The Unaccusativity Puzzle
Oxford Studies in Theoretical Linguistics General editors David Adger, University of York; ...
94 downloads
1130 Views
6MB Size
Report
This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form
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 taltaa
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;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 mal4 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 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