Grammar in Progress: GLOW Essays for Henk van Riemsdijk [Reprint 2010 ed.] 9783110867848, 9783110130645

Citation preview

Grammar in Progress

Studies in Generative Grammar The goal of this series is to publish those texts that are representative of recent advances in the theory of formal grammar. Too many studies do not reach the public they deserve because of the depth and detail that make them unsuitable for publication in article form. We hope that the present series will make these studies available to a wider audience than has hitherto been possible. Editors: Jan Koster Henk van Riemsdijk Other books, still available in this series: 9. Noam Chomsky Lectures on Government and Binding 10. Robert May and Jan Koster (eds.) Levels of Syntactic Representation 13. Hagit Borer Parametric Syntax

25. Pieter Muysken and Henk van Riemsdijk Features and Projections 26. Joseph Aoun Generalized Binding. The Syntax and Logical Form of Wh-interrogatives

14. Denis Bouchard On the Content of Empty Categories

27. Ivonne Bordelois, Heles Contreras and Karen Zagona Generative Studies in Spanish Syntax

16. Richard S. Kayne Connectedness and Binary Branching

28. Marina Nespor and Irene Vogel Prosodic Phonology

17. Jerzy Rubach Cyclic and Lexical Phonology: the structure of Polish

29. Takashi Imai and Mamoru Saito (eds.) Issues in Japanese Linguistics

18. Sergio Scalise Generative Morphology

30. Jan Koster Domains and Dynasties. The Radical Autonomy of Syntax.

19. Joseph E. Emonds A Unified Theory of Syntactic Categories

31. Maria-Lulsa Zubizarreta Levels of Representation in the Lexicon and the Syntax

20. Gabriella Hermon Syntactic Modularity

32. Lars Hellan Anaphora in Norwegian and the Theory of Grammar

21. Jindfich Toman Studies on German Grammar

33. Jon Ortiz de Urbina Parameters in the Grammar of Basque

23. S.J. Keyser/W. O'Neil Rule Generalization and Optionality in 34. Läszlo Maräcz and Pieter Muysken Language Change Configurationality 24. Julia Horvath FOCUS in the Theory of Grammar and 35. Günther Grewendorf the Syntax of Hungarian Ergativity in German

Joan Mascaro and Marina Nespor (eds.)

Grammar in Progress Glow Essays for Henk van Riemsdijk

1990 FORIS PUBLICATIONS Dordrecht - Holland/Providence Rl - U.S.A.

Published by: Foris Publications Holland P.O. Box 509 3300 AM Dordrecht, The Netherlands Distributor for the U.S.A. and Canada: Foris Publications USA, Inc. P.O. Box 5904 Providence Rl 02903 U.S.A.

CIP-DATA Grammar Grammar in progress / Joan Mascaro and Marina Nespor (eds.) - Dordrecht [etc.]: Foris. - (Studies in generative grammar, ISSN 0167-4331 ; 36) With bibliogr. ISBN 90-6765-417-5 paper ISBN 90-6765-493-0 bound SISO 805.1 UDC801.5 Subject heading: generative grammar.

ISBN 90 6765 4175 (paper) ISBN 90 6765 493 0 (bound) © 1990 Foris Publications - Dordrecht No part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, without permission from the copyright owner. Printed in The Netherlands by ICG Printing, Dordrecht.

Contents

Preface

ix

Acknowledgments

xi

Werner Abraham A note on the aspect-syntax interface

1

Josef Bayer What Bavarian Negative Concord Reveals about the Syntactic Structure of German

13

Adriana Belletti On the morphosyntactic nature of the sequence "Aux+Past Participle" in Italian

25

Hans Bennis Ή: A note on modal passives

33

Hans den Besten The syntax of nominatives in SOV Germanic

41

Reineke Bok-Bennema On the COMP of Relatives

51

/. Bordelois and Arn. Ever s Verbal chain and verbal cluster: a discussion between linguist A and linguist B

61

Anna Cardinaletti Subject/object asymmetries in German null-topic constructions and the status of specCP

75

vi

Contents

Guglielmo Cinque On a difference between English and Italian 'Complement Object Deletion' contractions

85

Peter Coopmans A Note on Bars and Barriers

93

Carmen Dobrovie-Sorin Auxiliaries and sentence structure in Romanian

101

G. Drachman Onset clusters in Greek

113

Martin Everaert NP-movement 'across* secondary objects

125

Giuliana Giusti Floating Quantifiers in Germanic

137

Giorgio Graffi Generative grammar in Italy

147

J. Gulron Particles, Prepositions, and Verbs

153

Liliane Haegeman Non-overt subjects in diary contexts

;

167

H. Haider Datives in German "ECM"-constructions

175

Lars Hellan A principle of global binding

187

Ben Hermans On the Fate of Stray Syllables

199

Teun Hoekstra Agreement and variables

211

Tilman N. Höhle Assumptions about asymmetric coordination in German

221

Contents

vii

Anders Holmberg On bare infinitivals in Swedish

237

Harry van der Hülst The segmental spine and the non-existence of [±ATR]

247

Jonathan Kaye What ever happened to dialect B?

259

Katalin E. Kiss Why noun-complement clauses are barriers

265

Jan G. Kooij Some thoughts on the cycle

279

Jaklin Kornfilt Remarks on headless partitives and case in Turkish

285

Jan Koster Pork without pigs

305

David Lightfoot Old heads and new heads

317

M. Rita Manzini A new formalization for locality theory

323

Christer Platzack On empty theta-marked subjects in Romance and Germanic languages . . .

331

A. Puglielli and S. Scalise Readjustment rules in Somali plural formation

343

Eric Reuland Reflexives and beyond: non-local anaphora in Italian revisited

351

Gemma Rigau The semantic nature of some Romance prepositions

363

Luigi Rizzi Speculations on Verb Second

375

viii

Contents

Ian Roberts Some notes on VP-fronting and head government

387

Norval Smith and Harry van der Hülst Züritüütsch umlaut and the non-existence of the feature [tense]

397

Anna Szabolcsi Across-the-board binding meets verb second

409

Knut Tarald Taraldsen D-projections and N-projections in Norwegian

419

Alessandra Tomaselli COMP° as a licensing head: an argument based on cliticization

433

Irene Vogel The clitic group in prosodic phonology

447

Xu Liejiong Are they parasitic gaps?

455

Preface

"One thing about GLOW is that there is a sense of common purpose. People may have different ideas and approaches [...] but [...] one has the feeling [...] that they think they are working towards a common goal [...] There is a sort of common enterprise." These thoughts were expressed by Noam Chomsky in an interview conducted by Riny Huybregts and Henk van Riemsdijk in 1979-1980 and they can be applied to any point in the by now long period of existence of GLOW (19771990). A common enterprise might be viewed as a set of beliefs spontaneously shared by a group of people, but it can be - and often is - the consequence of an act of will: the result of a willingness to make a project real, to share actively and to help others to share and to take part in a common enterprise. Whether or not this is an adequate description of GLOW, it is certainly true that Henk van Riemsdijk has been the member of GLOW who has taken this task most seriously, year after year. The fact that the eleven years of Henk's chairmanship coincide with the first eleven years of the life of GLOW may be seen as a reflection of his dedication to this common project. Research is not just a matter of personal struggle with linguistic problems: it is influenced by contacts with others, stimulated by personal interchange and by the sense of sharing a common goal. One often fails to realize one's debt to those who sacrifice time and effort from their own research in order to create these indispensable elements for their fellow colleagues. Hence, the Fest of this Schrift for Henk does not mark the end of a scholarly accomplishment (this would make no sense given Henk's age and promise), but the end of a period of active dedication to the organization and functioning of GLOW. We have limited the length of the contributions not only to be able to have as many participants as possible, but also because we believe short papers better show the present state of research, of grammar in progress. The list of contributors is intended to be representative of GLOW: generative linguists in the old worlds. Among these we limited the participation to those people who have been most active in the creation and in the organization of GLOW up till the point of Henk's request not to be considered eligible any longer for the position of chairman. Most papers included in this volume are therefore written either by linguists who have been board members at some

χ

Preface

point, or by local organizers of one of the GLOW Conferences. A few people were added because they felt particularly close to Henk because of what he has meant in their scientific career as linguists. In the name of GLOW, Henk, we wish to offer you this volume to thank you for the atmosphere of enthusiasm and dedication you, more than anybody else, have created amongst the people of the old worlds engaged in the progress of generative grammar. Joan Mascaro Marina Nespor

Acknowledgments

We wish to thank Harry van der Hülst for his cooperation in different moments and ways during the period in which Grammar in Progress took shape. Thanks also to Hans den Besten and to the graduate students of the University of Leiden: Marcel den Dikken, Marjan Grootveld, Clara Levelt, John van Lit, Reno Mulder, Rint Sybesma, Pirn Wehrman, Jeroen van de Weijer. During the busy time of proof-reading, their help has been essential. Joan Mascaro Marina Nespor

A note on the aspect-syntax interface Werner Abraham University of Groningen

1. THE PROBLEM

In recent linguistic research the question as to how verbal aspectual properties relate to the syntactic description has been taken up in several papers (Pustejovsky 1988, Tenny 1989, Grimshaw/Vikner 1989a). The main common tenet in these three papers is that a proposition needs to be described in terms of some distinct event structure in order to achieve the envisioned goal. The present paper takes up this issue. Its specific goal is to describe six distinct distributional properties and relate them to a specific distinction in terms of event structure. It is assumed here that this account in terms of event structure is superior to any syntactic account, at least at the present state of the syntactic theory. In Abraham (1989) it has been argued that the crucial distinction in order to cope with aspectual (or Aktionsart) questions is that between monophasic and biphasic event structures. Diphasic event structures relate to the Aktionsart properties of terminative verbs, while monophasics reflect those of duratives. The crucial interrelation with syntactic properties is a two-fold one: (a) ergatives (in Burzio's sense) are always terminatives and, consequently, biphasics; and (b) the θ-role of an AGENT can never serve to identify the second phase of a biphasic, terminative event structure (Abraham 1989). See (1) for the pertinent event structural distinction. (1)

terminatives or biphasics (both transitives and intransitives, or one-place verbs): t, ... reference point in the event; ElfE2 ... event of the approach phase/of the resulting state. >»»»» I ti E,

2 (2)

W. Abraham non-terminatives (monophasic):

======== E

| == ======tm E 1

E

\ for activiti' «n t l1 lor statals

E

«B

Terminatives include transitives as well as ergatives. According to the Aktionsart characterization, ergatives (of any of the three syntactic types sketched in (1) and (2) above) are thus inchoative intransitives. For the ensuing discussion it is not necessary to develop the identification of the Aktionsart phases in the event structures in terms of argumental -roles in any detail (see Abraham 1989 for modifications with respect to Pustejovsky 1988 and Grimshaw/Vikner 1989a).

2. WHAT IS REALLY AMENABLE TO AN ACCOUNT IN TERMS OF EVENT STRUCTURE

In what follows six seemingly unconnected complex phenomena are presented whose distributional behavior can be accounted for in terms of the event structure distinction made in (1) above. These are the 6 distinct types of data: (1) verbal affixes and their role in event structure; (2) directional vs. locative Stative adverbiale; (3) statal (or adjectival) passive vs. process passive; (4) the distribution of grammatical (default) accent and its relation to contrastive accent within NP and S; (5) defmiteness effects and object-incorporation; and (6) the status of the direct object with semantically weak verbs. 2.1. Verbal affixes and their role in event structure In Abraham (1986) evidence has been presented in some detail that in German a class of verbal affixes (most of which carrying main word stress) has an "ergativizing" effect (in Burzio's sense). In other words, prefigating a simple verb with a member from the set of ergativizing affixes will bring about ergative properties for the derived verb. In terms of event structure, what such a specific affix causes is biphasic terminativity. It adds an approach phase, thus reanalysing the original single phase of the non-terminative verb as a result state; see (1) above. Suffice it here to run through the list of ergative affixes and attach them to just one simple verb, the intransitive movement verb laufen "run". The following list of derived verbs shares all distributional assets of ergative verbs: (her/hin) ab-, her/hin) auf, (her/hin) ein, (her/hin) aus-, (her/hin) über-, entgegen-, ent, (hin-) durch-, zu. Note that all of these prefixes have

A note on the aspect-syntax interface

3

either a directional or else telic/finalizing locative meaning, one which the basic simple verb laufen (as well as English run) does not share. Prefixes which have no such clearly delimiting and result-producing meaning do not yield ergative verbs; cf. ver-, vor-, zurück, at least in combination with laufen. 2.2. Directional vs. Stative adjuncts and PPs Other than stative locative adjuncts/PPs, in construction with which the original distributional grid of intransitive movement verbs remains unaffected, directional/telic adjuncts/PPs (both locative and temporal) create ergative constituents. See the following opposition next to that in (3a,b). (3 denotes dative case, 4 denotes accusative.) (3)

a. *die im Saal (*-drin) getanzten Mädchen * die in den the in the-3 hall within danced girls the into-4 the Saal (hinein) getanzten Mädchen hall into danced girls b. *der unter Wasser geschwommene Hans * der durch the under water-3 swum John the through den Tunnel (hindurch geschwommene Hans. the-4 tunnel swum John c. *der auf dem Baum gekletterte Junge * der auf den Baum the on the-3 tree climbed boy the onto the-4 tree (hinauf) gekletterte Junge. up climbed boy d. *der in 2$ Stunden gelaufene Läufer * der die the within 2,5 hours-3 run runner the the erforderlichen 2$ Stunden gelaufene Läufer. necessary-4 2,5 hours run runner.

The transfer from iV to eV within the class of movement verbs is signalled morphologically by the transfer from the statal dative (=3) to the directional/telic accusative (=4). Movement verbs with adjunct constituents in the statal dative are non-terminative, whereas movement verbs with the directional accusative constituent (often governed by the very same preposition; see (3a,c) where a result phase is induced emerging from a previous durative, monophasic event structure are terminative. Results are statal, but presuppose an anterior vectorial (approach) phase, as opposed to unstructured states or non-vectorial, durative events without an ensuing resulting state. It is crucial to see that the result state of a terminative event can never be the recipient of the thematic AGENT-projection; rather, it can only host the one remaining non-AG -role (most frequently THEME). The

4

W. Abraham

incompatibility of the AGENT-role with the result phase is, to all appearance, the aspectual reason for the assignment of iA (the internal structural argument) to the only S-structure argument position in the case of ergative verbs (eV). What remains to be discussed is the structural position for the aspectual information of ergativeness. What we would want, furthermore, is a unified account for the ergative effect of ergative affixes (under 2.1.) as well as ergative PPs. Consider, in this context, the distribution of the adverbial copies of the prepositions in (3a,c): they are licit in construction with the telic accusative, but illicit with the statal dative. In line with the ergative (directional) affixes in 2.1., the assumption is legitimate that an aspectual clitic to V° is the host for the lexical information of ergativity. Such, and only such, information may be copied to a PP position within VP. It must, however, remain identifiable as an ergative by means of the accusative. Any other case will not relate to the ergative clitic and will therefore not be identified as copying ergative (= resultative) information from a lower sister to V°.

VP

(4)

V1

NP Subject

V°

Adv/Prep.object

1 ci

1

3

NP

| I Del n

den [+result]

I N

1 v-1

hineiii laufen [+res ult]

Garten

By way of convention, the feature [+result] on PO triggers the accusative to be realized on the P-governed NP. If, in the case of the statal dative adjunct, we assume a structural node outside the maximal projection of VP, the mechanism sketched above follows automatically: no feature can be processed across VP, and, consequently, the canonic accusative-resultative trigger is blocked.

3. STATAL OR ADJECTIVAL PASSIVE VS. PROCESS PASSIVE

The statal (adjectival) passive in German is unequivocally signalled by the AUX sein "be", as opposed to the process-passive AUX, werden, "become". The statal passive principally disallows the AGENT-PP with the canonical structural

A note on the aspect-syntax interface

5

AG-prepositions von and durch "by (from)". This receives a natural account on the basis of the event structure for statal passives, or adjectives as well as that for sein "be" (Abraham 1989). See (4) and (5) below. (4)

a. gestorben "died" = active past participle (APP) I »»»» I - 1 event identification: eA (!„,) n eA(t„) where ^(Ej) = ^(Ej) as well as well as tJEj) = b. herausgezogen "extracted" = passive past particple (PPP) I »»»» I - 1 »1

1»

«n

event identification: iA^) n...n iA(t„) where ^(Ej) = ^(Ej) as well as (5)

a. ist/sind "is/are": - 1 state identification^^) b. war(en) = ist/sind gewesen "was /were" state identification: A(tm) η c. wird/werden "become(s)" I »»»» Ι τ- I

ti

Ε!

^

Ej

t„

event identification: A^) n A(t„,) where !„,(£]) = ^(Ε^ as well as and ^(EJ * The state structure of (5a) is compatible with (4b) only if the link with the process phase (tj-t,,,) in (4b) is erased. This in turn disallows reference to the AGENT-argument by way of the grammatical passive prepositions von or durch ("by").

6

W. Abraham

On the other hand, it is obvious why either terminative verb, the one-place eV or the two-place tV, permits a statal passive: all there is to do is cut the link to the process (approach) phase (formally, by deleting the condition tm(E2) = ^(Ej) in the event identification). It is typical of adjectives that they represent states without implying an anterior approach phase. Why is it that non-terminatives cannot form statal passives in German? Consider (6). (6)

a. b.

Der Sack wird/ *ist getragen .... [-term] tV the bag becomes is carried Es wirdJ*ist fleißig getanzt .... impers.[-term]"passive" it becomes/is a lot danced

c. Es wird/*ist durch

den Tunnel geschwommen .... impers.

it becomes/is through the tunnel swum

[+term] constr.

It is easy to account for (6a,b). sein presupposes a statal (result or not) event structure. See (5). However, there is simply no state phase in the event structure of non-terminatives such as tragen "carry" and tanzen "dance". (6c), however, seems to be tricky. Note that the result phase induced by the telic, delimiting accusative would permit the prediction that the state "passive" is possible. However, it is not. The solution lies in the required reinterpretation of what have been called "impersonal passives" in (6b,c). Note, in the first place, that impersonal passives are not really passives, by any standard assumption holding for true passives (notably promotion of the basic direct object NPs: there is no direct object to promote to the passive subject function such that the event is predicable of such a subject-NP). The event characteristic in terms of approach and/or state phase can thus not be mapped on an NP. es is an expletive constrained to the topic node. No such expletive surfaces as soon as this topic node is filled by any other sentential element such as adjuncts (manner, temporal, or local). In other words, no mapping mechanism of argument identification within the event structure can be invoked, in the absence of an external argument NP.

2.4. The status of the direct object and semantically weak ("relational") verbs As Grimshaw/Vikner (1989b:3) have correctly observed there is a wide class of "constructive accomplishment" verbs disallowing passives without semantically adequate adverbial modifications. These are their examples (Grimshaw/Vikner nos. (5a-c)), among others.

A note on the aspect-syntax interface (7)

7

a. *The house was built/assigned/erected/constructed b. *the built/designed/constructed house c. The house was built yesterday/in ten days/in the south d. the newly/recently/built//badly/cleverly designed house

As opposed to (7), an equally large class of "inherent accomplishment" verbs does prompt adjunctless passives. (8)

a. b. c.

The house was destroyed (yesterday). The soldiers were shot. The resistance was broken.

For Grimshaw/Vikner's account of the distribution in (7) and (8) (which I will not dispute here) see (9a,b). (9)

a.

event

activity eA,iA

b.

state iA

event

activity eA

state iA

Let us, however, assume two structural types of direct objects: one where reference to the direct object NP is a prerequisite to arriving at a semantically satisfactory reading of the simple verb; and a second one, where reference to the object-NP is at much larger variance and, consequently, is no requirement for a semantically satisfactory reading of the simple verb. See (10a,b) for the structural distinction. (10)

a.

for "inherent accomplishments" like destroy, kill, arrest, and break: V1 I

V°

b.

NP

for "constructional accomplishments": V° I

NP

An NP of the type in (lOb) will not be independent structurally as an NP in (lOa).

8

W. Abraham

Given (10a,b), another type of data receives a ready account. Note that idiomlike verbal complexes (such as kick the bucket, hit the sack, hold a position; make a statement) disallow passivation, or they allow it only under specific restrictions. It is plausible to assume for them, on purely semantic grounds, a structure such as in (lOb), since the information collected by the complex [NP + V] is carried essentially by the NP (in the case of the metaphors just as well as with idioms). It is to be noted that the structural distinction in (10a,b) can accommodate an extended set of data, and, simultaneously, that it offers an account for the distribution distinguishing "constructional" and "inherent" accomplishments. Cf. (7) and (8) above. In 2.5. we will extend the empirical observations under one unified perspective and address new facts. 25. Definiteness effects and the syntactic status of the rhema The account offered in 2.4. is incomplete to the extent that it provides no explanation for the grammaticalizing effect of the adjunct Cf. again (6c,d) as opposed to (6a,b). The question is thus: What do adjuncts do for passives of "constructional" accomplishment verbs that they need not do for "inherent" ones, under the structural assumption of (10) above? My assumption is that the observed distinction has to do with the degree of semantic satisfaction provided for the position of the discourse function of the rhema as opposed to that of the thema. See (11) below as an illustration of what is meant by this distinction. We shall have to turn to German since in this language the thema-rhema distinction is borne out more clearly in terms of structural constraints while, at the same time, allowing for a considerable degree of linear variation such that discourse functions are reflected by word order. Note that the grammatical (default) accent (GA) within a sentence is always on the object-NP or PP immediately preceding V-last, or else on V-last in the case that there is no PP or object-NP. Within NPs, the rightmost N (constructional head) carries the grammatical accent By common assumption, the default accent together with the basic word order distribution signals the grammatically organized rhematic (or non-thematic) information in the sentence. Take bauen "build", one of Grimshaw/Vikner's class representatives. (11)

a. ?Das Haus wurde (heute) GEBAUT ... (GA=grammatical accent) the house became today built b. *Ein Haus wurde (heute) GEBAUT (GA) a house became today built c. ??Das HAUS wurde (heute) gebaut ... (CA=contrastive accent) d. DIESES Haus wurde (heute) gebaut ... CA this house became today built e. Ein HAUS wurde (heute) gebaut CA

A note on the aspect-syntax interface

9

f. Es wurde (heute) ein HAUS gebaut .. GA g. *Es wurde ein Haus GEBAUT .. CA "A house was built (today)" .... preseiitative sentence (= rhematic) information Note that the adjunct heute "today" has no grammaticalizing effect, nor does it carry default stress. Let us now turn to zerstören "destroy", which behaves differently according to Grimshaw/Vikner's observation. Accordingly, their account is different, too. Our examples, however, show that it is not the classification of the verbal that motivates the distributional distinctions, but definiteness effects together with a uniform focus position. (12)

a. Die Brücke wurde zerstört the bridge became destroyed b. *Eine Brücke wurde ZERSTÖRT

GA

c.

CA

d. e. f. g.

Die BRÜCKE wurde zerstört

...

DIESE Brücke wurde zerstört ... Eine BRÜCKE wurde zerstört ... Es wurde eine BRÜCKE zerstört ... *Es wurde eine Brücke ZERSTÖRT ..

GA CA CA GA CA

The distribution is obvious. Indefinite subjects are allowed only in presentative sentences (subject in rhema position, i.e. left-adjacent to V-last, where it receives GA), as in the case of the (inversions, or else with unchanged linear order under CA, as in the (e)-versions. Indefinites are deictic rhemata. They can only be tolerated in canonical rhematic positions (left-adjacent to the verb and thus under GA), or under contrastive stress (CA) in other positions. Now note the only two versions distinct between (11) and (12), namely (a) and (c). Our assumption is that the differences in the (a,c)-versions are taken care of by the structural distinction in (10a,b), bauen "build" being a relational verb as opposed to zerstören "destroy", an absolute verb. No adjunct influence is at play. In Abraham (1988b) it has been demonstrated in some detail what the parametric variation between rhematic structures in English, German, and Dutch is like. The parametric differences are induced by the fact that the V-second/V-last structure of the sentence (opening the "middle field") in German and Dutch leaves more room for a discourse-functional structure of the sort that the distinction between thema and rhema information presupposes. Suffice it to say here that the observed obligatory adjunct information required for "weak" (or relational) verbs as well as idiomatic verbal complexes appear to be a remnant of this information structure, which became gradually suppressed along with the emergence of the fixed SVO-structure in late Middle English. In other words, the adjuncts postulated for English "weak" verbs fill an original

10

W. Abraham

rhematic position in originally canonic left-adjacency to an original Germanic V-last position: [ +V°]V.. See Koster 1989 for an extensive argument to this effect. In the case of "inherent" accomplishment verbs the semantic rhema requirement is satisfied by the semantically satisfied verb alone. 2.6. Grammatical vs. contrastive stress The conclusions drawn in 2.5. carry over directly to the present observations. What contrastive stress does is open a contextual variable of identical formal structure constrained by a semantics of oppositeness. A pronominal passive participle carries no stress by default. If it does, however, a variable in the identical pronominal position is implicated in the semantic relation of oppositeness. Cf. (13) with inherent accomplishment verbs as distinct from (14) with "constructional" accomplishments. (13)

a. b.

the arrested MAN/hidden SOLUTION ... gr.(deflt.)accL (GA) the ARRESTED man/HIDDEN solution .... contrastive acct. (CA)

(14)

a. b.

a *built HOUSE /»written PAPER /*held BELIEF .... a *BUILT house /WRITTEN paper /*HELDbelief ....

GA CA

The grammatically judgments are those by Grimshaw/Vikner (1989a,b); the accents have been added for the present line of argumentation. See the irregularity in (14b) between built/held and written. Other examples betraying unsystematic heuristic validations are (again Grimshaw/Vikner's judgments): ??a loved man; Ithe destroyed house', Ithe killed chicken, line admired/respected man, line understood solution, Ithe constructed example, Ithe painted wall (Grimshaw/Vikner 1989a,b). The heuristically undecided status presumably has several sources. One is that the verbs in the prenominal position as above are not terminative (loved, admired, respected). Their PPs have to be made resultative, in this attributive function, by virtue of a result-implying, or adjective-selecting, adjunct Another reason is the concealed stress ambiguity. Note that it is easy to render grammatical a WRITTEN paper on the basis of such paradigmatic opposites as read, oral, copied. But it is much harder to legitimize built in (14b). Note further that a built house gives rise to a feeling of pleonasm, with house necessarily presupposing build. Finally, and probably foremost, the verbal adjective in prenominal position is by far not as canonical in English as it is in SOVlanguages. The main argument in support of this conclusion is the fact that the preterite participle in English has lost almost all of its verbal government properties. Compare German (15) (Dutch siding fully with German) with what would be absolutely non-understandable in English.

A note on the aspect-syntax interface (15)

11

der durch den seiner Frau geschenkten Gaul vor 2 Jahren the by the (to)his wife presented horse 2 years ago getretene Mann kicked man "the man kicked 2 years ago by the horse which he had given to his wife"

German and Dutch preserve the full sentential structure under canonical leftward-government of the verb irrespective of its participial-form. In English, this government relation is severely pruned. This permits the conclusion that pronominal PPs in English are not open to clear grammatically based evaluations.

3. CONCLUSION

The main goal of this paper was to show to which extent syntactically motivated verb classes and their distributional properties (see above all Grimshaw/Vikner's data in section 2.4.) correlate with, and can be explained by, aspectual (or Aktionsart) distinctions. This is what we have called at the outset of this paper, the interface between the semantic event structure and the syntactic agreement structure. We have tried to show that the "obligatory adjunct restriction" postulated by Grimshaw/Vikner (1989a,b), while essentially correct, does not provide a unique account for what is a heterogeneous set of data. Methodologically we have become aware of the restrictions characterizing English with respect to distinctions such as adjectival vs. event passive, morphological elements of word-formation (affix, case) yielding resultativeness and, consequently, ergativeness, and the unimpaired leftward government properties of verbals with telic PPs.

REFERENCES Abraham, W. (1986) "Transitivitätskorrelate und ihre foimale Ein bindung in die Grammatik." Groninger Arbeiten zur germanistischen Linguistik (GAGL) 26: 1-60. Abraham, W. (1988a) "Unaccusatives in German." Groninger Arbeiten zur germanistischen Linguistik (GAGL) 28: 1-72. Abraham, W. (1988b) "Ergative subjects, the partitive solution, and the NP/DP question." Groninger Arbeiten zur germanistischen Linguistik (GAGL) 29: 161-84. Abraham, W. (1989) "Modalverben und ihre Ercignisstruktur." (to appear in GAGL 30) Grimshaw, J. and St. Vikner (1989a) Obligatory adjuncts and the structure of events. Ms. Brandeis University and University of Geneva (April 1989). Grimshaw, J. and St. Vikner (1989b) Handout for the Workshop on Lexical Structure and Language Acquisition. University of Groningen, May 25, 1989. Koster, J. (1989) The residual SOV-structure of English. Paper read at the Vienna Round Table,

12

W. Abraham

October 6-8, 1989. Pustejovsky, J. (1988) Event semantic structure, Ms. Brandeis University. Vikner, St. (1988) "Modal verbs and event structure." Working Papers in Scandinavian Syntax. Lund.

What Bavarian Negative Concord Reveals about the Syntactic Structure of German* Josef Bayer Max-Planck-institute for Psycholinguistics, Nijmegen

0. INTRODUCTION

Since the dawn of generative linguistics, there has been an ongoing debate about the question whether German has a truly configurational syntactic structure like English, French, and Italian or whether it is more on the side of socalled 'non-configurational' or 'free word order' languages like Hungarian or Latin. More recently, a consensus seems to have been reached in favor of at least a weakly configurational structure, whatever this may be when observation comes down to a theoretical implementation. The. question at present is not so much whether German makes use of a maximal V-projection which does not include the subject-NP, but rather how one can account for the 'scrambling' phenomena which make German markedly different from English or even from such a closely related language as Dutch. In this contribution, I will consider some hitherto undiscussed data which turn out to be relevant both for the configurationality debate and for the German scrambling facts. The present data come from Bavarian, a non-standard variety of German, which is spoken in Bavaria and parts of Austria. Like many other languages of the world, including substandard English, Bavarian shows the negative concord phenomena, i.e. there may be more than one carrier for the negative element in a single sentence while its semantics is that of simplex negation. This phenomenon is well studied in languages in which negative concord (NC) or double negation belongs to the standard language, as in Italian (see Rizzi, 1982). It is less often studied in languages, however, where it is considered substandard. An exception is Labov (1972) which contains a detailed analysis of NC in Black American English. In section 1, a subject/object asymmetry will be discussed which arises in Bavarian sentences with NC. In section 2, an account for NC will be proposed which will then serve as a basis for the evaluation of the scrambling phenomena toe be introduced in section 3.

14

J, Bayer

1. A SUBJECT/OBJECT ASYMMETRY

In Italian one usually finds NC only in those cases in which a negative quantifier such as nessuno ('nobody') or niente ('nothing') appears inside the VP. Rizzi (1982) discusses cases such as the following: (1)

a.

b.

(2)

a.

b.

Mario *(non) ha visto nessuno Mario not has seen nobody 'Mario hasn't seen anybody Nessuno (*non) ha visto Mario 'Nobody has seen Mario' Non pretendo ehe nessuno sia arrestato not (I) demand that nobody be arrested Ί don't demand that nobody should be arrested' Non pretendo ehe sia arrestato nessuno Ί don't demand that anybody should be arrested'

Provided that the semantics represents the reading given in the glosses, (la) is illicit when non is absent, while (Ib) is illicit when non is present. (2) shows that what matters is not just the status of the negative quantifier as subject or object, but rather its position in the clause: In (2a), the negation nessuno cannot be 'absorbed' by the negation in the matrix clause, but when subject inversion has applied as in (2b), it can. Rizzi, following Kayne (1981), explains these cases with an application of the Empty Category Principle (ECP) (see Chomsky, 1981) at the level of Logical Form (LF): when the negation inherent in the quantifier is associated with the negative element non, an ECP-violation will arise in those cases where the negative quantifier is not properly governed. Since nessuno in (la) is governed by the verb, no ECP-violation will arise. In (2a) the negation inherent in nessuno cannot be associated with non, because it is not in a properly governed position. When nessuno appears in the government domain of the verb, however, the association is licit, as shown in (2b). Rizzi gives independent evidence from the syntax of movement which corroborates his analysis. In order to account for (Ib) one has to say that non must c-command the negative quantifier in order to make the association possible. It is this case which we also find in most varieties of Black English and in Bavarian. According to Labov, Black English as well as the majority of white dialects with NC require a negated auxiliary to c-command the negative quantifier. The following examples are drawn from Labov (1972):

Syntactic Structure of German (3)

a. b. c. d.

15

I don't measure nothing You don't have to know too much of nothing He didn't take nothin' offa nobody I ain't never had no trouble with none of 'em

For many speakers, NC with the preverbal position is considered to be ungrammatical. But Labov also found a group of speakers who produce sentences such as: (4)

a. b. c. d.

Nobody don't like a boss hardly Down here nobody don't know about no club None of our friends don't fight 'im None of 'em can't fight

Similar facts are reported from first language acquisition in Bellugi (1971). In Bavarian, NC from inside VP is always possible and for certain speakers it is the rule, while NC from the VP-external position is perceived as ungrammatical. The data in (5) are drawn from Ludwig Thoma's satirical works 'Briefwexel eines bayrischen landtagsabgeordneten' (1909) and 'Jozef Filsers Briefwexel' (1912):1 (5)

a.

b.

c.

d.

ich bin froh, das ich keine Rede nicht halden brauch,... l am glad that I no talk not give need am glad that I don't have to give a talk' Also brauchst keine Angst nicht haben thus need(2sg) no fear not have "Thus, you don't need to worry' sondern ich geh auf keine Rehdutt nicht mehr but I go to no camevall-ball not more 'But I will never go to a carneval-ball again' und keinen ändern nichd leihden wiel and no other not tolerate wants 'and does not want to tolerate another one'

In theses examples the negative quantifier is undoubtedly in the VP in which nicht occurs. However, the sentences in (6) are generally rejected by speakers of NC-dialects unless true double negation in the logical sense is intended. (6)

a. *das koa Mensch des Zeich ned ooschaung woidd that no man this stuff not look-at wanted 'that no one wanted to look at this stuff

16 /. Bayer b. *das that 'that c. *das that 'that

koana in des Audo ned eig'schdieng is no one in that car not entered is no one got in this car' neamads da Soffi ned wehdua woidd no one the Sophie not hurt wanted no one wanted to hurt Sophie'

In these examples, the negative quantifier NP remains outside the VP, given a configurational structure of German. Things are, however, not always that straightforward. There exist cases in which NC with the subject-NP does occur. Here are some more of Thoma's examples: (7)

a.

das keine Unanstendikeit nichd bassirt isd that no indecency not happened is 'that no indecency has occurred' b. und keine Todsiende nichd forfiel and no mortal sin not occurred 'and no mortal sin occurred' c. hobwohl jez in der kuhltur keine arbeid nicht zum ferrichten ist although now in the culture no work not to do is 'although there is at present no cultural work to do' d. intern das keine keischheid nichd dabei sein mus in that no chastity not involved be must 'since chastity does not have to be involved'

In (7a) and (7b) the verbs passieren en vorfallen are ergative verbs in the sense of Burzio (1986). Let us assume that their nominative argument must arise inside VP for principled reasons. Keine arbeid in (7c) is the underlying object of the verb verrichten, which appears in a quasi passive-construction (see Haider, 1984). There are reasons to attribute VP-internal status to such NPs (see Fanselow, 1987, ch.3). (7d) involves the copula sein; it can be analyzed as an expletive construction (see Safir, 1985, who argues for an empty expletive element in subject position). Jarich Hoekstra (p.c.) informs me that Frisian, another NC-language, exhibits a very similar subject/object asymmetry. Since Frisian has an overt expletive element (der), one can test whether, when it is present, NC with the subject is possible; and indeed, NC with the subject is permitted only in this case i.e. when the negative subject-NP arguably occurs inside the VP. Provided that this is so, we have found a subject/object asymmetry which can be captured with the following generalization: (8)

NC in Bavarian can only hold between a (primary) negative element X and a negative quantified constituent Υ if both X and Υ are VP-internal.

Syntactic Structure of German

17

In the next section we will discuss some of the consequences of this generalization for the syntax of Bavarian, and, by extension, for the syntax of German.

2. TWO RULES OF NEGATIVE CONCORD

As in Italian and in most varieties of substandard English with NC, Bavarian requires a command-relation between the negative element nicht (or ned) and the negative quantifier. We will assume that nicht is a syncategorematic expression (somewhat like a focussing particle, see Jacobs, 1982; 1983), which in the unmarked case adjoins to V0.2 The element NEC and V together form an new V. It appears that a negative quantifier must be c-command by this negative verb. C-command is to be understood in the strict sense of Reinhart (1976). (9)

α c-commands β iff the first branching node dominating α also dominates β and neither α dominates β nor β dominates a.

As shown in the example in (5), the negated V always c-commands the XP containing kein. (5c) shows that what counts is the verb's D-structure position. Given that what we have said about the examples.in (7) is tenable, the nominative-NP is also in the VP and will be c-commanded by the negated V.3 In (6), however, this is crucially not the case. The adoption of the definition in (9) correctly predicts that the following constructed examples are ungrammatical as well: (10)

a. *das da Bene koan Mensch im Haus ned g'seng hod that the Bene no man in the house not seen has that Bene hasn't seen anyone in the house* b. *das da Bene koam oanzign d'Soffi ned vorg'schd d hod that the Bene no single (daL) the Sophie not introduced has 'that Bene hasn't introduced Sophie to a single person'

We assume the following LF-rule for NC:

18 /. Bayer (11)

NC-Rule (l? SS: [„ ... [χρ NEG ...]j ... NEG + V] -> LF: [a ... [χρ ...]j ... NEGj + V] where α is a domain Y in which NEG + V c-commands XP.

In all cases to be discussed here, XP will be reprented at LF as an existentially quantified NP.5 At LF, NEG raises to VP. Since XP is coindexed with NEG and since it is a quantified phrase (QP), it will also raise to VP. This gives the intermediate structure in (12):

(12)

[w NEG QPi [yp ... [χρ e; ] ... V ]]

From this position NEG + QP can raise to IP, which corresponds to the previously used sentence 'S' (see Chomsky, 1986). The question is whether there is empirical motivation for this particular proposal. We think there is. Recall that we have found an asymmetry with respect to the position of the QP in an NC construction. Consider now the following examples by Thoma: (13)

a. Zu der weidlichen Obrikeid had kein Mentsch kein Ferdrauen to the mundane authority has no man no trust 'No one trusts mundane authority' b. Gozeidank had keine Zeitung nichz erfarren Thank God has no paper nothing experienced 'Thank God, no newspaper has caught wind of anything'

In these examples there is no NEC-element nicht, but two negative QPs. NC applies here as well, but it is always possible for the subject-NP to absorb the negation of the object-NP. We have seen before that NC is not licensed when NEG does not c-command the negative QP. In this case, however, the lower QP will never c-command the subject-QP. We therefore adopt a second NCrule. This may be seen as a subcase of the rule of Quantifier Raising (QR): (14)

NC-Rule (II) SS: [„> NEG QP, ... NEG QP2 ... NEG QPn ...] -» LF: [IP NEG QP, [IP QP2 [IP ... [„> QPn [„, e, ... % ... en ...]]]]]

(14) says that any NEG QPk (with k>l) will transmit its NEG element to QPj when QR applies. This derives the NC-reading of the examples in (13). Coming back to (12), rule (14) predicts that negative QPs whose negation was

Syntactic Structure of German

19

previously absorbed according to rule (11) will also be affected. If (14) would raise a negative QP before (11) has applied, readings would emerge which do not conform to NC. Cases such as (15), however, get a natural NC-reading: (15)

das koa Mensch [de Jager koa Bier ned zoid hod] that no man the hunter no beer not paid has 'that no one paid the hunter a beer'

Once (11) has applied to the VP, (15) will be as in (12), i.e. the negation inherent in koa Bier is absorbed by ned and both NEG and the QP (corresponding to "a beer") are adjoined to VP. Rule (14) will affect this newly formed negative QP and raise it to IP where its inherent negation is absorbed by the negation inherent in the subject-NP koa Mensch. The LF of (15) is: (16)

NEG QP! [„, QP2 [„> ej [yp e2 [yp de Jager % zoid hod]]]]

This seems to be a proper basis for interpreting (15) in Bavarian. In the next section we will return to rule (11) and see what it predicts for subject/object scrambling and object-cliticization to COMP.

3. NEGATIVE CONCORD AND SCRAMBLING

Consider the following sentences of Standard German (both of which have an equivalent in Bavarian): (17)

a.

b.

daß den Postboten der Hund gebissen hat that the postman (ace) the dog(nom) bitten has 'that the dog has bitten the postman' daß ihn der Hund gebissen hat 'that the dog has bitten him'

Consider furthermore that both of these examples are represented at S-structure with a trace to the immediate left of gebissen. This trace would be bound by the object-NP in (17a) and by the object-clitic in (17b): (18)

[n, [NP/CLITIC]; [IP NPNOM [w ^ V ...]]]

In the previous sections we have seen that NC between nicht (ned) and a negative QP is only permitted when the negated V c-commands QP. In (18), V crucially does not c-command the NPNOM. If V is an agentive verb like beissen, we can exclude the possibility that its subject-NP is an underlying object. With this prerequisite, (18) should disallow a NC-reading when NPNOM is a negative

20

/. Bayer

QP and V is negated. This prediction is wrong, as shown by the following pairs of examples: (19)

a. *das koa Hund an Bosdboon ned beisd that no dog(nom) the postman (ace) not bites b. das an bosdboon koa Hund ned beisd

(20)

a. *das koa Hund earn ned beisd that no dog(nom) him (ace) not bites b. das'n koa Hund ned beisd

While the canonical a-sentences are sharply ungrammatical (under an NCreading), the scrambled b-sentences are perfect. I conclude from this that (18) cannot be the right representation for scrambling or clitic movement in German. In order to acknowledge the fact that nicht (ned) has to c-command the QP whose negation it will absorb, the negated V will have to c-command the subject-NP. This could be achieved when we allow sentences like those in (17) as well as (19b) and (20b) to be base-generated. The question is not so much whether this can be done, but whether there are independent reasons to do so. I believe a major reason for advocating a base-generation account is that I(NFL) is not a syntactic category in German, but a morphological element that projects 'in tandem' with the verb. In this sense, an inflected verb is both V and I at the same time. Let us assume that the elementary projection of an inflected V looks as follows: (21)

In (21), I is the formal head of V°. Attachment to V°, which is stem, results in Γ. Only at the mother node in (21) is the word-level reached. This mother node is the minimal element that counts as a syntactic category in German. What is special about this node is that it combines two functors, a lexical functor an a functional one. In the same way as V seeks an NP to its left to which it will assign an object Case and a theta-role, Γ will seek a specifier-NP to which it can (via 1°) assign nominative Case. If this natural assumption is made, (21) is allowed to project as follows:

21

Syntactic Structure of German

(22)

VP

NP

IP

v° NP.,

Γ

v°

\

Lack of space prevents me from going into this any further. It should, however, be clear that such a system allows for at least a descriptive account of scrambling and clitic movement, without making reference to 'movement' in the nonmetaphorical sense. This sketch should suffice to show how the Bavarian NC-facts in (19) and (20) can be derived: since only the mother node in (21) is a syntactic word, NEG will adjoin to this node and never to anything dominated by it In the following structure of (19b) the negated [I'.V]-category c-commands the Q-NP koa Hund, and the predictions of the NC-rule in (11) go through.

(23)

beisIt is interesting to see that "movement" of an object-clitic across a negative subject-QP makes the use of NC also possible in Frisian (Jarich Hoekstra, p.c.).

22

/. Bayer

4. CONCLUSION In this brief investigation of Bavarian Negative Concord (NC), we have proposed two rules, one by which the negation on a quantified phrase gets absorbed by the negative element NEG which is attached to the verb, and another by which the raising of a negative QP to another negative QP leads to absorption. The first rule, which is central to this investigation, can only apply when NEG c-commands QP. This captures the fact that in Bavarian NC usually applies when Qp is an object or another element immediately to the left of NEG+V. Then we have shown that in those instances of scrambling where a negative subject-QP appears immediately to the left of NEG+V, NC is fully operative. This is a strong reason against scrambling as an instance of the rule Move-Alpha. If at S-structure NEG+V would c-command an object-trace, the NC-rule could not apply. Since the NC-rule does apply, we have concluded that the subject-NP can be base-generated to the immediate left of V. This conclusion is supported by the fact that I(NFL) in German is a morphological affix which projects together with the verb. This finding ties in with Rizzi's (1982) account of negation in Italian as well as with the facts of Frisian which were mentioned in the text While we have not tried to reduce the Bavarian NC-phenomenon to the ECP,6 it should be clear that both mechanisms, subject-postposing in Italian and the licensing of the subject in the VP in German and Frisian, achieve approximately the same goal.

NOTES * I want to express my gratitude to Jarich Hoekstra, Joachim Jacobs and Jaklin Komfilt for discussing issues of negative concord with me. David Pesetsky and Tom Roeper drew my attention to the work of Labov. Melissa Bowerman informed me about relevant data from language acquisition. Allard Jongman and Joan Sereno were kind enough to check my English. 1. Thoma's work is representative in that it ridicules the way in which a halfway illiterate peasant tries to write Standard German. The naive use of NC gives a reliable picture of the dialect. Whenever I quote from this work. I use Thoma's original transcription, which for artistic reasons is different from my own. Since we are dealing only with syntactic issues here, no confusions will arise from this. The quotes are drawn from the following edition: Thoma, L. (1986), Jozef Fibers Briefwexel. München: Piper. 2. An exception is that it must adjoin to [PP+V] when the PP is subcategorized by V. I have nothing to say about this peculiarity. 3. In the spirit of Kratzer (1984), Reuland and Kosmeijer (1988) and Reuland (1988) we take I(NFL) as a morphological feature on the verb, i.e. it is crucially not a terminal syntactic node which takes a VP-complement Rather, I projects together with the verb as a complex category. As such, it can license nominative Case inside the VP. This explains why the subject-NP does not have to move to pre-VP position in order to receive nominative Case. We will return to this issue in more detail in section 3.

Syntactic Structure of German

23

4. What counts as the input to LF is an S-structure (SS) with traces into which moved constituents can reconstruct. Otherwise we could not explain grammatical NC-examples such as (i), which is represented with traces: (i)

[Keinen Wiederspruch]; gibd: es e; nicht : no objection exists there not 'There is no objection' or 'Objections are not allowed'

5. Seen from a somewhat different perspective, NC is a means to signal at S-structure which constituents are affected by negation. In languages without NC, this relation has to be inferred. 6. One reason not to do this is that in German lexical government does not seem to play as important a role as it does in other languages. For example, WH-island violations are equally ungrammatical when a subject-WH-element or an object-WH-element is extracted. See Bayer (1989).

REFERENCES Bayer, J. (1989) "Notes on the ECP in English and German", ms. Max-Planck-Institute for Psycholinguistics, Nijmegen. Bellugi, U. 1971) "Simplification in children's language", in Huxley, R. and E. Ingram (eds.), Language Acquisition: Models and Methods, Academic Press, London. Burzio, L· (1986) Italian Syntax. A Government-Binding Approach, Reidel, Dordrecht. Chomsky, N. (1981) Lectures on Government and Binding, Foris, Dordrecht. Chomsky, N. (1986) Barriers, Press, Cambridge (Mass.). Fanselow, G. (1987) Konfigurationalität, Narr, Tübingen. Haider, H. (1984) "Was zu haben ist und was zu sein hat", Papiere zur Linguistik 30, 23-35. Jacobs, J. (1982) Syntax and Semantik der Negation im Deutschen, Fink, München. Jacobs, J. (1983) Fokus and Skalen. Zur Syntax und Semantik der Gradpartikeln im Deutschen, Niemeyer, Tübingen. Kayne, R. (1981) "Two notes on the NIC", in Belletti, A., L· Brandi and L. Rizzi (eds.). Theory of Markedness in Generative Grammar, Scuola Normale Superiore, Pisa. Kratzer, A. (1984) "On deriving syntactic differences between German and English", ms. Technical University Berlin. Labov, W. (1972) Language in the inner City. Studies in the Black English Vernacular, University of Pennsylvania Press, Philadelphia. Reinhart, T. (1976) The Syntactic Domain of Anaphora, doctoral dissertation, MIT, Cambridge (Mass.). Reuland, E. (1988) "Relating morphological and syntactic structure", in Everaert, M., A. Evers, R. Huybregts and M. Trommelen (eds.). Morphology and Modularity, Foris, Dordrecht. Reuland, E. and W. Kosmeijer (1988) "Projecting inflected verbs", Groninger Arbeiten zur Germanistischen Linguistik 29, 88-113. Rizzi, L. (1982) Issues in Italian Syntax, Foris, Dordrecht. Safir, K. (1985) Syntactic Chains, Cambridge University Press, Cambridge.

On the morphosyntactic nature of the sequence "Aux+Past Participle" in Italian Adriana Belletti University of Geneva - Scuola Normale Superiore, Pisa

Consider the following negative clauses in Italian: (1)

a. b. c.

Gianni non ha parlato piu lit: G. not has talked anymore (G. hasn't talked anymore) Maria non έ uscita mai lit: M. not has gone out ever (M. has never gone out) (Quel lavoro) Non 1'ho finite ancora lit: (that work) (I) not have finished it yet (I haven't finished it yet)

Let us call non the negation proper and the adverbs piu, mai, ancora, which (can) cooccur with it, negative adverbs. All the sentences in (1) display the order "non Aux+PstPrt piu, mai, ancora". Compare (1) with (2): (2)

a. b. c.

Gianni non ha piu parlato lit: G. not has anymore talked Maria non έ mai uscita lit: M. not has ever gone out (Quel lavoro) Non 1'ho ancora finite lit: (that work) (I) not have yet finished it

The sentences in (2) have exactly the same meaning as those in (1); they just instantiate the option of a different word order. The negative adverbs here are located between the auxiliary and the past participle; the resulting order is: "non Aux piu, mai, ancora PstPrt". This order is exactly the same as the one found in negative clauses in French containing the negation ne and the negative adverb pas: (3)

II n'a pas parlo lit: he not has pas talked (he hasn't talked)

In Belletti (forthcoming) I argue that Italian sentences like those in (2) can be analyzed in the same way as the French negative sentences exemplified by (3). The analysis I adopt assumes that the negation proper is a head (as also in Kayne (1989)), giving rise to its own projection in terms of X'-theory, call it

26

A. Belletti

NegP. The existence of a NegP in negative sentences is postulated in many recent works (Pollock (1989), Chomsky (1989), Laka (1989), Moritz (1989)). The particular version of the proposal that I am adopting is the one developed by Moritz (1989) for French, according to which, in the clause structure, NegP is located between the highest functional head, which I take to be AGR, and the next functional head containing verbal morphology, which I take to be T (see Belletti (forthcoming) for detailed discussion). The negative adverb pas is analyzed as the Specifier of the NegP. Extending this proposal to Italian, the following representation results: (4)

AGRP ~"

with non the head of NegP and the negative adverbs piü, mai, ancora, its specifiers. As is assumed for French ne in the references quoted, this proposal includes the further hypothesis that the negation non is a clitic. Suppose that, as such, it must move to the AGR position, with a head-to-head type of movement. (Assume, for concreteness that this instance of head-movement is leftadjunction to AGR). Consider then the derivation of a sentence like (5). It includes the following processes: Neg to AGR; V to T to AGR (let the latter two processes be substitution in the sense proposed by Rizzi & Roberts (1989)): (5)

Gianni non parla piu G. does not talk anymore

Given the combination of the processes just described, the order "non inflected V negative adverb" is directly obtained, with the negative adverb remaining in the Spec of NegP position. Notice that the same analysis holds for French negative sentences like the one in (6): (6)

Jean ne parle pas J. does not talk

According to this proposal, a significant difference between Italian and French consists in the fact that the Specifier position of the NegP is obligatorily filled by pas in French and optionally filled by the negative adverbs in Italian. Thus, when the Spec position of NegP is empty in Italian the result is a clause

"Aitx+Past Participle" in Italian

27

involving simple sentential negation. The diagram in (7) illustrates the proposed derivation for a sentence like (5); when the adverb piü is not present, a possibility indicated by the parenthesis, the result is sentence (8):

(7)

(8)

Gianni non parla G. does not talk

As for the question why the derivation in (7) does not give rise to a violation of the Head Movement Constraint / ECP, I assume (Belletti (forthcoming)) a version of the proposal discussed in Moritz (1989). According to this proposal, although derivationally problematic, (7) is in fact representationally wellformed. This is so because, once the negation is incorporated within the inflected verb, they constitute, together, the head of the same chain. At the appropriate level, then, no principle is violated. Let us now consider sentences containing a complex tense formed by an auxiliary and a past participle like those presented at the outset in (1) and (2). The proposal just reviewed makes a direct prediction concerning them. Before discussing it in detail, let me briefly present the (essential aspects of the) structure I am assuming for past participles. As is proposed in Pollock (1989), Chomsky (1989), Belletti (1989), a past participle can be viewed as an AGRP whose AGR head is the past participial morphology (-/-in Italian) taking the VP as complement. (9) illustrates this proposal:

28 (9)

A. Belletti AGRPX

AGR -t-

...V... where V has to move to AGR to incorporate with the past participial morphology. Given this analysis, a sentence containing a complex tense formed by an auxiliary and a past participle will have the Aux taking the past participial AGRP as a complement, in the sense of X'-theory. Let us now go back to sentences containing a complex tense, the negation proper and a negative adverb like those in (1) and (2). The abstract structure of such sentences will have the shape of the diagram in (10):

If we now fill this abstract representation with words, roots and inflectional endings corresponding to the sentences in (1) and (2) we notice immediately that a direct prediction is made by the analysis of negation we discussed once it is combined with the general head movement processes deriving the inflected forms of the verb: only the order in (2), "non Aux piu PstPrt", is expected to be wellformed. In a representation like (10), non cliticizes onto AGR, the auxiliary moves to T and then to AGR, and V moves to the AGR past par-

"Aux+Past Participle" in Italian

29

ticipial head. The result is (2), never (1). If we were considering French, this would be the desired result. Next to (3) French does not display any equivalent sentence (11), which is ungrammatical and which would correspond to the Italian (1) in terms of word order: (11)

*Jean n'a parlo pas

The question then arises as to how to derive the immediate adjacency of Aux and PstPrt in the Italian examples in (1). We now concentrate on this issue. It seems that if negative adverbs can only fill the Spec of NegP position, we are left with only one possible analysis for (1): the past participle must be allowed to incorporate within the Aux. If this is done, it would then be the complex word "Aux + PstPrt" to move to the highest AGR position generating the desired word order. Notice that from the point of view of the theory of morphosyntactic incorporation (Baker (1988)), this instantiation of the process would be perfectly legitimate. Alternatively, we could suggest that negative adverbs are also allowed to fill a different position in the sentence structure, beside the Spec of NegP position we discussed. A possible candidate is the VP-initial position (which is independently a possible adverbial position). If this is the case, there is no need to assume the occurrence of an incorporation process in order to obtain the desired word order. We could assume that the NegP has no overtly realized Spec (as in sentences containing simple negation like (8)) and that the negation regularly moves to the AGR position, the Aux as well and the V moves to the past participial AGR head past the VP-initial negative adverb. The proposed structure is given in (12) for (la) and the associated derivation is indicated by the arrows:

AGR' Gianni i AGR l { Spers >

TTie immediate adjacency of Am and PstPrt is thus obtained. (I take VP adverbs to be adjoined to VP; the infinitival form of the auxiliary is used as the non-inflected, "citation" form). The question whether the analysis in (12) is more adequate than the analysis assuming the Aux+PstPrt incorporation is empirical. It cannot be answered unless further empirical material is taken into account. The two analyses make two very different general predictions, though: given a sequence "NP Aux Adv PstPrt", whatever the nature of the adverb involved, the incorporation hypothesis predicts that the order "NP Aux PstPrt Adv" will always be available as well, no matter which position the adverb fills, provided that it is a position lower than the (highest) AGR head. On the other hand, if no process of "Aux+PstPrt" incorporation is assumed to be available, the prediction is that the order "NP Aux PstPrt Adv" can only be obtained if the adverb in question fills the VP initial position; if it files any position higher than VP, the final order of constituents will always be "NP Aux Adv PstPrt". A domain which provides the empirical material we look for is given by the distribution of sentence adverbs like probabilmente, evidentemente, etc. It can be

"Aux+Past Participle" in Italian

31

argued that these adverbs are generated in clause initial position (adjoined to the highest AGRP). This is in fact their most typical location: (13)

a. b.

Probabilmente Gianni telefonerä probably G. will call Probabilmente Gianni ha telefonato probably G. has called

However, if we consider sentences containing a complex tense formed by an auxiliary and a past participle like the one in (13b), we notice that, beside clause initially, the adverb can also be found in a position between the auxiliary and the past participle: (14)

a. b.

Gianni ha probabilmente telefonato G. has probably called Maria e evidentemente partita M. has evidently gone away

What the correct analysis of sentences like (14) might be is a very interesting question per se, which I address in detail in Belletti (forthcoming). As for the issue which concerns us here, we notice that the very existence of sentences like (14) provides us with the test case we were looking for. If the hypothesis assuming the existence of a morphosyntactic process of Aux+PstPrt incorporation in Italian were correct we would expect the following sentences in (15) to be wellformed on a par with (14); but this is not the case: (15)

a. *Gianni ha telefonato probabilmente G. has called probably b. *Maria e partita evidentemente M. has gone away evidently

((15) is possible only if a perceptible pause separates the adverb from the rest of the sentence. See also Jackendoff (1972) on the similar English facts. We disregard this case here, which is amenable to an analysis in terms of right dislocation of the adverb.) With continuous intonation, the sentences in (15) are totally impossible. We can then take these facts as direct evidence against the postulation of the existence of a morphosyntactic process of Aux+PstPrt incorporation in Italian. This in turn implies that negative sentences like those in (1) should be analyzed in the terms of the structure and derivation given in (12). As a final remark we should also point out that a further difference between French pas and the Italian negative adverbs has to be assumed, given the proposed analysis. Beside its being the obligatory specifier of the French NegP

32

A. Belletti

headed by ne, and presumably somehow linked to this property, pas, contrary to the Italian negative adverbs, has no other available location in the clause structure. Specifically, the VP initial adverbial position is not a possible base position for it. (French plus might behave somewhat differently in this respect according to evidence discussed in Pollock (1989) and taken into account in Belletti (forthcoming)). This is how the contrast with Italian illustrated by the ungrammaticality of (11) is accounted for.

NOTE * I thank Luigi Rizzi and Ian Roberts for their comments.

REFERENCES Baker, M. (1988) Incorporation. A Theory of Grammatical Function Changing, Chicago University Press. Belletti, A. (1989) Agreement and Case in Past Participial Clauses in Italian. Ms. Universil6 de Geneve/SNS, Pisa. Belletti, A. (forthcoming), "Generalized Verb Movement". Chomsky, N. (1989) "Some Notes on the Economy of Derivation and Representation", MIT Working Papers, 10, 43-74. Jackendoff, R. (1972) Semantic Interpretation in Generative Grammar. MIT Press, Cambridge (Mass.). Kayne, R. (1989) "Null Subjects and Clitic Climbing", in Jaeggli, O. and K. Safir (eds.), The Null Subject Parameter, Kluwer Academic Publishers, 239-261. Laka, I. (1989) "Constraints on Sentence Negation", MIT Working Papers, 10, 199-216. Moritz, L. (1989) "Apercu de la syntaxe de la negation en francais et en anglais", Memoire de Licence, UniversiU de Geneve. Pollock, J.Y. (1989) "Verb Movement, Universal Grammar and the Structure of IP', Linguistic Inquiry, 20, 365-424. Rizzi, L. and I. Roberts (1989) "Complex Inversion in French", Probus, l, 1-30.

ΤΙ A note on modal passives Hans Bennis University of Leiden

1. INTRODUCTION*

In "A Note on Case Absorption" van Riemsdijk (1982) discusses the construction in (1). He calls it the TI-construction (Te + Inf). Sentences of the type in (1) are also known as modal passives (cf. Hoekstra & Moortgat 1979). (1)

a.

b.

Dat bier is niet te drinken That beer is not to drink 'That beer is undrinkable' Henks argumenten zijn te begrijpen Henk's arguments are to understand 'Henk's arguments are understandable'

Rather convincingly van Riemsdijk demonstrates that the construction in (1) is derived by the application of NP-movemenL In this paper I shall first discuss the properties of this construction with respect to the internal structure of the infinitival clause, the distribution of TI, the obligatoriness of NP-movement and the modal interpretation. After a presentation of the facts, I shall give an outline of a proposal in which all the apparently extraordinary properties of TI follow from one assumption with respect to the status of the inflectional morpheme te. For a more detailed and complete discussion of the modal passive in Dutch I refer to Bennis, Neyndorff & Rats (in prep.).

2. TI AND PASSIVE

As van Riemsdijk argues, TI-constructions of the type in (1) contain a gap in object position. Moreover, he shows that this gap is created by NP-movement. He presents two arguments in favour of this. First, P-stranding is possible if a gap is the result of Wh-movement, but not in the case of NP-movemenL It is clear that P-stranding is not allowed in Ή. This is shown in (2a). The second observation which favours an NP-movement analysis of (1) is the fact that the external argument of the embedded infinitival verb is normally not lexically realized, but may appear in a fry-phrase, as in (2b).

34 (2)

H. Benins a. *Deze blikjes zijn niet uit te drinken These cans are not from to drink b. Dat bier is door Henk te drinken That beer is by Henk to drink "That beer must be drunk by Henk'

Additional arguments in favour of a passive analysis of Ή are: - intransitive verbs which may appear in impersonal passives show up in Ή as well. This is shown in (3a). - ergative verbs do not appear in TI, just as they do not appear in passives. This is shown in (3b). - verbs with Small Clause complements appear in TI without the SC having a lexical subject.1 This is shown in (3c). (3)

a.

In Tilburg is niet te werken In Tilburg is not to work 'It is impossible to work in Tilburg' b. *In Tilburg is niet te vallen In Tilburg is not to fall c. Henk is niet gemakkelijk kwaad te maken2 Henk is not easily angry to make 'It is not easy to make Henk angry'

We may conclude from this discussion that Ή has all the characteristics of a normal verbal passive, with the exception of passive morphology on the verb. If we accept the idea that in passives the external argument of a verb is absorbed by the passivized verb, the question arises as to how the external argument of the verb disappears in the TI-construction. If we can find a plausible answer to this question, the NP-movement properties are in line with the Burzio-generalization.

3. THE DISTRIBUTION OF Π

The distribution of TI is basically identical to the distribution of AP and substantially different from the distribution of VP, IP or CP. Let us discuss some of these distributional properties. - Ή occurs prenominally in NP. This position is typical for AP and not open to any other verbal projection. An example is given in (4). (4)

Dit is een (door / voor Henk) te betalen rekening This is a (by / for Henk) to pay bill 'This is a bill which can/must be paid by Henk'

ΤΙ: Α note on modal passives

35

- Ή occurs in the complement of those raising verbs which allow AP-complements (zijn (to be), lijken (to seem), blijken (to appear)). The raising verb schijnen (to seem) does not take AP-complements, but does take infinitival complements (cf. Bennis 1986). TI is not allowed with schijnen, as is shown in (5). The verb zijn does not take infinitival complements, but does occur with AP-complements. It occurs with Ή quite frequently. (5)

Dat bier is/lijkt/blijkt/*schijnt niet te drinken That beer is/seems/appears/seems not to drink

- Ή appears in constructions in which we find an AP Small Clause with a lexical subject, such as the complement of vinden (to consider) and the complement of absolute met (with), but not in constructions in which only a verbal Small Clause is allowed, as in the complement of perception verbs. (6)

a.

Henk vindt dat bier niet te drinken Henk considers that beer not to drink b. Met nog drie biertjes te drinken trok Henk zijn jas aan With still three beers to drink put Henk his coat on c. "Henk zag de varkens te slachten Henk saw the pigs to slaughter

- another AP-like property of Ή is the fact that the infinitival verb does not have to undergo Verb Raising in complex infinitival constructions. Moreover, Verb Raising in the case of TI is very restricted. The infinitival verb in Ή may be moved to the next verb, in which case acceptability decreases. Moving the infinitive to a more complex verb cluster leads to clear ungrammaticality (cf. van Riemsdijk (ibid.), note 1) (7)

a.

Henk Henk b. ?Henk c. Henk Henk d. *Henk

vindt dat dat bier niet te drinken is (-VR) thinks that that beer not to drink is vindt dat dat bier niet is te drinken (+VR) zegt dat dat bier niet te drinken schijnt te zijn (-VR) says that that beer not to drink seems to be zegt dat dat bier niet schijnt te zijn te drinken (+VR)

The marginal acceptability of (7b) does not argue against an adjectival status of Ή. Adjectives do not undergo Verb Raising, with the exception of those adjectives which look like participles, such as behept (afflicted) and geschikt (fit). Those adjectives can undergo Verb Raising only minimally, as the TIconstructions in (7). This is illustrated in (8).

36 (8)

H. Bennis a.

dat Henk daarmee behept is that Henk there-with afflicted is b. ?dat Henk daarmee is behept c. *dat Henk daarmee schijnt te zijn behept

The parallel between (7) and (8) suggests once more that TI is a construction with adjectival behaviour. - A final property which points to the non-verbal character of the TI-construction is the fact that no aspectual or temporal auxiliaries appear in TI, as is shown in (9). (9)

*Dat probleem is niet te hebben / te zijn begrepen That problem is not to have / to be understood

4. THE INTERNAL STRUCTURE OF Ti

In all relevant respects the internal structure of Ή is similar to the internal structure of IP and completely different from the structure of AP. This is apparent from the nature of the categories (te, V, ..), the order of constituents, the type of adverbial modification and the lack of adjectival morphology. The last two issues are illustrated by a comparison between TI and a comparable -boar adjectival in (10)-(12). These facts strongly suggest that TI is verbal with respect to its internal structure. (10)

een zeer kwetsbare jongen Λββη zeer te kwetsen jongen a very vulnerable boy / a very to hurt boy

(11)

een kwetsbare jongen, een kwetsoaar meisje /*een te kwetsene jongen a vulnerable boy a vulnerable girl a to hurt boy

(12)

een onkwetsbare jongen /*een on-te-kwetsen jongen, *een te onkwetsen jongen an invulnerable boy an un-to-hurt boy

5. THE MODALITY OF TI

The ΤΙ-construction in (13a) receives two interpretations, as indicated in (13b,d). It is interesting to note that one interpretation of (13a) corresponds to the epistemic interpretation of (13b), as is shown in (13c), whereas the the other interpretation of (13a) corresponds to the deontic interpretation of (13d), as is illustrated in (13e).

ΤΙ: Α note on modal passives (13)

a. b. c. d. e.

37

Deze som is op te lessen This sum is to solve This sum can be solved It is possible that this sum is solved (not: Someone is able to solve this sum) This sum must be solved Someone has to solve this sum (not: It must be the case that this sum is solved)

We may thus argue that Ή has both an epistemic and a deontic interpretation. There are several differences corresponding to the two interpretations in (13). First of all, the distribution determines to a large extent which interpretation has to be selected. In the complement of raising verbs and in the complement of vinden (consider) the Ή is most naturally, or, in the latter case, even obligatorily interpreted in the epistemic CAN-interpretation. In prenominal position and in the complement of absolute met (with) the MUST-interpretation is favoured or obligatory, respectively. I shall not discuss these observations here. Another interesting fact is that the external argument of the verb is optionally expressible in a fry-phrase in the MUST-interpretation, but in a /or-phrase in the CAN-interpretation. Consequently the sentences in (14a,b) are not ambiguous. (14)

a.

Deze som is door Henk op te lossen 'This sum must be solved by Henk' b. Deze som is voor Henk op te lossen 'This sum can be solved by Henk' c. *In Tilburg is door gastarbeiders hard te werken 'In Tilburg foreign workers must work hard'

As is shown in (3), TI is possible with intransitive, non-ergative verbs. However, this is only true for the CAN-interpretation and not for the MUSTinterpretation. This is illustrated in (14c). The modality of Ή and the properties of the two modal interpretations must follow from a property of Ή. Because modal verbs take fe-less infinitival complements it seems reasonable to take the modality of Ή to be a consequence of the presence of te.

6. ANALYSIS

In contrast to other modal constructions, Ή contains the I-constituent te (to). Given most recent analyses (Chomsky 1986, Pollock 1989, Bennis & Hoekstra 1989a) te is the head of IP. te is a verbal head taking a VP-complement. Now suppose that te, apart from being [+V], can be either [-N] or [+N]. If the [-N] option is selected, te is a verb-like head of IP. This is the situation in 'normal'

38

H. Bennis

infinitives. In that case te must be part of a Tense-chain (Bennis & Hoekstra 1989b). If the [+N] option is selected the fe-projection will be adjectival. The assumption that te is [oN] gives us a straightforward account for - the adjectival distribution of Ή (te is [+V.+N]) - the verbal internal structure of TI (te takes a VP-complement) - the non-occurrence of temporal auxiliaries in TI (adjectives do not belong to a Tense-chain) - the non-occurrence of Verb Raising of te+V (adjectives do not belong to a Tense-chain) This analysis is in several respects similar to Van Riemsdijk's analysis of Ή. van Riemsdijk claims that Ή is a [+V]-projection, which is somewhere in between a verbal and an adjectival projection. An important question is how the non-appearance of the external argument in the complement of te can be made to follow from the assumption that te is [oN]. In regular passives the external argument of the verb becomes an implicit argument as a consequence of absorption by V. We might consider the fact that the distribution of passive participles is basically similar to the distribution of adjectives to be a consequence of this absorption. If we hypothesize that the absorption of an external thematic role turns a category into a [+N]-category, we are able to relate the [+V,+N] status of te to the absence of an external argument in the complement of te. I shall leave the technicalities concerning the way in which the external argument of the embedded V can be absorbed by te out of consideration here. The general idea is that [+V]-categories such as verbs and te can become [+V.+N] by absorption of an argument. If that is correct, the passive behaviour of Ή follows. The Burzio generalization expresses that the verb is not able to assign objective case if it does not assign its external role to an argument position. The fact that ergative verbs do not occur in TI follows as well, te can only become [+N] by absorbing an external argument In ergatives no external argument is present. The final question is the most complicated one: how can we relate the modality of Ή to the analysis given above? The complexity of the question derives from the fact that no complete and satisfactory account of 'normal' modality is available.3 Without explicit motivation I shall assume that modal verbs are basegenerated in I and that the subject of a modal verb is raised from the complement VP to [SPEC.IP]. This implies that I shall assume that the external role of the deontic modal is an adjunct θ-role. We can now take modality to be a possible lexical property of verbal constituents in I, i.e. modal verbs and te. The differences between the two interpretations of Ή follow from the theory. The non-occurrence of intransitive verbs in the MUST-interpretation (deontic) (cf.(14)) follows from the θ-criterion since the adjunct θ-role of the modal

77: A note on modal passives

39

must be assigned to an NP, no such NP being available in a Tl-construction such as (3a). It is a central claim of Roberts (1987) that in verbal passives V assigns case to the absorbed external argument. We might then assume that a case-marked implicit argument can be lexically represented in a fry-phrase. The other situation in which we find implicit arguments which are optionally realized in a PP are APs with /or-PPs. Adjectives do not assign case. Suppose that an implicit argument is realized in a fry-phrase if the implicit argument is casemarked and in a /or-phrase if the implicit argument has no case. If that is correct, we expect te to be able to assign case with deontic Tls and not with epistemic Tls. We now find a situation which closely resembles the Burzio generalization. Deontic Tls assign an external (adjunct) θ-role and case to their implicit argument, whereas epistemic Tls assign neither an external θ-role nor case. If we make use of the theory on ergativity (Burzio 1986) and of Roberts' ideas on case assignment in passives in a somewhat unorthodox way, we are able to relate all the properties of Ή discussed here to the assumption that te becomes [+N] if it has absorbed the external argument of its complement.

NOTES * I gratefully acknowledge the many contributions made to this article by Joan Neyndorff and Sabine Rats. 1. In this respect the Tl-construction differs from -boar adjectives, which do not appear in this construction: *kwaadmaakbaar. It has often been claimed that the passive-like process in -boorformation is a lexical process. The occurrence of (3c) indicates that Tl-formation is not a lexical process, but in all relevant respects similar to the verbal passive. 2. For arguments against an easy-to-please derivation of (3c), I refer to van Riemsdijk (1982). One major argument against an analysis of the infinitival clause as an infinitival relative within AP is the fact that infinitival relatives of this type are always introduced by the complementizer om (cf. Bennis & Wehrmann 1987). 3. For modal verbs in Dutch one might claim that they are main verbs with an infinitival complement of the type CP, IP or VP. Alternatively, one might argue that modal verbs are basegenerated in I as auxiliary verbs. I shall adopt the latter analysis here. Furthermore, it is not clear whether the subject of a modal is base-generated as the subject of the modal or as the subject of the complement of the modal. The sentence in (ia) may have two representations. (i)

a. b. c.

Henk kan naar GLOW gaan Henk kan to GLOW go Henk; kan [PRO; naar GLOW gaan] Henlq kan [4 naar GLOW gaan]

It might even be argued that an epistemic interpretation of (ia) corresponds to the raising structure in (ic), whereas the deontic interpretation of (ia) corresponds to the control structure in (ib). Analysing the epistemic interpretation as in (ib), we have to assume that only in constructions with modals expletive PRO is possible. The contrast in (ii) strongly suggests that at least the epistemic interpretation corresponds to a raising structure.

40

H. Bennis (ü)

a. b.

Er kan [in Tilburg worden gedanst] liiere can in Tilburg be danced *Er wordt geprobeerd / Het is leuk [in Tilburg te worden gedanst] There is tried / It is nice [in Tilburg to be danced]

REFERENCES Bennis, H. (1986) Gaps and Dummies, Foris, Dordrecht. Bennis, H. and T. Hoekstra (1989a), Generative Grammatics, Foris, Dordrecht. Bennis, H. and T. Hoekstra (1989b), "Why Kaatje was not heard sing a song", in Jaspers, Klooster, Putseys and Seuren (eds.), Sentential Complementation and the Lexicon, Foris, Dordrecht, 22-41. Bennis, H. and P. Wehrmann (1987) "Adverbial Arguments", in Beukema and Coopmans (eds.), Linguistics in the Netherlands 1987, Foris, Dordrecht, 1-11. Bennis, H., J. Neyndorff and S. Rats (in prep.), "Modale Passieven". Burzio, L. (1986) Italian Syntax, Reidel, Dordrecht. Chomsky, N. (1986) Barriers, MIT Press, Cambridge (Mass.). Hoekstra, T. and M. Moortgat (1979) "Passief en het Lexicon", Forum der Letteren 21, 138-162. Pollock, J.-Y. (1989) "Verb Movement, UG and the Structure of IP", Linguistic Inquiry 20, 365-424. Riemsdijk, H. van (1982) "A note on Case Absorption", Wiener Linguistische Gazette 28-29, 72-83. Roberts, I. (1987) The Representation of Implicit and Dethematized Subjects, Foris, Dordrecht.

The syntax of nominatives in SOV Germanic* Hans den Besten University of Amsterdam

1. INTRODUCTION

In earlier work I have claimed that certain arguments put forward against assuming Move NP in SOV languages such as Dutch and German are invalid (den Besten 1981). The differences that hold in English between adjectival (statal) and syntactic (verbal) passives also hold in Dutch and German and it is shown how the difference in Move NP potential between English and SOV Germanic can be derived. The latter mainly derives from a basic distinction in Case-assigning properties between SVO and SOV languages. Accusative (Objective) Case will go to the NP nearest the verb. Since it is the Accusative and not the Dative (Oblique Case) that must disappear under passivization, an SVO language with double NP constructions will freely passivize the Indirect Object NP, whereas in SOV languages an IO NP will become the passive Subject only if it is lexically marked for Accusative Case in the active structure. Furthermore, prepositional passives are excluded in German and Dutch because a combination of 0V and prepositional word order does not meet the requirements for reanalysis of P+V sequences. Finally, the near-absence of Raising passives is due to the fact that nearly all of the pertinent ECM verbs are Verb Raising verbs, which always (Dutch) or practically always (German) resist participle formation, which happens to be a morphological prerequisite for verbal passives. In subsequent papers, eventually incorporated in den Besten (1985), I came to the conclusion that a DO NP (or: (NP.V)) does not always have to move to the syntactic Subject position in passive and ergative structures, due to the fact that the SOV languages Dutch and German are able to assign "external" Cases (Nominative or Accusative) to "case-less" NPs in situ through chains of governers: a chain (COMP, V) in the case of simple passives and ergatives (or, to update the analysis a little: (I, V)), a chain (COMP, V lt ... Vn), where n > 1 in Raising contexts (or I instead of COMP), and a chain (V1( ..., Vn) in ECM structures. Although I still think that Case-assignment through chain-government is available in Dutch and German syntax, and even that such a mechanism is universally available for I-final SOV languages (without knowing yet how to derive this parameter), I now believe that we can do without Move NP al-

42

H. den Besten

together, and that apparent cases of Move NP in Dutch and German can be derived in a different way. This does not imply that we should view the German-Dutch dynamic passive as a lexical passive, because the arguments of den Besten (1981) still stand. We rather have to say that a defining characteristic of the verbal passive is: absence of structural Case (cf. Chomsky 1981). In languages with Case-assignment through government, i.e. in languages without chain-government (e.g. English), a case-less NP has to move (usually) to the structural Subject position. In languages with chain-government (the SOV languages) no such movement is needed. These considerations again prove that to conclude from the absence of (some cases of) Move NP that the pertinent passive is lexical is a mistake - one of the topics addressed in den Besten (1981). In the following sections I will first discuss some technical and/or aesthetic problems with the chain-government analysis of den Besten (1985) (section 2). In section 3 I will show that problems even get worse if we consider the possible interactions between Move NP and the Adjunction or Scrambling operation which must be assumed for SOV Germanic syntax. Fortunately, all of these problems vanish, if we give up Move NP altogether and attribute the apparent cases of Move NP to Adjunction, which happens to make some correct predictions. Further discussion is relegated to section 4. As will become clear in the course of this paper, I concur with Broekhuis (1988: 68-72) in deciding against Move NP in favor of Adjunction, although I do not necessarily subscribe to all of his arguments.

2. CHAIN-GOVERNMENT AND MOVE NP

In den Besten (1985: (51)) the following definitions are introduced for Caseassignment under chain-government: (1)

a.

b.

If NPj is governed by a category α which cannot or may not assign Case, NPi will acquire its Case from the first Case-assigner up by which it is chain-governed. α chain-governs β iff α governs γ1( YJ governs γ2 γ^ governs γη, and γη governs β (η > 1).

These definitions presuppose a definition of government according to which an element Υ governs the head of a phrase XP, provided XP is governed by Υ (den Besten 1985: (11)-(12)). For a possible improvement, excluding [+N] elements from such chains of govemers, see Broekhuis (1988: 71-72). Chain government might be further restricted by adding unidirectionality of government to the definition, which implies that the pertinent remarks following (59) in den Besten (1985), including n. 6, have to be revised. But I will not go into that matter here.

The syntax of nominatives in SOV Germanic

43

Chain-government permits the occurrence of Nominative NPs in DO position. Section 3.2. of the paper under consideration tries to show on the basis of the wat voor (O)/was für (G) Split phenomenon that this may indeed be the case. (Henceforth D will be used for 'Dutch' and G for 'German'.) Let me add here that I now believe that the wat voor/was für test has to be revised, because the constellation of facts is more complicated than I was willing to admit in 1985. Yet, in my opinion a revised test will also lead to the conclusion that in sentences like (2) below the remnant NP is in DO position: (2)

D

Wat zijn er daar voor wegen aangelegd? What have-been there there for roads constructed? 'What sort of roads have been constructed there?'

It is furthermore assumed that there are several ways to deal with a Subject position (or: Spec,IP, to use more recent terminology) which is not assigned Nominative Case. Either the dummy locative er 'there' licenses the empty Spec,IP, most probably from an enclitic position following C (= Vf in main clauses), or an argument NP moves into SpecJP. The latter option is supposed to be shown in example (3), where the Dative NP is assumed to be in SpecJP, provided er is not expressed: (3)

D

.... dat (er) mijn oomDAT een prijsNOM toegekend is ..., that (there) (to) my uncle a prize awarded has-been

Now note that it is in principle possible to first assign Nominative to an NP in DO position and then move it to SpecJP. This implies that there are two ways to derive example (4) unless further restrictions are introduced: (4)

G

..., daß dieser PreisNOM meinem OnkelDAT i gefällt .... that this prize (to) my uncle pleases

Either Nominative is assigned to SpecJP under government and the DO NP moves there to acquire Case, or Nominative is assigned to the DO NP under chain-government and the Nominative NP is then moved into the Spec,FP position. This indeterminacy in the theory is discussed on pp. 44-45 of den Besten (1985), but no definitive stance is taken until section 5.2, where it is shown that the problems multiply as soon as Raising structures are taken into account For practical reasons it is assumed that Nominative Assignment applies at S-structure, so that the DO NP either moves and acquires Nominative in SpecJP position or stays in DO position and is assigned Nominative there (provided the case-less SpecJP can be saved by er or by another argument NP moving there). However, note that this assumption can hardly be called a principled solution.

44

H. den Besten

Yet, there is more that can be criticized in the chain-government hypothesis as expounded in den Besten (1985). First, note the dual status of Dutch er in the above examples. On the one hand it is clear that expletive er is related to indefinite Nominative Subjects, because er is out if the relevant Subject is definite (or Accusative). On the other hand er is supposed to license an empty Spec,IP. These assumptions do not necessarily exclude a configuration in which er occupies the enclitic position and is related to an indefinite Nominative Subject, while another NP is in Spec,IP - unlike what is assumed in den Besten (1985). Since er is also used in impersonal passives of the type Er werd gelachen '(lit.) there was laughed', the possibility that example (3) has two syntactic analyses (as is suggested above) cannot be rejected out of hand. Finally the chain-government hypothesis as it stands requires that a definite DO NP in passive or ergative structures be moved to SpecJP unless another argument NP is available. One may wonder whether it should not be possible to keep such DO NPs in place even in the absence of another argument NP. Below it will be shown that such structures exist indeed. What the above considerations boil down to is that the chain-government hypothesis as it stands is not simple enough. The next section will provide one more argument to that effect, while offering at the same time a solution that will save chain-government.

3. CHAIN-GOVERNMENT AND ADJUNCTION

In recent years research on the so-called Scrambling (or: Adjunction) phenomenon in SOV Germanic, which was started by de Haan (1979), has gained momentum; witness studies such as Hoekstra (1984), Bennis (1986), Broekhuis (1988), and Webelhuth (1989). There are various names for the pertinent operation, and I will use the relatively neutral name of Adjunction. The basic idea is that XPs may leave their D-structure positions and adjoin to the left of a dominating projection (VP, IP; maybe also I'?). Such XPs will be argument NPs and PPs in the first place, although also non-argument XPs (Small Clause predicates, adjuncts) seem to be able to undergo Adjunction provided they contain a deictic element. Due to Adjunction, XPs may invert with sentence adverbials and with one another without undergoing Topicalization. Consider example (5): (5)

D

Zij zal jouw liedj waarschijnlijk t{ kopen. She will your song probably r buy

If jouw lied is not moved it acquires a focal reading, as is the case in (6): (6)

D

Zij zal waarschijnlijk jouw lied kopen.

The syntax of nominatives in SOV Germanic

45

That is why (unspecific) indefinites have to stay in situ, i.e. an indefinite object such as een lamp 'a lamp' prefers the order exemplified in (6). Given such considerations we have to assume that the Dative NP meinem Bruder in (7a) and the Nominative NP deine Geschichten in (7b) cannot be in their D-structure positions: (7)

a.

G

b.

G

..., da deine GeschichtenNOM meinem BruderDAT nicht gefielen ..., that your stories (to) my brother not pleased ..., da meinem BruderDAT deine GeschichtenNOM nicht gefielen

These examples have been taken from den Besten (1985: (25)), where it is assumed that the pertinent NPs are in their D-structure positions. However, the position of nicht indicates the opposite: they have been adjoined - probably to Γ if the other NPs are in SpecJP. This does not mean that the pertinent NPs cannot stay in their D-structure positions. They can, and will then acquire a focal reading - also if we substitute other adverbials for nicht. Similar observations can be made for passive structures. (See also Broekhuis (1988: 69-70) who uses similar ergative structures in the context of a different type of argumentation against Move NP and in favor of a universal use of Adjunction in Dutch syntax.) The above discussion only shows that things are more complicated than I thought when I wrote my 1985 paper. But things are even worse. A grammar that makes use of chain-government and Move NP and Adjunction cannot exclude an analysis according to which it is the Dative in (7a) and the Nominative in (7b) that have been moved to SpecJP while the other NPs have undergone Adjunction. Thus there are in principle two possible analyses for (7a) and (7b) respectively, as is indicated under (7') (the adverbial is left out): (7·)

a. a. b. b.

1 2 1 2

C C C C

[„ fo h [,ρ

ΝΟΜ; NOMj DATj DATj

[r [jp [,. [,ρ

DATj DATj NOM, NOMi

[,. [,. [r [,.

[w t, [v. i,· V ]] I ]]] [w tj [v- ',· V ]] I ]]] [yp tj tv. t, V ]] I ]]] [w tj [v. r,. V ]] I ]]]

Even if we do not feel disturbed by adjunction to Γ, we have to admit that there will be a proliferation of possible syntactic structures which cannot be justified, and which is caused by the fact that there are two movement rules: one with a fixed landing site (Move NP) and one with a large set of possible landing sites (Adjunction). Now note that more than one XP may leave the VP via the Adjunction operation, as is clear from the following examples:

46 (8)

H. den Besten a.

G

b. G

..., daß sie meinem OnkelDAT das BuchACC nicht /DAT /ACC ..., that she (to) my uncle the book not fDAT iACC zuschicken wollte send wanted ..., daß sie das BuchACC meinem OnkelDAT nicht fDAT /ACC zuschicken wollte

Therefore the solution to the problem indicated by (71) is simple: all structures under (7') are wrong. In a grammar with chain-government and Adjunction there is no need for Move NP (cf. Broekhuis 1988: 72). This conclusion also rids us of the problems mentioned in section 1. If the above idea can be upheld, it follows that in the grammar of a language with Case-assignment via government as well as via chain-government, but without Move NP, Case is assigned to theta-marked positions only. Such Case-marked NPs (also Nominatives) may leave the VP and adjoin to a higher projection, which will yield Move NP-like effects. Note that we could get rid of adjunction to if we assume that external Subjects are base-adjoined to VP, and so may have to undergo Adjunction, too. But this is an issue that does not concern me here. More important is the question of whether the evidence for Move NP presented in den Besten (1985) can be reinterpreted in accordance with the above hypothesis. In so far as I can see, no problem will arise here. The sole evidence for Spec.IP as a landing site for Nominatives in NOM DAT passive and ergative structures mentioned (section 3.2. in that study) is the observation that such Nominatives do not allow wat voor/was für Split However, in den Besten and Webelhuth (to appear) it is noted that it is not possible to extract elements out of XPs that have been moved off their D-structure positions and adjoined to higher projections, and hence the wat voor/was für Split facts are not an argument for Move NP any longer. The evidence for Datives moving to Spec,IP presented in den Besten (1985: 3.3.) is even weaker. The pertinent argument runs as follows: "(...) if we follow de Haan (1979) and van Riemsdijk (1978) in assuming that there is an R-landing site between the Subject and the Indirect Object position, the following examples demonstrate that the Dative may show up in (NP,S) position;" (p. 41) (9)

a.

D

b.

D

.... dat Mary er{ het boek f; voor gegeven is .... that Mary itj the book t{ for given has-been ..., dat Mary cr; een mooi boek t{ voor gegeven is ..., that Mary itj a nice book t{ for given has-been (= (49a) and (49b) of den Besten (1985))

The syntax of nominatives in SOV Germanic

47

This argument is not particularly strong because the R-pronoun er may also show up between the Indirect Object and the Direct Object - contrary to what has been claimed in the earlier literature on R-pronouns (cf. Bennis 1986). Compare the following active examples (where zy is 'they' and hebten 'have'): (10)

a.

D

.... dat zij Mary er{ het boek r; voor gegeven hebben ..., dat zij Mary er{ een mooi boek t{ voor gegeven hebben

Therefore the examples in (9) do not militate against the assumption that Mary is not in Spec.IP. We may now turn to an interesting consequence of the hypothesis under consideration: by assuming that the NOM DAT order in passive and ergative structures is derived by Adjunction rather than by Move NP, we can reduce one of the lesser known properties of these alleged Move NP structures to properties of Adjunction. Lenerz (1977) in his book on the order of NPs in German has shown that if nominal Objects are in their unmarked order (i.e. Dative before Accusative) either Object may b e - i n terms of "communicative dynamism" - the Rheme (Rhema) of the sentence, whereas in the marked order (Accusative before Dative) the Indirect Object has to be the Rheme. Something similar holds for passive and ergative constructions. In this case the unmarked order is Dative before Nominative, which permits rhematic stress-assignment to either NP. In the marked order the Dative NP must be rhematic. Some examples taken from Lenerz (1977) may illustrate this point. Rhematic stress will be indicated by upper case for the relevant syllables. First, two examples for double Object constructions: (11)

(12)

Q

Wem hast du das Geld gegeben? To-whom have you the money given? Ich habe dem KasSIErer das Geld gegeben. I have the cashier the money given Ich habe das Geld dem KasSIErer gegeben

A

a.

A

b.

Q

Was hast du dem Kassierer gegeben? What have you the cashier given? a. Ich habe dem Kassierer das GELD gegeben. I have the cashier the money given b.?*Ich habe das GELD dem Kassierer gegeben

A A

(Examples taken from Lenerz (1977: 43).) The following examples from Lenerz (1977: 115) illustrate the point for ergative (and passive) structures:

48 (13)

(14)

H. den Besten Q

Wem ist der Coup gelungen? To-whom has the coup succeeded? Ich glaube, daß der Coup einem BaRON gelungen ist I think that the coup to-a baron succeeded has Ich glaube, daß einem BaRON der Coup gelungen ist

A

a.

A

b.

Q

Was ist dem Baron gelungen? What has to-the baron succeeded? a. Ich glaube, daß dem Baron ein COUP gelungen ist I think that to-the baron a coup succeeded has b. *Ich glaube, daß ein COUP dem Baron gelungen ist

A A

Note that I have somewhat simplified the central thesis of Lenerz (1977). It is evident that we may interpret the unmarked orders DAT ACC and DAT NOM as sequences corresponding to the respective D-structure orders (although both NPs may have undergone Adjunction.) Now if both marked orders (i.e. ACC DAT and NOM DAT respectively) are derived via Adjunction (rather than via Adjunction in one case, and via Move NP in the other case) the communicative properties of the NOM DAT order in (13Aa) and (14Ab) will be determined by the same module of grammar that is responsible for the communicative properties of the ACC DAT order in (11 Ab) and (12Ab) Note that it cannot be excluded that in a system with Adjunction and Move NP a similar module determines the communicative properties of the products of both Adjunctions and Move NP. Thus, my sole claim is that in a system with Adjunction only, all cases of inversion by Adjunction will necessarily be affected by the same "communicative" module. Additional data showing that passive and ergative structures evidence properties of Adjunction can be found in Broekhuis (1988), who demonstrates that definite Nominative NPs prefer to move out of the VP while indefinite Nominative NPs stay inside the VP. I only disagree insofar that I do not believe that the pertinent examples can be used as an argument against Move NP. Yet, his data shows again that if we reinterpret Move NP in the Germanic SOV languages as Adjunction, the properties of the alleged cases of Move NP will derive from the properties of Adjunction.

4. DISCUSSION

We may ask ourselves whether there is more data pointing into the same direction. Broekhuis (1988: 72) claims that the absence of adjectival Raising constructions in Dutch is a good argument against Move NP. However, the absence of such structures is also a problem for a grammar without Move NP, since it must be explained why Nominative Case cannot be assigned. The

The syntax of nominatives in SOV Germanic

49

answer may be that chain-government does not apply in such cases (cf. section 2). Furthermore there are Raising constructions in Dutch in which Move NP seems to have applied: (15)

..., dat Hansj geacht wordt [/; dat vroeger al bewezen ..., that Hans supposed is / that before already proven te hebben] to have

Yet, this example does not prove that the Nominative is in Spec,IP. First of all, as is pointed out in den Besten (1985: n. 8), indefinite NPs may stay inside the infinitival clause: (16)

Er wordt geacht [geen verschil te bestaan tussen man There is supposed no difference to exist between man en vrouw] and woman

These properties can be easily derived if we assume that the surface position of Hans in (15) is derived by (Long) Adjunction. This is in accordance with the fact that in (15), even the embedded Direct Object may be moved up to the matrix clause: (17)

..., dat HanSj datj geacht wordt [f; fj vroeger al bewezen te hebben]

(For Long Adjunction and the geacht worden construction see den Besten et al. (1988) and den Besten and Rutten (1989).) Although the above considerations do not yet solve all of the problems of the analysis of the geacht worden construction, it seems reasonable to assume that we can do without Move NP. Yet, the hypothesis put forward in this paper creates new problems for the theory of control - but that is a topic for another paper. Therefore I would like to warn those who are willing to buy the above hypothesis: caveat emptor.

NOTES * This paper repeats, and elaborates upon, remarks concerning Move NP and Adjunction in den Besten (1989). Unfortunately, while writing the pertinent pages I overlooked the partially similar argumentation in Broekhuis (1988: 68-72), which I can only make amends for by quoting him in this study.

50

H. den Besten

REFERENCES Bennis, H. (1986) Gaps and Dummes, Foris, Dordrecht. Besten. H. den (1981) "A Gase Filter for Passives", in A. Belletti, L. Brandi and L. Rizzi (eds.), Theory of Markedness in Generative Grammar. Proceedings of the 1979 GLOW Conference. Pisa, Scuola Normale Superiore, 65-122. Besten, H. den (1985) "The Ergative Hypothesis and Free Word Order in Dutch and German", in J. Toman (ed.). Studies in German Grammar, Foris, Dordrecht, 23-64. Besten, H. den (1989) Studies in West Germanic Syntax, Doctoral dissertation, KUB, Tilburg. Besten, H. den, et al. (1988) "Verb Raising, Extrapositie en de Derde Constructie", unp. research report, University of Amsterdam. Besten, H. den, and J. Kutten (1989) "On Verb Raising, Extraposition and Free Word Oder in Dutch", in D. Jaspers et al. (eds.). Sentential Complementation and the Lexicon. Studies in Honour of Wan de Geest, Foris, Dordrecht, 41-56. Besten, H. den and G. Webelhuth (to app.) "Stranding", in G. Grewendorf and W. Stemefeld (eds.), Scrambling and Barriers, Benjamins, Amsterdam/Philadelphia, 77-92. Broekhuis, H.J.W.M. (1988) "A- en A'-binden in het Nederlands", MA thesis. University of Amsterdam. Chomsky, N. (1981) Lectures on Government and Binding, Foris, Dordrecht. Haan, GJ. de (1979) Conditions on Rules. The Proper Balance between Syntax and Semantics, Foris, Dordrecht. Hoekstra, T. (1984) Transitivity. Grammatical Relations in Government-Binding Theory, Foris, Dordrecht Webelhuth, G. (1989) Syntactic Saturation Phenomena and the Modern Germanic Languages, PhD Diss., UMass.

On the COMP of Relatives' Reineke Bok-Bennema University of Groningen

1. QUESTIONS ON RELATIVES

The present squib will be based on the assumption that in relatives such as those in (1) the complementizer "takes over" the function of a relative pronoun (cf. Pesetsky (1982), Chomsky (1981), Kayne (1984)). (1)

a. b.

the linguist that founded GLOW la fille que tu as invite

My purpose is to answer a number of questions raised by this postulated "function shift". The first of these concerns the possibility for the shift to apply at all. I relate this possibility to the specific nature of relative complementizers, so that the first question can be formulated as in (2). (2)

question (1): What are the intrinsic properties of relative complementizers which allow them to take over the function of relative pronouns?

The contexts in which the function shift applies can be characterized as "recoverability contexts". Recoverability contexts are those in which the relative clause is preceded by a referential antecedent and in which the relativization gap can be construed as the trace of a pure relative pronoun. Thus, neither free relatives, which lack a referential antecedent, nor relatives where the gap must be interpreted as the trace of a fronted relative pronoun with pied-piped material allow for function shift. This raises my second question, formulated in (3). (3)

question (2): Why is the function shift restricted to recoverability contexts?

Another fact about the function shift is that it saves the constructions in which it applies from the the effects of the Doubly Filled COMP Filter (DFCF) of Chomsky and Lasnik (1977). DFCF effects, however, also occur in relatives without function shift (as well as in other ννΛ-constructions). One of my purposes is to present an analysis which is general enough to explain the DFCF

52

R. Bok-Bennema

effect in shifted and in non-shifted relatives (other ννΛ-constructions will be considered briefly at the end). I formulate this as question (3) in (4). (4)

question (3): Is there a uniform explanation for the DFCF effect in relatives with function shift and relatives where the shift does not apply?

My fourth and last question concerns the variation among languages w.r.t. the function shift. Whereas in English, which I will consider first, the shift is optional, in other languages it must apply once the recoverability context is present. Clear examples of languages with obligatory shift are the Romance languages (see e.g. Cinque (1982), Schroten (1984) and Smits (1989)).1 On the other hand, there are also languages - one example is Dutch - in which it does not apply at all. (5) contains the question under consideration. (5)

question (4): Is there a simple explanation for the language variation w.r.L the function shift in relatives?

The questions raised here nave not been neglected in the literature. In fact, all of the works mentioned above contain answers to one or more of them. What is more, most of these anwers have greatly influenced the analysis to be developed here. A new aspect of this analysis is, however, that it shows how all four questions can be answered in a closely interrelated way and in line with the nowadays standard conceptions on phrasal structure (cf. Chomsky (1986)). I must mention that I will confine myself to tensed relatives: infinitival relatives will not be considered.

2. THE COMPLEMENTIZER OF RELATIVES

It cannot be denied that if a complementizer can take over the function of a relative pronoun it must be able to act as an operator. I propose, however, to go a little further and to assume that the complementizer of relative clauses is an operator by definition. (6) highlights this proposal, which, incidentally, implies that relative clauses are generated as "operator phrases". (6)

The complementizer of relatives is inherently an operator

Suppose now that only elements which are lexically specified as operators such as wA-words, quantifiers and - now - relative complementizers, but not e.g. normal complementizers or personal pronouns - are able to act as such. It is then, basically, the operator status of the relative complementizer which enables it to assume the same role as a relative pronoun. (6), therefore, contains

On the COUP of relatives

53

my answer to the first question of the preceding section. Apart from being operator phrases, relative clauses are referentially connected to the antecedent of the relative. It is standard practice to assume that this connection is established by a rule of wA-movement, which preposes a relative pronoun to spec.CP, coupled with a rule of predication (Williams 1980), which equates the index of the antecedent to that of the relative pronoun (and its trace). I adopt this standard assumption, but, in addition, I also adopt the idea presented in Kayne (1984) that there is a rule which can assign the index of the antecedent to the complementizer of the relative.2 This rule may be considered as forming part of the rule of predication. It is plausible that predication assigns the index of the antecedent to the relative clause (CP) as a whole and that from there on it percolates down to the fronted relative pronoun, as a consequence of which the relative pronoun's index is equated to that of the antecedent Suppose now that index percolation not only involves the relative pronoun but also (optionally in English, see below) the head of the relative, i.e. the complementizer (7). This gives us the co-indexation of the complementizer with the antecedent (7)

Percolate the predication index to C

As will become clear below, it is crucial that the predication indices be present at S-structure.3 Given the assumptions made so far, it is possible to derive S-structures such as the one in (8), with that as operator. (8)

the linguistSj who; thatj I met X; in G ttingen

As who, being a ννΛ-word in spec.CP, is also an operator, (8) contains two operators, which are both co-indexed with x. Notice, however, that that, which occurs in C, is closer to χ (in terms of c-command) than who (in spec.CP); cf. (9). (9)

54

R. Bok-Bemema

It is plausible that when a configuration like the one in (9) arises, the operator closest to the variable is its actual operator. This has as a consequence that the other operator - who, here - is vacuous. Vacuous operators are excluded in Universal Grammar, either by Chomsky's (1982) Constraint on Vacuous Operators (VOC), or by Koopman and Sportiche's (1982) Bijection Principle. who, therefore, has to be eliminated in some way. Let us suppose that Universal Grammar has an S-structure rule which deletes vacuous operators that arise during derivations. I formulate this rule in (10). (10)

Delete a vacuous operator at S-structure

(10) eliminates who, as required.4 We have now obtained a (function shifted) relative with only one element in "COMP" (=pre IP), in accordance with the DFCF effect. As we shall see, it is also possible to explain the DFCF effect in other English relatives using the assumptions we have made. Moreover, the restriction of function shift to recoverability contexts will be shown to follow quite naturally. Consider, first, the S-structure in (11). (11)

the linguistSj who; that I met X; in Göttingeh

The only difference between (8) and (11) is that in (11) the predication index has not percolated to C: I am assuming that (7) is optional in English. As that is non-referential it has no index of its own, so that accidental co-indexation is excluded. Given (6), that is nevertheless an operator. It is, however, an operator that does not bind anything, i.e. a vacuous one. The VOC (or the Bijection Principle) requires it to delete, which it can by the rule deleting vacuous operators in (9). The result (the linguist who I met in Göttingen) is correct and respects the DFCF effect Consider now the - non-indexed - structure in (12), which contains a relative pronoun with pied-piped material and is therefore a non-recoverability case. (12)

the linguists [to whom] that I spoke

in Nijmegen

W.r.t to the indexing of that there are two possibilities: either percolation to C does apply or it does not. If it does not, that is a vacuous operator, just as in (11), and deletes. But if it does apply, that is also vacuous and must delete as well. The reason for this is that the index which that acquires from the antecedent is not the same as the one on the trace of the fronted relative constituent. Predication equates the index of the relative pronoun with that of the antece-

On the COM? of relatives

55

dent, but not that of the pied-piped phrase as a whole. In (13) I reproduce the relevant S-structure version of (12), with indices. (13)

the linguists; [to whom^ thatj I spoke Xj in Nijmegen

The result is that deletion of that is the only possibility for relatives with piedpiping, in accordance with both the recoverability generalization and the DFCF effect. In (14) I present an example of the S-structure of a free relative, the second non-recoverability case. I assume, following Groos and van Riemsdijk (1979), that the antecedent of a free relative is a non-referential empty NP. (14)

0 whoever; that you met X; in York

Given the non-referentiality of the antecedent and its concomitant lack of index, there is no way for that to be co-indexed with it. Consequently, that remains indexless and is therefore a vacuous operator, that is again forced to delete. The answer to the question about the restriction of the function shift to recoverability contexts (question (2)) is now quite obvious: it is only in these contexts that the complementizer-operator can acquire the same index as the variabletrace of the relative pronoun and index-sharing between these two elements is a conditio sine qua non for the function shift The DFCF effect in function shifted and non-shifted relatives (cf. question (3)) is also explained: both fronted relative pronouns and relative complementizers have operator status, but one of them is always vacuous, either because it is not the local operator (the relative pronoun in recoverability contexts with indexed complementizers), or because it lacks an index (the non-indexed operator in cases where index percolation to C cannot or does not apply), or because its index differs from that of the variable-trace (indexed complementizers in cases with pied-piping). Vacuous operators lead to a violation of the VOC (or the Bijection Principle) and must delete, which they do by the rule of vacuous operator deletion, at S-structure.

3. LANGUAGE VARIATION

The present section deals with the question about language variation in question (4).

56

R. Bok-Bennema

As I have already mentioned, Romance languages manifest an obligatory function shift in recoverability contexts. In other contexts, however, the facts in Romance are the same as those in English. The differences and similarities between the two types of languages are illustrated in the English/Spanish paradigm in (15).5 (15)

a b.

the linguists that went to Budapest los lingüistas que fueron a Budapest

c. the linguists who went to Budapest d. *los lingüistas quienes fueron a Budapest e. *the liguists with that I spoke in Paris f. *los lingüistas con que hablo en Paris g. h.

the linguists with whom I spoke in Paris los lingüistas con quienes hablo en Paris

i. *that went to Venice learned a lot j. *que fue a Venecia aprendio mucho k. 1.

who(ever) went to Venice learned a lot quien fue a Venecia aprendio mucho

The difference between English and Romance can quite easily be explained by the assumption that index percolation to C is obligatory in Romance (if the antecedent provides an index, i.e. is referential) (16). (16)

In Romance, index percolation to C (7) is obligatory

One consequence of (16) is that a relative pronoun in recoverability contexts is a vacuous operator and must delete, hence the ungrammaticality of (15d), as opposed to (15b). In non-recoverability contexts the derivation in Romance is the same as in English, except that Romance pied-piping relatives lack the version in which the deletion of the complementizer follows from it having acquired no index. In section 1, I mentioned Dutch as an example of a language in which the complementizer never substitutes for a relative pronoun. I illustrate this with the English/Dutch contrast in (17). (17)

a. the linguists that he met in Copenhagen b. *de lingmsten dat hij ontmoette in Kopenhagen

On the COMP of relatives c. d.

57

the linguists who he met in Copenhagen de Ungutsten die hij ontmoette in Kopenhagen

In pied-piped relatives and free relatives the relevant facts are the same as those in English and Romance. Within the system I am presenting, the explanation for the Dutch variant is obvious: Dutch lacks index percolation to C altogether (18), so that the complementizer is always vacuous and, consequently, always deletes. (18)

Dutch has no index percolation to C

If we assume that predication indices which percolate to C are 'stronger' than those which do not, we can say that predication indices in Dutch are 'weak', whereas in Romance they are 'strong'. In English they are of course either weak or strong. I conclude that, by choosing one of the values [+strong], [-strong] or [+/-strong] for predication indices, the variation between the language types under consideration is automatically accounted for.6

4. CONCLUDING REMARKS

Obviously the present squib leaves many questions unanswered, both of a factual and of a theoretical nature. I may mention, for example, that I have no explanation for the facts in a language like Danish which manifests an optional function shift with relativized subjects and an (almost) obligatory one with relativized objects (cf. Smits (1989)). One advantage of the system presented here is that it relates the DFCF effect to the function shift between relative pronouns and complementizers, in a way which is both explanatory and in line with the present insights in phrasestructure. However, specifically w.r.L the DFCF effect, there are two important points I have ignored. The first concerns the fact that this effect is not restricted to relatives. The DFCF, as formulated in Chomsky and Lansik (1977), was intended to exclude all initial combinations of wA-words and lexical complementizers. This implies that my proposal would probably miss an important generalization if it could not be extended to wA-questions and other constructions involving w>A-movement. Though more research is clearly needed here, I think that such an extension is not implausible. With respect to w>A-questions with potentiality for a

58

R. Bok-Bennema

lexical complementizer - i.e. wA-questions that do not require verbs in C - it would imply that their complementizer has operator status. Given the lack of an antecedent, the complementizer-operator will always be vacuous, and delete. In fact, the treatment of this case would be exactly parallel to that of free relatives. Note, incidentally, that wh-questions with verbs in C pose no problem for this approach. Verb-containing C might also be considered an operator. As, however, it acquires the index of V and binds the trace of V in I, no vacuousness is involved and, consequently, no deletion applies. A second point concerning the DFCF effect is that there are languages or language variants in which this effect does not hold or where it holds in some wA-constructions, but not in all of them (e.g. substandard Dutch allows for "doubly filled COMPs" in indirect questions, but not in relatives; whereas there are variants of Flemish which allow them in both constructions). My analysis forces me to conclude that whenever the DFCF effect is absent C lacks operator-status at the level at which the VOC (or the Bijection Principle) applies.

NOTES * I would like to thank the students of the Hispanic Department of Groningen University who participated in the 1987 seminar on Spanish relative clauses, as well as the participants of the Romanistendag 1987 (Utrecht), for comments and suggestions. Of the above-mentioned students special thanks go to Riet Vos for extensive discussion. The internal publication GILDE-RUG (1988) presents a (shorter) Spanish version of the present squib. For application of the ideas presented here to Spanish non-restrictive relatives see Vos (1988). 1. I assume, contrary to Srnits (1989), but in line with Kayne (1976), that qui in French subject relativization is a special form of the complementizer. 2. Pesetsky (1982) and Chomsky (1981) also assume the presence of an index on the complementizer. In their approach, however, this index comes (directly) from the relative pronoun and not from the antecedent. 3. According to Chomsky (1982) predication indices are not assigned before LF'. The reason he gives is the lack of Weak Crossover (WCO) effects in relatives: if at LF the index of the relative pronoun is not yet equated to that of the antecedent, Bijection (Koopman and Sportiche, 1982) will not exclude WCO cases (Chomsky (1982) does not discuss indexed complementizers). I am forced to assume that the lack of WCO in relatives, i.e. their immunity to Bijection, follows for other reasons. One possibility is that the fact that the range of the operator involved is uniquely restricted by the referentiality of the antecedent plays a role here. Note that the standard deletion-up-torecoverability treatment of relatives also requires index-equation at S-structure, at least if "recoverability" is taken to imply referential identification. 4. Instances of relatives with neither that nor a relative pronoun can be accounted for by a general rale of //mi-deletion, see e.g. Aoun, Hornstein, Lightfoot and Weinberg (1987). 5. The Spanish a, inserted before (specific) human direct objects, patterns with other prepositions, as can be seen in (i).

On the COUP of relatives (i)

59

El linguists a *que/quien encontrf en Girona "The linguist a *that/whom I met in Girona"

6. English non-restrictive relatives, which are generally felt to be more loosely connected to their antecedent than restrictive ones, do not permit function shift and must be assumed to involve weak indices. This indicates that the degree of strongness of a predication index is sometimes related to the degree of 'tightness' of the relation between the antecedent and the relative pronoun.

REFERENCES Aoun, J., N. Homstein, D. Lightfoot and A. Weinberg (1987) "Two Types of Locality", Linguistic Inquiry, 18, 537-577. Chomsky, N. (1981) Lectures on Government and Binding, Foris, Dordrecht. Chomsky, N. (1982) Some Concepts and Consequences of the Theory of Government and Binding, MIT Press. Cambridge (Mass.). Chomsky, N. (1986) Barriers, MIT Press, Cambridge (Mass.). Chomsky, N. and H. Lasnik (1977) "Filters and Control", Linguistic Inquiry, 8.3. Cinque, G. (1982) "On the Theory of Relative Clauses and Markedness", The Linguistic Review, 1, 247-294. GILDE-RUG (1988) "Las Construcciones Relatives Espanolas", unpublished manuscript, Hispanic Department, Rijksuniversiteit Groningen. Groos, A. and H. van Riemsdijk (1981) "Matching Effects in Free Relatives: a Parameter of Core Grammar", in A. Belletti, C. Brandi and L. Rizzi (eds.), Theory of Markedness in Generative Grammar, Scuola Normale Superiore di Pisa, 171-216. Kayne R. (1976) "French Relative 'que'", in F. Hensey and M. Lujän (eds.), Current Studies in Romance Linguistics, Georgetown University Press, Washington, D.C., 255-299. Kayne, R. (1984) Connectedness and Binary Branching, Foris, Dordrecht. Koopman, H. and D. Sportiche (1982) "Variables and the Bijection Principle", The Linguistic Review, 2, 239-160. Pesetsky, D. (1982) "Complementizer-Trace Phenomena and the Nominative Island Condition", The Linguistic Review 1, 297-343. Schroten, J. (1984) "Two Approaches to the Distribution of Spanish Relative Pronouns", in Estudis Gramaticals I, Universität Autonome de Barcelona, Bellaterra. Smits, R. (1989) The Relative and Cleft Constructions of the Germanic and Romance Languages, Foris, Dordrecht Vos, R. (1988) "Algunas Construcciones Relatives Espanolas", unpublished manuscript, Hispanic Department, Rijksuniversiteit Groningen. Williams, E. (1980) "Predication", Linguistic Inquiry, 11.1.

Verbal chain and verbal cluster: a discussion between linguist A and linguist B I. Bordelois Arn. Evers University of Utrecht

A: Dutch, and more generally the West Germanic languages except English, exhibit a well known transition: a verbal chain in D-structure may wind itself up in an S-structure V-cluster. All types of verb with an infinitival complement take part (1)

als Johan het geld aan zijn vader over [had zien [had proberen [had lijken [had kunnen [was blijven

te te -

dragen] dragen] dragen] dragen] dragen]

if John the money to his father over

had see - turn try to appear to 'can' continue (if John had seen/tried to/appeared to/been able to/continued to turn the money over to his father) The auxiliary 'had' (had) should have been followed by a Past Participle, but for some reason this does not happen within a V-cluster. The phenomenon has been observed in grammars from the nineteenth century on, cf. Hoeksema (1988). Another symptom of the V-cluster is the separation of the most embedded verb 'dragen' (turn over) and its idiomatic panicle Over'. I am willing to defend the arguments in favor of such a V-cluster that have been formulated in Evers (1975) in spite of criticism, such as raised against them in Kroch and Santorini (1987). But I would rather discuss the question of what might trigger the V-to-V movement Comparable verbs in Romance, aspectuals, modals and a.c.i. verbs, may show a verb-verb adjacency requirement and an unusual transparency of the verbal complement. It is improbable that West Germanic and Romance correspond in language specific idioms that are triggered by a rule feature . A major S-structure principle must be at work. Which S-structure well-formedness condition leads to the V-cluster?

62

/. Bordelois and Arn. Evers

B: Maybe none. I do not assume V-clusters, at least not for Romance, but the derivation of such clusters by means of a matrix verb feature like or admits the fact that the direct evidence for a V-cluster comes from a quite restricted set of verbs within quite idiomatic constructions. A: One should make a distinction between the evidence for a representation and the possibility to derive that representation from general principles. For example, there is hardly any direct evidence for the existence of a VP in the Dutch/German S-structure. Structuralist analyses did not acknowledge such an entity. The generativist grammarians imposed the VP upon the Dutch/German structure, mainly because the VP is implied by the D-structure principles of X-bar projection and subcategorization. B: But I see no comparable V-cluster principle on the horizon for S-structures. Assuming for the sake of argument such V-clusters, why derive them from a transformation and an S-structure principle? Suppose they are deep structure objects as in Emonds (1978). A: That would not fit a strict X-bar scheme, in which each head has to originate within its own projection. Moreover, the V-cluster cannot be a D-structure unit, because verbs that are part of the V-cluster, like the verbs of perception and causation above, assign the same theta-structure within a nonrestructured environment. D-structure must generalize over the restructured and the non-restructured variants of the causative projection and it will do so by not having the V-cluster. That follows from a beautiful D-structure definition proposed in Baker (1987): Unitary Theta Assignment. The V-cluster cannot be put off to P-structure, as was once suggested in Tiersch (1978), since it has clear effects in L-structure (scope of Negation, scope of Nominalization, scope of Gapping). Nor is it possible to delay the V-cluster formation to L-structure, as is erroneously assumed by Baker (1987). The V-cluster has clear effects on the linearization in P-structure. B: Let me accept that. Still assuming for the sake of argument the actual existence of V-clusters, why not derive them after ah1 as the effect of a grammaticalized set of verbs ? A: V-cluster formation in West Germanic is a more general phenomenon than V-cluster formation in Romance. A rule feature is less interesting than a V-to-V Raising principle. The latter cannot express a difference between Romance and West Germanic unless there is an interaction with some other fundamental difference, e.g. the SOV nature of West Germanic.

Verbal chain and verbal cluster

63

B: That is true again, but to my mind Romance languages do not need a transition of verbal chains into verbal clusters at all. Rizzi (1982) proposed a derived V-cluster but it will probably suffice to attribute transparency properties to the verbal chain. In general, apparent evidence for a V-cluster in Romance may be reinterpreted as an effect of a verbal chain. At present there are good reasons to derive long clitic movements from a notion of verbal chains rather than from the assumption of a verbal cluster. Let a verbal chain be a set of verbal heads that successively govern each other's projections and let a verbal chain break off with any projection of a non V. The chain is governed by a common I head and the clitic moves up to the highest V+I head. The domain of the accessible subject is extended for clitic movement to the domain of the first dominating I projection, as proposed in Schroten (1986) or Bordelois (1988). Suppose there are parameters for V-chains: some may tolerate internal I elements, or else they would not require string adjacency of the heads in the chain. Such entities may affect the analysis of West Germanic. Dutch is claimed in Evers (1975) to be almost addicted to V-clusters, but I got a new picture of the relevant constructions from Den Besten et al. (1988). Den Besten and his group accept the existence of a V-cluster, but in a more restricted set of cases. A whole group of constructions is no longer analyzed by V-to-V Raising. Consider their key example. Extraposition

(2)

a.

I

[NP t-I"

i

V [

t-NP PP V] ]

ι

Pseudo-scrambling

b. *als hij het geld probeert [ om t-NP aan zijn vader te geven] als hij het geld probeert [ - t-NP aan zijn vader te geven] if he the money tries (for) to his father to give if he tries to give the money to his father Here Den Besten et al. (1988) extrapose a sentential object As they point out, the extraposed complement is obviously lexically governed by the matrix verb. It does not react as a barrier. Subsequently they prepose a downstairs argument (het geld/the money) into the matrix construction. Rather unfortunately, they call the reordering rule LDS (Long Distance Scrambling). Let me at least add the element 'pseudo' to the name and speak about Pseudo-scrambling.

64

/. Bordelois and Arn. Evers

The examples of the rule demonstrate a complement transparency that follows only from the absence of a head C, without requiring V-to-V Raising. Pseudoscrambling reduces the domain of V-to-V Raising. Other embedded arguments can be scrambled out and we end up with an extraposed complement that contains the head only. The net result comes down to an extraposition of the complement head only if one disregards the empty places: [ t-NP t-PP V]. Such an extraposed V reminds of the proposal in Kroch and Santorini (1987). A: The framework in Den Besten et al. is different. Moreover it seems to me that the analysis by Den Besten et al. (1988) is so strong just because both derivations, V-to-V Raising and Pseudo-scrambling, exist next to each other for the same strings. The insistence of the system to produce a V-cluster although an identical string can be derived by means of Pseudo-scrambling seems to me to indicate the working of an S-structure principle. (3)

a.

The complement extraposed and the downstairs arguments leftwardpseudo-scrambled. Extraposition [ NP PP M" V t PRO t-NP t-PP V] ]

^

t

—ι

I

Pseudo-scramblings [het geld aan zijn vader probeert] [ t-NP t-PP te geven] the money to his father tries to give b.

The complement not extraposed and the downstairs V raised out. [ NP PP t-V] [ V V]

I

T

V-to-V Raising [het geld aan zijn vader t-V] [probeert te geven] the money to his father tries to give c.

tries to give the money to his father

B: That is a rather suspicious pair of derivational alternatives. The outcome of V-to-V Raising can nearly always be reached by means of a complement Extraposition followed by repeated Pseudo-scrambling. Is it not preferable to get a uniform derivation by extending the possiblities of either pseudo-scrambling or raising?

Verbal chain and verbal cluster

65

A: Both rules, V-to-V Raising and Pseudo-scrambling, are well motivated. The strong point for Pseudo-scrambling is that such a rule, an adjunction of arguments to VP/IP, is needed anyway for the monoclausal structure, assuming a VP. (3)

d.

als hij [het geld [gisteren [ t-NP aan zijn vader gaf] ] ] if he the money yesterday to his father gave if he gave the money to his father yesterday

A plausible path condition may automatically enlarge the scope of the rule. Den Besten et al. (1988) propose that the rule may take the long step in a biclausal setting without C. Insertion of the infinitival complementizer Om' blocks the long variant of pseudo-scrambling, cf. the pair of examples in (2b). B: The consequences are more serious though than some simple and plausible path condition. The controlled complements without C imply that the ungovemed PRO is abandoned for Koster's (1978, 1986) long standing proposal to consider an obligatorily controlled PRO as an anaphor governed by the matrix verb. I also wonder how Pseudo-scrambling deals with the accessible subject condition. A: I get the impression that they see the Pseudo-scrambling rule as WH-movement The difference is one of landing sites. Pseudo-scrambling never lands into the specifier-C position. The empirical advantage of having two possible derivations, one by V-to-V Raising and one by Pseudo-scrambling, follows from two additional phenomena, Infinitivus-pro-participio and Particle-verb separation. Both phenomena have a lot of borderline cases and native speakers may easily have different judgements about specific examples. The weak border of the V-to-V Raising class is an interesting problem as such, but this happens to strengthen the present argument Native judgements tend to agree that whatever the grammatical quality of an Infinitivus-Pro-Participio, it corresponds for each matrix verb with the grammatical quality of Particle-Infinitive-Separation. This parallellism follows from the assumption that the rules for Infinitivuspro-participio and Particle-verb separation apply within a V-cluster only. In this way Infinitivus-pro-participio and Particle-verb-separation form a subtle argument for the structural difference between V-to-V Raising and Pseudo-scrambling. The partial fogginess of the two crucial phenomena follows from an uncertain membership for the class of V-to-V Raising verbs. Verbs that resist Infinitivus-pro-participio and Particle-verb-separation clearly head a pseudoscrambling construction only.

66 (4)

/. Bordelois and Arn. Evers a.

V-to-V Raising and Infinitivus-pro-participio.

NP PP particle t-V V heeft V Infinitive *pastparticiple

V infinitive

V-to-V Raising (als Jan) het geld aan zijn vader over heeft zien/*gezien dragen (if John) the money to his father over has see/*seen turn if John has seen (someone) turn the money over to his father The infinitival complement of 'zien' (see) is obligatorily restructured and this blocks the Past Participle formation. (4)

b.

Extraposition of I" and leftward-pseudo-scrambling of NP and PP

past-part.

PRO

t-NP

t-PP 1

Particle V inf

Pseudo-scramblings

(als Jan) geld aan zijn vader heeft *verzuimen/verzuimd over te dragen money to his father has *fail/failed over to turn if John has failed to turn money over to his father

Verbal chain and verbal cluster

67

The infinitival complement of 'verzuimen' (fail) may allow pseudo-scrambling but resists V-to-V Raising. B: Maybe there is a way to bring Pseudo-scrambling and V-to-V Raising in the same perspective. Haegeman and Van Riemsdijk (1986) derive the 'scrambling' of downstairs constituents into the matrix construction by raising a small VP. Does this not offer an opportunity to get a uniform analysis for (4a) and (4b)?

(5)

Small-VP-to-V Raising (als Jan) het geld t-V [probeert [aan zijn vader over te dragen] ] (if John) the money tries to his father over to turn if John tries to turn the money over to his father A: I do not know. Haegeman and Van Riemsdijk (1986) analyze different dialects and one should know how the more subtle phenomena around particles and past participles appear in that context. They were primarily interested in the question, whether the West Germanic V and V* raisings are compatible with Chomsky's Projection Principle. Somewhat maliciously, one might summarize their answer as 'West Germanic V-to-V-Raising is incompatible with any standard interpretation of that principle'. As a mere speculation, let us suppose that Leftward-(pseudo)-scrambling is relevant for the Belgian and Swiss city dialects and constitutes an alternative to the raising of small VPs. Then there is an interesting perspective. Pseudoscrambling only adjoins maximal projections and V-to-V Raising only adjoins X° categories. Hence, there is a possibility to get rid of small VP adjunction, in West Germanic at least B: The hallmark of V-to-V Raising in Dutch is now a participation in two completely language specific phenomena, Infinitive-pro-participio and Particleverb-separation. How does that square with the expectation that V-clusters may represent the standard S-structure form of verbal chains?

68

/. Bordelois and Arn. Evers

A: Excellently, I think. Verbal chains that huddle themselves up in S-structure V-clusters have their language specific symptoms, but they may have their universal causes as well. What we need is a general trigger for V-to-V Raising. B: Baker's (1987) striking analysis of Head-to-Head movement used morphology as a trigger. Certain verbal forms carry morphemes that mark them e.g. as causatives. At the same time the constructions headed by these forms exhibit certain biclausal properties. The theta-role assigning causative morpheme is taken as a D-structure matrix. It would trigger a restructuring at S-structure, to attain word status. This obviously does not work for the West European languages. There are the same biclausal/monoclausal paradoxes, arguably the same V-to-V movement, but the verbs involved do not lack word status at all. One may assign a descriptive feature to the V-to-V Raising verbs, but that is a rather artificial extension of an S-structure principle. A: Baker considers the theta assigning morpheme as the atom of D-structure, whereas the word is a relevant entity in S-structure. Di Sciullo and Williams (1987) consider words as the syntactic atoms of D-structure and S-structure alike, but their proposal runs into difficulties for derived causativization and nominalization (gerunds) and they have to weaken the Projection Principle for these constructions. So let me follow Baker's proposal, which implies that each systematic level has its own atoms. D-structure atoms are lexical items, not necessarily words, S-structure atoms are words, not necessarily phonetic units and finally there will be P-structure atoms, the phonetic words. Elaborating this somewhat, I claim that the 'Stray Affix Filter' has to apply at S-structure, rather than at P-structure, because a head in S-structure needs word status to function as a governor, e.g. in Case assignment. B: But you have no verbal affix to start with, so I see no immediate reason to embellish the Stray Affix Filter at this moment. A: My point was that the Stray Affix Filter would not apply at S-structure, unless there were some S-structure function to be performed. Let us say that we have two types of S-structure government, Abstract Case government and Abstract Mood government, Abstract Case government for the S-structure licensing of NP arguments and Abstract Mood government for the licensing of clause-like complements, the same idea as Fabb's (1988) 'verbal case'. Both types of S-structure licensing (Mood and Case government) would require a governor with word status. Lack of word status disqualifies a head as an S-structure governor, e.g. it cannot be a mood governor. The West European heads do not lack word status, but assuming a licensing principle for clausal complements, we may look for other factors that disqualify

69

Verbal chain and verbal cluster

a matrix verb as a mood governor in S-structure. Directionality seems to be relevant. The V-to-V Raising in Dutch and German applies to non-extraposed complements. This may be taken as evidence that sentential objects in West Germanic are mood governed if extraposed. If Extraposition is not applied, there is no mood government and V-to-V Raising is the only way out to license the structure, in standard Dutch at least. It is either the mood index that moves down or the govemee that moves up. The govemee, i.e. the dependent infinitive, will invariably end up in a cluster configuration that signals its mood value in S-structure. The cluster consists of the govemee sister-governed either by the mood index or by the matrix governor. (6)

a.

Mood-cluster and a Head-to-Head relation by mood government.

S-structure license of I" by mood marking b.

Mood cluster and a Head-to-Head relation by antecedent government.

S-structure license by V-to-V Raising

70

/. Bordelois and Arn. Evers

B: I see. Dutch or German are of the SOV type but in a weak sense. The NP object is case-governed to the left as in a strict SOV language. The sentential object is mood-governed to the right Extraposition linearizes the S-structure that way. If Extraposition fails to apply, the sentential complement is not mood governed and can be licensed only by a V-to-V Raising. Romance is SVO. Hence it can hardly be expected to disqualify S-structure governors for reasons of directionality. In fact you seem to predict now that V-to-V raising is impossible in Romance and a permanent option in West Germanic. As you can imagine, I welcome the first part of your prediction, while I have my doubts about the empirical value of the second part. A: Directionality alone will not do everything and some additional tinkering with mood government will be necessary, but I would like to preserve the directionality parameter. It builds a relation between V-to-V Raising and the SOV nature of West Germanic. Suppose, however, that in universal grammar there is an alternative for S-structure forms, either raising the dependent head or marking it by percolation government. In view of such an S-structure principle, it seems strange that Romance grammars require an S-structure adjacency of such verbs, but do not restructure towards a V-cluster. B: Maybe, if only so much of this were not speculation. The empirical value of your position remains unclear, even for Dutch. If things were as you propose, Dutch would display a free switch for all sentential objects between either Extraposition and mood indexing or restructuring by V-to-V Raising. In fact, I see very little of that. The free option is the exception rather than the rule. A small group of ten verbs listed by Den Besten et al.(1988, par 7.6) has the free option between Extraposition and Restructuring. A: There are problems but something of a solution shines through. The Den Besten group (1988) assumes selection and subcategorization to hold for matrix verbs in D-structure, in such a way that we get C-headed, I-headed and plain V-headed complements. (i) The C-headed complements are finite clauses introduced by the complementizer 'dat', or non-finite clauses introduced by the complementizer Om/0'. These complements will be extraposed and never be restructured for some unknown but probably quite general and trivial reason. (ii) The plain V-headed complements are projections of infinitives without 'te'. These complements are selected by modal verbs, aspectuals and a.c.i. verbs. They will be restructured and never be extraposed. The absence of 'te' in the complement may be related to the theta-structure, but it is still unclear why

Verbal chain and verbal cluster

71

standard Dutch and German resist the extraposition of such complements, (iii) The I-headed complements are projections of 'te' marked infinitives not introduced by the complementizer Om/0'. Depending upon further properties of the matrix verb, they are obligatorily restructured, obligatorily extraposed or they have the option between restructuring and extraposition. The last group, a large class according to Evers (1975), is now severely reduced by the discovery of Pseudo-scrambling. B: The complements in the third group which is 'te' marked and hence supposed to be I-headed, make me sceptic about your approach. These verbs are predicted to have a free option between V-to-V Raising and Pseudo-scrambling. Some matrix verbs have that option, but there are two other and larger groups that lack the option. One group blocks Extraposition and requires V-to-V Raising. These are the subject-tQ-subject raising verbs. The other group blocks V-to-V Raising but allows Extraposition and Pseudo-scrambling. This is the group of Den Besten's mere pseudo-scramblers. The question for an acceptable trigger proposal is what blocks Extraposition for the first group and what blocks V-to-V raising for the second group. A: An answer has been given for the first group. Reuland (1983) has proposed that the complements of verbs such as 'schijnen te' (seem to) are never extraposed since the subject of the embedded complement needs to be governed by the matrix verb. Let us maintain this idea. For clarity within the present set of assumptions, we may add that the non-controlled embedded subject needs to be governed like an NP argument, i.e. it needs the governing matrix verb on the right hand side. This requirement blocks extraposition, and consequently leaves V-to-V Raising as the only way out for an S-structure licensing of the complement. An example and a diagram: (7)

(als) Johan [ t-NP het geld aan zijn vader over t-V] [schijnt te dragen] John [ the money to his father over ] seems to turn (if) John seems to turn the money over to his father

72

(7)

/. Bordelois and Arn. Evers

b.

C'

t-NP het geld aan zijn vader over t-V Nominative Lowering or Subject Raising

V-to-V Raising

(i) The complement I" has not been extraposed, hence is not mood governed, and this causes a restructuring by V-to-V Raising. (ii) The V-cluster NP-govems into the complement, hence licenses a promotion of the embedded subject to the function of matrix subject, represented by a subject-to-subject movement, c.q. a nominative lowering. Had the complement been extraposed, mood government by the matrix verb would have followed. However, the promotion of the embedded subject would have been blocked, since the matrix verb cannot NP-govem to the right. The ungrammaticality follows from the failure to license the embedded subject. B: This leaves us with the other group of I-headed complements that lack a free option between V-to-V Raising and Extraposition. Den Besten's set of mere pseudo-scrambling verbs. How are they to be kept out of the V-to-V Raising set? Is an additional rule feature needed in order to encircle a minus V-to-V Raising subset within the class of control verbs? A: It is kind of insulting for the 'Sprachgeist' which has suggested so many deep regularities for me to come up now with a mere listing of peculiar cases. However, at the moment I must admit that I can not see how to fit Den Besten's set of mere pseudo-scramblers in a trigger theory. B: Don't you worry. Some beautiful morning in the next century, all grammatical puzzles will have been solved. We have looked at standard Dutch only. The

Verbal chain and verbal cluster

73

variation in the West Germanic languages is likely to reveal why a verbal chain should wind up in a V-cluster. Such things seem to happen for systematic reasons.

REFERENCES Baker, M. (1987) Incorporation, University Press, Chicago. Besten, den H., J. Kutten, T. Veenstra, J. Veld (1988) "Verb Raising, Extrapositie en de derde constructie", ms. University of Amsterdam. Bordelois, I. (1988) "Causatives: From Lexicon to Syntax", in Natural Language and Linguistic Theory. Di Sciullo, A. and E. Wuliams (1987) On the Definition of Word, ΜΓΤ Press, Cambridge (Mass.). Emonds, J. (1978) "The verbal complex V'-V in French", Linguistic Inquiry 9. Evers, A. (1975) The Transformational Cycle in Dutch and German, diss. Utrecht. Fabb, N. (1988) "Doing Affixation in GB Syntax", in Everaert, M. et al. (eds.). Morphology and Modularity, Foris, Dordrecht Haegeman, L. and Riemsdijk, H. van (1986) "Verb Projection Raising, Scope and the Typology of Rules Affecting Verbs", Linguistic Inquiry 17. Hoeksema, J. (1988) "A Constraint on Governors in the West Germanic V-cluster", in Everaert, M. et al. (eds.), Morphology and Modularity, Foris, Dordrecht Koster, J. (1978) Locality Principles in Syntax, Foris, Dordrecht Koster, J. (1986) Domains and Dynasties, Foris, Dordrecht Kroch, A. and B. Santorini (1987) "The derived constituent structure of the West Germanic Verb Raising Construction", in Freidin, R. (ed.). Proceedings of the Princeton Workshop on Comparative Grammar, MIT Press, Cambridge (Mass.). Reuland, E. (1983) "Government and the search for Aux, a case of crosslinguistic category identification", in Heny, F. and B. Richards (eds.), Linguistic Categories: Auxiliaries and Related Puzzles, Vol. I, Reidel, Dordrecht. Rizzi, L. (1982) "A Restructuring Rule", in Issues in Italian Syntax, Foris, Dordrecht Schroten, J. (1986) "Ergativity, Raising and Restructuring in the Syntax of Spanish Aspectual Verbs", in Unguisticae Investigationes. Tiersch, C. (1978) Topics in German Syntax, diss. ΜΓΓ.

Subject/object asymmetries in German nulltopic constructions and the status of specCP* Anna Cardinaletti University of Venice

In this paper, I will be concerned with the so-called null-topic construction exemplified in (1), where no lexical topic appears in specCP and the finite verb superficially occurs in first position:1 (1)

a. b.

Habe have Habe have

ich I es it

gestern gekauft. yesterday bought ("I have bought it yesterday") gestern gekauft. yesterday bought ("I have bought it yesterday")

In (la), a direct object is realized as zero, (Ib) represents an instance of null subject. The construction is typical of the spoken language and only possible with contextually salient elements, i.e. the reference of the null argument must be recoverable either from the linguistic or the extralinguistic context. The main goal of this paper is to show that the sentences in (1), although apparently instances of the same phenomenon, are indeed two different, syntactically unrelated constructions: (la) must be analysed along the same lines as apparent extraction out of islands, i.e. as involving a base-generated empty operator locally binding a null pronominal variable (vbl) in object position (cf. Cinque 1986); on the other hand, (Ib) represents an instance of "pro-drop" in a non-null-subject language, involving the null pronoun pro (cf. Chomsky 1982) moved to specCP. The discussion will also lead to a proposal concerning the status of specCP in a V/2 language.

1. COMMON PROPERTIES OF NULL TOPICS

The S-structure representations of (la) and (Ib) both contain two empty categories (ec), one in the sentence-internal Α-position and one in specCP: (2)

a. b.

eel habe ich gestern ec2 gekauft, eel habe ec2 gestern es gekauft.

The presence of ec2 is forced by the projection principle, which requires an argument to be structurally represented at every syntactic level. That specCP is

76

A. Cardinaletti

not just empty, but contains a syntactically present element is supported by strong empirical evidence. No null argument is in fact possible when specCP is occupied by another element, either lexical or empty. The presence in specCP of a topicalised XP, (3), of a wA-phrase, (4), and of an empty operator (Op) giving the modality of the sentence, for instance "interrogative" in (5), leads to ungrammatically: (3)

a *Gestem habe ich ec2 gesehen, b. *Gestem habe ec2 es gesehen.

(4)

a. *Wann hast du ec2 gesehen? when have you (her /him/ it) seen b. "'Ich weiß nicht, wann du ec2 gesehen hast I know not when you (her/ him/ it) seen have c. *Wann habe ec2 ihn eingeladen? when have (I) him invited d. "'Ich weiß nicht, wann ec2 ihn eingeladen habe.

(5)

a. *Op hast du ec2 gesehen? b. *Op habe ec2 ihn eingeladen?

Interestingly, Huang (1984) also observes that only one null argument per sentence is possible in German, which confirms that both the null object in (la) and the null subject in (Ib) require the presence of an ec in specCP: (6)

a. *Habe gestern gekauft. b. *ecl habe ec2 gestern ec2 gekauft.

As expected, both constructions are also impossible in embedded contexts, where no element in specCP can co-occur with the complementizer: (7)

a. *Du you b. *Ich I

weißt, eel daß ich ec2 gesehen habe. know that I (her/ him/ it) seen have bin sehr müde, eel weil ec2 gestern viel getanzt habe. am very tired because (I) yesterday much danced have

Although (la) and (Ib) pattern alike with respect to the presence of an ec in sentence-internal position bound by an ec in specCP, it is possible to show that the nature of the ecs and the syntactic relation between them are different in the two sentences. For ease of exposition, I will begin with the discussion of the object case.

Subject/object asymmetries

77

2. THE NULL-OBJECT CONSTRUCTION

In the tradition of Huang (1984),2 the ec in specCP and the one in A-position enter an Op/vbl configuration. This configuration can be said to arise through movement, given that the construction in (la) displays the same properties as movement to specCP. Typically, it obeys island constraints, which apply to movement operations: (8)

a. b. c. d.

*ecl glaube ich an die Möglichkeit, ec2 zu sehen. believe I to the possibility to see *ecl kenne ich den Jungen, der ec2 gebracht hat. know I the boy who brought has *ecl weiß ich nicht, wer ec2 gesehen hat. know I not who seen has *ecl ging er weg, nachdem er ec2 gesehen hatte. went he away after he seen had

However, a deeper look at the German facts suggests that an analysis in terms of an Op/vbl configuration arisen via movement to specCP is inadequate. Sentences like (la) are subject to further constraints which are unexpected under a movement analysis. First, whereas successive cyclic wh- and XP-movement to specCP are possible, (9), the supposed movement of the empty Op of (la) is ungrammatical from an embedded clause, (10). The same contrast is found in southern German, where extraction out of a do/3-clause is also possible,

(9)

a. b.

Wen sagt Hans, i hat/ habe er t gesehen? whom says H. has/ hasSUBJ he seen Den Professor sagt Hans, t hat/ habe er i gesehen. the professor says H. has/ hasSUBJ he seen

(10)

*ecl sagt Hans, hat/ habe er ec2 gesehen. said H. has/ hasSUBJ he seen

(11)

a.

(12)

*ecl glaube ich, daß er ec2 gesehen hat.

Wen glaubst du, i daß er t gesehen hat? whom think you that he seen has b. Den Hans glaube ich, t daß er t gesehen hat.

78

A. Cardinaletti

However, long extraction is not a priori excluded. The extraction out of infinitival clauses shows that it is ungrammatical if the embedded clause is postverbal, (13a,c), but grammatical when the clause occurs in preverbal position, i.e. in the so-called Mittelfeld, (13b,d):3 (13)

a. *ecl habe have b. led habe c. *ecl habe have d. eel habe

ich I ich ich I ich

noch nicht versucht [ec2 zu k ssen]. yet not tried to kiss [ec2 zu k ssen] noch nicht versucht. beschlossen [ec2 zu kaufen]. decided to buy [ec2 zu kaufen] beschlossen.

These facts are quite puzzling and unexpected within a movement analysis of (la). However, they are not surprising and indeed represent the expected situation if one adopts the analysis proposed by Cinque (1986) for the apparent extraction out of strong islands, for parasitic gaps and complement object deletion (COD) constructions. He suggests that these structures are not derived by wA-movement, and, crucially, the gap in Α-position is not a pure vbl, but a pronominal vbl, i.e. a null pronoun which comes to be A'-bound by a basegenerated Op at S-structure. Along the same lines, the D-structure and S-structure representations of (la) are (14a) and (14b), respectively: (14)

a. b.

[cp Op [[p ich gestern pro gekauft habe]] [CP Op habe [w ich gestern pro gekauft]]

Following Cinque (1986), I regard Op as a [-anaph,-pron] ec, a non-argument in a non-θ position. The feature content of the pronominal vbl is locally determined by the Op, which, in turn, is sanctioned by the preceding discourse or by pragmatics. I further assume that pro is subject to the connectedness condition of Kayne (1983) (as revised by Longobardi (1985), who imposes a requisite of proper government on all intermediate projections between the gap and the antecedent; cf. also Bennis and Hoekstra 1984/85). Adopting this analysis, we are now in a position to explain the ungrammaticality of (10), (12) and (13a,c). First notice that a postverbal clause in German is always on the wrong side with respect to the verbal head, i.e. it is not canonically governed by the verb. This forces the successive cyclic application of Move-α in SOV-languages (cf. Kayne (class lectures, Univ. of Venice, 1983) for German, Bennis and Hoekstra (1984/85:37f) for Dutch), which gives rise to two separate subtrees, each of which is a well-formed g-projection set However, if the construction in (la) is base generated (or, in other words, does not include a real trace), no successive cyclic derivation will be available, and no intermediate trace will be present in (10), (12) and (13a,c). This amounts to saying that the g-projection of the governor of the null pronominal would stop

Subject/object asymmetries

79

at the embedded CP (which is located on a non-canonically governed branch), hence a violation of the connectedness condition.4 (13b,d), on the other hand, are grammatical, given that the infinitival clause occurs on a canonically governed branch. Some further evidence supports the suggested analysis. There is a categorial restriction on the phrases that can be realized as null in sentences like (la). Only NPs are possible, although movement to specCP is not restricted to NPs (see (15)) and, also, no condition on recoverability should exclude the recovery of a contextually salient object PP (cf. Cinque (1986) for the same restriction in the constructions which exhibit an Op/ pro configuration cited above):5 (15)

a. b.

(16)

(17)

Auf wen hast du gewartet? for whom have you waited Auf Hans habe ich lange gewartet, for H. have I long waited Speaker A: Hast du auf Hans gewartet? have you for H. waited

a. b.

Speaker B: *Habe ich lange gewartet Speaker B: Auf ihn habe ich lange gewartet.

The ungrammaticality of (17a) follows straightforwardly if NP is the only category which has a null pronominal form. A further restriction is observed, which also is not expected under a movement analysis plus recovery from the context of the feature content of the null object in (la). The null object cannot be a pronoun of the 1st and 2nd person. This restriction is not found in the movement to specCP (cf. the grammaticality of (19b)). Notice also that 1st and 2nd person pronouns are very salient in the communicative situation since they refer to the participants in the speech act, hence very easily recoverable: (18)

(19)

Speaker A: Habe ich dich gestört? have I you disturbed a. b. c.

Speaker B: *Hast du sehr gestört. Speaker B: Mich hast du sehr gestört. Speaker B: Du hast mich sehr gestört

The ungrammaticality of (19a) receives a principled account under the suggested analysis. Notice that 1st and 2nd person pronouns are intrinsically either t+pronominal] or [+anaphoric], whereas the null Op in specCP is a [-pronominal], [-anaphoric] ec, hence only compatible with 3rd person NPs.

80

A. Cardinaletti

Two further properties of the construction in (la), which could suggest an analysis in terms of movement to specCP and consequent Op/vbl configuration, are also compatible with the account proposed here. First, the null object cannot be coreferential with the subject of the clause, (20). This reminds of the strongcrossover effect found in WA-interrogatives like the one in (21): (20) *Op liebt er sehr pro. loves he much (21) *Wen liebt er i? whom loves he However, sentence (20) can also be excluded if the gap is a pronoun (either according to Principle Β of the Binding Theory which requires that pronouns be Α-free in their governing category, or because of a ban against the A-binding of an A'-bound pro, cf. Cinque (1986: §2.5.)). Second, the construction allows parasitic gaps: (22)

a. b.

Op Habe have Op Habe have

ich I ich I

[ohne e zu erkennen] pro gesehen. without to recognize seen [ohne e zu kaufen] pro gelesen, without to buy read

If a parasitic gap is licensed by an Op in specCP which locally binds it at S-stmcture, the fact that the Op is base generated in this position should not play any role. As the grammaticality of (22) shows, this expectation is borne out I have thus far discussed evidence supporting the analysis of (la) in terms of a base-generated Op/pro configuration. However, the question of why movement of pro to specCP is excluded is still unanswered. I will try to answer it after discussing the null-subject construction of (Ib) in the next section.

3. THE NULL-SUBJECT CONSTRUCTION

Let us now turn to (Ib), where a subject is realized as an ec. The analysis suggested in §2. for the null-object construction cannot, however, be extended to (Ib). Language-internal evidence supports the view that (Ib) cannot be represented in terms of an Op/pro configuration. Contrary to the object case (see (19a)), pronouns of the 1st and 2nd person can be realized as null subjects:

Subject/object asymmetries (23)

a. b.

81

Habe sie nicht bestanden. (I) have it not passed Hast sie sehr gut bestanden, (you) have it very well passed

Furthermore, cross-linguistic studies suggest that a pronoun cannot be locally bound by an Op. Borer (1984) and McCloskey (1989) observe that in Hebrew and Irish, respectively, the resumptive pronoun strategy is ungrammatical in the case of a subject6 (24)

a.

Ha-?ish she (*-hu) ?ohev Parayot nePelam. the man that (he) loves lions disappeared b. *An fear a raibh s6 breoite. the man that was he sick

I would like to suggest that the sentence (Ib) represents an instance of null subject similar to the null subjects found for example in Italian, i.e. it involves the null pronoun pro, which, like any XP in German, is moved to specCP. The S-structure representation of (Ib) is therefore the following: (25)

pro habe t es gestern gekauft.

However, pro in (25) differs from the null subject found in Italian in that the recovery of its feature content does not depend on the agreement specification of the verb, but on the linguistic or pragmatic context. This kind of pro is in fact also possible in languages like Swedish (cf. Sigur sson 1989) which do not display any subject-verb agreement. With respect to the formal requirement of the licensing conditions of pro (cf. Rizzi 1986), I would like to propose that in a V/2 language like German, it is satisfied by the fact that pro occupies specCP, which is a position available for any (topic) XP. This analysis can explain the restrictions observed in the null-subject construction. Non-arguments and quasi-arguments, being non-referential and therefore not contextually recoverable, cannot enter the construction in (Ib): (26)

(27)

a. *pro wurde t viel getanzt was much danced b. *pro ist i ein Mann da. is a man there *pro regnet i gerade./ *pro hat i den ganzen Tag geregnet. rains now has the whole day rained

82

A. Cardinaletti

A last remark concerns a null subject linked to an extraposed clause. The construction is possible, with the only restriction that V/2-clauses cannot occur: (28)

a. pro ist t allen bekannt, da er kommt is to-everybody known that he comes b. *pro ist t allen bekannt, er kommt.

Given that V/2-clauses, contrary to dq/8-clauses, cannot occur linked to an argumental pronoun (cf. Cinque 1988 and Cardinaletti 1989) and given that the non-argumental pronoun with which they could co-occur cannot appear as null in specCP (see (26) above), the ungrammatically of (28b) follows.7

4. ON THE STATUS OF specCP

If a movement strategy is possible for a subject, the natural question arises as to why the same does not hold true of an object As I said above, no movement analysis is in fact possible for the object case. The question is then: Why is (29) ungrammatical? (29) *pro habe ich gestern t gekauft My tentative suggestion is the following: in a V/2 language, specCP can qualify for XP-movement both as an Α-position and an A'-position,8 independent principles of the grammar choosing the only grammatical possibility for each configuration. A pronoun moved to specCP creates an Α-chain with the baseposition,9 which is ungrammatical in the case of an object given the intervening subject position (cf. Rizzi 1987). The movement of pro to specCP in (29) is therefore as ungrammatical as the topicalisation of its (unstressed) lexical counterpart:10 (30) *Es/ *Ihn habe ich schon gekauft On the other hand, an Α-analysis of specCP is available for a subject pronoun if empty, (25), or unstressed, (3 la), as well as an A'-analysis, (31b): (31)

a. b.

Er hat i angerufen. he has phoned Er glaube ich, t hat t angerufen.

Notice that non-referential subjects, which are not semantically salient enough to be "topicalised" under A'-movement, can only be Α-moved. This offers a

Subject/object asymmetries

83

principled account of the fact that movement of non-arguments and quasiarguments to specCP is strictly local:11 (32)

a. b. c.

Es wurde t viel getanzt. Es ist / ein Mann da. Es hat t den ganzen Tag geregnet

(33)

a. *Es b. *Es it c. *Es

glaube ich, t wurde i viel getanzt. sagte ein Mann, t wären t schon viele Leute da. said a man wereSUBJ already many people there sagte ein Mann, t hätte t den ganzen Tag geregnet

NOTES * For many helpful discussions, I would like to thank Josef Bayer, Gerhard B rugger, Guglielmo Cinque, Thomas Herok and Giuseppe Longobardi. 1. "Null topic", construed on the more usual terms "null subject" and "null object", is intended to replace terms like "pronoun zap" (Ross 1982) and "topic-drop" (Sigurdsson 1989). For (la), cf. Huang (1984:546ff), Stemefeld (1985:406f). For (Ib), cf. also Bayer (1983/84:fn.4). 2. Cf. Huang (1984) for Chinese, Raposo (1986) for Portuguese, Campos (1986) for Spanish, Cole (1987) for Korean and Thai, Authier (1988) for KiNande, Sigurdsson (1989) for Scandinavian languages. For a different analysis for Quiteno, see Suner and Yepez (1988). 3. Wh- and XP-movement are of course possible in both structures. 4. A further constraint forces the presence of the intermediate trace in (9), given that successive cyclic movement out of V/2 complements is also required in a SVO language like Yiddish. 5. Cf. Raposo (1986) for the same observation for Portuguese. 6. The ungrammaticality of gaps in subject position in apparent extractions out of strong islands, in parasitic gap and COD constructions are reduced in Cinque (1986) to the same phenomenon: (i)

a.

*Someone who John expected [t would be successful though believing [e is incompetent]]... b. *TThe student that Susan left because John said [e was intelligent]... c. *!Mary is hard for me to believe [e kissed John].

7. For reasons not clear to me, 3rd person pronouns referring to humans are more marginally realized as null than pronouns linked to an extraposed clause: (i)

(?)?/»O hat t gerade angerufen, has now phoned

8. Cf. also Taraldsen (1986). 9. For a similar suggestion, see Holmberg (1986). 10. Stressed pronouns are R-expressions, hence their topicalisation (via A'-movement) is freely allowed like that of any XP. 11. For the analysis of the sentence-initial es in (32a,b) as a non-argument moved from specIP, cf. Cardinaletti (1989).

84

A. Cardinaletti

REFERENCES Authier, J.-M.P. (1988) "Null Object Constructions in KiNande", Natural Language and Linguistic Theory, 6, 19-37. Bayer, J. (1983/84) "Comp in Bavarian Syntax", Linguistic Review, 3, 209-274. Bennis, H. and T. Hoekstra (1984/85) "Gaps and Parasitic Gaps", Linguistic Review, 4, 29-87. Borer, H. (1984) (1984) "Restrictive Relative Clauses in Modem Hebrew", Natural Language and Linguistic Theory, 2, 219-260. Campos, H. (1986) "Indefinite Object Drop", Linguistic Inquiry, 17, 354-359. Cardinaletti, A. (1989) "Es, pro and Sentential Arguments in German", to appear in Linguistische Berichte. Chomsky, N. (1982) Some Concepts and Consequences of the Theory of Government and Binding, MIT Press, Cambridge (Mass.). Cinque, G. (1986) "A'-Bound pro vs. Variable", in Types of A'-Dependencies, ΜΓΤ Press, to appear. Cinque, G. (1989) "On Embedded Verb-Second Clauses and Ergativity in German", in Jaspers and Kloosters (eds.), Sentential Complementation and the Lexicon, Studies in Honour of Won de Geest, Foris, Dordrecht Cole, P. (1987) "Null Objects in Universal Grammar", Linguistic Inquiry, 18, 597-612. Holmberg, C. (1986) Word Order and Syntactic Features in the Scandinavian Languages and English, Department of General Linguistics, University of Stockholm. Huang, C.-T.J. (1984) "On the Distribution and Reference of Empty Pronouns", Linguistic Inquiry, 15, 531-574. Kayne, R. (1983) "Connectedness", Linguistic Inquiry, 14, 223-249. Longobardi, G. (1985) "The Theoretical Status of the Adjunct Condition", ms., Scuola Normale Superiore, Pisa. McCloskey, J. (1989) "Resumptive Pronouns, A'-Binding and Levels of Representation in Irish", to appear in R. Hendrick (ed.), The Syntax of the Modern Celtic Languages, Academic Press. Raposo, E. (1986) "On the Null Object in European Portuguese", in O. Jaeggli and C. SilvaCorvilan (eds.), Studies in Romance Linguistics, Foris, Dordrecht, 373-390. Rizzi, L. (1986) "Null Objects in Italian and the Theory of orb", Linguistic Inquiry, 17, 501-557. Rizzi, L. (1987) "Relativized Minimality", ms, Universiti de Geneve. Ross, J.R. (1982) "Pronoun Deleting Processes in German", paper presented at the annual meeting of the LSA, San Diego, CA. Sigurdsson, A. (1989) Verbal Syntax and Case in Icelandic. In a Comparative GB Approach, University of Lund, Lund. Stemefeld, W. (1985) "Deutsch ohne grammatische Funktionen: Ein Beitrag zur Rektions- und Bindungstheorie", Linguistische Berichte, 99, 394-439. Sufler, M. and M. Yepez (1988) "Null Definite Objects in Quitefio", Linguistic Inquiry, 19, 511-519. Taraldsen, K.T. (1986) "On Verb Second and the Functional Content of Syntactic Categories", in H. Haider and M. Prinzhom (eds.), Verb Second Phenomena in Germanic Languages, Foris, Dordrecht, 7-25.

On a difference between English and Italian 'Complement Object Deletion' constructions Guglielmo Cinque University of Venice

If we compare English and Italian 'Complement Object Deletion' (COD) constructions1, we find a lack of perfect correspondence. Whereas English 'tough' constructions such as (la) and purpose clause constructions such as (Ib) find a (near) exact counterpart in Italian (cf. (2a) and (2b), the English COD constructions in (lc)-(le) have no COD analogue in Italian. Rather, the object ec of the English sentence must be filled in Italian by an overt pronominal (see (2c)-(2e): (1)

a. b. c. d. e.

The probten^ is not easy [to solve e; immediately] They gave it; to me [to review e; for tomorrow] This boolq is too biased [for us to adopt ej2 This job; isn't remunerative enough [for us to accept &{ straightaway] I bought itj [for you to wear Cj tonight at the party]

(2)

a. b.

II problemaj non e' facile [da risolvere β; subito] Me lOj hanno dato [da recensire e; per domani]

c.

Questo libro e' troppo di parte [ per \

d.

Questo lavoro non e abbastanza remunerative [ per J , , .. , I accettar/o subito ] l > . .. J*indossare e l „ , r L ho comprato [per \ . , . ( stasera alia festa ] F F I mdossar/o J

e.

I ^iifiof fiirp ί* I

}

noi]

Why should the two languages display this particular asymmetry? The obvious thing to note, concerning (1) and (2), is that where the correspondence holds between the two languages, the infinitival complement in Italian is introduced by the preposition da (lit. "from"), while where the correspondence fails, it is introduced by a different preposition (per "for"). It seems highly unlikely that the difference is due to some idiosyncratic lexical difference between the two prepositions. More reasonable seems the conjecture that the two types of complement differ as to their structure (beyond the difference in the choice of preposition). There are in fact indications that the complement introduced by da in (2a)

86

G. Cinque

and (2b) is a 'bare' CP complement (with da a prepositional complementizer filling the head position of CP) while the complement introduced by per in (2c)-(2e) is a PP containing a CP (with per head of the PP). Certain prepositions in Italian (di "of, a "to") are unquestionably infinitival complementizers in such contexts as (3a) and (3b): (3)

a. b.

Cercavo [di riposare] I was trying to rest Ho provato [ad aggiustarlo] I attempted to mend it

For example, they cannot be followed by an NP or finite CP complement:3 (4)

a. *Cercavo di questo I was trying of this b. *Ho provato a questo I attempted to this

(5)

a. *Cercavo di ehe rimanesse 1 was trying that he stay b. *Ho provato a ehe l'aggiustasse I attempted that he fix it

Other prepositions, instead, must be analyzed as genuine heads of PPs taking an infinitival CP as complement (in free alternation with NP and finite CP complements). The prepositions senza "without" and dopo "after" in (6) belong to this second class. (6)

a. b.

Se ne e andato [ senza [ salutarci ]] He went away without saying goodbye to us Se ne e andato [ dopo [ averci salutato ]] He went away after saying goodbye to us

To determine whether da and per in (2) above are prepositional complementizers or genuine prepositions heading a PP, we must look for properties which distinguish the clear cases of prepositional complementizers such as di and α from the clear cases of genuine prepositions like senza en dopo, and then check how the da and per of (2) behave with respect to such properties. There are at least two properties which distinguish the prepositions senza and dopo from the prepositions di and a of (3). The first is the fact that senza and dopo allow for an auxiliary to move to the left of a lexical subject of the infinitival complement, while di and a do not. See:

Complement object deletion (7)

a.

b.

(8)

87

La societa fu sciolta senza aver i soci avuto la possibilita di salvarla The partnership was dissolved without having the partners had the possibility of saving it La nave fu abbandonata dopo esser stato anche ultimo passeggero tratto in salvo su en battello pneumatico The ship was abandoned after having been even the last passenger rescued on a life-boat

a. *Ho cercato di aver anche il piü piccolo di voi la possibilita di parteciparvi I tried to nave even the smallest among you the possibility to take part b. *Ho provato ad esser anche mio figlio ammesso a quel corso I tried to be even my son accepted in that course

The construction in (7) and (8) is an instance of the Aux-to-COMP rule studied in detail in Rizzi (1982, chapL 3). The contrast in grammaticality between the two pairs of sentences is, in this light, not surprising. It supports the proposed analysis of (3) and (6) to the extent to which the contrast follows directly from that analysis. An auxiliary can move to C in (7) because senza and dopo do not occupy the head of the infinitival CP, but it cannot move to C in (8) because that position is already filled by the prepositional complementizers di and a. The contrast is thus directly linkable to the "Doubly Filled COMP Filter". The second property distinguishing the two classes of prepositions is the possibility of interpolating (certain) adverbials between the preposition and the following infinitive. Senza and dopo appear to allow for it, while di and a do not (at least in the ordinary style):4

(9)

a.

b.

_,,.... Gh parlero senza

J forse \ . \ . t nuscire a convmcerlo (per altro J I will talk to him without perhaps/in any case managing to convince him

Lo hanno convinto dopo { . ( averlo piü volte minacciato [ per altro J They convinced him after perhaps/in any case having menaced him many times

88

G. Cinque

v(10)

'

I forse. Ϊ vederlo I per altro J I will try to perhaps/in any case see him

a. *Cercher di

b. *Provero a

< ... > aggiustarlo \ per altro J I will attempt to perhaps/in any case mend it

We may now ask how the prepositions da respect to such properties. What we find is that per, but not da, interpolation of adverbials. This suggests followed by an infinitival CP, while da is a tizer, like the di and a of (3). See:5

and per of (2) above behave with allows for Aux-to-COMP and the that per is a genuine preposition prepositional infinitival· complemen-

(11)

a.

Gianni e troppo astuto per poter suo figlio sperare di ingannarlo G. is too smart for being able his son to deceive him b. Gianni e troppo astuto per forse non sapere ehe lo volevano ingannare G. is too smart for perhaps not knowing that they wanted to deceive him

(12)

a. *Glielo hanno dato da poter anche suo figlio utilizzare They gave it to him for being able his son too to use b. *Glielo hanno dato da forse recensire per domani They gave it to him to perhaps review for tomorrow

We may thus reasonably conclude that the relevant structures of (2a)-(2b) and (2c)-(2e) above are as indicated in (13) and (14), respectively (taking (2a) and (2c) as representatives of the two cases): (13)

II problema non fc facile [Q, da risolvere e subito ]]

(14)

Questo libro έ troppo di pane [PP [p. per [Q, C [^ adottarlo ]]]]

If so, we have some evidence for the following correlation: the complement object of a COD construction in Italian can be null only if the complement is introduced by an infinitival complementizer, not if it is introduced by a real preposition. Interesting support for the correctness of this correlation (and against the alternative view that the possibility of an empty object be tied to the lexical choice of the preposition da vs. per) comes from the following minimal contrast provided by the preposition da.

Complement object deletion

89

Along its use in such contexts as (2a) and (2b) above, and in relative clauses, da is also used to introduce infinitival result clauses. Cf., e.g., (15) (15)

Mario έ talmente distratto [ da dimenticarsi perfino di mangiare ] M. is so absent-minded that he even forgets to eat (lit.: to even forget to eat)

Now, with respect to the properties of Aux-to-COMP and interpolation of adverbials, this use of da gives results opposite to those of the da in (2). Compare (12a) and (2b) with (16)

a.

Gianni talmente furbo da non poter neanche lei sperarc di imbrogliarlo G. is so smart that not even she can hope to deceive him

b.

Carlo έ talmente ignorante da \ . \ non sapere ehe.. a v [sicuramentej C. is so ignorant to perhaps/surely not know that.

As it appears to allow for both Aux-to-COMP and interpolation of adverbials, we are led to conclude that da in this use is a genuine preposition, head of a PP, just like per in (2c)-(2e). Interestingly, this da behaves like per, and differently from the da of (2a) and (2b), also in disallowing Object deletion* in its complement, a property supporting the correlation noted above. See: (17) *Gianni έ talmente furbo da non poter neanche lei sperare di imbrogliare e G. is so smart that not even she can hope to deceive (lit.: to not be able not even she to hope to deceive) We may note, finally, that nothing principled prevents the complementizer da from introducing the complement of such adjectives qualifiers as abbastanza ("enough") or troppo ("too") in (2c)-(2e) above. It seems to be simply a matter of lexical selection that they take a PP headed by per. If they selected a 'bare' CP headed by the complementizer da (in some dialect), the prediction would be that a null object would then be possible, and Aux-to-COMP and interpolation of adverbials impossible. Having tentatively established the correlation that the complement object of a COD construction in Italian can be null only if the complement is introduced by an infinitival complementizer, we may raise the deeper question of what it is that causes the correlation to hold. Or, in other words, what is it that prevents the wellformedness of the configuration (18) below?

90

G. Cinque

(18) *Questo libro fe troppo di parte [PPper [Q,O [jpadottare e ]]] We suggest that the answer to this question lies in the predication requirement that these structures must satisfy (cf. Chomsky (1986, 109)). Their status as Open sentences', created by the null operator/variable structure, must be satisfied through predication in order for the null operator (ultimately, the variable) to acquire a value (See Chomsky's (1986) notion of 'strongly bound'). If the predication relation is a reindexing process which identifies the index of the adjunct XP with that of the predicated NP (cf. again Chomsky (1986)), then the predication relation will be correctly established only if the adjunct XP to be reindexed is the CP immediately dominating the operator. For, only this will inherit the index of the null operator, via SPEC-HEAD agreement and index percolation from head to maximal projection and will be in a relation of mutual m-command with the subject of the predication, as required:6 (19)

Me Vna dato NPj [Q, Ok [ cdaj fo PRO recensire ^ ]] i =k

If the XP to be predicated is instead a PP containing the indexed CP, as in (20), no correct predication relation will be established, as the two terms to be reindexed will not stand in a relation of mutual m-command: (20)

Lj'ho comprato NP; [„, [pper ] [CPkOk Ck [IP PRO indossare % ]]]

NPi and CPk m-command each other in (19), and in (Ib), (le), assuming CP to be dominated by VP, as is the case with other adjuncts, but not in (20), nor in (2c) and (2d).7

NOTES 1. The tern», drawn from Lasnik and Fiengo (1974), is used here in a purely descriptive fashion. We follow Chomsky's (1977) original analysis of these null object constructions in terras of covert wh-movement 2. The for which is found to optionally introduce COD constructions in English is uniformly analyzed as a complementizer in Chomsky (1977. 1981, 1982). For the possibility that the for which introduces easy-to-please constructions is a matrix "benefactive" PP, see Kayne (1984, Chapter 3). 3. CF. Rizzi (1982, Chapter 3), Cinque (1989, chapter 1). 4. See Rizzi (1982, Chapter 3), Cinque (1983) and references cited there for a more careful discussion and analysis of this property. 5. This converges with Rizzi's (1982, Chapter 3) conclusion that the da which introduces infinitival relatives in Italian is an infinitival complementizer. 6. For evidence that the command requirement should be mutual m-, rather than c-command, see Rizzi (1989, sect 2.4).

Complement object deletion

91

7. I further assume that in (la), (Ic), (Id) and (2a) the CP is in relation of mutual m-command with a subject position inside AP.

REFERENCES Chomsky, N. (1977) "On Wh-movement", in P. Culioover, T. Wasow and A. Akmajian (eds.), Formal Syntax, Academic Press, New York, 71-132. Chomsky, N. (1981) Lectures on Government and Binding, Foris, Dordrecht. Chomsky, N. (1982) Some Concepts and Consequences of the Theory of Government and Binding, MIT Press, Cambridge (Mass.). Chomsky, N. (1986) Knowledge of Language, Praeger, New York. Cinque, G. (1983) "Su una differenza tra 1'italiano e 1'inglese nelle construzioni 'ad ellissi dell'oggetto'", Rivista di Grammatica Generativa, 8, 127-151. Cinque, G. (1989) Types of A'-Dependencies, Universita di Venezia (to be published by MIT Press). Kayne, R. (1984) Connectedness and Binary Branching, Foris, Dordrecht. Lasnik, H. and R. Fiengo (1974) "Complement Object Deletion", Linguistic Inquiry 5, 535-571. Rizzi, L. (1982) Issues in Italian Syntax, Foris, Dordrecht. Rizzi, L. (1989) Relativized Minimality, Universiti de Geneve (to be published by MIT Press).

A Note on Bars and Barriers* Peter Coopmans University of Utrecht

One of the more explicit research questions on the Barriers agenda is how the island character of adjuncts can be squared with the escape hatch effects of adjunction. After all, if adjunction is possible to maximal projections in nonargument positions, then, as Chomsky (1986; 65) notes, the Adjunct Condition is voided entirely under the barriers system. On the core assumptions of this theory a representation like (1) is possible without w/i-movement crossing any barriers. (1)

Which concert did [JP you sleep [PP t [pp during t]]]

Adjoining which concert to the adjunct PP circumvents barrierhood of the latter category. The derived PP doesn't induce any barriers, and the IP cannot become a barrier by itself. A theory of adjunction will need to specify in which configurations it is possible to adjoin. If in (1) PP cannot be adjoined to, the structure will violate subjacency; PP itself will become a barrier and will also turn the immediately dominating node IP into one. No adjunction allows one to treat adjunct and subject islands in a uniform way, more in line with Huang's (1982) Condition on Extraction Domains (CED). In Coopmans (1988) I proposed a revision of the barrier system that allowed such a uniform treatment of subject and adjunct islands by incorporating the CED effects even more directly into the system. The general idea was that it appears desirable to prohibit intermediate adjunction for wA-movement altogether. In the theory outlined in Chomsky (1986), VP is the only maximal projection that it seems to be possible to adjoin to without further problems. The example in (1) shows that adjunction to phrases in adjunct positions is problematic; adjunction to arguments is prohibited as it will result in a violation of the θ-criterion; but the prohibition against adjunction to IP requires a special stipulation (cf. Chomsky 1986). Hence only VP seems to freely allow adjunction without any harm. It is not difficult, though, to imagine a way of accounting for the transparency of VP without recourse to unrestricted adjunction. VP can be made inherently transparent if the element that VP is selected by, INFL, is strong enough to make it transparent One way to achieve this is by assuming that θ-government is enough to make a maximal projection transparent.1

94

P. Coopmans

Clearly, the view that adjunction to VP is prohibited is too strong in view of the many proposals concerning such processes as extraposition (PP or clausal extraposition, Heavy NP Shift etc.) for which precisely adjunction of non-wAphrases to VP seems appropriate. But the present proposal should be interpreted as claiming that a wA-element should go en route to COMP directly, not barred by VP as an intrinsic barrier. With these changes the following picture emerges. Because adjunction as an escape route for wA-movement is prohibited, extraction from adjunct and subject domains is prohibited because these are not θ-governed. Extraction from complements (and perhaps certain selected adjuncts) is allowed because they do not form intrinsic barriers. Extraction from VP is now allowed for the same reason because it is θ-governed. For IP we still need to resort to the special proviso. While the barriers proposal is too lenient with respect to extraction from adjuncts, the present proposal is too strict. It prohibits extraction from any nonselected domain. This problem can be remedied if a distinction is made between certain kinds of non-selected domains. By the standard reasoning adjuncts are not selected, i.e. not properly governed, and thus they form islands for extraction. However, it turns out that certain complements may end up in adjunct-like positions, where they will count structurally as non-selected domains. Hence, from a representational point of view, they should have an island character as strong as normal adjuncts. It turns out that is incorrect In Coopmans (1988) I discussed a number of cases which provide evidence for the view that a distinction should be made between derived adjuncts and base-generated adjuncts, where the former but not the latter are transparent for syntactic movement. In order to derive this distinction between transparent and opaque adjuncts, I proposed to replace the bar notation in the categorial module with Muysken's (1982) features [±maximal; ±projection], and by doing this to make the barrier algorithm sensitive to the derivational history of the "adjunct". Such a change allows for a structural answer to the question why certain adjuncts are transparent and paves the way for an approach in which adjunction as an intermediate step for H>A-movement should simply not be an option. VP is inherently transparent because θ-government by INFL is claimed to be sufficient This proposal, in conjunction with the prohibition against intermediate adjunction to VP, makes exactly the same predictions as the original barriers account of complement extraction. Since complements satisfy the ECP locally via θ-government, moving them can at most result in a subjacency violation. But for the purpose of subjacency, the change proposed here has no effect The change does make a difference empirically for instances where the ECP has to be met via antecedent government. The crucial example in this respect is (2), taken from Chomsky (1986; 21). (2)

? Who did they wonder whether to consider t (to be) intelligent

A Note on Bars and Barriers

95

Chomsky observes that (2) is at most a weak violation of the w/i-island condition, not a strong ECP violation. His theory assigns it the structure in (3). (3)

Who ... [CP whether [w to [yp t' [yp consider [IP t .. intelligent]]]]]

On Chomsky's assumptions, the t in subject position is antecedent governed by t', and the ECP is satisfied. Further movement to the matrix clause crosses IP and CP, which will give rise to one barrier (by inheritance), and this will only weakly affect subjacency.2 If, as in the present proposal, intermediate VP adjunction is not an option, (2) is predicted to be as ill-formed as a standard ECP violation. The element that is needed to antecedent govern the trace in the complement subject position is the wA-phrase itself; but in its matrix COMP position it is too far away to do so. The intermediate CP (a barrier by inheritance from IP) will block this relation.3 The fact that (2) is only marginally ill-formed suggests that VPadjunction is needed to allow this example to at least satisfy the ECP. I will argue, however, that this conclusion is unwarranted, by means of certain adjunct extraction facts discussed by Stevenson & Coopmans (1989). The proposal advanced there suggests a way out of the present problem. Consider the following extraction possibilities of how, showing a surprising difference between infinitival and finite complements. (4)

a. b.

*I wonder how to believe [John fixed the car t] *I wonder how to believe [John to have fixed the car t]

(5)

a. b.

I wonder how Mary believed [John fixed the car t] I wonder how Mary believed [John to have fixed the car t]

What is surprising is that extraction of how across an infinitival is worse than extraction across a finite clause. It is important to note that the tense value of the clause in which how originates has no effect on the acceptability of these sentences; the sentences in (4) are equally bad, those in (5) equally acceptable. This contrast remains unexplained in the barriers theory. It is clear that the illformed examples of (4) must be ruled out as violations of the ECP rather than subjacency. One simple reason is that the corresponding extraction of a complement is, of course, perfect, as shown by / wonder what to believe John fixed. However, the structure that the barriers theory assigns to such examples as (4a) gives no indication that something is wrong with the links between how and the original adjunct trace: (6)

*.. how [to [yp t'[yp believe [Q, t[John [yp t[yp fixed the car t]]]]]]]

96

P. Coopmans

Stevenson & Coopmans (1989) argue that what underlies the difference between (4) and (5) is minimality (7), construed in such a way that in (4) the infinitival verb believe gives rise to an extra barrier, while its finite counterpart in (5) does not Minimality should be held responsible for blocking antecedent government of the t in COMP by f in (6). (7)

In the configuration .. α .. [τ . .8. .β..] .. α does not govern if τ is a (immediate) projection of δ excluding α Chomsky (1986;42)

This claim itself clearly does not distinguish the examples in (5) from those in (4). Stevenson & Coopmans (1989) propose that V-to-INFL movement eliminates the ability of the verb to serve as a minimal governor, they argue that the verbal trace is not lexical enough to induce minimality. As is well-known, lexical verbs in English do not move to INFL in syntax, rather this movement remains an option at LF. Stevenson & Coopmans assume, following Chomsky (1988), that V-to-INFL does apply as an LF process in finite clauses (to support a visible Tense node), but that it is not required in infinitivals. This process has the effect that in finite clauses there is no minimality barrier hampering antecedent government, but that in infinitivals the verb-in-situ blocks antecedent government of the trace in COMP by ( in (6). This gives the desired effect for chains resulting from adjunct movement, since these are checked for the ECP at LF. Minimality is blocked when a lexical head is present to induce it, movement of this head voids minimality. This scenario also makes redundant Chomsky's (1986) suggestion that intermediate bar levels only exist if required. On his assumptions such a special statement is necessary to allow Aow-extraction from clausal complements. This does not suffice to capture the difference between (4) and (5). The alternative proposal is that minimality is always induced by a lexical head; if this head has been moved out, its trace is not strong enough to induce minimality. The account in terms of minimality will only have an effect on adjunct chains, not on argument chains. The reason should be clear: the relevant distinction between finite and infinitival verbs is not made until LF. Since syntactic movement of an argument leaves behind a trace that has to be properly governed at S-S (cf. Chomsky 1986; Lasnik & Saito 1984), V-to-INFL raising at LF should have no effect. The following cases of subject extraction, where the subject trace has to obey the ECP through antecedent government, show that the tense nature of the intermediate verb has no effect on the syntactic movement of arguments.4

A Note on Bars and Barriers (8)

a. b. c. d.

They They They They

wondered wondered wondered wondered

97

who Mary believed [t* [ t fixed the car] who to believe [t' [ t fixed the car] who Mary believed [ t to have fixed the car] who to believe [ t to have fixed the car]

Notice that the subject trace in (8a) and (8b) is antecedent governed by the t' in COMP, which can subsequently be deleted at LF (if necessary). The proper government requirement is met at S-S. In (8c) and (8d), however, the subject trace has to be governed "across" the verb believe(d), which, as we have seen, imposes a minimality barrier. Since these sentences are grammatical, the conclusion must be that a closer antecedent governor is involved, analogous to the f' in COMP in (8a) and (8b). This conclusion is perfectly in line with Kayne's (1981) original proposal that such exceptional case marking constructions are CPs, with a case-marking null complementizer. In his analysis there will be an intermediate COMP position hosting a trace which will antecedent govern the subject trace at the relevant level. Assuming that the subject position of a small clause complement is not Θgoverned, we have to draw the same conclusion for small clauses, as is illustrated by the sentences in (9). Independent evidence to support the claim that small clauses have to be analyzed as CP is provided by Pesetsky (1984) and Kitagawa's (1986) discussion of certain subject-object asymmetries in parasitic gap constructions. (9)

a. b.

They wondered who Mary considered [t'[ t foolish]] They wondered who to consider [t'[ t foolish]]

If Stevenson & Coopmans' account is the correct way of explaining why adjuncts but not arguments are sensitive to the tensed/non-tensed nature of the verb they move across, it provides evidence for the view that ECM and small clause complements are CPs. With this conclusion we can now turn to the original problem of this note. If the sentential complement following consider in (2) is a CP rather than an IP, there is room for an intermediate trace in the COMP of this CP. This trace will operate as the antecedent governor for the trace in subject position, so that the ECP will be satisfied. The trace in COMP itself may be deleted at LF, since it is no longer necessary nor required by any principle. This means that no VP adjunction (such as in 3) is required to satisfy the ECP; in other words, from a proper government perspective, there is no theoretical evidence for adjunction to VP. And, as observed before, the fact that (2) is at most a weak subjacency violation can be accounted for under an "adjunction-less" approach to barriers as well, given the following revised S-structure:

98

P. Coopmans

(10)

?Who .. (cp whether [w to [yp consider [Q, t'[u> t .. intelligent]]]]]

If VP is inherently transparent under θ-government, movement of who from the embedded COMP position straight to the matrix COMP crosses CP as the only barrier for movement, exactly similar to the Chomsky's original treatment I conclude that in an overall explanation of what counts as an island it is perhaps possible to reduce the burden on an adjunction theory to allowing for the distinction between opaque and transparent adjuncts. In this note I have shown that a locality theory without intermediate adjunction requires a small reinterpretation of minimality along the lines proposed by Stevenson & Coopmans (1989) to account for adjunct extraction. The escape route via VP-adjunction seems to do more harm than good. The extraction facts are equally well explained without intermediate adjunction if θ-government by INFL is sufficient to make VPs intrinsically transparent.

NOTES * This is the original title I had in mind for this note. Due to various limitations, I will say little on how bars fare under the barrier proposal, particularly Muysken's (1982) interpretation of bars in terms of the features "maximal" and "projection", an approach which Henk van Riemsdijk has so often discussed favourably, in both oral and written work. It is best considered as a working title, to be replaced by the more appropriate, though less catchy: "adjunction, θ-government and minimality". 1. Alternatively, the same effect is obtained if INFL is taken as lexical enough to count as an "Lmarker". The difference between the two approaches involves different predictions with respect to the transparency of categories which are L-marked but not θ-govemed, e.g. the subjects of ECM complements. For some discussion, see Coopmans (1988). 2. If whether is base-generated as the head of CP and moved to spec CP at LF, then subjacency is met all the way through (cf. Chomsky 1986; 50). The marginal status of (2) will then be caused by the extra cost of vacuous LF movement. 3. Unless, as noted before, whether occupies the head position of CP, and who may have moved through the spec position. Still an ECP violation will result, caused by the fact that the presence of whether will then give rise to a minimality effect, analogous to (Ao/-trace. 4. I only give here cases of subject extraction, for the simple reason that object extraction "almost vacuously" satisfies the ECP if θ-government is one of the two ways in which a trace can be properly governed.

REFERENCES Chomsky, N. (1986) Barriers, MIT Press, Cambridge (Mass.). Chomsky, N. (1988) "Some Notes on Economy and Derivation of Representation", in Laka, I. and A. Mahajan (eds.), MIT Working Papers in Linguistics 10. 43-74. Coopmans, P. (1988) "On Extraction from Adjuncts in VP", in Borer, H. (ed.), Proceedings of WCCFL 7.53-65. Huang, J. (1982) Logical Relations in Chinese and the Theory of Grammar. PhD Diss. MIT.

A Note on Bars and Barriers

99

Kayne, R. (1981) "On Certain Differences Between French and English", Linguistic Inquiry 12, 349-371. Kitagawa, Y. (1986) "Barriers to Government", in Berman, S., J-W. Choe and J. McDonough (eds.), Proceedings of NELS 16, 249-273. Lasnik, H. and M. Saito (1984) "On the Nature of Proper Government", Linguistic Inquiry 15. 537-577. Muysken, P. (1982) "Parametrizing the Notion 'Head'". Journal of Linguistic Research 2, 57-75. Pesetsky, D. (1984) "Extraction Domains and a Surprising Subject/Object Asymmetry", GLOW Newsletter 12, 58-60. Stevenson, S. and P. Coopmans (1989) "How Extraction from Finite and Infinitival Complements: A Surprising Asymmetry", ms. University of Maryland/University of Utrecht

Auxiliaries and sentence structure in Romanian Carmen Dobrovie-Sorin University of Paris 7

Romanian auxiliary constructions (namely the "present perfect", future and conditional paradigms) show outstanding properties that set this language apart among the other Romance languages. The aim of this paper is to show that these characteristics can be derived as a consequence of the particular sentence structure of Romanian.

1. THE DATA

1.1. Romanian auxiliaries are strictly adjacent to the lexical verb:1 (1)

a.

b.

Am (*adesea) väzut (adesea) filme bune '[I] have (*frequently) seen (frequently) good movies' Voi (*adesea) vedea (adesea) filme bune '[I] shall (*frequently) see (frequently) good movies' A§ (*des) merge (des) la cinema '[I] would (*frequently) go (frequently) to the cinema' Elevii täi au (*to{i) citit (tofi) un poem de Verlaine 'Your students have (*all) read (all) a poem by Verlaine' Elevii täi vor (*toü) ein' (toti) un poem de Verlaine 'Your students will (*all) read (all) a poem by Verlaine' Elevii täi ar (*to(i) citi (toti) un poem de Verlaine 'Your students would (*all) read (all) a poem by Verlaine'

Assuming that certain adverbs and floating quantifiers are generated in a pre-VP position (cf. Emonds (1978), Pollock (1989)), the strict adjacency between auxiliaries and lexical verbs can be captured by assuming that the verb has raised out of VP, as in (2): (2)

S -» NP Infl Aux V [Adv [yp tv...]]

t i t

However, this representation cannot be adopted, because there is no available landing position for the raised verb. The fact that (2) is illicit accounts for a

102

C. Dobrovie-Sorin

quite well-established generalization, according to which the presence of an auxiliary forces the lexical verb to stay under VP (on this see English modals and have, as well as French avoir/etre).2 Romanian falls outside the scope of this generalization, but this does not mean that we are allowed to assume the representation in (2), which is illicit by virtue of the general principles of grammar (structure preservation). Our task will be to find an adequate landing position for the raised verb. 1.2. The second peculiarity of Romanian auxiliaries is their incapacity to carry Tense features. Consider (3)-(5) and compare the corresponding English glosses: (3)

am/ai/a/am/ali/au mincat (plecat) [I/you/he/...] have/has eaten (left) Ι ate'

(4)

vor/ar fi mincat3 [they] will/would be eaten 'they will/would have eaten'

(5)

mincasem/mincasei/mincase eat-Past Perfect-AGR 'I/you/he had eaten'

We are not going to provide a detailed analysis for these examples, which are relevant here only from a comparative point of view: in English and French (and in fact in Romance and Germanic languages in general) certain compound tenses are built with the "Perfect" auxiliaries have/be, which take Present, "Past Tense/Imparfait", and Future morphological inflections (or auxiliaries, see the English case); this strategy is absent in Romanian, which uses instead a bound morpheme (for the "Past Perfect") and two distinct Perfect auxiliaries (ami aila ..., historically derived form habeo, and invariable fi "be"), both of which are used with any kind of verb (unaccusatives, transitives or intransitives), but in different contexts (see (3) and (4)). The incapacity of Romanian auxiliaries to combine with Tense inflections also characterizes the future and conditional paradigms: Romanian does not present any "future in the past", nor any auxiliary comparable to would (obtained from will and -ed). 1.3. Romanian presents no context in which the subject NP (be it nominal or pronominal) can intervene between an auxiliary and the lexical verb: in other words we find no rule comparable to the French subject-clitic inversion (see (6)), nor the Aux to Comp rule characteristic of Italian and Portuguese (cf. Rizzi (1982), Raposo (1987)):

Auxiliaries and sentence structure in Romanian (6)

a. *a Ion/el plecat? b. *va Ion/el pleca? c. *ar Ion/el pleca?

103

'has John/he left' 'will John/he leave' 'would John/he leave'

2. MONOCLAUSAL AUXILIARY STRUCTURES ARE UNAVAILABLE IN ROMANIAN

By the characteristics presented above, Romanian auxiliary structures contrast with Romance4 and Germanic Perfect auxiliaries and with English modals. We are going to show that this contrast derives from two different parametrical choices concerning sentence structure, namely (7), which according to Sportiche (1988) is the underlying sentence structure for English and French, and (8), which we assume for Romanian (cf. Dobrovie-Sorin (1987, forthcoming)): (7)

S (= IP) -> [SpecJ.NP8] [j.V-Infl [Vm„ t. [vp tv NP0]]]

(8)

S (= ff) -> Spec,!' [,-V-Infl [Vm„NPs [^ tv NPJ]]

τ

ι

(7) and (8) are alike with respect to the base position of the NP subject,5 but differ with respect to the S-structure position of the subject: in (7), but not in (8), the NP subject necessarily raises to (Spec.11).6 The following definition (cf. Gueron and Hoekstra (1988), Rouveret (1987)) will be assumed for Tense, Mood and Aspect auxiliaries:7 (9)

Auxiliary verbs take VP complements (or functional projections of V) and do not assign any th-role (in particular, not even the external th-role).

Let us assume that auxiliaries defined as in (9) could in principle show up in any type of language, and in particular in the ones corresponding to (7) and (8), respectively. (10a,b) correspond to the structure in (7): (10)

a.

S (= IP) -> tspecj-NP,] [rlnfl [AuxPAux [Vm„ t. [WV NPJ]]]

b.

S (= IP) -> tspecj-NPJ [,-Aux [Vm„ L tvpV NPJ]]

t

1;

t-

J

|

I

A structure such as (lOa) can be assumed for the English have and for the French etre/avoir, which are generated outside Infl and move to Infl between D-structure and S-structure;8 (lOb) is currently assumed for English modals, which are directly generated under Infl. These two structures correspond to

104

C. Dobrovie-Sorin

grammatical sequences, and as such they must be ruled in. The condition in (9) requires that Vmax should not count as a predicational constituent. We may assume that this condition is satisfied because the position occupied by t, is not an "Α-position", but just a th-position (see footnote 6), Vmax thus reducing to VP. Consider next the structure in (8), where the auxiliary could be inserted either above VP or above Vmax: (11)

a. b.

0 S (= Π>) -> Spec,!' [rlnfl [Vm„NP, UuxpAux [ypV NPJ]]] t_ I 0 S (= IP) -> Spec,!' [,.Infl [AuxPAux [Vm„NPs [WV NPJ]]] T I

These configurations are both illicit (as indicated by the diamond), because the auxiliary takes Vmax as a complement (at both D- and S-structure in (lib), and at S-structure in (lla)). Since by hypothesis the NP, position dominated by Vmax is an Α-position in Romanian (see (8)), Vmax counts as a predicational constituent, and not as a verbal projection, in violation of (9).

3. BICLAUSAL AUXILIARY STRUCTURES IN ROMANIAN

3.1. The only abstract possibility that we may still consider for a language like Romanian, characterized by (8) and (9), is a structure in which Aux is inserted above IP; in other words Romanian auxiliary structures are biclausal configurations. Since in the general case ΓΡ projections are necessarily governed by Comp, it is reasonable to assume that Romanian auxiliaries take CP complements rather than IP complements. Biclausal auxiliary structures are indeed instantiated in Romanian, as shown in (12a), which should be kept distinct from (12b). (12a) is a future paradigm constructed with the verb a avea 'to have' followed by a id subjunctive, while (12b) is a subject raising construction: (12)

a.

b.

Ion are s -1 conduc la gar John has scP -him take^ to the station 'John will take him to the station' Ion urmeaz sa-1 piece la munte John follows sa leaveMbj leave for the mountains 'John is to leave for the mountains'

We can understand why in (12a) floating quantifiers and VP adverbs cannot intervene between the auxiliary and the verb (this is also the case for (12b)):

Auxiliaries and sentence structure in Romanian (13)

a.

b.

105

Ion are (*adesea) sä einte (adesea) la plan John has (*frequently) sä play^j (frequently) the piano 'John will frequently play the piano* Copiii au (*toti) sä einte (toti) la pian the children have (*all) play^ (all) the piano "The children will all play the piano'

Assuming that VP adverbs and floating quantifiers are generated in the embedded pre-VP position, the subjunctive verb necessarily raises past them (and also past the embedded NP subject, see the structure in (8)). However, we do not understand why (14a) is ungrammatical, as opposed to (14b): (14)

a. *Miine are (§i) Ion sä piece la munte tomorrow has (also) John sä leave^ for the mountains b. ?Miine urmeazä §i Ion sä piece la munte10 tomorrow follows also John so leave^ for the mountains

The impossibility of subject inversion illustrated in (14a) brings to mind the behavior of Romanian auxiliaries we mentioned under 1.3. Moreover, the verb a avea 'to have' cannot take Tense inflections when it enters the composition of the future. Again, the verb a urma 'to follow' does not show this restriction: (15)

a. *Ion avea sa-1 conduca la gara a doua zi b. Ion urma sa-1 conduca la garä a doua zi

(In (15a-b) avea and urma are in the "imparfait"; the glosses are otherwise as in (12a-b).) How can we explain the contrast in (14a) vs (14b)? Assuming the structure in (8) and V raising to Infl in both the embedded and the main clause, we would get (16), where C" dominates sä:

106

C. Dobrovie-Sorin

Given the grammaticality judgments in (14a-b), the structure in (16) should be ruled out for (14a) and ruled in for (14b). This contrast can be explained if we assume minimally that a avea, unlike a urma, is an auxiliary verb, i.e., a verb that must obey (9). This requirement is violated in (16): in the position it occupies in Infl, after V raising, are takes a predicational type of complement,11 the upper Vmax. We are thus led to assume that when it belongs to the future paradigm, are 'has' cannot raise to Infl. It is reasonable to assume that in a language characterized by (8), Nominative Case can only be assigned rightwards, by the verb raised to Infl. Since in (16) auxiliaries are not allowed to move to Infl, we are led to assume that there is no A position for the subject in the main clause of (12a); the pre-auxiliary position that the NP subject occupies in (12a) must be assumed to be an A'-position, accessible to subjects and non-subjects alike. In other words, the Romanian future construction illustrated in (12a) is "biclausal", because the auxiliary is outside the clause in which the main verb appears; but this biclausal structure is defective, in so far as the main clause presents no NP subject position and no Infl node (since V-to-I movement cannot apply, the Infl node cannot surface). We then end up with the structure in (16'):

Auxiliaries and sentence structure in Romanian

107

(16') IP Aux

CP

f\ IP

C„

(16') presents no slot between Aux and C", occupied by ία, which is itself strictly adjacent to the inflected verb (cf. Dobrovie-Sorin (forthcoming)). This structure explains (14a) quite straightforwardly; (15a) is also accounted for, if we assume that the only way for the verb to get Tense inflections is to raise to Infl.12 Let us now turn to (12b). A urma is a raising verb (cf. urma s vino to{i elevii a doua zi, 'it followed that come all the students the following day, all the students were to come ,..'),13 which is like auxiliaries in so far as it does not assign an external th-role. But unlike auxiliaries, raising verbs do not fall under (9), and by way of consequence they are not only allowed to, but necessarily raise to Infl.14 The word order in (14b) is expected,15 as well as the possibility of a urma to take Tense inflections. 3.2. We have so far shown that the structure in (16'), which is the only abstract configuration that we can assume for auxiliary structures in a language with a structure as indicated by (8), is instantiated by the future paradigm in (12a). This structure directly accounts for the data in (13a), (14a) and (15a), which appear to be parallel to the properties presented in 1.1-1.3. for the apparently monoclausal auxiliary structures. Let us then come back to these paradigms. The distribution of clitics that characterizes them seems to constitute important evidence in favour of a mono-clausal structure; clitics precede the auxiliary, and cannot precede the lexical verb: (17)

a. b.

L-am rugat L-a§ ruga16

[I] him-have asked [I] him-would ask

However, the distribution of clitics in Romanian is somewhat more complicated. Compare in particular (17a,b) to (18) and (19):

108

C. Dobrovie-Sorin

(18)

a. b.

(19)

a. il a§tept/L-astept b. Π am in buzunar/*L-am in buzunar c. *Ϊ1 am rugat/L-am rugat

Am rugat-o A§ ruga-o

[1] have asked-her [I] would ask-her [I] him wait [I] it have in the pocket [I] him have asked

Examples (18a,b) illustrate a well-known "idiosyncracy" of the Romanian clitic system: unlike all the other clitics, the feminine singular ο must follow the lexical verb in compound tenses (the examples (18a,b) are ungrammatical if we insert ο in front of the auxiliary); in simple tenses ο precedes the verb, just like the other clitics. Quite obviously the contrast in (17)-(18) cannot be assumed to be "syntactic": rules of syntax are probably not sensitive to specifications such as "feminine singular". By way of consequence the only solution to the problem is to assume that one of the two positions in (17)-(18) is "basic", or "syntactic", the other one being derived in Phonological Form, by phonological rules. This hypothesis is probably undebatable; the only question to be settled then is the position that we assume to be "syntactic": the one in (17) is currently assumed by Romanian linguists (including Dobrovie-Sorin (1987)) to be "regular", the one in (18) is considered to be "idiosyncratic". We think that this is incorrect. First because downgrading rules are very marginal (if they exist at all). Second, the clitics in (17) do not show the phonological shape they normally can show in pre-verbal sequences, namely the first alternative, il, given in (19a,b). The second alternative in (19a,b) is possible, but not obligatory, in certain phonological environments (/- is possible instead of il, if the verb following the clitic starts by a non-stressed vowel; hence the contrast between *l-am '[I] him/it-have' and l-a§tept '[I] him-wait [for]' in (19a,b)). The impossibility in (19c) thus appears to be isolated if we assume this clitic to be generated on a par with the clitics in (19a,b). We can understand (at least partially) this impossibility if we assume that in (19c) the clitic does not occupy the same syntactic position as the clitics in (19a,b), but rather the postverbal position which can be directly observed for o. When they appear in the postverbal position (see positive imperatives, gerunds and certain subjunctives), Romanian clitics show necessarily their "weak" forms (m-, t-, I-, i, etc.; compare ma, te, il, ii, etc.). The compulsory weak form in (19c) (/- is allowed, and U impossible) would then be due to the fact that in the syntax the clitic occupies the postverbal position characteristic of o. Thus, if we take the position of o in (18) to be the relevant one, the main evidence in favour of a mono-clausal structure disappears, and we can assume that these auxiliary structures are also bi-clausal, on a par with the compound future examined in 3.1. Let us be somewhat more explicit concerning the derivation of (17). Its abstract structure would be essentially the same as the one underlying (12a), i.e. (16*). There is just one question to answer: why is the clitic preverbal in (12a),

Auxiliaries and sentence structure in Romanian

109

but postverbal in (18)? In Dobrovie-Sorin (forthcoming) we propose that in Romanian the order V-cl is obtained by a rule of V proposing that moves the inflected verb to C,17 leaving object clitics behind. Interestingly, V preposing generally correlates with the absence of lexical complementizers; and indeed, in (17), the particle a characteristic of Romanian infinitives is absent18 (compare (18b) with inainte de a o Intllni, before that to her meet, 'before meeting her'), which constitutes important evidence in favour of V preposing and the biclausal structure given in (16'), that we propose for Romanian auxiliary structures. This structure directly accounts for the characteristics presented in 1.1.-1.3.: the embedded Infl node is a licit landing position for the obligatory movement of V out of VP that must be assumed to account for the data in (1); the auxiliary cannot raise to Infl, which accounts both for the fact that the NP subject cannot intervene between the auxiliary and the main verb, and for the inability of Romanian auxiliaries to take Tense inflections (the discussion of (14a) and (15a) directly transposes to the data in 1.2. and 1.3.). Under our analysis, the structure in (16') depends on the constituent structure of Romanian; more precisely, (16') is the only possible configuration for auxiliary structures in a language characterized by (8).

NOTES 1. The only elements that can intervene between Aux and V are monosyllabic adverbs belonging to an extremely restricted class (five items), which are arguably clitics (cf. Rivero (1988) and Dobrovie-Sorin (forthcoming)). Note that in structures with no Aux, these adverbs show up between the clitic - or the negative adverb, in case the clitic is absent - and the verb; this position is not accessible to other adverbs. 2. This matter is somewhat more complicated: in certain Aux VP constructions, the lexical verb may carry inflections (participial or infinitival), and we may assume that they are generated in a syntactic position outside VP (cf. Baker, Johnson and Roberts (1989)). 3. The auxiliary fi 'be' is also used in perfect infinitivals and perfect subjunctives: a fi mincal/aj vrea s ft mlncat 'to have eaten ΤΙ would like that I have eaten'. 4. The data in 1.2. and 1.3. distinguish Romanian from French, Italian, Spanish, and Portuguese, while 1.1. also characterizes Spanish and Italian. Note however that in Italian floating quantifiers are allowed to intervene between the Perfect auxiliary and the past participle. 5. The hypothesis according to which the NP subject is generated as a sister of VP (Vmax is some kind of small clause constituent) can be found (with certain technical variations, which concern the categorial status of Vmax and VP) in Kuroda (1986), Contreras (1987), Kitagawa (1986), Speas and Fukui (1986), Speas (1986), Zagona (1982), and Dobrovie-Sorin (1987). 6. The (SpecJ1) position thus has a clearly different status in (7) and (8): in (7), (SpecJ') is an Aposition (the "canonical" structural position of the subject), while the base position of the subject is a th-position, but not an A-posiuon; in (8), on the other hand, (SpecJ') is an A'-position (the landing site of left dislocations), which is accessible to subjects and non-subjects alike, while the position in which the subject is generated is its canonical A-posilion. 7. The passive auxiliary, as well as modals such as the Romanian putea 'can, may', do not fall under this definition (cf. Lema and Rivero (1989) and Dobrovie-Sorin (forthcoming)).

110

C. Dobrovie-Sorin

8. The structure may in fact be somewhat more complicated: Vmax may be governed by an Infl node (cf. Kayne (1987), and Baker, Johnson and Roberts (1989)). 9. We have preserved id in the glosses, because this subjunctive particle has no exact counterpart in English. The reader may translate s by thai, but the two elements are syntactically different 10. The question mark indicates a quite low degree of acceptability, which is presumably due to Topic-Focus requirements; the contrast with (14a) is nevertheless very clear for all Romanian speakers, who cannot even interprete (14a). 11. We must thus assume that the condition on auxiliaries stated in (9) applies not only at D-structure, but also at the level of representation immediately following the rules triggered by morphological subcategorization (cf. Lieber (1980)), i.e. V raising to Infl. However, this condition does not hold at later levels of representation, obtained by the application of purely syntactic rules such as Aux-to-Comp. 12. Note that we must assume that AGR inflections can attach to the verb in the absence of V raising to Infl (avea is inflected for AGR), but we shall leave the technical details open for further research. Note also that AGR features are not required by any principle. We therefore expect them to be absent, which is indeed the case in another Romanian future construction, which is identical to (11 a) with respect to the embedded clause, but differs from it by the use of an uninflected element ο: ο sa pledplecilpiece ... o s leave-lpAeave-2p, ... 'I'll leave, etc., I am going to leave'. 13. On raising (and control) in sa subjunctives cf. Rivero (1987) and Dobrovie-Sorin (forthcoming), where a hypothesis different from the one in Dobrovie-Sorin (1987) can be found. 14. Correlatively, a (non-thematic) structural position for the subject is projected, which at Sstructure appears on the right of the auxiliary. In (14b) this position is occupied by Ion, which has been raised from the subject position of the embedded clause. The difference between raising and auxiliary verbs thus appears to reside in the number of structural subject positions (two vs. one). Note that this is also true of English and French (recall that the VP internal subject position does not count as a structural subject position in these languages). The difference between these languages and Romanian concerns the location of the missing position (subject of V vs subject of Aux). 15. Note that under our analysis of Romanian, this order is a result of V-to-Infl, while in Romance and Germanic languages the same surface order is obtained by "V second" (V-to-Infl-to-Comp). 16. The future construction in voi/vei/va ... is underlied by the same structure as (17a,b), but a number of phonological rules should be specified, to explain the fact that ο is allowed to appear in two positions, see voi vedea-o and o voi vedea '[Ϊ] will see her'. 17. This rule also applies in Romance positive imperatives (cf. den Besten (1983)), as well as in an important number of other (old and modem) Romance constructions, cf. Dobrovie-Sorin (forthcoming). 18. This particle occupies the same position as the subjunctive particle s , presumably C°, or some nearly equivalent node, cf. Dobrovie-Sorin (forthcoming) and Farkas (1989).

REFERENCES Baker, M., K. Johnson and I. Roberts (1989) "Passive Arguments Raised", Linguistic Inquiry 20, 2, 219-253. den Besten, H. (1983) "Root Transformations and Lexical Deletive Rules", in Abraham, W. (ed.), On the Formal Syntax of the Westgermania, Benjamins, Amsterdam/Philadelphia. Contreras, H. (1987) "Small Dauses in Spanish and English", Natural Language and Linguistic Theory 5, 225-244. Dobrovie-Sorin, C. (1987) Syntaxe du roumain. These de Doctoral d'Etat, Universit£ de Paris 7.

Auxiliaries and sentence structure in Romanian

111

Dobrovie-Sorin, C. (forthcoming) The Syntax of Romanian: Comparative Studies in Romance, Foris, Dordrecht. Emonds, J. (1978) "The Verbal Complex V'-V in French", Linguistic Inquiry 9, 151-175. Farkas, D. (1989) "On the Morpho-Syntax of Subjunctive Clauses in Rumanian", International Journal of Rumanian Studies, 6,2, 7-19. Gu£ron, J. and Hoekstra, T. (1988) "T-chains and the constituent structure of auxiliaries", paper presented at GLOW 1987, Venice. Kayne, R. (1987) "Facets of Romance Past Participle Agreement", ms., ΜΓΓ. Kitagawa, Y. (1986) Subject in Japanese and English, PhD Diss., University of Massachussets, Amherst. Kuroda, Y. (1986) "Whether We Agree or Not", ms., University of California at San Diego. La Jolla, California. Lema, J. and M.L. Rivero (1989) "Inverted Conjugations and V-second Effects in Romance", paper presented to the XIXe Linguistic Symposium on Romance Languages, Columbus, Ohio, 1989. Lieber, R. (1980) On the Organisation of the Lexicon, PhD Diss., MIT, Indiana University Club. Pollock, J.-Y. (1989) "Verb-Movement, UG and the Structure of IP", Linguistic Inquiry 20,3. Raposo, E. (1987) "Case Theory and Infl-to-Comp: the Inflected Infinitive in European Portuguese", Linguistic Inquiry, 18,1. 85-109. Rivero, M.-L. (1987) "Barriers and Rumanian", in J. DeCesaris and C. Kirschner (eds.). Proceedings of the XVIIlh LSRL, Benjamins, Amsterdam. Rivero, M.-L (1988) "The Structure of IP and V-movement in the languages of the Balkans", ms.. University of Ottawa. Rouveret, A. (1987) Syntaxe des dipendances lexicales. Identiie" et identification dans la Theorie Syntaxique. These de Doctoral d'Etat, Universit£ de Paris 7. Rizzi. L. (1982) Issues in Italian Syntax, Foris, Dordrecht. Speas, M. (1986) Adjunctions and Projections in Syntax, PhD Diss., MIT. Speas, M. and N. Fukui (1986) "Specifiers and Projections", MIT Working Papers 8, MIT, Cambridge, Massachussets. Sportiche, D. (1988) "A Theory of Floating Quantifiers", Linguistic Inquiry 19, 3, 425-449. Zagona, K. (1982) Government and Proper Government of Verbal Projections, PhD Diss., University of Washington, Seattle.

Onset clusters in Greek G. Drachmen University of Salzburg

0. INTRODUCTION1

This paper opens, though it by no means fully answers, the question why clusters are the way they are and not otherwise in Standard Modem Greek (SMGk) syllables. As a preliminary report, it is confined largely to word-initial clusters. After taxonomising the distribution of consonants in such clusters, I briefly survey relevant parts of a number of contemporary theories in this domain. I then offer a principled extension of an X-Bar-Syntax based analysis of the notion Syllable, viz., an attempt to define a constituent "Onset", to be generalised to "Consonant" in a later work. The resultant account makes crucial use of the notions Sonority and Government It re-elucidates the classical distinction between systematic-gap vs. accidental gap in distribution. And it characterises part of the notion "soundchange" as a change in Conditions on syllabification. I have concentrated here on simply establishing the framework, drawing examples almost entirely from the analysis of Word-Onsets. This framework will be justified more fully elsewhere, wit alternations in syllabification (especially those word-medially), alternations apparently "forced" on segmentsequence Representations by the very principles and parameters established here.

1. DESCRIPTION

1.1. For a preliminary, taxonomic survey, let us assume the elementary syllable structure: Syllable = Onset Nucleus Coda 1.2. Then for SMGk a) An Onset may be any consonant from among [ p t t s k f 9 s x v 6 z y m n r 1]; or any of certain clusters of these, such as Obstruent+Consonant, Spirant+Obstruent+Resonant. b) A Nucleus is at least a Vowel. c) A Coda-consonant, if present, consists of [s] or a Resonant

114

G. Drachman

1.3. Our interest in the structure of the SMGk syllable will in the first place be directed to word-initial (Onset) clusters. With this analysis as a first step, and taking into account some possibly exceptional features of Onset clusters (as well as Codas) in SMGk as in languages in general, we will in a later report discuss the remaining cases, viz., the representation of [j, b/d/dz/g] as well as the cases of medial clusters.

2. THEORIES ON SYLLABLES

What follows by no means constitutes summaries of the relevant theories; rather, only those points immediately relevante to the present discussion are mentioned. 2.1. Traditional accounts, e.g., those in Sievers (1881), Saussure (1916) and Jespersen (1904), concern the Sonority effect on syllable structure. 2.2. Steriade )1982) a) Assumes there is no fixed universal sonority scale.2 b) Rather, the possible inclusion of a particular Distinctive Feature is a parameter of variation. Thus, Classical Gk uses Voicing3 but Latin uses Place of Articulation. c) Minimal Sonority Distance constitutes a cross-language parameter. d) Extra-syllabicity is a further cross-language parameter. e) Stray-adjunction is employed for consonants not handled by "core" syllabification. 2.3. Clements (1988). a) Core syllabification employs a DF-based Universal Sonority hierarchy. Thus (with sonority increasing Left-to-Right): { Obstruent > Sonorant >Approximant > Vocalic > Syllabic } b) Peripheral exceptions are allowed for, but on the assumption that implicational universals will be honoured. 2.4. Kaye-Lowenstamm-Vergnaud (KLV) (1988). a) There are no syllables, only Onset-Rime sequences. b) A three-valued property "charm" is invoked, with the distribution: (i)[-charm] segments are Obstruents except [s]; (ii)[neutral charm] segments are all Resonants, plus [s]; and (iii)[+charm] segments are e.g. most vowels. c) Within a legitimate branching Onset, Left-to-right strictly local government must obtain: thus a branching Onset may only have two members, viz., a [-charm] segment followed by a [neutral-charm] segment.

Onset clusters in Greek

115

d) A Coda is not a separate constituent but the second part of a Rime; governed by the [4charm] Nucleus, it can only be a [neutral charm] consonant. 2.5. Michaels (1988). a) A modest X-Bar Syllable structure is assumed, with obligatory Spec (Onset) position but no internal structure for consonant clusters. b) Language-specific Coda-licensing and Coda-content conditions may be superposed on syllable structure. c) Move-α may apply. d) On the other hand, the notions Sonority, phonotactics, and even Inventory have no theoretical status in Michaels' Syllable Phonology.

3. CODAS

We first clear up the matter of Codas in SMGk with a preliminary statement, to be elaborated on elsewhere. Here, only Coda-content is treated, while Michaels' Coda-licensing through stress is for the moment not taken into account. 3.1. For SMGk, Codas are optional. The coda-consonants found are [s,n,l,r]. And the corresponding Coda-conditions proposed are: i) [+Coronal]4 ii) if [-Voi], then [+Cont] iii) if [+Voi], then [Resonant] Note that if any of the conditions cannot be met, then either an adaptation follows, or the consonant is uninterpretable and is deleted. 3.2. The operation of the parts of this constraint is clearly illustrated in certain Noun paradigms of SMGk. Thus: 3.2.1. For the first part, i.e., "if [-Voi], then [+Cont]", note that a stem like (Neut.Nom.Sg.) [kreat] "meat" retains its final [t] when followed by a vowel, as in Gen.Sg. [kreat-os], or Nom.Pl. [kreat-a], but shows the alternation with [s] when no suffix follows, as in Nom.Sg. [kreas]. 3.2.2. And for the second part, i.e., "if [+Voi], then [Resonant]", note that a stem like (NeuLNom.Sg.) [to paron] "the present time" exhibits a final [d] when followed by a vowel, as in the Gen.Sg. [parond-os]; but shows no [d] when no suffix follows, as in the Nom.Sg. Assume the [d]-final form to be the dictionary stem. Then, since the stem contains a pre-fmal nasal, no alternation in fact occurs; instead, loss of [d] follows as soon as the Coda-condition applies, simply leaving the [n] in final position. 3.3. But the Condition on Codas will almost certainly prove to be too general. Indeed, medial clusters such as -kpt-, -fkl-, or even -fstr- or -mbr-, where the

116

G. Drachman

first element is quite outside the potential of our Onset, and is thus necessarily a coda to the preceding syllable, suggest we constrain the Coda Condition to "word-final". As a word-periphery condition, this recalls Steriade's notion of "Word-Template". But such modifications and extensions can hardly be treated here.

4. EXTENDING THE NOTION "SYLLABLE"

4.1. Suppose we set up an X-Bar syllable. The literature discloses a wide spectrum of proposals on syllable-internal constituent structure. Representative syllable make-ups might be: a) No hierarchical structure - from Kahn (1976), as also Clements-Keyser (1983). b) (Onset Peak Coda) - from Hockett (1955). c) (Onset (Rime (Peak + Coda))) - from Kurylowicz (1948). d) ((Onset+Peak) Coda) - from Yoshiba (1983). e) Demi-syllable + demi-syllable (overlapping at the Peak) - in Clements (1988). 4.2. Here, according to X-Bar theory, we shall in the first place allow (in what follows, X° refers to a Head, X' to the immediate Projection of a Head, and X" to a maximal Projection): Syllabic" -> Spec-Syu" + Syll' SylT -> Syllabic0 + Complement which corresponds to the tree: -SylTSpec-Syll'

ι S° Head-V

Syll' Complement

Note that we allow optimally one element per position. Thus far we follow Michaels. 4.3. Coming now to Spec-SylF, the Onset position, we will begin by assuming a certain minimal structure (cf the Determiner-projection DP inside the Inflexion-projection IP in syntax). The (Maximal) Onset projection is thus:5 Onset" -> Spec-Onset' + Onset' Onset' -> O° + Complement again corresponding to:

Onset clusters in Greek ι

117

Onset"

Spec-Onset'

1

ι—Onset'i—unset O°

1

Complement

We assume that a single element occupies the Head position. Note that, though we do not discuss this in detail here, it is a clear implication of X-Bar theory that, since a Head must be present, every Onset will in principle contain at least that Head consonant - even if its position is occupied by "zero".

5. SOME CONSEQUENCES FOR SMGK

5.1. Consider the configurations available. a) In the single-consonant configuration, only the Head position is (indeed, as the basis of the projection, must be) occupied. b) In the two-consonant configuration, the Head position is occupied, and so is the Complement position. Here the barest of theories would perhaps not need to appeal to Sonority at all (the position of Michaels), but simply allow clusters like {sp} (ps), as well as those like (pr) or (pn) without further comment. 5.2. However, even at this level of analysis, the three-member clusters will require more elaborate treatment a) Recall that these clusters contain (e.g.)(s + stop + R}. b) We could simply assign [s] to Spec without further ado, on the minimal assumption of "one element per position". c) This would simply be the formal equivalent of Steriade's extrasyllabicity for initial [s] in the case of anti-sonority clusters. d) However, we have so far not allowed ourselves the luxury of a Sonority hierarchy, so that assigning the [s] of (s + stop + R) clusters to Spec-Cons is in a sense quite ad-hoc - recalling that in principle we allowed binary clusters of type {s p} to occupy the Head & Complement positions under Onset'.

6. FURTHER ASSUMPTIONS

6.1. We thus assume a rather more articulated structure for the constituent Onset", viz., a) that the Spec-Onset' position, as in syntax (cf. Fukui-Speas 1986), is underlyingly an empty one, one to which an element may in certain circumstances migrate if it cannot be licensed in its original position - cf. the case of Sentence-Subjects, originating under VP, but whose Norn-Case is only available

118

G. Drachman

in Spec-IP; or that of Genitive NP-Subjects such as "John" in the structure [NP* [ 's [book John]]], giving "John's book" by movement to Spec-DP. Notice that we have assumed, in contrast to Michaels wrt Spec-Syu", that the Spec-Onset' position is only optionally occupied. It seems that the "Maximize onsets" Principle is at least partially6 reducible to the obligatoriness of SpecSyll', without the need for obligatory Spec-Onset' too. Note too, that (cf. 4.2. above) in leaving Spec-Syll' obligatory we indeed intend to imply that there are no onset-less syllables. 6.2. We add one very simple and informal Phonological-Sonority postulate: Heads demand more sonorant Complements. We do not debate the issue of the phonetic basis for such a principle. Rather, we see that though the syllable-internal Sonority-increase of Sievers-JespersenSaussure still applies, it is here a purely phonological principle; in fact it holds only within Onset', but not between Spec-Onset' and Head of Onset'. This domain constraint on Sonority probably applies because the phonological principle of Sonority obtains only under government; whereby we assume that Heads govern uniformly Left-to-Right within their constituents,7 so that a Specposition is not governed by a Head. Cf., however, the argument of 7.5.3. below. The Sonority scale is presumably universal, and rises from Stop to Vowel. Thus:8 Stop < Fricative < Resonant < Semi-vowel < Vowel 6.3. With these principles, we assume free assignment of consonants to positions within Onset", allowing the principles themselves to decide whether a particular configuration is properly licensed (hence, allowed) or not - including the possibility (given the fixing of whatever parameters prove to be relevant) of distinguishing between systematic vs. accidental gaps in SMGk Onset clusters. We assume that the least marked situation will be preferred, viz., the one in which both Head and its complement position are first taken up, reserving the occupation of Spec for (e.g.) anti-sonority clusters. 6.4. Note that the present account probably makes unnecessary the distinction (e.g., in Clements 1988) between "core" and extrasyllabic, later adjoinable (or deletable) material of a syllable. Since all syllabification procedures are simultaneous, there can be no ordering. We have, on the other hand, no constituency-based solution for preferential densities in clusters across languages. Thus we cannot so far replace a) Steriade's linking condition for an obstruent coda, accounting for the difference in permissibility of (e.g.) the SMGk clusters pt/kt vs. *tp/*tk.9

Onset clusters in Greek

119

b) Steriade's "Minimal Sonority Distance"; or Clements' substitute for it, viz., "demisyllable complexity".

7. APPLICATIONS

7.1. Take first a SMGk cluster like [p s], [ p r] or even [p n]. Here [p] is the Head, and the more sonorant [s], [r] or [n] is its complement The interpretation of each of these clusters seems quite unproblematic, and the non-occurrence of such clusters with rightward [r] may therefore be looked upon as automatic, due to the lack of a further position. However, although [t s] would be straightforward, there is some reason to hold that [ts] is a "unit", viz., an affricate.10 But if [ts] were a (single-consonant) Head, the Complement position would be unoccupied, and the nonoccurrence of a cluster such as [ts r] would for us no longer be a principled gap. Thus, we must consider [t s] a cluster rather than a complex segment 7.2. Now consider [sp]. The Sonority Principle forbids what we have called (6.3. above) the least marked assignment, viz., [s] as Head plus [p] as complement And just as NP movement is catalysed by lack of Case and hence interpretation for the relevant NP, so here we will now suppose that to resolve the Sonority violation one consonant will move. In fact, we must suppose that [s] moves. How can we motivate this movement as obligatory? Suppose [s] as Head moves to Spec-Onset' as a kind of alpha-movement. Then it is a clear example of neither Head-movement, nor maximal-projection movement; for maximal projections can move to Spec, but not Heads. On the other hand, that this move would leave O' headless is not necessarily an undesirable outcome, since the antecedent in Spec-Onset' would at least c-command its trace. It should not discourage us that the analog with syntax is very imperfect Since both are modules of the same cognitive sub-system, we naturally seek analogs between syntax and phonology. But phonology, insofar as it is a less richly structured module, will perhaps in principle lack some parts of the syntactic analog; and insofar as it is differently and autonomously structured, there will be negative parts to that analog.11 7.3. Consider next the cluster-types pt ft as distinct from the non-occurring types *ptr, *ftr.12 Suppose O' contains [p] as Head, and [t] as Complement This will effectively block the undesirable [ptr], [ftr], for whose last member no position is in fact available. The sonority-clash triggered by Heads is avoided in the usual way, by moving the first element to Spec; though the cost is that we have, on

120

G. Drachman

the surface at least, headless Onsets. But we reconsider this solution immediately below, in 7.5.3. 7.4. We come to [spr]. As before, we cannot assign [s] as Head with [pr] as Complement; for the [s]-[pr] sequence would violate the Sonority condition. To circumvent this, [s] must move to Spec-Onset'. 7.5. The question now arises, why the following data distribution obtains: a) sp, and ps b) spr, but *psr.

That is, what blocks onsets of the type *psr? Note first that while the occurring cluster [spr] in fact violates the Sonority Condition, the putative but non-occurring cluster *[psr] is perfectly well-formed wrt the Sonority Condition, each element being more sonorous than the one to its left. The impression might thus arise that the Sonority Condition itself is disconfirmed. We shall nevertheless show that the Sonority Constraint is crucial (rather than disconfirmed), and also that the cluster *psr in fact constitutes not an "accidental" but a "principled" gap, at least for SMGk. 7.5.1. On the principle "one element, one position" in the unmarked case, let us try first the analysis (Left-to-Right) O'[Head p [Complement s]]. Since this provokes no Sonority-clash, [p] remains as Head. So far as the leftover r is concerned, there is simply no position available. 7.5.2. Assume now again, as claimed above, that we may try any possible assignment of segments to the positions available, checking later for Sonoritylicensing of a Complement, etc. Then we might as a last resort try the marked (move-alpha) assignment: [ Spec-Onset' p [O' s [Complement r]]]. But were such an assignment allowed, it would of course immediately legitimize this forbidden cluster. How, then, are we to block it in a non-ad-hoc fashion, recalling that the Sonority Condition is not violated here? 7.5.3. Up to now, we have allowed "movement to Spec" fairly freely, as a "rescue" strategy for anti-sonority sequences. But the present case suggests that, just as there are conditions on Coda-content, so there may also be a condition on Spec-content. Taken together with the Coda-condition of 3. above (better, the

Onset clusters in Greek

121

Word-final condition, in the revision suggested in 3.3.). this would of course generalise to the notion that Syllable-Periphery is subject to special conditions. For SMGk, we claim now that there is such a condition on Spec-Onset', viz., that Spec-Onset' may contain only a fricative. This of course allows [s[pr]]. But, since [p] is not a fricative, it blocks *[p[sr]], and of course *[p[sl]] too, as desired. Thus, contra the argument in 6.2., we will after all need to refer to Spec-Head Agreement as the left-oriented parallel to government, as in syntax. 7.5.4. It is obvious that the earlier proposal on *ptr (7.3. above) must be rejected in favour of the present proposal; [p] could not move into Spec, as it now seems, and *ptr is now no longer an accidental gap but rather a principled one, since the only remaining assignment is O'[p [Complement t]], with no position left for [r]. 7.5.5. But then what of "simple" [pt], in which the same sonority violation occurs, and which is thus surely a candidate for "p-in-Spec", especially since this cluster is in fact perfectly licit? In particular, does this invalidate the whole argument concerning *[ptr]? We note that there is an alternation to be accounted for here; the Katharevusa style allows [pt], but the corresponding lexical items in the Demotic style show only the cluster [ft]. Seen as a historical change, there has been a dissimilation, whereby {stop+stop} clusters gave {fricative + stop}: but of course the synchronic outcome [f t] may well be represented under Onset'. For [f] would then move to Spec-Onset', where (as claimed above in 7.5.3.) it would be licensed by the Spec-Onset' Condition. As for the "older" [pt], we might now claim that Katharevusa has retained the more permissive Spec-Onset' Condition obtaining word-initially in Classical Greek, viz., one allowing stops as well as fricatives in word-initial Spec-Onset', while Demotic narrowed that Condition. We thus arrive at the rather interesting reinterpretation of one kind of sound-change, namely that of [Stop-Stop] > [Fricative-Stop] as "change in Peripheral Conditions on syllables".13 7.6. Note that in generalising the [s] in the cluster [s stop] to fricative as we have already done, we automatically cover the sources of [ft] [xt] clusters. But this would then over-generate a) to non-occurring [θ-t], as also b) to the case of O' containing [fricative +stop + Res]. We would then theoretically get as clusters not only the occurring [str] but also the nonoccurring *[ftr], *[xtr]. However, the present theory suggests that we should simply allow all of these as permissible clusters of Greek, their non-occurrence simply coming

122

G. Drachman

under the notion of "accidental gap". And indeed, the "non-occurring" Onsets do occur medially.

NOTES 1. This paper grew out of David Michaels' Seminar in Syllable Phonology held at Salzburg University in summer 1989. Thanks also to A. Malikouti-Drachman. Exculpations for bothl 2. Drachman 1976 had earlier come to a similar conclusion on typological/processual grounds. 3. Following Steriade, Malikouti-Drachman 1984 shows that SMGk also includes the DF Voicing in its Sonority scale. 4. It is hardly an accident that all the possible SMGk Coda (or, Word-Final) segments share the identical Place of Articulation, viz., [+Coronal]. Cf. fn. 12 below. It must also be noted that the set of coda-consonants for SMGk corresponds exactly to KayeLowenstamm-Vergnaud's class of consonants with [neutral charm]. 5. And we add that there may well be an Onset-Max as complement to the Spec position (cf. NPmax within DP for Greek, in Drachman, 1988). We expect the constituent Onset" to generalise straightforwardly to Consonant"; but we leave it as Onset" for this article rather than complicate the exposition. 6. It guarantees non-arbitrary syllable division at least for the basic case, that of [CV.CV]. 7. In Kaye et al. (1988), Ultra-constituent government is L-R (say, in Onsets), inter-constituent government is R-L, say, between an Onset and a preceding Coda; while inter-Nucleus government, i.e., between (projected) Vowels in adjacent syllables, is neutral to direction. 8. We know of no data which would forbid us to reverse this sonority scale simply by labelling it "strength", though we acknowledge Clements' caveat wit Foley's (1977) Inertial Development Principle. All other things being equal, we might have used "decreasing strength" instead of "increasing sonority" in defining the conditions under which a Head licenses its complement. 9. One theory professing to explicate this particular preference, viz., that of Kaye-LowenstammVergnaud (1988), must suppose that, since (for them) stop sequences may not be tautosyllabic and must thus fall under (Right-to-Left) inter-constituent government, [p] and [k] must be governable and thus possess [neutral charm]. 10. The different behaviour of [t s] as against [p s] or [ k s] clusters across morpheme-boundaries suggests that [t s] is not a cluster. 11. A follow-up paper will examine the full extent of the positive, negative, and neutral parts of the syntactic analog. On the syntactic side, it is important to isolate the motivations for movement. Thus comparing motivations, move-alpha in phonological clusters seems to parallel NP-movement best, as a matter of licensing an element in a given position. Thus compare: a) Head-movement and »free-affix. Head-to-Head movement is motivated by the need to "complete" or "attach" otherwise freeremaining affixes, or the need to lexically govern certain empty categories. b) NP-movement and licensing through case. NP-movement is motivated by the Case Filter, perhaps as well as the need to absorb the case discharged to Spec position. c) WH-movement and Scope. WH-movement is perhaps motivated by considerations of Scope-assignment. A cased NPmaxOperator-element moves to non-Cased/non-A Spec-CP position. In addition to the parallels wrt α-movement, success of the programe behind the present theory would favour a purely Representational, as against a purely Derivational approach to alternations in Phonology.

Onset clusters in Greek

123

12. In the case of *[ftr] there is the additional factor that [f] is a marked spirant as compared to the unmarked spirant [s]. And the fuller attestation of voiceless fricative clusters than voiced ones is likewise unremarkable, since [-voice] is the unmarked value of voicing for fricatives. Likewise, Markedness consideration of rather specific kinds may be responsible for other preferences: thus a preference for the [+Coronal] segments [t n s l & j], clearly among the ten commonest non-vowel segments attested in languages of the world. 13. The explanation of [pt] of course carries over to the similar cluster [kt]. Notice, however, that there are a few old pt-initial loans from the Katharevusa into the Demotic style, each of which we must mark as allowing a stop in word-initial Spec-Onset". For the different nature of the block on *[tp]/*[tk], however, see 6.4. above.

REFERENCES Clements, G.N. (1988) "The role of the Sonority Cycle in Core Syllabification", in Working Papers of the Cornell Phonetics Laboratory, No. 2, 1-68. Clements, G.N. and SJ. Keyser (1983) "CV Phonology: a Generative Theory of the Syllable", Linguistic Inquiry Monograph 9, ΜΓΓ Press, Cambridge (Mass.). Drachman, G. (1976) "On the notion 'Phonological Hierarchy'", in Dressier, W.U., and O.E. Pfeiffer (eds.), Phonologica 1976, 85-102. Drachman, G. (1988) "NP Structure in Greek", paper read at Annual Meeting of Austrian Linguists, Salzburg (to appear). Foley, J. (1977) Foundations of Theoretical Phonology, Cambridge University Press, Cambridge. Fukui, N. and M. Speas (1986) "Specifiers and Projections", MIT Working Papers in Linguistics 8, 128-72. Hockett, C.F. (1955) A Manual of Phonology. IJAL Memoir 11, reproduced by University of Chicago Press 1974. Hulst, H. van der and N. Smith, eds. (1982) The Structure of Phonological Representations. Parts 1 and 2, Foris, Dordrecht. Jespersen, O. (1904) Lehrbuch der Phonetik. Leipzig and Berlin Kahn, D. (1976) Syllable-based Generalisations in English Phonology, PhD Diss, MIT. 1980 Garland Publications, NY. Kaye, J., J. Lowenstamm and J-R. Vergnaud (1988) Constituent Structure and Government in Phonology, ms. Kurylowicz, J. (1948) "Contribution la theorie de la syllable", Biuletyn Polskiego towarzstwa jezykoznawczego 8, 80-114. Malikouti-Drachman, A. (1984) "Syllables in Modem Greek", in Dressler et al. (eds.), Phonologica 1984, 181-186. Michaels, D. "Natural and unnatural Phonology", paper given at the 6th. International Phonology Meeting, Krems, Austria, 1988. Saussure, F. de (1916) Cours de linguislique gέnerale, Payot, Lausanne and Paris. Selkirk, E. (1982) "The Syllable", in van der Hulst. H. and N. Smith (eds.). The Structure of Phonological Representations, Π, 337-83. Sievers, E. (1881) Grundz ge der Phonetik. Breitkopf and Hartel, Leipzig. Steriade, D. (1982) Greek Prosodies and the nature of Syllabification, PhD Diss., MIT, Cambridge (Mass.). Stevens, K.N. and SJ. Keyser (1989) "Primary features & their enhancement in consonants", Language 65, 1, 81-106. Yoshiba, H. (1983) Moric Phonology: Towards the Establishment of a New Phonological Unit, PhD Diss., University of Washington, Seattle.

NP-movement 'across' secondary objects Martin Everaert University of Utrecht - Royal Netherlands Academy of Arts and Sciences

1. THE PROBLEM1

Within the Government-Binding framework passive and raising verbs are generally taken as clear examples of NP-movement. Ever since Burzio's (1981, 1986) extensive study of unaccusative verbs, a new area of NP-movement phenomena has opened up. Several 'unaccusative' constructions have been claimed in the literature as involving NP-movement. They are presented in (1-3): (1)

Idiosyncratic unaccusatives a. The man arrived _

(2)

Inchoatives (also called ergatives/causatives) a. I broke the glass b. The glass broke _

(3)

Middles a. She painted the floor b. The floor paints _ nicely

It is argued that in all these cases the direct object of the verb is moved to the subject position. However, if we look at NP-movement of the secondary object, the parallel between passive and unaccusative constructions breaks down:2 (4)

a.

Passive He was given _ the book b. Idiosyncratic unaccusatives (psych verbs) *He pleases _ the book c. Middle *Children teach _ French easily d. Inchoative *The child bakes _ a cherry pie

126

M. Everaert

Several restrictions on the lexical specification of these ditransitive verbs have been suggested to explain why the secondary object of a ditransitive verb cannot be moved to subject position in constructions other than passive. Belletti & Rizzi (1988), for instance, take (4b) to violate Case Theory. The direct object the book is not assigned Case by the verb because it is unaccusative while the secondary object receives Case inherently. Moving the latter to subject position would lead to the direct object violating the Case filter. Furthermore, the secondary object he would be assigned nominative Case in subject position and (dative) Case inherently, resulting in incompatible Case-assignment Others have proposed that restrictions on predicate formation (Zubizarreta 1987, Afarli 1987), theta-assignment (Roberts 1987) or the principles mapping lexical structure onto syntactic structure (Grimshaw 1989) should be held responsible for the ungrammaticality of (4b-d). I will show that this difference between the passive construction and unaccusative constructions is not only manifested in unability of the secondary object to move. Even in cases of direct object movement, unaccusative constructions sometimes behave differently from passive. Kayne (1982) shows that there are cases of (direct) object / subject alternations where the presence or absence of a secondary object makes a difference. He observes that in English there are no inchoative pairs with ditransitive verbs (cf. 5), and that the same seems to hold for middle constructions (cf. 6): (5)

a. John showed the child the window b. *The window showed the child _

(6)

a. He sold linguists these books b. These books don't sell linguists _

Note, however, that the blocking effect of the secondary object appears to be absent in the case of idiosyncratic unaccusatives. Psych-verbs have been analyzed as involving movement of the direct object of a ditransitive verb (cf. Belletti & Rizzi 1988): (7)

The books pleased him _

One might be tempted to attribute the ungrammaticality of (5b) and (6b) to a violation of the Case filter by the secondary object, just as is the case in (8) where the secondary object remains Caseless because passive morphology has absorbed the structural Case: (8)

"The book was given him _

NP-movement 'across' secondary objects

127

However, based on observations in Dutch similar to those presented in (6,7), it can be shown that Case considerations may not be held exclusively responsible for the impermissible object/subject alternations in English.

2. THE FACTS

In Dutch ditransitive passives, only the direct object may be moved to subject position: (9)

a.

DC bezorgde hem het eten (I delivered him the food) b. Het eten werd hem _ bezorgd (door mij) The food was him delivered (by me) (The food was delivered to him (by me)) c. *Hij werd _ het eten bezorgd (door mij) He was the food delivered (by me)

It is generally assumed that the secondary object is inherently Case-marked (Den Besten 1981), and can therefore not be moved to subject position (cf. the discussion of (4b) above). Passivization of the verb absorbs structural Case assignment of the direct object but does not influence inherent Case assignment of the secondary object. A similar pattern of admissible NP-movements holds, for instance, in the case of idiosyncratic unaccusatives and raising verbs: (10)

a.

De teugels onglipten hem _ The reins slipped him (The reins slipped out of his hands) b. *Hij onglipte _ de teugels He slipped the reins

(11)

a.

Het lijkt hem [ _ te gaan sneeuwen] It seems him to go snow (It seems to him that it is going to snow) b. *Hij lijkt _ [ net te gaan sneeuwen] He seems it to go snow

Let us now return to the discussion of NP-movement in ditransitive verbs across a secondary object The examples (12-14) show how the presence of a secondary object influences the alleged object to subject NP-movement in unaccusative constructions:

128

M. Everaert

Idiosyncratic unaccusatives (12) a. Het boek bevalt _ b. Het boek bevalt hem _ (The book pleases (him)) Inchoatives (13) a. De croupier betaalt (hen) de winnende combinatie uit (The croupier pays (them) the winning combination out) b. De juiste combinatie betaalt _ uit (The right combination pays out) c. *De juiste combinatie betaalt hen _ uit The right combination pays them out Middles (14) a. De verkocht (the Studenten) het boek (I sold (the students) the book) b. Het boek verkocht _ goed (The book sold well) c. *Het boek verkocht de Studenten _ goed The book sold the students well As in English, NP-movement across a secondary object is blocked in the case of inchoatives and middles, while idiosyncratic unaccusatives allow for this type of movement. The pattern becomes somewhat more complicated if we take another type of unaccusatives into consideration, i.e. inherently reflexives. Burzio (1981, 1986) argued that inherently reflexive verbs in Italian are unaccusatives. In Everaert (1986), I claimed that the same holds for Dutch. In fact, I showed that Dutch has inherently reflexive idiosyncratic unaccusatives, inchoatives and middles. In all these cases a secondary object prohibits an object/subject alternation (cf. Everaert 1986:144): Idiosyncratic unaccusatives (15) a. Jan brandde zieh _ aan de kachel Jan burnt himself on the fire (Jan bums his hands on the fire) b. *, no attested examples of NP V zieh NPio _ Inchoatives (16) a. Jan uitte (mij) zijn verbazing (John expressed his surprise (to me)) b. Zijn verbazing uitte zieh _ in gestotter (His surprise expressed itself in stammering) c. *Zijn verbazing uitte zieh mij _ in gestotter His surprise expressed itself me in stammering

NF'-movement 'across' secondary objects

129

Middles (17) a.

DC laat (de kinderen) het verhaal voorlezen door de gouvernante I let (the children) the story read by the governess (I let the governess read the story (to the children)) b. Het verhaal laat zieh _ gemakkelijk voorlezen c. *Het verhaal laat zieh de kinderen _ gemakkelijk voorlezen The story lets itself (the children) easily read (The story is easily read to children)

Note that there are no cases of ditransitive idiosynractic unaccusatives which are inherently reflexive (cf. 15b) while there are such cases without a reflexive (cf. 12b). This pattern is also illustrated in (18-19): (18)

a.

Dat voordeeltje ontgaat hem _ (That windfall escapes him) b. *Hij ontgaat _ dat voordeeltje He escapes that windfall

(19)

a.

Hij laat zieh _ dat voordeeltje niet ontgaan Het let himself that windfall not escape (He will not let that windfall pass him by) b. *Dat voordeeltje laat zieh hem _ niet ontgaan That windfall lets itself him not escape

The verb ontgaan belongs to a class of unaccusative ditransitive verbs in Dutch which allows only the direct object to move to subject position in the 'bare' unaccusative variant while only the secondary object may move to subject position in the inherently reflexive variant (cf. Everaert 1986:59-60). Observe that it is not simply the presence of a second complement that blocks movement in these cases.3 As Kayne (1982) observed, the PP-variant of (6b) is grammatical: (20)

These books don't sell _ to linguists

For Dutch, the PP-variants are judged considerably better (for some even perfectly grammatical) in the case of middles: (21)

(?)Het boek verkoopt goed aan Studenten (The book sells well to students)

130 (22)

M. Everaert a.

b.

Deze kranten laten zieh eenvoudig _ aan abonnees versturen These newspapers let themselves simply to subscribers send (These newspapers are easily sent to subscribers) Het verhaal laat zieh gemakkelijk _ aan de kinderen voorlezen The story let itself easily to the children read (The story is easily read to the children)

The same seems to hold for French (Ruwet 1976:88):4 (23)

Ce genre de livre se vend _ surtout aux bonnes soeurs (This sort of book sells mainly to nuns)

For English, judgements apparently differ for middles. While Kayne takes (20) to be grammatical, Roberts gives the examples in (24) as ungrammatical:5 (24)

a. *Presents give _ to orphans most at Christmas b. *Books send _ to libraries best in boxes

As for inchoatives judgements vary for the PP-variant. In (25-26) the PP-variant seems to increase acceptability considerably, but that is not case in (27) which offers the PP-variants of (13c) and (16c): (25)

a. *De realiteit openbaarde zieh hem _ als een apocalypse The reality manifested itself him like an apocalypse b. De realiteit openbaarde zieh _ aan hem als een apocalypse (The reality manifested itself to him like an apocalypse)

(26)

a. *Die gedachte drong zieh mij _ op The thought forced itself me b. Die gedachte drong zieh _ aan mij op (The thought forced itself on me)

(27)

a. *De juiste combinaties betalen _ aan hen uit The right combinations pay to them out b.?*Zijn verbazing uitte zieh _ aan mij in gestotter His surprise expressed itself to me in stammering

For idiosyncratic inherently reflexive unaccusatives the judgements are clear-cut; the PP-variant is fully acceptable:

NP-movement 'across' secondary objects (28)

(29)

131

a. *Hij gaf zieh haar _ volkomen b. Hij gaf zieh _ volkomen aan haar He gave himself (to) her completely (He fully deferred to her) a. *Hij vertoonde zieh mij _ niet He showed himself me not b. Hij vertoonde zieh _ niet aan mij He showed himself not to me (He did not show himself to me)

3. QUESTIONS (AND ANSWERS?)

Two questions arise in connection with the Dutch data discussed above. First of all, one might ask why the presence of a secondary object would make a difference for NP-movement. Answering this question obviously raises the second question; why do passives and bare unaccusatives behave differently from inchoatives, middles and all inherently reflexives on this point? Below I will suggest some (very) tentative answers. As it stands, the theory does not seem to provide an obvious answer to the first question. In all cases of section 2, an NP is moved from a Case-less direct object position to a theta-less subject position, as is required by Case theory and Theta theory.

(30)

[s e [vp NP NP V ]] -» [SNP; [yp NP e; V ]] -θ +θ +Θ +C +C -C

Furthermore, subjacency- and ECP-requirements are satisfied in a structure like (31). Suppose, however, that we adopt Kayne's (1981) proposals about the structural properties of ditransitives. In his view the structure of ditransitives with a secondary object would be as in (31):

(31) t«, NP [vp [s NP NP] V] In this small clause structure, the secondary object has subject-like properties. Movement of the direct object would result in a violation of the Binding Theory because the NP-trace, subject to Binding Condition A, would be bound outside the first category containing a subject. Huybregts (1988), crucially making use of such a small clause analysis of ditransitives, suggests an intricate account of why the inherently reflexive constructions are excluded. Unfortunately this approach does not give us a clue as to why there should be a difference between passives and idiosyncratic unaccusatives on the one

132

M. Everaert

hand and inchoatives, middles and inherently reflexives on the other. It would be totally ad hoc to assume a small clause analysis for ditransives in the latter case but not in the former. In fact, in such an analysis one might even expect grammaticality judgements to be the opposite of what they are. For instance, if inchoatives, contrary to passives, do not involve NP-movement - as has frequently been argued - there are no NP-traces which would violate Binding Theory. Consequently one would expect the object/subject alternation to be grammatical in cases where no NP-movement is involved, i.e. in inchoatives, and ungrammatical in cases of NP-movement, i.e. in passives. Kayne (1982) has noted the dichotomy. He assumes that the presence of a secondary object influences the process of theta-assignment. The subject in the b-examples of (5) and (6) cannot be assigned the theta role that is assigned to the direct object in the respective a-examples. Apparently, however, the secondary object does not interfere with transmittance of the theta role as a result of NP-movement. I would like to pursue this line of reasoning and claim that the object/subject alternations discussed above look similar but are in fact fundamentally different, because they are the result of processes that apply at different levels of grammar. But, contrary to Kayne, I want to sketch an analysis which is not necessarily based on his assumptions about the small clause structure of ditransitive verbs, because it is not evident to me that his analysis holds for Dutch. Let us assume that, at least in Dutch, the object/subject alternations in passives and idiosyncratic unaccusatives are the result of syntactic NP-movement while the object/subject alternations in (inherently reflexive) inchoatives and middles are due to lexical processes.6 As noted above, there is no reason to assume that object/subject alternations which are the result of NP-movement are sensitive to the presence of a secondary object However, this secondary object does make a difference when these alternations are the result of lexical processes, as is the case with middles and inchoatives. These lexical processes both involve promotion of the (internal) theme argument and some sort of demotion of the (external) agent argument: (32)

Middle Formation/Inchoativization: (i) a. Erase the underlining on the external argument b. Delete the external argument (ii) Underline the theme argument

There seems to be no immediate reason why 'demotion' of an argument (cf. 32i) would be directly sensitive to the presence of a secondary object, but one can very well imagine that promotion (cf. 32ii) is. Suppose that we rephrase (32ii) as in (33,34):7

NP'-movement 'across' secondary objects

133

(33)

Underline the most prominent argument

(34)

Prominence of thematic roles is defined according to the following hierarchy:8 0lgmt > eeiperiencer > Qgou,souree / locition > θΛβηβ

It will be clear that (34) excludes all cases of object/subject alternations in middles and inchoatives when secondary objects are involved. In the case of ditransitive verbs the secondary object will be linked to a thematic role other than theme and, given the hierarchy as presented in (34), promotion of the direct object will always be promotion of a non-prominent internal argument in violation of (34). I will assume that promotion of the secondary object itself will be excluded by the Case theory (cf. Belletti & Rizzi 1988). But are we still able to account for Kayne's original obervation that object/subject alternations are allowed when the PP-variant of the distransitive verb is involved? Given the thematic hierarchy as formulated in (34) there seems to be no way to account for this dichotomy. In both cases the secondary object is assigned a more prominent thematic role than the direct object. However, Jackendoff (1989) proposes a refinement of theta theory which may make it possible to make a difference between the [NP NP]- and the [NP PP]-variant of ditransitive verbs. In Jackendoff s theory NP's can be assigned combinations of theta-roles. In such cases, one thematic role is indicated as dominant Jackendoff notes that, for most ditransitive verbs, there is difference in meaning between the [NP NP]-variant and the [NP PP]-variant The secondary object is assigned a (dominant) thematic role higher than Theme in the [NP NP]-variant while it is assigned a thematic role lower than theme in the [NP PP] variant. The relevant part of the thematic hierarchy that Jackendoff adopts is summarized in (35): (35)

Actor > Patient/Beneficiary > Theme > Source/Goal/Ref. Obj.

What about idiosyncratic inherently reflexives? In Everaert (1986) I argued for an unaccusative analysis for these constructions triggering NP-movement from object to subject position. In view of what is said above, however, these constructions should be analyzed as lexically derived. Everaert (to appear) argues for such a lexical approach, essentially following a type of analysis of inherent reflexivity as proposed in Chierchia (1989), Hellan (1988) and Reuland & Reinhan (1989). Inherently reflexive verbs are formed by a lexically specified rule coindexing an internal thematic role with the external thematic role.9 We can formulate this rule as in (36) under the assumption that this process of lexical binding is subject to the thematic constraint (34): (36)

Coindex the most prominent internal argument

134

M. Everaert

Huybregts (1988) observes that even for non-inherently reflexive constructions making use of the reflexive anaphor zieh, the presence of a secondary object leads to ungrammaticality:10 (37)

a. MariCj raadde Jan zichzelfj aan b. *Mariei raadde zichj Jan _ aan Marie recommended Jan herself (Marie recommended herself to John)

(38)

a. Hijj stelde mij zichzelfj voor b. *HiJi stelde zichj mij _ voor He introduced me himself (He introduced himself to me)

As is to be expected, the PP-variants are grammatical: (39)

a. b.

MariCj raadde zichj _ aan Jan aan (Marie recommended herself to Jan) Hiji stelde zichj _ aan mij voor (He introduced himself to me)

This provides evidence for the claim in Everaert (to appear) that there is no fundamental difference between the inherently reflexive constructions discussed above, and 'normal' reflexive constructions as in (39) which make use of the reflexive anaphor zieh. Both are the result of a lexical rule like (36).

4. CONCLUSION

I have shown that the presence of a secondary object blocks subject/object alternations in inchoatives, middles and inherently reflexive constructions in Dutch, while passives and idiosyncratic unaccusatives defy this restriction. To account for this dichotomy, I proposed that the object/ subject alternations of the former are lexically derived, while the object/ subject alternations of the latter are the result of the (syntactic) process of NP-movement. The lexical rules responsible for the object/subject alternations are subject to a thematic constraint blocking promotion when a secondary object is present However, arguments have been offered that all types of unaccusative constructions discussed above are derived by NP-movement (cf. Hoekstra 1984, Everaert 1986, among others). Clearly then, those arguments need to be revaluated. As has frequently been noted, not all arguments for an unaccusative analysis necessarily imply an analysis in terms of syntactic NP-movement.

NP-movement 'across' secondary objects

135

NOTES 1. I would like to thank Peter Coopmans for his valuable comments. Riny Huybregts and Ian Roberts contributed to this paper indirectly and unconsciously. Let me stress that I alone am responsible for errors and shortcomings. This research was made possible by a fellowship from the Royal Netherlands Academy of Arts and Sciences. 2. Cf. Roberts (1987:191) and Zubizarreta (1987:147) for examples like (4d) and Afarli (1987:2) for examples like (4b). 3. Ian Roberts, however, informed me (p.c.) that for English middle constructions this seems to be a fairly accurate description of the facts (cf. also Hoekstra & Roberts 1988). 4. Note that the indirect object cannot be cliticized because the construction would then violate restrictions on clitic combinations: (i) *Ce genre de livre se leur vend surtout This son of book sells them mainly 5. All native speakers I consulted found (i) considerably better than (ii), though not perfect: (i) ??Long stories don't tell very well to young children (ii) *Long stories don't tell young children very well 6. There seems to be little doubt in recent literature that inchoatives are lexically derived (Keyser & Roeper 1984, Hellan 1988, Pitz 1988, Roberts 1987, Zubizarreta 1987). For middles, both syntactic (Roberts 1987) and lexical (Pagan 1988, Zubizzareta 1987) approaches have been suggested. With respect to Dutch, inchoatives and middles have been argued to be syntactically derived (Hoekstra 1984). Everaert (1986) proposes a syntactic account for all inherently reflexive constructions, including inchoatives and middles. 7. In terms of Grimshaw's (1989) framework, middle formation would involve an operation on both the aspectual and thematic hierarchy of a verb. Similar constraints as to what is presented in the text can be formulated in her framework. 8. Cf. Grimshaw (1989) for this specific formulation of the thematic hierarchy. Only the rightmost part of the hierarchy is relevant for the present discussion. 9. I will assume that, in the case of inherently reflexive inchoatives and middles, (36) also works on the outcome of (32i). 10. The reflexive anaphor zicA is a clitic and, therefore, moved leftward. Huybregts (1988) shows that this movement cannot be the cause of the ungrammaticality of (37b) and (38b).

REFERENCES Besten, H. den (1981) "Government, Syntaktische Struktur und Kasus", in M. Kohrt and J. Lenerz (eds.), Sprache, Formen und Strukturen, Max Niemeyer Verlag, Tübingen. Burzio, L. (1981) Intransitive Verbs and Italian Auxiliaries, Doctoral dissertation MIT, Cambridge Massachusetts. Burzio, L. (1986) Italian Syntax, Reidel, Dordrecht Chierchia, G. (1989) "A Semantics for Unaccusatives and its Syntactic Consequences", ms. Cornell University. Everaert, M. (1986) The Syntax of Reflexivuation, Foris, Dordrecht. Everaert, M. (1989) "Contextual Determination of the Anaphor/Pronominal Distinction", to appear in J. Koster and E. Reuland (eds.), Lang Distance Anaphora, Cambridge University Press.

136

M. Everaert

Fagan, S. (1988) "The English Middle", Linguistic Inquiry 19. Grimshaw, J. (1989) "Argument Structure", ms. Brandeis University. Hellan, L. (1988) Anaphora in Norwegian and the Theory of Grammar, Foris, Dordrecht Hoekstra (1984) Transitivity, Foris, Dordrecht Hoekstra, T. and I. Roberts (1988) "Middles in Dutch and English", paper presented at the GLOW Lexicon Workshop, Budapest, 31-3-1988. Huybregts, M.A.C. (1988) "Chapter 9 'Verb Raising' and Chapter 10 'Clitica* of course book Grammatische Analyse", ms. Tilburg University. Jackendoff. R. (1989) "Semantic Structure", ms. Brandeis University. Kayne, R. (1981) "Unambiguous Paths", in R. May and J. Koster (eds.), Levels of Syntactic Representation, Foris, Dordrecht, 143-183. Kayne. R. (1982) "Comments on Chomsky's Chapter 'On the Representation of Form and Function*", in J. Mehler, E. Walker and M. Garret (eds.). Perspectives on Mental Representation, Lawrence Erlbaum Associates, Hillsdale, New Jersey, 453-456. Keyser, J. and T. Roeper (1984) "On the Middle and Ergative Construction in English". Linguistic Inquiry 15, 381-416. Pitz, A. (1988) "Middle Constructions in German", Working Papers in Linguistics, University of Trondheim. Reuland, E. and T. Reinhart (1989) "Anaphora and Logophors: A Theta-Theory Perspective", to appear in J. Koster and E. Reuland (eds.), Long Distance Anaphora, Cambridge University Press. Roberts, I. (1987) The Representation of Implicit and Dethematized Subjects, Foris, Dordrecht Zubizarreta, M.L. (1987) Levels of Representation in the Lexicon and in the Syntax, Foris, Dordrecht Afarli, T. (1987) "Lexical Structure and Norwegian Passive and Ergative Constructions", Working Papers in Scandinavian Syntax 29, Lund University.

Floating Quantifiers in Germanic1 Giuliana Giusti University of Venice

0. INTRODUCTION

German is more liberal in the distribution of floating quantifiers than other Germanic languages such as English and Danish. In all three languages it is possible to have a quantifier in distant position linked to a subject NP. But only in German is it possible to observe floating of a quantifier from the direct and from the indirect object position: (1)

(2) (3)

Ge. En. Da. Ge. En. Da. Ge. En. Da.

a. b. c. a. b. c. a. b. c.

die Schüler haben ohne Zweifel alle einen Preis erhalten the students have undoubtedly all received a prize Eleverne fik uden tvivl alle en praemie der Lehrer hat die Schüler (ohne Zweifel) alle gelobt *the teacher has praised the students (undoubtedly) all *Laereren roste eleveme (uden tvivl) alle der Lehrer gab den Schülern (ohne Zweifel) allen ein Buch *the teacher gave the students (undoubtedly) all a book *Laereren gav eleverne (uden tvivl) alle en bog

As Sportiche (1988) has shown, the phenomenon of floating should be derived from independent principles of the grammar. It seems therefore highly desirable that the difference between the three Germanic languages shown in (2)-(3) be also derived from some independent difference among them. In this paper it will be argued that the contrasts in (2)-(3) are to be reduced to the possibility in German vs. impossibility in English and Danish of scrambling (that is displacing) certain NPs (from their basic position to another position) inside the clause. In section 1, we will first review the basic features of Sportiche's (1988) proposal. In section 2, we will briefly see some effects of clause-bound movements in various Germanic languages. In section 3, we will turn to the contrast in (2)-(3) and we will explain other facts related to the occurrence of distant quantifiers in other Germanic languages, such as Dutch and some members of the Scandinavian family.

138

G. Giusti

1. A RECENT THEORY OF FLOATING QUANTIFIERS

Sportiche (1988) assumes an independently motivated structure like (4), for French and English, to derive the position of the quantifier in (5b). In (4), the subject of the sentence is generated in the NP* position and then moved (for reasons of case assignment) to the Spec of IP. He further assumes that a modifier is generated (left-)adjacent to the element which is modified by it:2

(4) ΝΡΛ

QNP

(5)

a. b.

tous les enfants ont dormi all the children have slept les enfants ont tous dormi the children have all slept

Since (5a-b) have roughly the same interpretation, they should be assigned the same structure, namely (4). In (5a) the entire NP* is moved to ΝΡΛ where Nominative case is assigned in French. In (5b) only the internal NP* is moved while tous remains in place. As the glosses show, this account holds for English as well. In assuming (4), Sportiche can account for the anaphoric behaviour of distant quantifiers, which is now a mere effect of NP-movement, and for their apparent adverbial position, which is due to the assumption that subjects, as well as adverbs, are both generated in positions adjoined to VP. In so doing, he dismisses two mysterious properties previously attributed to floating quantifiers: their downward movement into an adverbial position and their anaphoric behaviour, neither of which would be allowed in a GB framework. (4) implicitly assumes that the quantifier is base-generated left-adjacent to NP* in English. This should be assumed for German as well. In other words it would be theoretically and empirically incorrect to assume that die Sch ler alle in (2a) and den Sch lern allen in (3a) are base-generated sequences. From a theoretical point of view, it would be difficult to admit that the definite quantifier alle 'all', but not other determiners such as the article or an indefinite quantifier, such as for example viele 'many', can appear on either side of an NP.

Floating Quantifiers in Germanic (6)

a. b.

(7)

139

der Lehrer lobt alle die Schüler the teacher praises all the students der Lehrer lobt die Schüler alle

a.

der Lehrer lobt viele Schüler the teacher praises many students b. *der Lehrer lobt Schüler viele

From an empirical point of view it is possible to demonstrate that die Schüler alle, in (6b), contrary to alle die Schüler in (6a), does not form a constituent. It is well-known that the sentence-initial position in a V/2 clause in Germanic can be taken to be a test for constituency. (8)

a.

alle die Schüler haben ein Buch bekommen all the students have received a book b. *die Schüler alle haben ein Buch bekommen c. die Schüler, alle haben ein Buch bekommen

The relevant contrast is illustrated in (8a-b). In (8c) die Schüler is left-dislocated, while alle functions as a resumptive element in sentence-initial position. They do not form a constituent, as is shown by the obligatory presence of a pause between the two. The structure in (4) also accounts for the fact that a distant quantifier can only refer to a subject in French and English, since this is the only case in these two languages in which an NP receives -role and case in two different positions. In the following sections, we will see why this restriction does not hold for German.

2. SOME PROPERTIES OF "SCRAMBLING" IN GERMANIC

Although German is often considered to have a quite liberal word order, there are some interesting constraints, as the seminal work of Lenerz (1977) shows. Indirect object - direct object can be taken to be the unmarked word order. The reverse is possible provided that the direct object is definite: (9)

a. b.

ich I ich I

habe have habe have

einem/dem Studenten ein/das Buch gegeben a/the student-DAT a/the book-ACC given (*ein)/das Buch einem/dem Studenten gegeben (*a)/the book-ACC a/the student-DAT given

Let us take the existence of an unmarked word order to be evidence for a structured VP. This excludes a priori the possibility of generating the NP and

140

G. Giusti

the quantifier related to it in a discontinuous fashion.3 In (9b), the definite object must have moved from its base position. The existence of such a process has been discussed from several viewpoints (cf. Webelhuth (1985), Webelhuth and den Besten (1987)). Some consensus has been reached on the assumption that, although it is clause bounded, it must be an -bar movement, since it licenses parasitic gaps (cf. Bennis and Hoekstra (1984)). Since the theory of floating quantifiers adopted here predicts that whenever an NP is moved, it can leave a quantifier in its base position, its interaction with scrambling straightforwardly explains (2-3a), which will have the structures in (lOa-b), respectively: (10)

a.

der Lehrer hat [die Schüler^ (gestern) [yp [alle tj gelobt] the teacher has the students yesterday all praised b. der Lehrer gab [den Schülern]; (gestern) [w [allen tj ein Buch] the teacher gave to-the students yesterday all a book

Among the other Germanic languages, Dutch is considered to have some scrambling effects as well (cf. de Haan (1979), Hoekstra (1984), Bennis and Hoekstra (1984)). (11)

a. b.

ik heb I have ik heb I have

gisteren een kind/de kinderen gezien yesterday a child/the children seen de kinderen/*een kind gisteren gezien the children/a child yesterday seen

The sentences in (11) are completely parallel to those in (9). For our purposes we can say that Dutch behaves on a par with German. This will allow us to check our hypothesis in section 3. Let us now look at the Scandinavian family. Norwegian appears to behave like Danish, in that no scrambling is possible: (12)

No. a. lareren gav elevene b0kene teacher-the gave students-the books-the b. "lareren gav b0kene elevene

According to Vikner (1989), and references cited there, all Scandinavian languages display some kind of NP-movement inside the clause, which he refers to as "object shift", to distinguish it from scrambling.4 The basic difference between scrambling and object shift that concerns us here is that in German and Dutch all definite NPs can be moved, while in Scandinavian only NPs with morphological case can. This amounts to saying that only in Icelandic full NPs can be displaced, since in this language morphological case is present in NPs,

Floating Quantifiers in Germanic

141

while in Mainland Scandinavian only pronouns are marked for case, and only they can move: (13)

Ic. a. hvers vegna lasu studentarnir ekki greinina b. hvers vegna lasu studentarnir greinina ekki why read the students the article not

(14)

Da. a. hvorfor laeste studenteme ikke artiklen/*den b. hvorfor laeste studenteme den/*artiklen ikke why read the students them/the article not

If we take the negation ekki to be VP initial, in (13a) we have the object NP greinina in base position, while in (13b) it has been shifted over the negation, that is outside of VP. In Danish, always following Vikner (1989), this is possible (and indeed obligatory) only if the object is a pronoun, as shown in (14). It is not clear which deep principle allows and differentiates these two minimally different types of clause-internal movement in Germanic. But we have seen in this section that it is possible to isolate their effects. This will enable us to check their interaction with Sportiche's theory of floating presented in section 1.

3. ON THE EFFECTS OF VARIOUS TYPES OF MOVEMENT ON FLOATING QUANTIFIERS

The nature of the movement is actually irrelevant to our research. We expect that any time an NP is moved, a quantifier can be left in place. This expectation is borne out. Let us start with the simple cases in (15a-b), which are completely parallel to (2a)-(3a), respectively, except for the presence of the VPinitial adverbial wahrscheinlich: (15)

a. b.

der the der the

Lehrer hat teacher has Lehrer hat teacher has

die Schüler wahrscheinlich alle gelobt the students probably all praised den Schülern wahrscheinlich allen ein Buch gegeben the students probably all a book given

The presence of wahrscheinlich 'probably' permits us to say that the NP has been moved outside of the VP, while the quantifier has remained in place. Of course, a definite NP can be scrambled over a quantifier apparently yielding an inappropriate position for ic

142

G. Giusti

(16)

a.

die the b. *die the

Kinder haben children have Kinder haben children have

gestern das Buch alle bekommen yesterday the book all received gestern ein Buch alle bekommen yesterday a book all received

In (16a) the object das Buch precedes the quantifier alle linked to the subject position. This could be taken as a counterexample to the claim that the quantifier never moves to a lower position. But if we consider (16b), we see that if the direct object is indefinite, it cannot precede the quantifier linked to the subject, showing that it is the definite object in (16a) that is scrambled over the quantifier, and not the quantifier that is lowered. Both floating and scrambling have been noticed to be clause-bounded phenomena. In our proposal this is not accidental: (17)

a.

weil because b. *weil because

niemand sagt, daß ich die Bücher nobody says that I the books die Bücher niemand sagt, daß ich the books nobody says that I

(alle) all (alle) all

gestohlen habe have stolen gestohlen habe have stolen

When clause boundaries are weakened as in clause union cases, scrambling and floating both become possible again: (18)

a.

weil die berühmte Friedlandia in Wien alle diese Lieder because the famous Friedlandia in Vienna all these songs singen versuchte sing tried b. weil diese Lieder die berühmte Friedlandia in Wien alle because these songs the famous Friedlandia in Vienna all singen versuchte sing tried

zu to

zu to

In (18) die berühmte Friedlandia 'the famous F.' is the subject of the matrix predicate versuchen 'try' and can be crossed in (18b) by the object diese Lieder 'these songs' of the embedded predicate singen 'sing' because versuchen is a clause union trigger.5 Scrambling from the embedded clause is allowed and floating goes along with it. Since topicalization and wh-movement are not clause bounded, the floating quantifier connected to a topicalized NP or a wh-phrase can appear in the embedded clause:

Floating Quantifiers in Germanic (19)

a. b. c.

143

die Studenten sagt Maria hat der Lehrer alle gelobt the students says Mary has the teacher all praised Studenten sagt Maria hat der Lehrer viele gelobt students says Mary has the teacher many praised welche Studenten sagt Maria hat der Lehrer alle gelobt? which students says Mary has the teacher all praised?

The perfect acceptability of (19) confirms that clause-boundedness is not an intrinsic feature of floating but an effect of scrambling. If the movement that gives rise to floating is not clause bounded, then neither is floating. Since scrambling applies only to definite NPs, we expect no floating of indefinite quantifiers, as we have already seen in (7b). This cannot be trivially due to the fact that indefinite quantifiers take an intermediate projection of N that cannot be moved, since bare NPs can be topicalized leaving a Q in place: (20)

a. b.

der Lehrer hat viele Studenten gelobt the teacher has many students praised Studenten hat der Lehrer viele gelobt

As van Riemsdijk (1988) has shown, the topicalized element in (20b) behaves as a full NP.6 This shows that the reason why a quantifier in distant position cannot be linked to an indefinite NP does not reside in its intrinsic structure but in the simple fact that an indefinite NP cannot be scrambled. Since we have seen in (19) that topicalization and wh-movement can leave a quantifier in place also in the case of a definite NP, there is no reason to assume that (20b) is different from (21): (21)

a. b.

die Studenten hat der the students has the welche Studenten hat which students has

Lehrer alle gelobt teacher all praised der Lehrer alle gelobt? the teacher all praised

This is not the place to present a theory of the internal structure of DP/QP in German. However, it is interesting to observe that at least extraction and floating quantifiers do not hint at any structural difference between definite and indefinite quantified nominals. All the properties of German displayed in (16)-(21) above have been first noticed by Link (1974) and subsequent studies by Vater (1980). Both studies are in the EST framework and present several shortcomings that I will not discuss here.7 Since it was noticed that Dutch behaves in a parallel fashion to German with respect to scrambling, all things being equal, we expect to find a parallel distribution of floating quantifiers.

144

G. Giusti

(22)

a. b.

de leraar the teacher de leraar the teacher

heeft has heeft has

de kinderen the children de kinderen the children

gisteren yesterday gisteren yesterday

allen all allen all

geloofd praised een boek gegeven a book given

This is actually the case. The quantifier alien can refer to the direct object in (22a) and to the indirect object in (22b). In both cases, the NP de kinderen is clearly scrambled over the adverbial gisteren 'yesterday'.8 Turning now to Scandinavian, we expect that object shift, but not scrambling, interacts with the possibility of leaving a quantifier in place. (12b), repeated here as (23a), is as bad as (23b): (23)

No. a. *lareren gav b0kene elevene b. *lareren gav elevene b0kene alle the teacher gave the students the books all

(23b) shows that b#kene cannot scramble over its quantifier, even remaining in a lower position with respect to the indirect object This shows that the impossibility of scrambling in Norwegian is not simply due to some minimality effect that can arise when the indirect object intervenes between the moved NP and its trace. On the contrary, object shift gives rise to floating, as argued in Vikner (1989): (24)

Da. a. *hvorfor laeste studenteme artiklen uden tvivl alle why read the students the article undoubtedly all b. hvorfor laste studenteme den uden tvivl alle why read the students it undoubtedly all

(25)

Ic. a. hvers vegna lasu studentamir ekki allir greinina b. hvers vegna lasu studentamir greinina ekki allir why read the students the article not all

The unacceptability of (24a) is due to the fact that scrambling of artiklen is impossible in Danish. But as soon as the NP is pronominalized, it is moved for independent reasons and the quantifier can remain in place. In (25), greinina in Icelandic, having morphological case, can object shift and it can leave a quantifier in place, as in (25b). In conclusion, we have seen that the occurrence of distant quantifiers in Germanic can be fully derived from the interaction of independent principles. In all languages observed here a subject NP can move inside the clause from its base position inside VP to the Spec IP position where it receives case, leaving a quantifier in place. In languages with scrambling, such as German and Dutch,

Floating Quantifiers in Germanic

145

a distant quantifier can be linked to any NP scrambled from its base position. In Scandinavian, object-shift can be the cause of the distant position of a quantifier. No additional rule is therefore needed to account for the apparently quite erratic distribution of distant quantifiers in Germanic.

NOTES 1. I thank Guglielmo Cinque and Giuseppe Longobardi for long discussion and support. I am also grateful to Liliane Haegeman, Tarald Taraldsen, and Sten Vikner for having provided the sentences in Dutch, Norwegian, and Danish, respectively, and for having discussed them with me. However, none of the scholars mentioned above is responsible for my mistakes. 2. Sportiche (1988) does not take a stand on the status of VP°, or on the status of the node dominating Q NP*, if there is one. He proposes that the generation of the Q undergoes the Adjunct Projection Principle (his (7)): If X "modifies" some (semantic) type Y, and X and projected as adjacent to b, or to the head of b.

are syntactically realized as a and b, a is

But I think it would be interesting to suppose in the spirit of the DP hypothesis that Q heads its own maximal projection and selects NP*. A definite Q such äs alle 'all' will select a definite NP, an indefinite Q such as viele 'many* will select an indefinite NP. We will touch upon this in section 2, although this is not the place to develop this idea. 3. In Giusti (1989), I discuss the configurationality issue in German with respect to floating quantifiers in more detail. 4. Vikner (1989) argues that object shift is different from scrambling in that it does not license parasitic gaps. The difference would ultimately be in the type of the target position: An A-bar position would be the target in German/Dutch, while an A position would be the target in Scandinavian. This differentiation, however, does not concern us here. What is relevant to our research is that both A- and A-bar-movement can leave a floating quantifier. 5. Cf. Thiersch (1978), McKay (1985) for the effects of clause union. 6. It is extracted by wh-movement, since it is sensitive to islands and to the same dialectal variation we find in extractions in German; and it can even have an (indefinite) determiner. Cf. also Bayer (1986) for a different treatment of this phenomenon. 7. For a discussion, cf. Giusti (1988). 8. It is possible to substitute the invariable allemaal for alien. Since allemaal looks more like an adverbial than a nominal element, I suggest that it is adjoined to the NP in Sportiche's fashion (cf. note 2). However, more research is needed to settle this point.

REFERENCES Bayer, J. (1986) "The syntax of scalar particles and so called 'floating quantifiers'", ms. Max Planck Institut Bennis, H. and T. Hoekstra (1984) "Gaps and Parasitic Gaps", Linguistic Review 4, 29-87. Evers, A. (1975) The transformational cycle in Dutch and German, PhD Diss. University of Utrecht. Giusti, G. (1988) "The syntax of floating alles in German", to be published in the proceedings of the Vm Workshop in Scandinavian Syntax, W. Abraham and E. Reuland (eds.), Linguistik Aktuell, Narr, Tübingen.

146

G. Giusti

Giusti G. (1989) "Floating Quantifiers, scrambling and configurationality", ms. University of Venice. Haan, G. de (1979) Conditions on Rules, Foris, Dordrecht. Hoekstra, T. (1984) Transitivity, Foris, Dordrecht Lenerz, J. (1977) Zur Abfolge nominaler Satzglieder im Deutschen (Studien zur Deutschen Grammatik S), Narr, T bingen. Link, G. (1974) "Quantoren Floating im Deutschen", in P. Kiefer and D. Perlmutter (eds.), Syntax und Generative Grammatik vol. 2, 105-127, Athenaion, Frankfurt. McKay, T. (1985) Infinitival Complements in German, Cambridge University Press. Riemsdijk, H. van (1988) "Movement and Regeneration", in P. Beninca' (ed.), Dialect Variation and the Theory of Grammar, Foris, Dordrecht Sportiche, D. (1988) "A theory of floating quantifiers and its corollaries for constituent structure", Linguistic Inquiry 19, 425-449. Thiersch, C. (1978) Topics in German Syntax, PhD Diss. ΜΓΤ. Vater, H. (1980) "Quantifier floating in German", in J. van der Auwera (ed.). The Semantics of Determiners, pp. 232-249, Groom Helm, London. Vikner, S. (1989) "Object Shift and Double Objects in Danish", talk delivered at the DC Workshop in Comparative Germanic Syntax, to appear in the Proceedings, Working Papers in Scandinavian Syntax, University of Lund. Webelhuth, G. (1985) "German is configurational", Linguistic Review 4, 203-246. Webelhuth, G. and H. den Besten (1987) "Remnant topicalization". paper delivered at the XI GLOW Conference.

Generative grammar in Italy Giorgio Graffi University of Pavia

I hope I won't be accused of chauvinism (an attitude which I consider to be among the most detestable in the world) if I assert that the prestige of Italian generative linguistics is remarkable, not only on a European level, but on a world-wide level as well; as they say, "the facts speak for themselves": the current president of GLOW is Italian, the European editors of "Linguistic Inquiry" are Italian, many books by Italian authors are present in the major international collections of literature on generative grammar (those published by Reidel and by Foris for example); there is practically no international conference or seminar on generative grammar which doesn't have Italian scholars among the speakers, and so on. To these exterior elements (which are in any event significant and which testify to the large contribution already made and currently being made by Italian scholars to the research programme of generative linguistics) are to be added more substantial aspects such as the fact that many theoretical developments of recent years are by now linked to Italian generative linguists. But at this point I can leave off the commemorative overtone (even if justified, as I've said) and begin to bring out a paradox: many of the most famous Italian generative linguists do not work in Italian universities, but in foreign ones, and one certainly cannot say that generative grammar has a strong presence in the Italian universities. From among approximately 60 "full professors" (professori ordinari - the highest level of the Italian academic system) of linguistics, not more than four or five conduct (in all or in part) research in the theoretical framework of generative grammar. To these are to be added three "associate professors" and very few "researchers". As one can see, it's not a very consoling picture. Nevertheless, generative research in Italy, rather than languishing, is in constant expansion. The number of students who wish to specialize in generative linguistics is increasing, as well as the number of "final theses" (lesi di loured), and just in recent months, the first doctoral theses either have been completed or are in the process of completion. The paradox of Italian generative linguistics thus seems to lie in the enormous disproportion between both the remarkable results therein achieved and the vivaciousness revealed on the one hand, and its rather limited institutional presence on the other. This paradox is however only apparent, once one takes into consideration the characteristics of linguistic studies in Italy and, consequently, the cultural context in which Italian generative grammar had to be

148

G. Graffi

developed. And it is exactly this subject (i.e., the beginnings of generative grammar in Italy) that I want to address here, because I consider it a curious example of cultural history while at the same time instructive in a certain way. In the often slightly forced schematization which is usually imposed on the history of linguistics (derived in large part from a rapid and not always wellassimilated reading of Kühn) generative grammar is presented as the theory which arose from the reaction against structuralism; the latter was fought and then defeated while generative grammar subsequently imposed itself as the dominant paradigm. This synthetic representation of the facts may be true for countries such as the United States or Holland, but it certainly can't be true for Italy; if in fact, between the 1950s and the 1960s, the rising generative linguistics of the United States fought with all its might against the structuralist paradigm then dominant, a conflict of this sort could never have occurred in Italy for the simple reason that in Italy during this time no kind of structural linguistics existed, or any kind of theoretical linguistics for that matter. The analysis of the causes of this cultural situation would probably require not merely an article, but an entire book, and I can therefore but limit myself to describe the main points. The type of linguistics that was taught in Italy at that time (almost always under the non-translatable name of "glottologia") belonged exclusively to Indo-European historical-comparative linguistics. Within such a linguistic domain, the Romance languages (as well as their respective dialects, and above all the Italian ones) were obviously studied with particular attention, and according to the specializations of the various teachers, other groups of Indo-European languages: from German to Celtic to Armenian; in most universities, great importance was also placed upon the study of Sanskrit A common characteristic of almost all of these teachings was the "allergy" to theoretical discussions, which were generally considered to be "philosophical stuff'. Paradoxically (but not very) the school of philosophy which had dominated the Italian scene for over half a century (the "neo-idealism" of Benedetto Croce) negated the possibility of an autonomous scientific linguistics and identified linguistics with aesthetics (see B. Croce, Estetica come scienza dell'espressione e linguistica generale, Bari, Laterza, 1909). In reality, Croce's philosophy was in perfect harmony with the kind of linguistics then dominant, given their common disinterest for a scientific theory of language. And it is completely unjustified to assert that such a theoretical disinterest was the necessary consequence of the fact that linguistic research was restricted to the historical-comparative sphere. The theoretical concerns of one Hermann Paul (just to cite one linguist who is often considered to be a symbol of the historical-comparative method) were for the most part foreign to the Italian linguists of the epoch (with a few partial exceptions such as Giacomo Devoto, Luigi Heilmann, Giovanni Nencioni and Antonino Pagliaro, for example).

Generative grammar in Italy

149

If therefore, in other countries, as just explained, structuralism presented an obstacle which generative grammar was to overcome, in Italy in the 1960s, the at first timid and later tumultuous manifestation of an interest in structural linguistics undoubtedly prepared the way for generative grammar, even if in a non-definitive manner, because it aroused interest in linguistic theory. In reality, one cannot speak of a genuine structuralist period in Italy because so little work was carried out, and the work which was done in the sphere of this paradigm was almost exclusively phonological, according to the framework of the Prague School (the most important work belonging to Giulio Lepschy). It was an age of authentic frenetic activity as far as regards all that was (more or less correctly) labelled "structuralist". This frenetic activity materialized in, among other things, a series of translations in Italian which were carried out without any coordinated plan. First to be published was the translation of Cows by Saussure in 1967, followed by that of Prolegomena by Hjelmslev in 1968, Language by Sapir in 1969, Grundzüge der Phänologie by Trubeckoj in 1971, and only in 1974 that of Language by Bloomfield. At the same time between 1969-71, three volumes of Saggi linguistici ("Linguistic Essays") by Chomsky came out. These included, among other things, the translations of Aspects of the Theory of Syntax. Cartesian Linguistics and Language and Mind. The interest in generative grammar had been bom, even if in a confused way and little appreciated by the academic "establishment" which labelled (disfavourably) all those who were involved in linguistic theory as "structural linguists". However, some exceptions had already begun to appear such as L. Heilmann, who in 1966 founded the journal "Lingua e Stile" (originally intended to be entitled "Structura"). Moreover, with the increase in the number of translations of Chomsky's essays, the first Italian works in generative grammar came to light. In effect, research in generative grammar began to be conducted already in the 1960s in Italy, even if in a largely "clandestine" manner. Such research most likely had its start in the Faculty of Magistero ("Education") at the University of Rome, where, at the time, no institutionalized teaching of linguistics was offered, either under the traditional name of "Glottologia" or under the name (which was by this time considered almost revolutionary) of "Linguistica generale", but where courses (for reasons largely due to chance) were held in which themes of generative grammar were dealt with, at least in part In the early 1960s in fact, an English Program for Italy was active. It was funded by the Ford Foundation and the Fulbright Commisssion and provided for some "free" courses of English grammar (i.e., courses offered outside the curriculum required to obtain a university degree). These courses were held by S. Sacks of Berkeley who, among other things, had his students read Syntactic Structures. More or less in the same epoch and as part of the same programme, P. Roberts was also present in Rome. Among the students that participated in these courses was Annarita Puglielli who, fascinated with generative grammar, after having

150

G. Graffl

received her degree in English language and literature, went abroad to perfect her preparation in generative grammar. Finally, she received her Ph. D. at Cornell University (under the guidance of F. Agard) with a thesis on The Predicate Phrase in Italian (which was subsequently published under the title Strutture sintattiche del predicate in italiano, Bari, Adriatica, 1970). In the second half of the 1960s, the generative approach to the study of language had also begun to interest a group of psychologists (gathered around Domenico Parisi) who were active at the Institute of Psychology of the CNR ("Consiglio Nazionale delle Ricerche") in Rome. These scholars associated themselves with Annarita Puglielli upon her return from the United States in the autumn of 1968. Thus was bom the first Italian research group in generative grammar. One year later, at the proposal of Puglielli and Parisi, the "Societa di Linguistica Italiana" ("Italian Society of Linguistics", founded only in 1967; Robert B. Lees took part in its first meeting, presenting a paper entitled On very deep grammatical structure.) organized a conference in Rome that was to be dedicated solely to "Italian transformational grammar". The participants at this conference included the group from Rome, some foreign scholars (Herwig Krenn, Pieter Seuren), some Italian linguists working abroad (Mario Alinei, Mario Saltarelli, Paolo Valesio) and some Italian scholars who were approaching generative grammar on a self-taught basis (Gualtiero Calboli, Adriano Colombo, Edgardo T. Saronne and others). There thus seemed to be dawning in Italy a period of remarkable development of studies in generative grammar, but unfortunately the reality was otherwise. First of all, as was obvious, the major part of "official" Italian linguistics remained firmly bedded in a strongly antitheoretical position; the Italian Society of Linguistics (where almost all linguists seeking to break with the old cultural patterns had originally gathered) assumed a progressively decisive antigenerativist attitude, favouring instead research projects of a sociolinguistic nature. At last the "battle" between "generative semantics" and "extended standard theory" had arrived in Italy, provoking various divisions within the rising generativist movement; Parisi and his collaborators (Francesco Antinucci, Cristiano Castelfranchi and others) enthusiastically went the way of generative semantics and developed their own original model (see D. Parisi - F. Antinucci, Elementi di grammatica, Torino, Boringhieri, 1973). The possibility of development of the standard theory in Italy seemed truly very slight at the end of the 1960s. Nevertheless, the activity of divulgation of the theoretical linguistics of the twentieth century which, as we have seen, had so thoroughly characterized the Italian linguistic scene in the 1960s, was destined to produce fruitful results. As regards this, it is impossible not to mention the activities of Giulio Lepschy, both for his publications and on a level of personal contacts. His work La linguistica strutturale (Torino, Einaudi, 1966; English version A Survey of Structural Linguistics, London, Faber, 1970), which put Italian linguists in

Generative grammar in Italy

151

contact with subjects hitherto unknown, dedicated a chapter to "transformational grammar" (which once again demonstrates that, in the Italian situation, structuralism and generative transformational grammar were not seen as antithetical). Furthermore, Lepschy, Venetian by birth, kept in contact with some of the scholars from the University of Padova and from the "Scuola Normale" of Pisa (where he had studied) even after having relocated in England around the midsixties. Among these scholars, two must be mentioned: Lorenzo Renzi of the University of Padova and Alfredo Stussi of the Scuola Normale of Pisa, both Romance philologists (as can be seen, being alien to institutional linguistics was a constant feature) who were nevertheless interested in theoretical and descriptive linguistics. To Renzi and Stussi (who were both little more than 30 years old at the close of the 1960s) the development of Italian generative linguistics undoubtedly owes much. Their policy was to encourage the study of the most important works in generative grammar and above all, that of urging some of their more promising students to relocate abroad in those centres where a generative linguist could truly learn to work, as well as come into contact with the international generative linguistic community and thereby get away from the provincial narrow-mindedness which unfortunately was then characteristic of the Italian culture in general as well as of the Italian linguistics in particular. Thus, a seminar was held on Syntactic Structures in Padova in 1969, whose participants included, along with Renzi, various other scholars (among these Guglielmo Cinque, who left Padova for Berkeley in 1972). At the Scuola Normale of Pisa, during Alfredo Stussi's seminar of the academic year 1973-74 (which was officially dedicated to the "history of Italian language and grammar" and in which also Leonardo Savoia, who subsequently became one of the very few Italian linguists dealing with generative phonology, took part), the standard theory, the extended standard theory and generative semantics were presented and discussed. Leaving Pisa, Luigi Rizzi, Claudio Bracco, Adriana Belletti, Giuseppe Longobardi and Rita Manzini went, successively, to Vincennes for some years beginning in 1974 in order to attend the courses of Richard Kayne and Nicolas Ruwet. Also in the first half of the 1970s, an important role in diffusing the theory and practice of generative research was played by the International Summer School for Computational Linguistics, organized in Pisa by Antonio Zampolli (Director of the Institute of Computational Linguistics of CNR) and funded by IBM. This school, in reality, was not limited solely to computational linguistics, but also offered courses in generative grammar held by some of the most respected scholars in the field. For example, in 1974 E. Bach and B. Partee were to be counted among these scholars. One of the first tangible results of this diffusion of generative grammar among Italian researchers, was the foundation of the "Rivista di grammatica generative" in Padova in 1975, by Francesco Antinucci and Guglielmo Cinque. Some of the people who first had conducted research in generative grammar in

152

G. Graffi

Italy tried for the first time to coordinate and give rise to a common initiative (which, in spite of economic difficulties, proceeded and presently proceeds). Simultaneously with the founding of the journal, it was decided to initiate a series of "informal meetings on generative grammar", which continue regularly to this day. More or less in the same period, GLOW made its appearance on the European (and extra-European) linguistic scene; in 1976 in Amsterdam, Luigi Rizzi participated in its first meeting. Through these meetings, Amsterdam also became known in Italy as an important centre of research in generative grammar, and therefore, in the autumn of 1978 Sergio Scalise (having been a student of L. Heilmann and L. Rosiello, and to whom generative linguistics had always been of interest) and myself (having been a student of philosophy at the Scuola Normale of Pisa and therefore being very familiar with the group of young generative linguists headed by Alfredo Stussi) decided to go to Amsterdam to the Institute of General Linguistics, seeking to learn generative grammar "for real". There we found Henk van Riemsdijk and many other esteemed generative linguists (including another Italian, Marina Nespor, who, by chance, had studied philosophy rather than linguistics in Italy and had subsequently obtained her Ph. D. from Chapel Hill) and undoubtedly... we succeeded in learning something. With the occurrence of the GLOW conference in 1979 in Pisa, Italian generative linguistics made its definitive entrance into the international linguistic scene. The academic careers of the best known Italian generative linguists, as well as the fundamental lines of their scientific production, are well-known, and for that matter, beside the scope of this brief and substantially "memoristic" article, which is intended to bring to light the substantially "extra-institutional" character (where "institution" is taken to mean Italian linguistic academy) of the development of generative studies in Italy. This should explain the paradox to which I alluded in the beginning, that is to say: the disproportion between the actual scientific prestige and the vivaciousness of Italian research in generative grammar on the one hand, and the relatively low number of academic positions occupied by generative linguists in Italy on the other.

Particles, Prepositions, and Verbs J. Gu6ron University of Paris Vlll

1. Emonds (1976) classifies the particle (pit) as an "intransitive preposition" (P). Indeed the two categories have much in common. 1.1. Both are non-productive.1 1.2. Like a PP, a pit can denote the end point of a directional path, as in (la-c). (1)

a. b. c.

We put the food out/on the table. Off/to the garden he ran. Out/to the devil with him.

1.3. Pits "strand" like Ps in passive Ss. (2)

a. Johiii was talked about q. b. The informationi was looked up ^.

1.4. Both Prts and Ps can be construed compositionally or idiomatically, creating ambiguity, as in the Ss of (3). (3)

a. b.

She stood up the young man. He decided on the boat.

2. In other ways, particles are more like verbs than like prepositions, however. 2.1. A prt can raise to Ml in Dutch, like V and unlike P. 2.2. In Dutch, a P governs its object to the right, like a Noun, but a prt governs its object to the left, like a Verb (cf. van Riemsdijk 1978). In English, the object of a P appears only to its right, but the object of a prt appears either to its right or its left. While (3a) above is grammatical with the alternative order NP Prt in (4a), (3b) is ungrammatical with the order NP P in (4b).

154 (4)

J. Guaron a. She stood the young man up. b. *We decided the boat on.

The ordering alternation of Pit Phrases in English recalls that of causative VP complements in French or Norwegian (cf. Taraldsen 1983, Afarli 1984, Gu6ron 1985). (5) (6)

a. b. a. b.

Nous laissons (partir Marie). (We let leave Mary) Nous laissons (Marie partir). Vi lot (skuret rive). (We let demolish the shack) Vi lot (rive skuret).

2.3. At first glance, pits appear unable to govern a sc, like Ps. (7) (8)

a. b. a. b.

*We talked about (John a fool), *We sent to (John the tools). "They made out (John a fool). *We handed down (John the tools).

Closer examination shows, however, that prts are like the unaccusative verbs in (11) and (12). They can govern a sc, but cannot, in English at least, assign case to the sc subject Raising the sc subject derives an acceptable S with a Prt governor, although not with a P governor (cf. (9) vs (10)).2·3 (9) (10) (11) (12)

a. ?Johnj was made out (tj a fool.) b. Johiij was handed down (tj the tools). a. "John; was talked about (t; a fool), b. *Johnj was sent to (tj the tools). Johnj became (tj sick) Mary; has (tj brothers) (cf. Gueron 1986b)

2.4. While a PP can be extracted by e.g. topicalization or clefting, a PrtP, like a VP complement, must remain contiguous to its d-structure governor throughout the derivation. (13) (14) (15)

a. (On the table) we put the books, b. It was (to Mary) that we spoke. a. *(Up the information) we looked. b. *It was (the package up) that he brought. a. He let (the child cry) b. *It was (the child cry) that he let.

Particles, prepositions, and verbs

155

2.5. Nouns take PP but not PrtP or VP complements. (16)

a. A call to your brother would be nice. b. *A call up your brother would be nice. c. *A let leave would be nice.

2.6. (Non-idiomatic) verb-particle constructions have a resultative construal (cf. Hoekstra 1988). So do causative constructions. In all the Ss of (17), the embedded sc denotes a situation caused by the activity described in the matrix. (17)

a. He pushed (p„p Bill out). b. He hammered (^ the nail flat). c. He made (w Mary cry).

2.7. A particle phrase projects an unaccusative small clause in t%3 d-structure, like a causative complement with unaccusative verb (cf. Burzio 1986; Gueion and Hoekstra 1988). Kayne (1985) proposes that a particle projects a sc of the form NP Prt (a predicative sc) in d-structure. The order Prt NP is derived by adjunction of the sc subject to its predicate. Adopting Kayne's proposal that particles are predicates, we argue in Gueion (1986a) that the pit projects an unaccusative sc, of the form Prt NP. The NP Prt order is derived by NP raising, as with other unaccusative predicates.4 Kayne shows that a Prt P of the form NP Prt is subject to the "left-branch" constraint and failure to genitivize, like an NP or AP sc. (18)

(19)

a. b. c. a. b. c.

We find (^ Mary intelligent). *WhOj do you find (the mother of e; intelligent)? *The finding of (Mary intelligent). We put (p^ the book back). *Whatj did you put (a book about ^ back). *The putting of (a book back).

Guoron (1986a) shows, however, that while these contraints still hold of an NP, AP, or PP sc with the inverted order Pred NP, they do not apply to a PrtP with the order Pred NP. (20)

(21)

a. We found happy (many people). b. *WhOi did you find happy (the mother of t;)? c. *We finding happy of (many people). a. We wrote up a paper. b. WhOj did you write up a paper about tj? c. The writing up of the paper took a month.

156 /. Guaron The contrast between (20) and (21) suggests that the sc subject is adjoined to its predicate in (20), leaving the underlying violation-inducing predicative structure intact, but is governed by its predicate in (21), within a subjectless sc free of violations. Like PrtPs, causative complements are subject to sc constraints with the order NP V but not with the order V NP. Thus English (22b) and French (23b) with NP V order violate the left-branch constraint, but French (24b) with V NP order does not. (22) (23)

(24)

a. We let (John's children leave). b. *Who did you let (the children of ^ leave)? a. Je laisse (le fils du chef partir). (I let the child of the chief leave) b. *J'eni laisse (le fils tj partir). a. Je laisse (partir le fils du chef). b. J'eiij laisse (partir le fils Q.

Nor do the left-branch and genitivization constraints apply to those (rare) AP, NP, or PP scs which allow the order Pred NP. (25) (26) (27)

a. *The turning of the leaves red. b. ?The turning red of the leaves. a. *Who did they elect the brother of president? b. ?Who did they elect president the brother of? a. *The turning of Lot's wife to stone. b. ?The turning to stone of Lot's wife.

These data suggest that only scs with subjects are liable to the left branch and genitivization constraints, and that non-finite verbs in French and Norwegian, and particles in English (and Norwegian) can project a subjectless sc, while Ns, ADJs, and Ps normally cannot. This partitioning of syntactic categories with respect to sc structure will be accounted for below. 3. In spite of shared properties, particles are not verbs. 3.1. English has unaccusative PrtPs but no unaccusative causative complements, as shown in (28). (28)

a. We put (p^ down the dog). b. *We made (yp whimper the dog).

Particles, prepositions, and verbs

157

3.2. A prt cannot govern a pronoun, unlike P or V.5 (29)

a. We wrote about it. b. We wrote it up. c. *We wrote up it

3.3. Prts "strand" in middle Ss like (30) and ergative Ss like (31), but Ps and Vs don't, as shown in (32) and (33). (30) (31) (32) (33)

a. b. a. b. a. b. a. b.

This articlej writes up tj easily. These expenseSj write off tj as tax deductions. The treCj lit up tj. The plane; took off tj. "John; speaks to tj easily. (Hoekstra 1984) *That shirtj lets wash tj easily. t%3 *Philadelphia arrived in. *That shirt; made dry tj in the sun.

3.4. Prts have aspectual import the sequence imperfective verb + (non-idiomatic) Prt is construed as perfective (or iterative), as its incompatibility with the imperfect adverbial in (34) shows. Neither PPs nor causative complements have this aspectual property. (34)

(35) (36)

a. b. c. a. b. a. b.

She wrote (*up the paper) for hours She turned (*up the bed cover) for hours. They threw (*out) the ball for hours. She ran (towards the city) for hours, He worked (on the boat) for hours. She made John sing for hours, She let John fall for hours.

4. The unaccusative sc as Tense Phrase. 4.1. Nouns, Adjectives and Verbs. Gueron and Hoekstra (1988) claim that the construal of a non-finite XP as nominal or verbal, ultimately as argument or predicate, depends on its syntactic environment (37)

XP is nominal if case-marked and verbal if T(ense)-marked. XP is T-marked if governed by Tense or a T-marked verb.

158 J. Gutron In a simple S, finite tense T-marks the main verb (and intervening auxiliary verbs) under government T-marking creates a tense chain, of the form 7* - (Aux)*k - Vk (where k is a T-mark), as illustrated in (38). The VP governed by Tense defines a tense domain (in italics in (38)). (38)

a. John loves Mary. b. John 1* (wk loves* Mary)

We proposed that certain verbs can T-mark a clausal complement, creating an extended tense chain with two tense nodes, each licensing a tense domain. The complement XP in an extended chain is construed as a VP in LF, by (37), even if it is a CP in syntax. Thus the inflected infinitive in Portuguese is nominal in (39a), where it is case-marked by lamentar (cf. Raposo 1987), but t%3 verbal in the Aux-to-Comp structure (39b), where it is T-marked by pensar. (39)

a.

EU T, lamento (JP/NP os deputados T2 terem trabalhado pouco) CASE b. EU T! penso (CP/VP T2 + teren^ dp os deputados tj trabalhado pouco) T-mark (I regret/think the deputies to have worked little)

In what follows we situate (37) with respect to the theory of Phrase Structure. Adapting Abney (1987) and Fukui and Speas (1986), we propose (40). (40)

Every maximal lexical projection must be governed by a functional head; every functional head must govern a maximal lexical projection.

If (37) is correct, and if every lexical XP is governed by a functional category, then functional categories must themselves be either nominal or verbal. NP is governed by D (Abney 1987). Agreement phenomena in many languages show that D contains the same φ-Fs of number and gender as Nouns, so we assume D to be a nominal functional category. Lexical verbs are governed either by Tense or an auxiliary. This suggests that Tense and auxiliaries are verbal functional categories. If every lexical projection is governed by a functional category, then a sc is not NP, VP, etc. as in Stowell (1983), but more like AGR-P or T(ense)-P as in Pollock (1989), with AGR now construed as a nominal category, like D, and T as a verbal category. We propose that a functional category identifies the lexical category it governs by realizing (or binding) its inherent functional Fs, T or agr. A lexical item thus selects the F node(s) which govern it by means of its inherent functional features, just as it selects its argument structure by its inherent semantic or theta-role Fs. We define the notion "lexical item" in (41).

Particles, prepositions, and verbs (41)

159

X is lexical iff it contains (i) a semantic (θ-role) F and (ii) either agr or a Tense F.

N and A have φ-Fs (= agr), so NP and AP select an AGR node. If NP/AP functions as a predicate, then agr, ΝΙΑ. is construed as an anaphor, and a sc subject is necessary to bind it.6·7·8 (42)

We made Mary happy. TP

made

Maryj

Finite verbs are lexically + T, + agr, and select both TP and AGRP (cf. Pollock 1989, Belletti 1988, Chomsky 1989). A non-finite verb may lack agr, in which case it directly selects a subjectless TP, as in (43).

160 /. Gulron (43)

Nous faisons partir les enfants (We make leave the children) TP

les enfants

A predicative sc is thus projected by a lexical item containing an agr morpheme, such as N or A, while an unaccusative sc is projected by a lexical item lacking an agr morpheme, such as a non-finite V in French. 4.2. Particles and Prepositions. Ps and Pits are functional categories, lacking agr and T Fs. Prts differ from Ps, however, in having an inherent LOCATIVE semantic content.9 An ungoverned pit cannot be construed as lexical, for it lacks the +T or agr F needed to license the functional node which identifies its syntactic class. Nor is it functional: its inherent semantic content presumably prevents it from identifying another XP. We propose that a particle governed by a verb is T-marked in s-structure. T-marking of the pit has the same effect in LF as an inherent T feature would have had in D-structure: it licenses a Tense node which identifies the particle as a verbal category and as a predicate. In the absence of agr, the pit projects an unaccusative sc, as in (44), like the unaccusative verb in (43).

161

Particles, prepositions, and verbs (44)

We brought up the package. TP

the package

T-marking is a form of L-maiking (cf. Chomsky 1986): it removes the barrier status of the PrtP, allowing it to function as a link of the matrix T-chain, and licensing pooling of the semantic content of the V and Pit. If the governing verb is an activity verb denoting movement, the particle adds DIRECTIONAL semantic content to its inherent LOCATIVE content.10 Levin and Rappaport (1989) argue that DIRECTIONAL semantic content characterizes unaccusative verbs. We propose that the acquisition of DIRECTIONAL semantic content allows the prt to assign the same theta role -THEME- as an unaccusative verb. The particle thus acquires the syntactic and semantic status of a verb in the course of the derivation, as a function of T-marking.11 The inherent LOCATIVE content of the particle accounts for the telic aspect of the derived S: a verb plus prt combination is t%3 semantically equivalent to a telic unaccusative verb such as arrive, discussed in Levin and Rappaport (1989). 4.3. Aspect as temporal anaphora. T-marking of an embedded sc licenses a second T node and a second tense domain, and gives rise to a resultative interpretation. Finite tense is analogous to a NAME: it refers to a specific moment in the discourse world, just as the name John refers to a specific person. Non-finite tense is anaphoric; it cannot refer to a time unless bound by finite tense.12 Bound anaphoric tense denotes "another occurrence" of the "same" moment that the finite tense refers to, just as the anaphor himself denotes "another occurrence" of the "same" individual that its antecedent, say, John, refers to. Now, although it is reasonable to refer to "another occcurence" of the "same" individual, since individuals are (conceived of as) constant through time,

162 /. Gufron it is unreasonable to refer to "another occurrence" of the same moment, for a moment is not constant through time. We propose that the logical contradiction implied by the notion anaphoric tense is resolved by aspect. The introduction of a second instance of the same time is compatible with a construal of the reference time of S as an interval or sequence of points rather than as a single point The "second occurrence" of the "same" time then refers to a medial or final point of the interval which begins with the matrix tense.13 When the matrix tense domain contains an activity verb, the situation denoted in the embedded tense domain is construed, plausibly simply by the causative fallacy characteristic of human cognition - post hoc ergo propter hoc - as the result of the action denoted by the matrix verb.14 If an embedded sc predicate is imperfective, it can T-mark the XP it governs, licensing still another tense domain. If the predicate is perfective, it denotes the final point of a temporal interval and cannot initiate a new tense domain. The imperfectivity of causative verbs thus accounts for iteration of causative TPs shown in (45), while the inherent LOG content of the particle determines perfective aspect and accounts for the failure of PrtP iteration in (46). (45) (46)

a. She made Mary let Joe make Sue speak to Bill. b. Elle a laisse" faire faire cette robe par une couturiere ä Marie. (She let make make this dress by a dressmaker to Mary) a. *She took Bill out the book down, b. *She put the book back out down.

5. Reanalysis. Under reanalyis, the transformation of the prt from functional to lexical node is incomplete. T-marking still licenses an embedded T-node which identifies the PrtP as verbal, but the embedded TP does not define a semantically independent tense domain. If the matrix verb does not denote a movement, the particle does not receive DIRECTIONAL semantic content It doesn't license a theta role, and the embedded TP is not construed as a result clause. The matrix and embedded Tenses are interpreted as literally the "same" moment in time, and the embedded TP has no aspectual content The inherent LOG content of the PrtP is preserved by its telic contribution to the denotation of the joint verb under idiomatic semantic symbiosis of verb and particle. Many interpretive properties distinguish reanalyzed from non-reanalyzed V - V or V - Prt constructions in addition to idiomaticity. For example, a causative construction is incompatible with stative aspect in the embedded VP, as shown in (47a). Under reanalysis, there is no embedded aspect so this constraint is not pertinent as shown in (47b).

Particles, prepositions, and verbs (47)

163

a. *Je fais (savoir 1'anglais ä Pierre). (I make know English to Pierre) b. Je fais (savoir ä Pierre que je ne viendrai pas). (I make+know (=tell) to Pierre that..)

In a S with an embedded TP, both Tt and T2 normally initiate a tense domain. A stative adjunct, which takes scope over a tense domain, can then refer ambiguously to the matrix or sc subject, as in (48). (48)

a. b. c.

We brought (the baby up) tired. We hammered (the nail flat) hot. We made (John imbibe) already drunk.

Under reanalysis, since there is no independent embedded tense domain, there is no ambiguity. In the Ss of (49), the adjunct can refer to the sc subject, but only under a non-idiomatic reading. (49)

a. b. c.

She brought (the child up) tired. She stood (the fellow up) half-dressed. John let (her down) unrepentent.

A Pit P is construed as resultative only if the matrix verb denotes an activity. In English, a middle construction must denote a change of state rather than an activity (cf. Fellbaum and Zribi-Hertz 1989). Consequently, only idiomatic verbpit constructions appear in English middles. (50) (51)

a. "The package brings up easily, b. *The cat lets out easily. a. These forms fill out easily, b. This article writes up easily.

The interpretative differences between compositionally construed and noncompositionally construed causative and/or particle constructions are unaccounted for by the numerous linguists who have proposed that all such Ss contain a single verb, derived by reanalysis in syntax or by incorporation in LF.15 We have argued that all such structures are, on the contrary bi-clausal, in that they contain two Tense Phrases. The differences between the compositionally construed and the idiomatically construed structures is a function of the aspectual role of the embedded TP. Aspect is in turn based on the interaction of the lexical content of the matrix verb and that of the embedded predicate within the domain defined by a Tense chain.

164

/. Gulron

NOTES 1. Unlike a P, a Prt can combine with a veib to form a new compound item, as in "laid-back" or "off-putting". This supports the text hypothesis that the prt has inherent semantic content 2. Kayne (1985) first proposed that a particle can take a sc argument For Kayne, the sc in (8a) or (8b) is the d-structure subject of the particle rather than, as we claim, its object. 3. In Norwegian, a prt can assign case to the sc subject it governs, as in (our analysis of) (i) or (ii), from Afarli (1984). (i) Vi kledde pa (gc han frakken) (We put on (him the coat)) (ii) Vi sparka ut (§£ Kunden av huset) (We chased out (the dog from the house)) 4. This derivation is also proposed in Taraldsen (1983), who claims, however, that the Prt is a P. 5. A Prt which does govern a pronoun functions like a P. It cannot take a subject While (i) below is ambiguous between a directional P construal and an idiomatic prt construal, (ii) has only the P construal, and (iii) and (iv) only the prt construal. (i) We looked up the periscope. (ii) We looked up it. (iii) We looked the periscope up. (iv) We looked it up. 6. We will say nothing about PP predicates here. 7. In (42), AGRP is selected "from below" by the AP predicate, while the TP which governs it is selected "from above" by make. Auxiliary verbs, like causative make, are obligatory T-markers and select a TP (cf. Guiron and Hoekstra 1988). The sc AGRP in (42) is equivalent to the AGRO of Chomsky (1989) or the ip of Fassi Fehri (1987); the sc TP is equivalent to the Aspect Phrase of Carstens and Kinyalolo (1989). 8. AGRO is usually governed by T, as in (42), but this is not always the case. While TP is verbal and resists case, AGRO is nominal and must have case, for θ-role visibility. So a sc which appears in a position to which case is obligatorily assigned, such as subject of a tensed S or object of a P, must be an AGRP, not a TP. The contrasts below suggest that a sc with Stative adjectival predicate can function as an AGRP. They thus provide evidence for the existence of a functional sc node with nominal status. (i) *(John intelligent) is a surprise. (ii) (John sick) is a sad sight (iii) With John sick, we will lose the game. (iv) *With John intelligent, we will win the game. 9. While Ps are perhaps not devoid of all semantic potential, their construal depends on their syntactic environment The Ps at or on introduce a locative phrase when the NP complement denotes a place and a temporal phrase if the complement denotes a time (cf. (i) - (iv)). But a (nonidiomatic) particle always has a LOCATIVE construal. (i) he works at the farm, (ii) he came at three, (iii) he sat on the sofa

Particles, prepositions, and verbs

165

(iv) the trains comes on the hour. 10. For the F LOG, cf. Emonds (1989). Hoekstra and Mulder (1989) also discuss the change from LOCATIVE to DIRECTIONAL· 11. The theta-role assigned by a particle is licensed in s-structure, where the prt is semantical])1 like a verb, but not in d-structure, where it is a P-like functional node. This violates the theta-criterion. We suggest that a revision of the criterion may be necessary. Note that the situation of the Prt is not unique. P also assigns no theta-role in isolation but only when governed by a verb, as in (i) vs (ii) or (iii) vs (iv). (i) To John, life is a dream (ii) We spoke to John, (iii) John read On Binding. (iv) John worked on the article. 12. On temporal anaphora, cf. Partee (1973). (1984). 13. The hypothesis that aspect is an anaphor in the domain of temporal reference accounts for the structural constraint between morphological tense and aspect morphemes: the first must c-command the second. The opposite structural relation violates principle C of the binding theory. 14. Hoekstra and Mulder (1989) make essentially the same point. Note that in (39b), the matrix tense domain denotes a mental activity while the embedded CP denotes a proposition which results from that activity. The same causal cognitive mecanism triggered by anaphoric tense seems to function here as in the Ss of (17). Resultative interpretation fails only when Tj and T2 are pragmatically ("We saw John leave") or under reanalysis (cf. 15. Arguments against reanalysis in syntax reanalysis (incorporation) is proposed in Baker

construed as simultaneous, as with perception verbs section 4). are provided in Gueron and Hoeksta op.cit. LF (1988).

REFERENCES Abney, S. (1987) The English Noun Phrase in its Sentential Aspect, Diss. ΜΓΤ. Afarli, T. (1984) "Norwegian Verb Particle Constructions as Causative Constructions", Working Papers in Scandinavian Syntax 11. Baker, M. (1988) Incorporation: A Theory of Grammatical Function Changing, U. of Chicago Press. Belletti, A. (1988) "Generalized Verb Movement" Lecture, GLOW Budapest. Burzio, L· (1986) Italian Syntax: A Government-Binding Approach, Reidel, Dordrecht Carstens and Kinyalolo (1989) "On IP structure: Tense, Aspect and Agreement", ms. Cornell U. and UCLA. Chomsky, N. (1986) Barriers, MIT Press. Chomsky, N. (1989) "Some Notes on Economy of Derivation and Representation", in I. Laka and A. Mahajan (eds.), Functional Heads and Clause Structure: ΜΓΓ Working Papers 10, 43-74. Emonds, J. (1976) A Transformational Approach to English Syntax, Academic Press, NY. Emonds, J. (1989) "The Syntactic Basis of Semantics and the Non-Existence of Theta-Grids", ms. U. of Washington. Fassi Fehri, A. (1987) "Generalised IP Structure, Case and VS Word Order", ms. U. of Rabat Fellbaum, C. and A. Zribi-Hertz (1989) "La construction moyenne en francais et en anglais", Recherches Linguistiques 18, U. Paris Vffl, 19-57. Fukui, N. and M. Speas (1986) "Specifiers and Projections", MIT Working Papers 8.

166 /. Gudron Gu6ron, J. (1985) "Opacity in small clauses", GLOW lecture, Brussels. Gu6ron, J. (1986a) "Clause Union and the Verb-Particle Construction in English", talk presented at NELS 16, ms. U. Paris Vm. Gu6ron, J. (1986b) "Le verbe avoir", in P. Coopmans, Y. Bordelois and B. Dotson Smith (eds.), Formal Parameters of Generative Grammar U: Going Romance, Foris, Dordecht, 83-106. Gueron, J. (1989) "Subject, Tense, and Indefinite NPs", to appear in NELS 19. Gueron, J. and T. Hoekstra (1988) "T-Chains and the Constituent Structure of Auxiliaries", Constituent Structure: Papers from the 1987 Glow Conference, Foris, Dordrecht. Hoekstra, T. (1984) Transitivity, Foris, Dordrecht. Hoekstra, T. (1988) "Small Clause Results", Lingua 74, 101-139. Hoekstra T. and R. Mulder (1989) "Unergatives as Copular Verbs: Locational and Existential Predication", ms. U. of Leiden. Kayne, R. (1985) "Principles of Paricle Constructions", in J. Gueron, H.-G. Obenauer and J.-Y. Pollock (eds.). Grammatical Representation, Foris, Dordrecht. Levin, B. and M. Rappaport (1989) "An Approach to Unaccusative Mismatches", to appear in NELS 19. Partee, B. (1973) "Some Structural Analogies between Tenses and Pronouns in English", The Journal of Philosophy 70, 601-609. Partee, B. (1984) "Nominal and Temporal Anaphora", Linguistics and Philosophy 7, 243-286. 83 Pollock, J.-Y. (1989) "Verb Movement, UG, and the Structure of IP", Linguistic Inquiry 20. Raposo, E. (1987) "Case Theory and Infl-to-Comp: the Inflected Infinitive in European Portuguese", Linguistic Inquiry 18, 85-109. Riemsdijk, H. van (1978) Λ Case Study in Syntactic Markedness, Peter de Ridder Press, Lisse. Stowell, T. (1983) "Subjects across Categories", The Linguistic Review 2, 285-312. Taraldsen, T. (1983) Parametric Variation in Phrase Structure, Diss. U. of Troms0.

Non-overt subjects in diary contexts Liliane Haegeman University of Geneva

\. NON-OVERT PRONOMINAL SUBJECTS

Languages vary with respect to whether they allow non-overt pronominal subjects, the so-called pro-drop parameter. Italian allows null subject pronouns (la) represented as pro (Ib), while English and French do not. (1)

(2)

a. b.

Italian: (Ιο) parlo italiano (I) speak Italian [pro] parlo italiano

a. b.

English: *(I) speak Italian French: *(Je) parle Italien

Interestingly, the underlined finite verbs in the English, and French diary extracts in (3) lack an overt subject In (3a), the verbs will be interpreted as having a first person subject, in (3b) a contextually salient third person subject is involved.1 (3)

a.

A very sensible day yesterday. Saw no one. Took the bus to Southwark Bridge... Saw a flight of steps down to the river. I climbed down. [The diary of Virginia Woolf, vol 5, 1936-41. pp. 203]

b.

Apres-midi discuter, puis agreable. M'accompagne au Afternoon to discuss, afterwards agreeable. Me accompanies to Mercure, puis la gare... s'est αοηηέ souvent the Mercure, then to the station... herself is given often l'illusion de ramour P... en peasant moi,... the illusion of love to P... thinking of me,...

168

L. Haegeman Revient ä l'affairc Alb... Me demande si ... Returns to the affaire Alb. Me asks if ... [Paul Leautaud, Le Fteau. Journal particulier 1917-1930, pp. 69-70]

In the diary register non-overt subjects alternate with overt subjects. From now on I refer to null subjects in diary contexts as NSDCs. The label diary for the register involved is an approximation: informal letters and postcards cf. 'Wish you were here' also allow null subjects. My assumption is that special registers may represent non-core grammatical properties and that non-core grammars may differ from core grammars with respect to certain parametric settings (cf. Haegeman 1987 and 1988). At first glance, the constraints on the occurrence of pro in the grammar of English, and French, might seem to be relaxed in the diary register. The data suggest, though, that the grammar of the diary register in these non pro-drop languages does NOT involve a different setting of the pro-drop parameter, but rather that, unlike the core grammar, the peripheral or non-core grammar allows for a zero discourse operator in [Spec.CP] which binds an empty category in the [NP.IP] position. The constraints on the distribution of NSDCs follow from this analysis. In the last section of the paper I will extend the data to other languages and raise some questions for future study.

2. THE NON-OVERT SUBJECT IN DIARY CONTEXTS

The NSDC interacts with other sentence constituents exactly like an overt subject. I assume, therefore, that the NSDC is syntactically represented as an empty category ([ec]). In the constructed examples in (4)2 the NSDC, [ec], binds a reflexive (4a), controls a PRO subject in an non-finite clause (4a), and is a subject for a predicative adjunct (4b). Similar examples can be constructed for French (cf. the reflexive se in (3b) above). (4)

a. b.

[eCj] Hurt myself while PRO; trying to cut the roses [ecJ Came home quite exhausted;

The GB-framework identifies four empty categories: the base-generated pronominals PRO and pro, and the non-pronominal traces: NP-trace and vvA-trace. Before addressing the problem of which type of empty category NSDCs represent, I shall identify some constraints on their distribution. Although NSDCs appear regularly in finite root clauses, they cannot occur in subordinate clauses: (5) illustrates adjunct clauses, (6) complement clauses and (7) relative clauses:

Non-overt subjects in diary contexts (5)

a. b.

ENG. *[ec] Saw noone after [ec] had left the party FR. *[ec] Voyait Jean quand [ec] quittait la fete3

(6)

a. b.

ENG. *[ec] Don't know when [ec] will see her again FR. *[ec] Ne sail pas quand [ec] me reverra not knows when me will see again

(7)

a. b.

ENG. *[ec] Don't find the letter which [ec] need FR. *[ec] Trouve pas la lettre dont [ec] a besoin

169

Moreover, not all root clauses allow NSDCs: (8)

a. b.

ENG. *When will [ec] be able to meet him? FR. *Quand me reverra [ec]?

(9)

a. b.

ENG. *This book, [ec] did not approve of FR. *Ce livre, [ec] Faime

(8) and (9) suggest that NSDCs are disallowed when [Spec.CP] is filled by an operator, be this a wA-operator or a TOPIC operator. The yes-no questions (10) can be aligned with (8) if we assume that there is a non-overt wA-operator in [Spec.CP]. (10)

a. b.

ENG. *Will [ec] be able to meet him? FR. *Me reverra [ec]?

3. THE IDENTIFICATION OF THE NSDC

It is unlikely that the NSDCs are PRO. NSDCs occupy [NPJP] in finite clauses, a governed position. Unlike PRO, NSDCs alternate with overt subjects (cf. (3a)). NP-trace can also be discounted. NP-traces are typically caseless and the NSDC occurs in a position marked NOMINATIVE. Moreover, if NSDCs were NP-traces, they would be subject to principle A of the Binding Theory. If the GC of the NSDC is the clause immediately dominating it, then there will be no Α-binder available. Extending the GC to a higher clause (in a stipulatory fashion) is irrelevant since NSDCs do not occur in subordinate clauses. This leaves us with the two [-anaphoric] zero elements to consider: pro and wA-trace. pro might seem the natural choice. In (la), for instance, the understood subject of the underlined verbs corresponds to T. If NSDCs are identified as pro one might say that in specific registers of non-pro-drop languages the requirements for the licensing of pro are somewhat relaxed and that, due

170 L. Haegeman perhaps to functional reasons, even a relatively poor inflection suffices to identify the subject However, this hypothesis is inadequate: the constraints on NSDCs ((5)-(10)) do not apply to Italian pro-subjects: (11)

a.

pro non ho visto nessuno quando pro sono partita not have seen noone when am left Ί did not see anyone when I left' b. pro mi domando quando pro ti rivedr me ask when you will see again Ί wonder when I will see you again' c. pro non trovo la lettera di cui pro ho bisogno not find the letter of which have need Ί don't find the letter which I need' d. (Quando) pro potr rivederti? (when) can see again you '(When) can I see you again?' e. Questo libro, pro non lo voglio This book, not it want 'This book, I don't want'

This leaves wA-trace. If there is a trace in the [NP,IP]-position of the relevant clauses in (3) and (4) above, it will have to be A'-bound by a null topic operator in [Spec, CP], to be represented here as TOP, along the lines of Huang (1984). The representation of (12a) would be roughly (12b). (12)

a. b.

Saw noone [cp TOPj [a, tj saw noone]]

The range of TOP in (12b) is fixed by the 'discourse topic', a salient discourse element, (cf. Reinhart 1981). In (3a), for instance, the discourse topic is T, the diary writer; in (3b) it is 'she', the writer's mistress, the theme of the 'journal particulier'.4 I assume that the interpretation of TOP is only partially fixed in the syntax and is completed at the level of the discourse, where it is assigned a range. For the mechanisms determining the contextual interpretation of null operators we could turn to theories of utterance interpretation such as Relevance Theory (Sperber and Wilson 1986, and Haegeman (1989), for further data).5 If the NSDC is a wA-trace bound by a zero topic operator, the observed constraints on its distribution follow. Work on zero-discourse operators in Chinese (Huang 1984), Portuguese (Raposo 1986), Spanish (Campos 1986), Mauritian Creole (Syea 1988), and colloquial German (Ross 1982, cited in Huang 1984) etc. shows that the zero topic operator in [Spec, CP] (see Authier 1989) is incompatible with a second operator in [Spec, CP]. This predicts the ungrammatically of (8)-(10). The ungrammaticality of NSDCs in subordinate

Non-overt subjects in diary contexts

171

clauses ((5)-(7)) also follows. The a-examples, for instance, will be ruled out as ECP violations analogously to (13):6 (13)

a. b. c.

*WhOj did you see noone after tj left the party *WhOj do you wonder when tj will see you again *Whoi can you not find the letter which ^ needs

4. EXTENDING THE DATA: DUTCH. GERMAN, AND DANISH

An interesting observation is that NSDCs seem to be a general property of non pro-drop languages. Dutch, and German, for instance, which do not freely allow null subjects, allow a non overt subject in the diary register. (14)

a.

Dutch Vind, in Gryll Grange, ook een erg aardige uitdrukking die find, in Gryll Grange, another nice expression which ik niet kende en die ik ook nergens vinden kan I not knew and which I also nowhere find can [C. Buddingh', Een mooie tijd om later te worden, p. 99]

b.

German Freut mich und macht mich traurig,... Makes me glad me and makes me sad... [Kafka, Tagebücher 1910-1923, 264] Beschloss, ihn zu kaufen, änderte diesen Beschluss .... Decided it to buy, changed this decision ..., [Kafka, Reisetagebücher, 436].

Again, the non overt subject is restricted to main clauses, and to clauses which do not have material in [Spec, CP]. (15H20) are the equivalents of (5H10). (15)

a. b.

DU. *[ec] Zag Jan toen [ec] net feest verliet GE. *[ec] Sah Hans wann [ec] das Fest verliess

(16)

a. b.

DU. *[ec] Weet niet wanneer [ec] haar terug zal zien GE. *[ec] Weiss nicht wann [ec] sie wiedersehen werde

(17)

a. b.

DU. *[ec] Vind de brief niet die [ec] nodig heb GE. *[ec] Finde den Brief nicht den [ec] brauche

172

L. Haegeman

(18)

a. .b.

DU. *Wanneer zal [ec] hem kunnen zien? GE. *Wann werde [ec] ihn sehen k nnen?

(19)

a. b.

DU. *Dit boek, vind [ec] niet goed GE. *Dieses Buch, finde [ec] nicht gut

(20)

a. b.

DU. *Zal [ec] hem kunnen zien? GE. *Werde [ec] ihn sehen k nnen?

An analysis along the lines sketched for English and French seems possible: TOP is moved to [Spec.CP] and binds a trace in the [ΝΡ,ΙΡ] position. The finite verb in Dutch and in German occupies the C-position, both of these being Verb Second languages. Note, though, that in informal German (and also in Dutch), a null operator7 in [Spec, CP] can bind both subject traces and object traces, the so-called pronoun zap (Ross 1982) phenomenon. With respect to the NSDCs in V2 languages the question arises whether they can be reduced to an instantiation of pronoun zap and are perhaps a function of the V2 phenomenon. (21)

a. b.

GE. [OPJ habe [w ich schon t^ gesehen] have I already seen, GE. [OPJ habe [^ ^ es schon gesehen]

Danish, a V2 language lacking pronoun zap, also has NSDCs, though; this is illustrated in (22). This suggests that the phenomenon described here is one that relates to a specialized register and may be differentiated from pronoun zap. (22)

Danish8 Jeh har nu ikke studiet over me, ... Har blot ingen tid at scrive i I have now not study-the over me... have just no time to write in Ί am not one for studying, ...just don't have any time to write in' [Jorgen-Frantz Jacobsen, in W. Heinesen ed. (1958) Dei Dyrebare Liv. Copenhagen: Gyldendal].

5. QUESTIONS FOR FURTHER RESEARCH

The analysis of NSDCs in English and French as traces bound by a zero topic operator raises the immediate question why this posibility should be restricted to subjects, i.e., why the null operator cannot also bind object traces. I hope that further research into the nature of null operators and register linked syntactic properties may throw light on this issue.

Non-overt subjects in diary contexts

173

NOTES * I wish to thank the various colleagues who have discussed this paper with me: Margaret Clayton, Corinne Grange, Michel Jeanneret, Ray Lurie, Greg Polletta, Genoveva Puskas, Ian Roberts, Bonnie Schwartz, Carl and Sten Vikner and Rick Waswo, and other participants in the work in progress seminar of the English Department Needless to say the shortcomings of the paper are my own. Special thanks to Hedwig De Pauw for help with the material. 1. So far I have not come across any examples of non-overt first person subjects in French diaries. However a non-overt first person subject is possible in informal writing: (i)

Suis allee au magasin. Am gone to the shop.

2. Speakers of the languages in question, diary writers or not, share the intuitions given. Thanks to Michel Jeanneret and Corinne Grange for discussing the French data with me. Thanks to Margaret Clayton and Rick Waswo for help with the English data. 3. Following the attested French examples I take third person non-overt subjects for French. 4. See Haegeman (1989) for more examples. Third person pronominal subjects may also be nonovert in English diary contexts. See note 1 on French non-overt first person subjects in colloquial writing. 5. TOP differs from the null operator in purpose clauses, infinitival relatives, infinitival complements to adjectival degree specifiers, infinitival complements to adjectives of the tough-dass, represented from now on as OP, as shown convincingly by Authier (1989). Authier assumes that TOP is interpreted, i.e. assigned a range, at the level of discourse. Raposo (1986: 380) offers a syntactic account of the interpretation of the discourse topic. He proposes that TOP is coindexed with a non-overt discourse topic at LF. He suggests (384-6) that the parameter distinguishing languages with zero discourse topics from those without such topics is that in the former languages the syntactic rule of predication that will assign a range to TOP has access to pragmatics while in the latter languages it does not. This proposal is unattractive because it implies that the sentence grammar proper, i.e., the language specific module, has access to the level of discourse, which presumably is part of the central processing mechanisms. This move seems undesirable in principle. Given a theory of utterance interpretation like Relevance Theory it is probably not a necessary move. 6. If (ii) is an ECP violation, (i) should be grammatical: (i) Bill thought would pass my driving test, (ii) *Bill thought that would pass my driving test Ray Lurie (p.c.) has provided me with the following example: (iii) Mary in hospital. Bill thinks [ec] will live. 7. I shall use the representation OP for the null operator in the pronoun zap phenomenon, because it is not clear to me whether the topic operator can be assimilated to it (see discussion below). 8. Thanks to Sten Vikner for pointing out the importance of the Danish data. Thanks to Carl and Sten Vikner for help with the examples.

174

L. Haegeman

REFERENCES Authier, J.-M.P. (1989) "Two Types of Empty Operator", Linguistic Inquiry 20, 1, 117-125. Campos, H. (1986) "Indefinite Object Drop", Linguistic Inquiry 17, 354-59. Haegeman, L. (1986) "Complement ellipsis in English or how to cook without objects", in A Festschrift for Professor Dr. Ren£ Derolez on the occasion of his 65th birthday. Department of English and Old Germanic philology, State University of Ghent, 348-361. Haegeman, L. (1987) "Complexity and Literary Prose", Language and Style 20, 3. 214-222. Haegeman, L. (1988) "Register Variation in English: some theoretical observations". Journal of English Linguistics 20, 2, 230-248. Haegeman, L. (1989) Understood subjects in English diaries. On the relevance of theoretical syntax for the study of register variation, ms. University of Geneva. Huang, J. (1984) 'On the distribution and reference of empty pronouns.' Linguistic Inquiry 17, 531-74. Massam, D. (1989) Middles, Tough and Recipe Constructions: Licensing of Null Objects and NonThematic Subjects, ms. Massam, D. and Y. Roberge (1989) "Recipe context null objects in English", Linguistic Inquiry 20, 1. 134-139. Raposo, E. (1986) "On the Null Object in European Portuguese", in Jaeggli, O. and C. SilvaCorvalan (eds.), Topics in Romance Syntax, Foris, Dordrecht, 373-390. Reinhart, T. (1981) "Pragmatics and linguistics: an analysis of sentence topics", Philosophica 27, 53-94. Rizzi, L. (1986) "Null objects in Italian and the theory of pro", Linguistic Inquiry 17, 3, 501-551. Ross, J.R. (1982) "Pronoun deleting Processes in German", paper presented at the annual meeting of the Linguistic Society of America, San Diego, California. Sperber, D. and D. Wilson (1986) Relevance, BlackweU. Stowell, T. (1985) "Null Antecedents and Proper Government", Proceedings of the Fifteenth Annual Meeting ofNels, GLSA, University of Massachusetts, Amherst Syea, A. (1988) "Null objects in Mauritian Creole", unpublished ms.. University of Manchester.

DIARIES CITED Buddingh', C. (1978) Een mooie tijd om later te worden, de Bezige Bij, Amsterdam. Jacobsen, Jorgen-Frantz (1958) in William Heinesen (ed.), Det dyrebare Live. Jorgen-Frantz Jacobsen i Strejflys af hans Brefe, Gyldenhal, Copenhagen. Kafka, Franz. (1976) Tageb cher. 1910-1923, in Franz Kafka. Gesammelte Werke. Herausgegeben von Max Brod, Fischer Taschenbucher Verlag, Frankfurt am Main. Leautaud, Paul (1989) Le Fleau, Journal particulier 1917-1930, suivi d'un fragment ΊηέάΛ 1932, Preface d'Edith Silve, Postface de Pierre Michelot. Mercure de France, Paris. Woolf, Virginia (1985) The diary of Virginia Woolf, Volume 5: 1936-41. Eds. Anne Olivier Bell assisted by Andrew McNeillie. Penguin Books, Middlesex. First published by Chatto and Windus, The Hogarth Press 1984.

Datives in German "ECM"-constructions H. Haider University of Stuttgart

0. INTRODUCTION

The following set of examples illustrates a well known difference between German-(ic) and Romance causative constructions: The subject-argument of an embedded transitive verb may surface as an accusative (or as an agent-phrase). In French or Italian, it may surface as dative (or as an agent-phrase). (1)

a.

II fera boire un peu de vin ä son enfant He'll have his child drink a little wine b. *I1 fera boire son enfant un peu de vin c. II a fait partir son ami He made his friend leave d. *I1 a fait partir ä son ami

(2)

a. *Er ließ seinem KindDAT etwas WeinACC trinken

b. Er ließ sein KindACC etwas WeinACC trinken c. Er ließ seinen FreundACC abreisen d. *Er ließ seinem FreundDAT abreisen From the point of view of comparative syntax, one should be able to answer at least the following two questions: 1. Why is the construction of the type (Ib) absent in Romance? 2. Why is the construction (2a) missing in German? A preliminary answer for (1) seems to be straightforward. The existence of (la) presupposes the possibility of a double object construction. If this construction is precluded on independent grounds, it will be precluded for the causative construction as well (cf. Comrie 1981: 171). Since German (and English), but not French or Italian, allows double accusatives independent of causatives (cf. 3), there will be no restriction against a construction like (2b). (3)

a.

Er fragte ihnACC etwasACC He asked him something

176

H. Haider b.

Er lehrte ihnACC etwasACC He taught him something

As for the second question, the empirical fact is puzzling. In terms of UG, there should be no absolute restriction against applying the Romance strategy in a Germanic "EMC"-construction as well. It is unclear why a German Dative should be unable to serve the same grammatical function as the French or Italian one. Hence the difference must be a fairly low-level one, presumably a lexical one.1 In the subsequent sections, I will show that Dative-"ECM"-subjects are widespread in German dialects (sect. 1) as well as in previous stages of German (sect. 2) in constructions with the verb lassen (= let/make2). They are missing, however, in those dialects which are close to the standard variety of modern German, i.e. northern dialects, but also in Alemannian dialects, another major dialect group of German. In section 3, the Dative-"subject" will be shown to differ significantly from an ECM-subject. Section 4 sketches a tentative explanation.

1. DATIVE CAUSATTVES IN AUSTRO-BAVARIAN DIALECTS

In a brief informant-survey, I checked four dialects of Austria and Bavaria and found that my own dialect conformed with the other dialects, each one from a distinct dialectal area, with respect to dative "ECM"-constructions: Dialect l (Waldviertel)3 (4) a. Löß iajjAj 'nACC do heiradn, waon's heiradn wii Make her him marry, if she wants to marry him b. Löß des waiDAT de eabfiACC schöin, waon 's de eabfi schöin wü Make this woman the potatoes peal, if she wants to peal the potatoes in this dialect, the Dative-causative is the genuine construction. Double accusatives are accepted, however, in a register compromising on the standard usage, but are avoided in the dialect proper. The following examples are taken from geographically nonadjacent areas: Dialect 2 (Seewinkel)4 (5) a. Löß iauAT 'n do hoiradn, woea's 'n hoiradn wiil b. Löß dean woe^y d'grumbbian schöin, woea's es schöin wül

Datives in German ECM -constructions

177

Dialect 3 (Murtal, Styria)5 (6) a. Löß iraQAT earn heiradn b. Löß dean weiblDAT d'eadebfl schöön Dialect 4 (Bavaria, Obe falz)6 (7) a. Lou'n ihraDAT heiradn, wen's'n wü b. Lou(s) dem waeDAT de eadöbfi schöön, wen's wü In all these dialects, the Dative and the Accusative are morphologically distinct, i.e. there is no case syncretism. The following example is representative for all the dialects: Dialect 1: (8) a. I I b. I I

hob iaDAT/*'sACC ghoifn have helped her hob 'SACC have seen her

Like in Romance languages, the Dative occurs only on the subject of a transitive verb. If there is no complement, the only possible case is the accusative. Again, the following example is representative for other dialects as well. Dialect 1: (9) a. Löß 'SACC schlofa Make her sleep b. *Loß ia schlofa Western (Suabian) and northern dialects (Hessen, Sachsen, etc.) lack the dativecausative construction, according to the judgements of the informants I consulted.

2. DIACHRONIC OBSERVATIONS ON DATIVE CAUSATTVES IN GERMAN

Most of the comprehensive descriptive grammars of German (Behagel, 1923 vol. I, p. 623ff.; Dal 1966:37; Paul, 1919 vol. , p. 394ff.) note that from the 17th to the 19th century dative causatives are found in the texts. Paul (1919, vol. Ill: 394) claims that in the 18th century the dative became nearly obligatory and that only in the 19th century the modem German double accusative construction became predominant. The following examples are taken from the above mentioned sources:

178 (10)

H. Haider a.

b.

c. d.

unt liez iuDAT daz sehen an (middle high German; Behagel 1923: 623) and let her that look-at (and let her watch that) wenn er dem Auge^T glauben machen wollte,.. (17th cent.; Behagel p. 624) if he the eye believe make wants (if he wants to make the eye believe,...) Ich ließ dirDAT spätere Zeiten sehen (19th cent.; Dal 1966:36) I let you coming times see lasset mirDAT doch den brieff lesen (17th cent; Paul 1919:394) let me anyhow the letter read

All the examples listed above (10b)-(10d) are ungrammatical in modern German, unless the dative is replaced by an accusative. However, this is not a particular feature of standard German, since there are other dialects which are like the standard in this respect, as was mentioned above. For instance there is no Dative causative construction in Suabian7 or the dialects of Hessen8 and Saxonia9. Unfortunately, I do not have sufficient information to be able to decide whether the construction is confined to the Bavarian-Austrian dialects of German. With regard to other Germanic languages, I have only anectotal evidence. The following example illustrates a Dative causative construction in Dutch: (11)

(Aan) hem laat ik dit werk niet doen (to) him let I this work not do (I do not let/make him do this work)

In Dutch, aan is the preposition for indirect objects, comparable to English to.

3. THE SYNTACTIC STATUS OF THE DATIVE NP: "ECNT-DATIVE?

In this section I will compare the dialectal Dative "ECM"-subject with the standard Accusative ECM-subject in order to determine whether they are instances of the same kind of construction. Let us review first some distributional properties of the "ECM"-Dative: 1. It is found only with transitive verbs that govern an object. (12)

a. b.

Löß *maDAT/miACC schlofa (intransitive verb) Let me sleep Löß *iaDAT/siACC mia net auf d zechn foin (ergative verb) Let her/she (to) me not on the toes drop

Datives in German ECM-constructions c. d. e. f.

179

Löß *iaDAT/siACC lesn (optionally transitive, intransitive usage) Let her read Löß iaDAT/*siACC des biachl lesn (transitive) Löß iaoAT miaDAT do hööfn (intransitive with DAT-object) Let her me but help (Let her but help me) Löß *iaDAT/siACC a kechin wean/sae/bleim (copula) Let her a cook become/be/remain

2. Unlike the ACC-ECM-subject in standard German, the Dative does not induce opacity effects. In German, pronouns usually appear in the so-called Wackemagel-Position, which is analyzed as a position following the COMP-position.10 (13)

a.

daß ihn ihr Fritz vorstellte that him her Fritz introduced (that Fritz introduced him to her)

As was noted by Grewendorf (1987), ECM-subjects block this kind of pronoun distribution. (14) *daß ihn ihr Fritz niemanden vorstellen ließ that him her Fritz nobody introduce let (that Fritz did not let anybody introduce him to her) In Haider (1987), I described that ergative ECM-subjects do not block the fronting of the object pronoun: (15)

daß ihr Fritz einen Stein auf die Zehen fallen ließ that her Fritz a stone on the toes fall let

The subject of an unergative verb like introduce blocks the fronting of the pronoun across the matrix subject in (14), while the subject of the ergative verb fall in (15) does not. That the presence of the ECM-subject is the crucial factor becomes clear in the contrast between (14) and (10). (10) differs from (14) only with respect to the presence/absence of the subject-NP: (16)

daß ihn ihr Fritz vorstellen ließ that him her Fritz introduce lets (that Fritz lets him (be) introduce(d) to her

The contrast between (14) and (15) or (16) follows, if we assume that only the local Wackernagel-position is accessible, and that local is defined in terms of

180

H. Haider

accessible subject. The local WP is the WP in the governing category that contains an accessible subject. The ECM-subject counts as an accessible subject11 for an object pronoun. The following dialectal examples illustrate the lack of opacity effects with the dative 'subject': (17)

a.

b.

c.

wei des, da fritz den wai e, ned mocha loßt since this Fritz the woman not do lets since Fritz does not let the woman do this wei deSj da fritz ^ ned moch loßt since this Fritz not do lets (since Fritz does not let this (be) do(ne)) worum loßt'nj do da Fritz net sein wai e{ schreibn why let it there Fritz not his wife write why doesn't Fritz then let his wife write it

In (17b) the subject is missing, which leads to a passive equivalent just like in standard German. 3. Unlike ECM-ACC, the ECM-DAT cannot bind a reflexive. (18)

a. b.

(19)

ET{ ließ die LeutCj sich.^ Schnaps besorgen He let the people (for) them/him-self schnaps buy hat sichj Schnaps besorgen lassen He has himself schnaps buy let (He got bought schnaps)

a. *Eai hod den leidnj si^ an schnobs midbringa lossn (cf. 18a) b. Eaj hod sij an schnobs midbringa lossn (cf. 18b)

In (19a) the anaphor is unbindable: Neither the matrix-subject nor the Dative can serve as an antecedent. Given that the Dative-NP represents the ECMsubject, the matrix-subject is not in the local domain, hence not able to bind. The same is true for (18a). What prevents the Dative from acting as the local binder? This particular restriction need not to be taken as a peculiarity of this construction. In general, datives do not bind reflexive co-arguments: (20)

a.

Ich habe die Leute; € einander; vorgestellt I introduced the people to each other b. *Ich habe den Leutenj DAT einander; vorgestellt I have (to) the people each other introduced

Datives in German ECM-constructions

181

Although the word order with respect to the DOsequencing is free for a verb like introduce in German, a Dative-NP, irrespective of its linear position, cannot bind an Accusative anaphor. The restriction illustrated in (19) provides crucial evidence for any explanatory attempt, since we should expect that a Dative on a subject argument c-commands an object and hence would qualify as a binder. Hence either the Dative is not an ECM-subject or the fact that Dative-objects cannot bind an anaphor cannot be derived from a c-command violation, i.e. by hypothesizing that direct objects are not c-commanded by indirect objects on the level where binding principles apply. 4. Unlike ECM-ACC, the ECM-DAT cannot control an infinitival adjunct. In German, infinitival adjuncts are controlled by the subject Thus and ECMsubject may act as a controller. (21)

Er ließ ihn; [ohne PRO; instruiert worden zu sein] den Raum betreten He let him [without having been instructed] the room enter

Of course, (21) is ambiguous, since the matrix subject is a potential controller as well. (22)

a.

Er ließ die Frau, [nicht [ohne gegessen zu haben]] die Arbeit verrichten b. *Er hod dera fraUj [ned [ohne wos g'essn z'haom]] de oawad mocha lossn He has the woman not without having eaten the hard work do let (He did not let the woman do the work without having eaten)

(22b) is unambigous: it is grammatical only with the matrix subject as controller. 5. Unlike ECM-ACC, the ECM-DAT cannot trigger a predicative nominative. In standard German, a predicative NP in a copula construction either surfaces in the nominative or with the case of the NP it is predicated of: (23)

a. b. c.

ErNOM ist [ein großer Held]NOM He is a big heroe [PRO [ein großer Held]NOM zu sein] ist anstrengend A big heroe to be is hard Laß ihnACC [einen großen Helden]ACC sein Let him a big heroe be

182

H. Haider d.

Laß ihnACC [ein großer Held]NOM sein Let him a big heroe be

The crucial cases are (23b) and (23d). In (23b) the nominative appears in an infinitival context, which precludes the assignment of nominative. In (23d) the predicative NP shows nominative case although it is predicated on an ECMAccusative in an ECM-construction. Whatever mechanism is responsible for the assignment of a nominative to a predicative NP, it is triggered by the subject property. Predication on an object invariably produces case assignment by agreement (24)

a.

Sie nannten ihn einen großen HeldenACC They called him a big heroe b. *Sie nannten ihn ein großer HeldNOM

The same patterning we find in predication constructions with the particle als: (25)

a. b. c.

PRO [als ein großer Held]NOM zu sterben As a big heroe to die (to die as a big heroe) Ich lasse ihn [als ein großer Held]NOM sterben I let him as a big heroe die Ich lasse ihn [als einen großen Helden]ACC sterben I let him as a big heroe die

Since copula structures do not appear with 'ECM-Datives' in the dialects, we can check only the latter construction: (26)

a.

Da vodda had iaDAT [ois oanziga dochda]DAT des ned mocha lossn The father has her as the only daughter this not do let As the only daughter, father would not have let her do that) b. *Da vodda had iaDAT [ois oanzigi dochda]NOM des ned mocha lossn

(26b) shows that the 'ECM-Dative' patterns rather like an indirect object than like an ECM-subject, since it does not trigger a predicative nominative, which is triggered by the accusative both in the dialect as well as in standard German: (27)

a.

Löß den maon ois easchtaNOM/easchtnACC schpringa Let/make the man as the first one jump

In summary, the evidence points to the conclusion that the Dative on the external argument of a verb embedded under lassen does not have the subject properties associated with ECM-subjects: the ECM-subject in standard German appears in the accusative case, it is found with transitive, intransitive and

Datives in German ECM-constructions

183

ergative verbs, it binds a reflexive, controls a PRO-subject in an adjunct clause and triggers NOM in predicative constructions. The dative under lassen lacks these properties.

4. TOWARDS AN EXPLANATION

Elsewhere (Haider 1988, 1989) I developed a technical concept of a 'unified argument structure' associated with a verbal complex. Lassen is one of those verbs which occur only in a verbal cluster, a property sometimes referred to as Obligatory V-raising'. Argument unification provides the technical means for base generating a verbal cluster without violating the projection principle. Auxiliaries and ECM-verbs, especially lassen, are characterized as verbs with a (partially) unspecified argument structure. Put informally, these verbs have an empty argument structure, i.e. thematically unspecified slots. Auxiliaries do not have any specified argument, whereas a causative verb has a single specified argument, the external one, and the rest of the argument structure is unspecified. Unification with a transitive verb leads to a passive effect, because the external argument of the embedded verb cannot be unified, since in the verbal format there is room for one external argument only. For details of this analysis I refer to Haider (1989). The crucial consequence of this analysis is that clauses with an oven ECMsubject of a transitive verb cannot have the same kind of structure as clauses without it (28)

a. b.

Er ließ; [[mich das Haus streichenle;] He made me paint the house Er ließ [das Haus streichen ej He made the house paint (= be painted)

Ergative verbs, which lack an external argument, unify their argument structure completely and hence the ECM-subject of an ergative verb does not produce the kind of opacity effect (cf. examples (14) and (15) above) triggered by a transitive subject, the reason being that ergative verbs yield a structure like (28b) whereas transitive verbs with an overt external argument as subject require a structure as in (28a). Now we are in the position to sketch the relevant distinction between dialectal and standard constructions with the verb lassen: in the dialects under discussion, lassen has an external argument of its own. In addition, dialectal lassen is a potential dative assigner:

184 (29)

H. Haider a. Worum lossn's des net ia? b. % Warum lassen sie es nicht ihr? Why don't they leave it (to) her?

(29b) occurs only as a hypercorrect attempt to translate (29a) into standard German. In standard German proper it is ungrammatical, unless it is construed as an elliptical expression with the infinitival verb missing. It is instructive to note that lassen in (29a) has only a permissive reading, not a causative, i.e. the indirect object of lassen is interpreted as the individual which receives something under its own control. It is this very argument slot which inherits the thematic specification of the external argument of the infinitival verb governed by lassen. This might be responsible for the fact that the predominant reading of dialectal /auen-constructions is permissive rather than causative. An immediate consequence of this analysis is that the dative-NP unlike the ECM-accusative is not a structural subject: it is structurally an indirect object, whence its syntactic object properties. Another consequence is that it is impossible to combine a dialectal /ossen-construction with a benefactive dative, like in the standard variety, since the dative-slot is occupied already: (30)

a.

Wir lassen dirDAT [jemanden es ihm überbringen] We make (for) you someone it (to) him transmit b. *Mia lossn diaDAT ia des earn bringa We let (for) you her it him bring

Finally, it follows that an 'ECM-dative' does not occur with verbs of perception. These verbs lack a dative-argument position, hence cannot assign a dative: (31)

a. *I hob iaDAT des lesn g'seng I have her that read seen b. I hob siACC des lesn g'segn

5. CONCLUDING REMARKS

The Romance causative construction with a Dative-marked external argument of the infinitival verb finds a close parallel in dialects of German. The analysis of this construction points to the conclusion that the external argument marked with the Dative is not a subject constituent in structural terms but rather an indirect object of the matrix verb. What this amount to is reminiscent of Object raising'. The particular analysis I presented does not assume a syntactic process of object raising. The apparent effect of Object raising' is the consequence of an operation on the argument structure in German verb-cluster constituents.12

Datives in German ECM-constructions

185

NOTES 1. Pace Chomsky 1988, I assume that subparametric differences are to be localized in the lexicon. 2. In all dialects as well as in standard German, the "ECM"-verb lassen (with its dialectal variants) can receive either a causative or a permissive reading, i.e. either cause someone of permit someone. 3. This dialect, the native language of the author, is spoken in the north-west part of Lower Austria, a highland area north of the Danube. The transcription I use is very broad, basically using the nearest equivalent of the standard German writing conventions. 4. This dialect is spoken in the east-most area of Austria. The informant, Ch. Reinprecht, speaks the dialect of the village Apetlon. 5. This dialect is spoken in Styria, the south-east part of Austria, in the upper part of the valley of the river Mur. The informant, H.C. Luschutzky, speaks the dialect of the area of Judenberg. 6. This is a North-Bavarian dialect, spoken in the Oberpfalz, the north-east part of Bavaria. The informant, Josef Bayer, speaks the dialect of the area around the village Dielfurt. 7. This dialect is spoken in and in the vicinity of Stuttgart, 8. Hessen is the area around Frankfurt. 9. The area around Leipzig. 10. The details of the analysis are immaterial for our present concern. The WP seems to be an adjunction position to the complement of C. Up to now it is unclear what forces the NOM-DATACC order on a series of pronominals. 11. For a more detailed analysis I refer to Haider (1989). 12. In technical terms, this process is a syntactically controlled instance of functional composition: first the two functors are combined and only the functional application takes place, which is called theta-assignment in GB-terms. Functional composition provides a means of respecting the thetacriterion without the need for extra levels of structure.

REFERENCES Chomsky, N. (1988) "Some Notes on Economy of Derivation and Representation", ms. . Comrie, B. (1981) Language universal! and linguistic typology, Blackwell, Oxford. Behagel, O. (1923) Deutsche Syntax, Bd. Hi, Heidelberg, Winter. Paul, H. (1919) Deutsche Grammatik. Bd. Hl, Teil IV: Syntax, Max Niemeyer. Halle a.S. Dal, I. (1966) Kurze deutsche Syntax, Niemeyer, Tübingen. Grewendorf, G. (1987) "Kohärenz und Restrukturiering. Zu verbalen Komplexen im Deutschen", in B. Asbach-Schnitker and J. Roggenhofer (eds.). Neuere Forschungen zur Wortbildung und Historiographie der Linguistik, Narr, Tübingen. Haider, H. (1987) "Nicht-sententiale Infinitive", Groninger Arbeiten zur germanistischen Linguistik 28. 73-114. Haider, H. (1988) "PRO-Neme", ms. Univ. Stuttgart. (In print in: G. Fanselow and S. Felix (eds.), Strukturen und Merkmale grammatischer Kategorien, Narr, Tübingen.) Haider, H. (1989) "Against Raising", in Jaspers, D. et al. (eds.). Sentential Complementation and the Lexicon, Foris, Dordrecht, 173-187.

A principle of global binding* Lars Hellan University of Trondheim

1. INTRODUCTION

This paper makes two proposals, based on phenomena involving reflexives in Norwegian: (i) Binding theory should be split into two components: (a) the component of individual binding; (b) the component of global binding. Relative to some construction K of arbitrary size, individual binding in K consists of binding relations in K seen individually, each one in isolation from the others. This is the perspective of 'classical' binding theory as formulated, e.g., in Chomsky (1981). Global binding in K consists of sets of binding relations in K, the members of such sets seen relative to each other. The globality aspect of binding has been addressed in Reinhart (1983) and Higginbotham (1983), although not in the form of the present proposal. (ii) At the level of global binding, principles of subject-oriented locality of binding hold with much less exception than they do at the level of individual binding: an example to be adduced is what we call the Local Subject Selection Principle (LSSP), formulated in (13) and (15) below, based on Norwegian. Given the existence of long distance anaphors, as well as non-subject oriented anaphors throughout many languages, there is no universally valid counterpart to this principle at the level of individual binding, even though many anaphors also observe subject-oriented locality. Although we cannot claim universality for a principle like LSSP on the basis of only one language, the fact that individual level binding in Norwegian deviates quite clearly from it, whereas global level binding observes it, gives an indication that this global level principle has more of a chance to stand up as a universal principle than any principle formulated of individual level binding. The notion of binding to be adopted is the same as what is called 'linking' in Higginbotham (1983). According to this conception, binding is a syntactic relation holding asymmetrically between two expressions, a binder A and a bindee B. The relation is not transitive: if A binds B and B binds C, it does

188 L. Hellan not follow that A binds C. Moreover, a bindee has a unique binder (whereas a given binder can bind more than one element). The representation of binding should reflect the asymmetric and non-transitive nature of the relation. For these purposes, traditional co-indexing is ill-suited, and we instead assume an interplay between token-indices and binding indices, token indices being assigned to all nodes in a structure distinctively, so that no two nodes have the same token index, and binding indices being features associated with bindees, marking exactly which node acts as their binder. This feature really functions like an arrow, with the token index of the binder corresponding to the point of the arrow, and so, for perspicuity, this feature can be represented by a line going from the bindee to the binder. Binding may be contrasted with coreference, which is a semantic notion, the expression A being coreferential with the expression B whenever A and B have the same denotation. This relation is both symmetric and transitive. What traditional coindexing is suited for is the representation of coreference, and for expository purposes, we will use the device in that way throughout. When the bindee is an anaphor or a pronominal, it will, of course, often be the case that the elements of a binding relation are coreferential, but not always, as when the bindee has no denotation at all, like in cases of anaphors with non-argument status. Moreover, there are many instances of coreference which do not involve binding, such as the coreference between John and he in Three weeks after John came home, he got shot; hence there is no coextensionality between the two notions either, confirming their distinctness. As a matter of terminology, we define an anaphor as a phrasal element that has to be bound, whereas a pronominal is a phrasal element that may, but need not, be bound. An anaphoric element is a word or morpheme whose occurrence inside a phrase X induces the status of X as an anaphor. A binding configuration of an anaphor B is a minimal structure containing both a binder for B and B, where by a 'minimal structure containing Χ, Υ, Ζ, ...' we mean a structure containing X, Y, Z, ..., and no item c-commanding any of these which is not already contained in one of them. The sum of possible binding configurations of the anaphor B is the domain of B. The binding conditions of B are the conditions delimiting the domain of B. Classical binding theory is a theory about possible binding conditions on anaphors, at the level of individual binding.

2. THE SYSTEM OF ANAPHORS IN NORWEGIAN

As the data to be considered come from Norwegian, we give a very brief introduction to the essential properties of the Norwegian anaphoric system, reflecting views of Hellan (1988).

A principle of global binding

189

Norwegian has two reflexive (anaphoric) elements, the words seg (with a distinct reflexive form only in third person) and selv (same form for all persons). They combine to form the anaphor [seg selv^. [seglfjp can also occur as an anaphor, whereas selv, in order to form part of a normal reflexive phrase, has to combine with some version of seg. The reciprocal anaphor is hverandre; here we will deal only with the reflexives. It is argued in Hellan (1988) that the binding conditions of a Norwegian reflexive anaphor are induced by its anaphoric elements, and we will assume so in the following.1 Briefly, the element selv imposes on a reflexive B the requirement that the binder of B be a co-argument of it, where X and Υ count as co-arguments under the following circumstances: (a) they get their theta-roles from the same role assigner, or (b) they have a direct syntactic relation to the same head (thus, the subject and object of an accusative-with-infinitive construction count as co-arguments). The element seg imposes on a reflexive anaphor Β the requirement that its binder A must predication-command B, where X predication-commands Υ if Υ is contained in a constituent Ζ which is predicated of X. A reflexive of the form '[seg selv],^' thereby has to satisfy two conditions: that of being a co-argument with its binder, induced by selv, and that of being predication-commanded by its binder, induced by seg. (In addition come certain other conditions, such as a tensed S -requirement, but the two mentioned will be the essential ones for the purposes of this discussion.) Clearly, the co-argument condition produces local anaphors, i.e., anaphors whose binding configuration is a 'local', or 'small', configuration; in the absence of selv, on the other hand, the binding configuration of the reflexive (i.e., [segjj^p) will be non-local, also called 'long distance'. Accompanying the binding conditions induced by each element, there is a complementarity principle to the effect that once an element E induces D as a binding domain for the reflexive containing E, then the binding domain of any anaphor not containing E has to be disjoint from D. This situation obtains both with regard to seg and selv, as argued in Hellan (1988). Examples (1) and (2) serve as illustrations of these conditions and principles. (la) illustrates the long distance potential of the anaphor [seg],^, and (Ib) illustrates the locality requirement induced by selv in the anaphor [seg selv],^: the local subject PRO being 1st person, the 3rd person form of the reflexive leads to conflicting person features. Coindexing marks coreference, but when binding relations exactly match coreference, the indexing also serves to indicate binding: (1)

a. Joiii ba megk PROk snakke om seg; b. *Jonj ba megk PR(\ snakke om seg sety Jon asked me PRO to talk about REFL

190

L. Hellem

(2) illustrates the complementarity induced by selv: once the binding configuration in (2a) satisfies the conditions induced by selv, this is not a possible binding configuration for an anaphor lacking selv?· (2)

a. Jorij forakter seg selvj b. "Joiii forakter seg; Jon despises REFL

Similarly, (3) illustrates the complementarity induced by seg: since the reflexive here is predication-commanded by the binder, only the version with seg is possible: (3)

a. Jorij snakker om seg sety b. "Joiii snakker om ham Jon talks about REFL

2. A GLOBAL BINDING CONDITION

Consider first the following question: why is it that the occurrences of in (5) and (6) are prevented from being long distance bound, given that this possibility is available in constructions like (la) and (4)? Lines in (5) and later examples mark binding relations, when they are not predictable from the coreference relations: (4)

Jonj h0rte megt snakke om seg{ Jon heard me talk about REFL

(5)

ι I "Joiij lovet megk PROj snakke om se& Jon promised me PRO to talk about REFL

(6)

ι *Jonj h0rte seg selvj snakke om Jon heard REFL talk about

REFL

As shown by (7), a reflexive in the position of [segl^, in (5) and (6) is acceptable when it is locally bound:

A principle of global binding (7)

a.

Joiij lovet megk PRO; ä snakke om seg selv; Jon promised me PRO to talk about REFL

b.

Jon; h0rte seg sety snakke om seg selv; Jon heard REFL talk about REFL

191

Adapting a proposal for this type of facts in Hellan (1988) (p. 178, building partly on proposals from Higginbotham, 1983), we may first consider a principle like (8): (8)

Tentative principle: a. No item A can serve as binder for two elements B and C if B c-commands C, even if the individual binding conditions associated with B and C are thereby fulfilled. The content intended can only be expressed by B binding C (or the appropriate alternative to C as dictated by individual binding theory, given B as the binder). b. (Same principle, from the perspective of C) Given a bindee position C, if there are two semantically and binding-theoretically eligible potential binders A and B for the bindee in C, then the one of A and B which is closest to C (and c-commands C) must be chosen as binder for that bindee.

Clearly, (8) is a global binding condition, since it makes a preference between two alternative sets of binding relations over the otherwise same configuration, modulo only the form of C as determined by the choice of binder. (8) makes no other structural specification than c-command. As we will now see, there are reasons to enrich C with more structural specification, at the cost of the reference to mere 'closeness'. First, when the binding distance gets longer than it is in (5) and (6), data no longer provide motivation for the closeness stipulation in (8). In each of (9a,b), both the interpretation of the construction and the form [seg^ are compatible with both of the binding relations indicated, whereas (8) predicts that only the shortest relation in each case will be possible:3 either:

or. (9)

a.

Joiij pr0vde PRO; ä h0re meg snakke om segj Jon tried to hear me talk about REFL

192

L. Hellan either:

or: b.

Jonj forestilte seg seg selv; h0re meg snakke om seg{ Jon imagined REFL hear me talk about REFL

Faced with such a situation, one might of course regulate that only the closest possible binder should count, since the data would not falsify that stipulation. But the data are also compatible with a principle which does not fix a binder in cases where both potential binders give long distance binding. Secondly, as illustrated in (11), even though the reflexive [ham selv^ can in principle be bound by a preceding non-subject, as shown in (10),

(10)

I

I

Vi fortalte Jon om ham selv we told Jon about himself

it cannot have a binder in this position when this binder in turn is bound by the subject:

(11) "Joiii fortalte Jon told

seg selv; om ham REFL about REFL

In this case, only (12) is possible, the exact opposite of what (8) would predict:

(12)

I Jon Jon

I I fortalte seg selv om seg selv. told REFL about REFL

In the case of (11)/(12), one possible strategy of saving (8) might be to say that in (11), seg selv does not really c-command ham selv. The structure of such constructions not being quite clear, one might propose, e.g., that fortalte seg selv in (11) forms a constituent There is hardly any independent evidence for such an analysis, however: this alleged constituent cannot be moved on its own (cf. the illformedness of *fortelle Jon bj>r du ikke om ham selv 'tell Jon should you not about himself, where forteile Jon turns out to be unable to undergo the kind of VP-fronting otherwise generally available for verb phrases in Norwegian); and if one adheres to something remotely like a c-command restriction on binding, the wellformedness of (10), of course, argues against this analysis. (It is well known that ham selv can be bound from within a PP, as in vj snakket med Jon om ham selv 'we talked with Jon about himself, where Jon

A principle of global binding

193

and ham selv however are coarguments. If also the 'small VP' analysis in question could be made compatible with counting Jon and ham selv as coarguments in (10), the latter being sufficient to license the reflexive, then the wellformedness of (10) cannot obviously be held against the combination of the 'small VP' proposal and (8) as an account of why (12) is possible but not (11)·) Given the moot status of a small VP analysis of (10)/(11)/(12), a more radical proposal excluding seg selv as a possible binder in (11) is to say that if one of the semantically eligible binders is a subject, then it is to be chosen. We will adopt that solution. As for the closeness clause of (8), we will modify it according to what data such as (5), (6), and (9) positively show, and thereby add that the preference for a subject as binder holds only when this subject is a local subject, that is, a coargument of the bindee. This leads to the reformulation in (13) of (8), where the notion 'subject* is construed in terms of 'predication-command' as defined above: (13)

a.

b.

No item A can serve as binder for two elements B and C if B predication-commands and is a coargument of C, even if the individual binding conditions associated with B and C are thereby fulfilled. The content intended can only be expressed by B binding C (or the appropriate alternative in the position of C as dictated by individual binding theory, given B as binder), (Same principle, from the perspective of C) Given a bindee position C, if there are two semantically and binding-theoretically eligible potential binders A and B for the bindee in C, then if one of them both predication-commands and is a coargument of C, then it must be chosen as binder for the bindee in C.

We will contemplate below some aspects and consequences of (13) and subsequently its status as a global binding condition. First we consider a possible alternative to the analysis now proposed. The definitions of binding laid out in the beginning are not the only ones conceivable. The main alternative will be to treat binding as a transitive notion, so that if A binds B and B binds C, then C is also bound by A. On this view, an item can then be bound by more than one NP. The traditional coindexing notation is compatible with this construal. The asymmetry of binding can still be maintained via an extra requirement (or definition) to the effect that a binder must c-command the bindee (modulo the type of case mentioned above where a binder is embedded in a PP). This construal will allow for a partly alternative account of the data considered here, but it is doubtful whether it will be an improvement. Regarding (9) first, and assuming it has the freedom of choice of binder opted for above,

194

L. Hellem

the way to allow for this freedom will be to say that any pair of coindexed nodes, one c-commanding the other, represents an admissible binding relation. To account for the illformedness of (11), one may add a principle to the effect that when an item is bound by two NPs, each binding relation must induce the same morphological form of the anaphor; what is wrong about (11) is then a morphological conflict, the long relation demanding the form seg selv, the short relation demanding the form ham selv. The ill-formedness of (5) and (6) is subject to the same account, since here, the long relation induces the form seg, the short relation induces seg selv. What 'saves* the double binding in (9) is thus simply the fact that both relations are of the long distance type, both yielding the form seg. This is not the whole story, however. In the case of the wellformed (7) and (12) (repeated below with those binding relations which induce wellformedness), one has to prevent the induction of a representation of binding between Jon and seg selv in (7a,b), and seg selv and seg selv in (12), since such a representation would lead to the same conflicts in morphological shape as were assumed to rule out (5), (6), and (11):

(7)

(12)

a.

Jorii lovet megk PROj ä snakke om seg Jon promised me PRO to talk about REFL

b.

Jorij h0rte seg selvj snakke om seg sely; Jon heard REFL talk about REFL

Jon fortalte seg selv om seg selv. Jon told REFL about REFL

In the case of (12), this is straightforward, as long as it is binding and not coindexing which has the transitivity property: from the circumstance that Jon binds both occurrences of seg selv, it does not follow by transitivity that the first seg selv should bind the other, something which would otherwise have resulted in an extra binding configuration demanding ham selv instead of the last seg selv, i.e., conflicting morphological forms. For (7a,b), however, the situation is worse: transitivity of binding here entails that Jon should be able to bind a reflexive in the position of the last seg selv in these constellations, a reflexive which would get the form seg, leading to conflicting morphological shapes. It seems that in order to prevent such a state of affairs, one will have to prevent transitivity of binding from applying under certain circumstances, namely when the intermediate link of the transitivity schema (i.e., the 'B' in 'if

A principle of global binding

195

A binds B and B binds C, then A binds C) predication-commands and is a local subject of C. (To allow for the freedom of choice of binder in (9), we cannot block the transitivity when the 'B'-link is at a further distance away (i.e., PRO in (9a), seg selv in (9b)).) Clearly, such a constraint on the transitivity of binding violates the spirit of the approach in question, and it reflects exactly the content of (13). It therefore seems that (13), with its underlying conception of the nature of binding, provides the most consistent approach to the problems at hand.

3. CONCLUDING REMARKS 3.1. Global binding The reference to sets of binding relations, the characteristic of a global binding condition, is only implicit in (13), but can be made explicit through the reformulation in (IS), assuming the definitions in (14): (14)

Definitions. a. A binding network in a construction K consists of a consistent set S of binding relations in K involving NPs with one and the same referent such that for each such relation, if at least one of the elements in the relation is part of another binding relation in K, then that relation is a member of S as well. b. A binding network base in a construction K consists of a maximal set of NPs which may constitute the domain of the relations of a binding network in K.

(15)

Local Subject Selection Principle (LSSP): Given a bindee C and a binding network base F in a construction K such that C is a member of F, whenever possible, associate F with a binding network in K where the binder of C both predication-commands and is a coargument of C.

Thus, among the possibly many binding networks in K which are constructive over the set F as far as obedience of individual binding theory is concerned, (15) enforces the choice of networks where 'local subjects' are used as binders. This construal captures the intention of (13), and is clearly a global type binding condition.

196 L. Hellan 3.2. Conference as a grammatically relevant notion A binding network base is an equivalence class of NPs, and the relation coreference holds among all of them, according to the definitions in (14). Given the grammatical significance of LSSP (i.e., (IS)), this may give rise to the question of whether coreference as such thereby acquires a certain importance as a grammatical notion. Clearly, coreference is not the distinguishing criterion of the equivalence class in question: if I is the shared referent of the members of a binding network base F, then these members may well be just a subset of the total set of NPs having I as referent in the construction in question. For instance, in (16), all of the italicized NPs may be read as sharing a referent, but only the boldfaced ones belong to one and the same binding network base: (16)

After John had arrived, he realized that he had forgotten PRO to wash himself.

It is still clear that coreference is what provides the intuitive hold on the notion of binding network base, so even if it is formally represented only at some level of semantic representation, it is among those semantic notions which are constitutive to important parts of the grammar.4 3.3. LSSP as a possible universal As noted in the introduction, we clearly cannot conclude universality from results from a single language. As noted there, the fact that individual level binding in Norwegian can be both long distance and non subject-oriented, whereas global level binding observes LSSP, still gives an indication that this global level principle may have more universality than any principle formulated of individual level binding. Moreover, from an intuitive learning or acquisition viewpoint, it seems that LSSP is hardly as directly accessible as the individual binding principles are - global principles will seem to be inherently less amenable to observation. For that reason, it is not unreasonable to speculate that LSSP may have some kind of universal status. The issue clearly invites to further investigation.

NOTES * I am grateful to Lars Johnsen and audiences at the University of Trondheim and at the conference on Logical Form in Tilburg in the fall 1988 for comments on ealier versions of this paper. 1. Possibly, this kind of dependence may hold universally (for argumentation, see Hellan, to appear), but it is not part of the definition of 'anaphoric element': all it takes to be such an element is that it induces anaphoric status, regardless of which domain that anaphor gets.

A principle of global binding

197

2. There are apparently systematic exceptions to the complementarity illustrated by (2b), one type exemplified by constructions like Jon vasker seg 'Jon washes himself. It is argued in Hellan 1988 that these exceptional constructions have in common that the reflexive lacks argument status, while reflexives with selv uniformly require argument status, and that the complementarity only holds with regard to bindees with argument status. As these non-argument anaphors do not affect the points under consideration here, we will say nothing more about them. 3. Needless to say, these alternative binding relations can be attested as coexisting possibilities when more than one referent is involved, such as in (i): either:

(i)

Jon; ba Perk PR

1

/ ' \\

2

3

adjunction

I LD3ND I I

- —>

1 *

2 *

3 *

Yer Devocalization

*----->

Of course, in practice this prediction is very hard to test, because it requires a combination of relatively rare phenomena. Firstly, we need a rule that is able to eliminate a stress bearing syllable; secondly, this syllable must be followed by another syllable which is stray for a reason that has nothing to do with extrameticality. If it were extrametrical, one could argue that after the deletion of the head syllable the corresponding marker on the stress plane cannot move to the right because the only possible landing site is invisible. One could then say that for this reason the marker is deleted together with its constituent In order to have an unambiguous testcase, the stray syllable must therefore be outside the domain of an accent constituent for a reason that has nothing to do with extrametricality. In the next section I will show that Serbo-Croatian exhibits all the relevant phenomena. It has a rule that in certain circumstances eliminates a stressed syllable. Also, some syllables are outside the domain of accentual constituents, although they are not extrametrical.

3. A SKETCH OF THE SERBO-CROATIAN ACCENT SYSTEM

Like all Slavic languages, Serbo-Croatian has a so called Yer. It is generally agreed upon that a Yer is only realized if it is followed by another Yer. In all other environments it is deleted (see Rubach (1986) for a recent treatment of Slavic Yers). Slavic languages differ with respect to the realization of the Yers. In Serbo-Croatian Yer happens to be realized as a. In (7) I give two highly informal rules that account for the distribution of the Yers (see Hermans (1990) for a detailed account).

On the fate of stray syllables (7)

203

Yer Vocalization Turn a Yer that is followed by another Yer into a Yer Devocalization Delete a Yer that is not followed by another Yer

For the present discussion it suffices to say that the nom. sg. masc. and the gen. plur. contain a Yer. The rules in (7) explain why morphemes that end in a Yer syllable exhibit an α - 0 alternation, where a shows up in the nom. masc. sg. and gen. plur. Some examples are given in (8) (data are taken from Benson 1979). (8)

jec'am jec'ma

nom. sg. gen. sg.

'barley'

jec'mi jec'arma:

bakar bakra

'copper'

bakri baka:ra:

odar odra

'catafalque'

odri odarra:

nom. plur. gen. plur.

In the nom. sg. the Yer of the root morpheme surfaces as a, because it is followed by a Yer syllable. The same is true for the gen. plur. Before the remaining inflectional endings it is deleted because these suffixes do not contain a Yer. The gen. plur. is special in two respects. Firstly, it lengthens the nucleus of the preceding syllable. Secondly, its Yer is not deleted. In Hermans (1990) the latter fact is claimed to be a consequence of Compensatory Lengthening which bleeds Yer Devocalization and lengthens the Yer. For the present discussion Yer Devocalization is a crucial rule because it is able to eliminate stress bearing syllables. To see this, we must come to grips with the essentials of the inflectional accent system. In Serbo-Croatian two classes of roots can be distinguished. In the first class the root has the accent in a fixed position. In the second class the accent position is variable. In the nom. sg. it surfaces on the final syllable of the root; in the voc. ending it shows up on the initial syllable of the root; in all remaining cases it surfaces on the inflectional ending. Some examples are given in (9). Accented syllables are italicized. It should be noted that Serbo-Croatian has a rule which retracts all non initial accents one syllable to the left, resulting in a peculiar tonal contrast. To save space I have abstracted away from the effects of this rule. The accent notations thus have a fair degree of abstractness. For a discussion of the retraction phenomenon (the so called Neo S'tokavian Accent Shift) see Hermans (1990).

204

B. Hermans Fixed Accent

(9)

narrod narroda na:rode

nom. sg. gen. sg. voc. sg.

sa:bor sa:bora sa:bore

'people'

na:rodi na:ro:da:

'parliament'

sa:bori sa:bo:ra:

k/.'cos' ki.-cos'a k/xos'e

nom. plur. gen. plur.

kixos'i ki/cors'a:

Movable Accent z'ivot z'ivota z'/vote

'life'

z'ivoti z'ivo:to.·

svjedok svjedoka svjedoc'e

'witness'

svjedoc'i svjedo:ka:

Traditionally, it is assumed that roots which belong to the 'fixed accent class' are specified with an accent on the relevant syllable. The roots of the second class are accentless at the underlying level. In my thesis I follow this tradition. Furthermore I argue that underlying accents are specified with the features LD.-BND. In this approach the constrast between na:rod and z'ivot is expressed in the following way. (10)

underlying accent "LD-BND na-rod

no underlying accent

z ivot

In Serbo-Croatian the default construction rule inserts left dominant feet Inflectional endings are a systematic exception to this; they trigger right dominant constituents. This explains why in the nom. sg. masc. roots which belong to the 'movable class' have an accent on the final syllable. After the loss of the Yer, due to Yer Devocalization, the head of the right dominant constituent moves to the left. The vocative ending is exceptional in that it does not trigger an RD foot. Consequently the unmarked LD constituent is inserted.

On the fate of stray syllables

205

One of the most important claims of my thesis is the following: the construction of left dominant constituents is blocked by the SCC (Strict Cycle Condition). This condition is revised in such a way that it only blocks insertion of distinctive material. Distinctness is defined in terms of underlying constrasts. From this it follows that insertion of RD constituents is possible, whereas construction of LD constituents is blocked in the cyclic (but not in the postcyclic) component because they are constrastive in Serbo-Croatian (cf. the contrast in (10)). One last rule must be mentioned. In those cases where a sequence of accents is generated, all accents except the first are wiped out. Let me refer to this as the BAP principle. In (11) I give a highly informal version of this principle (see Hermans (1990) for a formal treatment). (11)

Basic Accentuation Principle Delete all accents except the first

The derivation of some crucial forms proceeds as follows. The gen. sg. naroda has an underlying LD foot in the base. The inflectional ending receives an RD foot. This is allowed by the SCC because it is a non distinctive operation. Due to the BAP the inflectional ending is deleted again. In the vocative narode the BAP does not need to apply because the construction of the LD foot is blocked by the SCC; the rule operates in a non derived environment and it carries out a distinctive operation. Likewise, no LD foot can be constructed in the first cycle of z'ivot (nom. sg.). On the second cycle an RD foot is constructed at the instigation of the nom. sg.. After Yer Devocalization the head moves to the left Hence, the final syllable of the root becomes accented. The voc. z'ivote is derived in the following way: On the first cycle construction of the LD foot is blocked by the SCC, because the rule operates in an underived (monomorphemic) environment However, on the second cycle the SCC allows construction of the LD foot, because now it operates in a derived (polymorphemic) environment Notice that we cannot account for the exceptional behaviour of the vocative ending by marking it extrametrical. If it were extrametrical, we would expect penult stress because inflectional endings trigger the construction of right dominant feet. We thus may conclude that Serbo-Croatian has all the necessary ingredients. It has a rule capable of deleting syllables; it also has non-extrametrical syllables. In principle we should therefore be able to answer the question: what happens in the representation given in (6)? Does the accent shift to the right, or is it deleted? In the next section I will supply the relevant data.

206

B. Hermans

4. UNDERLYING ACCENTS ON YERS

Consider the following paradigm. (12)

kosac kosca kos'c'e

nom. sg. gen. sg. voc. sg.

'reaper'

kosci kosa:ca:

nom. plur. gen. plur.

The distribution of the accent can fully be explained on the assumption that the base is accentless at the underlying level. Consequently, the BAP is inapplicable. Therefore, the inflectional endings receive accent. In the nom. sg. the Yer of the base is vocalized because it is followed by another Yer. After the loss of the final Yer the stress is shifted to the left, so that the vocalized Yer appears as accented on the surface. Compare the pattern in (12) with the following data (cf. also (8)). (13)

jec'am jec'ma jec'me

'barley'

jec'mi jec'arma:

bakor bakra txzkre

'copper*

bakri baka:ra:

odor odra odre

'catafalque*

odri oda:ra:

Most forms in (13) suggest that the base is accentless because the inflectional ending is stressed and not the root In the gen. plur., however, something strange happens. On the basis of this ending one would conclude that the root is stressed, rather than the inflectional ending. How can we reconcile these contradictory facts? The key to the solution is provided by the representative forms in (14).

On the fate of stray syllables (14)

207

jelen jelena

'deer'

jeleni jele:na:

konop konopa

'cord'

konopi kono.-pa:

kovc'eg kovc'ega

'chest'

kovc'ez'i kovc'e:ga:

podrum podruma

'cellar'

podrumi podrurma:

With respect to these examples most case endings point to the conclusion that the accent is immovably fixed to the second syllable of the base. However, the gen. plur. is special again. In the environment of this suffix some kind of accent retraction seems to take place; the stress shifts one syllable to the left This retraction is restricted to bisyllabic roots that are underlyingly accented on the second syllable and that have a short vowel in the initial syllable. Forms with a long vowel in this position do not undergo the rule (cf. the first two examples in (9)); neither do accentless roots undergo the rule (cf. the last two examples in (9)). For a formal treatment of the gen. plur. accent shift I refer to Hermans (1990). Meanwhile it suffices to say that the phenomenon operates on accents located on the second syllable of the root at the underlying level. Suppose now that we postulate an underlying accent on the second syllable in the forms in (13). In that case we can explain why all of a sudden an accent appears on the initial syllable in the gen. plur.. Notice that we now have a left dominant foot which is monosyllabic on the first cycle. The crucial question now is: what happens with the accent after the application of Yer Devocalization? Does it move to the right, or is it deleted? Recall that the traditional framework predicts that the accent is shifted to the right This is a consequence of the fact that stray syllables are adjoined, which enlarges the domain of the originally monosyllabic foot. In the inflectional paradigm stray syllables arise as a result of two phenomena. Firstly, the BAP deaccentuates inflectional syllables if they follow a base accent. In (13) all inflectional endings except the vocative undergo the BAP, because we are forced to postulate an underlying base accent. Secondly, in the vocative the SCC blocks construction of a left dominant foot Hence, this suffix remains stray. As far as the BAP rule is concerned, the adjunction theory makes the correct predictions. Stress does indeed shift to the right However, no stress shift is found in the vocative. Within the adjunction theory it cannot be explained why there is no shift to the right after the loss of Yer, because adjunction makes the preceding monosyllabic foot bisyllabic. A derivation using adjunction is given in (15) (the Yer syllable is represented by means of a dot).

208 (15)

B. Hermans I LD.-BND ο d.r * *

output of first cycle

I LD,-BND I ο d.re vocative cycle; construction of LD foot is BLOCKED * *

LD-BND ο d.re

adjunction

I LD.-BND I o dr e Yer Devocalization followed by * * automatic stress shift

This prediction is completely wrong. The underlying accent should not go to the right. In a theory without adjunction the facts in (13) are expected. Since there is no adjunction, a monosyllabic constituent remains monosyllabic even if it is followed by a stray syllable. After the loss of its head, due to Yer Devocalization, the entire constituent is removed. At the postlexical level, the constituent construction rules reapply. All inflectional endings, except the vocative, trigger the construction of RD feet. Hence, the accent shows up on the final syllable. However, the vocative ending triggers an LD foot Consequently, after the loss of the root accent, due to Yer Devocalization, the accent ultimately shows up on the initial syllable. The relevant part of the derivation of the vocative is given in (16).

On the fate of stray syllables (16)

209

I LD-BND I ο d.re output of vocative cycle; construction of * * * LD foot is blocked; there is no adjunction

ο dr e * *

Yer Devocalization; elimination of monosyllabic foot

/\ LD-BND o dr e * *

Reapplication of accent rules; insertion of LD foot at the instigation of vocative

5. CONCLUSION

In this paper 1 have shown that the theory of adjunction makes the wrong predictions when it is confronted with the theory of stress shift. This is caused by the fact that adjunction is able to enlarge the domain of a constituent (cf. the derivation of Arizona in (2)). The problems can be solved simply by assuming that there is no such thing as adjunction. Although I have primarily argued against a Hayesian framework, one should keep in mind that the argument of this paper is also relevant for other notations, like the one proposed by Halle and Vergnaud. Also in the bracketed grid framework one can adopt adjunction. Most recently, this is done in Kager (1989), for instance.

NOTE * Many thanks to Hans Verhulst for correcting my English.

REFERENCES: Benson, M. (1979) Serbocroatian-English Dictionary, Belgrade, Prosveta. Halle, M. and J.R. Vergnaud (1987) An Essay on Stress. MIT Press, Cambridge (Mass.). Hayes, B. (1981) A Metrical Theory of Stress Rules, distributed by Indiana University Linguistic Club, Bloomington Indiana. Hermans, B. (1990) The Phonological Interpretation of Mora Accent. A Theory of Stress and Accent based on Limburgian and Serbo-Croatian, PhD Diss., Free University, Amsterdam. Kager, WJ. (1989) A Metrical Theory of Stress and Destressing in English and Dutch. PhD Diss., University of Utrecht

210

B. Hermans

Rappaport, M. (1984) Issues in the Phonology of Tiberian Hebrew, PhD Diss., MIT, Cambridge (Mass.). Rubach J. (1986) "Abstract vowels in three dimensional phonology: the yers". The Linguistic Review 5.3.

Agreement and variables Teun Hoekstra University of Leiden

1. THE PROBLEM*

The problem I want to discuss involves participial agreement in impersonal constructions in Swedish and some varieties of Norwegian. I shall refer to del as the expletive, and to the "notional subject" as the correlate NP. We find participial agreement with the correlate NP in (la),1 but not in (Ib) and (Ic). (1)

a. b. c.

Del var ingen fuglar skotne/*skote It was no birds shot+AGR3pl Del var skote/*skotne mange fuglar Mange fuglar var det skote/*skotne

If agreement is explained in terms of some locality condition, e.g. to provide a local antecedent for the trace in order to avoid an ECP-violation, it is a mystery that long movement to sentence initial position, as in (Ic), does not trigger the same agreement as does the shorter movement in (la). A solution to this problem might be to assume that vaere in (la) takes a small clausal complement, which I shall take to be an AGR-phrase, while it takes a VP complement in (Ib-c), yielding the following structures: (2)

a. b. c.

det var [AGRP NP; AGR [w V t{ ]] det var [^ V NP] ΝΡ; var det [yp V /J

(la) (Ib) (Ic)

The postverbal trace in (2a) would then be an anaphor, locally Α-bound, while in (2c) it would be an Α-bar bound trace, i.e. a variable. Clearly, however, this solution really begs the question, rather than answering it, unless independent motivation can be given to support the different selectional properties of vaere. On the contrary, I will show that vaere takes an AGRP complement throughout First, however, I will briefly discuss the impersonal construction. Then I establish that vaere (and similarly bli in impersonal passives) takes an AGRPcomplement. Finally, I proceed to give an explanation for the problem in (1).

212

T. Hoekstra

2. DCT-IMPERSONALS Impersonal constructions in mainland Scandinavian languages (and dialects) usuallly feature the pronominal element DET. The sentences in (3)-(4) illustrate this use: (3)

a. b. c. d.

(4)

a. b.

Det it Det it Det it Det it

vart was var was bli was var was

telefonert telephoned snakket om Jon talked about John troet at Jon ikke var pä kontoret believed that John not was in the office kjöpt en hund bought a dog

Det it Det it

kommer en mann came a man starte en brann started a fire

(3) shows passive sentences: (3a) with no internal argument, (3b) with a PP internal argument, and (3c) with a tensed CP internal argument. In neither case is there an internal "Case-needy" NP, hence the "expletive" det may be inserted. We shall assume that det, being a pronominal element, "absorbs" the Nominative Case. A problem arises in (3d) and the ergative constructions in (4): there is an internal NP in these constructions, which ought to be Case-marked. A standard assumption is that the NP in these constructions enters into an expletive chain with det, which would then explain the defmiteness effect (i.e. the requirement that the postverbal NP be indefinite). The NP is Case-licensed through this expletive chain. This can either be regarded as Case-transmission, or, alternatively, as replacement of the expletive by the correlate NP at LF. I will return to this "expletive replacement" below. I want to argue that the indefinite NP does not receive Case by entering into a chain with the expletive. Rather, the NP is Case-marked directly by the verb (the same claim is made by Äfarli (1989), but he does not motivate it on the same grounds as I do). In this respect, the Norwegian system is like the French impersonal construction, which like Norwegian, features a pronominal "expletive", il, rather than an adverbial NP like English there. In the French ^-construction, the indefinite NP does not receive Nominative Case, as it does not determine the person/number agreement on the finite verb. I want to claim that agreement is always with the Nominative NP. This is true even in a language like Icelandic in sentences with so-called oblique subjects,

Agreement and variables

213

i.e. in sentences where the Nominative NP does not have the behavioral properties of "subjects". That the postverbal NP in French impersonate is assigned Case directly by the verb is also claimed by Pollock (1986). Norwegian provides two pieces of evidence that the postverbal NP receives Case from the verb. The first piece of evidence is the fact that in a sentence such as (5), the internal NP has the accusative form. (5)

Del vil altid vaere dem som sier noe annet it will always be those(ACC) who say something different

A more striking piece of evidence is found in sentences such as those in (6), which are mentioned in Askedal (1986). These sentences parellel those in (7), with a tensed clause instead of an infinitival complement. (6)

(7)

a. b.

Del ble pästott ä komme mange tilhörere it was claimed to come many listeners Det var ikke lett ä fä det til & komme sä mange tilhörere it was not easy to get it to to come so many listeners "it wasn't easy to manage to get many listeners to come"

a. b.

Det ble pästott at det kommer sä mange tilhörere Det var ikke let ä fä det til at det kommer sä mange tilhörere

Assuming that Nominative Case is not available in tenseless complements, the only way in which the postverbal NPs can pass the Case filter is for them to get Case directly from the verb. In this respect, then, the sentences in (6) are similar to those in (8), with a transitive (i.e. Case-marking) verb instead of an ergative verb. (8)

a. b.

Det it Det it

ble was var was

pästott ä ha repariert bilen claimed to have repaired the car ikke lett ä fä det til ä repariere bilen not easy to get it to to repair the car

It should be noted that the translation equivalents of (6) are totally impossible in languages such as German, Dutch and English, where the correlate NP must be assumed to receive Nominative Case, since it determines number agreement We conclude from this that the postverbal NP in tier-constructions in Norwegian receives internal Case, while det itself is marked with Nominative Case. I propose that V may always assign Case to an NP it governs if Nominative is "used", i.e. either assigned to some chain (of the external argument) or to a pronominal element such as Norwegian det or French //, leaving open the question of whether or not these pronominal elements receive some thematic

214

T. Hoekstra

role. With respect to Case assignment by the verb, then, det behaves as if it were an external argument.

3. DEK-IMPERSONALS

There are dialects which allow der instead of det in these impersonal constructions. I now want to argue that, although the difference is not visible on the finite verb, the postverbal NP in those cases is licensed by Nominative Case. The relevant evidence for this claim comes from those dialects that also have participle agreement. This agreement system is described and discussed in Christensen & Taraldsen (1989). We may summarize the description as follows: in clauses with auxiliary vaere ('be') and a participle, the participle agrees with the subject NP. The phenomenon of participial agreement is illustrated by the examples in (9): (9)

a. b.

Gjestene er komne guests-the are come+AGRpl Mange fuglar er skotne many birds are shot+AGRpl

I shall adopt, without further argumentation (cf. Hoekstra & Mulder 1989), the there-raising analysis proposed by Moro (1989). According to this analysis, there is not inserted, but rather raised from a predicative position in a small clausal complement to the verb be, i.e. the analysis of (lOa) is as in (lOb): (10)

a. b.

there is a problem there{ BE [sc a problem ij

Assuming further that predicate and subject in the SC share features for person, number, agreement and Case, we have an immediate account of why in the agreeing dialects, raising of der triggers the same agreement as raising of the subject NP. This is shown in (11). (11)

a. b.

Der There Der There

er are er is

komne come+AGRpl skotne mange shot+AGRpl

nokre gjester some guests fuglar birds

If in these dialects det fills the subject position, there is no agreement with the correlate, with an exception to which we return immediately.

Agreement and variables

215

The agreement facts follow if we accept that vaere takes an AGRP-complement, i.e. if the structure of (9) is as in (12). (12)

NPi vaere [AGRP t AGR [yp V fj]

As the structure indicates, the NP moves through the SPEC of an AGR-phrase on its way to the subject position. It is the intermediate trace that determines the agreement on the participle. Movement of der in the sentences in (11) will likewise go through the SPEC of the AGR-phrase. The intermediate step is required, since direct movement is impossible because a barrier intervenes,2 and an escape by means of adjunction would yield a violation of the improper movement constraint. Interestingly, no agreement is found if, instead of vaere, the auxiliary ha 'have' is used, which is possible in Norwegian with ergative verbs (i.e. in (9a) and (lla)). This would follow if ha takes a VP-complement, whereas vaere takes an AGRP-complement. This supports the claim made in section 1.

4. SUPINE AND ΡΑΚΉΟΡίΕ

The absence of participial agreement with the postverbal NP when del is used instead of der suggests that det does not raise in the same way.3 From our perspective, that is what we would expect: der is a locative adverbial which can function as a predicate, but det is pronominal, and is therefore unfit to be a predicate. However, if we maintain that vaere always takes an AGRP-complement and that det is inserted in [SPEC.IP] to meet the extended projection principle, the question arises as to whether or not the SPEC of its AGRP-complement is empty in def-constructions such as (Ib). This in turn raises the question of how the postverbal NP can receive Case, since now the verb would not seem to take an external argument (not even a quasi-argument like det). I would like to argue that the SPEC of vaere's AGRP-complement is not empty, but rather contains a trace of det, i.e. the structure of (Ib) is as in (13): (13)

det, var [AGRP t{ AGR [^ V NP]]

In def-constructions, then, the participle agrees with det itself, something which is possible for det, as it is an inherently pronominal element, but not for der, as it lacks the inherent properties to determine agreement Put differently, a participle can only agree with the NP that der is correlated with. The relevant evidence for the claim that the participle agrees with det in del-constructions comes from Swedish (as well as certain Norwegian dialects), where a distinction is made between a supine, used in the complement of ha, and a participle,

216

T. Hoekstra

used with vaere, among others. This constrast is displayed in (14), taken from Christensen & Taraldsen (1987): (14)

a. b. c.

Jens har skjutrt ett lejon Jens has shot a lion Lejonet blev skjutef lion-the was shot+AGR3Nsg Djuren blev skjut/ia animals-the were shot+AGR3pl

The relevant point is that in Swedish ^-constructions, we do not find the supine form, which we would expect if we were dealing with a VP complement Rather, we find the participial form, taking neuter third person singular agreement, irrespective of the number and gender of the correlate NP, as is shown in (15). This suggests that the structure of (15) is as in (13). (15)

Del blev skjutef/*skjutif en älg/ett lejon/djur it was shot+AGR3Nsg/shot an elk/a lion/animals

This conclusion lends further support to our claim that vaere always takes a AGRP-complemenL

5. OPERATORS AND VARIABLES

At this point, we might think we have solved the problem illustrated in (1), by positing the structures in (16): (16)

a. b. c.

det var [AGRP NP; AGR [w V 4 ]] det var [AGRP i; AGR [yp V NP]] NPi var detj [AGRP t} AGR [w V ij]

(la) (Ib) (Ic)

The structure in (16a), however, is not the correct structure for (la), as becomes apparent if we undo the disturbing effect of Verb Second. The embedded clause variant of (la) is given in (17). (17) immediately shows that ingen fuglar cannot be in the SPEC of the complement of vaere. Nevertheless, the agreement of the participle suggests that the NP does bind a trace in that SPECposition, yielding a representation of (17) as in (18). (17)

at det ingen fuglar var skotne that it no birds were shot+AGR3pl

(18)

at det NP{ vaere [AGRP ^ AGR [yp skotne ij]

Agreement and variables

217

The position of the NP ingen fuglar results from a process, studied by Christensen (1989), which I shall call wgen-movemenL The pre-VP negative adverbial ikke may optionally attract an indefinite NP, and yield the form ingen N, i.e. (17) is derived from (19) by the process of i'n^€«-movement (cf. Christensen 1986,1989 for further details). (19)

at det ikke var skote noe fuglar that he not has shot any birds

The agreement is again a consequence of vaere taking a AGRP-complement, as is evident from the contrast with (20), where the NP also precedes the participle, but without triggering agreement. This follows from our previous assumption that ha, unlike vaere, takes a VP complement (cf. the absence of agreement in der-sentences in section 3). (20)

at ho ingen fuglar har skote that he no birds has shot

The problem in (1) can now be phrased as follows. Assuming that ingen fuglar in (la) and mange fuglar in (Ic) both occupy an Α-bar position, i.e. both have the basic structure in (21), why does ingen fuglar have to go through the [SPEC, AGRP], and why can't mange fuglar move through it? (21)

Α-bar [AGRP np* AGR [yp V e ]]

If we define the notion of variable as being locally Α-bar bound, it would follow that e is a variable in (Ic), but not in (la), since in the latter, e would be bound from np*, an Α-position. This conclusion is only correct for (Ic) if np* could not itself be a variable, with e being an anaphor, bound by np* (as in "Who np seems e to be winning"). We can avoid the possibility of np in (21) being a variable, if it is furthermore required that a variable be Casemarked so that the definition of variable is as in (22): (22)

α is a variable iff locally Α-bar bound and in a Case-marked position

The suggestion that np is not Case-marked is independently motivated by the ungrammaticality of (23): (23) *Jeg sp rte om det var [AGRP I asked whether it was

mange fuglar skote/skotne] many birds shot

If (22) is correct, it enforces our earlier conclusion that the correlate NP in detconstructions receives Case in situ from the governing verb. We now have an

218

T. Hoekstra

explanation for the impossibility of agreement in (Ic). This account raises questions about (la), however. If (22) is correct, it implies that e in (la) is not a variable. Rather, it should be an anaphor, being locally Α-bound, from the position np. The question we then face concerns the status of np in (la): it is bound from an Α-bar position and hence should be a variable, according to standard assumptions. But then (22) would require that the position be Casemarked, contrary to what we claimed before. Clearly, t'ngen-movement is subject to a locality requirement which is typical of Α-chains, rather than Α-bar chains. This is evident if we look at a construction such as (24), which is ungrammatical: (24) *at jeg ingen fuglar har sagt at del var skote/skotne i dag that I no birds have said that it was shot to day There are many instances of such "local" Α-bar chains. Clitic chains constitute one case, Q-movement constructions like "Jean a beaucoup lu de livres" constitute another. In those cases, the local nature of these Α-bar chains might be captured if appeal is made to the head status of the leftward-shifting elements, head movement being subject to locality requirements that are identical to Α-movements. However, in the case of "ingen"-movement we are not dealing with head movement. Another example of the same sort is the leftward shift of objects and R-pronouns in Dutch, which, in so far as non-wh forms are concerned, are similarly "clause" bounded. An example is given in (25): (25)

a.

dat Jan had gedacht dat ik er met Piet [over e] zou praten that John had thought that I there with Peter about would talk b. *dat Jan er had gedacht dat ik met Piet [over e] zou praten

What appears to be crucial here is not the Α-bar status of the binder of the empty category per se, but rather that a distinction must be made between two types of Α-bar binders: those that are in [SPEC.CP] and those that are in other Α-bar positions - specifically those that are adjoined to VP. Such a distinction has been proposed in several places in the literature (cf. Kayne 1984, Taraldsen 1986, Holmberg 1986), viz. the distinction between operator binding and nonoperator binding. So, the definition of variable in (22) should be replaced by that in (26): (26)

α is a variable iff locally operator bound and in a Case-marked position

Assuming that the ingen-NP is a non-operator, e in (la) could not be a variable. According to Chomsky's (1982) functional approach to empty categories, a non-variable empty category is an anaphor; i.e. e of (21) would be an anaphor

Agreement and variables

219

in (la), but a variable in (Ic). Being an anaphor, it must be locally bound, hence the agreement on the participle. Two questions remain at this point: a. when is an Α-bar binder an operator? I suggested that this might be determined by the Α-bar position, [SPEC, CP] being an operator position. There are reasons to doubt the correctness of this assumption, but it would take too long to go into that here. I conclude that a clear definition of operator is lacking at this point, a topic for future research. b. if an empty category is an anaphor because of failure to meet the requirement of being operator-bound, does it follow that it also cannot have Case? This again is a complicated issue, that cannot be settled in this paper.

7. CONCLUSIONS

The agreement pattern in (1) raises interesting descriptive and theoretical questions. The descriptive problem was solved by postulating an AGRP-complement to vaere in all cases, where the occupant of [SPEC, AGRP] determines agreement of the participle. We saw that ingen fitglar at some point in the derivation passed through this specifier. The theoretical question then becomes: why is such an intermediate step required in the case of (la), but impossible in (Ic)? I argued that a solution to this problem can be found in the definition of variable, which mentions two requirements of a variable: it must be in a Caseposition, and it must be locally operator bound. Assuming ingen fuglar not to be an operator then yields the correct results for (la), while (Ic) is explained if mange fulglar is an operator.

NOTES * Both the problem and my insights into it owe a great deal to Kirsti Koch Christensen. I thank her for kindly discussing these issues with me. I also would like to thank Amfinn Vonen for talking with me about these matters, as well as the students in my class in Oslo in the Fall of 1989. 1. Sentences of the type (la), the result of /«gen-movement, have a fairly marginal status. Participial agreement is found in Swedish, Nynorsk and certain Norwegian dialects. I have drawn my data mainly from the works of Christensen and Christensen & Taraldsen, with some additions provided by native informants. 2. I will not go into the question as to what constitutes a barrier in this case. We may either assume the AGRP to inherit barrierhood from VP, or adopt a version of minimality, either Chomsky's (1986a) or Rizzi's (1988) relativized minimality ([SPEC.AGRP] being an intervening Α-position). A choice between these alternatives is not relevant at this point

220

T. Hoekstra

3. More specifically, if agreement is forced as a way to escape minimality effects, i.e. it is required to meet the ECP, and ECP is a condition on LF, the absence of agreement suggests that there could not be something like expletive replacement at LF (unlike what Chomsky 1986b suggests).

REFERENCES Afarli, T. (1989) The syntax of Norwegian passive constructions, Trondheim diss. 1989. Askedal, J.O. (1986) "On ergativity in Modem Norwegian", Nordic Journal of Linguistics 9, 25-45. Chomsky, N.A. (1982) Some concepts and consequences of the theory of government and binding, MIT Press, Cambridge (Mass.). Chomsky, N.A. (1986a) Barriers, MIT Press, Cambridge (Mass.). Chomsky, N.A. (1986b) Knowledge of language: its nature, origin and use, Praeger, New York. Christensen, K.K. (1986) "Norwegian ingen: a case of postsyntactic lexicalization", in O. Dahl and A. Holmberg (eds.), Scandinavian syntax. University of Stockholm. Christensen, K.K. (1989) Partisipkongruens, A'-argumenter og HV-flytting, ms. University of Bergen. Christensen, K.K. and T.K. Taraldsen (1987) "Expletives, chains and agreement", in V. Rosen (ed.), Papers from the 10th Scandinavian conference of linguistics, Dept of Linguistics and Phonetics, University of Bergen, 116-131. Christensen, K.K. and T.K. Taraldsen (1989) "Expletive chain formation and past participle agreement in Scandinavian dialects", in P. Beninca (ed.), Dialectology, Foris, Dordrecht Hoekstra, T. and R. Mulder (1989) Unergatives as copular verbs: locational and existential predication, ms. University of Leiden. Holmberg, A. (1986) Word order and syntactic features, diss. University of Stockholm. Kayne, R. (1984) Connectedness and binary branching, Foris, Dordrecht. Moro, A. (1989) tnerelci as raised predicates, ms. MIT. Pollock, J.-Y. (1985) "On case and the syntax of infinitives in French", in J. Gueron, H.-G. Obenhauer and J.-Y. Pollock (eds.), Grammatical representation, Foris, Dordrecht, 293-248 Rizzi, L. (1988) Relativized minimality, ms. University of Geneva. Taraldsen, K.N. (1986) "Som and the binding theory", in L. Hellen and K.K. Christensen (eds.), Topics in Scandinavian syntax, Reidel, Dordrecht

Assumptions about asymmetric coordination in German Tilman N. Höhle University of Tübingen

1. SYMMETRIC AND ASYMMETRIC COORDINATION

Consider an S-structure configuration containing a coordinate structure such as (1): (1)

"A ... kA [(&) 'B ... & "B]

k+1

A ... »A

where each *B is a conjunct, each & is a coordinating particle such as und 'and', and each 'A is an element external to the conjuncts (m > 0; > 2). The fundamental principle of coordination that I will assume is (2): (2)

Conservation Condition Each *B is a constituent (i) whose structure and whose combinatorial properties follow from general rules that are independent of coordination, or (ii) which conforms to the coordination scheme (1).

Prototypical coordinate structures are symmetric in the sense that they comply with some principle like (3): (3)

External Homogeneity Condition The combinatorial properties of each >B are satisfied by !A, ... , mA in the same way as the combinatorial properties of every JB are.

I will not attempt here to make this rather vague statement precise. Under appropriate specifications it should follow from (3) that in the typical case all conjuncts are members of the same syntactic category and that each conjunct stands in the same grammatical relations to the external elements just as every other conjunct. Specifically, the Coordinate Structure Constraint with its 'Across-The-Board exception' (4) should follow from (3): (4)

CSC/ATB: If there is an 'A in a nonA-position that binds a trace in one JB, it binds a trace in every kB.

222

T.H. Höhle

It follows, then, that each single conjunct 'B may be substituted for the whole constituent "(&) 1E ... "B" salva grammaticalitate. See Neijt (1979: ch.l) and Sag et al. (1985) for detailed discussion of symmetric coordination.1 There are, however, certain types of coordinate structures in German that are asymmetric in that they do not comply with (3) and its corollaries. Compare, e.g., (5a) and (6a): (5)

a.

b. c. (6)

wenn [[jemand nach Hause kommt] und [da when someone to home comes and there der Gerichtsvollzieher vor der Tür steht]], ... the bailiff at the door stands wenn [jemand nach Hause kommt], ... wenn [da der Gerichtsvollzieher vor der Tür steht], ...

a.

wenn [[jemand nach Hause kommt] und [da steht der when someone to home comes and there stands the Gerichtsvollzieher vor der Tür], ... bailiff at the door b. *wenn [da steht der Gerichtsvollzieher vor der Tür], ...

The coordination in (5a) is symmetric; and consequently the first conjunct (5b) or the second conjunct (5c) can be substituted for the whole coordinate structure. The coordination in (6a) is asymmetric; hence, substituting the second conjunct for the coordinate structure, as in (6b), is not possible. There is also another type of asymmetric coordination. Compare (7a) and (8a): (7)

(8)

a.

wenn jemand [[nach Hause kommt] und [den Gerichtsvollzieher when someone to home comes and the bailiff sieht]], ... sees b. wenn jemand [den Gerichtsvollzieher sieht], ... a. wenn jemand [[nach Hause kommt] und [sieht den when someone to home comes and sees the Gerichtsvollzieher]], ... bailiff b. *wenn jemand [sieht den Gerichtsvollzieher], ...

The coordination in (7a) is symmetric, and the second conjunct can be substituted for the whole coordinate structure, as in (7b). The coordination in (8a) is asymmetric, and the coordinate structure cannot be replaced by the second conjunct

Assumptions about asymmetric coordination in German

223

In this contribution I will sketch a set of assumptions that seem jointly to be able to account for the major syntactic properties of asymmetric coordinations such as (6a) and (8a).2

2. GERMAN CLAUSE STRUCTURE

Traditionally three major typological types of clauses are distinguished according to the position of the finite verb. Embedded clauses typically (but not invariably) conform to the scheme (9): (9)

CMP - X - VK

where CMP corresponds to what used to be called the 'COMP-position' in English. The finite verb as well as infinite verbs (if any) are in VK. All other elements of the clause (if any) are in X. (Extraposition will be disregarded throughout.) Ordinary declaratives and direct wh-interrogatives deviate from (9) in the way indicated in (10), where FINIT is the position of the finite verb, and K is a wh-phrase in the case of interrogatives and some other kind of phrase in the case of declaratives: (10)

K - FTNIT - X - VK

Direct polar interrogatives, certain conditionals and concessives, and some other functional types differ from (10) in that they lack a K-position: (11)

FINIT - X - VK

To fix terminology, I dub clauses of type (9) Ε-clauses' ("E" for 'elementary'); those of type (10), 'FZ-clauses'; and those of type (11), Tl-clauses'. Fl-clauses and F2-clauses are 'F-clauses', i.e., clauses where the finite verb is fronted. As to the categorial structure of clauses, I will assume that FINIT can always be identified with 1°, in one of the current conceptions of I, and that the base position of the subject is contained in Vm. In German this position can be case-marked. The abstract structure of clauses then is as given in (12). Examples are analysed accordingly in (13). (12)

a. b. c.

Ε-clauses: [cm CMP [vm X VK ]] F2-clauses: tf Κ [,1 1° [vm X VK ]]] Fl-clauses: [,1 1° [vm X VK ]]

224

T.H. Höhle

(13)

a. b. c.

[cm [ mit wem]; [vm Karl gestern ^ gesprochen hat]] with whom yesterday spoken has [r2 [mit wem]; [j1 hatj [vm Karl gestern it gesprochen [j1 hatj [vm Karl gestern mit dir gesprochen /;]] has yesterday with you spoken

Some of these assumptions will be modified later on. Following arguments by Safir (1984), I will assume that the subject position of the clause is obligatory. This implies that German has an expletive pro, as in (14): (14)

[cm daß [vm pro vielen Leuten geholfen wurde ]] that many people (DAT) helped was

I furthermore assume that the K-position in (12b) is neither governed nor casemarked. It follows correctly that ordinary declarative variants of (14) can appear as (15a-c), but not as (15d), since pro must be governed (and case-marked) in S-structure: (15)

a. [j2 b. [j2 c. [j2 d. *[j2

[vielen Leuten]; [j1 wurdej [vm pro ^ geholfen /j]]] Geholfen [j1 wurdej [vm pro vielen Leuten t{ tfi] Es [j1 wurdej [vm pro vielen Leuten geholfen fj]] prOj [j1 wurdej [vm fj vielen Leuten geholfen ij]]]

(The es in (15c) is an expletive particle whose occurrence is restricted to the Specl-position.) With this terminology at hand we can describe asymmetric coordination in slightly more detail. In (16) (= (6a)) the first conjunct is a Vm. The second is an F2-structure, hence an I2: (16)

Wenn [^ [vm jemand nach Hause kommt] und [j2 da [j1 stehtj [vm der Gerichtsvollzieher vor der Tür

This is unusual in two ways: First, wenn usually must introduce -clauses, cf. (6b). Second, conjuncts typically are of the same category type. Coordinate structures whose conjuncts are of the same category are symmetric. Following traditional insights (e.g., Hockett 1958), I assume that in symmetric coordination each conjunct is a head of the coordinate structure. Making the natural complementary assumption for asymmetric coordination, I suggest that only the normal conjunct is a head of the coordinate structure, whereas the asymmetrically added second conjunct is a non-head. Since the category of the head and the category of the head's mother are identical, I will assume that in (16), = Vm. Hence the combinatorial properties of wenn are satisfied in the way we

Assumptions about asymmetric coordination in German

225

would expect them to, in that wenn here is a sister of Vm. Coordinate structures with a non-head F2-conjunct I will call (asymmetric) F2-coordination. In (17) (= (8a)) the first conjunct is a V1, hence α = V1: (17)

Wenn [vm jemand [„ [γ1 nach Hause kommt] und [ sieht; [vm da den Gerichtsvollzieher

The second conjunct is similar to an F-clause, but it lacks the subject that is required by the predicate. It is not obvious whether β is I1 or I2 (or something else). Coordinate structures with a non-head conjunct of this kind I will call SLF-coordination ('Subject Lacking in F-structure').

3. ASYMMETRIC Fl-COORDINATION

In (16) an F2 conjunct occurs in an Ε-clause. This construction is very common with conditional we/in-clauses and temporal a/i-clauses. With other types of Ε-clauses it is less common, although there does not seem to be any type where it is strictly excluded. Asymmetric F2-coordination is also very common with conditional Fl-clauses, as in (18a), and similar types: (18)

a.

b.

kommst du nach Hause und da steht der Gerichtsvollzieher come you to home and there stands the bailiff vor der T r, ... at the door [j1 komms^ [vm [vm du nach Hause i;] und [i2 da steht der Gerichtsvollzieher vor der T r]]]

I will assume that the coordinate structure of (18a) is basically identical with the structure of (16), i.e., I2 conjoined with Vm, as shown in (18b). Notice that the first conjunct - but not the second - contains a trace nonA-bound by the finite verb kommst, thus violating the CSC/ATB (4). We expect this to be possible, because (4) is a corollary of the External Homogeneity Condition (3), adherence to which would constitute a defining property of symmetric coordination, not of asymmetric coordination. With many speakers, asymmetric F2-coordination can also be observed with F2-clauses, as in the interrogative (19a): (19)

a. b.

Wann holst du die Fahrkarten und Heinz packt sein Zeug ein? when get you the tickets and packs his stuff up [,2 wannj [j1 holst; [vm [vm du t} die Fahrkarten fj und [j2 Heinz packt sein Zeug ein]]]]

226

T.H. Höhle

At this point one may ask why it is possible to conjoin I2 and Vm. Given the fact that it is possible, in principle, to asymmetrically conjoin categories of different types, we certainly expect this possibility to be restricted in accordance with some general principle. As a minimal restriction I assume that for any kind of coordination the constituents to be conjoined must be functionally similar with respect to their degree of saturation. In German, Vm and I2 are completely saturated in the sense that they are complete functional complexes. In (16) this is visibly true. In (18) and (19) it is true under the assumption that nonA-bound traces as they appear there are evaluated, for the purpose under discussion, in the way that overt linguistic expressions are. If these assumptions are correct, one may try and substitute a Vm without traces for I2 in (16), (18) and (19). For (16) the result of substitution is, of course, an ordinary symmetric coordination, i.e., (5a). For (18a), the result is (20): (20)

(21)

*kommst du nach Hause und da der Gerichtsvollzieher vor der Tür steht, ... a. b.

[ß kommst [vm [vm du nach Hause /J und [vm da der Gerichtsvollzieher vor der Tür steht]]] [ß [ß kommst du nach Hause] und [vm da der Gerichtsvollzieher vor der Tür steht]]

Why is (20) impossible? Consider some candidate structures for (20). In structure (21a) a Vm is conjoined with a V"1. Conjunction of like categories is a sufficient condition for symmetric coordination, and as such (21a) would have to comply with the CSC/ATB, which it does not. This is as it should be. But it seems that (21b) must be regarded as being wellformed, according to our assumptions. If it were, we would not have an account for the unacceptability of (20). Intuitively speaking, (21b) seems to be incorrect because ß (ß = I1) should be a full clause, which should not be able to be conjoined with Vm, a non-clause. If so, we have to specify assumptions that enforce this result. To this end I would like to adopt some suggestions made by Kathol (1989). Modifying and extending ideas of Fukui & Speas (1986) and Travis (1988), he proposes the set of assumptions (22): (22)

(i) Vm is a sister of 1°. In German, 1° is to the left of its sister, (ii) In German, 1° is empty if and only if I1 is a sister of C°. (iii) In German, 1° contains a lexical element if and only if there is a Specl, i.e., I1 projects to I2.

From (22i, ii) it follows that -clauses must contain an empty 1°, and from (22iii) it follows that in Fl-clauses the finite verb is in C°, with 1° empty.

Assumptions about asymmetric coordination in German

227

(12a,c) must then be replaced by (23a,b): (23)

a. b.

-clauses: [cm CMP [,1 1° [vm X VK]]] Fl-clauses: [cm C° [j1 1° [vm X VK]]]

I will assume, furthermore, that the special semantic properties of Fl-clauses - i.e., their being conditionals, direct polar interrogatives, etc. - are associated somehow with the verb being in C°. Under these assumptions (21b) is impossible, as it does not comply with (22i). We have now to consider two new structures for (20): (24)

a. b.

[c1 kommstj [j1 [j1 e{ [ym du nach Hause fj] und [^ e [vm da der Gerichtsvollzieher vor der Tür steht]]]] [c! [c1 kommst du nach Hause] und [j1 e [vm da der Gerichtsvollzieher vor der Tür steht]]]

Structure (24a) is ungrammatical as it violates the CSC/ATB. (24b) does not comply with (22ii), thus accounting for the intuition that the conjunction of the non-clause I1 with the full clause C1 is the source of this structure's ungrammaticality. With (21) and (24) the set of potential candidate structures for (20) is not exhausted, of course, but one can easily see how other structures fail to be grammatical. Basically the same results obtain when Vm is substituted for I2 in (19): (25)

a. *Wann holst du die Fahrkarten und Heinz sein Zeug einpackt? b. waniii [j1 hols/j [vm [ym du t{ die Fahrkarten /j] und [vm Heinz sein Zeug einpackt]]] c. wannj [j1 holsfj [vm [vm du t{ die Fahrkarten ij] und [j1 e [vm Heinz sein Zeug einpackt]]]] d. warnij [j1 [j1 holst du f; die Fahrkarten] und [j1 e [vm Heinz sein Zeug einpackt]]]

Structure (25b) is a CSC/ATB violation. Both (25c) and (25d) are in violation of (22ii). In addition, (25d) is a CSC/ATB violation.3 Much like an F2-clause (i.e., an I2) and a Vm, a full Fl-clause constitutes a complete functional complex. When we substitute an Fl-clause for I2 in asymmetric coordinations, we get examples such as (26):

228

T.H. Höhle

(26)

a. *als Karl nach Hause kam und stand da sein Vater when to home came and stood there his father vorder Tür, ... at the door b. kommt Karl nach Hause und steht da sein Vater vor der Tür, ... c. Wann holt Karl die Fahrkarten und packt Heinz sein Zeug ein?

None of these examples is a wellformed asymmetric coordination. An Fl-conjunct within an -clause as in (26a) is strongly unacceptable. An Fl-conjunct within an Fl-structure as in (26b) is unobjectionable; but this is an ordinary (symmetric) coordination of two conditional Fl-clauses, just as (27) is a (symmetric) coordination of two conditional wenn-clauses:4 (27)

[[wenn Karl nach Hause kommt] und [wenn da sein Vater vor der when to home comes and when there his father at the Tür steht]], ... door stands

The structure of (26c) is an ordinary symmetric coordination of two I^conjuncts, as shown in (28): (28)

Wannj [j1 [j1 holt Karl ^ die Fahrkarten] und [^ packt Heinz t{ sein Zeug ein]]

There is no reason whatsoever for assuming an asymmetric coordinate structure for (26c). Why is it that full Fl-clauses, as opposed to F2-structures such as (16), (18), and (19), cannot be asymmetrically conjoined with Vm? The reason, I propose, is the special semantics associated with the structure of Fl-clauses. When we try and give (26) analyses in accordance with (23b), the second conjunct will always be a C1. That C1 conjunct must receive the interpretation of a conditional (or a direct polar interrogative, and so on), but there is of course no way of integrating this interpretation into the containing structure of (26a) and (26c). The only way of integrating it into (26b) is by assuming that the first conjunct is a C1. There are exactly two possible structures for full Fl expressions: they can be analysed as a C1 clause, as in (26b), or as an I1 which according to (22iii) must project to I2, as in (28). Hence there is no asymmetric Fl-coordination.

4. SLF-COORDINATION

In (29) (= (17)) an SLF-conjunct occurs in an E-clause:

Assumptions about asymmetric coordination in German (29)

a. b.

229

wenn jemand nach Hause kommt und sieht da den Gerichtsvollzieher when someone to home comes and sees there the bailiff [cm wenn [j1 e [vm jemand [y1 [y1 nach Hause kommt] und [ß siehtj tvm da den Gerichtsvollzieher ij]]]]]

Just like asymmetric F2-coordination, SLF-coordination also occurs in Fl-clauses, äs in (30), and in F2-clauses, äs in (31): (30)

a. b.

(31)

a.

b.

kommst du nach Hause und siehst da den Gerichtsvollzieher, ... come you to home and see there the bailiff [c! kommst; [j1 ei [vm du [y1 [y1 nach Hause fj] und [ß siehsij [vm da den Gerichtsvollzieher /j]]]]]] Hoffentlich kommt keiner nach Hause und sieht da den I hope comes no one to home and sees there the Gerichtsvollzieher, bailiff [j2 hoffentlich [j1 kommt, [vm keiner [y1 [y1 nach Hause ij und [ß siehfj [vm da den Gerichtsvollzieher ij]]]]]]

The structures given under (b) are based on the assumption that constituents can only be (symmetrically or asymmetrically) conjoined if they are similar with respect to their degree of saturation. The ß conjunct in each case is unsaturated with respect to its grammatical subject; hence, I assume, it must be conjoined with a constituent that is similarly unsaturated, i.e., with V1. The assumption that the coordinate structure as a whole is a V1 is confirmed by recursive embedding of coordinate structures such as (32): (32)

a.

b.

Wenn jemand [„ in die Wüste zieht und lebt dort von when someone into the desert moves and lives there off Heuschrecken] locusts oder [