underived underlying verb. This diagnostic has allowed us to identify verbs as unaccusatives that either do not show up in the po-phrase test (although some do) or that do not show secondary imperfectivization simply because they are not prefixed perfectives. The range of verbs that can be identified in this way is similar, but not identical, to those found in such languages as Dutch and Italian. In the next section I will concentrate on the semantic properties of Russian unaccusative verbs.

8.7.

LINKING RULES

I have concentrated on establishing the argument structure of Russian underived verbs. It follows that whatever linking rules we may derive from this materia}>' will be incomplete and possibly even inaccurate. However, I think that some patterns are obvious, and that some are extremely interesting in view of the recent debate (in this volume and elsewhere) on the status oflinking rules as proposed by Levin and Rappaport Hovav (1995). Under the proposal, linking rules are rules that determine whether, in the event, a particular participant will project syntactically as an internal or external argument. Under this conception of linking rules, as opposed to the direct linking approach to argument projection, linking rules should not exclusively rely on semantic properties used elsewhere in the grammar. Instead, we expect semantic properties to playa role in linking rules that might not have any other syntactic effect. In this section I will argue that such a linking rule must be assumed for Russian. As in English, intransitive verbs whose subjects undergo a change of state project internally. Also-and again this corresponds to the situation in other languages, notably Dutch and Italian-verbs of directed motion project internal argument subjects. The same applies to verbs of existence (see n. I) and verbs of simple position (see (27)). Especially for the latter, the semantic criterion for projecting the subject internally

n Unaccusative verbs that cannot be identified by the methods applied here, potentially are unaccusative verbs whose subjects nevertheless do not show up as bare po-phrases and that have the following morphological characteristics: they are the reflexive (inchoative or otherwise) counterparts of transitive verbs, andlor they derive perfective forms with semantically empty prefixes and as a result do not form secondary imperfectives.

Maaike Schoorlemmer

240

seems to be just that-this feature is not visible to the syntax: in a way that allows it to be used as a condition for other phenomena. This is also true for a specific class ofRussian unaccusative verbs, the cognates of which in other languages do not share their unaccusative properties. I will refer to them as 'essential verbs'. There is a class of verbs that projects internal argument subjects which cannot be characterized by previously established linking rules. I will refer to them as verbs expressing 'the nature of the essence of existence' or 'essential' verbs. They include the following (see (27) and section 8.6.2.2): (71)

a. b. c.

d.

bolet' be ill kipet' boil (liquid) goret' be ablaze zjabnut' freeze, be cold

xvorat' be ill, ail cvesti flower, be blooming pret' rot, stew

I will contrast them to a number of related classes and show that in each case it is the semantics of qualifYing existence that is crucial in deriving unaccusativity. The semantics of these verbs is very closely related to change of state, in that they all very easily derive change-of-state verbs through prefixation. However, they do not themselves express a change of state, as can be illustrated by their inability to be modified by the adverb postepenno ('gradually'); see the examples in (72b-c). Change-ofstate verbs do allow such modification, as shown in (72a).

(72)

a.

Deti postepenno nemelilzdoroveli/rosli. children gradually grew. silent/grew. healthy/ grew b. *Deti postepenno zjabli/boleli. children gradually were.cold/were.sicldy c. *Bljuda postepenno kipeli/goreli/preli. dishes gradually boiled/burn/smouldered

Like existential verbs, these are all non-volitional predicates. Unlike existential verbs, however, which assert the subject's actual existence, these verbs characterize the nature of the subject's existence. Subjects of essential verbs have a particular existence or life (when animate), as a patient, a flower, or boiling water (kipjatok), rotting, burning, or subsistence fighting the cold. Whereas existential verbs in most cases are stative, essential verbs are not. They are durative predicates, which very easily combine with temporal modifiers. In order to belong to the class of essential verbs, the verb must be able to describe the condition of non-human subjects. Thus, verbs like bedstvovat' ('to be poor, to lead a miserable life') or xuliganstvovat' ('to live, behave like a hooligan'), which can apply only to humans, do not qualifY. Observe that, in contrast, essential verbs expressing sickness may easily apply to humans, animals, and plants.

Syntactic Unaccusativity in Russian

24I

Verbs that express the way a subject vocalizes or makes noise fulfil this criterion, since they occur with human subjects as well as subjects like doors, teapots, birds, etc. Nevertheless, they are unergatives. Examples are sipet' ('hiss'), skripet' ('squeak'), and sumet' ('be noisy'). The distinction between them and essential verbs again lies in the property of the latter to qualifY the subject's entire existence, as opposed to noise verbs which concentrate on a particular property of the subject-not its existence. A further closely related class of verbs is the class of property verbs. Such verbs in Russian include xromat' (,limp, be lame'), okat' ('talk with a northern accent'), and zimovat' ('hibernate'), which are unergative verbs by our criteria. These verbs express properties that do characterize their subjects and may affect their lives, but they do not express properties that have been conceptualized as qualifYing the actual existence of the subject. This concludes my brief discussion of the class of essential verbs and the semantic property of qualifYing existence that derives unaccusative properties in Russian. Observe that in many cases, no distinction in the extralinguistic world can be made between the type of eventuality expressed by an essential verb and a non-essential verb. The distinction is all about conceptualization-in other words, whether the concept expressed by these verbs includes a particular property or not, independently of whether this is observable in reality. Semantic factors that have been argued to playa role in argument projection directlyare relevant for the characterization of this class of verbs. They are non-volitional, eventive, and they lack the expression of change of state. However, other factors could not possibly determine argument projection directly, as, for example, the verb's ability to talce a non-human subject. If the subject is human, there is no obvious way in which the verb's ability to take a particular type of alternative-that is, non-human-could determine whether it projects as an internal or external argument. Additionally, the difference between 'describing the subject' and 'describing the subject's existence' seems to go beyond these features, but does not add any that are otherwise relevant to syntax. The notion of 'qualifYing existence', although it plays a role in determining the syntactic behaviour of a verb expressing it, does not otherwise playa role in the grammatical system. It seems to be relevant to conceptualization of events in terms of verballexical semantics, but not to the linguistic expression of temporality, telicity, or any other related phenomenon. If argument projection is nevertheless sensitive to it, it can only mean that argument projection is driven by an explicit pre-syntactic mechanism.

8.8.

SUMMARY OF CONCLUSIONS

I have argued for an approach to unaccusativity based on syntactically internal projection of arguments of monadic verbs. This is in line with Perlmutter's original proposal as well as Levin and Rappaport Hovav's (I995) approach. The evidence presented

242

Maaike Schoorlemmer

consists of the ability of these arguments to be replaced with bare po-phrases and to participate in the derivation of compositional aspect. In either case, it cannot be maintained that the special unaccusative properties are due to the semantics of either the argument or the construction it occurs in. Bare po-phrases show different restrictions than other quantificational or even distributive expressions in Russian. Participation of an argument in deriving telic aspectuality in transitive verbs is uniquely restricted to direct objects, and cannot be related to the semantics of either the verb or the argument in a predictive way. What is certainly possible is that the correlations that do exist between lexical semantics and internal or external projection of an argument function as a trigger in language acquisition, allowing the learner to switch to other indicators (such as the existence of secondary imperfective forms) once the categories have been firmly established. The chapter also provides evidence to argue against the so-called direct-mapping approach, by which the unaccusative-unergative distinction manifests itself by a difference in syntactic projection of the subject argument, which is nevertheless directly conditioned by verbal lexical semantics. Under this view, we predict the relevant semantic feature also to show up in other syntactic phenomena, not just in determining argument projection in non-transitive verbs. No evidence for alternative effects is available, from which I conclude that the conditioning of argument projection proceeds through lexical linking rules as proposed in Levin and Rappaport Hovav (1995), and not directly. The other main conclusions are that both distributive po and secondary imperfectivization are solid diagnostics for unaccusativity in intransitive verbs in Russian. Further results are that on the basis of Borik's distinction, by which numeral po-phrases are distinct from bare po-phrases, numeral po-phrases can be identified as DPs. I have argued that similar quasi-PPs in other languages are also DPs. Also, we have been able to relate the unaccusative behaviour of prefixed verbs to unaccusative properties of the verb's morphologically and semantically underlying form. Finally, we have been able to confirm the validity in Russian of the semantic factors that are involved crosslinguistically in linking rules deriving unaccusative verbs: change of state, directed motion, simple position, and existence. A further factor has been identified that is referred to in a linking rule in Russian but not in many other languages, namely, qualification of existence.

9 Gradience at the Lexicon-Syntax Interface: Evidence from Auxiliary Selection and Implications for Unaccusativity AntoneLLa Sorace

9.1.

INTRODUCTION

The Unaccusative Hypothesis (UH), as a syntactic explanation of split intransitivity, has generated a large number of studies since it was first proposed (Perlmutter 1978; Burzio 1986). The original formulation of the hypothesis incorporates two claims. One is that the single argument of unaccusative verbs is an underlying direct object, and thus displays many syntactic properties of direct objects of transitive verbs; in contrast, the single argument of unergative verbs is a subject at all levels of representation, and thus displays the same syntactic behaviour as the subject of transitive verbs. The other claim is that the distinction is also systematically related to the semantic characteristics of the predicate: agentivity correlates with unergativity and patienthood correlates with unaccusativity (Perlmutter 1978; Dowty 1991). The UH is thus conceptually simple, elegant, and broad in scope, encompassing both the syntax and the semantics of split intransitivity; it assumes a relationship of almost complete

244

Antonella Sorace

predictability between the two, as expressed by the Universal Alignment Hypothesis (Perlmutter and Postal I984). However, the UH has recently been challenged on two fronts. First, linguistic theory has moved away from formal models of grammar that inspired the hypothesis (Relational Grammar and Government and Binding). Recent models of generative grammar do not include some of the fundamental tenets of the UH, such as the distinction between internal and external arguments (see the introduction to this volume for details). The most recent syntactic accounts of unaccusativity (Kayne I993, for instance) regard apparent auxiliary 'selection' as an epiphenomenon of a syntactic operation of incorporation of an abstract preposition, thus detaching it completely from the UH. Second, a vast body of empirical research (see Levin and Rappaport Hovav I995 for a review) has repeatedly shown inconsistencies in the alignment between the syntactic and semantic properties of split intransitivity: some verbs with similar semantics have different syntactic behaviour across languages (for example, blush is unaccusative in Italian but unergative in Dutch), and some verbs are classified as both unaccusative and unergative by the same diagnostic (for example, continuare 'continue' can take both auxiliary essere 'be' and auxiliary avere 'have' in Italian). It has become apparent that these 'unaccusative mismatches' are problematic only to the extent that one expects unaccusative and unergative verbs to represent syntactically and semantically homogeneous classes, as in the original formulation of the UH. Most of the syntactic diagnostics of unaccusativity and unergativity (such as auxiliary selection in Italian, impersonal passives in Dutch, resultative constructions in English) tend in fact to identify semantically coherent subsets of verbs within the unaccusative and unergative classes (Levin and Rappaport Hovav I995), suggesting that a proper explanation of these phenomena has to be placed at the syntax-semantics interface. From this perspective, the main endeavour of the theory of split intransitivity has thus become the identification of the syntactically relevant components of meaning in different languages and the search for an account of their interaction with the syntactic configurations in which a verb can appear. The principle is that neither a verb's ability to be found in the unaccusative or unergative syntactic configuration, nor the verbs' particular semantic characteristics are, by themselves, sufficient conditions to satisfy particular diagnostics: split intransitivity is 'syntactically encoded and semantically determined' (Levin and Rappaport Hovav I995).1

1 Several purely semantic models of split intransitivity have been proposed (such as Centineo 1986, Van Valin 1990, Cummins 1996, among others), which assume that a syntactic characterization of the distinction is unnecessary. However, much research has shown that a level of syntactic explanation is necessary to account for phenomena not easily reducible to purely semantic explanations: for example, the resultative construction in English (Levin and Rappaport Hovav 1995), scrambling involving quantifiers in Japanese (Tsuijmura 1990, 1991), the distribution of possessive and reflexive datives in Hebrew (Borer 1994), and necliticization in Italian (Belletti and Rizzi 1981). In alI these cases the distribution of the constructions that separate unaccusative from unergative verbs can be captured at the most general level by configurational factors, regardless of the semantics of the verb.

Gradience at the Lexicon-Syntax Interface

245

Various theories of argument structure (focused on the syntactically relevant lexical properties of verb arguments) and event structure (focused on the temporal and aspectual organization of the event described by a verb) which have been developed in recent years have set out to pursue the goal of explaining how lexical semantic or aspectual representations underlying individual verbs are mapped onto the binary syntactic representations underlying split intransitivity (Grimshaw I990; Pesetsky I995; Pustejovsky and Busa I995; van Hout I996; Rappaport Hovav and Levin I998, among others). Following Levin and Rappaport Hovav (I996), two main perspectives can be distinguished (see also Rappaport Hovav and Levin I998 for in-depth discussion): the 'projectionist' approach and the 'constructional' approach. The defining features of the two approaches will be briefly illustrated in the following section.

9.2.

PROJECTIONIST THEORIES

The projectionist approach maintains that lexical entries deterministically project onto syntactic positions according to universal linking principles which map particular arguments onto particular syntactic positions; this in turn produces the syntactic behaviour associated with unaccusativity or unergativity (Hale and Keyser I986, I993; Levin and Rappaport Hovav I992a, I994, I995, among others). Since unaccusativity and unergativity are lexical properties of verbs, verbs exhibiting variable behaviour are assumed to have different lexical-semantic representations, each of which is mapped onto syntactic representations in regular ways. One of the most comprehensive accounts of this type is Levin and Rappaport Hovav's (I995) model, in which a small number of linking rules map lexical semantic components of verb meaning (such as Immediate Cause, Directed Change, and Existence) onto positions at argument structure. The Immediate Cause Linking Rule is the only rule that maps the single argument of the verb onto the position of external argument. The Directed Change and the Existence Linking Rules map the argument onto the position ofinternal argument. The external and internal argument positions map 'trivially' onto the syntactic positions of subject and direct objects respectively, thus classifYing a verb as unergative or unaccusative. A number of problems are immediately apparent, of which only two will be mentioned here (for a full discussion see Sorace 2000). First, linking rules are language-specific: the existence Linking Rule, for example, produces unaccusative verbs in English and Italian (as shown by their ability to select auxiliary BE in Italian and appear in the there-construction in English, but not in Dutch or French):

(I)

I vampiri non sono mai esistiti. the vampires not are never existed 'Vampires never existed.'

(2)

There exist three versions of the manuscript.

(Auxiliary BE)

( There-insertion)

Antonelfa Sorace (3)

Die Dinosaurier haben/*sind wirklich existiert. the dinosaurs have/were really existed 'The dinosaurs really existed.'

(Auxiliary HAVE)

(4)

Il *estla ete al'universite. 'He is/has been at the university.'

(Auxiliary HAVE)

The cross-linguistic variation exhibited by stative verbs is in net contrast with the relative uniformity and invariance of verbs governed by the other two linking rules, a fact for which the model does not have an immediate explanation. Second, linking rules are too broad. No distinction is made between directed change and inherent telidty; even though there is an asymmetric relationship between the two. All telic verbs involve a directed change, but directed change does not necessarily imply telidty: degree-achievement verbs such as rise and cool imply an indefinite change in a particular direction but they do not denote the achievement of a final state. The asymmetry is relevant for split intransitivity because verbs of telic change behave differently from verbs of directed change in a number of languages. This can be illustrated by a comparison of verbs of directed motion in French and Italian. It is possible to distinguish four types of verb of movement (c£ Donadio 1996), according to the extent to which they express telidty, and whether they express it lexically or syntactically. A verbs denoting telic and inherently delimited movement: arriver ('arrive'); B verbs denoting directed, but not delimited, movement: monter ('rise'); C verbs lexically denoting atelic, non-directed movement, which can telidze compositionally in some contexts: courir ('run); D verbs denoting atelic, non-directed movement that cannot telidze in any context: vagabonder ('stroll'). The distribution of auxiliaries with these verbs in the two languages is summarized in Table 9.1, where E stands for essereietre andAfor avere/avoir.2 Only inherently delimited verbs of motion consistently select E in modern French, whereas auxiliary choice is variable for non-delimited verbs of directed change; E cannot be selected by verbs belonging to the other two classes. In Italian, both inherently delimited and directed non-delimited movement verbs select E, atelic verbs that telidze compositionally may take both auxiliaries depending on the context, and atelic verbs that cannot telidze in any context tal(e A. Inherent telidty, and not just directed change, therefore appears to be the determining factor in the consistent selection ofE in French. 3 2 Here the selection of auxiliary etre is taken as a marker of unaccusativity in French. This is somewhat controversial. There are other syntactic diagnostics of unaccusativity that have been reported in the literature which, arguably, identifY a larger class of verb. Legendre (1989) lists nine tests of unaccusativity, which single out distinct and only partly overlapping subsets of verbs. In her view, a verb is unaccusative if it satisfies at least one of these tests; conversely, a verb is unergative if it fails all the tests. Labelle (1992) discusses, in addition to auxiliary selection, six diagnostics (impersonal constructions, en-cliticization, infinitival relatives, tough construction, adjectival passives, and participial constructions). See Legendre and Sorace (in press) for an updated discussion. 3 Table 9.1 shows that Italian has more E-selecting verbs than French; see also Sorace (2000) and Legendre and Sorace (in press) for a fuller discussion.

Gradience at the Lexicon-Syntax Interface

247

TABLE 9.1. Distribution of auxiliaries for verbs of motion in Italian and French

Italian French

Telic, delimited

Telic Non-delimited

Atelic Can telicize

Atelic Cannot telicize

E E

E

EfA A

A A

EfA

Similarly, agentivity and internal causation are not differentiated in Levin and Rappaport Hovav's model, but there are languages in which agentive verbs are consistently unergative whereas non-agentive internally caused verbs fluctuate in their syntactic behaviour. For example, the so-called verbs of emission, which are internally caused but not agentive, are classified by Levin and Rappaport Hovav as 'variable-behaviour verbs' in English because they are basically unergative but behave like unaccusatives if they are interpreted as verbs of directed motion, as shown in (5). Agentive activity verbs never show this alternation-see (6). These verbs are also unstable in their selection of auxiliary in Italian, whereas agentive activity verbs are not: (5)

The curtains creaked open.

(Resultative construction OK)

(6) *John laughed down the stairs. (meaning: John laughed his way down the stairs) (7)

reco ha risuonato/e risuonato nella valle. 'The echo has resounded/is resounded in the valley.'

(8)

Paolo ha riso/*e riso fino a sentirsi male. 'Paolo has laughed/is laughed until he felt sick.'

(Auxiliary HAVE/BE) (Auxiliary HAVE)

Third, it appears that some of the rules can be arranged in order of priority: both the Directed Change and the Existence Linking rules take precedence over the Immediate Cause rule. But Levin and Rappaport Hovav offer no explanation of why such an ordering should obtain. A fourth, more general, problem with this type of projectionist model is that it is unable to account for variation without positing double entries in the lexicon and elaborated lexical rules which change the basic classification of verbs by mapping members of one class onto a different class. Verbs of manner of motion and verbs of emission, which are the only two classes showing rule-governed behaviour in the model, are basically unergative verbs in that they fall under the Immediate Cause Linking Rule, but they can become verbs of directed motion in the presence of particular adverbials and be reclassified as unaccusative. These verbs therefore have a double lexical-semantic representation, each corresponding to a distinct lexical entry that deterministically projects onto the syntax in a regular way. This solution is viable only as long as variable-behaviour verbs are regarded as the exception to a general pattern of deterministic mapping. If variation is the rule rather than the exception, a lexicon burdened with a proliferation of entries becomes uneconomical and unlearnable.

Antonella Sorace Finally, Levin and Rappaport Hovav's projectionist model is unable to predict which verb or verb classes are consistent and uniform in their mappings and which ones are variable. In their words, 'we do not have any explanation for the fact that only verbs from certain semantic classes can become verbs of directed motion or for the fact that they can become verbs of directed motion but not verbs of change of state. These explanations must await a full theory of possible and impossible meaning shifts.' (Levin and Rappaport Hovav 1995: 202).

9.2.1. Probabilistic versions The question of systematic variation in verb behaviour is addressed by other projectionist accounts in a probabilistic way. Dowty's (1991) analysis treats the thematic roles of Agent and Patient as clusters of semantic entailments, with no discrete boundaries. These inherently fuzzy concepts (called Proto-agent and Proto-patient) are characterized by open lists of semantic features (Dowty 1991: 572): • Proto-Agent: volitionality; sentence andlor perception; causer of an event or a change of state in another participant; movement; exists independently of the event named by the verb. • Proto-Patient: undergoes a change of state; incremental theme; causally affected; stationary; does not exist independently of the event named by the verb. Variation in the unaccusative-unergative classification arises from cumulative effects. Dowty in fact predicts that In any language which manifests unaccusativity, predicates that are 'high' in agentivity AND 'low in patient properties are invariably unergative, while those low in agent properties and high in patient properties are invariably unaccusative; only those high in both kinds of entailments, or low in both, should be unstable. (Dowty 1991: 608)

These predictions are imprecise and difficult to substantiate empirically: which verbs would be characterized by a high, or by a low, number of both agent and patient entailments? Are there verbs with two maximally agent-like, or two maximally patient-like properties? Tenny (1994: I02-5) demonstrates that such verbs, though not excluded in principle, are not attested. In an attempt to substantiate Dowty's predictions, Zaenen (1993) suggests that the crucial difference is not that between verbs high in agentive properties and low in patient-like properties, on the one hand, and verbs low in agentive properties and high in patient-like properties, on the other. Rather, what matters is the difference between verbs that have an equal number of proto-Agent and protoPatient entailments and verbs that have no entailments. The former, exemplified by verbs of inherently directed motion (which have both the agent property volition and the patient property incremental theme), are usually unaccusative; the latter, exemplified by verbs of emission (which, being stative, have no agent or patient properties; although they do have at least the property of being stationary), are usually unergative. However, there are two problems with this characterization of variation. One is that it

Gradience at the Lexicon-Syntax Inteiface

249

is unmotivated: it is unclear why there should be a relation between having an equal number of entailments and unaccusativity, or between having no entailments and unergativity. The other problem is that Zaenen's observations are correct for Dutch, but happen to be wrong for Italian, in which stative verbs denoting existence, including positional verbs in their 'simple position' sense, are unaccusative. As Zaenen suggests, Dowty's system does not pay sufficient attention to the notion of state, it being only one among the Proto-patient entailments: It is unlikely, however, that across languages, verbs like staan (to stand) and verbs like bloeden ('to

bleed') behave in the same way. In general it seems that among the stative verbs there are more semantic distinctions to be made than Dowty's list of properties allows for. (Zaenen 1993: 150)

As in Levin and Rappaport Hovav's model, what is still lacking is an explanation of why stative verbs tend to be variable across languages, whereas verbs denoting inherently directed motion are not. It seems that consideration of thematic factors alone, even when couched in probabilistic rather than deterministic terms, does not lead us sufficiently close to such explanation. As Grimshaw (1990) notes, Dowty's lists of entailments include both thematic and aspectual properties but do not consider how they combine with each other in determining argument realization. As she states, Dowty's proposal is a response to the failure of purely thematic theory to provide illuminating accounts for argument realization [... J. Nevertheless it is undesirable to retreat to a probabilistic theory [... J because some of the restrictions are absolure and can never be overridden. (Grimshaw 1990: 31)

9.3.

CONSTRUCTIONAL APPROACHES

'Constructional' approaches have gained ground in recent years as alternatives to lexical theories of linking (cf. Borer 1994, 1998; McClure 1995; van Hour 1996, 2000; Arad 1998a). These approaches have two main characteristics:

(a) unaccusativity and unergativity are considered to be a sentence-level property of the predicate, rather than a lexical property of the verb; (b) there is a closer and more immediate link between the aspectual interpretation and the syntactic configuration of unaccusative and unergative predicates: specific aspectual readings are determined by the appearance of the verb argument in particular syntactic configurations. In Borer's constructional model (1994, 1998, in this volume), lexical entries are 'bare': they contain only an unordered list of arguments. The 'core' (lexical) meaning of verbs serves as a modifier, rather than as a determinant of structural properties. Unaccusativity and unergativity become constellations of phenomena derived from the verb's ability to appear in particular syntactic configurations, which in turn determine aspectual interpretations. Syntactic structure and lexical specification are thus

Antonella So race divorced and may follow different developmental timetables in language acquisition (Borer 2003 and in this volume). A telic reading is derived by the presence of an argument in the specifier position of a functional projection labelled AspE; an activity reading is derived by the verb's appearance in the specifier position of AspP.4 Since the lexical entry of verbs does not contain any specification of whether an argument is internal or external, any verb is free to enter into more than one syntactic configuration and, consequently, to receive multiple aspectual interpretations. For example, an ambiguous verb like wilt may be interpreted as a telic, completed event (as in At that temperature, the plant wilted in a few hours) or as a process (The plant wiltedfor a few hours but then recovered): whether it receives one or the other interpretation depends on whether the verb argument is positioned in AspE or AspP. Optionality is therefore built into the system: the unaccusative-unergative classification of intransitive verbs is inherently unstable. This approach, unlike the projectionist model, predicts flexibility in the syntactic realization of arguments, but at the price of massive overgeneration. The problem is that variation is not unconstrained: some verbs can appear in only one configuration; others can appear in more than one but to different degrees. This is a concern for Borer herself It is thereby predicted that a verb such as run is perfectly ambiguous between a so-called unergative and unaccusative reading [... J this prediction appears problematic [... J clearly one would need to explain why some intransitives are much more susceptible to the MEASURE/NONMEASURE alternation than others. A possible explanation may be found in the appropriate characterization of particular verbs and their contribution to the meaning of the predicate in which they are embedded. Specifically, it may be that the meaning o/some verbs entails delimitation much more strongly that other verbs. (Borer 1994: 32; emphasis added)

What mechanism can, then, prevent such verbs as arrive from ever appearing in an unergative syntactic configuration? Constraints on overgeneration have to be present to rule out impossible matches, but this type of model does not specifY exactly how such constraints operate and at what level. The model, furthermore, focuses only on the aspectual distinction between events and processes, but is completely silent on stative verbs which, as seen earlier, tend to be inherently ambiguous in many languages.

9.3.1. Feature-checking versions Different versions of the constructional approach, closer to the Minimalist model, incorporate a feature-checking component. The assumption in this case is that aspectual features of the predicate, such as telicity, have to be discharged through the 4 Other proposals use different labels for what essentially is the sarne specifier position. For example, Agro in van Hout (1996) and den Dikken (1994), APinn" in McClure (1995). These proposals share much common ground with recent, purely syntactic, theories of auxiliary selection (Kayne 1993; also Cocchi 1994), in which the choice of auxiliary with intransitive verbs also hinges on the presence vs. absence or inactiviry of an AgrObjP projection (which is, however, not characterized as the locus of aspectual interpretation).

Gradience at the Lexicon-Syntax Interface

251

movement of arguments to some specifier of a functional projection. In van Hour's approach (1996, 2000, and in this volume), the lexicon-syntax mapping system is sensitive to event types, instead of deriving them as the outcome of lexical or syntactic operations on the arguments and their positions. This model shares with Borer's the assumption that the mapping system is defined on the event structure of the whole VP in which the verb appears, rather than projecting up from purely lexical properties of the verb alone. What characterizes van Hour's model is the requirement that unaccusative verbs incorporate telicity, which is introduced in the syntactic computation as an interpretable feature that needs to be checked in AgrOP, thereby triggering movement of the object to the Specifier of AgrO. This model therefore is, at least in principle, more constrained than Borer's, since telicity checking requires that the verb's (or predicate's) event-type properties must match or be compatible with the telic feature; it is unclear, however, how compatibility is checked and how incompatible matches are ruled out. Moreover, nothing is said about how mapping works with stative verbs.5

9.4.

PREDICTIONS FOR LANGUAGE ACQUISITION

Language acquisition, of both first and second languages, is an important testing ground for theories of the lexicon-syntax interface. An explanatory theory, in fact, should be powerful enough to account for the rapid acquisition of interface constraints, and particularly for the fact that language acquirers are inherently conservative and tend to make generalizations from which they are able to retreat. Projectionist and constructional theories of the lexicon-syntax interface malce different predictions for language acquisition, which are not always easy to distinguish empirically on the basis of current methodologies. While an exhaustive review of research in this area is beyond the scope of the chapter, the main positions may be summarized here as follows (see Borer, in this volume, for a more detailed overview).

9.4.1. The projectionist view: semantics has developmental priority Projectionist theories assume that children are predisposed to notice the syntactically relevant meaning components of verbs through their interaction with the environment; and that they also have lmowledge of the set of (potentially universal) linking rules that map semantic components onto syntactic positions. Children are therefore endowed with innate lmowledge of the shape of both semantic and syntactic

5 Van Hout's model may be modified in such a way as to incorporate rhe notion of underspecification. Verbs may be lexically specified or unspecified for telicity: a configuration with AgrO would unambiguously attracts verbs with a [+telic] feature, whereas underspecified verbs (statives, for instance) may project in more rhan one way (van Hout, personal communication).

Antonella So race representations, and of the constraints that rule the interface between the two. The combination of semantic knowledge and linking rules allow them to break into the syntactic system: this position has become lmown in the literature as the 'semantic bootstrapping' hypothesis (see Pinker 1989 and Gleitman 1990, among many others). The hypothesis leaves a substantial amount of learning for children to be accomplished: the semantic components of individual verbs, the verb structures permitted by the ambience language, the verb (sub)classes to which lexical rules apply, and the morphosyntactic expression of lexical alternations, all have to be learned on the basis of exposure to the language. Intriguing learnability questions arise. Consider the subcategorization requirements of locative verbs in English, which sometimes alternate between two different argument structures (I'm stuffing the turkey with breadcrumbs; I'm stuffing breadcrumbs into the turkey) and sometimes do not (I filled the jar with cookies; *1filled cookies into the jar). Or the well-Imown case of transitivity alternations, in which only a particular semantically defined subset of transitive verbs have intransitive alternants, that is, allows the possibility of mapping the Theme argument onto the subject position: the price increased is well-formed, but the paper cut is not. How does the child figure out which alternations are possible and which are not? The problem is quite complex: alternations with non-alternating verbs simply do not occur, so in the absence of negative evidence the child will have no reliable indication that they are disallowed; furthermore, verbs do not consistently occur with all arguments, since some arguments are optional in both alternating and non-alternating verbs. Unless it is assumed that the child actually keeps track of the non-occurrence of certain alternations, one has to conclude that the child comes to the task already equipped with knowledge of the possible ways in which human languages can organize meaning in lexical categories .. Some projectionist models assume an in-built bias to tryout hypotheses in a given order. In Pinker's version of the model, for example, children first acquire general, broad-range linking rules based on the basic semantics of the event; these rules are later re-analysed with the incorporation of finer semantic distinctions and narrowrange rules are added, in such a way as to differentiate the possible, narrow-range domains of application from the impossible ones. There is evidence that some meaning components have a privileged status over others: for example, children display an earlier sensitivity to 'change of location' than to 'change of state' (see Pinker 1989 and Gropen et al. 1991 for details). However, the origin oflinking rules and of their underlying semantic primitives is still an unresolved question (Gleitman 1990). For L2 acquisition, the relevant research questions revolve around the well-Imown distinction between universal developmental paths and transfer: can adult L2 learners acquire properties of the L2lexicon that are not deducible from the input and that cannot be transferred from the LI? Research on the acquisition of lexical alternations in a second language has shown that the problems encountered by L2 learners in the initial stages, and the overgeneralizations they mal(e, are consistent with those found in LI acquisition, regardless of the native language (see Juffs 1996 and Montrul 1997

Gradience at the Lexicon-Syntax Interface

253

for overviews). Both Lr and L2learners have to acquire universal argument structure alternations, the narrow semantic constraints operating on them, and the morphological expression of the alternation instantiated by the target language. While the 'syntactically relevant' components of meaning are determined by Universal Grammar, the morphological expression of alternations is language-specific. The former are acquired through the same developmental path across languages, both in Lr and in L2 acquisition, whereas the latter are initially approached by L2 learners through the particular analyses imposed by the native language. For example, languages that do not mark the transitivity alternation with overt morphology, such as English, are more difficult to acquire than languages, such as Spanish or Turkish, that differentiate members of this alternation morphologically (Montrul r997). Similar developmental patterns are found in both children and adults. One such pattern is an asymmetry in the directionality of overgeneralization errors involving transitivity: there are more causative errors involving the causativization of intransitives (lm going to disappear the ball) than anticausative errors, involving the intransitivization of transitives (The machine will fix). Montrul's account of this asymmetric pattern of overgeneralization is partly consistent with Pinker's (r989) distinction between broad- and narrow-range rules: adult L2learners, like Lr acquirers, start from a default transitive template, which represents all the basic sub events and semantic primitives (causative, inchoative, and stative) and the canonical realization of arguments. This template is initially applied to non-alternating verbs. At a later stage, learners have to learn the specific aspects of meaning (i.e. external causation and unspecified agency) that characterize alternating verbs: in Pinker's terms, the narrowrange rules that further constrain transitivity. Knowledge of these narrow-range rules involve 'turning off the CAUSE node from the basic template only for verbs that instantiate these meaning components. These convergences may be interpreted as evidence that L2 learners do not start from their Lr, but somehow revert to default universal principles. They indicate that L2 learners, again like Lr acquirers, are initially constraineq. by syntactic principles (i.e. the canonical alignment of thematic roles with syntactic positions that the default template represents), rather than purely semantic ones (such as the relative simplicity of intransitive forms, which lack the subevent CAUSE, compared to transitive forms). Hence, the L2 data may be construed as compatible with both a projectionist and a constructional account. What seems established is that both Lr and L2 learners rely on lmowledge of universal syntax-semantics correspondences; L2learners have the Lr as an additional source of hypotheses, but they use it in a conservative and modular fashion.

9.4.2. The constructional view: syntax has developmental priority Constructional theories (so far applied only to Lr acquisition) turn the relationship between syntactic and lexical knowledge upside down (see Borer, in this volume, for a detailed illustration; Gleitman r990). Rather than acquiring the syntactic expression

254

Antonelfa So race

of verb arguments from knowledge of their syntactically relevant meaning components, children deduce word meaning from the semantically relevant syntactic structures associated with verbs. If the learner uses in-built knowledge of subcategorization frames and of basic syntax-semantic relations, the range of configurations in which a verb appears may narrow the hypothesis space for acquiring the verb meaning: this is, in essence, the syntactic bootstrapping hypothesis. For example, verbs denoting change of possession tend to appear in sentences with three NPs, which canonically express the old possessor, the new possessor, and the entity that goes from one to the other. Verbs denoting perception and cognition are associated with sentences with an object NP, representing specific entities that can be perceived, or with sentential complements, representing whole events. There is evidence that visually impaired children acquire lexical and semantic distinctions underlying verbs of vision (look and see) and colour terms, which they could not possibly acquire through experience (Landau and Gleitman 1985). These authors suggest that the syntactic contexts in which the verbs are used are the most probable basis for their acquisition, along with innate syntactic and lexical categories. The role of experience does not seem to be a crucial one. Recent constructionist models (see Borer, this volume) reinterpret the evidence from child overgeneralizations in terms of imperfect lexical learning. Lexical verb meanings act as modifiers of the aspectual meaning carried by the merger of the verb with a specific functional head: thus, they are gradually memorized as vocabulary lists. The discrepancy between syntactic lmowledge (which is in place early on) and lexical lmowledge (which takes more time to develop) suggests a 'dual mechanism' model of the kind envisaged by Pinker and collaborators for the acquisition of past tense forms (see Marcus 2000 for a comprehensive overview): children make more overgeneralizations than adults because they have shorter vocabulary lists and weaker memory traces. When memory fails them, children will resort to the more general computational, rule-governed behaviour which consists of projecting verbs onto any syntactic configuration: so they produce Daddy disappear rabbit because they have not been exposed to a sufficient number of exemplars of this verb to lmow that it does not allow this type of projection in English. It is unclear, however, why only certain kinds of configurations are generalized by children-why is the transitive configuration more likely to be extended to intransitive verbs rather than the opposite? Can statistical frequency be the only explanation? Constructional models, in their most extreme form, predict protracted and unconstrained projection errors in language learners, just as they predict massive variation in the syntactic behavious of verbs. It seems plausible to assume, with Gleitman (1990), that semantic and syntactic bootstrapping do not exclude each other, and that the evidence from acquisition is not entirely consistent either with a purely projectionist or with a purely constructional model.

9.4-3- Summary To review so far, both projectionist and constructional approaches to split intransitivity attempt to explain the fact that verbs may vary in the syntactic realization of their

Gradience at the Lexicon-Syntax Interface

255

arguments. Projectionist models do so either by assuming lexical operations on the lexical-semantic representation of verbs, which create multiple lexical entries, or by positing probabilistic mappings based on a number of semantic entailments incorporated by the verb's lexical entry. These explanations have two disadvantages: they do not explain why certain verbs are more susceptible to variable behaviour, and they lead to a proliferation of lexical entries. The common denominator of constructional approaches is their emphasis on syntax as the main determinant of interpretation, rather than the interpretation affecting syntactic behaviour. The verb's syntactic specification is reduced to a set of simple features that have to be discharged in canonical checking positions, or to unstructured sets of arguments. While freeing the lexicon from syntactic specification allows more flexibility in mapping, it cannot deal with the fact that some verbs do not exhibit flexibility at all. A problem shared by both the projectionist and the constructional approaches is their relatively limited empirical basis. Data from language acquisition at the present stage do not unambiguously support either approach. Recent research on auxiliary selection has begun to fill this gap, and it is this research that will now be examined in detail.

9.5.

THE AUXILIARY SELECTION HIERARCHY

An empirical challenge to both the projectionist and the constructional views has come from a series of studies by Sorace and her collaborators (Sorace 1993a, b,I995b, Keller and Sorace 2003; Sorace and Cennamo 2000). The starting point of these studies is the set of facts that are long-established in the literature: (a) across languages, some verbs tend to show consistent unaccusative-unergative behaviour, whereas others do not; (b) within languages, some verbs are invariably unaccusative-unergative regardless of context, whereas others exhibit variation. These studies provide supporting evidence for these generalizations, mostly based on experiments testing native speakers' intuitions about auxiliary selection (perhaps the best lmown diagnostic of unaccusativity) in various languages that have a choice of perfective auxiliaries (such as Dutch, German, French, Italian, and Paduan). In all these languages, unaccusative verbs tend to select the counterpart of English auxiliary be and unergative verbs tend to select the counterpart of auxiliary have. However, native intuitions on auxiliaries are categorical and consistent for certain types of verb, but much less determinate for other types. The suggestion is that the systematic differences within the syntactic classes of unaccusative and unergative verbs may be captured by a hierarchy in which 'core' monadic verbs are distinguished from progressively more 'peripheral' verbs. This hierarchy is shown in (9). 2000;

256 (9)

Antonella Sorace The Auxiliary Selection Hierarchy CHANGE OF LOCATION

Selects BE (least variation)

CHANGE OF STATE CONTINUATION OF A PRE-EXISTING STATE EXISTENCE OF STATE UNCONTROLLED PROCESS CONTROLLED PROCESSES (MOTIONAL) CONTROLLED PROCESS (NON-MOTIONAL)

Selects HAVE (least variation)

Verbs at the extremes of the hierarchy ('core' verbs) are change oflocation verbs at the BE end and non-motional process verbs at the HAVE end. They are characterized by the following properties: • categorical and consistent syntactic behaviour across languages; • consistent behaviour within individual languages; insensitivity to compositional properties of the predicate; • determinacy of native spealcers' intuitions; • primacy in acquisition; • diachronic stability; Let us examine some evidence in support of these properties.

9.501. Core verbs Core verbs tend to be categorical and consistent in auxiliary selection across languages/ language varieties. This is exemplified in (10)-(II), which show that the auxiliary selected by change of location verbs in the present perfect is BE, and that selected by nonmotional process verbs is HAVE, in all the languages that have a choice of auxiliary.

(10)

a.

b. c.

d. e.

(II)

a.

b. c.

Paolo e venuto/*ha venuto all' appuntamento. Paolo is come/has come to the meeting Ma seeur est arrivee/*a arrive en retard. my sister is arrived/has arrived late De brief is/*heeft aangekomen. the letter is/has arrived Der Zug ist/*hat spat angekommen. the train is/has late arrived Maria est!*at arrivata a domo. Maria is/has arrived at home delegati hanno parlato/*sono parlati tutto il giorno. the delegates have talked/are talked whole the day Les ouvriers ont travailIe/*sont travailles toute la nuit. the workmen have worked/are worked whole the night De trompettist heeft/*is met bolle wangen geblazen. the trumpeter has/is with all his might blown

(Italian) (French) (Dutch) (German) (Sardinian)

Gradience at the Lexicon-Syntax Interface

d.

257

Kurt hat/*ist den ganzen Tag gearbeitet. Kurt has/is the whole day worked

Core verbs display consistent behaviour within languages; in particular, they tend to select the same auxiliary regardless of the contribution of other aspectual or thematic elements in the sentence in which they appear. So in (12) arrivare ('arrive') selects BE even though the predicate is atelic; cadere ('fall') in (13a) selects BE despite the fact that the event described by the verb clearly denotes intentionality, just as it does when the event is clearly unintentional (13b). Conversely, lavomre ('work') selects HAVE regardless of the telicity of the predicate, as in (14). (12)

a. b.

(13)

a.

b.

Sono arrivate lettere in continuazione. are arrived letters continuously Sono apparse decine di imitazioni per anni are appeared dozens of imitations for years Maria e caduta apposta per non andare a lavorare. Maria is fallen on purpose to not go to work 11 bicchiere e caduto dal tavolo. the glass is fallen from-the table I poliziotti hanno lavorato fino all'alba. the policemen have worked until the dawn

(Atelic predicate)

(Agentive) (Non-agentive) (Telic predicate)

1he data from studies on other languages (e.g. Paduan; see Sorace and Cennamo 1999) confirm that, in general, inherent lexical aspect determines auxiliary choice with core verbs, whereas compositional aspect (the event structure of the whole predicate) affects auxiliary selection with peripheral verbs. 1he data support the conclusion that auxiliary selection with core verb types is a lexical phenomenon and is relatively insensitive to compositional factors. 1he degree of sensitivity to these factors increases for non-core verb types as they get more distant from the core. Core verbs tend to elicit categorical intuitions from native speakers of languages with auxiliary selection, who categorically accept sentences in which these verbs appear with the 'correct' auxiliary and reject those in which they appear with the 'wrong' auxiliary. Evidence of differential judgements is particularly strong for Italian (see Sorace 1993a, b, 1995b; Bard, Robertson, and Sorace 1996 for experimental evidence). Furthermore, descriptive studies ofItalian (Berruto 1987 and Rohlfs 1969, for instance) indicate that there is more variation in auxiliary usage for peripheral verbs than for core verbs, which is consistent with the predictions of the hierarchy. Supporting evidence also comes from Germanic languages. Experiments on Dutch (Sorace and Vonk 1998) show orderly gradience in the judgements of native Dutch spealcers on zijn and hebben largely corresponding to the intransitive hierarchies identified for Italian. In addition, they show that the acceptability of impersonal passives (a construction traditionally regarded as a diagnostic of unergativity) is affected by semantic factors, particularly agentivity, which cut across the unaccusative-unergative distinction (a fact that had already been established by Zaenen 1993). For German, Keller

Antone/la So race and Sorace (2002) provide similar findings for native judgements on sein, and haben, and also show that inter-dialectal variation in auxiliary usage between Northern and Southern varieties is mostly found with peripheral (but not with core) verbs. Core verbs are the first ones to be acquired with the correct auxiliary both in firstand second-language acquisition. Data from the acquisition ofItalian as a non-native language show that the syntactic properties of auxiliary selection are acquired earlier with core verbs and then gradually extended to more peripheral verb types (So race 1993a, 1995a). Moreover, Italian learners of French find it more difficult to acquire avoir as the auxiliary for verbs closer to the core than for peripheral verbs (So race 1993b, 1995b), and do not completely overcome this difficulty even at the advanced level. These developmental regularities can be explained by assuming that the acquisition of the syntax of unaccusatives crucially depends on the internalization of two elements: one is the hierarchical ordering of meaning components, and the other is the lexiconsyntax mapping system instantiated by the target language. 6 The pattern uncovered by these data is consistent with an enriched constructional model, equipped with a checking mechanism that is sensitive to the degree oflexical specification ofverbs. As it is the position of verbs on the ASH, rather than their frequency, which determines the order of acquisition, it seems that L2 learners do rather more than engaging in the kind of statistical learning envisaged by a basic constructional model. Finally, core verbs tend to be diachronically stable. There is evidence from studies on the historical development of auxiliaries in Romance (Benzing 1931; Tuttle 1986, for instance) showing that core verb types tend to be the last to be affected by the replacement of EssE-reflexes by HABERE-reflexes, whereas peripheral verb types are the most vulnerable to the change. A recent study by Cennamo (1999) suggests that the development of reflexives selsibi in Late Latin as markers of split intransitivity followed a path largely consistent with the unaccusative-unergative hierarchies.

9.5.2. Intermediate (non-core) verbs While core verbs are categorical in their auxiliary selection behaviour, non-core verbs show increasing variation. The greater flexibility of these verbs will be illustrated here by Italian examples (for cross-linguistic evidence see Sorace 2000). 6 The primacy of (overt/inherent) telicity characterizing change-of-location verbs is also shown by Dutch children who, even at 5 years old, differ markedly from the adults in their interpretation of full transitives, allowing more atelic readings than adults do. A sentence is telic only if it includes an overt marker of telicity such as a particle. A quantized direct object on its own is not sufficient (van Hout 2000).

(r)

a. b. c.

Heeft de rode muis kaas gegeten? has the red mouse cheese eaten Heeft de rode muis zijn kaas gegeten? has the red mouse his cheese eaten Heeft de rode muis zijn kaas opgegeten? has the red mouse his cheese eaten up

Children learn the overt and transparent telicity markers before the more indirect ones.

Gradience at the Lexicon-Syntax Interface

259

A class that exhibits regular alternations is that of verbs denoting 'indefinite change' in a particular direction (such as rise), change of condition (wilt), and appearance (appear). BE is strongly preferred by these verbs, but HAVE is not completely rejected. The strength of preferences is a function of the (±) inherent telicity of the verb: as the Italian sentences in (15)-(16) show, many of these verbs allow two readings-one telic and one atelic-which may be disambiguated by the context.

(15)

a.

b. c.

d. (16)

a.

b. c.

La popolarita del governo e scesa/ha sceso the popularity of the government is gone down/has gone down notevolmente. noticeably La pianta e cresciutal?*ha cresciuto molto da quest'inverno. the plant is grown/has grown a lot since last winter II fantasma e apparso/?*ha apparso in soffitta. is appeared/has appeared in the attic the ghost E' successa/*ha successo una tragedia. is happened/has happened a tragedy :Lalbero e/ha fiorito due volte quest'anno. (Indefinite change) the tree is/has blossomed twice this year Le arance sono marcite/hanno marcito al sole. the oranges are rotten/have rotted in the sun II grano e/ha finalmente germogliato. the plant is/has finally blossomed

Verbs denoting continuation of a pre-existing condition (such as stay) are less determinate: BE is preferred but HAVE is not rules out categorically, and is in fact accepted with many of these verbs. The agentivity of the subject correlates with the degree of acceptance of HAVE, suggesting that these verbs, unlike core verbs, are sensitive to the feature contributed at the predicate level. (17)

a. b. c.

d. e.

f

Ancora una volta sono/?ho rimasto senza soldi. again one time am/have remained without money La discussione eJ?ha durato a lungo. (Non-agentive) the discussion is/has lasted for long II primo ministro e/ha durato in carica tre mesi. (Agentive) the prime ministert is/has lasted in post three months I miei genitori sono/?hanno sopravvissllto alia guerra. my parents are/have survived to the war Questa situazione e/?ha persistito per troppo tempo. this situation is/has persisted for too long Gianni *e/ha persistito nel suo atteggiamento. paolo is/has persisted in his attitude

Antonella Sorace

260

Stative verbs (including both verbs of physical and abstract existence and psychological verbs) are the most indeterminate in Italian, consistent with the findings from other studies. Auxiliary alternations (some restricted to regional or non-standard varieties) are shown in (18) and (19).

(18)

a.

b. c.

d. (19)

a. b.

c.

d.

I mammut sono esistiti/??hanno esistito molti milioni di anni fa. the mammoths are existed/have existed many millions of years ago Ii libro e piaciuto/?*ha piaciuto soprattutto ai bambini. the book is pleased/has pleased especially to the children Lo zucchero non e bastato/??ha bastato per fare i biscotti. not is lasted/has lasted to make the cooldes the sugar Ii film e sembrato/??ha sembrato noioso a tutti gli spettatori. the film is seemed/has seemed boring to all the spectators La villa ha appartenuto/e appartenuta alla mia famiglia. the villa has belonged/is belonged to my family Le medicine sono scarseggiate/hanno scarseggiato tra i the medicines are/have run in short supply among the terremotati. earthqual(e victims Ii partito e/?ha sussistito senza Ie sovvenzioni dei politici. the party is/has subsisted without the contribution of the politicians Ii suo testamento non e/?ha servito a nulla. his will not is/has served to nothing

The use of HAVE induces an agentive reading, whereas BE does not. So in (20), the verb mancare is understood as intentional in (b) and non-intentional in (a). (20)

a. b.

ealunno e mancato all'appello. the pupil is missed at the roll call Ii presidente ha mancato all' appuntamento. the president has missed at the appointment

(Non-agentive) (Agentive)

Peripheral verbs closer to the 'unergative' core include verbs denoting motional processes (e.g. swim). Native intuitions are less determinate: HAVE is preferred but BE is not completely rejected. (21)

a.

b.

Gli atleti cinesi non hanno corso/?*sono corsi alle run at the the athletes,-Chinese not have run/are Olimpiadi. Olympic Games I bambini hanno saltato/?*sono saltati suI letto tutto il the children have jumped/are jumped on the bed all the pomeriggio. afternoon

Gradience at the Lexicon-Syntax Interface (22)

a. b. c.

d.

Giovanna ha corso/?e corsa pili velocemente di tutti. Giovanna has run/is run faster than everyone else Piera e corsa/?*ha corso al supermercato. Piera is run/has run to the supermarket Paola ha nuotatol?*e nuotata a stile libero. Paola has swumlis swum freestyle Paola ha nuotatol?*e nuotata fino all' altra sponda. Paola has swum/is swum to the shore

In German, these verbs are preferred with BE (see Keller and Sorace 2003 for an interpretation in terms of the feature 'locomotion' being a stronger determinant of unaccusativity in German than in other languages). The effects of agentivity on auxiliary selection are shown in (23), where HAVE is the preferred auxiliary with a human subject, BE is the preferred one with an inanimate subject. (23)

Il pilota ha/?e atterrato sulla pista di emergenza. the pilot has/is landed on the runway of emergency tetto del grattacielo. :Lelicottero e/?ha atterrato suI the plane is/has landed on the runway of emergency

Next, the hierarchy includes various types of uncontrolled processes (such as bodily functions; sweat, for instance), involuntary reaction (tremble), and emission (rattle). These verbs are internally caused but tend to be non-agentive. (24)

a.

b.

c.

d. e.

f (25)

a.

Il convincimento politico ha tentennatol?e tentennato anche nei the political belief has wavered/is wavered even in the pili anziani. oldest people Paolo ha tentennato/*e tentennato a lungo prima di prendere una Paolo has wavered/is wavered for long before maldng a decisione. decision *Paolo ha tentennato apposta prima di decidersi. Paolo has wavered on purpose before of decide-self La terra ha tremato/?e tremata. the earth has trembled/is trembled Mario ha tremato/*?e tremato dallo spavento. Mario has trembled/is trembled of the fear Il mendicante ha rabbrividito/e rabbrividito dal freddo. the beggar has shivered/is shivered from the cold Linnesto non e attecchito/ha attecchito. the transplant not is caught/has caught

Antonella Sorace

262

b. *I.:innesto c.

d. (26)

a. b. c.

d. e.

j

ha attecchito apposta. the transplant has taken root on purpose I.:acqua ha/?e scarseggiata. the water has/got scarce La bicicletta hal?e sbandata senza preavviso. the bicycle has/is skidded suddenly

La sveglia hale squillata. the alarm clock has/is rung I.:eco hale risuonato. the echo has/is resounded II tuono hale rimbombato. the thunder has/is rumbled I.:orologio ha ticchettato/?e ticchettato. the clock has ticked/is ticked La campana hal?e rintoccata. the bell has/is tolled La stella ha brillato/??e brillata. the star has shone/is shone

9.5.3. Typological predictions The hierarchy makes it possible to advance some specific typological predictions. Note that it does not predict that all languages differentiate among all verb classes, but only that there should not be complete reversals of the hierarchical order of verb types (for example, languages in which stative verbs select BE most categorically, or verbs denoting involuntary processes select HAVE more consistently than non-motional activity verbs). The data on auxiliary selection suggest that within any given language there is a cut-off point between verbs that select auxiliary BE and verbs that select auxiliary HAVE. The cut-off point cannot be identical in all languages, since if it were, all languages with a choice of auxiliary would have exactly the same system of auxiliary selection. Thus, the locus of variation must be in the mapping governing the interface between the lexicon and the syntax. Mapping must be language-specific because the location of the cut-off point along the hierarchy may be different. However, variation in the location of the cut-off point is found among the verbs in the middle of the hierarchy, but does not affect the core.

9.5+ A split-intransitivity hierarchy? Since the data reviewed so far pertain to auxiliary selection, the question of relevance to a theory of split intransitivity is whether the hierarchy is a peculiar property of this construction, or whether it underlies not only auxiliary selection but split intransitivity in general. To substantiate the latter claim it is necessary to demonstrate that:

Gradience at the Lexicon-Syntax Interface • other syntactic reflexes of split intransitivity in Romance (and Germanic) languages are affected by the hierarchy; • syntactic reflexes of split intransitivity in languages without auxiliary selection are affected by the hierarchy; • core verbs satisfY a greater number of syntactic diagnostics of split intransitivity than non-core verbs. While more investigation is needed, some research findings are already available with respect to each of these points. Lack of space permits only a brief summary. (a) Other diagnostics in languages with auxiliary selection. There is some evidence that other syntactic manifestations of split intransitivity might also be sensitive to this hierarchy. Sorace (I995a, 1995b) shows that ne-cliticization in Italian displays similar systematic variation as auxiliary selection, that is, there is a scale of acceptability for this construction depending on the position of a verb along the ASH. This is shown in order of increasing acceptability in (27). (27)

a. *Ne b. c.

d. e.

f g.

h.

hanno lavorato molti, di studenti, in questo ristorante. restaurant of-them have worked many, of students, in this ?*Ne hanno camminato tanti, di turisti, su questa strada. of-them have walked many, of tourists, on this street ??Ne hanno squillato/?sono squillati tanti, di telefoni. rung many, of telephones of-them have rung/are ??Non ne ha risuonatol?e risuonata nessuna, di voce. not of-it has resounded/is resounded any, of voice ?Ne sono bastati due, di documenti. of-them are sufficed two, of documents Ne sono sopravvissuti pochi, di soldati. few, of soldiers of-them are survived Ne sono passati tanti, di anni. of-them are gone by many, of years Ne sono venuti molti, di turisti. of-them are come many, of tourists

(b) Diagnostics in languages without auxiliaries. Sorace and Shomura (2001) indicate that syntactic diagnostics such as Quantifier Floating with intransitive verbs in Japanese may be sensitive to the semantic distinctions represented on the hierarchy. In Japanese, an NP and its numeral quantifier must be adjacent in order to enter a relation of reciprocal C-command. So (28a) and (29a) are both grammatical. If the quantifier is separated from the NP that it modifies, only (29b) with an unaccusative verb is grammatical, but not (28b) with an unergativeverb; the reason is that the quantifier c-commands the trace left behind by the apparent subject of the unaccusative verb.

(28)

a.

Kodomo-ga hutari [yp inu-to yukkuri aruita]. child-NOM two dog-with slowly walked 'Two children walked slowly with a dog.'

Antonelfa Sorace

b. *Kodomo-ga [yp inu-to

yukkuri hutari aruita]. child-NOM dog-with slowly two walked 'Two children walked slowly with a dog.'

(29)

a.

b.

Kodomo-ga hutari [yp inu-to (galdm-ni) child-NOM two dog-with school-at 'Two children arrived at school with a dog.' Kodomo-ga [yp inu-to hutari (galdm-ni) child-NoM dog-with two school-at 'Two children arrived at school with a dog.'

tuita]. arrived tuita]. arrived

Native Japanese speakers do not distinguish between verbs denoting non-motional processes and verbs denoting a motional process (unlike Italians), but seem to judge both categories as core, to the extent that they have clear and determinate judgements about the ungrammaticality of these verbs with QF. In contrast, they have significantly less determinate intuitions about the ungrammaticality of QF with other unergative verb types, and express the least determinate judgements on verbs of emission. (c) Consistency across a range ofdiagnostics for the same verb. Evidence from French (a language in which auxiliary selection is no longer a strong diagnostic of unaccusativity), shows that other, arguably stronger, diagnostics such as participial constructions, are satisfied more consistendy by core verbs and less so by non-core verbs (see Legendre and Sorace, in press, for discussion). From a study by Labelle (I992) it also emerges that the verbs selecting etre also behave like unaccusatives with respect to the other tests, and the verbs selecting avoir also fail the other unaccusativity tests (see also Zubizarreta I985). While it is true that there are some verbs that usually select avoir and pass one or more tests of unaccusativity, such verbs are inconsistent in their behaviour, or are less felicitous in these tests than the etre-selecting verbs: an example mentioned by Labelle is disparaitre, which can appear in the impersonal construction (II a disparu des douzaines de livres) but sounds less natural than II est arrive trois hommes. In Labelle's words 'the verbs constructed with etre form the core cases of unaccusative verbs in French' (p. 380).

9.6.

TOWARDS A MODEL OF GRADIENCE

The generalization that is beginning to emerge from these studies is that as soon as one moves from the core one finds substantial but predictable indeterminacy in the syntax-semantics mapping with intransitive verbs. This indeterminacy is difficult to accommodate within a projectionist model of the lexicon-syntax interface, since it would require multiple lexical semantic classifications for a great number of verbs (see van Hout 1996 and Rappaport Hovav and Levin 1998 for discussion). It is also problematic for a constructional model, since core verbs display categorical behaviour and the other verbs are variable, but to different degrees.

Gradience at the Lexicon-Syntax Interface

With respect to auxiliary selection, telicity is the primary factor that separates BEverbs from HAVE-verbs. Agentivity is a secondary factor that differentiates among HAVE-verbs. Core verbs (those at the extremes of the hierarchy) are inherently specified for telicity and agentivity, respectively, and their syntactic behaviour is insensitive to non-lexical properties contributed by the predicate. Intermediate verbs, which are neither telic not agentive, are the most variable and least determinate in many languages; unlike core verbs, they vary in their syntactic behaviour depending on the properties of the predicate in which they appear. Ideally, a theory that accounts for these facts is a constructional model which identifies the factors underlying the differential flexibility exhibited by verbs and incorporates a set of compatibility constraints capable of ruling out inappropriate lexicon-syntax mappings. These constraints would be a crucial component to be acquired in language development. Such a model has not yet been proposed, although potential elements of it already exist. For example, the greater 'elasticity' displayed by stative verbs has been addressed in some detail by McClure's (1995) analysis, which is essentially couched within the constructional approach. McClure addresses the ambiguity exhibited by stative verbs in the context of a general theory of aspect. His theory combines Parson's situation semantics with a modified version of the Vendler-Dowty classification, which includes only three logical event types: states, achievements ('changes'), and activities ('processes')'! McClure's analysis revolves around the structure of predicates internal to events. The notion of state is the basic aspectual component in the system, in agreement with Dowty (1991) and Van Valin (1989). Unlike the latter, however, McClure argues that there exists an aspectual hierarchy such that a state component is part of the logical semantic structure of both achievements and activities: achievements are pairs of states, whereas activities are open-ended collections of spatio-temporally connected achievements controlled by the same individual (the Locus of Change). The aspectual hierarchy is represented in (30). (30)

a.

b.

Aspectual structures States = s, a situation Achievement =(s, s') Activities = {(s, s'), (s', s"), (s", Sill), (Sill . ..} Aspectual types BECOME: sets of states -? sets of pairs of states DO: sets of be comings -? sets of sets of becomings having the same protagonist (McClure 1993: 316)

The aspectual hierarchy establishes a basic distinction between verbs denoting activities and achievements, which specifY a Condition of Change in their lexical entry, and stative verbs, which do not. The pairs of changes characterizing achievements are directed and temporally ordered, whereas the changes characterizing activities are 7 Accomplishments are not considered as a separate class because they are regarded as a particular type of activity leading to a conclusion. See Pustejovsky (1995) for similar arguments.

Antone/La So race linked head to tail, in the sense that the final state of one change is the initial state of the next change, and all states have equal importance. Further, the changes are nondirected: every state can be either an onset or a result. These logical aspectual structures are essentially constant across languages, although event conceptualization, which leads to the classification of individual predicates in one aspectual class or another, may vary from language to language. Variation (both within and across languages) arises from aspectual ambiguities, which are a common feature of many verbs, and is particularly frequent with statives. Statives, in fact, 'begin with almost no aspectual content' (McClure 1995: 131) and can therefore take on the aspectual properties of either achievements or activities compositionally, as indicated by the English examples in (31) and the Italian examples in (32).

(31)

SPARKLE

a. b. c. (32)

This kind of paint sparkles the most. I've just seen that star sparkle. The stars sparkled all last night. ('continue') 11 saggio e continuato alIa pagina seguente. the essay is continued on the page following 'The essay continued on the following page.' Lo spettacolo e continuato alle tre. the show is continued at three o'clock 'The show continued at three 0' clock.' 11 discorso ha continuato per ore. the speech has continued for hours 'The speech continued for hours.'

(State) (Achievement) (Activity)

CONTINUARE

a.

b.

c.

(State)

(Achievement)

(Activity)

Many of the aspectual ambiguities exhibited by stative verbs, such as those exemplified in (33)-(35), may also be accounted tor on the basis of their aspectual underspecification.

(33)

Paola e vissuta/ha vissuto per tre anni a Parigi. Paola is lived/has lived for three years in Paris

(State/process)

(34)

r; aiuola

e fiorita/ha fiorito. the plant is blossomed/has blossomed

(State/process)

(35)

La moda grunge e attecchita/ha attecchito in Italia. the grunge fashion is caught on/has caught on in Italy

(State/process)

Thus, states do not have a fixed mapping, as in Levin and Rappaport Hovav's model: rather, their syntactic status varies across languages according to the conceptualization they are given in a particular language, and within languages according to the interpretation they receive in a particular context. s 8

A distinction between lexical and compositional unaccusativity is also central to Pustejovsky's Gen-

Gradience at the Lexicon-Syntax Intel/ace A step in the same direction is Rappaport Hovav and Levin's (1998) sketch of a generative theory of verb meaning, aimed at establishing the constraints underlying the flexibility exhibited by some verbs. Assuming a conventional theory of predicate decomposition, they provide a (potentially universal) inventory of lexical semantic templates corresponding to the basic event types:

(36)

a. b. c. d. e.

[x ACT (MANNER)] [x STATE] [BECOME [x (STATE)]] [[x ACT] CAUSE [BECOME [y(STATE)]]] [x CAUSE [BECOME [y (STATE)]]]

Constants may be paired with one of these event-structure templates via canonical realization rules, which must satisfy certain compatibility constraints. For example, the constant 'lengthen' is mapped onto the template (36d) to produce [[x ACT] CAUSE [BECOME [y (LONG(ER)]]] Complex event structure may be built from simpler ones in an incremental, monotonic fashion. The mechanism responsible for these derived verb meaning is called Template Augmentation (Rappaport Hovav and Levin 1998): 'Event structure templates can be freely augmented up to other possible templates in the basic inventory of event structure templates.' The operation of Template Augmentation is constrained by well-formedness conditions on syntactic realization, such as the Subevent Identification Condition, which requires that each subevent must be identified by a lexical head, and the Argument Realization Condition, which involves the presence of an argument in the syntax for each event structure participant. Accomplishments may be derived from activity through Template Augmentation (Mary swept the floor clean from Mary swept the floor). The systematic ambiguity of some change-of-state verbs such as blossom, wilt, and decay between a stative and an eventive interpretation may also be regarded as an effect of Template Augmentation, which derives an achievement template from a basic state template. In contrast, achievement verbs have a rigid event template structure which cannot be augmented further (for example by adding another state) or turned into an activity template (which would create a non-existent structure [BECOME [x ACT(MANNER)]], which does not belong to the universal inventory of event structure templates). What this suggests is that stative verbs, by virtue of their less specified event structure, are more erative Lexicon Theory. Within this theory, a set of generative devices operate on semantic representations of verbs, determining their 'event headedness', which in turn causes the foregrounding of particular event arguments: unaccusatives are right-headed predicates, for which the result state is the focus of interpretation; unergatives are left-headed predicates, for which the agentive cause of the event is in focus. While these devices can operate on unspecified lexical representations (such as sink) or inherently unergative predicates (run) to obtain derived unaccusative predicates compositionally, other unaccusatives (arrive, for instance) are specified as right-headed in the lexicon and are not subject to event-type shifts via semantic rules (see Pustejovsky and Busa I995). I argue that it is precisely these verbs that are at the core of unaccusativity.

Antonelfa Sorace

268

susceptible to shifts in interpretation, consistent with McClure's account and with the cross-linguistic facts. The unresolved question is whether Template Augmentation and the other conditions on argument realization are lexical or interpretive operations. Ultimately, both projectionist and constructional theories recognize the need for a theory of the relationship between the meaning of verbs and the structures in which they can appear. Within a projectionist model, event structure templates are part of the lexical entry of verbs but, as Rappaport Hovav and Levin (1998) suggest, it may be incorporated in a constructional model as a post-lexical, checking mechanism.

9.7.

CONCLUSION

The Auxiliary Selection Hierarchy provides a generalization that captures the systematic variation exhibited by intransitive verbs in their choice of auxiliary across a number of languages. By doing so, it offers a stronger empirical basis to a range of observations and data that had been presented in the literature on split intransitivity. The ASH also accounts for the developmental paths followed by second-language learners ofItalian and French, who start acquiring auxiliary selection from core verbs and are more likely to retain non-native intuitions with respect to non-core verbs at advanced stages of development. Further research is needed to corroborate the still limited evidence that the ASH may underlie not only auxiliary selection but also other syntactic reflexes of split intransitivity. The ASH suggests that both a syntactic and a lexical characterization of split intransitivity is necessary to account for the complexity of the phenomenon: a syntactic characterization is needed to account in a general way for the distributional properties of unaccusative and unergative verbs; a lexical account is needed to explain the constraints that govern the lexicon-syntax interface and the variation that results from mapping one level onto the other. Neither existing projectionist nor constructional theories are able to do full justice to the pattern of variation represented by the ASH, because they do not incorporate a fully workedout mechanism for checking possible and impossible pairings oflexical meanings and structural configurations. The evidence available so far indicates that a variant of the constructional approach that includes a specification of such mechanism may well become a solution to the problem: future research will tell whether this is the right track to follow.

10 Unaccusativity in Saramaccan: The Syntax of Resultatives Tonjes Veenstra

10.1.

INTRODUCTION

Unaccusative diagnostics come in two types (Levin and Rappaport Hovav 1995). Ne-cliticization, there-insertion, and locative inversion count as diagnostics of surface unaccusativity; auxiliary selection and the resultative construction are diagnostics of deep unaccusativity. Here I only deal with one of the diagnostics of the latter type: the resultative construction. This construction is said to be indicative of unaccusativity because only the object can be predicated by the resultative XP (Simpson's 1983 directobject generalization). In Saramaccan' (as well as in many other serializing languages), resultatives are expressed by means of serial verb constructions. An example to illustrate the construction is given in (1). I would like to thank the audience at the I998 Unaccusativity Workshop at the ZAS in ~erlin for comments. The usual disclaimers apply. Special thanks to Feledi Pold and other Saramaccans for sharing their knowledge with me. The data reported on was collected in Surinam during a field trip made possible by the Netherlands Foundation for the Advancement of Tropical Research (WR 39-2°7) which I hereby gratefullyaclmowledge. The chapter was written during a stay at the University of Potsdam during an NWOfellowship (TALENT Stipend: S 3°-485). I Saramaccan is a Creole language spoken by some 20,000 Maroons. It is the heritage of people who were courageous enough to escape the barbarism of the plantations to live freely in the South American

Tonjes Veenstra

270 (I)

De jaka hen puu. 3PL chase 3SG pull 'They chased him away.'

They typically have the format shown in (2), in which NPI is interpreted as the 'subject' of both verbs, NP2 as the 'object' of both verbs: NPI V, NP2 V2

(2)

One of the burning issues in the study of serial verb constructions is how to account for the argument-sharing effect. Several interrelated questions arise. First, is argument sharing taken care of in the lexicon, as it would be in a lexical analysis, or is the argument-sharing effect mediated through empty categories, as it would in a syntactic analysis? And if the latter is the case, what is the nature of this empty category (see Veenstra 1996a; Collins 1997; den Dild{en and Sybesma 1998 for relevant discussion). Second, is NP 2 part of the first VP or of the second VP? Third, is the external argument ofV2 covertly present? And, fourth, which type of verb (active transitive, passivized transitive, unaccusative, unergative) can occur in the VI as well as the V 2 position? The main objective here is to present an analysis of resultative constructions. This analysis involves a Larsonian VP-shell configuration, in which NP 2 is part of the first VP and the second VP is an adjunct to the first VP (c£ Larson 1991; Veenstra 1996a). The argument-sharing effect is obtained through operator movement inside the second predicate. At the end of the chapter I will discuss the implications of this analysis for the analysis of resultatives as found in Germanic languages. Most of the discussion will be concentrated on the internal make-up of the second VP, thereby adressing the following issues:

(3)

a. b. c.

d.

Is the second VP an unaccusative VP? Is it a passivized VP? Is it a full-blown VP? Is NP2 part of the second VP?

With respect to the first three issues, I will argue for option (3c). Evidence against option (3a) comes from the discussion of unaccusative verbs in Saramaccan. Evidence against option (3b) comes from a detailed discussion of passivization in Saramaccan. As to (3d), I will present syntactic evidence that NP 2 is not part of the second VP, but part of the first one. The evidence stems from the comparison of serial verb constructions with perception verb constructions. The discussion centres on ECM-configurations, showing that serial verb constructions do not reduce to them. Additional evidence comes from the behaviour of ideophones. The discussion of unaccusativity allows me to make two additional points. First, it shows that there must be an aspectual projection inside the second VP in resultative rain forest. It is a tone language. There are two surface tones-high and low-and three underlying tones-high, low, and unspecified. Here, high tones are marked with an acute accent (0, low tones with a grave accent (l). Only lexically specified tones are marked-the tones that show up due to sandhi phenomena are not (except in sect. 10.2). The data are from the Basulio dialect.

Unaccusativity in Saramaccan constructions. Second, it shows that it is not the case that only transitive verbs can participate in resultative serial verb constructions (contra Veenstra 1996a). The chapter is organized as follows. Section 10.2 introduces serial verb constructions. In section IO.3 I discuss un accusative verbs and their relevance for the structure of serial verb constructions. Section IO.4 examines passives. Section 10.5 concerns the status of the 'shared' object. Finally, in section 10.6 the analysis is presented.

10.2.

SERIAL VERB CONSTRUCTIONS

Serial verb constructions are considered to be constructions containing at least two (main or independent) verbs in what appears to be a single clause. Only one subject and no overt markers of subordination or coordination are present Gansen, Koopman, and Muysken 1978; Sebba 1987; Muysken and Veenstra 1995). They have a single (accomplished) event interpretation and all the verbs have the same specification for tense, mood, and polarity. There are two diagnostic tests for the verbal status of the elements in serial verb constructions: aspect marking and predicate clefts (cf. Veenstra and den Besten 1995). Applying these tests to the resultative construction gives the following results. Aspect is normally only marked on the first verb; see (4a). It then expresses nonpunctual, that is, durative, habitual or iterative, aspect, and has scope over all subevents. If aspect is marked only on the second verb, as in (4b), it expresses durative aspect and has scope over only the second sub-event. If, on the other hand, aspect is marked on both verbs, it can express durative and iterative aspect, as in (4c). The iterative reading is preferred, however. (4)

a.

A

ta faa pau tue.

chop tree throw 'He is felling a tree/the tree/trees.' A faa pau ta tue. 3SG chop tree ASP throw 'He is felling the tree (Le. at this very moment the tree is falling).' A ta faa pau t3. tue. 3SG ASP chop tree ASP throw 'He is (constantly) felling trees.' 3SG ASP

b.

c.

Predicate-cleft constructions are constructions in which a copy of a verb appears in sentence-initial position. The main function is to focus on the verbal action (event quantification, cf. Larson and Lefebvre 1991). All the verbs in the construction can undergo predicate cleft:

(5)

a.

Suti a suti hen klL shoot 3SG shoot 3SG kill 'He shot her dead.'

Tonjes Veenstra

272

b.

Kii a

sun hen Wi. kill 3SG shoot 3SG kill 'He shot her dead.'

This shows that all the elements in the construction under discussion are real (eventive) verbs. Verbs are subject to a tonal phenomenon in Saramaccan, in particular a sandhi rule that operates depending on certain syntactic conditions, which gives an important insight into the structure of the resultative construction and shows the way towards its proper analysis. As noted earlier, Saramaccan is a tone language. It has underlyingly three types of tone: changeable low (or unspecified), unchangeable low, and unchangeable high. There are four classes of words: words with only high tones, words with only unchangeable low tones, words with high and unchangeable low tones, and words with high and changeable low tones. The sandhi rule relevant to the latter class of words is defined as follows: All changeable low tones between the highs in successive morphs in certain syntactic positions are changed to high, unless there are also unchangeable lows between the highs. (Rountree 1972 : 314)

Rountree discusses the whole range of sandhi phenomena in Saramaccan and shows that adjacency is required in this process of assimilation. Not all adjacent elements undergo this sandhi rule, however. For instance, verbs and nominal objects (as opposed to pronominal objects) do not undergo sandhi. Interestingly enough, the sandhi rule does apply when two adjacent verbs are involved. This being the case, it is to be expected that sandhi does not take place in the resultative construction, since an object intervenes between the two verbs. This object would destroy the adjacency between the two verbs and, thus, block the application of the sandhi rule because it only operates on adjacent elements. 1his is not the case, however. The two verbs do undergo sandhi (even when an object with unchangeable low tones intervenes, as in (6b)), as if they were adjacent: (6)

a.

b.

A niki di dagu lai. A naki di digu lai. 3SG hit DET dog kill 'He struck the dog dead.' A niH di logoso kli. A nata di logoso kli. 3SG hit DET turtle kill 'He struck the turtle dead.'

The object that stands between the two verbs in (6a) is not affected by the tones of either the first verb or the second verb, as predicted (that is, dagu does not become dagu), nor does it block the application of the sandhi rule in (6b), but the changeable low tone of the first verb does change into a high one (ndki becomes ndkt). Thus, two

Unaccusativity in Saramaccan

273

apparently non-adjacent verbs can form a context for the sandhi rule. Since in all other instances adjacency is required, the sandhi rule should be maintained. This implies that the two verbs in serial verb constructions are adjacent at a level prior to Spell-out and that subsequent movement operations create the observed surface order. Two possible underlying orders have been proposed to accomodate this adjacency effect: (7)

a. b.

NPI v NP2Yr V, NPI Yr V2 NP2

In (7a) the observed surface order is derived by movement of Yr to the v-position. In (7b) it is NP2 that moves to a position between the two verbs to create the surface order. Both options are schematized in (8). (8)

a.

b.

VI-to-v Movement (c£ Veenstra 1996a) [vp NPI [Yr b NP2 [t [VP2 .. ,V2 ... m]] l A-movement (c£ den DiH:en and Sybesma 1998)2 [vp NPI [VP I Yr [AspP NP2 b 2 (... ) V2 t]]]]

Crucial differences between these two analyses are: (i) in (8a), NP2 is part ofVP" in (8b) it is part ofVP2; (ii) in both analyses NP2 is Case-marked by Yr, but while in (8a) it is regular accusative Case-assignment, (8b) involves an Exceptional Case Marking-configuration; (iii) in (8a) VP2 is an active transitive VB in (8b) it is not, but rather a passivized VP (adjectival or verbal) or an unaccusative VP. I will argue that the analysis in (8a) is superior to the one in (8b). In the next section I show that VP 2 cannot be an unaccusative VP. Section 10.4 discusses passives in Saramaccan, showing that VP2 is not a passivized VP either. In section 10.5 it will be shown that NP2 is part ofVP" and not ofVP2.

10.3.

UNACCUSATIVE VERB S

It is not easy to classifY verbs as unaccusative in a language such as Saramaccan, in which all the 'classic' tests of unaccusativity-auxiliary selection, ne-cliticization, there-insertion, and locative inversion-do not apply. Additionally, semantic criteria should be applied with the greatest care, because it is well known that verbs with similar meanings can be classified differently in and across languages. For example, in 2 It is only fair to say that den Dikken and Sybesmas analysis was restricted to serial verb constructions with 'take' as V,. Here I examine whether their analysis can be extended to account for the resultative construction.

Tonjes Veenstra

274

Italian, the equivalent of die is unaccusative, but in Choctaw it is unergative (Levin and Rappaport Hovav 1995). In spite of these considerations, I will take kai ('fall') to be a prototypical unaccusative verb in Saramaccan (as will become clear in what follows, this assumption has advantages when it comes to making sense of the patterns in the data with respect to harmony effects). In addition, I will assume that the intransitive variant of the Causative-Inchoative (Ergative) Alternation is also syntactically characterized as unaccusative (cf. Alexiadou and Agnostopoulou, in this volume). Saramaccan has two classes of alternating verbs (Veenstra 1998a): ClassA

booko ('break'), sinki ('sink'), jo ('melt'), jabi ('open'), puu ('pull'), etc.

(9)

A

a.

jabi di doD. open DET door 'He opened the door.' Di doD jab!. DET door open 'The door opened.' 3SG

b.

Class B etc.

(10)

a.

siki ('(be) sick'), hanse ('(be) beautiful'), limbo ('(be) clean'), be ('(be) red'), De siki hen. 3PL sick 3SG

b.

'They sickened him.' A siki. 3SG sick 'He is sick.'

Turning now to resultative serial verb constructions, it is relatively easy to determine that the second VP is not an unaccusative VP. First, the V 2 position is not confined to unaccusative verbs only: verbs that are definitely not unaccusative, such as kii ('kill') and tue ('throw'), can head VP2 (see examples below). Second, there is a transitivity restriction on the first verb, as in (II), as well as on the second verb, as in (12):

(II)

a.

De sikopu hen kii. kick 3SG kill 'They kicked him dead.' b. *De kai kii. 3PL fall kill c. *De kuie (de-seei) kii. 3PL run 3PL-REFL kill

(Transitive)

3PL

(12)

a.

pau tue. tree throw 'They felled the tree.' De faa

di

3PL chop DET

(Intransitive (unaccusative)) (Intransitive (unergative)) (Transitive)

Unaccusativity in Saramaccan

b. *De faa 3PL

chop

di DET

pau kat. tree fall

275

(Intransitive (unaccusative))

This shows that both verbs in the resultative construction have to be transitive (as argued in Veenstra 1996a, but see below). Therefore, the VP2 in the resultative construction cannot be unaccusative. The behaviour of the two classes of alternating verbs in the resultative construction shows two things: (i) there is an aspectual projection in VP 2; Oi) it is not the case that only transitive verbs can participate in the construction (contra Veenstra 1996a). Class-B verbs, such as be ('(be) red'), cannot appear on their own as the second verb:

(13)

*Mi fefi di w6su be. ISG paint DET house red

These verbs must be accompanied by an aspectual verb k6 ('come'), which signals the completion of the event: 3

(14)

Mi fefi di w6su k6 be. ISG paint DET house come red 'I painted the house red.'

Together with the data in (4) on aspect marking, this shows that there is an aspectual projection inside the second VP in the construction. Class-A verbs are possible both in VI and in V 2 position:

(15)

a.

A

b06ko hen puu. break 3SG pull 'He broke it off.' b. *A b06ko hen kat 3SG break 3SG fall 3SG

Both bo6ko ('break') and puu ('pull') are class-A verbs. If the first verb in the construction is the transitive variant of a class-A verb, then the second verb must also be transitive in order to account for the difference in grammaticality between (15a) and (15b). If, on the other hand, the first verb is the intransitive variant, then the second verb is also intransitive. In this case, the unaccusative kat (,fall') is possible as the second verb: (16)

a.

A b06ko puu. 3SG break pull 'It broke off.' or 'It was broken off.'

J This aspectual verb is best analysed as an auxiliary verb (Le. part of the (functional) TMA-complex above the VP), since it cannot undergo predicate cleft: (i) a. A k6 be. 3SG come red 'It has become red.' b. *K6 a k6 be. come 3SG come red

Tonjes Veenstra

276

b.

Di w6su bo6ko kai. house break fall 'The house is fallen apart.' DET

The generalization with respect to possible verb combinations in the resultative construction can be stated as follows: (q)

a.

NPI V TRANS NP2 V TRANS

(=

lIa)

b. *NPI V TRANS NP2 VUNACC-(= I5b) c. *NPI VUNACC NP2 V TRANS (= lIb) NPI VUNACC NP2 V UNACC (= I6b)

d.

I do not have an explanation for this harmony effect or co occurrence restriction in resultative constructions. In the remainder of the chapter I will be concerned only with the pattern in (I7a), leaving the pattern in (qd) for further research. The implications of this section for the analysis are (i) VP 2 is not (in all cases) unaccusative; (ii) all the arguments of the second verb are syntactically represented; (iii) there is an aspectual projection present in the second predicate. The minimal structure ofVP2 , therefore, is:

(18)

[vp SUB

[AspP

ASP [yp V OBJ]]]

10.4. PASSIVE CONSTRUCTIONS I should begin with a word of caution. It is not the case that all speakers of Saramaccan accept all examples of the passive construction. Much depends on the context provided with the examples. For instance, examples such as (19), when uttered in isolation, were often judged as unacceptable or strange (as indicated by the bracketed asterisk):

(19)

(*)Di gania lui. chicken kill 'The chicken was killed.' DET

The following contexts significantly enhance the acceptability of such examples: (i) the addition of an (aspectual) adverb (e.g. kdd 'already'), as in (2oa); (ii) embedding in a subordinated temporal clause, as in (20b): (20)

a.

b.

Di gania kii kaa. DET chicken kill already 'The chicken had already been killed.' [Di di gania kii], di onko6ku k6 ldi. then DET chicken kill DET misfortune come fall 'At the same time the chicken was killed, misfortune started to happen.'

Unaccusativity in Saramaccan These contexts supply aspectual information on the event expressed by the 'passivized' verb, thereby putting focus on the endpoint of the event. Other restrictions (independent of variation between speakers) are also semantic in nature. These can be formulated in terms of 'transitivity' and 'agentivity' of the verbs involved, 'affectedness', and 'animacy' (see Veenstra (1998b) for details). The construction, therefore, may be best characterized as a 'basic' passive, in the sense of Keenan (1985). A by-phrase is not possible:

Di womi bi naki (*ku/u di mUJee f'en). DET man TNS hit with/for DET woman for.3SG 'The man was hit (by his wife).'

(21)

Due to the lack of passive morphology mentioned earlier, there is no formal difference between passives and unaccusatives (or anticausatives); see (22).

Di bangi booko. DET stool brealc 'The stool broke.' or 'The stool was broken.'

(22)

One way to distinguish the two constructions is to look at implicit argument effects (Roeper 1987; Verrips 1996). I discuss three tests-the addition of an agent-oriented adverbial, an infinitival purpose clause, and an instrumental prepositional phrase-to show that 'basic' passives in Saramaccan have an implicit argument. But first, I turn to a very basic question-namely, are 'basic' passives in Saramaccan adjectival or verbal? Consider the following examples: (23)

a. b.

The letter was stolen (by John). The pillow remained stuffed (*by John).

(Verbal) (Adjectival)

Verbal passives differ from adjectival ones in (at least) three respects (see, for instance, Levin and Rappaport Hovav 1986; Verrips 1996); (i) adjectival passives generally cannot have a by-phrase, verbal ones can; (ii) verbal passives have an implicit argument (that is, the external argument is syntactically represented in the passive), as opposed to adjectival ones (the external argument is not represented in the syntax); (iii) adjectival passives generally refer to a state, verbal ones refer to a dynamic event. Note that the first two differences are interrelated. Following proposals by Wasow (1977) and Borer (1984), the basic proposal to account for these different thematic properties has been (and still is) that adjectival passives are formed in the lexicon, and verbal passives are formed in the syntax. Lexical derivations may involve argument deletion, hence there is no implicit argument present in adjectival passives. In syntactic derivations, on the other hand, this is ruled out (due to the Projection Principle, or its successor), so an implicit argument is present in verbal passives. Applying these tests to the passive in Saramaccan gives the following result. As shown in (21), repeated in (24) for convenience, a by-phrase is not possible:

278

Tonjes Veenstra

(24)

Di womi bi nalu (*ku/u di mujee f'en). DET man TNS hit with/for DET woman fOf.3sG 'The man was hit (by his wife).'

On the face of it, this constitutes evidence for an adjectival status of the passive. The issue is not as crystal-clear, however. First, as pointed out in Roeper (1987), the presence vs. absence of a by-phrase is not a 'hard' property distinguishing verbal and adjectival passives in English (see Cornel is and Verhagen 1995 for the same observation for Dutch). Sentence (25) is an example of an adjectival passive talung a by-phrase:

(25)

The code remained unbroken by the Russians.

Second, it may be the case that the impossibility of a by-phrase in Saramaccan is independent of the categorial status of the passive. For instance, there is simply no preposition available that can introduce external arguments. Potential candidates are ku ('with') and If)u (,for'). The former introduces adjuncts (such as instruments and comitatives), while the latter only marks non-agentive participants. This is what I will assume here. Thus, the unavailability of ry-phrases is not an argument for the adjectival status of passives in Saramaccan. Turning now to the third difference (event vs. state), the passive construction has an eventive interpretation, as the following example illustrates: (26)

Iniwan daka biifi ta fufUu a postkantoor. every day letter ASP steal at post office 'Every day letters get stolen at the post office.'

Verbs in a passive construction also interact differently with the aspect marker ta than stative verbs do. Whereas a stative verb, when preceded by the aspect marker, is interpreted as an inchoative, as in (27a), a passive construction is interpreted as a nonpunctual (including iterative and durative), just like eventive verbs, as shown in (27b) and (27c), respectively: (27)

a.

b.

c.

A ta siid. 3SGASP sick 'He is getting sick.' Di dagu bi ta nald gaanduwe. DET dog TNS ASP hit long. time 'The dog was being hit for a long time.' A bi ta mbei tembe. 3SG TNS ASP make wood 'He was malung wood-carvings.'

Thus, passives in Saramaccan have an eventive interpretation, not a stative one. Another argument for the verbal nature of the passive is the categorial status of the 'passivized' element (verb or adjective). Saramaccan has a class of adjectives. They are reduplicated forms, and cannot occur without a copula:

Unaccusativity in Saramaccan

279

Di boto *(de) sinki-sinki. boat BE sink-sink

(28)

DET

They also cannot undergo predicate cleft: *Sinki-sinki di boto de sinki-sinki. sink-sink DET boat BE sink-sink

(29)

In the passive construction the copula cannot appear (30a). The 'passivized' element can undergo predicate cleft, as in (30b). (30)

a.

Di biifi (*de) fufUu. letter BE steal 'The letter was stolen.' FufUu di biifi fufuu, naa Jasi. steal DET letter steal NEG lose 'Stolen was the letter, not lost.' DET

b.

Thus, the categorial status of the 'passivized' element is verbal, not adjectival. Implicitargument effects point towards the verbal status of the passive construction, too. An implicit argument is the suppressed external argument of the verb which plays a role in the interpretation of verbal passives without by-phrases. It has a bare-plural prointerpretation-that is, unspecific, non-generic (c£ Baker, Johnson, and Roberts 1989; Borer 1996). Roeper (1987) argues that implicit arguments are represented syntactically in English. The empirical core of his argument is that the implicit argument can control the non-overt PRO subject of an infinitival purpose clause. This is illustrated in (31). (31)

The glass was broken [PRO to hurt Mary].

Furthermore, adverbial phrases can predicate over the implicit argument:

(32)

a. b.

The glass was broken on purpose. The glass was broken with a hammer.

The Agent-oriented adverbial in (32a) modifies the implied 'breaker', not the glass. The instrumental prepositional phrase in (32b) is interpreted as the instrument used by the 'breal{er'-there is a PRO in these adverbial phrases, controlled by the implicit argument. The passive construction in Saramaccan shows all the implicit-argument effects. This is shown in (33):

(33)

a.

Di bangi booko. stool break 'The stool was broken.' Di bangi booko ku sdbi. DET stool break with knowledge 'The stool was broken on purpose.' DET

b.

(Agent-oriented adverbial)

Tonjes Vt-enstra

280

c.

Di bangi booko u fee df muFe. (Infinitival purpose clause) stool break for frighten DET woman 'The stool was broken to frighten the woman.' Di bangi booko ku pau. (Instrumental prepositional phrase) DET stool break with stick 'The stool was broken with a stick.' DET

d.

One might argue (cf. Lasnik 1988) that these effects are due to event control, but the next example clearly shows that they must be caused by the presence of an implicit argument: (34)

a.

Di biifi fufuu [u tei moni puu n'en]. letter steal for tal<:e money pull LOC:3SG 'The letter was stolen to take the money out of it.' b. *[Di fufuu u di biifi] tei moni puu nen. DET steal for DET letter take money pull LOC:3SG DET

The conclusion, therefore, is that passives in Saramaccan are verbal, and have the following properties: (i) the external argument of the verb is not expressed; (ii) the internal argument of the verb appears in subject position; (iii) implicit arguments are syntactically represented. 4 The implications of the discussion for the analysis of (resultative) serial verb constructions amount to the following. First, the second VP cannot be an adjectival passive, since passives are shown to be verbal in Saramaccan. Second, the second VP cannot be a verbal passive either. If this were the case, we expect the external argument OfV2 to have a bare-plural pro-interpretation. This is not the case. The external argument ofV2is (uniquely) interpreted as the external argument of V,. The conclusion of this section, therefore, is that VP2 in the resultative construction is not passivized. Note that this conclusion is relevant for both analyses of the resultative construction. It rules out a covert passive analysis ofVP 2 in the V,-to-v movement analysis of (8a), and an overt one in the A-movement analysis of (8b).

10.5.

STATUS OF THE 'SHARED' OBJECT

A crucial difference between the two analyses of the resultative construction concerns the status of the 'shared' object. In the V,-to-v-movement analysis of (8a) it is part of 4 Saramaccan passives are a problem for Baker, Johnson, and Roberts's (1989) theory; in which all properties of the passive construction can (and, in fact, should) be derived from the lexical properties of the 'passive morpheme', since these passives have, on the one hand, all the syntactic characteristics of syntactically formed passives, but lack all the mophologicalhallmarks: (i) no overt passive morphology is present (or possible); (ii) the absence of an overt 'passive' auxiliary verb. Passive morphology, therefore, is not a central feature of the construction, but merely a side-effect of a syntactic process. See Veenstra (1998b) for discussion and an analysis along the lines of Borer (1996).

Unaccusativity in Saramaccan VPI' in the A-movement analysis of (8b) it is part ofVP 2 • In the former analysis it is assigned accusative Case in a regular fashion, but in the latter one it is exceptionally Case-marked. Exceptional Case Marking (ECM) occurs in Saramaccan in perception verb constructions (Veenstra 1994, 1996a), and I will now contrast perception verb constructions with serial verb constructions. Essentially, both constructions have the same surface order; see (35a-b). (35)

a.

b.

NP, V; NP2 V2 (XPJ ) Mi sf df f6u lui kia. ISG see DET bird kill already 'I saw the bird was killed already.' Mi naId df f6u lui kia. ISG hit DET bird ldll already 'I already hit the bird dead.'

(Perception verb construction)

(Serial verb construction)

Both allow aspect marldng on the second verb, but not tense (or mood) marldng. Superficially, they seem to be different with respect to selection properties. Although it is reasonable to claim that in perception verb constructions the first verb selects the clause consisting of both NP 2 and the second VP (as in 'I saw [God perform miracles] '), this is not as clear in the case of serial verb constructions. The first verb (ndki in (35b), for instance) does not straightforwardly select a constituent consisting ofNP 2 and the second VP in the latter case. I first show that perception verb constructions exhibit ECM. The distinction between finite and non-finite complements of perception verbs involves the simultaneity of the events as expressed by the two verbs. When the complement is finite, in which case a tense marker can occur on the embedded verb, the events are non-simultaneous. Furthermore, the finite complementizer tda is (optionally) present. Thus, in (36a) the moment of seeing is not at the same time as the moment of sleeping. In (36b), on the other hand, both events take place at the same time. The tense marker cannot occur on the second verb and the finite complementizer is obligatorily absent. In this case only aspect can be (optionally) marked on the embedded verb: (36)

a.

b.

Mi sf raa a bi duumf. ISG see say 3SG TNS sleep 'I saw that he had slept.' Mi sf (*raa) a ta duumf. ISG see say 3SG ASP sleep 'I saw him sleeping.'

(Full complement)

(Bare complement)

A major difference between the two types of complement involves binding possibilities of pronouns. The presence of Tense (finiteness) is an opacity factor for binding domains (c£ Chomsky 1981). Consider the following examples from English: (37)

a. b.

Hei expects him*i to win the race. Hei expects hei will win the race.

Tonjes Veenstra In (37a) the embedded subject of the non-finite complement him cannot be coreferential with the subject of the matrix verb. Note that the embedded subject is Casemarked by the matrix verb. In (37b), on the other hand, coreference of the embedded subject and the matrix subject is possible. In this case, the embedded subject is Casemarked by a finite INFL inside the complement. In Saramaccan we find the same pattern with respect to the binding possibilities of the embedded subject. In full complements, coreference of the matrix subject and the embedded subject is possible, but is prohibited in bare complements. The relevant data are given in (38).

(38)

a.

b.

Ai jei a*i ti fin. 3SG hear 3SG ASP talk 'He heard him talking.' Ai jei tia ai bl ta maa tongo da sembe d1 a bl abi 3SG hear say 3SG TNS ASP sharpen tongue give people then 3SG TNS have dii jaa. three year 'He heard that he had (already) been ordering people around at the age of three.'

The data in (38) constitute strong evidence for the clai~ that bare complements are non-finite. Additional evidence has been presented in Veenstra (1994). Based on the distribution of time adverbs, the non-availability of tense and negation, and the object-like properties of the embedded subject with respect to negation, quantification, and their interaction, it is shown that in the case of bare complements we are dealing with a non-finite clausal complement selected by the verb of perception. Since nominative Case depends on the feature-specification ofTense, it follows that the embedded inflectional complex cannot assign nominative Case to the embedded subject. Instead it is dependent on a Case-assigner from the outside-here, the perception verb. The matrix verb assigns accusative Case to the embedded subject-a 'straightforward' case ofECM. We conclude, then, that ECM occurs with perception verbs that select a bare complement. But we are left with the question why the form a surfaces instead of the expected (non-emphatic) form hen. After all, it is traditionally assumed (Byrne 1987, for instance) that a signals nominative case, and hen, accusative case:

(39)

Mi S1 a/*hen go. see 3SG go 'I saw him leave.' ISG

Veenstra (199Gb) identifies the 'pronominal' a as being a member of the set of subject clitics in Saramaccan, and argues that subject elitics are in fact Agreement heads. As such, they are independent of Tense and structural Case-marking. They function solely as identifiers of null subjects. The analysis of perception verb constructions, then,

Unaccusativity in Saramaccan runs as follows. The embedded null subject is Case-marked by the verb of perception. Since null subjects have to satisfy two licensing conditions (Rizzi 1986a), ECM from the matrix verb alone does not suffice. They also need to be identified. The identification is handled by the AGRs of the embedded complement. Thus, we have the following configuration for Case-assignment to embedded null subjects: (40)

[Y' VA

[AGRsP

pro AGRso

[AspP' •• J]]

In (40), the null subject pro is formally licensed by VA and identified by AGRso. Resultative serial verb constructions are different from perception verb constructions in this respect. A subject clitic cannot occur in the NP 2 position of a serial verb construction (see (41)), as it can in perception verb constructions in (39).

(41)

A

naki *a/hen lUi. 3SG kill 'She struck him dead.'

3SG hit

This shows that serial verb constructions do not pattern with perception verb constructions with respect to Case-marking. From the fact that perception verb constructions involve an ECM configuration, I conclude that such a configuration is not present in serial verb constructions. Instead, what takes place in serial verb constructions is regular accusative-Case assignment by the first verb. Later I will return to another difference between serial verb constructions and perception verb constructions. An independent argument against NP2 being part of the second VP concerns the distribution of ideophones. Ideophones are adverb-like elements, used to modulate more closely the meanings of verbs, and are selected by particular verbs. They demarcate the right-hand edge of the VP (cf. Rountree 1992): (42)

a.

A weti faan/*njaa. 3SG white IDEO PHONE 'It is snow-white.' b. A naki hen gb66-gb66. 3SG hit 3SG IDEO PHONE 'She hit him really hard.' c. *A lui hen gb66-gb66. 3SG kill 3SG IDEO PHONE

In serial verb constructions, the ideophone is placed between NP 2 and V., not between

VI and NP 2 , although it is selected by VI and not V2 • This is shown in (43). (43)

a.

A naki hen gb66-gb66 kn. 3SG hit 3SG IDEO PHONE kill 'She struck him dead with really hard blows.' b. *A naki gb66-gb66 hen lui. 3SG hit IDEOPHONE 3SG kill

The A-movement analysis would predict the order in (43b), contrary to fact:

Tonjes Veemtra

284

[vp NP, [YP, Y, IDEOPHONE

(44)

[?P

NPl

[YP2 ( • • • )

V2 t]]]]

On the other hand, the ideophone facts follow naturally from the V,-to-v-movement analysis:

[vp NP, [Y, [vp, NP2

(45)

[IDEOPHONE

t

[VP 2

•••

V2

• •• ]]]]]

Therefore, the placement of ideophones provides strong evidence for the claim that NP2 is part ofVP" and not ofVP2 • In summary, two arguments have been presented against the claim that NP 2 is part of the second VP in serial verb constructions: (a) serial verb constructions do not involve ECM; (b) the placement of ideophones. In the remainder of this section I will discuss another difference between serial verb constructions and perception verb constructions that points towards a difference in the structure of the two constructions, namely, extraction patterns. Perception verb constructions do not show island effects at all, while in the serial verb construction the second VP is a weak island. Thus, although both constructions allow argument extraction out of the embedded domain, as shown in (46) and (47), the embedded constituent is transparent for adjunct extraction in perception verb constructions (PVCs), in contrast to serial verb constructions (SVCs), as shown in (48) and (49): (46)

a.

b.

C.

(47)

a.

b.

(48)

a.

b.

A kandi stroop buta a di batao 3SG pour syrup put P DET bottle 'He poured the syrup into the bottle.' Andi a kandi buta a di bata? what 3SG pour put P DET bottle 'What did he pour into the bottle?' Naase a kandi stroop bud? where 3SG pour syrup put 'Where did he pour the syrup into?'

(SVC; argument extraction)

Ambe i si (d) dis a di konde go? (PVC; argument extraction) who 2SG see (ASP) leave DET village go 'Who did you see leaving the village?' Ambe i si di womi (d) naki gboo-gboo? who 2SG see DET man (ASP) hit IDEO PHONE 'Who did you see him hitting really hard?' Ufa a faa di pau tue? how 3SG chop DET tree throw 'How did he fell the tree?' #Di pau kai saapi. DET tree fall slow 'The tree fell slowly.'

(SVC; adjunct extraction)

Unaccusativity in Saramaccan (49)

a.

b.

Ufa i sf df w6mi ta kuIe? how 2SG see DET man ASP run 'How did you see the man running?' A bl ta kule taanga. 3SG TNS ASP run fast 'He was running fast.'

(PVC; adjunct extraction)

Sentence (48b) is not a possible answer to the question in (48a), but (49b) is possible in reply to (49a). I propose that these different extraction possibilities follow from a difference in structure. In perception verb constructions, the first verb selects the clause consisting of both NP 2 and the second VP. In serial verb constructions the first verb NP 2 and the second VP is an adjunct: (50)

a.

b.

Serial verb constructions: NP, TNS MOOD ASP V; NP2 [ASP V2 .. ']ADJUNCT Perception verb constructions: NP, TNS MOOD ASP V; [NP2ASP V2 .. ']COMPLEMENT

Apart from accounting for extraction patterns, these different structures explicate a number of other properties as well, such as the Case-marking patterns discussed earlier. 5

10.6.

ANALYSIS

I have discussed the internal make-up of the second predicate in resultative serial verb constructions in Saramaccan. I have shown that (i) VP 2 is not unaccusative; (ii) VP 2 is not passivized; (iii) the 'shared' object, NP 2, is not part ofVP 2; and (iv) VP2 is an adjunct to, and not a complement of, the first predicate. I have argued that these facts do not support the A-movement analysis of this construction. Instead, they argue for the V,-to-v movement analysis. From (i) and (ii) it follows that VP 2 is a (active) transitive VP with all arguments syntactically present. The question is how these are represented. The discussion in the literature has mainly focused on the representation of the internal argument of V 2. Two proposals have been made, both assuming that the object is Case- and thetamarked. The object is either pro or a null constant, associated with an empty operator; see (51).

(51)

a. d.

[vP 2 PRO [AspP ASP [VP 2V 2 pro]] [vP2 0pi [vP 2 PRO [AspP ASP [VP 2 V 2 ncJ]

(c£ Collins 1997) (c£ Veenstra 1996a)

, For additional arguments for the adjunct status ofVP, in serial verb constructions, see Bickerton and Iatridou (1987), Seuren (1991), and Veenstra (1996a).

Tonjes Veenstra

286

A major problem for the pro-analysis is that the obligatory argument-sharing cannot be derived, pro being a covert pronominal, free within a binding domain. Pronouns are always optionally, never obligatorily, bound. An argument for the presence of an empty operator inside the second VP comes from parasitic gap constructions. These constructions are still not a well-known area of Saramaccan grammar. The judgements of the informants vary highly on this topic. Nevertheless, it appears that wh-movement is a prerequisite for parasitic gaps for all speakers: (52)

a.

di fakai sondo wasi *(heni) bifo. sharpen DET knife without wash 3SG before 'He sharpened the knife without washing it first.' Un fakai a saapu ti sondo wasi ?(hen) bifo? which lmife 3SG sharpen without wash 3SG before 'Which knife did he sharpen without washing first?'

A saapu 3SG

b.

In (52a) a pronoun has to be present in the parasitic-gap position, in (pb) it can (marginally) be left out. The difference between the two examples is thatwh-movement has taken place in the (a)-example, but not in the (b)-example. Resultative serial verb constructions marginally allow parasitic gaps, as shown in the next example: (53)

A naki hen kn sondo sik6pu ?(hen) bifo. 3SG hit 3SG kill without kick 3SG before 'She hit him dead without kicking him first.'

Despite its marginality, I take this to be positive evidence for the presence of an empty operator in the second VP (c£ den Dikken 1991; Law and Veenstra 1992). This, then, is my analysis. Initially, the two verbs are adjacent (to accomodate the sandhi facts). The first verb moves up to a higher position (v position), giving rise to the observed word order at spell-out. The argument-sharing effect is established by the movement of a null operator from its base position inside the second predicate to a position adjoined to that predicate. The object of the first verb is co-indexed with the null operator by principles of predication (cf. Muysken 1989). Semantically, the operator movement functions as A-extraction (Hornstein 1992), thereby turning the (otherwise saturated) second VP into a predicate. This analysis treats resultative serial verb constructions as instances of secondary predication. Several researchers, for example, Muysken (1988) and Larson (1991), have observed that both constructions share a number of semantic features. Serial verb constructions display a variety of readings, ranging from conjunctive, modificatory to causative. The same range of variation is found with secondary predicates. Subject-oriented depictive predicates have a conjunctive flavour in their interpretation, object-oriented depictive predicates a modificatory reading and resultative secondary predicates come quite close to causativelresultative SVCs. If this is the case, then the study of verb serialization can shed some new light on an old controversy in linguistic

Unaccusativity in Saramaccan theory, namely, the syntactic structure of secondary predication. Two positions have been extensively argued for-the small clause analysis (Hoekstra 1988) and the complex-predicate analysis (Neeleman 1993). Both analyses are problematic in the light of the facts of verb serialization discussed here (cf. Veenstra 1996c). The small-clause analysis cannot account for the weak island effects ofVP2 • Moreover, it is crucial in this type of analysis that NP2 is part of the second predicate, which I have shown not to be the case. In the complex predicate analysis the observed word order is derived by rightward movement of the second predicate. fu such, it cannot account for the extraction facts in resultative serial verb constructions in general, since rightward movement typically has a freezing effect on the moved constituent. The overall conclusion, then, is that resultative constructions in serializing as well as non-serializing languages are best analysed as Larsonian VP-shell configurations, in which the second predicate is an adjunct to the first one.

II The Grammar Machine Hagit Borer

Il.l.

INTRODUCTION

Wnat mechanisms does the child bring to bear on the early acquisition of argument structure and grammatical relations? At least two competing hypotheses have emerged concerning this issue. The first, Semantic Bootstrapping (Grimshaw 1981; Pinker 1984; Grimshaw and Pinker 1990; see also Randall, in this volume) reduces early syntactic knowledge to the lexical semantics of particular verbs, learned from situations. Within that system, the predicate-argument structure 0f verbs, as determined by their lexical semantics, projects into the syntactic structure in accordance with universal linking principles which associate particular arguments, as specified in the lexical entry, with particular syntactic positions. In contrast, the second approach, Syntactic Bootstrapping (Gleitman 1990, 1995, and subsequent work) relies heavily on the early knowledge of the syntax of argument structure to help the child acquire the meaning of specific verbs associated with that structure. Gleitman explicitly challenged the ability, presupposed by Semantic Bootstrapping, to learn the meaning of verbs from situations, and argues that it is the syntactic structure (specifically, the subcategorization environment) which suggests to the child what the meaning of the verb may be in isolation.' I AI; is clear from later writings within both approaches, both sides acknowledge that some measure of learning from situations as well as some structural contribution are essential to the acquisition of lexical items. These approaches continue to differ, however, in the relative weight that they assign to these components, and crucially, in their assumptions as concerning the relationship between the acquisition of

The Grammar Machine Both approaches agree that there is a relationship between the interpretation of arguments and their syntactic position. They differ, however, as regarding the direction of the implication here. While Syntactic Bootstrapping subscribes to the view that it is the syntactic position which determines the interpretation of arguments and the interpretation of the lexical verb, Semantic Bootstrapping holds that it is the lexical verb which determines the nature of the arguments and their syntactic placement. Consider now these two views from the perspective of syntactic theory. Semantic Bootstrapping finds its roots in a well-established syntactic tradition, reducing argument structure to lexico-semantic information associated with single lexical items. The assumption that the syntax associated with argument structure is thus determined is shared by many approaches to argument structure developed in the 1980s and early 1990s.2 These approaches, differ as they may on other matters, share the assumption that the appropriate lexical representation of the verb contains information on the syntactic projection of its arguments, making the latter deterministically dependent on the former. A syntactic theory of argument projection which is compatible with Syntactic Bootstrapping, on the other hand, could be, potentially, quite different. Specifically, if we assume that children do have in their possession the knowledge allowing them to assign interpretation to arguments independently of lexical entries, as Gleitman assumes, it raises the distinct possibility that for adults, too, argument structure is computed syntactically, and independently oflexical information, thereby stripping the verbal lexical entry ofits crucial role in the determination of the projection of arguments for adults as well as for children. Instead, the interpretation of arguments would proceed along 'constructionist' lines: the syntactic configuration of the arguments would determine their interpretation, and the verb, rather than being the determinant of structural properties, would serve as the modifier of the resulting event structure. 3 Now, if this is the correct approach to the adult projection of arguments, an interesting prediction emerges concerning acquisition. If the projection of arguments is not related to lexical knowledge of any sort, we expect it to be available in the absence of lexical knowledge. Specifically, we expect the child, potentially, to go through a developmental stage in which the syntax of argument structure is known, but knowledge of the properties of specific vocabulary items may be missing or fuzzy. If such a stage indeed exists, it would cast serious doubt not only on Semantic Bootstrapping, lexical items and the acquisition of structure. See discussion later in this section. It should be noted that Gleitman does not assume that argument structure, for adults, is independent oflexical entries. Rather, she suggests that Syntactic Bootstrapping is the means by which information concerning argument structure associated with specific verbs (and hence the precise meaning of those verbs) is acquired. 2 See, among others, Williams (1981b); Baker's (1985, 1988) Universal Theta-Assignment Hypothesis; di Sciullo and Williams (1987); Conceptual Structure, as developed by Jackendoff (1990) and subsequent work; the linking approach developed in Levin and Rappaport Hovav (1995 and previous work); and, more recently, Reinhart (1996, 2000). l See Goldberg (1995) and Fillmore and Kay (1997) for discussion of Construction Grammar. See Marantz (1997) and Borer (2000, forthcoming) for some discussion of the role of 'constructions', in this specific sense, in Universal Grammar. We note that crucially, in the approach put here, as well as in Marantz (1997), the interpretational role of constructions is compositional and universal, in a departure from traditional Construction Grammar approaches.

Hagit Borer but also on any linguistic model which projects, for adults, argument structure on the basis of information in the lexical entry. Instead, it would support Syntactic Bootstrapping, alongside a model of argument structure in which the link between the syntactic position of arguments and their interpretation is independent of the properties of anyone particular vocabulary item. Within such a model, the traditional lexicon must be eliminated and replaced, rather, by a vocabulary list of some sort, in which some morpho-phonological and semantic information may be associated with vocabulary items, but little or no syntactic information. Systems assigning interpretation to arguments entirely independently or partially independently of information in lexical entries have been developed by a number of grammarians in the past few years, largely under the influence of the seminal work of Hale and Keyser (1993). Kratzer (1994, 1996), as well as Harley (1995) and Marantz (1997), among others, assumes that external arguments are assigned structurally, through the mediation of a functional head (VoiceP for Kratzer, v for Harley and Marantz). Both external and internal direct arguments are assigned structurally in proposals made by van Hout (1992, 1996), Borer (1994, 1998), and Ritter and Rosen (1998), among others. Specifically, in Borer (1994, 1998, forthcoming) I argue that arguments are assigned interpretation in functional specifiers of nodes associated with event structure. Event structure, within that system, is not determined by properties of the vocabulary, but rather, by the optional merger of specific functional heads with particular semantic values. Substantive vocabulary items, in turn, function as modifiers of the emerging event structure. The optional merger of nodes which give rise to varying event structures, together with the modifYing nature of substantive vocabulary items, gives rise to the emergence, for any particular vocabulary item, of multiple event structures and multiple argumental interpretation. In fact, it predicts massive 'ambiguity' for anyone verb. In view of this prediction, consider the paradigm in (I) (from Clark and Clark 1979).

(I)

a. The factory horns sirened throughout the raid. b. The factory horns sirened midday and everyone broke for lunch. c.

d. e.

The police car sirened the Porsche to a stop. The police car sirened up to the accident site. The police car sirened the daylight out of me.

We note that if the syntax of the arguments and the event structures in (la-e) are to be attributed to the properties of some verbal lexical entry siren, we would have to assume that there are five distinct entries for siren, the one in (la) associated with an atelic agentive reading, and meaning 'to emit a siren noise', the one in (Ib) associated with a telic agentive (and theme?) meaning 'to signal through emitting a siren noise', the one in (IC) associated with a telic agent-patient and meaning 'to force by emitting a siren noise', the one in (Id) associated with telic-agentive, and subcategorizing a particle, meaning 'to hurry while emitting a siren noise', and finally, in (Ie), siren would be associated with a stative and an experiencer, and would mean 'to frighten byway of emitting a siren noise'. Of course, the common denominator here is the emission of a

The Grammar Machine siren noise, which, indeed, appears to be the meaning of to siren, but it is entirely clear that in each of (la-e) the event denoted is modified by the emission of a sound, rather than determined by that emission. Thus at least in (la-e), we must assume that the syntax of the event (and the syntax of the event's arguments) does not emerge from five different lexical entries for siren. Rather, it is the syntax which determines the interpretation of the event and its arguments, as well as the specific nuance contributed to that interpretation by the vocabulary item siren which modifies that event. If it is correct to assume that argument structure is largely syntactically determined, rather than lexically, and if it is further correct to assume that the building blocks of argument interpretation are nodes associated with event structure, rather than thematic structure, then we derive the result that the relationship between structure and argumental interpretation must be fixed, but nevertheless, different verbs need not occur with the same syntax. Thus, for instance, if destructible is embedded within a stative event structure, and is not itself associated with the assignment of any semantic roles, we may think of its subject in the piano is destructible, for example, as subjectof-state. On the other hand, in I destroyed the piano, the same stem, destroy/destruct, is embedded within a telic event structure, and the piano is subject-of-result, or subject-of-change, plausibly occupying a distinct syntactic position from that occupied by subject-of-state. In contrast, in approaches driven by the Universal Theta-Assignment Hypothesis (UTAH), destroy/destruct is lexically associated with a theme which must always project in an identical syntactic position, a restriction that has proven difficult to reconcile with the differing syntactic properties of, for example, the subject of [A destructible] and the object of [vdestroy] (see Baker 2001 for a weakening of UTAH so as to accommodate this difficulty). That the interpretation of arguments is indeed dependent on event structure, and that it is independent of properties of verbs is argued in detail in van Hout (1992, 1996), Borer (1994, 1998, forthcoming), and Ritter and Rosen (1998). These researchers focus on the well-known correlation between the syntax of the unergative-unaccusative distinction and its event interpretation (see Dowty 1991), alongside the fact that most intransitive verbs occur in both contexts, exhibiting variable behaviour. Van Hour (1992,1996) further discusses transitivity alternations associated with single verbs (such as move-move; drop-drop), arguing that here, too, it is the syntax of the argument structure which determines the event structure, rather than lexical information associated with distinct (related) lexical entries for move. TRANS and move.intrans, for example. 4 Acknowledging the challenge to the projection of arguments from lexical entries posed, specifically, by the correlation between structure and interpretation for intransitive variable-behaviour verbs, Levin and Rappaport Hovav note: Verbs which show variable behaviour [between unaccusative and unergative1are always associated with more than one meaning; each meaning turns out to be correlated with the predicted syntactic properties ... the question ... is whether the change of meaning ... is to be attributed 4 Although van Hout (I996) does assume that verbs may be lexically marked as to whether or not they are telic.

Hagit Borer to [the verb's] appearance in a particular construction ... or to the existence of some lexical rule which gives rise to multiple semantic classifications of verbs, which then license the appearance of these verbs in more than one construction. 5 Levin and Rappaport Hovav (I992b: 12, 13)

Wishing to preserve the projection of arguments from lexical entries, Levin and Rappaport Hovav (1992b, 1995) opt for the second solution, assuming that variable-behaviour verbs, occurring in more than one syntactic environment with distinct interpretation in each configuration, do so due to the existence of some lexical rule which gives rise to multiple semantic classifications of verbs (and see also Reinhart 1996,2000).6 Returning to Semantic vs. Syntactic Bootstrapping, note that the question posed by Levin and Rappaport Hovav is the same question posed here concerning the type of acquisition device used by the child and its relation to the adult grammar. From the perspective of an acquisition theory which assumes Semantic Bootstrapping, there must exist a lexical rule which gives rise to multiple semantic classification of variablebehaviour verbs, which then licenses the appearance of these verbs in distinct syntactic contexts. In a theory that views the lexical entry as the sole source of information on the syntactic projection of argument structure, no other possibilities exist, as no other source is available, for child or for adult, for the syntax of arguments. On the other hand, if Syntactic Bootstrapping is on the right track, it is indeed possible, in Levin and Rappaport Hovav's terms, that 'the change in meaning ... [is] to be attributed to [the verb's] appearance in a particular construction, maldng the existence of lexical rules of the sort discussed by Levin and Rappaport Hovav (op. cit.) or Reinhart (op. cit.) unnecessary. Instead, it would be the syntax of the arguments that would determines their interpretation (as part of the event interpretation), and that interpretation would shift when the syntactic position of the arguments is different, regardless of the specific verb used. In turn, if it is correct to assume that it is the syntax of event struc-

5 And see also Levin and Rappaport Hovav (1995: 208), where it is stated that 'The question is whether multiple meaning are handled via principles or rules specific to the lexicon, or whether they can be shown to reduce to properties of syntactic configurations'. 6 Levin and Rappaport Hovav (1992b) propose that variable-behaviour intransitives involve a lexically-marked telicity alternation. Levin and Rappaport Hovav (I995) depart from that assumption, proposing instead that the distinction should be characterized as involving an external vs. an internal causer (and see Reinhart, 2000 for a detailed criticism of this latter assumption). The main point we wish to make, however, is independent of any particular lexical account for variable-behaviour intransitives. Rather, it concerns the lexicon-syntax-interpretation interface. Specifically, we ask whether in the presence of two syntactic structures, each associated with a distinct interpretation, one should reduce the distinct interpretation to the distinct syntactic structures, or rather, assume that the distinct interpretation is to be traced back to two (related) lexical entries, which, in turn, project distinct structures. Whether the interpretational difference involves telicity or any other appropriate semantic classification is clearly orthogonal to our main interest here. Levin and Rappaport Hovav (1992b, 1995) as well as Reinhart (1996, 2000) opt for the lexical solution, regarding the lexical entry as the ultimate source of information on syntactic projection. Here, as well as in Borer (1994, 1998, forthcoming), the opposite view is taken, advocating the determination of the interpretation by the structure, independendy of lexical properties. While the specific syntactic distinctions used here relate ro event structure, other executions compatible with syntactic, rather than lexical, projection of arguments are possible and have, indeed, been proposed. For the detailed justification of the event structure approach, see Borer (forthcoming).

The Grammar Machine

293

ture, rather than lexical entries, which determine the interpretation of arguments for adults, then Syntactic Bootstrapping does not just become a plausible hypothesis, but the only hypothesis compatible with the adult grammar. Quite simply, the child could not project argument structure from a vocabulary item, regardless of her knowledge of that item, as such an item does not contain information concerning the syntax and the interpretation of arguments. Such information is available exclusively through the syntactic structure'?

11.2.

SYNTACTIC EVENT STRUCTURE IN A NUTSHELL

From here on, I will assume-without further justification-the structures in (2), following Borer (forthcoming), where they are justified in great detail. Importantly, however, the logic of the argument to be put forth here is independent of the particular syntactic structures used here. To the extent that it can be established that the projection of arguments is independent of the properties of substantive vocabulary items, clearly some syntactic structure must exist such that it represents the unique linking between structural positions and argument interpretation. The specific properties of that structure could then be the subject matter of a separate debate. 8 (2)

a.

Transitive, telic [EP DP, [TP BP, [ASpQ DP2 [ypV]]]] (in two hours/*for two hours) NOM

ACC

7 Randall (in this volume) puts forth a specific execution of the Semantic Bootstrapping hypothesis with respect to intransitive verbs. We note, however, that the experimental results reported by Randall test the knowledge, on the part of children and adults, of the syntactic contexts which are associated with specific events, and is hence compatible, in principle, with an approach that attributes this knowledge to the understanding of the meaning of the verb. as well as to the understanding of the entire event. S The structutes in (2) are partial syntactic representations, focusing on the syntactic placement of arguments. Facets of the syntactic structure which are not directly relevant-such as verb movement or the status of the VP-are largely ignored here for the sake of simplicity. As for functional structure, in addition to the event structure marked by the aspectual nodes EP and ASP Q> only TP is marked. I take no position on the necessity of additional functional structure between ym'" and CPo In Borer (2000) I argue that lexical category labels such as V, N, and A are determined by the functional structure dominating substantive vocabulary items. Slightly simplifYing. a TP or an EP would render an (underived) vocabulary item dominated by it a verb, while a DP would render an (underived) vocabulary item dominated by it a noun etc. In other work I also suggest that arguments which are not interpreted in functional specifiers must project as PPs. As these claims are not crucial to the acquisition discussion in the rest of the chapter, they are largely ignored, as is the relations between the numeration and the subsequent emerging phrase structure. Following much literature (see Verkuyl 1989 and Parsons 1990, among others), I assume that there is no structural distinction between achievements and accomplishments. Consequently, the structures put forth here do not reflect such a distinction, and the main dividing line, within non-stative events, is between events which are either telic (accomplishments and achievements) or atelic (processes, at times known as 'activities'). See Borer (forthcoming) for some discussion, as well as the demonstration that so-called achievements are not a unified class. Stative events are not treated here.

Hagit Borer

294

b.

Transitive, atelic DP, [YP BP, [FP DP2 [ypV]]]] (*in two hours/for two hours)

[EP

NOM

c.

PRY

Intransitive, telic [EP DP, [YP BP, [ASpQ BP, [ypV]])] (in two hours/*for two hours) NOM

d.

Intransitive, atelic [EP DP, [yp BP, [ypV]]]] (*in two hours/for two hours) NOM

In (2), EP is an eventive (non-stative) node and ASP Q is a quantity node which induces telicity.9 Unless the DP in [Spec, EP] is already assigned an interpretation in some other functional specifier ([Spec,ASPd, for instance), it is interpreted as the originator of the (eventive) event headed by E. The DP in [Spec,ASPQ] is interpreted as the subject of a quantifiable change. Following Verkuyl (1972, 1989), Tenny (1987, 1994), and Krifka (1992), among others, I will assume that a quantity DP, in turn interpreted as subject of quantifiable change, is necessary to give rise to telicity.1O Finally, I assume that ASP Q (may) check accusative Case for the DP in its specifier (the subject-ofquantifiable-change), as in (2a). Based on the aspectual properties of partitive Case in Finnish as described by Vainikka and Mailing (1993) and Kiparsky (1998), I assume that partitive Case (PRY) marks the absence of telicity, and that it is the Case assigned to the direct argument in atelic configurations, as in (2b). In English and Hebrew, objective case marking does not differentiate morphologically between accusative and partitive, but such a distinction is marked in Finnish. Likewise, Slavic languages mark morphologically the projection of ASP Q (as perfective, and see Borer, forthcoming, for a detailed discussion). Considering, specifically, an English verb such as move, it may be embedded in all structures in (3), with the following result:

(3)

a.

Transitive, telic Kim, [YP ffimI

[EP

[ASPQ

NOM

originator

b.

Transitive, atelic [EP Kim, [yp ffimI NOM

originator

the piano 2 [yp move ]]]] (in two hours) ACC

subject-of-quantifiable-change [FP

the piano 2 [yp move]]]] (for two hours) PRY

default participant*

9 ASP Q corresponds structurally (but notinterpretationally) to ASPE , in Borer (1994, 1998). EP ('Event Phrase') largely corresponds to ASP p of Borer (1994, 1998). See Borer (forthcoming) for the relevant dis-

cussion. to This is again simplifYing somewhat. See Borer (forthcoming) for cases in whichtelicity emerges without a DP altogether, or with a non-quantity DP. For a discussion of telicity with non-quantity DP see also Mittwoch (1991). I am setting aside here issues concerning the role of the internal argument as measuring out the event as proposed by Tenny (1987, 1994) (see also Krifka's 1992 'measure theme'). For a review of some problems, see especially Schein (2002), Rothstein (2000), and Borer (forthcoming).

The Grammar Machine c.

Intransitive, telic [EP Kim, [TP IBm"

[ASPQ

295

Ifun, [yp move ]m (in two hours)

NOM

subject-of-quantifiable-change

d. Intransitive, atelic [EP Kim, [TP Ifun, [yp move]]]] (for two hours) NOM

originator (*default participant = a pragmatically appropriate participant (see Borer I994, forthcoming, for discussion.)

Crucially, the structures in (3) all exist independently of the verb inserted in them, and each has a fixed event structure regardless of that verb. Recall that the verb in this system acts, essentially, as a modifier. Combined with the arguments in (3a-d) being relations with the event, rather than of the verb, the resulting interpretation is best captured by the (neo-Davidsonian) representations in (4). (4)

a.

b.

c. d.

Transitive, telic :3e (eventive, e) & originator (Kim, e) & subject-of-quantifiable-change (the piano, e) & (move, e) ) Transitives, atelic :3e (eventive, e) & originator (Kim, e) & participant (the piano, e) & (move, e) Intransitive, telic (unaccusative) :3e (eventive, e) & subject-of-quantifiable-change (the piano, e) & (move, e) Intransitive, atelic (unergative) :3e (eventive, e) & originator (the piano, e) & (move, e)

Given the modifier status ofvocabulary items in this system, cases ofmismatch between the syntax of an event and the specific verb inserted into it (*Kim ran in two hours) are to be ruled out using the very same system that would rule out inappropriate modification-for example, *John ran gradually, *Kim deliberately understood the solution, *Pat feared the storm energetically, etc. (see Borer 2000, forthcoming, for discussion). We note now that the transitive derivations available for English (3a-b) are not straightforwardly available for adult Hebrew. With a few exceptions, Hebrew marks intransitive-transitive pairs morphologically. For zaz, 'move.INTRANS', the transitive form would be heziz, 'move.TRANs'. For that reason, an utterance such as (5a), the equivalent of either (3a) or (3b) in English, is ungrammatical. But as predicted, (5b), the correct transitive form, heziz, is ambiguous between a telic and an atelic reading. (5)

a. *Ranzaz 'et ha-ricpa. Ran moved-INTRANS OM the-floor b. Ran heziz 'et ha-ricpa (be-mdek saCatayimltok saCatayim). Ran move-TRANS OM the-floor (for two hourslin two hours)

Hagit Borer Likewise, zazis ambiguous between an unaccusative/telic and an unergative/atelic reading. That this is indeed the case can be illustrated by using the unaccusative-unergative tests suggested in Borer and Grodzinsky (1986), where it is shown that the possessor dative in Hebrew must possess a DP within the classical complementation domain. So when it occurs in intransitive contexts, it is compatible only with unaccusatives (and passives), while a reflexive dative in Hebrew may only be coindexed with a true 'external argument'. When reflexive datives occur in intransitive contexts they are compatible only with unergatives. In Borer (1998) it is further shown that these tests correlate with event structure. Thus, in (6b), in the presence of a reflexive dative, atelicity is obligatory; the modification with the telic modifier in two hours is ungrammatical. Similar logic predicts the possessor dative in (6c) to induce telicity, and hence the ungrammaticality which results in the presence of an atelic modifier such as for two hours:"

(6)

a. b. c.

ha-ricpa zaza (be-mdek saCatayim/tok saCatayim). the-floor moved (for two hours/in two hours ha-ricpa zaza h! (be-mdek saCatayim/*tok saCatayim). the floor moved to-it (for two hours/*in two hours) ha-ricpa zaza Ie-rani (*be-mdek saCatayim/tok saCatayim). the floor moved to-Rani (*for two hours/in two hours)

We note, then, that zaz and heziz, intransitive and transitive 'move', respectively, are each associated with two syntactic event structures-not with four, as is the case with English move. Rather, zaz may only modifY intransitive structures, while heziz modifies transitive ones. In section 11.3.2 I return to the characterization of the zaz-heziz alternation in the grammar of Hebrew, and to the manner in which this additional restriction is to be characterized. Focusing for the time being on the ways in which verbs leave the syntactic projection of arguments undetermined, as in English, or underdetermined, as in Hebrew, we note that if the system sketched here for the adult syntactic representation of argument structure is correct, it has clear consequences for language acquisition. First, it makes the acquisition of syntactic structures based on verb meaning impossible. In this system, a vocabulary item is categorially and syntactically undetermined (or underdetermined), and specifically, it makes no reference to the syntactic projection of arguments. The full interpretation of the arguments is computed on the basis of the syntactic structure of the entire predicate. While it is headed by a substantive vocabulary item, in turn verbalized by the functional structure, the meaning of that vocabulary item is syntactically uninformative. Rather, in such a system, the acquisition of syntactic structures must proceed independently of the acquisition of vocabulary. In reference to our concrete examples, even if children have U Importantly, the respective morpho-phonological form of zaz and heziz do not suffice to determine the number of arguments associated with either. The morpho-phonological template associated wirh zaz (binyan I) is often associated wirh transitive verbs, at times extremely close to the meaning of transitive heziz (e.g. daxat, ('push'); sarat ('burn.TRANs')) while rhe morpho-phonological form of heziz is often associated wirh intransitive verbs, including inchoative ones (e.g., higlid ('form-scab'); hi!zri' ('get healthy'), etc.). See Tables 11.1 and 11.2 and related discussion.

The Grammar Machine

297

acquired the meaning of English move or Hebrew zaz in isolation, it would provide them with no information on whether or not to project a single argument in the specifier of ASP Q ('internally'), as in (3c) and (6c), or in the specifier of EP ('externally'), as in (3d) or (6b). For English, it would further leave learners with no information on whether to project move in nominal structure or in verbal structure, and in the latter case, whether to project it within a dyadic or monadic event structure. But if the child has knowledge of the projection of arguments independently of the properties of the verbs associated with the resulting structure, an interesting prediction emerges. We predict that it should be possible, in principle, for the child to pass through a stage where the syntactic event structure is fully in place, but vocabulary knowledge is impaired. If such a stage turns out to exist, it would lend strong independent support not only to Syntactic Bootstrapping, but also to the independence of argument structure from vocabulary and to a syntactic approach to the projection of arguments. The remainder of the chapter is devoted to showing that there is indeed such a stage in language acquisition. I will argue that children acquiring Hebrew pass through a stage in which their performance has precisely these characteristics: they appear to have full knowledge of the syntax of argument structure, complete with nominal and verbal syntactic functional structure in place, as attested by word order, case markers, and tense and agreement inflection. Nevertheless, that knowledge could not possibly be coming from the actual verbs used, as these are often employed in incorrect syntactic contexts, or are altogether non-existent in the adult vocabulary. At that stage, I will suggest, children often make decisions on vocabulary insertion based on morphophonological factors alone, being oblivious to the way in which the correct selection of a morphological template is conditioned, for adult Hebrew, by syntactic factors. Far from projecting the syntax as based on syntactic properties of vocabulary items, their knowledge of vocabulary items is deficient precisely in that respect. Instead, they often embed a morpho-phonologically correct, but morpho-syntactically flawed, form in a nearly perfect syntactic structure, complete with tense and agreement markings, providing evidence that their ability to do so is entirely independent from their vocabulary knowledge. '2 Specifically, I will propose the following developmental sequence: (7)

Naming> the morpho-phonology stage> the morpho-syntax stage> adults

t l We note, before proceeding, that by necessity, any data on flawed vocabulary knowledge on the part of the child is limited both by comprehension and by the lack of access to direct introspective data. For instance, if a child utters a totally inappropriate verb--existing or non-existing-in the context of some arguments (e.g., mommy poured (at) the table), comprehension fails altogether. Any reasoning concerning the knowledge of argument structure is hampered by our inability to grasp whether the placement of arguments does or does not correspond systematically to the interpretation the child might have had in mind. Similarly, if the child says mommy touched the ball, we assume, plausibly, knowledge of both verb and argument structure, although, of course, it is possible that the child has mis-lexicalized touch as meaning 'throw-at', 'kick', 'move', etc. It therefore emerges that 'most 'errors' that can actually be studied are those which exhibit at least partial vocabulary knowledge-and are therefore transparent to our investigation-rather than cases of radical misuse ofvocabulary. The latter may not be studied systematically on the basis of spontaneous production data.

298

Hagit Borer

Il.3.

THE MORPHO-PHONOLOGY STAGE

II.p. Valence neutralization Berman (1982, 1993, 1994), observes that there is a stage in the early acquisition of Hebrew in which errors such as those in (8) are quite common:

(8)

a.

b.

c.

ra'iti.I 'et ha-ciyurim le-'aba. saW.ISG OM the-paintings to-daddy. 'I showed the paintings to Daddy.' (c£ adult her'eti.V 'showed') 'ani roca se-'aba yokal.I 'oti cakSaQ. I want that-daddy eat.FUT.3.SG OM.me now 'I want Daddy to feed me now.' (cf. adult ya'akil.Y 'feed.FuT.3.se') 'ima zuzi.I li 'et ha-kise. mommy move.INTRANS.IMPERATNE to-me OM the-chair 'Mommy, move the chair for me.' (cf. adult tazizi.V 'move.TRANs.IMPERATNE')

In (8), the child is using the wrong morpho-phonological template, binyan, in each case. The binyan the child is using is marked immediately following the form used (ranging from I to VII). For example, ra'iti.I is the root RE used in binyan I. The gloss in (8) indicates what the adult interpretation of the form used by the child is (for instance, ra'iti means 'saw.IsG' for the target language). The translation gives the (presumed) meaning of the utterance for the child ('show'). Correct adult forms, for the meaning intended by the child are given under each form, together with the adult binyan membership (such as her'eti.V, 'showed.lse', inflected in binyan V). We note, now, that in (8a-c) the child produces syntactically correct forms, placing the arguments a,ppropriately for the (presumed) meaning intended. Further, in (8a-c) the child uses the correct root to express the meaning she has in mind. However, the morpho-phonological form of the verb, the binyan, is wrong when matched against the syntax. For example, for adults, the form zaz.I is disallowed in transitive contexts, and the form associated with the root zz in transitive context must be heziz.V. The cases in (8) appear to cluster, in that all erroneous forms are of binyan 1, and all avoided correct forms are of binyan V. Further, all cases in (8) are cases of valency increase (intransitive form used transitively, dyadic form used triadically). However, as already shown by Berman (1982, 1993, 1994), and as is clear from the study of the CHILDES files ofNa'ama (see (9», this by no means characterizes early performance. In (9), dyadic mesader.III ('arrange'), madbiq.V ('stick TRANS') and moci.V ('talceout. TRANS') are used monadically. In the latter case, the child uses binyan V instead of binyan I, in direct opposition to the pattern in (8). Similarly, transitive forms are used intransitively in (10).

The Grammar Machine (9)

a.

b.

c.

(10)

a.

b.

299

ze 10 mesader.III Na'ama,2;2 this no arrange 'It doesn't fit/become arranged.' (c£ adult mistader.Yn 'get-arranged.INTRANS') ze 10 madbiq.V Na' ama, 2;2 this no stick. TRANS 'It doesn't stick.' (cf. adult nidbaq.II 'sticluNTRANS') ken hu moci.Y lebad Na'ama,2;3 yes he take-out alone 'It comes out by itself, too.' (c£ adult yoce.! 'come out') tir'i 'ek kol ha-xalab saj2.ak..I look how all the-milk spilled. trans 'Look how all the milk spilled.' (cf. adult nifpak.II 'spilled.INTRANs') lama ha-delet 10 potaxat.I? why the-door no open. TRANS 'Why doesn't the door open?' (c£ adult nij2taxat.n 'open.INTRANS'; Berman 1982)

Some additional cases, tal(en from Na' ama's CHILDES files, together with cases cited in Berman illustrate that errors occur, in fact, in all possible morpho-phonological directions, regardless of valence or binyan. Following Berman, I will refer to the (erroneous) use of one morphological form in more than one valence context as (Valence) Neutralization. (n)

Valence neutralization

a.

Intransitive forms used transitively (valence increase) Adult form

i. hi/i'i/z.V'hurt.cause' ii. hiS'ir.V'leave.TRANs' iii. liklek.III 'soil'

b.

Neutralized child form

Adult meaning of neutralized form

kaa/z.I niSar.n hitlaklek.Yn

'hurt.INTRANS' 'stay' 'become-soiled'

Transitive forms used intransitively (valence decrease) Adult form i. ii.

Neutralized child form

nizraq.II'thrown.pass' zaraq.I mitxabeq.Yn 'hug.redp.' mexabeq.III

Adult meaning of neutralized form 'throw. active' 'hug. TRANS'

These data already suggest that children are capable of projecting argument structure correctly, although their knowledge of the specific vocabulary item which that structure is paired with is flawed. If, indeed, the syntactic projection of arguments is

HagitBorer

300

independent of information associated with specific vocabulary items, the behaviour illustrated above can be readily explained. This has already been observed by Berman (1982, 1993), who proposes, correctly in my view, that some syntactic knowledge must precede fulllexicalization. What, however, does that syntactic knowledge consist of? In section 11.3.2, I will make a specific claim as concerning the nature of that early knowledge. In section 11.4 I turn to some questions which the model of early lmowledge must address before being complete. In section 11.5 I outline the next developmental stage, labelled in (7) as the morpho-syntax stage, and turn to speculations as concerning the passage from the morpho-syntax stage to adult lmowledge.

The Hebrew binyan system and the early knowledge of it

11.3.2.

Before proceeding to a description of the early grammar, a brief review of the Hebrew binyan system is in order. In Hebrew, a Semitic language, verbs are formed based on a consonantal root, mostly consisting of three consonants, but at times consisting of two or four. These roots, loosely associated with a meaning, are not in and of themselves associated with either syntactic category or argument structure. To illustrate, a root such as KTB, loosely associated with the meaning of writing, can occur in all forms in (12), associated with distinct categories and with different argument structures, when they are verbaP3 (12)

a. b. c. d. e.

f g. h. i. j.

k. I.

kata/z 'wrote' (verb.!, dyadic) nikta/z 'was-written' (verb.II, monadic) hikti/z 'dictated' (verb.V; triadic) hukta/z 'was-dictated' (verb.VI, dyadic) hitkate/z 'corresponded' (verb:vII, symmetrical) mikta/z 'letter' (noun) makte/za 'desk' (noun) kat/zan 'typist' (noun) kto/zet 'address' (noun) ktuba 'marriage contract' (noun) kta/z 'hand-writing' (noun) kata/z 'correspondent' (noun)

Roots are, in turn, embedded within different templates, consisting of vocalic melodies as well as affixes. We centre on verbal templates, the binyanim, of which Hebrew has seven. Each binyan is associated with specific morpho-phonological properties, IJ AI; is traditional in Semitic linguistics, forms are given as past tense, third-person masculine singular. The list in (12) does not contain nominals derived from verbs. AI; the text discussion focuses on syntactic properties of binyanim, the phonological representation is intended to highlight the relevant syntactic properties. While vocalic representation follows Modern Hebrew pronunciation, representation of roots is designed to highlight use of identical root across forms. For instance, spirantized Ikl and lxI, pronounced identically in Modern Hebrew, are nevertheless represented as Ikl and lxI, respectively, in order to highlight the fact that Ikl is the same consonant, in a root, as Ik/, but distinct from Ix/.

1he Grammar Machine

30I

and linked to its morphologically specific set of participial forms, derived nominals, agentive nominals, etc. as illustrated in (13) for binyan III and binyan V.

(13)

1he morpho-phonologicalparadigm, binyanim III and V Verb.psT.3.SG Infinitive Derived nominal Agentive nominal Passive participle/adjective

Binyan III biteax 'insure'

le-/zateax bituax me/zateax me/zutax

Binyan V

hibtiax 'promise' le-ha/ztiax ha/ztaxa ma/ztiax mu/ztax

No doubt you note that while the diachronic association between the meaning of 'insure' and 'promise' is rather obvious, nevertheless it is sufficiently different in current, synchronic use to warrant distinct root listing. And indeed, different roots occurring in differing binyanim may have quite diverse meanings. Further, even with respect to roots which have a consistent meaning acr.oss different binyanim, while some of the binyanim are canonically associated with particular argument structure configurations and particular interpretations (see Table 11.1), the link is pretty loose, and the bulk of Hebrew verbal vocabulary does not actually conform to these canonical generalizations, as Tables II.2 and 11.3 illustrate. These canonical generalizations do hold, however, for productive word-formation operations, guided, in Modern Hebrew, by syntactic and morpho-phonological regularities. '4 The pairing of a particular argument structure and a morpho-phonological token is almost always unique, which is to say, a particular form in a particular binyan is either transitive or intransitive, but is almost never both. This is in sharp contrast with English, for instance, where verbs such as move, drop, shake, etc. may be either transitive or intransitive (but see row 5 in Table 11.3 for an exception to this generalization in Hebrew). Nevertheless, the particular binyan associated with transitive or intransitive interpretation is not predictable from the argument structure alone. Thus any binyan may be intransitive. The degree of unpredictability in matching a particular binyan with argument structure configuration is such that we must assume that adult speal<:ers of Hebrew have access to a mental reservoir which contains listed pairings of roots and binyanim, associated with particular syntactic structures (for example, zz.I is only appropriate in intransitive contexts, DXP.I is appropriate in a transitive context). While the productive word-formation component of the adult grammar is overwhelmingly regular, pairing specific binyanim with predictable argument-structure configurations-subject only to some morpho-phonological restrictions-this productive generative system is peripheral, by and large, to the existence of a list (for the adults) of the particular morpho-phonological binyan associated with particular

14 For discussion, see especially Bolotsky (I978) and subsequent work. One comment is noteworthy from the morpho-phonological perspective: binyanim III and VII are (potentially) morpho-phonologically quadro-consonantal, malcing them the only possible choice for morphological innovations based on quadro-consonantal roots, regardless of their morpho-syntactic properties.

Hagit Borer

302 TABLE

ILL

Basic Hebrew binyanim and their canonical properties

Binyan

Canonical argument structure

MorphoProductive phonology

None

No

Tri-consonantal roots only

sal2ar- 'broke.TRANs'; najlal- 'fell'; caxaq- 'laughed'

Always intransitive (-OM); a. middle; unaccusative; b. passive of binyanI

No

Tri-consonantal roots only

a. namas- 'melt. INTRANS'; nijltax- 'open. INTRANS' , b. nora- 'shot.PASS

Transitive

Yes

Quadro-consonantal roots possible

pocec- 'blew Up.TRANS'; Jilem- 'paid'; sereq- 'combed'

Yes

Tri-consonantal roots only

a. hijlxid- 'scared'; he'edim- 'redden.TRANs' b. hif2.ri'- 'healed.INTRANs'; he'edim- 'redden.INTRANs'

Yes

Quadro-consonantal roots possible

a. hitpocec- 'blewUp.INTRANS'; hiftalem'paid Off.INTRANS'

CaCaC II

niCCaC

III

CiC(C)eC IV

CuC(C)aC V

hiCCiC

Internal passive of binyan III (-OM) a. transitive, causative b. intransitive, inchoative

VI huCCaC

Internal passive of binyan V (-OM)

VII

Always intransitive (-OM)

hitCaC(C)eC

Examples

a. inchoative, often related to binyan III

b. histareq'combed. reflexive'

b. reflexive/ reciprocal of binyan III

syntactic structures in the context of specific roots. This list is not fully predictable from a regular rule system. A full model of vocabulary insertion is outside the scope of this chapter. For concreteness, however, suppose we follow Anderson (1992) in assuming that functional structure is associated with (inflectional) features, and that substantive vocabulary items associated with such functional structure become marked with these inflectional features. In turn, such inflectional features trigger the application of particular

The Grammar Machine phonological operations, resulting in the insertion of phonological material, following the syntactic derivation. For concreteness sake, we may assume that the phonological component consists of a search for that phonological representation which matches in features the abstract properties of a particular syntactic node, reflecting its syntactic derivational history. I will diverge, specifically, from both Anderson (1992) and Halle TABLE 11.2. Intransitive, atelic motion verbs-binyan membership

Motion Verb

Interpretation

ZaZ hiStoBeB NaDaD RaC hitRoCeC QaPaC QiPeC DiLeG HaLaK hitHaLeK neXPaZ MiHeR hitQaDeM hiMSiK GaLaS hitGaLeC hiSMiL

moved wandered around wandered ran ran around jumped jumped repeatedly skipped walked walked around hurried rushed progressed continued slid slid veered left

Binyan

I VII I I VII I

III III I VII II

III VII VI I VII V

TABLE I 1.3. Inchoative-causative alternations-binyan correspondences

Alternation Inchoative

Causative

nisra)!.l1 'burn.INTRANs' ni)!tax.II 'open.INTRANS'

sara)!.1 'burn. TRANS' patax.l 'open. TRANS'

ta/zcf.l 'drown.INTRANS camax.l 'groW.INTRANS'

hitbiac.V 'drown.TRANs' hicmi'ax.V 'grow. TRANS'

3. VII-III

hitpareq.VII 'fall-apart' hitgalgel.VII 'roll.INTRANS'

pereq.III 'talce apart' gilgel.I1I 'roli.TRANs'

4. II-V

nifxat.II 'become-ruined' nirta/z.II 'wet.INTRANS'

hifxit.v 'ruin' hirti/z.v 'wet. TRANS'

5. V-V

he'edim.v 'redden.INTRANs' hi/zri''y 'heal.INTRANS'

he'edim.V 'redden. TRANS' hi/zri''y 'heal. TRANS'

6. I-III

gadal.I 'groW.INTRANS'

gidel.III 'groW.INTRANS'

7. VII-V

hit'adem.VII 'redden.INTRANs'

he'edim.v 'redden.TRANs'

1. II-I

2.

I-V

Hagit Borer and Marantz (1993) in assuming that at least some phonological material must be associated with vocabulary items throughout the derivation, and specifically, I will assume that the consonantal root is present in the structure throughout the derivation, functioning as a modifier of the structure (see sections lI.I and Il.2 for discussion). Consider, as an illustration, the structures in (2) in conjunction with the consonantal roots ZZ, ('move') and SRP, ('burn') (and see Arad 2001 for a recent discussion of similar issues within a constructionist approach): (14)

a.

Transitive, te/ic [EP DP,zz hp

BP,zz [ASpQ DP 2 zz [ypZZ]]]] NOM

ACC

zz,+ASPQ> +ACC, +PST, +E ~

b.

Ihezizl Transitive, atelic [EP DP, zz [TP BP, zz [FP DP 2 zz [yp zz]]]] NOM

ZZ, +PRT, +PST, +E ~

c.

PRT

Ihezizl

Intransitive, telic [EP DP, [TP BP, zz [ASpQ BP, ZZ [vp zz]]]] NOM

ZZ, +ASP Q , +PST, +E ~

d.

Izazl Intransitive, atelic [EP DP, zz [TP BP, zz [yp zz]]]] NOM

ZZ, +PST, +E ~

(IS)

a.

Izazl

Transitive, telic [TP BP, stt1' [ASpQ DP 2 stt1' [yp stt1']]]]

[EP DP, SRP

NOM

ACC

SRP,+ASP Q' +ACC, +PST, +E ~

b.

[EP DP, SRP hp

BP, SRP [FP DP 2 stt1' [yp stt1']]]] NOM

SRP, +PRT, +PST, +E ~

c.

Isarap/

Transitive, atelic PRT

Isarapl

Intransitive, telic [EP DP, [TP DP, SRP [ASpQ BP, stt1' [yp stt1']]]] NOM

SRP, +PST, +ASP Q +E ~

d.

Inisrapl

Intransitive, atelic [EP DP, SRP hp

BP, stt1' [yp stt1']]]] NOM

SRP, +PST, +E ~

Inisrapl

Scrutiny of the output forms in (14) and (IS) reveals immediately the inflectional features which the phonology does and does not choose to spell out. EP, we note, is con-

The Grammar Machine stant across all derivations, and hence its contribution cannot be evaluated. Tense (as well as agreement) are clearly distinctive. While the presence of ASPQ as such does not seem to result in a phonological distinctiveness, the availability of an objective case does. Thus transitive forms, regardless of whether telic or atelic, are phonological homophones, as are intransitive forms, regardless of their telicity. In turn, while the presence of objective case is always phonologically marked, that phonological marking need not be consistent across roots, realized as binyan V for zz, but as binyan I for SRP and binyan III for GLGL ('roll'). We may thus conclude that the relevant input to the phonological search consists of the representations in (16), containing the root itself, together with the phonologically relevant feature +OM (object marking), and whatever idiosyncratic morphological pattern is associated with the output for particular roots: (16)

a.

V(heziz) -7I(sarap) +PST (+E) -7 III (gilgel) ('roll', trans., telic, atelic)

ZZ,+OM, +PST (+E) -7 SRP,+OM, +PST (+E) GLGL, +OM,

etc.

b.

zz, (+ASP), +PST, (+E) -7I(zaz) SRP, (+ASP), +PST, (+E) -7 GLGL,(+ASP),+PST, (+E) -7

II(nisrap) VIIChitgalgel) ('roll', intrans., telic, atelic)

etc. '5 Let us highlight what is regular in (14)-(16) and what is idiosyncratic. Regular is the fact that almost without exception, the presence of the inflectional feature +OM in Hebrew is distinctively marked, although there is no one-to-one correspondence between +OM and a specific binyan. Nevertheless, the binyan system does have some inflectional properties that have no exceptions. Most strikingly, binyanim II, IV; VI, and VII never talce direct objects-that is to say, they never occur in the presence of a +OM feature. There are simply no vocabulary items in Hebrew which are an exception to this generalization. The +OM feature, thus, may only be realized as binyanim I, III, or V Yet, this is a fact that the child clearly does not know. Note specifically (naii, aiii), where the child uses binyanim II and VII transitively.'6 Further, binyan III is never used as an inchoative or a reciprocal in the adult language, yet the child uses it inchoatively in (9a) and as a reciprocal in (nbii). It seems safe to assume, then, that the child is oblivious to the information contained in column 2 of Table 11.1: she does not know of any syntactic regularities associated with particular binyanim, nor does she seem aware of any measure of relationship between the syntactic structure and the selection for the appropriate binyan. Finally, although the phonological distinctiveness of the '5 Quite possibly, the +ASP specification as well as the +E specification in the phonologization conventions in (r6b) are redundant, making the occurrence of the morphological forms in (r6b) the unmarked verbal instantiation of the roots in question. Wishing to leave this possibility open, subject ro further investigation, these features are placed in parenthesis. ,6 Passive binyanim IV and VI, uncommon in adult speech, are not attested in pre-school children altogether, clearly for independent reasons.

HagitBorer

306

+OM feature is almost without exception in the adult language, the child seems oblivious to that fact, too. In sum, children seem oblivious to the effect of syntactic feature marking on their choice of binyan, be it regular-from the perspective of the target use-or idiosyncratic. Yet, strikingly, all forms produced by the child are morpho-phonologically correct, in that they all represent morpho-phonologically-possible words and belong to wellformed binyanim. All attest to morpho-phonological knowledge of what are possible words in the adult grammar, but to the absence oflmowledge, on the part of the child, of the fact that the appropriate morpho-phonological output must be checked against the history of the syntactic derivation. This is not only true of the neutralized verbs in (8)-(u), attested in the adult language, and used by the child with the wrong valency. Novel forms which are not attested in the adult language, occur in the early speech as well, as illustrated by (17). And here, too, all forms are morpho-phonologically possible-although not attested-words.

(17)

a.

hu kines.II1 la-tanur. Na'ama 2;2 he . . . to-the. oven 'He entered the oven.' (c£ adult niknas.II (root KNs) 'entered')

b.

'ani me/z.ina.Y le-Dafna. Na'ama 2;2 I to-Dafna 'I explain to Dafna.' (lit. 'I understand to Dafna.') (cf. adult hisbir.V (root SBR) 'explain'; cf. adult he/z.in.V (root YBN) 'understand')

c.

'ani roca le-naheg.lII ba-fiesta Na'ama 2;3 I want to- . . . in-the. Fiesta 'I want to drive the Fiesta.' (cf. adult fi-nhog.l (root NHG) 'to-drive')

d.

kaka 'ani roca le-sa'er.III Na' ama 2;4 so I want to- ... 'I want to stay like this (?)' (cf. adult le-hiSa'er.II (root s'R) 'to-stay')

As in (8)-(n), all forms in (17) use an existing root, interpretationallyappropriate, in an existing, but wrong, binyan. For the adults, another binyan is used to convey that meaning with that root, or, as the case is for (17b), another root altogether is used (rather on a par with the eat/feed situation in English). All forms are morpho-phonologically correct, in that they belong to an existing binyan, but strongly suggest that the child could not possibly be projecting argument structure from an acquired token. As an illustration, the child has never been exposed to kines.II1, as in (17a), although it is likely that she has been exposed to the adult niknas.II, 'enter', sharing the same root. She has been successful in acquiring, from the input, the basic meaning of the root KNS, as pertaining to entry, but has failed to store in memory the particular morpho-

The Grammar Machine phonological binyan with which it is associated in the adult language. She is now proceeding to embed this root within a binyan which is morpho-phonologically correct, but which is not the one associated with the correct adult binyan. To the extent that the child is now projecting argument structure to go along with this creatively produced form, what could be its origin? It could not possibly be coming from the newly invented form, as that form has just been coined and does not have an argument structure in and of itself. It could not be emerging from the binyan used, as binyan III, used here, does not have a fixed argument structure associated with it and is compatible with both +OM and -OM. Finally, it could not be emerging from the meaning of the root, as roots, as such, are not associated with argument structure (or a particular event interpretation). All the roots in (17) occur in more than one binyan, with divergent meanings and argument structures. We are therefore driven to the conclusion that the argument structure associated with these newly invented forms must be available independently of the knowledge of vocabulary items, reflecting a computational knowledge of the structure of events that cannot be reduced to the acquisition of tokens. We note the particularly interesting case of (17 b), where the child is using an existing root in an existing binyan, but with an argument structure never attested, for adults, with this root. Yet, in the process the child is displaying a conceptual understanding of the meaning of the root, together with complete disregard for whatever grammatical context in which she may have heard the existing form previously. This behaviour thus casts serious doubt on the idea that lexical items are acquired from situations, together with their argument structure projection possibilities. To summarize, the child appears to have acquired some extremely important aspects of the morpho-phonology of Hebrew: she is successful in extracting roots from existing words, storing them with the relevant meaning and embedding them in morpho-phonologically correct templates, as the novel forms in (17) illustrate. She is further successful in projecting syntactic event structure to go along with its morphophonologically correct forms. However, the child is unsuccessful on two fronts. First, she has yet to learn that the particular binyan associated with a particular root is not just subject to morpho-phonological constraints, but also to morpho-syntactic ones, and that the appropriate binyan must be searched in accordance with the syntactically determined +OM feature. And second, paying no attention to the syntactic conditioning of the binyan system, she has also failed to acquire whatever measure of morphosyntactic regularities are associated with it.I?

'7 We note in this context that the child is not ignoring other components of the syntactic structure and the effect that they have on the morpho-phonological output, inflecting forms correctly for both tense and agreement. The problem here is associated with the choice between different binyanim, conditioned as it is by the +OM feature together with idiosyncratic, root-specific knowledge. The source of the delay here may be either the idiosyncratic, root-specific knowledge required, or, alternatively, a specific difficulty with marking event structure through the morphology. The choice between these options cannot be made without a fuller investigation of the development of argument structure inflectional marking in other grammars, but see the conclusion for some relevant considerations.

Hagit Borer

11.3.3. On root extraction and morphological early knowledge A few comments are in order here on the child's assumed abilities. I assumed that the child has attained the morpho-phonological aspects of the binyan system by the morpho-phonological stage. That ability consists of two crucial elements: the child knows all existing morpho-phonologically possible templates (relevantly, binyanim I, II, III, V, and VII, excluding passives), and further, has acquired the ability to extract the root from input words and store it independently of the morpho-phonological environment in which it was acquired. Evidence of fullimowledge of all morpho-phonologically possible binyanim has been independently argued for by Berman (1982) and Levy (1988). Table 11.4 shows the distribution of roots across binyanim in two children, Ruti and Arnon. All forms are singletons, or underived, in the sense of Levy. 18 Table 11.5 gives an similar distribution for Na'ama (singleton and non-singleton roots); Table II.6 gives the distribution of verb tokens and types in adults. '9 With the possible exception ofbinyan II, the distribution of both tokens and types for children across different binyanim is virtually identical to that of adults. Even for binyan II, Na' ama has thirteen distinct types. The children under consideration, then, have an accurate knowledge of the different morpho-phonological templates available in Hebrew. There are virtually no cases of 'invented' binyanim in the early speech, that is, morpho-phonological forms used by the child that cannot be traced back to an existing binyan, possibly with some phonological errors especially in cases which are phonologically irregular. Consider now the ability attributed here to the child to extract the root out of a morpho-phonological template, and embed it within a different one. While this

" Where by 'singletons' (Levy's 'underived') we mean roots which occur exclusively in a single binyan, and by 'non-singletons' we mean roots which occur in more than one binyan. By 'type', we mean a given root in a specific binyan, that is, the root zz in binyan I is a type (zaz, 'move.1NTRANs') distinct from the same root zz in binyan V (heziz, 'move.TRANs'), but they count as one root. Thus for singletons, the number of roots and the number of types is the same. For non-singletons, the number of types exceeds the number of roots. Levy's classification is based on her contention that for children at the relevant stage, a given root only occurs in a single binyan (and see also Berman 1982), which is ro say, children only have singlerons. For adults, of course, a particular root may occur in more than one binyan. As it turns out, the conclusion that children at the relevant age only have singletons is in error. I return to this issue at length in section 11.4, where I evaluate the claim that morphological biases or deficiencies account for the early neutralization facts. 19 It would have been best, of course, to conduct an analysis on the adult input to Na' ama. Unfortunately, such a direct analysis is impossible as the adult tier in the CHILDES files is rarely included. Adult data here is from a sample text containing a rotal of 519 verbs (tokens), taken from the Hebrew newspaper JaQua israeli ('An Israeli Week') of 19 July 1999. It represents three articles by different authors, so does not actually characterize the language of anyone particular individual. As this is a written, rather than a spoken, sample, it probably represents a higher linguistic register. In view of that, the almost identical distribution among binyanim ro that of the children is particularly striking. Distribution of verbs in the passive binyanim, IV and VI, is included for completeness, as is the one occurrence in the stylistically elevated variant of VII, marked here as VII', nitpa'el. Participial forms (requiring an auxiliary in past and future tenses), including adjectival passives, are not included in Na'ama's sample and in the adult sample.

The Grammar Machine TABLE I I -4- Distribution of verb tokens and roots (singletons) across different binyanim, Ruti and Arnon (2;0-2;4) Arnon Binyan

II III V VII

Ruti

Tokens (total 400)

Roots (total 8 I)

N

%

23 0 20

57·5 5. 0 13·75 20.0

55 80 15

3·75

N

%

Tokens (total 1,757) N %

N

%

43 4 15 15 4

53. 0 5. 0 18·5 18·5 5. 0

1,062 65 28 3 286 61

40 7 25 22 6

4 0 .0 7. 0 5. 0 22.0 6.0

60·4 3. 6 16.0 16.0 3·4

Roots (total 100)

Source: Levy (1988)

TABLE I 1.5. Distribution of verb tokens and types across different binyanim, Na'ama (singletons and non-singletons) (1;7.8-2;6.4) (root total: 146) Binyan

Tokens (total 1,199)

Types (total 174)

N

%

N

%

62·9 2.2 13. 8 18.8 2·3

75 13 37 32 16

43. 1 7·5 21.2 18·4 9. 2

754

II III V VII

a

26 16 5 225 b

28

a Including 170 occurrences of raca ('want') b Including

73 occurrences of hebi ('give, bring')

TABLE II.6. Distribution of verb tokens and types across different binyanim, Adults (singletons and non-singletons) (root total: 217) Tokens (total 519) Binyan

II III IV V IV VII VII' (nitpacel)

N

%

29 1 48 60 2 82

56 .1 9. 2 Il.6 0·4 15. 8 0.6 6.1 0.2

32

Types (total 2 54) N % 75 33 36 2

43-1 13·0 14. 2 0.8

49

19·3 1.2 9. 8 0·4

25

HagitBorer

310

appears as a very abstract operation, requiring considerable computational sophistication, it is also dear that the ability to do so must be assumed to be available at an early age. The reason here has to do with properties of the tense system, rather than the binyan system. Just like the binyan system, tense morphology is expressed through a combination of vocalic melodies and affixation. These vary from binyan to binyan, as Table 11.7 shows (only phonologically regular third-person singular masculine forms are given; binyanim IV and VI, absent in early language, are omitted. Root is PKD, roughly 'count' or 'command'). TABLE 11.7. Regular tense inflection in the binyan system

Past Present Future

I

II

III

V

VI

PaQaD PoQeD yiPQoD

lIiPQaD lIiPQaD yiPaQeD

PiQeD mePaQeD yePaQeD

hiPQjD maPQjD yaPQjD

hitPaQeD mitPaQeD yitPaQeD

As is evident from Table II.7, the appropriate use of tense inflection already requires the ability to extract the root and to embed it, in different tenses, in distinct vocalicaffixal melodies, which have distinct tense value (past, present, etc.), together with the explicit knowledge that different vocalic-affixal melodies may be associated with the same tense value in different binyanim. Thus the vocalic-affixal melody for past for binyan I is a-a, while the vocalic-affixal melody for past for binyan V is hi-i, etc. Interestingly, binyan I, the one the children and adult use most (see above), is the one which displays the widest vocalic variation across tenses (a-a; o-e; yi-olyi-a). And yet, as is well-established, Hebrew learners master tense morphology extremely early. At age 1;9, during one session, Na'ama uses nineteen tokens with the tense marking set out in Table u.8-vowels and affixes used are correct in at least seventeen forms. At age 2;2, at which Na' ama begins to use neutralized forms, she uses, in one session, eleven verbs in more than one tense, with vocalic melody fully correct, ranging over. three distinct binyanim (I, III, and V). The ability, however acquired, to extract TABLE II .8. Distribution of tensed forms, Na'ama, 2;2 Binyan I Past Present Future Imperative Truncated infinitives Truncated, unclear Inflection error

Binyan V

Unclear

3 (I root) 4 (3 roots) 2 (2 roots) 2 (2 roots) 5 (4 roots) I I

(sigor)

(bet)

1he Grammar Machine

3U

the root from vocalic/affixal templates and to embed it correctly within another, sensitive both to tense considerations and binyan considerations, is beyond dispute. It is in place before the stage in which children exhibit the errors illustrated in (8)-(rr), the first clear instance of which occurs in the Na' ama corpus at age 2;2.20

11.3.4. Constructing argument structure in the morpho-phonological stage Granting that the child did master the principles which govern the morpho-phonology of the binyan system, including the ability to extract the root and to embed it in a morpho-phonologically well-formed binyan, the errors exhibited by the child could not be attributed to morpho-phonological shortcomings. However, the child is presented with an interesting dilemma once she attained the morpho-phonology of the binyan system without having acquired the knowledge that the form selected must be matched, specifically, against the presence of the +OM feature. Suppose for a moment that she already knows that pairs such as yaca.Ilhoci. V(,come-out/take-out') and hidbiq. Vlnidbaq.II ('glue/become-glued') have the same root and hence the same basic conceptual meaning. The child also knows already that these are all morphophonologically well-formed. In the absence of the knowledge that the full grammaticality of her choice is dependent on the presence vs. absence of the +OM feature, what could possibly be the difference between them, from her perspective? The answer, it appears, is none. To the extent that the child has become aware of the existence of par- . ticular morpho-phonological pairs derived from the same root, such pairs could only be construed as synonyms-two equally felicitous outputs of the morpho-phonological system with the same interpretation. In the absence of any sensitivity to the +OM marking, there is no way for the child to know, in a given syntactic environment, which of them to use. As a result, she is guessing. This, I claim, accounts for the cases of neutralization with attested forms, illustrated by (8)-(u). In turn, memory of attested morphophonological forms, whether attested in derivational pairs or not, occasionally fails altogether. In those cases, the child resorts to a productive use of the morpho-phonological system. Having acquired the root with its basic meaning, the child embeds it in some possible morpho-phonological binyan, which quite plausibly will deviate from that actually attested in the language. The novel forms in (I7) are the result. Consider what such a system might look like, so that it involves knowledge of the root and storage of some already attested root-binyan pairs, but no knowledge of their syntactic conditioning. In that system, argument structure could not come from the w Berman (I993, I994) suggests that at the earlier stage-the stage corresponding here with the morphophonological stage-children have acquired specific tokens as un analysed amalgams. It is not clear, however, how that can be, or how the facts could possibly support such a claim, given the productive use ofverbs in different tenses. The ability to extract the root must be assumed, given the competence children show at the relevant age in prodUcing tense morphology. Further, the fact that the children err on binyanim but not on roots also indicates an ability to separate the root from the morpho-phonological template. None of these is expected if the children treat specific tokens as unanalysed amalgams.

Hagit Borer knowledge of vocabulary. Even if children have learned many verbs as isolated tokens, with root and vocalic template together, this knowledge does not interact with the way in which they project argument structure. In fact, as is clear from (S)-(n), they proceed to ignore the syntactic environment in which they presumably learned of the existence of the relevant token, and instead project whatever argument structure suits the propositional content they have in mind, inserting into it a morpho-phonological form with an interpretationally appropriate root, but not necessarily the correct binyan from the perspective of the morpho-syntax. Consider now this behaviour from the perspective of a syntactic model in which the interpretation of argument structure is independent of verbs. Suppose the child knows the syntactic structures which are associated with arguments and their interpretation. Specifically, the child is using the following innate linking correlations to help her into the projection of a preliminary argument structure: (IS)

a. b. c.

originator subject-ofquantifiable change

¢:} ¢:}

[Spec, EP] [Spec, ASP Q1

EP dominates ASP Q

From the perspective of production, the linking regularities in (IS) tell the child (and the adult) that whenever they wish to express the existence of an originator of an event, they must embed it within the structure in (ISa), and that whenever they wish to express the existence of a subject-of-quantifiable change of an event, they must embed it within the structure in (ISb). From the perspective of comprehension, the child will likewise understand a DP occurring in the structures in (ISa-b) as originator and subjecl-of-quantifiable change, respectively, regardless of the particular verb used." As an illustration, consider again the root zz, associated with the meaning of movement. Making use of the linking principles in (IS), the child will be able to construct the structures in (19), with the accompanying interpretation. EP

sp~ DP, E ASPQP

originator

sp~ DP 2 ACC

subject of q-change

ASP Q

VP

I

V

zz

" Of course, comprehension is dependent here on syntactic and semantic transparency. Thus in the presence of two DPs and a non-stative event, the first must be an originator. The second, however, may either be a subject-of-quantifiable change or just a default participant. Similarly, for intransitives, a single DP may be either originator or subject-of-quantifiable change. A successful parse, then, is dependent on

7he Grammar Machine ::Ie (eventive, e) & originator (DP" e) & subject-of-quantifiable change (DP"e) & (zz,e) (DP, originated a ZZ (non-stative) event which affected DP2 in a quantifia-

ble way). ((2a)(4a))

b.

EP spec DP,

E

originator

FP ~

spec DP 2 PRT

F

default*

VP

I

V

zz ::Ie (eventive, e) & originator (DP" e) & participant (DP" e) & (zz, e) (DP, originated a ZZ (non-stative) event. DP 2 was a (relevant, non-origina-

tor) participant) ((2b)(4b))

c.

EP

sp~ DP E ASP

Q

sp~ £?P

subject of q-change

ASP

Q

VP

I

V

zz ::Ie (eventive, e) & subject-of-quantifiable change (DP, e) & (zz, e) (DP

underwent a quantifiable change in a (non-stative) zz event) ((2C)(4c))

the degree to which the type of event involved is obvious from the communication act. The child, however, is not alone here in attempting to process a structurally ambiguous input on the basis of pragmatic clues, nor is event structure the only area in which structural ambiguities cannot be resolved without context. A similar problem is faced by adults here, as well as in many other well-discussed cases of structural ambiguity.

HagitBorer

d.

EP

sp~ DP E originator

VP

I

v

zz 3e (eventive, e) & originator (DP, e) & (zz, e) (DP originated a zz (non-stative) event) ((2d)(4d» The child now needs to phonologize, so to speak, the root, given that a string of consonants-even if related to some meaning--is not a phonological option. If at all retrievable, the child will assign to the root in (19a-d) a morpho-phonological token already attested with that particular root. As most roots appear to occur in the (adult) input only in one binyan (see section 11.4), the child will be largely correct. However, when she is confronted with roots attested in more than one binyan, or when she has forgotten the attested binyan, she will randomize, phonologizing the root in any of the binyanim attested with the root, or alternatively, embedding it within an unattested one. Consider an attested example. A particular root, say YC'" pertaining, in essence, to 'exit', is quite commonly attested in adult speech both as binyan V (transitive, meaning 'bring-out', 'take-out', in both a telic and an atelic sense) and as binyan I (intransitive, used in telic sense, meaning 'come-out', and in an atelic sense with the meaning 'go-out', for instance, 'for a walk'). We expect the child to phonologize the root yc'in more than one way, and in a manner quite oblivious to the argument structure configurations and the interpretations associated with the structures in (19). In short, we expect the child to randomize, mapping either binyan-root pair onto any of the structures in (19), although yaca.I is only appropriate for (19C-d), while hoci. V is only appropriate for (19a-b). The data bears out these predictions, as (20)-(21) illustrate: (20)

a.

b.

(21)

a.

'ima, taCazri li, ze 10 yoce.! 10. (19 c) 22 Na'ama 2;2. mommy, help to.me, it no come.outto.it 'Mommy, help me, it is not coming out.' caxsav 'ani raca la-cet.! cim 'imale. (19d) Na'ama 2;3 now I want to-go. out with mommy 'Now I want to go out with Mommy.' ken hu mociY lehad. (neutralized form) (19c) yes he alone lit: take out; intended meaning: come out 'It comes out by itself, too.'

Na'ama, 2;3

22 AI; an aside, we note that the child has not yet acquired the correct use of the reflexive dative, ungrammatical for adults in the context in (20a).

The Grammar Machine b.

hociY 'oto me-ha-bor. (19a) took-out it from-the-hole

Na'ama,2;3

'(He) got it out of the hole.' Apart from negation, the subject-verb and event interpretations of (20a) and (2Ja) are virtually the same, and the same root is used. Still, Na' ama uses two different binyanim here to express the same event. Further, when using the neutralized (adult-incorrect) form in (2Ja), she has already used both yaca.! ('come out') and hoci. V('take out') correctly, in the very same session. And in a particularly striking example of random use, consider the following sequence, in which hoci. Vform is used twice, once transitively and correctly, and immediately following it, intransitively, and incorrectly:

(22)

NAA: toci'iY

,et ze. take.out.2SGF. OM it. NAA: ken hu moci. V [=yoce.!] * lehad yes he (comes. out) by.itself 'It comes out by itself, too.' (*adult tier missing; comment in square brackets from original CHILDES transcript)

More than anything else, the case in (22) looks like a recency effect. The child has just used the root yc'in binyan V; and this use is quite straightforwardly repeated. And yet, such recency effects are not attested for tensed verbs: the child is never tempted to repeat the past tense with a future interpretation, just because it has been recently used, nor does she use the wrong root. That recency does, in fact, have an effect here is thus direct evidence for the fact that the root-binyan pairs do not register, with the child, as having differing values appropriate in some, but not other, syntactic environments. To conclude, neutralization errors stem from the fact that the child has extracted the root and stored it with its basic meaning, and possibly with already attested morpho-phonological forms. The child has further acquired the morpho-phonological aspects of the binyan system, pretty much in full. However, in the absence of the ability to select the correct binyan based on syntactic factors, types which for the adult are only distinguishable as based on syntactic factors are treated as synonyms by the child, and the choice between them becomes random.

11.4.

MORPHOLOGICAL DEFICITS?

MORPHOLOGICAL PREFERENCES?

As shown in Tables 11.4-6, it cannot be assumed that the early performance is driven by a morpho-phonological deficit which leads the child to avoid some binyanim and replace them by others. Children have no morpho-phonological problems with any binyanim as such (passive binyanim excluded), nor is their distribution of roots across

HagitBorer different binyanim substantially different from that of adults, as Table II.6 illustrates. Even roots in binyan II, occurring in the early speech only abour half as often as in adult speech, take part in neutralization in both directions. Thus binyan II, avoided in (lOa-b) is the very same one neutralized in (naii). Nevertheless, the child may be experiencing a variety of morphological difficulties of a more complex sort. Berman (1982, 1994) makes two claims as concerning the early behaviour. First, she suggests that the child rarely has roots attested in more than one binyan. Neutralization, then, could be simply the result of the child acquiring a single binyan for any particular root, and sticking to that binyan for all occurrences of that root. The difficulty, then, is not in recognizing morpho-phonologically distinct forms, but in mapping between them. Children are, indeed, predicted to have no problems with singletons, but to have problems with non-singletons. Yet another claim concerning morphological difficulties is made by Berman (1993, 1994). Berman suggests that the child neutralizes overwhelmingly towards binyan I, precisely because the child does not yet know how to match particular binyanim with particular syntactic contexts. As binyan I is a catch-all template without any canonical morpho-syntactic properties, the child gravitates towards it, thereby avoiding using binyanim with restrictions which she does not yet fully comprehend. Obviously, both claims can only be substantiated if it turns out that children have non-singletons significantly less than adults, and that their attested preference for binyan I, likewise, exceeds that of adults. We have already shown that the early distribution of roots across the binyanim does not, in fact, differ significantly between children and adults. Importantly, while Table 11.4 only contains singletons (underived forms, in Levy's terms), which is to say, roots that only occur in a single binyan, Table II.5, from Na' ama, includes both singletons and non-singletons (where by non-singletons we mean cases of roots attested in the corpus in more than one binyan). Yet, the distribution of forms when both singletons and non-singletons are included remains the same; there is no significant increase in the occurrence of anyone particular binyan as a result of the inclusion of non-singletons. This is not only true for Na'ama, but also for adults. Thus, when the distribution across different binyanim for adults or for Na' ama is compared to that for Ruti and Arnon, we find that the occurrence of root-binyan pairings in different binyanim is by and large the same with binyan II, showing a marked increase for adults, being the only possible exception. At first sight, then, the inclusion of non-singleton root-binyan forms does not alter the proportion of forms attested in each binyan. Even if children tend to favour one binyan for each root occurrence, this is a tendency, rather than an exceptionless situation. We already saw that Na' ama uses roots in more thaq one binyan, including YC' (I, V, 'come out/ take our'), 'KL (I, V, 'eat/feed'), and others. A closer scrutiny of non-singletons in Na'ama's speech (including erroneous forms, both neutralized and novel), as compared to non-singletons in an adult corpus, further reveals that there is no difference here between the adults and children, disproving the claim that children (tend to) have singletons and that they favour binyan I. A count of non-singletons pairs-that is, types which share a root but are inflected in a

The Grammar Machine different binyan, such as zaz-heziz-in the speech of Na'ama reveals that non-singleton pairs occur in 27 out of the 147 roots that she uses. Thatis, 18.3 per cent of the roots she uses occur in more than one binyan. Table 11.9 gives the distribution, by binyan, of non-singleton types, including neutralized forms and novel forms. In total, there were 55 non-singleton types in Na' ama's speech, to a total of 31.6 per cent of singleton and non-singleton types attested (where a non-singleton type is a rootI-binyani form with a corresponding, attested, rootI-binyan j form). TABLE I I.9. Distribution of non-singleton types by Binyan, Na'ama (total non-singleton roots: 18.3%)

Binyan Distribution

15 (27.3%)

II

III

7 (12·7%)

9 (16.4%)

v

VII 8 (14.5 %

)

Table II.IO sets out the distribution of erroneous types only (singletons and non-singletons), including neutralized forms and non-existing innovations. In total, there were eighteen erroneous types, 10.3 per cent of total types, or 30 per cent of all 'creative' forms-that is, correct non-singletons, neutralized, and non-existent ones (a total of 60, or 34.5 per cent). Table 11.10 should be read as follows: six of the erroneous forms used by Na'ama, or 33.3 per cent of total erroneous forms, were in binyan I. One, or 5.6 per cent, were in binyan II, etc. TABLE I I. 10. Distribution of erroneous types by Binyan, Na' ama

Binyan

I

II

v

III

Na'ama

VII

5 (27. 8%)

Consider now the distribution of non-singletons in the speech of adults. In the adult sample, of a total of 217 roots, 32 were non-singletons-that is, occurred in more than binyan-or 14.7 per cent of total roots, ironically, less than the number of non-singleton roots in Na' ama's sample. In Table II.II the distribution is presented, by binyan, of non-singleton types for adults. In total there were 65 non-singleton types, or 25.6 per cent of total types (254). TABLE

1 I. I I.

Distribution of non-singleton types by Binyan, Adults (total non-singleton

roots: 14.7%) Binyan Distribution

22 (33. 8%)

II

III

IV

V

20 (30.8%)

5 (7·7%)

I

II

(1. 5%)

(16.9 %

)

VI

VII

VII'

I

4 (6.2%)

(1.5%)

(1. 5%)

I

Hagit Borer As is evident from the comparison of the adult corpus to the performance ofNa' ama, there is, in the early speech, no preference for singletons, which distinguishes it from adult performance. The percentage of non-singleton roots, as well as the total of nonsingleton types, is in fact higher in the child's speech than in the adult corpus. Furthermore, preference for binyan I is attested neither for non-singletons nor for neutralized or novel forms. Binyan I makes one member of a non-singleton root group 27.3 per cent of the times for Na' ama (Table 11.9), to 33.8 per cent in adult speech (Table n.n), and its occurrence in neutralized and non-existent forms is virtually the same as its occurrence as a member of a non-singleton root group or in non-singletons for adults, at 33.3 per cent (Table n.IO). For Na'ama, the distribution across binyanim is largely the same for the entire corpus (non-singleton types only) and for erroneous types. If anything, this shows a slight decrease in use ofbinyan I in non-singleton and erroneous types as compared with the entire corpus. The only marked difference from adult behaviour involves binyan II, which for the adults makes up 30 per cent of non-singleton types, but only 12 per cent of non-singleton types for Na' ama. For adults, the distribution of binyan II in non-singleton types far exceeds its distribution in the adult corpus in general (at 13 per cent; see Table n.6), but not so much so for Na'ama, supporting directly the claim that Na' ama's choice ofbinyan is governed by whatever factors determine the statistical frequency of that binyan in the corpus at large, and not on her knowledge of the syntactic restrictions that may restrict the occurrence of that binyan in the context of a specific root. We conclude, then, that contra Berman (1993, 1994), the pattern of errors in the early grammar cannot be attributed either to a preference for singletons, or to a principled favouring ofbinyan I. Rather, as we claimed in the previous section, morpho-phonological knowledge at the relevant stage of development seems virtually identical to that of adults, and the reason for the early errors must be sought elsewhere, in a deficient morpho-syntactic component-that is, in the early inability to relate particular morpho-phonological forms to the syntactic structure in which they are inserted. 23 It might be worthwhile to consider at this point what would need to be assumed, for the early grammar, by approaches which project argument structure directly from traditional lexical entries. Considering, specifically, the multiple occurrences of the root YC', pertaining to exit, illustrated in (20)-(21) one would have to assume, on the part of the child, at least the following lexical entries: (23)

a.

YC: VI theme JJ, direct internal

come out illustrated in Na'ama's speech by (2Ia).

2) Higher numbers for neutralized forms in binyan I are reported in Berman (1994), where binyan I is reportedly used in 67 per cent of neutralized cases (although for individual children it is at times as low as 35 per cent). Berman only counts neutralization, and although other unconventional cases are considered, their distribution across binyanim is not given. It is also not clear whether the 67 per cent result is of tokens or of types. If the former is the case, it continues to reflect, quite closely, the overall ratio ofbinyan I tokens in both adult and early speech.

The Grammar Machine b.

YCT2 causer, theme

V, c.

YC~Ir

take out

V,

external direct internal

illustrated in Na'ama's speech by (21b).

theme

come-out

V, d.

yc~h

direct internal

illustrated in Na'ama's speech by (2Oa)

agent

go out

V, external

illustrated in Na'ama's speech by (2ob)

As telicity is not morphologically marked by the binyan system, YC',1 in Hebrew could be used in both telic and atelic intransitive contexts, making it a so-called 'variablebehaviour verb' (see section II.r). We noted earlier the theoretical problems which are presented by variable-behaviour verbs for models which project argument structure from information stored in lexical entries. This issue, however, is theoretically problematic quite independently of the issue under discussion here, as the child attempting to acquire variable-behaviour verbs need not make any assumptions here that would lead her to deviate considerably from the properties of the target grammar. We note, considerably more significantly, however, that the entries in (23) include homophones with a distinct transitivity value (see (23a) and (23b», as well as synonyms with distinctive morpho-phonological representations ((23a) and (23c». All these are derived with the same root, a situation never attested in adult Hebrew. Another puzzling thing is the emergence of the lexical entry in (23a), which is non-existent in the adult grammar. How could it have come to exist in the early grammar? Clearly not through positive evidence. Nor is the singleton route helpful here. Not only have we shown that this generalization does not in general hold for the early grammar, it most certainly does not hold for this particular root and for this particular child. The data in (20) and (21) show this: Na'ama plainly has both YC',1 and YC'.v. The only possibility for a lexicalist account, then, would be that YC'.V with its intransitive interpretation in (23a) came into existence through a wrongly formulated rule of word formation, essentiallyas in (24a) or alternatively, (24b). Rule (24a) would derive an intransitive form in binyan V from an intransitive form in binyan I, with an identical meaning and argument structure. Rule (24b) would derive an intransitive form in binyan V from a transitive form in binyan V. Neither rule represents a sub-regularity occurring in the adult grammar: (24)

a. b.

root.I (theme) =} root.Y (theme) (e.g., yaca.INTRANS =} hod.INTRANS 'come out.INTRANS =} come out.INTRANS') root.Y (causer, theme) =} root.V (theme) (e.g., hod. TRANS =} hoci.INTRANS 'take out.TRANS =} come out.INTRANS')

Rule (24b) would further require the child to assume the equivalent of0-affixation (or conversion) for valence decrease, otherwise not attested in the Hebrew binyan system.

HagitBorer

320

A second rule of 0-affixation, this time for valency increase, would have to be postulated to account for the derivations in (Sa-b). This rule, in essence as in (25a), would allow a transitive binyan I to be derived from an intransitive binyan 1. Rule (25b), on the other hand, would 0-derive valency decrease, as in (IOa). The child must now be hypothesized to have two rules of 0-affixation with a diametrically-opposed effect, none of which is attested in the input: 24 (25)

a.

b.

root.! (theme) ::::} root.! (causer, theme) (e.g., zaZ.INTRANS::::} zaz.trans 'move.INTRANS move. TRANS') root.! (causer, theme) ::::} root.! (theme) (e.g., sajl.ak.TRANs::::} sajl.ak.INTRANS 'spill. TRANS ::::} spill.INTRANS')

While (25) would require the child to assume 0-affixation with diametrically opposed effects, (24d) would require it to assume vacuous morphological affixation, mapping binyan I to binyan V without any modification in argument structure-again a situation otherwise not attested in the Hebrew binyan system. To make matters worse, once the mapping operations in (24a-b) and (25a-b) are assumed to exist in the early grammar, one must address the question of why the child abandons them, especially in the absence of negative evidence which would be required to exclude them. And finally, the child must also be assumed to have the operation in (26), deriving transitive yc:V 'take out' from intransitive yc:I, to give rise to the two correct forms, both already present in the early language, as attested by (2Ib): (26)

root.! (theme) ::::} root.V (causer, theme) (e.g., yaca.INTRANS ::::} hoci.TRANS 'come out.INTRANS take out.TRANS')

In short, assuming projection from a lexical entry requires the child to postulate four distinct mapping operations, of which only one is attested systematically in the adult language. Alongside the common correlation between binyan I and binyan V which involves overt affixation together with valence increase, the child must also assume affixation with no valence effect, as well as two rules of 0-affixation with valence increase and valence decrease respectively-three mapping operations not available in the adult system. We conclude by noting that in principle, Semantic Bootstrapping is inherently conservative in its assumptions about the relations between the acquisition of verbs and the projection of arguments. Assuming that argument structure canonically projects from the lexical entry once the verb and its lexical semantics have been appropriately acquired, Semantic Bootstrapping predicts that errors made by the language learner are restricted either to cases of wrongly acquired lexical semantics-as associated with a particular phonological representation-or, alternatively, to overgeneration-that '4 Interestingly, in the one case in which 0-derivation valency change might be involved, as in (15b.5), the relevant verbs in binyan V are always derived from adjectives, and never from verbs in any other binyan. See Borer (1991) for further arguments that each of the members of the inchoative-causative pair is independently derived from the source adjective, and that the forms are not directly derivationally related through 0-affixation or otherwise.

The Grammar Machine

321

is, the rule-governed use of possible, but non-existing, extensions (for example, a transitive, non-attested, use of an intransitive verb, on a par with the existing move-move, drop-drop alternation). The neutralization and novel-use errors here involve correct lexical semantics (of the root), together with the absence, in the target language, of any systematic rule-governed correlations, excluding the possibility that they are the result of overgeneration. Still, a mismatch occurs between the verb used and the argument structure. This is, in principle, a type of error which Semantic Bootstrapping predicts not to occur, and yet it does-a strong case against the association of argument structure with the lexico-semantic properties of vocabulary items. 25

I1.5. FIXING IMPERFECTIONS: THE MORPHO-SYNTAX STAGE A1; has been established by Bowerman (1982) and Clark (1982), children learning Eng-

lish pass through a stage in which they produce forms such as those in (27) and (28).

a. b.

It always sweats me. This is aching my legs.

(Bowerman 1982)

'5 Before turning to the next developmental stage, it is worthwhile to consider briefly the possibility that the child is suffering from a very specific type of computational overload problem. To illustrate, suppose the child has acquired both yara.I, as an intransitive, and hoci. Vas a transitive. However, when faced with retrieving the particular morpho-phonological form associated with each of these entries, she suffers a computational overload and resorts to guessing. That the child actually uses entirely novel forms, as in (17), based on roots which otherwise are available in the adult language as singletons with morpho-syntactic properties identical to the novel form, suggests that computational overload cannot possibly be the problem here. Specifically, the root nhg, having to do with driving, occurs as a singleton in the adult language in binyan I with the relevant meaning, 'drive'. The child coining nhg.III, as in (I7C), is associating with it exactly the same argument structure that is associated, for adults, with the correct nhg.I. Any attempt to make sense of this behaviour must allow the child to store roots with their basic meaning but without binyan specification, and to have independent access to the inventory of attested binyanim, allowing her to join the stored root nhgwith its stored interpretationwith binyan III morpho-phonology. Roots, however, cannot be assumed to be srored with information about argument structure, as they are completely underdetermined with respect to such information. Proponents of a lexical approach to the solution of the neutralization/novel binyan problem would thus have to explain why the retrieval of the root and the syntactic projection information, for a specific singleton entry, presents no problem, while at the same time the retrieval of the binyan used by the adults-the only binyan in which the child ever heard the token under consideration-is problematic. We note that even for non-singletons, computational overload must presuppose the psychological reality of the root, or forms such as root,-binyani and root,-binyanj would not be part of a comparison set to begin with. We must also presuppose that the child would not be tempted to use the same root in a different binyan, rather than a different root-binyan combination altogether. In turn, the psychological reality of the root and its knowledge by the child, together with the child's perception of the relatedness of nonsingleton pairs, entails computational knowledge of the morpho-phonological system. That the knowledge of roots and templates results in phonologically and syntactically correct input for tense inflection, but for phonologically correct and syntactically wrong input for binyan inflection, further suggests that simple morphological complexity would not do, and that the deficit here is more specific.

Hagit Borer

322 (28)

a. b.

I broomed her. (2;7) Mommy trousered me.

(2;3)

(Clark 1982)

Can the existence of cases such as (27) and (28) in early English be correlated with the type of neutralization errors made by the Hebrew-learning child? In order to investigate this question, let us consider briefly the nature of the 'errors' in (27) and (28). As is well known, in adult English the very same stem often occurs as a transitive verb and as an intransitive verb. Further, in adult English the very same (underived) stem may occur as a verb and as a noun, giving rise, in effect, to triplets such as movev/ move v,! moveN; walkv/wa1kv,!walkN; sinkv/sinkvJsink N; racev/raceV/raceN' In Borer (2000) I argue that this picture, in English, is a direct result of the category-neutral, unspecified nature of (un derived) substantive vocabulary items in English, and the fact that category and event structure are determined exclusively by the syntactic structure and are not reflected in any way in the phonology of (underived) stems. For the English learner, then, the marginality of sweatVt and achev or broomv and trouserv constitutes, in actuality, an inexplicable exception. What the child needs to learn here is that a general process-allowing the embedding of (underived) stems as the heads ofNP or VP (and realized as nouns or verbs respectively), and under any event structure-sometimes gives an 'unconventional' output. Consider now the picture in adult Hebrew. The equivalent of the English category-neutral stem in Hebrew is a consonantal root-by itself phonologically unpronounceable. The pronounceability of the root, in turn, depends on vocalic-affixal information provided through the syntactic structure-specifically, through the binyan system and the tense/agreement system, however structurally represented. Thus every phonologically well-formed word in Hebrew, but not in English, corresponds to a unique syntactic structure. To illustrate, English move, itself without category and devoid of information concerning event structure, may be inserted in nominal structure, in verbal transitive structure, or in verbal intransitive structure. On the other hand, a Hebrew word such as heziz, containing not only the consonantal root ZZ, pertaining to motion, but also the vocalic-affixal melody hi-i, is already a conglomerate of a category-neutral root, the structural information provided by +OM (realized here as binyan V) and by the structural information associated with past tense (determining here specifically the quality of the first vowel, see Table II.7). Thus heziz entails the projection of ASP Q (or FP) with the accompanying +OM feature, and ofTP. In turn, as heziz must be dominated by ASP Q (or FP) and by TP, it becomes categorized as a verb. Neutralization errors by the Hebrew learner do not constitute a failure to learn an exception associated with unconventional vocabulary items, but rather, a failure to have acquired a grammatical process-namely, that which links the vocalic-affixal melody to aspects of the grammatical event structure (and specifically, to the presence or the absence of the +OM feature). To the extent that Hebrew learners act like English learners, they do so in assuming, erroneously, that vocalic-affixal melody of Hebrew verbs-while necessarily reflecting the presence of a TP-is not sensitive to event structure in general and to the presence vs. absence of the +OM feature, in particular.

lhe Grammar Machine Suppose, then, that this is the case, and that learners of English, too, go through a stage in which they do not recognize the ability of event structure to affect the (morpho-)phonology of words. In Hebrew, neutralization errors emerge at this stage. In English, on the other hand, little would go wrong beyond the type of overgeneration illustrated by (27) and (28). This is simply because the task of the English learner is greatly simplified by the fact that in English, neither category nor event structure need affect the phonology of the word. The child's initial assumption, that the phonology is insensitive to event structure, works for English, although it is not correct universally; nor is it correct for Hebrew. We note here that the ungrammaticality of (27) and (28), if indeed they are ungrammatical, is of a different class than that of neutralization cases in Hebrew. While sweaty, and achey or broomv and trouserv do not occur in standard English, they are possible English words in the contexts in (27) and (28). Not so neutralization cases in Hebrew. These are not possible words. AB the earlier developmental stage just happens to correspond to the state of affairs in adult English, little further learning is required on the part of the English-learning child beyond that which is involved in the conventionalization of the vocabulary (and see section II.G). Hebrew speakers, on the other hand, have an extra task; they need to acquire the information that event structure does have a morpho-phonological effect on words. Let us turn, now, to this extra task, considering the next stage in the acquisition of Hebrew-a stage that we will refer to as the 'morpho-syntactic stage'. I suggest that at this stage the child not only becomes fully aware of the significance of event structure in determining the morphology of the verb, she is now also applying her newly found knowledge with a vengeance. This gives rise to frequent overgeneralizations and to the by now well-known U-shaped curve: 'exceptional' forms are now systematically replaced with 'regularized' ones, giving the appearance of a declined performance. Neutralization errors are attested, among learners of Hebrew, roughly between ages 2;2 and 2;II, at which point they stop. Neutralization errors are now replaced by errors of the type illustrated in (29) and (30).26 (29)

a.

b.

c.

lama 'at madxip.a.V'oti kaka. (3;2) (earliest attested occurrence) why you push me so (cf. adult doxej!.et.I) 'ani yodaCat leb.ad le-haxlic.V' et ha-nacalayim (3;3) I know alone to-tal{e.off OM the-shoes (c£ adult la-xloc.I) ze mamas masrip..V 'oti, ha-semd. (4;?) it really burns me the-sun cf. adult sorej!..I

26 Data and generalization in this section are all based on the original observations and conclusions reached in Berman (1982) and Berman and Sagi (1981). While the pervasive 'creativeness' on the part of children at the relevant age is amply demonstrated in these studies, quantitative data on overgeneration is not available. A developmental study of Na'ama, as compared to her earlier stage, is not possible, as the Na'ama corpus ends at age 2;6.4. For this reason the conclusions reached in this section remain tentative and must be subject to quantitative verification.

HagitBorer (30)

a.

b.

'uf, 'at kimCat hielalt.V 'oti. (4;7) expo you almost made-dive me (novel; derived from adult ealal.LINTRANs 'dive') ba-yam ' aba masxe.V ' oti. in-the-sea daddy make-swim me (novel; derived from adult saxa.LINTRANS 'swim')

The examples in (29) and (30) are of two kinds. In (29) the child replaces an existing word that occurs in the adult grammar in binyan I with a form derived from the same root, but in binyan V. In (30) the child is coining a novel causative form in binyan V, which is related to an existing intransitive form in binyan 1. The adult vocabulary does not contain a binyan V realization of these roots. The cases in (29) in (30) involve a 'creative' use ofbinyan V, with novel binyan V forms derived from binyan I (30), or with the replacement of existing binyan I forms with binyan V forms (29). Replacements ofbinyanim other than I and innovations in binyanim other than V, occur as well, as illustrated by (31) and (32).

(31)

a.

b.

(32)

a.

b.

ha-qosem cilem.III 'et ha-saj2.an. the-magician disappeared OM the-rabbit (c£ adult Melim.V) calJ;sav'at masmixa.V'oti. now you cheer-up me (cf. adult mesamaxat.III)

(4;5)

lama ze ha-kol nif!-raq.II. (3;n) why it the-all fell-apart 'Why did it all fall apart?' (cf. adult hitpareq.VII) Cani nora mit'aleIz..VII kse-'omrim li kalea. (5;2) I very offended when-saY:PL to-me that 'I become very offended when someone says that to me.' cf. adult nlelaIz..II

How is the pattern of errors exhibited by (29)-(32) different from the pattern of neutralization errors attested in younger children? Crucially, in all the 'errors' made in (29)-(32), the children respect the canonical morpho-syntactic properties of the different binyanim which they are using (cf. Table n.I). Recall, specifically, that binyan I does not have any canonical properties, binyanim II and VII are always intransitiveboth often corresponding to inchoative and middle interpretation-and binyanim III . and V are dominantly transitive, often with a causative interpretation. If we assume that at this stage, the children take the syntactic environment in which a verb form is inserted very seriously, quite possibly more seriously than adults, then the pattern of errors is readily explained. Binyan I was previously favoured, being the most common morpho-phonological template in the language. But now the children are avoiding it,

1he Grammar Machine

325

precisely because it has no well-defined morpho-syntactic properties. They replace it with those binyanim which best correspond to the particular syntactic structure which they wish to express: binyanim III and V for transitive configurations, binyanim II and VII for intransitives. Of that particular stage, Berman (1982) and Berman and Sagi (1981) make the specific claim that while children often confuse binyanim III and V, on the one hand, and binyanim II and VII on the other hand, mistakes involving the replacement ofbinyanim III/V by binyanim II/VII never occur, resulting in the generalization in (33). (Glossed forms in (33) are adult forms. Forms marked with are not attested in the sample, regardless of whether or not they are adult-correct. ¢::} indicates inter-changeability; ~ * ---? indicates absence ofinterchangeability.)

*

(33)

Root. III

¢::}

"item simeax

Root.Y

~

* ---?

Root.II

¢::}

hecelim

nlelam*

'disappear.TRANs'

'disappear. TRANS'

hismiax

nismax*

hiJzriq*

nij2.raq

Root.VII

hitcalem* histameax*

'make-happy'

pereq* 'take-apart'

cileH

hitpareq fall-apart

heceliJzj

necela/z.

'insult'

'be-offended'

mitCale/z.

What can account for this particular pattern of behaviour? Recall that we assumed a model of vocabulary insertion for adults, according to which the particular selection of a vocabulary item for a particular root is dependent both on the presence vs. absence of the +OM feature and on a list matching up the realization of the +OM featute with specific binyanim for specific roots. Such a list is made available at some point by the end of the syntactic derivation, after the relevant root becomes associated with the relevant structural properties. For the adults, both are needed: while the presence vs. absence of the +OM feature delimits the search, it does not fully predict the morpho-phonological appropriate form. It is an arbitrary fact of Hebrew vocabulary that the relation between burn.TRANs and burn.INTRANS is expressed as saraj2..I.TRANs and nisraj2..II.INTRANS; the relation between drown. TRANS and drown.INTRANs is expressed as ta/z.ac.I.INTRANS and hit/z.id.Y.TRANS; and the relation between grow. TRANS and grow.INTRANS is expressed as gidel.III.TRANs and gadal.l.INTRANs (see Table 11.3). While the syntactic structure would exclude some possible morpho-phonological binyanim from consideration (that is, binyanim II and VII, which are always intransitive, would never occur as causatives; binyan I is never the causative ofbinyan V, etc.), the presence or the absence of the +OM feature still leaves the final morpho-phonology rather underdetermined and subject to root-specific listed information. But suppose our child has now moved from the earlier stage, the morpho-phonological stage, to a new stage. From the morpho-phonological stage she comes equipped with the knowledge of roots and with the knowledge of the set of morpho-phonological well-formed outputs in the language. The child also appeared to be equipped

HagitBorer with some statistical guidelines concerning which of these morpho-phonological outputs is more common, leading her to favour binyan I, the most common in the vocabulary of the language and in its input. What the child could not do at that earlier stage were two things: she was not aware of the fact that the choice of a particular morpho-phonological form is conditioned by aspects of the syntactic event structure, and she did not lmow that some binyanim are absolutely excluded in some syntactic event structures. Without that lmowledge, all entries related to the same root were on an equal footing, and were equally appropriate in any syntactic contexts. Further, when memory of the particular binyan associated with a particular root failed, the child simply guessed, embedding that root in some possible binyan, giving rise to an existing or a non-existing form. Suppose that now the child has learned that the particular morpho-phonological form to be used is dependent on the syntactic event structure, and specifically, on the presence vs. absence of the +OM feature. Presumably the child will still avail herself largely of remembered tokens, now to be paired with the relevant syntactic features accumulated throughout the syntactic derivation. But as before, memory will often fail her!7 When that happens, she again must let the grammar guide her. Previously, equipped with morpho-phonological knowledge alone, the child simply chose an appropriate morpho-phonological output as based on its statistical distribution in the language. Now, equipped with morpho-syntactic knowledge as well, she lets that lmowledge guide her in selecting the right forms: binyanim II and VII for intransitive structures, binyanim III and V for transitive structures. Binyan I, so statistically dominant in the language otherwise, is now a liability. Having no morpho-syntactic properties, it provides the child with no guidance whatsoever, and is to be avoided in all productive operations. In short, just as in the morpho-phonological stage, in the morpho-syntactic stage the child could not possibly be projecting argument structure based on information associated with vocabulary items. In fact, the child often bypasses the vocabulary list altogether by associating morphology directly with the syntactic projection of arguments, and to a much larger degree than adults do. The result is the appearance of 'fixing the adult mess', regularizing, incorrectly, the correlation between the selection of the binyan and the syntax of event structure. In an approach to event structure such as the one outlined here, it is thus plausible to assume that at this stage, the child often does not merely check the appropriateness of a particular vocabulary item against the syntactic structure, as we assumed for the adults, but rather, allows the structure to directly determine the particular form used. Specifically, I propose that at the morpho-syntactic stage, the child is projecting exactly the same structures as she did in the morpho-phonological stage. She continues to assume the mapping in (18), repeated 27 In the absence of information on frequency of errors, the frequency of such memory failures cannot be evaluated. Clearly, however, it occurs frequently enough to differentiate the early performance from the adult one. Assuming, without direct evidence, that the child continues to use standard vocabulary, by and large, we maintain that the primary pool for vocabulary must remain that which is available through memory, or the child would be predicted to avoid just about all binyan-I forms.

The Grammar Machine here as (34), and the structures in (19). However, she now adds the morphological generalizations in (35) to guide her through productive word formation (and compare with the adult system in (16)). (34)

(35)

~ [Spec,EP] orlgmator subject-ofquantifiable change ~ [Spec, ASP Q]

a. b. c.

EP dominates ASP Q

a. b.

root, +OM => V,III root, (+E) => II, VII

The resulting trees are in (36), the input to the morpho-phonology in (37).

(36)

a.

Telic and atelic transitives, compare with (19a-b) EP

sp~ DP E ASPQ/FP r

sp~ DP 2

[ASPdF]+OM

OM

Y.III

VP

I

V Sample roots:

cLM, SMX,

eLL, SXH, SRP

b.

Telic and atelic intransitive, compare with (19c-d) EP

sp~ (ASP E

DP 2

II, VII

Q

sp~ BP2

ASPq)

VP

I

V Sample roots:

28

PRK, cIB 2s

Atelic intransitives are missing from the sample, a fact which we take to be accidental.

Hagit Borer

(37)

Mapping to phonological representation (forms in parentheses conjectured and not attested in child sample; underlined forms are correct adult forms)

a.

b.

cLM,+OM CLL,+OM SMX,+OM SXH,+OM SRP,+OM PRQ,+ asp cLB,+asp

=> => => => => => =>

m,v m,v III,V III,V

m,v II,VII II,VII

cilem; (heCelim) (eilel); hiclil (simeax), hismiax (sixah); hisxah (sirejl), hisrijl nijlraq, hitpareq ne Celab, hitCalef2.

Structure

Example

(35a) (35a) (35a) (35a) (35a) (35 b) (35 b)

(3 0a) (29 a) (3 0b) (29 b) (28c) (3 Ia) (3 Ib)

The child now associates directly the morpho-phonology of some binyanim with the syntactic event structure. This still leaves a certain degree of under-determination, precisely because the canonical properties ofbinyanim III-V and binyanim II-VII cannot be teased apart by using event structure criteria. Rather, both binyan III and binyan V are canonically compatible with +OM; and both binyan II and VII are incompatible with +OM. We therefore expect precisely the behaviour reported in (29)-(33): a measure of randomization when it comes to the selection between binyan II and binyan VII, or between binyan III and binyan V, but no errors across these types. To conclude, the child has learned to focus on syntactic argument structure configurations as determining the particular selection of an appropriate morphophonological form. However, she clearly continues to project argument structure independently of anyone particular vocabulary item. Rather, having at her disposal the syntactic structures in (19)/(36), she continues to project event structures as based exclusively on syntactic considerations, linking argumental interpretation with particular syntactic position. As before, whenever memory fails her, the child will let her grammar alone select the correct morphological form of the verb for her. Unlike the previous stage, however, she now considers not only morpho-phonology, but morpho-syntax, as reflected in the particular syntactic structure associated with particular argument structure configurations. It may be worth noting that in a sense, the morpho-syntactic stage presents even more of a problem for models which entail the projection of arguments from the information stored in lexical entries, than the morpho-phonological stage. While during the morpho-phonological stage one could at least try to make a case for a computational overload leading to randomization, such an account is patently implausible for the children at the morpho-syntactic stage. There is no confusion here, but rather, a very systematic case of overgeneralization, showing a rather sophisticated computational apparatus in place. The child that produces masxe or cilem is not suffering from a computational overload, but rather, is over-computing, in a sense, when attempting to match the morphology with the syntax of the event structure in a regular way not attested in the adult grammar. If, indeed, argument structure is projected from lexical

1he Grammar Machine entries, what is the relevant lexical entry from which the argument structure of masxe or citem are projected? These forms do not exist in the adult language, and the child did not hear them or list them in anyway. The child obviously knows the meaning of the root, for example, eLM 'pertaining to disappearance', but equally obviously does not have the knowledge that there is a lexical entry citem with transitive argument structure-as such a form does not exist. Nor can the child be assumed to remember that there is a transitive entry associated with this root, but fail to recall its morpho-phonology. Were that to be the case, we would not expect errors to conform to canonical argument structure possibilities in the second developmental stage, but not in the first one. Rather, the child's behaviour is directly predictable, if we assume that the binyan morphology is a kind of agreement for the child-a reflection of syntactic structure. It is the syntax of arguments which determines the agreement-the binyan distribution-and not vice versa. And it is the assumption that agreement is 'regular' which leads to overgeneralization. Let us finally return briefly to the English learner. Recall that to the extent that the English learner goes through a morpho-phonological stage, like the Hebrew child, such a stage is obscured by the fact that performance in the morpho-phonological stage-giving rise to words whose phonology is insensitive to syntactic informationis by and large compatible with the target language, English, in which (underived) stems are indeed rarely phonologically marked for such information (pairs such as eat andftedbeing the exception rather than the rule). Suppose now that, like the Hebrew learner, the English child, too, proceeds to the morpho-syntactic stage, fully marking stems for their event structure environment. As it turns out, that stage is obscured as well, as English rarely marks agreement of any sort-event structure agreement is no exception here. The English-speaking child will thus continue to produce forms such as those in (27) and (28) in the morpho-syntactic stage, not because she does not lmow that her stems are now marked by event-structure features, but because these eventstructure features are phonologically unrealized in her target language. We conclude, then, that the passage through these two developmental stages, in evidence in Hebrew, is obscured in English, quite simply because the phonological distinctions that make it possible to discern these two stages are never overt in English. Between ages 2;2 (the onset of the morpho-phonological stage in Hebrew) and roughly 6;0 (the stage at which the morpho-syntactic stage phases out) we expect the English learner to persist with 'errors' such as those in (27) and (28), obscuring her passage through two distinct, but inert in English, developmental stages.

1I.6.

CONCLUSION

The main purpose of the chapter was to investigate the ramifications, for acquisition, of a grammatical model in which argument structure and event structure are not based on properties of vocabulary items, but rather, are associated with specific

33 0

Hagit Borer

syntactic structures and projected independently of vocabulary. Within such an approach, vocabulary items dominated by a verb function as modifiers, rather than as determinants of event structure. Crucially, if this model is on the right track, children acquiring language could not resort to their understanding of the meaning of a particular verb or any other information relevant to argument structure listed in individual entries to guide them into the syntactic projection of arguments. Rather, they must have syntactic knowledge on the projection of arguments and event structure independently of their knowledge of vocabulary. In turn, such children are expected to produce utterances in which the syntax of the event and the arguments and the actual verb embedded in it do not match. Such cases of mismatch were indeed found and were shown to cluster in a way that gave evidence for the existence of two developmental stages:

Stage A. The morpho-phonological stage, in which children lmow the syntax of events and the morpho-phonology of the binyan system, but do not show knowledge of the fact that the particular morphological form used with a particular root is not just conditioned by the morpho-phonology, but also by the syntactic event structure. At that age, errors in binyan selection tended to be random, and their statistical distribution across binyan types mirrored vocabulary distribution across binyan types in the language in general. Stage B. The morpho-syntactic stage, in which children augment their lmowledge of the syntax of arguments and the morpho-phonology of the binyan system with the understanding that the selection of a particular binyan is conditioned by the syntactic event structure. However, unlike adults, who use the syntax primarily to delimit the selection of the correct binyan in the context of a specific root, children appear to consider the binyan morphology as agreement of sorts, associated directly with specific syntactic event structures. As a result, children continue to confuse binyanim which have the same event function and tend to favour, at times erroneously, binyanim with well-defined morpho-syntactic properties over binyan I, which lacks them. A final important question must concern the recovery from the morpho-syntactic stage leading to adult performance. At some point Hebrew-learning children and English-learning children do learn that the forms in (27)-(3I) are not the standard adult forms, and that vocabulary insertion involves primarily a search through a finite, conventionalized, list. This search returns unique and at times idiosyncratic items for a particular syntactic environment, and lacks-at times idiosyncratically-possible but non-attested forms. Viewed differently, however, the adult system and the child grammar, at the morpho-syntactic stage, are not very different. Both adult and child have a list which they match against a set of syntactic environments. The child must, in fact, be assumed to have such a system, or a phenomenal rate of errors would be expected, contraty to fact. We suggested that children resort to the system in (34)-(37) whenever memory fails them and they must be productive, or alternatively, whenever they have not been exposed to a specific token but have already learned the root, and must make do with

?he Grammar Machine

33 r

a productive strategy rather than appeal to memory. Such productive strategy avails itself of the system in (34)-(37). But adults, too, have a productive word-formation system, virtually identical in properties to that outlined in (34)-(37). Such productive word-formation knowledge, on the part of the adult, comes to the front in the adult ability to comprehend novel expressions and to produce them, in the context of in novative word formation, extremely common in contemporary Israeli society, in many varied social and cultural domains. Children at the morpho-syntactic stage and adults, then, both have a productive word-formation component. Both have a vocabulary list from which they select items in accordance with the guidelines in (r6). They differ on one issue, however: the size and the accessibility of the vocabulary list. The adult list is bigger, and the adult's access is easier. Children are more creative, quite simply, because they have a smaller vocabulary, and because their memory fails them more often, forcing them to resort to rule-governed behaviour. Forms produced in this fashion, however, are not stored. Rather, they are produced 'on line', like syntactic structures, which are not committed to memory. When children's ability to store vocabulary and to access it improves, 'improvised' forms such as those in (27)-(32), in both English and Hebrew, disappear, quite simply because they are no longer produced on-line. However, the ability to produce them stays intact and is at the core of all future productive word formation and comprehension. There remains an open question, concerning not just the grammar of Hebrew, but grammar in general. Phrasal idioms aside, why are syntactic structures produced and comprehended on-line, and why does syntactic knowledge not avail itself of a list? Put differently, why does word formation remain a generative system which exists only at the periphery of a conventionalized vocabulary list? The answer to this general question notwithstanding, the picture of the language learner which emerges from this study, shows them to acquire the generative, computational, rule-governed aspects of linguistic knowledge independently of that conventionalized vocabulary list and well before it is fully in place. Computationally, the child is sophisticated and adultlike at a very early age. It is exactly those aspects of the linguistic behaviour which are not computational in nature and which may very well interact with general cognitive development, which the child comes to acquire fully at a late stage, well past the solidification of the computational system. It is thus precisely in this respect that the child is a little automaton, computationally sound-but conceptually lacking: a grammar machine.

12 Acquiring Unaccusativity: A Cross-Linguistic look' Janet Randall Angeliek van Hout, Jurgen Weissenborn, and Harald Baayen

12.1.

INTRODUCTION

Imagine a language learner. She confronts an unfamiliar intransitive verb such as glorp in (I).

(I)

Cookie Monster saw that [Ernie glorped].

Now she has a problem. She knows that glorp is intransitive, but is it unergative or unaccusative? How does she decide? The problem is stated in (2).

(2)

The learning problem for unaccusativity How does the learner decide to which class a given intransitive verb belongs?

For their helpful discussion and input, we thank Melissa Bowerman, Ursula Brinkmann, Jill Carrier, Marlene Jonas, Zvi Penner, and Susan Powers, as well as the many colleagues who have commented on presentations of this material. We are also grateful to out wonderful assistants: Berdine Bodegom, Bianca Hettlich, Bart Hollebrandse, Christina Lamertz, Tina Lieb, Ellis van Lieshout, Suzanne Requardt, and Colinda Verlinde, for their help thtough all stages of this research. The research was supported by grant WE-1236-2-2 ftom the Deutsche Forschungsgemeinschaft to Jiirgen Weissenborn, by a Research Grant

Acquiring Unaccusativity

333

In a context such as (I), the surface syntax does not help her. In English, although unaccusativestake an underlying object, at the surface they look just like unergatives with an NP subject.

Unergative She laughed.

Unaccusative [Shel disappeared [t]j.

Now, if unaccusativity is determined on the basis of semantics (Perlmutter 1978), the learner will expect a correlation between the semantics of a given verb and its syntax, and assign it to one class or the other based on that. What this means is that there is a difference between the meanings of the two classes of verbs at the level of Conceptual Structure (CS). This is indicated in (4) with P and R.'

Unergative

(4) S-Structure:

[NP

she] [yp laughed]

D-Structure:

[NP

she] [yp laughed]

Argument Structure: Conceptual Structure:

Unaccusative

[yp disappeared

[NP

she]]

I I I I I

a

(

~ [ P (x)

( a

I

[ R (x)

Based on this meaning difference, the two x arguments at the Conceptual Structure level link to Argument Structure (AS) in different ways. In the unergative CS (laugh) on the left, the x argument links to an external position in the AS, and projects to the subject position in D-Structure, where it stays at S-Structure. In the unaccusative CS (disappea/~ on the right, the x argument links to an internal position in the AS, and projects to the D-Structure object slot. It then moves to the subject position at S-Structure, to satisfY the need in English for a surface subject. Now, if the linking difference between the two verb classes is based on semantics, we need to say specifically which semantic factors matter. What is it about the meanings of laugh versus disappear that makes the linking from CS to AS different? Going back to our learner trying to classifY the new verb glorp, the question is: (5)

What semantic factors is she looking for?

from the Alexander von Humboldt Foundation to Janet Randall, and by grant 300-75-025 from the NWO (Netherlands Science Foundation) to Angeliek van Hout. The Max Planck Institute for Psycholinguistics also contributed generous support. This chapter is an abridged version of a longer article (Randall, van Hout, Weissenborn, and Baayen, in progress); space constraints force us to limit our discussion of related work on the theory and acquisition of unaccusativity. However, we provide a brief bibliography of some of that work in the reference section. t To be clear, Perlmutter's (I978) claim was simply that semantics is relevant to syntax; he did not frame his claim in terms of Conceptual Structure.

Randall van Hout, Weissenborn, and Baayen

334

Secondly, should we expect learners across languages to look for the same ones? That is, (6)

Are the linking rules universal?

12.2.

LINKING RULES ACROSS LANGUAGES

To approach the second question first, when we look across languages, there are linking rules that exist in one language but are missing from another, as Carter (1988) pointed out. A linldng difference between English and French is illustrated in (7):

(7)

a. The bottle floated under the bridge. (location/movement reading) b. La bouteille a flotte sous le pont. (location/*movement reading)

In English, (7a) is ambiguous; it can mean either that the bottle was located under the bridge while it was floating, or that it moved to a position under the bridge in a floating manner. In French (and other Romance languages) this second reading is impossible. French does not allow [NP la bouteille] ('the bottle') to appear in the subject position of a sentence containing a manner verb that takes a directional PP.2This meaning has to be expressed in another way. Now, given that we find differences in the linking rules that languages contain, we might also find differences in how a particular linking rule looks in two different languages. Two languages could share a common linking rule, but in different versions. This, in fact, seems to be the case in Dutch and German, for the linking rules that determine whether a given intransitive verb is unaccusative or unergative, as we will see below. Before we can understand what the linking rules for intransitive verbs look like, we need an independent way to tell apart unaccusatives and unergatives. One diagnostic that has been proposed for these two languages is auxiliary-verb selection. As shown in (8), unergatives take HAVE and unaccusatives take BE.

(8) German Dutch

Unergatives

Unaccusatives

haben hebben

sein zijn

HAVE

BE

Most of the time Dutch and German agree on the auxiliary for a verb in a given context. In both languages, as shown in (9) and (10), for instance, dance on the table takes HAVE and dance into the room takes BE. (9)

Dutch: John heeft urenlang op de tafel gedanst. German: John hat stundenlang auf dem Tisch getanzt. 'John AUX been dancing on the table for hours.'

, The prohibition actually applies before the NP moves to subject position.

Acquiring Unaccusativity (10)

335

Dutch: John is in twee seconden de kamer in gedanst. German: John ist in zw'ei Sekunden ins Zimmer getanzt. 'John AUX danced into the room in two seconds.'

But, interestingly, Dutch and German disagree on the auxiliary in (II) for dance

around the room. Dutch takes HAVE, but German takes BE. (II)

Dutch: John heeft urenlang door de zaal rondgedanst. German: John ist stundenlang durch den Saal herumgetanzt. 'John AUX been dancing around the room for hours.'

The situation looks as in (12). The two languages draw the line between unaccusatives and unergatives in different places. Unaccusative verbs such as dance into the room are to the left of the lines for both languages, unergatives such as dance in the room are to the right of the lines for both. But dance around the room falls between the two lines. It falls on the unergative side in Dutch but on the unaccusative side in German. This means that, if we use auxiliary selection as our indicator of verb class, then Dutch and German must have slightly different linking rules for distinguishing between the two classes of intransitive verbs.

(12) Unaccusative

Unergative

BE

HAVE

Dutch

12.3.

German

THE SEMANTICS OF UNACCUSATIVES

How can we characterize the difference between the two intransitive verb classes in each language? Can we find a semantic explanation underlying the split? And can we explain, also in semantic terms, the difference we just found between Dutch and German? Notice that what differs in the three sentences in (9)-(II) are the prepositional phrases. In (10), john has danced into the room, where both Dutch and German select BE, the preposition is into. In this sentence, John moves from outside the room to inside and he must end up at a different point from where he started. In (9), john has been dancing on the table, where both languages select HAVE, John need not change his position; he can dance in place. In (II), john has been dancing around the room, just as in (9), John can travel around and finish in the same spot that he started. In Dutch this verb phrase is categorized with (9) and assigned HAVE. Only (10) is considered unaccusative and assigned BE.

Randall van Hout, Weissenborn, and Baayen Looking at the semantics of these three cases, what seems to matter for unaccusativity in Dutch is 'telicity', that is, whether or not an event comes to an endpoint. Although there have been slightly different uses of the term 'telic' in the literature/ for clarity, we will use it in its classical sense, in which a telic situation is one that has an endpoint (Comrie 1976; Smith 1997): [... J telic events have a natural final endpoint, or intrinsic bound. In contrast, atelic events are simply processes. They can stop at any time: there is no outcome. In other words, atelic events have arbitrary final endpoints. (Smith 1997: 19)

Verbs whose CSs contain an endpoint will link as unaccusatives, using what we will call the Telicity Linking Rule, in (13). (This rule is stated in the formalism of Jackendoff (1990). INC BE stands for INchoative BE, or 'comes to be'.) (13)

Telicity Linking Rule: disappear, arrive, dance into the room (unaccusative / BE) AS: CS:

( a INC BE (

I

x,

AT .•.

We can read the rule as follows: an x argument that comes to be AT a new place (or state) links to an internal argument position in AS. The rule in (13) applies to all verb phrases that contain INC BE: to verbs such as disappear and arrive and to more complex predicates such as dance into the room. But it does not apply to dance around the room or dance in the room, which do not contain INC BE, only BE. We can differentiate verbs that have INC BE in their CS from those that do not with the feature [±telic]. However, it is important to remember that the [±telic] feature is really only a shorthand for a type ofCS.4 What about German? German classifies (II), dance around the room, with (10) dance into the room. As such, it cannot be using the Telicity Linking Rule in (13) to make its unergative-unaccusative distinction. As we saw, (13) applies to (10), but not to (II), where there is no endpoint. In order to classifY this predicate as unaccusative, German must be using something else. One possibility is a linking rule based on what we will call 'locomotion'. Locomotion is not any kind of motion-it refers only to 'travelling' motion. Wiggling or stretching, for instance, is not locomotion, nor is dancing in place. But dancing around the room in (II) and dancing into the room in (10) are both instances oflocomotion. Ifwe classified predicates in terms of a [±locomotion] feature, and if German had a linking rule based on this feature, then both (II) and (10) would be unaccusative in German. 5 1 For example, van Hout (I998: 92) uses telic not to refer only to events that reach endpoints, or 'terminative' events; but more broadly, to refer to all event types with a moment of temporal transition, including resultative and also inchoative and causative events. 4 We are using features strictly as a shorthand device, to make the differences between the verb classes easy to see. Technically spealdng, though, in a framework like ours that uses full-blown CS representations, the features can be read off of the CSs, and are not additional tags assigned to them. , Of course, (10) would also be classified as unaccusative by the Telicity Linking Rule so it is tempting to suggest that we should just replace the Telicity Linldng Rule in German by a more inclusive Locomo-

Acquiring Unaccusativity

337

Once we add the locomotion feature, only (9), dance on the table, will be classified as both [-locomotion] and [-telic], and will be assigned an unergative syntax by both rules. We do not want to characterize the Locomotion Linking Rule here, but we want to stress that according to the data we have seen so far, Dutch and German split the intransitives using two different semantic factors. As shown in (14), Dutch uses telicity; German uses locomotion.

(14)

Dutch: Telidty Linking Rule German: Locomotion Linking Rule

We can add these factors into our picture of where the two languages divide the unergatives and unaccusatives:

Unaccusative

Un ergative

BE

HAVE

. +locomotion: -locomotion +telic

~ -telic

Dutch

~

German

Before we return to our learner, in (16) is a diagram of how the features 'telic' and 'locomotion' map onto other semantic features that have been proposed in the literature for unaccusatives, for example, directed change (Levin and Rappaport Hovav 1992b), and what Lieber and Baayen (1997) have called Inferrable Eventual Position or State (IEPS). (16)

+telic

I

-telic +directed change +IEPS

dance into the room

I I I I

German:

BE

Dutch:

BE

I I I I I

dance towards the room

I I I I

I I

I I I I I I I

-directed change -IEPS +locomotion

I I

-locomotion

dance around in the room

I I

dance in the room

German: BE Dutch:

HAVE

I

I I I I I I

German:

HAVE

Dutch:

HAVE

Let us now turn to one more semantic factor that has been suggested to be relevant to unaccusativity, which we can also state as a feature: [±actor]. Like [±telic], [±actor] has been characterized in slightly different ways, some of which are listed in (17). tion Linking Rule. However, we cannot do this, since German will still need the Telicity Linking Rule for change-of-state verbs such as disappear and break, which are [-locomotion].

Randall, van Hout, Weissenborn, and Baayen

338

(17)

±actor I control I internal cause I no direct external cause I agent

Although [actor] may, in fact, be reduceable to other semantic properties, we will use it here to malce the very rough semantic distinction between an animate participant actively doing something and a completely non-volitional inanimate participant. An independent diagnostic for [±actor] is JackendofFs (1990) 'actor test' in (18).

(18)

The Actor Test: What John did was ... (laugh, sing, sleep, dance, *disappear, *arrive)

Verbs that pass this test contain [ACT] in their CS, so the actor test distinguishes the subjects of laugh and sing from the subjects of disappear and arrive. A linldng rule based on [actor] is shown in (19). The x argument of a CS that contains ACT links to the external position in AS. When this linking rule applies, the x argument qualifies as an actor and the verb is classified as unergative.

(19)

Actor Linking Rule: laugh, sing, sleep, dance AS:

a

CS:

fDO (x)

(unergative/HAVE)

\

LACT How does the Actor Linking Rule operate alongside the Telicity Linking Rule in Dutch? For the two Dutch [-telic] cases, dance on the table and dance around the room, the Telicity Linldng Rule does not apply. The Actor Linking Rule is the only relevant rule, and it links these verbs as unergative with HAVE as the predicted auxiliary. This was shown in (9) and (n): (9)

(n)

Dutch: John heeft urenlang op de tafel gedanst. German: John hat stundenlang auf dem Tisch getanzt. 'John AUX been dancing on the table for hours.' Dutch: John heeft urenlang door de zaal rondgedanst. German: John ist stundenlang durch den Saal herumgetanzt. 'John AUX been dancing around the room for hours.'

In our [+telic] case, repeated from (IQ),John danced into the room, the Telicity Linking Rule does apply, linking the verb as an unaccusative.

(10)

Dutch: John is in twee seconden de kamer in gedanst. German: John ist in zwei Sekunden ins Zimmer getanzt. 'John AUX danced into the room in two seconds.'

However, notice that this sentence also contains an actor, John. So the Actor Linking Rule should apply too, which would make the verb unergative. What happens in this case, where the two rules conflict? We propose that the answer lies in the geometry of

Acquiring Unaccusativity

339

the CS for the verb phrase, which contains both the CS of dance in (20) and the CS of into in (21). The verb dance is a manner-of-motion verb that takes an actor but does not specifY an endpoint. But the preposition that it combines with, into, does encode an endpoint. In Carrier and Randall (1993), we propose that a preposition like into is a two-place predicate, containing INC BE. This is where the endpoint comes from. (20)

dance:

IDO[+motion]

(z)

LACT

(21)

into:

INC BE

(x, AT (y))

In (22) we can see how these two CSs combine. If we conceptualize dance into the room as meaning 'go into the room by dancing', then the CS for dance is subordinated under the CS for into. (22)

dance into:

INC BE (Xi' AT

(y))

VIA I DO[+motion]

(z)

LACT

Linking always begins at the top with the highest clause. Since this clause contains INC BE, the Telicity Linking Rule (repeated below), applies first. The x in (22) is linked to an internal AS position, maldng the entire phrase unaccusative, and the predicted auxiliary is BE. Since linldng begins at the top, the Actor Linldng Rule does not have a chance to apply, because ACT is lower in the representation than the INC BE clause. (13)

Telicity Linking Rule: disappear, arrive, dance into the room (unaccusative / BE) AS: CS:

( a INC BE (

I

x,

AT ••.

We have one more case to look at-namely, when the Telicity Linldng Rule applies but the Actor Linking Rule does not. Such a case is (23).

(23)

Dutch: De tennisbal is in twee seconden de baan op gerold. German: Der Tennisball ist in zwei Sekunden auf den Tennisplatz gerollt. 'The tennis ball AUX rolled onto the court in two seconds.'

Here there is no conflict between the two rules, and the sentence is unaccusative. To summarize, the three cases we have seen so far are shown in (24), In the lower left cell, only the Telicity Linldng Rule applies, and BE is selected. In the upper right cell, only the Actor Linking Rule applies, and HAVE is chosen. However, in the upper left cell, where both the Telicity and Actor Linldng Rules apply and lead to different outcomes (both HAVE and BE), the conflict is resolved by the geometry of the CS. Although BE is finally chosen, HAVE appears in parentheses to reflect its potential competition with the other linking rule.

Randall, van Hout, Weissenborn, and Baayen +telic

-telic

+actor

BE (HAVE)

HAVE

-actor

BE

II

We have not discussed what happens where neither linking rule applies, cases that would fall into the bottom right cell. At this point we have no prediction to make, but if there were a 'default' linking rule (as proposed by Levin and Rappaport Hovav 1995), this is where it would apply.

12.4.

THE LEARNER AND THE LINKING RULES

Let us summarize so far. We have proposed two linking rules that appear to be operating for Dutch intransitive verbs. One of these, the Telicity Linking Rule, is clearly insufficient for German. The other rule, the Actor Linking Rule, applies in both languages. Now let us go back to the learner. Our question was what factors she would be looking for in trying to determine whether a given intransitive verb, such as glorp in (I) (repeated here), is unaccusative or unergative. To see whether she pays attention to the two factors [factor] and [±telic] we can construct new verbs in which we systematically vary these factors and see how she behaves.

(I)

Cookie Monster saw that [NP glorped].

Moreover, we can look not only at Dutch, where both of these factors operate in the adult system but also at German, where a third factor, [±locomotion], plays a role. This is just what we did in a set of experiments. Before we turn to them, we want to address one other issue that could affect what our learners will do. Consider again how verb phrases link as unaccusatives. They can satisfY our Telicity Linking Rule in two different ways. One way is when the verb itself is telic, because it contains END in its CS. Telicity is an inherent property of the verb's CS, as in disappear in (25). We will call such cases 'inherently telic'. Their [Helic] feature is [+inherent] in the verb's CS. (a

disappear:

INC BE

!

(x, AT

) (END))

The second way that a verb phrase can link unaccusatively is when a [-telic] verb comines with a [Helic] PP. This is a common property of manner-oE-motion verbs such as dance. Though the verb is not inherently telic (it has no END in its CS), combining it with into the room produces a [Helie] Vp, [dance into the room] The combination, in terms of features, is [Helic] [-inherent]. (Of course, it is also possible for a [+inherent] [Helic] verb such as disappear to combine with a [+telie] Pp, but that case is not relevant here.)

Acquiring Unaccusativity

341

Now consider the effect of the [±inherent] feature by looking at our new verb,

glorp. Imagine that glolp is a [+telic] verb like disappear. And imagine that you observe someone 'glorping' while you hear a sentence such as (26a). In order to interpret glorping as a [+telic] event and assign the verb phrase to the unaccusative class, you need to observe that the glorping activity involves a clear endpoint. (26)

a.

b.

Ernie saw that [Bert glorped in the forest]. Ernie saw that [Bert glorped into the forest].

This is not the case with (26b). In this [+telic] case, it is not only the scene that can tell us that the verb phrase is unaccusative, the syntax does, too. English uses into as opposed to in. German and Dutch also both mark the PPs clearly. German uses casemarking; Dutch uses postpositions. (Cases without an endpoint are marked with a preposition in Dutch.) (27)

a.

b.

Dutch: Ernie liep in het bos. German: Ernie lief in dem Wald. (dem = dative) 'Ernie ran in the woods.' Dutch: Ernie liep het bos in. German: Ernie lief in den Wald. (den = accusative) 'Ernie ran into the woods.'

So for cases such as (27b) there are two sources of information about which class glorp belongs to; the visual information from the scene and the auditory information from the syntactic marking-either case or a postposition. Now consider the learner. We hypothesize that she will find it easier to recognize an endpoint when it is overtly marked, belonging to our [-inherent] class, than when it is part of the lexical meaning of the verb, or [+inherent]. That is, sentences that mark the endpoint with either case or a preposition or postposition will be more readily understood as having endpoints than sentences in which the endpoint is not overtly marked in the syntax, but is detectable only from the scene. We can add this prediction to the chart in (24) as shown in (28). Notice that it changes the chart only on the [+telic] side. It predicts that the [+inherent] [+telic] verbs (verbs like disappear, where the endpoint is inherent in the meaning, not marked with a PP or case) will be harder to classifY as unaccusative (and will get fewer BE auxiliaries) than the verbs whose endpoints are [-inherent], overtly marked with case or an unambiguous preposition or postposition. ?BE indicates fewer BE responses than BE. And remember that the (HAVE) -telic

+telic

+inherent +actor -actor

-inherent

+inherent

-inherent

'stretch'

'dance into the room'

'laugh'

'dance in the room'

?BE (HAVE)

BE (HAVE)

HAVE

HAVE

'disappear'

'roll into the room'

'sparkle'

'roll in the room'

?BE

BE

??

??

"

~

i:t

~

~

i:t 1:::

~

(29)

+actor

+inherent

-inherent

-telic -inherent

+telic +inherent

4

(

3

a

2

)

I

(

(x)

~

[DO [+motion] ACT

8

[DO (x) ACT

~

a

'dance in the room'

a )

'laugh'

( a

'dance into the room' )

~

(y)])

6

(x)

'roll in the room'

DO[+motion]

'sparkle'

(x)

7

DO

-

-

'roll into the room' ( a

(y)])

VIA [DO [+motion] (Z)

INC BE (Xi [AT

~ ~

a )

VIA[DO[+motion] (Zi) ACT

INC BE (~ [AT

~

'stretch' ( a

~

(z)

)

INC BE (~ [AT (END)])

ACT

VIA[DO

<::>

"

~ .... 5

;:!

( a

'disappear'

~

i:t

...s::,

-actor

INC BE (Xi [AT (END)])

~

~ i:t

~ ~

-...

i:t

~

~

""'

"T

N

-

)

Acquiring Unaccusativity

343

in the top two left cells comes from the fact that the actor is present along with the endpoint in these verbs' CSs, which might lead to some uncertainty in assigning them to the unaccusative class. At the top of each cell is an English verb that fits the category. In the expanded chart in (29) these verbs are shown with their ASs and CSs. 6 Let us sum up so far. We have focused on two semantic factors that are encoded in Linking Rules for classifYing intransitive verbs, [±telic] and [±actor]. These seem to capture the facts of Dutch adult grammar. A third factor, [±inherent], indicates whether or not a verb is inherently specified for [+telic]. Our [-inherent] verbs such as dance do not inherently encode an endpoint. They can switch from [+telic] to [-telic] depending on the PP they appear with. Our [+inherent] verbs do not switch. They are either like stretch or disappear with an endpoint inherent inside their CS, [+inherent, +telic], or like laugh or sparkle with no endpoint, [+inherent, -telic]. We have proposed that when a verb is [+telic], it will be easier for a learner to classifY it as unaccusative if it is overtly marked with a syntactic clue (Le., if it is [-inherent]), as in dance into the room, than if it is not overtly marked but has a [+inherent] endpoint, as in disappear. In all, we are looking at three variables: [telicity], [inherency], and [actor], which give us eight types of verb?

12.5.

AN EXPERIMENT TO TEST THE SEMANTIC FACTORS IN LINKING RULES

The question is: how do learners use the three semantic factors to categorize intransitive verbs that they have never seen before? Our study focused on Dutch and German, two languages in which auxiliary selection has been proposed as a marker of intransitive verb class. Although we have seen that these two languages divide up the intransitives in slightly different ways, in this study we expected similar results, because we used only those verbs on which the two languages overlap. In other words, we did not use any cases such as dance around the room, which are unaccusative in German but unergative in Dutch.

6 Because our two linking rules make no predictions about how the CS arguments in cells 7 and 8 link to AS, we do not show linking lines for these cells. As we noted in section I2.3, though, in some theories a Default Linking Rule stipulates the linking of arguments that do not fall under the scope of any other linking rule. Levin and Rappaport Hovav (I995), for example, stipulate that such arguments link internally. 7 It should be kept in mind that many verbs can qualifY for membership in more than one cell, depending on the context. For instance, a verb like drop can be used either with a volitional [actor] (When they heard shots, the soldiers dropped to the ground) or with an inanimate, non-volitional participant (The glass dropped out ofSarah's hand). We are not claiming that all verbs belong to one cell uniquely.

Randall van Hout, Weissenborn, and Baayen

344

The subjects who participated in our study are shown in (30).

(30)

Number of subjects in each age group Dutch German

4-5 years 14 19

7-8 years 18 18

Adult 16 15

Our methodology was a cloze task. Our subjects watch a series of movies, each one showing a new type of event that is describable by a new verb. Cookie Monster (manipulated by the experimenter) watches the movie alongside the subject, and his job is to learn the new verb from our experimenter. The subject is asked to help Cookie Monster learn the verbs, 'because he's not very good at it'. Each scene depicting a verb had a script similar to (31), which is for a scene containing a [-telic, +inherent, +actor] verb. In this scene, Ernie continually wiggles his mouth in a back-and-forth wavy motion, as he sits on a pile of books.

(31)

A sample Dutch script Experimenter:

Hier is Ernie. En een stapel boeken. En nu komt plurgen.

Here's Ernie. And a pile ofbooks. And now comes plurging. Can you say that? Subject:

plurgen (repeats the verb)

plurging Experimenter:

Dat was plurgen. Ernie plurgt op de boeken.

That was plurging. Ernie is plurging on the books. Can you say that? Subject:

Experimenter (to CM): Cookie Monster:

Experimenter (to subject): Subject:

Ernie plurgt op de boeken. (repeats the sentence)

Ernie is plurging on the books. Now, Cookie Monster, tell us what you saw: Ik zag dat Ernie op de boeken ge- uh, ... , ge-, uh ...

I saw that Ernie on the books ge- uh, ... , ge-, uh ... Can you help Cookie Monster? geplurgd is/heeft participle AUXILIARY

Cookie Monster (played by the experimenter) always falters on the participle, pretending not to remember the verb, but at the same time he fails to supply the word that follows the participle. That word is the crucial auxiliary verb, which tells us whether the subject classifies the new verb as unergative (HAVE) or unaccusative (BE). Subjects who saidgeplurgd heeft ('has plurged') chose the unergative auxiliary, those who said geplurgd

Acquiring Unaccusativity

345

is ('is plurged') chose the unaccusative auxiliary. We tested the three factors, [±telic], [±actor], and [±inherent], using the eight-cell design in the table in (28). There are sixteen scenes, two in each of the eight conditions. They are listed in the Appendix.

12.6.

RESULTS AND DISCUSSION

The results are set out in Tables 12.1-6, in two different ways. We begin with tables showing the percentages of HAVE and BE chosen in the eight cells in which we made our predictions. In each cell, the percentage of subjects that chose the predicted auxiliary is in bold. The HAVE responses are to the left in each cell, the BE responses, to the right. Recall that in each cell we encode our predictions as follows. In cells (3), (4), and (6), either HAVE or BE is clearly predicted, so the cell contains simply H or B. In cells (I) and (2), subjects should use the Telicity Linking Rule to choose BE, but may be influenced by the presence of an actor in the scene (even though the Telicity Linking Rule should 'win' over the Actor Linking Rule). So in these cells, both auxiliaries appear, with H in parentheses. In cells (I) and (5), BE is predicted by the telicity facts but telicity is inherent and therefore potentially not detectable to our subjects. So, here, BE is marked with a ? In cells 7 and 8, neither the Telicity Linking Rule nor the Actor Linking Rule applies, so there is no predicted preference for HAVE or BE. This is indicated with a'?'. Tables I2.1-3 show the Dutch results for adults, 7-8-year olds, and 4-5-year olds, respectively. 8 (To remind the reader which verb classes our nonsense verbs belong to, we have inserted some English verbs at the top of each cell. But of course, our subjects heard no real verbs at all, only nonsense verbs belonging to the same classes.)

TABLE 12.1. Dutch adults: per cent choosing HAVE or BE auxiliary (n=16) +telic

+actor

-telic

+inherent

-inherent

1 'stretch'

2

(H)

?B

97

3

5 'disappear' -actor

'dance into the room' (H) 12

6 'roll into the room'

?B

41

59

B 88

0

B 100

+inherent

-inherent

3 'laugh'

4 'dance in the room'

H 81

19

9

8 'roll in the room'

7 'sparkle'

? 88

H 91

?

?

?

12

78

22

8 For a more complete discussion of the results, together with statistical analyses, see van Hout (1998: ch. 6) and Randall et al. (in progress).

Randall van Hout, Weissenborn, and Baayen TABLE

12.2. Dutch 7-8-year oIds: per cent choosing HAVE or BE auxiliary (n=I 8) -telic

+telic +inherent

-inherent

+inherent

-inherent

I 'stretch'

2 'dance into the room'

3 'laugh'

4 'dance in the room'

+actor

(H)

?B

(H)

B

H

67

33

47

53

75

5 'disappear' -actor

6 'roll into the room'

?B

58

TABLE 12.3.

42

22

B 78

25

17

8 'roll in the room'

7 'sparkle'

? 64

83

?

?

?

36

69

31

Dutch 4-5-year oIds: per cent choosing HAVE or BE auxiliary (n=I4) -telic

+telic

+actor

H

+inherent

-inherent

+inherent

-inherent

I 'stretch'

2 'dance into the room'

3 'laugh'

4 'dance in the room'

(H) 79

?B 21

5 'disappear' -actor

(H) 54

6 'roll into the room'

?B

75

25

B 46

32

B 68

H 79

H 21

25

8 'roll in the room'

7 'sparkle'

? 68

75

? 32

? 82

? 18

What we see, for all three groups of subjects, is that on the [-telic] side of the tables, in cells 3, 4, 7, and 8, the predominant response is HAVE. The BE responses cluster on the [+telic] side. This suggests that we have evidence for an effect of our [telicity] feature. Notice that within the four [+telic] cells, the BES are strongest in cells 2 and 6, the cells in which the endpoint is detectable from the postpositional phrases. Although the endpoints were visually marked in the scenes in all four [Helic] scenes, hearing the syntactically marked directional PP increased the likelihood that the subject would classifY the verb phrase as unaccusative. The endpoints that were [+inherent], contained in the verbs but not mentioned in a Pp, in the scenes for cells I and 5, may not have been detected. So our [inherency] factor is playing a role here. What about [±actor]? This factor is exactly what we need in order to explain the difference in the adult responses between cells I and 5. Here, we just said that some of our subjects might not have detected the endpoints. But this is equally likely for cellI as for cell 5. The only difference between these scenes was in [±actor]. For the [+actor] scenes (in cellI), the adult subjects virtually always assigned the verbs to the HAVE

Acquiring Unaccusativity

347

class. For the [-actor] scenes (in cell 5), this happened much less often. Thus, [±actor] seems to be playing a role for Dutch adults, but not in those cases in which subjects are clearly using the Telicity Linking Rule. Now, we might have expected [±actor] also to distinguish cells 3 and 4 from cells 7 and 8. However, since we made no prediction about 7 and 8, it is not clear what is happening. This non-difference is not evidence against the Actor Linking Rule, though, since this is exactly what we would expect if there were an unergative default linking rule when neither the Telicity Linking Rule nor the Actor Linking Rule applied. To sum up, from the adult Dutch results we have evidence for all of our factors. And we also have evidence that when both the Telicity Linking Rule and the Actor Linking Rule apply, the Telicity Linking Rule wins. Now let us look more carefully at the youngest Dutch subjects, the 4-5-year olds. Here, again, though the results in every cell are less clear-cut than the adult results, the cells with the highest numbers of BE are cells 2 and 6. Again, cell 6 is the strongest, as we would expect if there is some competition in cell 2 from the presence of [+actor]. Basically, the only scenes that are categorized as unaccusative are those with a syntactically detectable endpoint, namely, cells 2 and 6-in fact, these children did no better in detecting the endpoints in cells I and 5 than they did where there were no endpoints, in cells 3, 4, 7, and 8. So, a syntactic PP is used even by 4-year oIds, in helping to decide what the semantics of a verb is. The 7-8-year olds basically fell between the youngest children and the adults. The differences just get stronger as we move from the youngest subjects to the adults. Overall, then, we see that differences in lexical semantics, unless they are accompanied by a detectable syntactic (or phonological) difference, do not help subjects to sort out verb type. With new verbs, it is harder to pick up on lexical telicity (inherent in the verb) than on compositional telicity, where the endpoint is clearly marked. While the charts show a clear pattern of results, in order to see which differences are significant, we can view the data using what are called 'classification trees' (Breiman, Friedman, Olshen, and Stone 1984).9 Each factor that plays a significant role adds structure to a tree. The more significant factors, the more structure. Note that

(32)

Classification tree for Dutch adults telic

~

H 84%

inherent ~ actor actor

~

B

IOO%

;/"'Z

B

B

H

88%

59%

97%

9 In this chapter we use simplified versions of the classification trees that emerged from our full analysis. In the more elaborate versions of the trees, the length of the tree branches indicates the relative proportions of the reduction in deviance that a split at any particular node brings about (see van Hout 1998; Randall et. al, in progress).

348 (33)

Randall, van Hout, Weissenborn, and Baayen Classification tree for Dutch 7-8-year olds telic

~

actor

inherent

~

B

(34)

~

H

actor 79% ~

B

B

78%

53%

H

Classification tree for Dutch 4-5-year olds telic

~

H

inherent

~

B

H

57%

77%

we have entered the percentages of HAVE and BE based on the majority of responses. Branches in which HAVE was chosen more often are represented with per cent H, and branches in which BE was preferred show per cent B, so that it is clear which auxiliary was preferred for each category. 10 What we see in the trees for each subject group confirms our earlier conclusions. Beginning with the adults, we see that they assign HAVE to the [-telic] cases, with no other factor having a significant effect. BE is assigned to the [+telic] cases, and within these, was chosen significantly more often in all of the [-inherent] cases, which correspond to cells 2 and 6. However, the 88 per cent BE for the [+actor] cases, as opposed to 100 per cent when no actor is present, is significant-it is evidence for the [±actor] factor. We were also correct that [±actor] played a role in distinguishing the responses in cells rand 5, where the endpoint may not have always been detectable, since we see a split, 59 per cent BE for the [-actor] branch (cell 5) versus 97 per cent HAVE for the [+actor] branch (cell I). Interestingly, (34), the tree for the Dutch 4-5-year olds, tells us that their important distinction is between the [+telic] [-inherent] cases, that is, the detectable endpoint cases, cells 2 and 6, versus all the others. And the difference between cells 2 and 6, which shows a trend towards an effect of [±actor] is not significant. So these children appear to be using the Telicity Linking Rule wherever they detect an endpoint. The German results follow, beginning with the tables. Again, in each cell the percentage of the predicted auxiliary is in bold. Ifwe look at the adults, in Table 12.4 again the highest numbers of BE cluster in cells 2 and 6, where the endpoints were detectable 10 Alternatively, we could have used only BE responses and converted all the HAVE responses to percentage of BE responses by subtracting from 100.

Acquiring Unaccusativity

349

from the syntax as well as the scenes, as we predicted. But there is something going on in cells 4 and 8, where we find higher numbers of BE than we found in the Dutch adults. Notice that these four cells (2, 6,4, and 8) are our four [-inherent] cells, which correspond to our flexible verbs like dance. That is, these four cells all contain verbs that can switch classes. And the verbs in this 'switchable' class correspond to [+locomotion] verbs. What we see is that the [+locomotion] verbs are being treated differently by German adults than the other verbs. This is not such a surprise when we remember that German, unlike Dutch, employs the Locomotion Linking Rule, using BE not only for [+telic] verbs, but for all [+locomotion] verbs. Now, the scenes that we showed with the verbs in cells 4 and 8 did not include endpoints, but the actions were clearly [+locomotion]. So a German speaker who is paying attention not to telicity but to locomotion could treat these scenes differently from the others, even though all of the

TABLE

12.4. German adults: per cent choosing HAVE or BE auxiliary (n=1 5) +telic

+actor

-telic

+inherent

-inherent

+inherent

-inherent

r 'stretch'

2 'dance into the room'

3 'laugh'

4 'dance in the room'

(H)

?B

(H)

B

H

93

7

7

93

93

5 'disappear' -actor

?B 37

TABLE

6 'roll into the room'

B 0

57

100

7

7 'sparkle'

? 50

63

37

8 'roll in the room'

? 50

? 30

? 67

12.5. German 7-8-year olds: per cent choosing HAVE or BE auxiliary (n=r8) +telic

+actor

H

-telic

+inherent

-inherent

r 'stretch'

2

(H)

75

?B 25

5 'disappear' -actor

'dance into the room' (H) 31

6 'roll into the room'

?B

44

56

B 69

24

B 76

+inherent

-inherent

3 'laugh'

4 'dance in the room'

H 78

H 22

7 'sparkle'

? 69

75

25

8 'roll in the room'

? 3I

? 61

? 39

Randall van Hout, Weissenborn, and Baayen

35 0 TABLE

12.6. German 4-5 -year olds: per cent choosing HAVE or BE auxiliary (n=14) -telic

+telic +inherent

-inherent

+inherent

-inherent

1 'stretch'

2 'dance into the room'

3 'laugh'

4 'dance in the room'

+actor

(H)

?B

74

26

5 'disappear' -actor

(H) 39

6 'roll into the room'

?B 47

B 61

B 26

53

74

H 82

16

29

8 'roll in the room'

7 'sparkle'

? 61

H 71

? 37

? 66

? 34

prepositions in these scenes were in or on. The subject could have assigned these scenes to the unaccusative class on the strength oflocomotion alone." One might now ask why, if the German adults are using [±locomotion], the BE responses are stronger in cell 8 than in cell 4, and similarly, in cell 7 than in cell 3. This is predicted by the presence of [+actor] in cells 3 and 4, which should pull responses in the HAVE direction. Confirming the adult responses in Table 12.4 is the corresponding tree in (35), which shows that the German adults are using the [±inherent] feature in classifYing the verbs. In other words, they seem to be basing their auxiliary choice on whether the verb is [±locomotion]. (Remember that the [-inherent] verbs are all [+locomotion], and the [+inherent] verbs are all [-locomotion].) Those [+locomotion] verbs that have a detectable endpoint are assigned BEs virtually all the time, 97 per cent. But even [+locomotion] verbs are assigned BE 52 per cent of the time when they occur with a [-telic] PP (corresponding to cells 4 and 8). On the [+inherent] side, which corresponds to cells I, 5,3, and 7 (where the verbs are [-locomotion]), [telic] plays no role, but [actor] emerges in the predicted direction. (35)

Classification tree for German adults inherent

~

telic

actor

~

/"Z

B

B

B

52 %

97%

54%

H

11 Because we are evaluating the effect of the three factors [telic], [inherent], and [actor], and not [locomotion], our predictions for German do not take into account the possible effect of the Locomotion linking Rule. If German adults treat all [+locomotionj verbs as unaccusative, this could dilute the expected HAVE responses in cells 4 and 8, as compared with Dutch adults. This appears to be the case.

Acquiring Unaccusativity (36)

35 1

Classification tree for German 7-8-year oids telic

~

H

inherent

~

B 73%

(37)

actor

/"Z

B

H

56%

75%

Classification tree for German 4-5-year olds telic

~

H

inherent

~

B

actor

/"Z

B

H

53%

74%

Interestingly, when we look at the trees for the children, the results look very different from the adults. For both groups of children, [telic] is at the top, just as it is for our Dutch subjects. So German children appear to be behaving like Dutch children in some respects: when the event has no endpoint, they choose HAVE, and choose BE most often when they detect an endpoint, that is, when the endpoint is [-inherent]. However, the trees of the German children differ from those of the Dutch children in one respect. For both groups of German children, [actor] shows up as a factor within the [+inherent] [+telic] verbs, that is, non-locomotion verbs with endpoints (nonsense verbs parallel to stretch or disappeal1. For the Dutch children, [actor] emerged only in the 7-8-year olds, and in a different set of verbs. The fact that this factor occurs in both groups of German children in the same verbs but not in the Dutch children in the same way, suggests that the effect for the German children is not an accident, and that it may relate to the difference in the two adult systems. Exactly how, though, is a question for further research.

12.7.

CONCLUSIONS

What do these results tell us about the questions we started with? First, syntactic unaccusativity is indeed determined by meaning in both German and Dutch. Two semantic factors appear to determine unaccusativity-[telicity] and [actor]. Subjects

352

Randall van Hout, Weissenborn, and Baayen

use the Telicity Linking Rule for verbs with detectable endpoints, classifying them as unaccusative. They also sometimes use the Actor Linking Rule to classify verbs with detectable actors as unergative. When both an endpoint and an actor are present for a given verb, subjects classify the verb as unaccusative. So the Telicity Linking Rule appears to take priority over the Actor Linking Rule. We proposed that this was related to the geometry of their Conceptual Structure representations. Looking more closely at the telicity factor, all subjects find it easier to detect an endpoint for a verb if it is expressed in an explicit PP (our [-inherent] cases). When the syntactic PP cue was lacldng (the [+inherent] cases), adults are more adept than either group of children at deducing an endpoint from the scene alone. With respect to [actor], while all subjects use the Actor Linking Rule to link at least a subset of the [+actor] verbs as unergative, for the Dutch 4-5-year-old children this is only a trend, not significant, as it is for all the other subjects. Why this is the case deserves further investigation. Although we did not predict any differences between Dutch and German (since we did not test any examples such as dance around the room, which would be unergative in Dutch but unaccusative in German), nevertheless, we do see differences in how our adult subjects classify verbs for scenes that are [+locomotion], for example, a hat shuffling around on a book, or Bert bouncing around on his head on a box. These cases are syntactically parallel to dance in the room, which has no [+locomotion] PP and no [+telic] PP either, so we expected unergative responses-which our Dutch adults gave. German adults, however, classify these verbs as unaccusative, presumably using the locomotion information in the scene alone. None of the German children do this. Like all of the Dutch subjects, they require a syntactically detectable endpoint in order to classify a verb as unaccusative. This suggests that German and Dutch both have a Telicity Linking Rule, used even by 4-year olds, and that, on the basis of positive evidence, Germans Locomotion Linking Rule is acquired later.I2 Taken all together, these results may be the beginnings of evidence that an event's telicity and the presence or absence of an actor are two semantic universals for determining unaccusativity. Studies on other languages, and more refined studies on Dutch and German, are waiting to be done. U See Randall (1990, 1992) for a proposal about how a learner can 'unlearn' an incorrect rule in their grammar solely from positive evidence.

Acquiring Unaccusativity

353

Appendix: Scenes and corresponding cells +telic, +inherent, +actor (cell I) A Bert straightens up under a picture. B Oscar shrivels up into himself on a red carpet (Le., he makes himself into a ball). +telic, +inherent, -actor (cell 5) C White blinds dose beside Ernie, when Ernie pulls the cord. D A blue balloon deflates in Bert's hand. -telic, +inherent, +actor (cell 3) E Ernie 'grimaces' (moves his mouth back and forth in a grimace), while sitting on a book. F Ernie makes scissor-motions with his hands inside a blue ring, behind his back. -telic, +inherent, -actor (cell 7) G A little ball and top jump around in a black saucer after a hand spins the top. H A green pitcher dangles on a rubber band held by a hand. +telic, -inherent, +actor (cell 2) I Bert 'mouths' his way along a table and into a paper bag at the far end. J Ernie walks on his arms stuck into blue tubes, onto a pile of books. +telic, -inherent, -actor (cell 6) K A red block flies off of a shovel when a fist hits the handle. L A yellow block tolls into a house when one end of the xylophone it is sitting on is lifted. -telic, -inherent, +actor (cell 4) M Some feet move back and forth on a grey floor, twisting in and out. N Bert bounces on his head on a purple box. -telic, -inherent, -actor (cell 8) o A black hat shuffles around on a book while a hand shalces the book. P A chocolate easter egg spins around on a dish after a hand gives it a spin.

References

ABRAHAM, W. (2000). 'The aspect-case typology correlation: perfectivity and Burzio's Generalization'. In E. Reuland (ed.), Arguments and Case: Explaining Burzio's Generalization. Amsterdam: John Benjamins: 131-93. ABUSCH, D. (1986). 'Verbs of change, causation and time'. Report 86-15°. Stanford, Calif.: CSLI Standford University. ACKEMA, p. (1995). 'Syntax below zero'. OTS Publications, University of Utrecht. -(2000). Issues in Morphosyntax. Amsterdam: John Benjamins. -and SCHOORLEMMER, M. (1994). 'The middle construction and the syntax-semantics interface'. Lingua, 93: 59-90. ---(1995). 'Middles and nonmovement'. Linguistic Inquiry, 2612: 173-97. ALEXIADOU, A. (1996). 'Aspectual restrictions on word order'. Folia Linguistica, 301I-2: 36-46. --(2001). Functional Structure in Nominals: Nominalization and Ergativity. Amsterdam: John Benjamins. -and ANAGNOSTOPOULOU, E. (1997). 'Tests for unaccusativity in a language without tests for unaccusativity'. Greek Linguistics 97: Proceedings ofthe Third International Conference on

Greek Linguistics: 23-31. ---(1999). 'Non-active morphology and the direction of transitivity alternations'. In P. Tamanji, M. Hirotani, and N. Hall (eds.), Proceedings ofNELS 29. ALSINA, A. (1996). The Role ofArgument Structure in Grammar. CLSI Publications. ANAGNOSTOPOULOU, E. (2001). 'On the distinction between verbal and adjectival passives: evidence from Greek'. MS. University of Crete. --and EVERAERT, M. (1999). 'Towards a more complete typology of anaphoric expressions'. Linguistic Inquiry, 30: 97-II9. ANDERSON, S. (1992). Amorphous Morphology. Cambridge: Cambridge University Press. ANDREWS, A. (1982). 'The representation of Case in Modern Icelandic'. In Bresnan (1982: 4275°3). ARAD, M. (1998a). 'VP-structure and the syntax-lexicon interface'. MIT Occasional Paper in

Linguistics, 16. --(1998b). 'Are unaccusatives aspectually characterized? (and other related questions)'. In H. Harley (ed.), Papers fi'om the UPenn-MIT round table on argument structure and aspect. MIT Working Papers in Linguistics, 32: 1-20. -(1999). 'On the nature of v: evidence from object experiencer verbs'. Paper presented at GLOW 22, Berlin. -(2001). 'Locality constraints on the interpretation of roots: the case of Hebrew denominal verbs'. MS. University of Geneva.

References ARENDS, J., MUYSKEN, P. c., and SMITH, N. (eds.) (1995). Pidgins and Creoles: An Introduction. Amsterdam: John Benjamins. ARNOTT, D. W (1956). 'The middle voice in Fula'. Bulletin ofthe School of Oriental andAfrican Studies, 28/1: 130-44. -(1970). The Nominal and Verbal Systems ofFula. Oxford: Oxford University Press. BABBY, 1. (1980). Existential Sentences and Negation in Russian. Ann Arbor: Karoma Publishers. BABYONYSHEV, M., FEIN, R., GANGER, J., PESETSKY, D., and WEXLER, K (2001). 'The maturation of grammatical principles: evidence from Russian unaccusatives'. Linguistic Inquiry, 32: 1-44BACH, E., and PARTEE, B. H. (1980). 'Anaphora and Semantic Structure'. In J. I
References

357

BERRUTO, G. (1987). Sociolinguistica dell'italiano contemporaneo. Florence: La Scuola. BHATT, R. (1999). 'Covert Modality in Non-Finite Contexts'. Doctoral dissertation, University of Pennsylvania. BICKERTON, D., and IATRIDOU, S. (1987). 'Verb serialization and empty categories'. MS. University of Amsterdam. BIERWISCH, M. (1996). 'Fragen zum Beispiel'. In G. Harras and M. Bierwisch (eds.), W'enn die Semantik arbeitet. Tiibingen: Niemeyer: 361-78. -(1997). 'Middles in German'. MS. Projektgruppe Strukturelle Grammatik, Berlin. BOLOTSKY, S. (1978). 'Word formation strategies in the Hebrew verb system: denominative verbs'. Ajroasiatic Linguistics, 5/3: 1-26. BORER, H. (1984). 'The projection principle and rules of morphology'. NELS 14,16-33. -(ed.) (1986). The Syntax o/Pronominal Clitics. Syntax and Semantics, xix. San Francisco: Academic Press. -(1990). 'V + ing: it walks like an adjective, it talks like an adjective'. Linguistic Inquiry, 21ir: 95-103. -(1991). 'The causative-alternation: a case study in parallel morphology'. The Linguistic Review, 8: II9-58. -(1994) . 'The projection of arguments'. In E. Benedicto and J. Runner (eds.), Functional Projectiom, University of Massachusetts Occasional Papers, 17: 19-47. -(1996). 'Passives without theta grids'. MS. University of Massachusetts. -(1998). 'Deriving passives without theta-grids'. In Lapointe, Farrell, and Brentari (1998: 60-99). -(2000). 'Exo-skeletal vs. endo-skeletal explanations: syntactic projections and the lexicori'. To appear in Explanation in Linguistics. -(2003). 'Computing argument structure'. In J. Shimron (ed.), Language Processing and Language Acquisition in a Root-BasedMorphology. Amsterdam: John Benjamins: 321-62. -(forthcoming). Structuring Seme. Oxford: Oxford University Press. -and GRODZINSKY, Y. (1986). 'Syntactic eliticization and lexical eliticization: the case of Hebrew dative elitics'. In Borer (1986: 175-217). BORIK, OLGA (1995). 'SintalQ;ieeskij priznalc neakkuzativnosti glagola'. MA thesis, Moscow State University. BORoVIKoFF, N. L. (2001). 'Unaccusativity and Movement in Russian: Integrating Formal Syntax and Discourse Functions'. Dissertation, Indiana University. BOUCHARD, D. (1984). On the Contento/Empty Categories. Dordrecht: Foris. BOWERMAN, M. (1982). 'Evaluating competing linguistic models with language acquisition data: Implications of developmental errors with causative verbs'. Semantica, 3: 1-73. BREIMAN, L., FRIEDMAN, J., OLSHEN, R., and STONE, C. (1984). Classification and Regression Trees. Belmont, Calif: Wadsworth International Group. BREKKE, M. (1988). 'The experiencer constraint'. Linguistic Inquiry, 1912: 169-80. BRESNAN, J. (ed.) (1982). The Mental Representation 0/ Grammatical Relations. Cambridge, Mass.: MIT Press. -(2001). Lexical-Functional Syntax. Oxford: Blackwell. -and ZAENEN, A. (1990). 'Deep unaccusativity in LFG'. In Dziwirek et al. (1990: 45-57). BURING, D. (1997). The Meaning o/Topic and Focus: The 59th Street Bridge Accent. London: Routledge. BURzIO, L. (1981). 'Intransitive Verbs and Italian Auxiliaries' . Ph.D. dissertation, MIT. -(1986). Italian Syntax: A Government and Binding Approach. Dordrecht: D. Reidel. BYRNE, F. X. (1987). Grammatical Relations in a Radical Creole. Amsterdam: John Benjamins. CARRIER,J., and RANDALL, J. (1993). 'Lexical mapping'. In Reuland and Abraham (1993: II9-42).

References CARRIER, J., and RANDALL, J. (forthcoming). From Conceptual Structure to Syntax. CARTER, R. J. (1988). 'On linking: papers by Richard Carter'. In B. Levin and C. Tenny (eds.), Lexicon Project Working Papers, 25. Cambridge, Mass.: Center for Cognitive Science. CENNAMO, M. (1999). 'Late Latin pleonastic reflexives and the Unaccusative Hypothesis'. Transactions o/the Philological Society, 97. CENTINEO, G. (1986). 'A lexical theory of auxiliary selection in Italian'. Davis Working Papers in Linguistics, I: 1-35. CHIERCHIA, G. (1985). 'Formal semantics and the grammar of predication'. Linguistic Inquiry, 16/3: 417-43. --(1988). 'Dynamic generalized quantifiers and donkey anaphora'. In M. Krifka (ed.), Genericity in Natural Language. University of Tubingen Technical Report. -(1989). 'A semantics for unaccusatives and its syntactic consequences'. MS. Cornell University. --(1988). 'Semantics and property theory'. Linguistics and Philosophy, U/3: Z61-302. CHOMSKY, N. (1981). Lectures on Government and Binding. Dordrecht: Foris. -(1986). Knowledge o/Language, its Nature, Origin and Use. New York: Praeger. -(1995). The Minimalist Program. Cambridge, Mass.: MIT Press. --(2000). 'Minimalist inquiries: the framework' . In R. Martin, D. Michaels, and J. Uriagereka (eds.), Step by Step: Essays on Minimalist Syntax in Honor 0/ Howard Lasnik. Cambridge, Mass.: MIT Press: 89-156. --(2001). 'Derivation by phase'. In Kenstowicz (2001: I-52). CINQUE, G. (1988). 'On si constructions and the theory of Arb'. Linguistic Inquiry, 19/4: 52181. -(1989). 'On embedded verb second clauses and ergativity in German'. In D. Jaspers, W. Klooster, Y. Putseys, and p. Seuren (eds.), Sentential Complementation and the Lexicon. Studies in Honour o/Wim de Geest. Dordrecht: Foris: 77-96. -(1990). 'Ergative adjectives and the lexicalist hypothesis'. Natural Language and Linguistic Theory, 8: 1-39. -(1999). Adverbs and Functional Heads: A Cross-linguistic Perspective. Oxford: Oxford University Press. CLARK, E. (1982). 'The young word maker: a case study of innovations in the child's lexicon'. In E. Wanner and L. Gleitman (eds.), Language Acquisition: The State 0/ the Art. Cambridge: Cambridge University Press. --and CLARK, H. (1979). 'When nouns surface as verbs'. Language, 55: 767-8u. COCCHI, G. (1994). 'An explanation of the split in the choice of perfect auxiliaries'. Probus, 6: 87-102. COHEN, A (1996). 'Think Generic: The Meaning and Use of Generic Sentences'. Ph.D. dissertation, Carnegie Mellon University. COLLINS, C. (1997). 'Argument sharing in serial verb constructions'. Linguistic Inquiry, 28/3: 461-97. -(1997). Local Economy. Cambridge, Mass.: MIT Press. COMRIE, B. (1976). Aspect: An Introduction to the Study o/Verbal Aspect and Related Problems. Cambridge: Cambridge University Press. COOPMANS, P., EVERAERT, M., and GRIMSHAW, J. (eds.) (2000). Lexical Specification and Insertion. Amsterdam: John Benjamins. CORNELIS, L., and VERHAGEN, A. (1995). 'Does Dutch really have a passive?' In M. den Dikken and K. Hengeveld (eds.), Linguistics in the Netherlands I995, Amsterdam: John Benjamins. CORVER, N. (1997). 'The internal syntax of the Dutch extended adjectival projection'. Natural Language and Linguistic Theory, 1512: 289-368.

References

359

CUMMINS, S. (1996). 'Meaning and Mapping'. Ph.D. dissertation, University of Toronto. DE Hoop, H. (1992). 'Case Configuration and Noun Phrase Interpretation'. Doctoral dissertation, Groningen University. [Publ. 1996 by Garland, New York]. DEN BESTEN, H. (1982). 'Some remarks on the ergative hypothesis'. Groninger Arbeiten zur germanisctischen Linguistik, 21: 61-81. -(1991). 'Serialisation in Europe: Norwegian empty object constructions and the analysis of serialisation'. MS. University of Leiden. -(1994). 'Auxiliaries and participles'. Proceedings of NELS 24,65-79. GLSA, Amherst. -(1997). 'The syntax of possession and the verb have'. Lingua, 101: 129-50. -and SYBESMA, R. (1998). 'Take serials light up the middle'. Paper presented at GLOW 21, Tilburg University. DI SCIULLO, A.-M., and WILLIAMS, E. (1987). Word. Cambridge, Mass.: MIT Press. DIMITR1ADIS, A., SIEGEL, L., SUREK-CLARK, C. and WILLIAMS, A. (eds.) (1997). Proceedings of the 2Ist Penn Linguistics Colloquium, UPenn Working Papers in Linguistics: 201-25. DONADIO, P. (1996). '.cipotesi inaccusativa e i verbi di movimento in italiano e in francese'. MS. Universita di Napoli. DORON, E., and A. MITTWOCH (1987). 'Negative polarity in Hebrew'. Talk given at the second IATL conference, Bar-Ilan University. DoWTY, D. (1979). WordMeaningandMontague Grammar. Dordrecht: D. Reidel. -(1982). 'Grammatical relations and Montague Grammar'. In P. Jacobson and G. Pullum (eds.), The Nature o/Syntactic Representation. Dordrecht: D. Reidel, 79-130. -(1991). 'Thematic proto-roles and argument selection'. Language, 67/3: 547-619. DZIWIREK, K., FARELL, P., and MEJlAs-BIKANDI, E. (eds.) (1990). Grammatical Relations: A Cross-Theoretical Perspective. Stanford: CSLI Publications. EHRICH, V. (1992). 'Verbbedeutung und Verbgrammatik: Transportverben im Deutschen'. In E. Lang and G. Zifonun (eds.), Deutsch---Typologisch. Berlin: Mouton de Gruyter: 229-60. EISENBERG, P. (1999). Grundriss der deutschen Grammatik, ii. Del" Satz. Stuttgart: Metzler. -(1989). 'Perfektbildung und Ergativparameter im Deutschen'. In J. Buscha and J. Schroder, Linguistische und didaktische Grammatik: Beitriige zu Deutsch als Fremdsprache. VEB Verlag Enzyldopadie Leipzig: II2-28. EMBICK, D. (1995). "'Mobile Inflections" in Polish'. Proceedings of NELS 25, 127-42. -(1997). 'Voice and the Interfaces of Syntax'. Doctoral dissertation, University ofPennsylvania. -(1998). 'Voice systems and the syntax-morphology interface'. MIT Working Papers in Linguistics, 32: 41-72. -(2000). 'Features, syntax and categories in the Latin perfect'. Linguistic Inquiry, 311z: 185-230. EVERAERT, M. (1982). 'A syntactic passive in Dutch'. Utrecht Working Papers in Linguistics, II: 37-73-(1986). The Syntax ofRejlexivization. Dordrecht: Foris. --(1995). 'Auxiliary selection in idiomatic constructions'. In L. Nash and G. Tsoulas (eds.), Actes du premier colloque Langues et Grammaire. University Paris-8, Paris. -(1996). 'The encoding of the lexical semantic structure of verbs: the case of auxiliaty selection in idioms'. In E. Weigand and F. Hundsnurscher (eds.), Lexical Structures and Languages Use. Tiibingen: Max Niemeyer. FAGAN, S. (1988). 'The English middle'. Linguistic Inquiry, 191z: 181-203. -(1992). The Syntax and Semantics ofthe Middle Constructions. Cambridge: Cambridge University Press. FANSELOW, G. (1985). Deutsche Verbalprojektionen und die Frage del" Universalitiit konfigurationaler Sprachen. Doctoral dissertation, University of Passau.

References FANSELOW, G. (1992). 'Ergative Verben und die Struktur des deutschen Mittelfelds'. In L. Hoffman (ed.), Deutsche Syntax: Ansichten undAussichten. Berlin: Mouron de Gruyter: 276-303. FILLMORE, c., and KAy, P. (I997). 'The formal architecture of Construction Grammar'. MS. University of California, Berkeley. Fox, D. (1993). 'A modification of Reinhart and Reuland's "reflexivity'''. MS. MIT. FRANKS, S. (1995). Parameters ofSlavic Morphosyntax. Oxford University Press. FRIEDEMANN, M. A., and SILONI, T. (1997). 'Agrobject is not AgrpmidpJe'. The Linguistic Review, 14: 69-96. GARTNER, H.-M., and STEINBACH, M. (I997): 'Anmerkungen 2ur Vorfeldphobie pronominaler Elemente.' In E-J. d'Avis and U. Lutz (eds.), Zur Satzstruktur des Deutschen. Arbeitspapiere des SFB 340, 90. Stuttgart and Tiibingen, 1-30. --(2000). 'What do reduced pronouns reveal about the syntax of Dutch and German?' LIP, 9: I-5 6. GERRITSEN, N. (1990). Russian Reflexive Verbs; in Search of Unity in Diversity. (Studies in Slavic and General Linguistics 15). Amsterdam: Rodopi. GLEITMAN, L. (1990)' 'The structural sources of verb meanings', Language Acquisition, I: 3-56. --(I995). 'When prophesy fails: how do we discard our theories oflearning'? Keynote address at the 20th Annual Boston University Conference on Language Development. Boston, November I995. GOLDBERG, A. (1995). Constructions: A Consti1tction Grammar Approach to Argument Structure. Chicago: University of Chicago Press. GREUDER, and BUTT, M. (eds.) (1998). The Projection ofArguments. Stanford: CSLI. GREWENDORF, G. (I988). Aspekte der deutschen Syntax. Tiibingen: Gunter Narr. -(1989). Ergativity in German. Dordrecht: Foris. -(2002). Minimalistische Syntax. Tiibingen: Francke. GRIMSHAW, J. (I98I). 'Form, function and the language acquisition device'. In C. L. Baker and J. J. McCarthy (eds.), The Logical Problem ofLanguage Acquisition. Cambridge, Mass.: MIT Press. -(I982). 'On the lexical representation of romance reflexive ditics'. In J. Bresnan (1982: 87148). -(1987). 'Unaccusatives: an overview'. Proceedings ofNELS, 17: 244-59. -(1990). Argument Structure. Cambridge, Mass.: MIT Press. --and PINKER, S. (1990). 'Using syntax to deduce verb meaning'. Paper presented to the 15th Annual Boston University Conference on Language Development, November 1990. GROENENDI]K, J., and STOKHOF, M. (1987). 'Dynamic Montague Grammar'. MS. University of Amsterdam. Paper presented at the Stuttgart Conference on Discourse Representation Theory. GROPEN, J., PINKER, S., HOLLANDER, M., and RICHARD, G. (1991). 'Affectedness and direct objects: the role oflexical semantics in the acquisition of verb argument structure'. In Levin and Pinker (I991: 153-96). HAIDER, H. (1985). 'Uber sein oder nicht sein: zur Grammatik des Pronomens sich'. In W. Abraham (ed.), Erkliire-nde Syntax des Deutschen. Tiibingen: Gunter Narr, 223-54. -(1990). 'Topicalization and other puzzles of German syntax'. In G. Grewendorf and W. Sternefeld (eds.), Scrambling and Barriers. Amsterdam: John Benjamins: 92-II2. -(I993). Deutsche Syntax-generativ. Tiibingen: Gunter Narr. -and RINDLER-SCH]ERVE, R. (1987). 'The Parameter of auxiliary selection: Italian-German contrasts'. Linguistics, 25: 1029-55. HALE, K., and KEYSER, S. J. (1986). 'Some transitivity alternations in English'. Lexical Project Working Paper, 7. Center for Cognitive Science, MIT.

w.,

References ---(1993). 'On argument structure and the lexical expression of syntactic relations'. In Hale and Keyser (1993: 53-109). --(eds.) (1993). The View from Building 20. Cambridge, Mass.: MIT Press. --(1998). 'The basic elements of argument structure'. MIT Working Papers in Linguistics, 32: 73-u8 . HALLE, M., and MARANTZ, A. (1994). 'Some key features of distributed morphology'. In A. Carnie, H. Harley, and T. Bures (eds.), Papers on Phonology and Morphology, MIT Working Papers in Linguistics, 21: 275-88. --(1993). 'Distributed Morphology and the pieces of inflection'. In Hale and Keyser (1993: III-76). HARDY, D. E. (1994). 'Middle voice in Creek'. IntemationalJoumal ofAmerican Linguistics, 601r: 39-68. HARLEY, H. (1995). 'Subject, Events and Licensing', Ph.D. dissertation, MIT. HARRIS, A. c. (1981). Georgian Syntax: A Study in Relational Grammar. Cambridge: Cambridge Universiry Press. HASPELMATH, M. (1990). 'The grammaticization of passive morphology'. Studies in Language, 14: 25-7 2. HAYES, B. (1990). 'Precompiled phrasal phonology'. In S. Inkelas and D. Zec (eds.), The SyntaxPhonology Connection. Chicago: Universiry of Chicago Press: 85-108. HElM, 1. (1982). 'The Semantics of Definite and Indefinite NPs'. Ph.D. dissertation, Universiry of Massachusetts. HIGGINBOTHAM, J. (1985). 'On semantics'. Linguistic Inquiry, 16/4: 547-93. HIGGINS, R. (1973). 'The Pseudo-Cleft Construction'. Ph.D. dissertation, MIT. HIRAKAWA, M. (2001). 'L2 Acquisition of Japanse Unaccusative Verbs'. Studies in Second LanguageAcquisition, 23: 221-45. HOEKSTRA, T. (1984). Transitivity: Grammatical Relations in Govemment-Binding Theory. Dordrecht: Foris. -(1988). 'Small clause results'. Lingua, 74: 101-39. -(1999). 'Auxiliary Selection in Dutch'. Natural Language and Linguistic Theory, 171r: 67-84. -and MULDER, R. (1990). 'Unergatives as copular verbs: locational and existential predication'. The Linguistic Review, 7: 1-79. --and ROBERTS, 1. (1993). 'Middle constructions in Dutch and English'. In Reuland and Abraham (1993: 183-220). HORNSTEIN, N. (1992). LOT Class lectures, Universiry ofNijmegen. lATRIDOU, S., PANCHEVA, R., andANAGNOSTOPOULOU, E. (2001). 'Observations about the form and meaning of the perfect'. In Kenstowicz (2001: 189-238). IWATA, S. (1999). 'On the status of implicit arguments in middles'. Journal of Linguistics, 35: 527-53· JACKENDOFF, R. (1990). Semantic Structure, Cambridge, Mass.: MIT Press. JACOBS, J. (1993). 'Integration'. In M. Reis (ed.), Wortstellung und InJormationsstruktur. Tiibingen: Niemeyer, 63-117. --YON STECHOW,A., STERNEFELD, w., and VENNEMANN, T. (eds.), Syntax. Ein internationales Handbuch del' zeitgeniissischen Forschung, i. Berlin: Mouton de Gruyter. JAEGGLI, O. (1986). 'Passive'. Linguistic Inquiry, 17: 587-622. JANSEN, B., KOOPMAN, H., and MUYSKEN, P. C. (1978). 'Serial verbs in the Creole languages'. Amsterdam Creole Studies, 2: 125-59. Institute of General Linguistics, Universiry of Amsterdam. JUFFS, A. (1996). Leamability and the Lexicon: Theories and Second Language Acquisition Research. Amsterdam: John Benjamins.

References KAMp, H. (1981). 'A theory of truth and discourse representation. In J. Groenendijk, T. Janssen, and M. Stokhof (eds.), Formal Methods in the Study 0/ Language, Mathematisch Centrum, Amsterdam. KARTTUNEN, L. (1977). 'The syntax and semantics of questions'. Linguistics and Philosophy, I: 3-44· KATHOL, A (1991). 'Unaccusative mismatches in German'. In J. Bobaljik and T. Bures (eds.), Proceedings o/the 3rd Student Linguistic Conference. MIT Working Paper in Linguistics, II530 . KAYNE, R. (1975). French Syntax: The Transformational Cycle. Cambridge, Mass.: MIT Press. -(1988). 'Romance selsi'. Paper presented at GLOW II, Budapest. GLOW Newsletter, 20. --(1993). 'Toward a modular theory of auxiliary selection. Studia Linguistica, 47: 3-31. KEENAN, E. (1985). 'Passive in the world's languages'. In T. Shopen (ed.), Language Tjpology and Linguistic Description, i. Cambridge: Cambridge University Press: 243-81. -and FALTZ, L. (1985). Boolean Semantics for Natural Language. Dordrecht: D. Reidel. KELLER, E, and SORACE, A (2002). 'Gradient auxiliary selection in German: an experimental investigation'. Journal 0/Linguistics, 39/r: 57-108. KEMMER, S. (1993). The Middle Voice. Amsterdam: John Benjamins. KENSTOWICZ, M. (ed.) (2001). Ken Hale: A Life in Language. Cambridge, Mass.: MIT Press,. KEYSER, S. J., and ROEPER, T. (1984). 'On the middle and ergative constructions in English'. Linguistic Inquiry, 15/3: 381-416. KrPARSKY, P. (1982). 'Lexical morphology and phonology'. In I.-S. Yang (ed.), Linguistics in the Morning Calm: Selected Essays from SICOL-I98I. Seoul: Hanshin: 1-91. -(1998). 'Partitive case and aspect'. In Greuder and Butt (1998: 265-307). KrTAGAWA, Y. (1986). 'Subjects in Japanese and English'. Doctoral dissertation, University of Massachusetts, Amherst. KLAIMAN, M. H. (1991). Grammatical Voice. Cambridge: Cambridge University Press. -(1992). 'Middle verbs, reflexive middle constructions, and middle voice'. Studies in Language, 16/r: 35-62. KOLLIAKOU, D. (1995). 'Definites and Possessives in Modern Greek: An HPSG Syntax for Noun Phrases'. Ph.D. dissertation, University of Edinburgh. KOOPMAN, H., and SPORTICHE, D. (1991). 'The position of subjects'. Lingua, 85: 2II-58. KRATZER, A. (1994). 'The event argument and the semantics of voice'. MS. University of Massachusetts. -(1996). 'Severing the external argument from its verb'. In J. Rooryck and L. Zaring (eds.), Phrase Structure and the Lexicon. Dordrecht: Kluwer: 109-37. -(2000). 'Building statives'. Berkeley Linguistic Society, 26. KruFKA, M. (1988). 'The relational theory of generics'. In M. Krifka (ed.), Genericity in Natural Language. University ofTiibingen technical report. -(1989). NominalreJerenz und Zeitkonstitution: Zur Semantik von Massentermen, Pluraltermen undAspektklassen. Munich: Wilhelm Fink Verlag. [Revised 1986 doctoral dissertation, University of Munich]. -(1992). 'Thematic relations as links between nominal reference and temporal constitution. In Sag and Szabolcsi (1992: 29-53). --PELLETIER, E J., CARLSON, G. N., TER MEULEN, A, CHIERCHIA, G., and LINK, G. (1995). 'Genericity. An introduction. In G. N. Carlson and F. J. Pelletier (eds.), The Generic Book. Chicago: University of Chicago Press: 1-124. WELLE, M. (1992). 'Change of state and valency'. Journal o/Linguistics, 28: 375-414. UKOFF, G. (1965). 'Irregularity in Syntax'. Ph.D. thesis, Indiana University. (Publ. 1970 by Holt, Rinehart and Winston).

References LANDAU, B., and GLEITMAN, L. (1985). Language and Experience. Cambridge, Mass.: Harvard University Press. LANDAU, (2001). 'The Locative Syntax of Experiencers'. MS. Ben Gurion University. LApOINTE, S., BRENTARI, D., and FARRELL, P. (eds.) (1998). Morphology and its Relation to Phonology and Syntax. Stanford, Calif.: CSLI Publications. LARsoN, R. (1988). 'On the double object construction'. Linguistic Inquiry, 19/J: 335-91. -(1990). 'Double objects revisited: reply to Jackendoff'. Linguistic Inquiry, 21/4: 589-632. -(1991). 'Some issues in verb serialization'. In C. Lefebvre (ed.), Serial Verbs: Grammatica~ Comparative and Cognitive Approaches. Amsterdam: John Benjamins: 185-210. -and LEFEBVRE, C. (1991). 'Predicate clefting in Haitian Creole'. Paper presented at NELS 21, UQAM, Montreal. LASNIK, H. (1988). 'Subject and the theta-criterion'. Natural Language and Linguistic Theory, 6: 1-17. LEGENDRE, G. (1989). 'Unaccusativity in French'. Lingua, 79: 95-164. -and SORACE, A. (In press). 'Split intransitivity in French: an Optimality-Theoretic perspective'. In D. Godard (ed.), Les langues romanes: problemes de la phrase simple. Paris: Editions duCNRS. LENERZ, J. (1977). Zur Abfolge nominaler Satzglieder im Deutschen. Tiibingen: Niemeyer. LEVIN, B. (1993). English Verb Classes and Alternations. A Preliminary Investigation. Chicago: University of Chicago Press. -and PINKER, S. (eds.) 1991. Lexical and Conceptual Semantics. Oxford: Blackwell. -and RAPPAPORT HOVAV, M. (1986). 'The formation of adjectival passives'. Linguistic Inquiry, 17: 623-61. --(1988). 'An Approach to unaccusative mismatches'. Proceedings ofNELS I8. --(1988). 'Nonevent -er nominals: a probe into argument structure'. Linguistics, 26/6: 1067-83. ---(1992a). 'The lexical semantics of verbs of motion: the perspective from unaccusativity'. In Roca (1992: 247-69). --(199Zb). 'Unaccusativity: at the syntax-semantics interface', MS. Northwestern University and Bar Ilan University [a revised version of this manuscript was published as Levin and Rappaport Hovav 1995]. --(1994). 'A preliminary analysis of causative verbs in English'. Lingua, 92: 35-77. --(1995). Unaccusativity: At the Syntax-Lexical Semantics Intelface. Cambridge, Mass.: MIT Press. --(1996). From Lexical Semantics to Argument Realization. Cambridge: Cambridge University Press. LEVIN, L. (1986). 'Operations on Lexical Forms: Unaccusative Rules in Germanic Languages'. Ph.D. dissertation, MIT. LEVY, Y. (1988). 'The nature of early language: evidence from the development of Hebrew morphology'. In Y. Levy, M. Schlesinger, and M. D. S. Braine (eds.), Categories and Processes in Language Acquisition. Hillsdale, N.J.: Lawrence Erlbaum Associates. LIDZ, J. (1998). 'Valency in Kannada: evidence for interpretive morphology'. Penn Working

r.

r.

Papers in Linguistics. LIEBER, R., and H. BAAYEN (1997). 'A semantic principle of auxiliary selection in Dutch'. NaturalLanguage and Linguistic Theory, 15: 789-845. LONZI, L. (1985). 'Pertinenza della struttura tema-rema per l'analisi sintattica'. In H. Stammerjohann (ed.), Theme-Rheme in Italian. Tiibingen: Gunter Narr: 99-120. MCCAWLEY, J. (1968). 'The role of semantics in grammar'. In E. Bach and R. Harms (eds.), Universals ofLinguistic Theory. New York: Holt, Rinehart and Winston: 125-70.

References MCCLURE, W. (1990). 'A lexical semantic explanation for unaccusative mismatches'. In Dziwirek et al. (1990: 45-57). --(1993). 'Unaccusativity and "inner" aspect'. Proceedings ofWCCFL, II: 313-25. -(1995). Syntactic Projections ofthe Semantic ofAspect. Tokyo: Hitsujishobo. MCGINNIS, M. (1997). 'Reflexive binding and lethal ambiguity'. Proceedings ofWCCFL 16: 30 3-17. MAcKRIDGE, P. (1985). the Modern Greek Language. A Descriptive Analysis of Standard Modern Greek. Oxford: Oxford University Press. MALING, J. (2001). 'Dative. The heterogeneity of the mapping among morphological case, grammatical functions, and thematic roles'. Lingua, III: 419-64. MANNEY, L. (2000). Middle Voice in Modern Greek: Meaning and Function of an Inflectional Class. Amsterdam: John Benjamins. MARANTZ, A. (1984). On the Nature of Grammatical Relations. Cambridge, Mass.: MIT Press. -(1993). 'Implications of asymmetries in double object constructions'. In S. A. Mchombo (ed.), 7heoretical Aspects of Bantu Grammar, i. Chicago: University of Chicago Press: II349· --(1997). 'No escape from syntax: don't try a morphological analysis in the privacy of your own lexicon'. U Penn Working Papers in Linguistics, 41z: 201-25. MARCUS. G. (2000). 'Children's overregularization and its implication for cognition'. In P. Broeder and J. Murre (eds.), Models of Language Acquisition. Inductive and Deductive Approaches. Oxford: Oxford University Press: 154-76. MARTIN, J. (1991). 'The Determination of Grammatical Relations in Syntax'. Doctoral dissertation, UCLA. MAVROMANOLAKI, G. (2002). 'Unaccusative Verbs, the Causative-inchoative Alternation and Aspectual Properties'. Master thesis. University of Crete. MElBAUER, J. (1995). 'Wortbildung und Kognition: Dberlegungen zum deutschen -er-Suffix'. Deutsche Sprache, 23: 97-123. MITTWOCH, A. (1991). 'In defense ofVendler's achievements'. Belgian Journal ofLinguistics, 6: 71- 84. MONTRUL, S. (1997). 'Transitivity Alternations in Second Language Acquisition: A Crosslinguistic Study of English, Spanish and Turkish'. Ph.D. dissertation, McGill University. MULLER, G. (1999). 'Optimality, markedness, and word order in German'. Linguistics, 37/5: 777-818 . MUNRO, P. (1996). 'Valence arithmetic in the Tolkapaya lexicon'. In V. Golla (ed.), Proceedings ofthe Hokan-Penutian Workshop (I994, I99S), Survey of California and Other Languages Report 9. U.e. Berkeley: II3-29. MUYSKEN, P. e. (1988). 'Parameters for serial verbs'. In K. Demuth and V. Manfredi (eds.), Niger-Congo Syntax and Semantics, i. Bloomington: IULe. --(1989). 'Predication chains: case and argument status in Quecha and Turldsh'. Linguistic Inquiry, 20/4: 627-45. -and VEENSTRA, T. (1995). 'Serial verbs'. In Arends et al. (1995: 289-303). NAPOLI, D. J. (1988). Review of Burzio 1985. Language, 64: 130-42. NARASIMHAN, B., Dr-ToMASO, v., and VERSPOOR, e. (1996). 'Unaccusative or unergative? Verbs of manner of motion'. Quaderni del Laboratorio di Linguistics, 10: 132-6. NEELEMAN, A. (1993). 'Complex predicates'. Ph.D. dissertation, Utrecht University. -(1997). 'PP-complemems'. Natural Language and Linguistic 7heory, 15: 43-87. NEIDLE, e. (1989). the Role of Case in Russian Syntax. Dordrecht: Kluwer. NERBONNE, J. (1982). 'German impersonal passives: a non-structure-preserving lexical rule'. Proceedings ofWCCFL: 341-52.

References NOYER, R. (1997). Features, Positions and Affixes in Autonomous MOIphological Structure. New York: Garland. -(1998). 'Impoverishment theory and morphosyntactic markedness'. In Lapointe, Brentari, and Farrell (1998: 264-85). OUHALLA, J. (1988). 'The Syntax of Head Movement: A Case Study of Berber'. Ph.D. dissertation, University College London. PARSONS, T. (1988). 'The role of events in semantics'. MS. University ofIrvine. [Rev. edn. pub!. as Events in the Semantics ofEnglish. Cambridge, Mass.: MIT Press, 1990]. -(1990). Events in the Semantics of English: A Study in Subatomic Semantics. Cambridge, Mass.: MIT Press. PARTEE, B. (1965). 'Subject and Object inModern English'. Ph.D. dissertation. MIT. -and ROOTH, M. (1983). 'Generalized conjunction and type ambiguity'. In R. Bauerle, C. Schwarze, and A. von Stechow (eds.), Meaning, Use and Interpretation of Language. Berlin: Walter de Gruyter: 361-83. PERLMUTTER, D. (1978). 'Impersonal passives and the unaccusative hypothesis'. Papers from the Annual Meeting ofthe Berkeley Linguistic Society, 4: 157-89. -(1989). 'Multiattachment and the unaccusative hypothesis: the perfect auxiliary in Italian'. Probus, I: 63-II9. -and POSTAL, P. (1984). 'The I-Advancement Exclusiveness Law'. In Perlmutter and Rosen (1984: 8I-I2 5)· -and ROSEN, C. (eds.) (1984). Studies in Relational Grammar, ii. Chicago: University of Chicago Press. PESETSKY, D. (1995). Zero Syntax. Experiencel'S and Cascades. Cambridge, Mass.: MIT Press. --(1982). 'Paths and Categories', Ph.D. dissertation, MIT. PINKER, S. (1984). Language Learnability and Language Development. Cambridge, Mass.: MIT Press. -(1989). Learnability and Cognition: the Acquisition of Argument Structure. Cambridge, Mass.: MIT Press. PINCJN, C. (1997). 'Verbs of motion in Polish, I: Paths and Processes'. In U. Junghanns and G. Zybarow (eds.), Formale Siavistik. Frankfurt am Main: Vervuert Verlag, 467-88. POLLARD, c., and SAG, 1. (1994). Head-Driven Phrase Structure Grammar. Chicago: University of Chicago Press. POLLOCK, J.-Y. (1989). 'Verb-Movement, Universal Grammar, and the Structure ofIP'. Linguistic Inquiry, 20/3: 365-420. PRIMUS, B. (1999). Cases and 7hematic Roles. Tiibingen: Niemeyer. PUSTEJOVSKY, J. (1991). 'The syntax of event structure'. Cognition, 41: 47-81. [Repr. in Levin and Pinker 1992]. -(1995). the Generative Lexicon. Cambridge, Mass.: MIT Press. --and BUSA, F. (1995). 'Unaccusativity and event composition'. In P. M. Bertinetto, V. Bianchi, J. Higginbotham, and M. Squartini (eds.), Temporal Reference, Aspect andActionality. Torino: Roserberg & SeIlier: 159-78. RADFORD,A. (1997). Syntax:AMinimalistIntroduction. Cambridge: Cambridge University Press. RAMCHAND, G. (1997). Aspect and Predication: the Semantics ofArgument Structure. Oxford: Oxford University Press. RANDALL, J. (1990). 'Catapults and pendulums: the mechanics oflanguage acquisition'. Linguistics, 28: 1381-406. --(1992). 'The Catapult Hypothesis: an approach to unlearning'. In J. Weissenborn, H. Goodluck, and T. Roeper (eds.), 7heoretical Issues in Language Acquisition. Hillsdale, N.J.: Lawrence Erlbaum Associates: 93-138.

References RANDALL, J., VAN HOUT, A., WEISSENBORN, J., and BAAYEN, H. (1994). 'Approaching linking'. Talk presented at 18th Boston University Conference on Language Development, Boston, January; Langues et Grammaire I, Universite Paris-8. MS. Max Planck Institute and Northeastern University. - - - - - ( i n progress). 'Learning linking: acquiring unaccusativity in Dutch and German'. MS. Northeastern University. RApPAPORT HOVAV, M., and LEVIN, B. (1998). 'Building verb meanings'. In W. Geuder and M. Butt (eds.), The Projection 0/ Arguments: Lexical and Compositional Factors. Stanford: CSLI Publications, 97-134. --(2000). 'Classifying single argument verbs'. In Coopmans, Everaert, and Grimshaw (2000: 269-304). --(2001). 'An event structure account of English resultatives'. Language, 77/4: 766-97. REINHART, T. (1983). Anaphora and Semantic Interpretation. London: Croom Helm. --(1996). 'Syntactic effects oflexical operations: reflexives and unaccusatives'. UiL OTS Working Papers, Utrecht University. -(2000). 'The theta system: syntactic realization of verbal concepts'. UiL OTS Working papers, University of Utrecht. -(200ra). 'A synopsis of the theta system'. UiL OTS Working Papers, University of Utrecht. --(2001b). 'Experiencing derivations'. UtL OTS Working Papers, University of Utrecht. -and REULAND, E. (1993). 'Reflexivity'. Linguistic Inquiry, 24/4: 650--720. --and SILON1, T. (1999). 'Reflexive reduction: lexical and syntactic application'. Handout, talk given at the conference 'On the Syntax of Semitic Languages'. University of UrbanaChampaign, and at lATL 15, The University of Haifa. RE1S, M. (1973). 'Is there a rule of subject-to-object raising in German?'. Proceedings o/CLS, 9: 519-2 9. -(1987). 'Die SteHung der Verbargumente im Deutschen'. In 1. Rosengren (ed.), Sprache und Pragmatik. Stockholm: Almqvist &WikseH: 139-77. REULAND, E. (ed.) (2000). Arguments and Case: Explaining Burzio's Generalization. Amsterdam: John Benjamins. -and AnRAHAM, W (eds.) (1993). Knowledge and Language, ii. Lexical and Conceptual Structure. Dordrecht: Kluwer. -and REINHART, T. (1995). 'Pronouns, anaphors and case'. In H. Haider, S. Olsen, and S. Vikner (eds.), Studies in Comparative Germanic Syntax. Dordrecht: Kluwer: 241-68. RITTER, E., and ROSEN, S. (1998). 'Delimiting events in syntax'. In Greuder and Butt (1998: 97-134). RIvERO, M.-L. (1992). 'Adverb incorporation and the syntax of adverbs in Modern Greek'. Linguistics and Philosophy, 15: 289-331. RIzZI, L. (1986a). 'Null objects in Italian and the theory of pro'. Linguistic Inquiry, 1712: 50158. -(1986b). 'On chain formation'. In Borer (1986: 65-95). ROCA, 1. M. (ed.) (1992). Thematic Structure: Its Role in Grammar. Berlin: Mouton de Gruyter. ROEPER, T. (1987). 'Implicit arguments and the head-complement relation'. Linguistic Inquhy, 18/3: 267-310. ROHLFS, G. (1969). Grammatica storica della lingua italiana e dei slloi dialetti. Turin: Einaudi. ROOTH, M. (1992). 'A theory of focus interpretation'. Natural Language Semantics, 1: 75-n6. ROSEN, C. (1981). 'The Relational Structure of Reflexive Clauses: Evidence from Italian'. Ph.D. dissertation, Harvard University. [Pub!. 1988 by Garland, New York]. -(1984). 'The interface between semantic roles and initial grammatical relations'. In Perlmutter and Rosen (1984: 38-77).

References ROTHSTEIN, S. (1983). 'The Syntactic Forms of Predication', Ph.D. dissertation, MIT.

-(2000). 'What are incremental themes?' MS. Bar-Han University. ROUNTREE, S. C. (1972). 'Saramaccan tone in relation to intonation and grammar'. Lingua, 29: 308- 2 5. -(1992). 'Saramaccan grammar sketch'. Languages of the Guianas, 13. Paramaribo: Summer Institute of Linguistics. RUWET, N. (1972). Theorie syntaxique et syntaxe du franfais. Paris: Editions du Seuil. -(1991). Syntax and Human Experience. Chicago: University of Chicago Press. SAG, 1. and SZABOLCSI, A. (eds.) (1992). LexicalMatters. Stanford: CSL1. SCHACHTL, S. (1991). 'Der Akkusativ in der Medialkonstruktion des Deutschen'. In G. Fanselow and S. Felix (eds.), Strukturen und Merkmale syntaktischer Kategorien. Tiibingen: Gunter Narr: I04-2o. SCHEIN, B. (2002). 'Events and the semantic content of thematic relations'. In G. Preyer and G. Peter (eds.), Logical Form, Language and Semantic Content: On Contemporary Developments in the Philosophy ofLanguage and Linguistics. Oxford: Oxford University Press. SCHMITT, C. (1995). 'Aspect and the Syntax of Noun Phrases'. Ph.D. dissertation, University of Maryland. SCHOORLEMMER, M. (1995). 'Participial Passive and Aspect in Russian'. Ph.D. dissertation, Utrecht University. -(1997). 'The role of the internal argument in the Russian aspectual system'. In U. Junghanns and G. Zybatow (eds.), Formale Slavistik. Frankfurt am Main: Vervuert Verlag: 229-38. -(1998). 'Parnyj glagol kale glagol s leksieeski nefiksirovannym vidom' ['The paired verb as a verb with no lexically fixed aspect']. In M. Ju. Certkova (ed.), Tipologija vida: problemy, poiski, re enija. Moscow: Jazyki russkoj kul'tury: 422-9. SEBBA, M. (1987). The Syntax ofSerial Verbs. Amsterdam: John Benjamins. SELLS, p. (1987). 'Aspects oflogophoricity'. Linguistic Inquiry, 18/4: 445-79. SEUREN, P. A. M. (1991). 'The definition of serial verbs'. In F. X. Byrne and T. Huebner (eds.), Development and Structures ofCreole Languages. Amsterdam: John Benjamins: 193-205. SHLONSKY, u. (1987). 'Null and Displaced Subjects'. Ph.D. dissertation, MIT. -and DORON, E. (1992). 'Verb second in Hebrew'. In D. Bates (ed.), Proceedings ofthe Tenth

West Coast Conference on Formal Linguistics, 431-5. SILONI, T. (1995). 'On participial relatives and complementizer DO: a case study in Hebrew and French'. Natural Language and Linguistic Theory, 13/3: 445-87. -(1997). Noun Phrases and Nominalizations: The Syntax ofDPs. Dordrecht: Kluwer. -(zooz). 'Reciprocal verbs.' Proceedings ofIATL, 17. SIMPSON, J. (1983). 'Resultatives'. In L. Levin, M. Rappaport Hovav, andA. Zaenen (eds.), Papers in Lexical-Functional Grammar. Bloomington, Ind.: Indiana University Linguistics Club. SMITH, C. (1997). The Parameter ofAspect (znd edn.). Dordrecht: Kluwer. SNYDER, w., HYAMS, N., and CRISMA, P. (1995). 'Romance auxiliary selection with reflexive clitics: evidence for earely lmowledge of unaccusativity'. In Proceedings ofthe 26th Annual Child Language Research Forum, zz-47. SORACE, A. (199Z). 'Lexical Conditions on Syntactic Knowledge: Auxiliary Selection in Native and Non-Native Grammars ofItalian'. Ph.D. dissertation, University of Edinburgh. -(I993a). 'Incomplete vs divergent representations of unaccusativity in non-native grammars ofItalian'. Second Language Research, 9: z2-47. -(1993b). 'Unaccusativity and auxiliary choice in non-native grammars ofItalian and French: asymmetries and predictable indeterminacy'. Journal ofFrench Language Studies, 3: 71-93. -(1995a). 'Contraintes semantiques sur la syntaxe: l'acquisition de l'inaccusativite en italien Lz'. Acquisition et interaction en langue etrangere, 5: 79-1I3.

References SORACE, A. (1995b). 'Acquiring argument structures in a second language: the unaccusativel unergative distinction'. In L. Eubank, L. Selinker, and M. Sharwood Smith (eds.), The Current State ofInterlanguage. Amsterdam: John Benjamins: 153-75. --(2000). 'Gradients in auxiliary selection with intransitive verbs'. Language, 76: 859-90. --and CENNAMO, M. (1999). 'Semantic constraints on unaccusativity: evidence from Paduan'. MS. University of Edinburgh and University of Naples. ---(2000). 'Aspectual constraints on auxiliary choice in Paduan'o MS. University of Naples and University of Edinburgh. --and SHOMURA, Y. (2001). 'Lexical constraints on the acquisition of split intransitivity: evidence from L2 Japanese'. Studies in Second Language Acquisition, 2312: 221-45. --and VONK, W. (1998). 'Gradient effects of unaccusativity in Dutch'. MS. University of Edinburgh and Max Planck Institute for Psycho linguistics, Nijmegen. SPORTICHE, D. (1998). Partitions and Atoms of Clause Structure: Subjects, Agreement, Case and Clitics. London: Routledge. STEINBACH, M. (2002a). 'The ambiguity of wealc reflexive pronouns in English and German'. In W. Abraham and J.-w. Zwart (eds.), Studies in Comparative Germanic Syntax. Amsterdam: John Benjamins: 323-48.

-(2002b). Middle Voice: A Comparative Study in the Syntax-Semantics Inteiface of German. Amsterdam: John Benjamins. STOWELL, T. (1987). 'As So, Not So As'. MS. UCLA. --(1991). 'The alignment of arguments in Adjective Phrases'. In S. Rothstein (ed.), Perspectives on Phrase Structure: Heads and Licensing. Syntax and Semantics, xxv. New York: Academic Press: 105-35. STROIK, T. (1992). 'Middles and movement'. Linguistic Inquiry, 23!I: 127-37. --(1999). 'Middles and reflexivity'. Linguistic Inquiry, 30: II9-3I. TENNY, C. (1987). 'GrarnmaticalizingAspect and Affectedness'. Ph.D. dissertation, MIT. -(1992). 'The aspectual interface hypothesis'. In Sag and Szabolsci (1992: 1-27). -(1994). Aspectual Roles and the Syntax-Semantics Intnface. Dordrecht: Kluwer. THEOPHANOPOULOU-KoNTOU, D. (zooo). '-O/-me alternations in MG patient oriented constructions: anticausatives and passives'. Proceedings ofthe 20th Annual Meeting ofthe Linguis-

tics Dept. ofthe University ofThessaloniki. TORREGO, E. (1989). 'Unergative-unaccusative alternations in Spanish'. MIT Working Papers in lingUistics, 10: 253-72. TRAVIS, L. (1994). 'Event phrase and a theory offunctional categories'. In P. Koskinen (ed.), Proceedings ofthe I994 Annual Conference ofthe Canadian Linguistic Society: 559-70. -(1999). 'The syntax of achievements'. MS. McGill University. TSUJIMURA, N. (1990). 'Unaccusative mismatches in Japanese'. Eastern States Conference on Linguistics, 6: 264-76. -(1991). 'On the semantic properties of unaccusativity'. Journal ofJapanese Linguistics, 13: 91- II6. --(1994). 'Resultatives and motion verbs in Japanse'. Studies in the Linguistic Sciences, 24: 429-40 . TUTTLE, E. (1986). 'The spread of esse as universal auxiliary in central !talo-Romance'. Medioevo Romanzo, II: 229-87. USZKOREIT, H. (1987). 'Word order and constituent structure in German'. Stanford, Calif: CSLI Lecture Notes. VAINIKKA, A., and MALING, J. (1993). 'Is partitive case inherent or structural?' MS. University of Massachusetts, Amherst and Brandeis University.

Reftrences VAN GEENHOVEN, V. (1996). 'Semantic Incorporation and Indefinite Descriptions: Semantic and Syntactic Aspects of West Greenlandic Noun Incorporation'. Doctoral dissertation, University ofTiibingen. VAN HoUT, A. (1992). 'Linking and projection based on event structure'. MS. Tilburg: Tilburg University. -(1996). 'Event Semantics of Verb Frame Alternations: A Case Study of Dutch and its Acquisition'. Doctoral dissertation, Tilburg University.

-(1998). Event Semantics of Verb Frame Alternations: A Case Study ofDutch and its Acquisition. New York: Garland. --(2000). 'Event semantics in the lexicon-syntax interface: Verb frame alternations in Dutch and their acquisition'. In C. Tenny and J. Pustejovsky (eds.), Events as Grammatical Objects. Stanford, Cali£: CSLI: 239-82. --RANDALL, J., and WEISSENBORN, J. (1992). 'On the acquisition of unaccusatives'. Talk presented at the 17th Boston University Conference on Language Development, Boston. ------(1993). 'Acquiring the unergative-unaccusative distinction'. In M. Verrips and F. Wijnen (eds.), The Acquisition ofDutch. Amsterdam Series in Child Language Development, I: 79-120. University of Amsterdam. VAN VALIN, R. D., Jr. (1987). 'The unaccusative hypothesis vs. lexical semantics: syntactic vs. semantic approaches to verb classification'. Proceedings ofNELS I7. --(1990). 'Semantic parameters of split intransitivity'. Language, 66: 221-60. VAN VOORST, J. (1985). 'Unaccusativity revisited'. Cahiers linguistiques d'Ottowa, 14: 97-109. -(1988). Event Structure. Amsterdam: John Benjamins. VEENSTRA, T. (1994). 'The acquisition of functional categories: the Creole way'. In D. Adone and 1. Plag (eds.), Creolization and Language Change. Niemeyer, Tiibingen:99-II5. -(1996a). Serial Verbs in Saramaccan: Predication and Creole Genesis. The Hague: HAG. -(1996b). 'The pronominal system of Saramaccan'. In D. Veronique (ed.), Materiaux pour !'etude des classes grammaticales dans les langues creoles. Aix-en-Provence: Publications de l'Universite de Provence: 259-71. --(1996c). 'Serial verb constructions as secondary predicates'. CONSOLE Proceedings, 279-92. --(1998a). 'Basic argument structure: transitivity alternations in Saramaccan'. In J. Arends and A. Bruyn (eds.), Mengelwerk voor Muysken. Publicaties van het Instituut ATW; Universiteit van Amsterdam: 13-19. --(1998b). 'The syntax of passives in Saramaccan: architectural implications'. MS. Freie Universitat Berlin. -and DEN BESTEN, H. (1995). 'Fronting'. In Arends, Musken, and Smith (1995: 303-14). VENDLER, Z. (1967). Linguistics in Philosophy. Ithaca: Cornell University Press. VERKUYL, H. J. (1972). On the Compositional Nature ofAspect. Dordrecht: D. Reidel. -(1989). 'Aspectual classes and aspectual composition'. Linguistic and Philosophy, 12: 39-94-(1993). A Theory ofAspectuality. Cambridge: Cambridge University Press. VERRIPS, M. (1996). 'Potatoes Must Peel: The Acquisition of the Dutch Passive'. Doctoral dissertation, University of Amsterdam. VOGEL, R., and STEINBACH, M. (1998). 'The dative: an oblique case'. Linguistische Berichte, 173: 65-90. VON STECHOW, A. (1991). 'Current issues in the theory of focus'. In Jacobs, von Stechow, Sternefeld, and Vennemann (1993: 8°4-25). VON STECHOW, A. (1995). 'Lexical decomposition in Syntax'. In U. Egli, P. Pause, C. Schwartze, A. von Stechow, and G. Wienold (eds.), Lexical Knowledge in the Organization ofLanguage. Amsterdam: John Benjamins: 81-117.

370

References

WANNER, A. (1999). Verbklassijizierung und aspektuelle Alternationen im Englischen. Tilbingen: Niemeyer. WASOW, T. (1977). 'Transformations and the lexicon'. In P. Culicover, T. Wasow, and A. Akmajian (eds.), Formal Syntax. New York: Academic Press. WEHRLI, E. (1986). 'On some properties of French ditic se'. In Borer (1986: 263-83). WILLIAMS, E. (1981a). 'Argument structure and morphology'. The Linguistic Review, 1: 81-II4. --(1981b). 'On the notions "lexically related" and "head of a word"'. Linguistic Inquiry, 1212: 245-74· -(1983). 'Against small clauses'. Linguistic Inquiry, 1412: 287-308. WUNDERLICH, D. (1985). 'Uber die Argumente des Verbs'. Linguistische Berichte, 97: 183-227. --(1993). 'Diathesen (valency changing)'. In Jacobs, von Stechow, Sternefeld, and Vennemann (1993: 73 0-47). WURMBRANDT, S. (1998). 'Infinitives'. Ph.D. dissertation, MIT. YUAN, B. (1999). 'Acquiring the unaccusative distinction in a second language: evidence from English-spealdng learners ofL2 Chinese'. Linguistics, 37: 275-96. ZAENEN, A. (1988). 'Unaccusative verbs in Dutch and the syntax-semantics interface'. CSLI-

ReportI23. -(1993). 'Unaccusativity in Dutch: integrating syntax and lexical semantics'. In J. Pustejovsky (ed.), Semantics and the Lexicon. Dordrecht: Kluwer: 129-61. ZRIBI-HERTZ, A. (1982). 'La Construction se-moyen du fran<;:ais et son statut dans Ie triangle moyen-passif-nSflechi'. Linguisticae Investigationes, 6: 345-401. -(1987). 'La reflexivite ergative en fran<;:ais'. Le franfais moderne, 55: 23-54. --(1989). 'Anaphor binding and narrative point of view. English reflexive pronouns in sentence and discourse'. Language, 65: 695-727. -(1993). 'On Stroik's analysis of English middle constructions'. Linguistic Inquiry, 24/3: 583-9. ZUBIZARRETA, M.-L. (1987). Levels o/Representation in the Lexicon and in the Syntax. Dordrecht: Foris. -(1985). 'The relation between morphophonology and morphosyntax: The case of Romance causatives'. Linguistic Inquiry, 1612: 247-89.

Index

A-chain 199-201, 203-4, 206 a-shell 92-3 adjective: ergative 4, 89-96 complex ergative 93-6, 98-9 simplex ergative 89-93, 98-9 unergative 4, 84, 89-96, 100 agentivity 12, 74, 75, II3 anticausative II6-18, 120-30, 133-5, 181-206; see also causative-inchoative alternation argument: reduction 187-8, 202, 236 saturation 187-8 sharing 270, 280-5 structure 10-12, 15, 234, 3II, 318, 333, 338 aspectuality: morphology 208,228-30 perfective 222 lexical vs. compositional 222-4 structures 265 telicity 208, 221-30 auxiliary selection 5, 8, 14, 17, 45, IIO, 168, 182, 241-66, 334-9 hierarchy 19-20, 255-63 binding principles 198 binyan 300-31, 302, 303, 309, 316 bootstrapping: semantic 15, 288, 292 syntactic 15, 288, 292 Burzio's generalization 77, 78, 85,152 causative-inchoative alternation II5-35, 181205,303; see also anticausatives constructional approach 249-50, 253

control 95-6, 280 detransitivization 29, II6 morphology II8, 121 Distributed Morphology 138 empty operator movement 285 Exceptional Case Marking 163, 170, 176, 178, 273, 281 expletives 32, 34 expletivization 32-5 Extended Projection Principle (EPP) 61, 83 extraction: argument vs. adjunct 284 root, see root ideophones 270, 283-4 implicit argument effects 95, 280 Lexical Conceptual Structure lQ-II, 333, 33 8,34 2 linking rules II, 75, 204, 239-41, 312, 334, 33 6-51 Locomotion Linking Rule 336, 337, 348 semantic factors in 343£ Telicity Linking Rule 336, 337, 348 middle: ending 147; see also voice morphology interpretation 147,148,186-7 marker 191, 203 template 196 morphophonological stage 297, 298-9, 3II, 330 morphosyntactic stage 297, 321-2, 330

37 2

Index

participle: modification 71 present 100-4 passive 33-9, 43-4, 46, 49-50, U6-18, 146-8, 276; see also voice morphology adjectival vs. verbal 153, 277 passivization 29,33,105,276-7 po-phrases 212 distributive 208-14 numeral 212 predicate cleft 279 predication 25, 32 principle 30-3, 35 prefixation 229, 234; see also aspect projectionist view 245-7,251 property theory 25-8, 35-6, 53 quasi PPs 218 reduced relative 156 reflexive: argument and non-argument 192-3 clitic 161-3, 165, 169 marker 191, 234-6 morphology 41; see also voice morphology process of reflexivization 29, 38, 50, 139, 147,148, 166, 176 pronoun interpretation of 200-1 reduction 154, 163-6, 176; see also process of reflexivization verbs u6, u8, 141, 143, 145, 161, 184, 185-8 afto-reflexives u8, 144-5

inherent reflexives U8,188 lexical and syntactic 176-9 unaccusatives as reflexives 36-146, 13954,159-79 resultative 269-71 root 139, 298, 300-2, 304-25; see also binyan extraction 308 syntactic event structure 293-7, 328 telicity 13,44, 60, 62, 69, u3, 223, 227; see also aspect checking 64, 67 transitivity alternations 138-49 truth conditions 25, 33, 52 unaccusative diagnostics 5; see also auxiliary selection mismatches 8 syncretism 142, 149-51 Underspecification 149 Universal Alignment Hypothesis II UTAH u,291 valence neutralization 298 verbs: complex ergative 104-8 deadjectival verbs 125,127-13 psych verbs 3, 4, 9, 82, 88, 108-12 serial verb construction 271-87 simple ergative Uo-I2 underived verbs 234-8, 308 voice morphology n6-j6, 142-9, 167