Studies on Scrambling: Movement and Non-Movement Approaches to Free Word-Order Phenomena [Reprint 2011 ed.] 9783110857214, 9783110135725

Citation preview

Studies on Scrambling

Studies in Generative Grammar 41

Editors

Jan Köster Henk van Riemsdijk

Mouton de Gruyter Berlin · New York

Studies on Scrambling Movement and Non-Movement Approaches to Free Word-Order Phenomena

Edited by

Norbert Corver Henk van Riemsdijk

Mouton de Gruyter Berlin · New York

1994

Mouton de Gruyter (formerly Mouton, The Hague) is a Division of Walter de Gruyter & Co., Berlin.

The series Studies in Generative Grammar was formerly published by Foris Publications Holland.

© Printed on acid-free paper which falls within the guidelines of the ANSI to ensure permanence and durability. Library of Congress Cataloging-in-Publication

Data

Studies on scrambling : movement and non-movement approaches to free word-order phenomena / edited by Norbert Corver, Henk van Riemsdijk. p. cm. — (Studies in generative grammar ; 41) "The present collection of articles grew out of a workshop on scrambling which took place at Tilburg University in October 1990"Introd. Includes bibliographical references and index. ISBN 3-11-013572-8 1. Grammar, Comparative and general—Word order. 2. Generative grammar. 3. Grammar, Comparative and general—Clauses. I. Corver, Norbert, 1 9 6 3 . II. Riemsdijk, Henk C. van. III. Series. P295.S748 1994 415—dc20 94-14409 CIP

Die Deutsche Bibliothek — Cataloging-in-Publication

Data

Studies on scrambling : movement and non-movement approaches to free word-order phenomena / ed. by Norbert Corver ; Henk van Riemsdijk. — Berlin ; New York : Mouton de Gruyter, 1994 (Studies in generative grammar ; 41) ISBN 3-11-013572-8 NE: Corver, Norbert [Hrsg.]; GT

© Copyright 1994 by Walter de Gruyter & Co., D-10785 Berlin. All rights reserved, including those of translation into foreign languages. No part of this book may be reproduced in any form or by any means, electronic or mechanical, including photocopy, recording, or any information storage and retrieval system, without permission in writing from the publisher. Printing: Werner Hildebrand, Berlin. Binding: Dieter Mikolai, Berlin. Printed in Germany.

Contents

Norbert Corner — Henk van Riemsdijk Introduction: approaches to and properties of scrambling

1

Josef Bayer — Jaklin Kornfilt Against scrambling as an instance of Move-alpha

17

Marguerite Browning — Ezat Karimi Scrambling to object position in Persian

61

Viviane Deprez Parameters of object movement

101

Daniel L. Finer On the nature of two A'-positions in Selayarese

153

Kenneth Hale Core structures and adjunctions in Warlpiri syntax

185

Katalin E. Kiss Scrambling as the base-generation of random complement order

221

Young-Suk Lee — Beatrice Santorini Towards resolving Webelhuth's paradox: evidence from German and Korean

257

Anoop Mahajan Toward a unified theory of scrambling

301

Gereon Müller — Wolfgang Sternefeld Scrambling as Α-bar movement

331

Ad Neeleman Scrambling as a D-structure phenomenon

387

Urpo Nikanne Notes on movement to the Spec(IP) position in Finnish

431

Ayumi Ueyama Against the A/A'-movement dichotomy

459

Sten Vikner Scandinavian object shift and West Germanic scrambling

487

Index

Introduction: approaches to and properties of scrambling Norbert Corver and Henk van Riemsdijk

The present volume deals with the issue of "scrambling", the phenomenon of variable word order within a clause. Ross (1967), who was one of the first to discuss this phenomenon within the generative paradigm, attributed the freedom of constituent order in languages such as German, Latin and Russian to the existence of a stylistic reordering rule, which stated that two adjacent constituents can be permuted if they are clause-mates. Ever since Ross's initial discussion of scrambling, various alternative analyses have been proposed within generative grammar to account for the occurrence of alternate word orders in various natural languages. In this introduction, we will give a brief overview of the various approaches towards scrambling (cf. also E. Kiss (this volume) and Grewendorf — Sternefeld 1988), and discuss several issues related to the free word order phenomenon which will be dealt with more extensively in the various contributions to this volume. A basic dichotomy in the current approaches towards scrambling is that between a movement approach versus a base generation approach. According to the former, there is one underlying word order and the variety of alternate word order arrangements in a clause is the result of movement. The latter approach claims that there is not one basic order of constituents and that the variable word order is the result of free generation of constituents in an arbitrary order at D-structure. In other words, free constituent order is a phenomenon already present at Dstructure (cf. Neeleman, Bayer — Kornfilt). The difference between the two approaches can be illustrated by means of the following examples from German: (l)a.

...weil ...because

Hans wahrscheinlich das Buch Hans probably that book

gekauft bought

hat has b.

...weil Hans das Buch wahrscheinlich gekauft hat

2 Norbert Corver and Henk van Riemsdijk

(2)a.

b.

...weil niemand dieses Buch gekauft ...because nobody this book bought

hat has

...weil dieses Buch niemand gekauft hat

The movement analysis considers (la) and (2a) to be the D-stnicture constituent orders, from which the inverted S-structure orders "Direct object - Adverb" (lb) and "Direct object - Subject" (2b) respectively are derived via "Move a". The base-generation analysis generates both constituent orders at the level of D-structure. In other words, the major constituents do not have a fixed syntactic position at D-structure. A central assumption in the base generation approach to scrambling is that theta assignment and case assignment do not universally presuppose adjacency between assigner and assignee (See Bayer — Kornfilt, E, Kiss, and Neeleman). That is at D-structure a direct object NP can receive a theta role from a verb when a subject as in (2b) or an adjunct as in (lb) intervenes. Hence, the non-adjacent direct object NP still occupies an A(rgument)-position (i.e. a syntactic position that can be assigned a theta role) in those scrambled structures. At S-structure, the verb is able to assign accusative case to the non-adjacent direct object NP in the OSV- or SO Adjunct V-sequences. Under a movement analysis, it is generally assumed that the direct object NP is adjacent to the verb at D-structure, from which it receives a theta role under sisterhood. So, there is one basic word order, and the alternate order(s) is (are) derived by means of some sentence-internal movement operation leading to a syntactic chain. The relevant question then is whether the scrambled constituent heads an Α-chain or an A'chain (see below for further discussion). Within the base generation approach, we can further distinguish a configurational and a non-configurational analysis of constituent structure. According to the former, the variable D-structure representation is a hierarchical, asymmetric (i.e. configurational) constituent structure in which the arguments and adjuncts are freely distributed (cf. Bayer — Kornfilt for German and Neeleman for Dutch). According to the latter the variable D-structure representation has a flat (i.e. non-configurational) phrase structure (cf. among others E. Kiss for Hungarian). Subject and object are not separated by a VP-boundary, and the arguments within a VP are not hierarchically organized (cf. also Haider 1988 for German).

Introduction: approaches to and properties of scrambling

3

The issue of scrambling (i.e. freedom of constituent order) and configurationality (i.e. hierarchy of constituent structure) is also discussed in Hale's paper on Warlpiri, a language with great freedom of constituent order. He argues that there is little evidence in support of a movement theory of Warlpiri free word order. He assumes that there are two levels of representation in Warlpiri: a core argument structure and an overt phrase structure. The first level of representation conforms to the configurational type and consists of a predicate and its associated argument slots which are occupied by non-overt (pronominal or anaphoric) elements. The second level of representation is less clearly configurational. The overt nominal expressions are analyzed as adjuncts which are linked via coindexation to the argument positions of the core argument structure of the clause, and are rather freely distributed within the clause. Obviously, the movement and the base generation analyses of scrambling make different predictions with respect to the properties that the phenomena of free word order (the scrambled structures) are expected to display. Under a movement approach, one would expect scrambling to display properties generally associated with movementderived structures: there is an atecedent-trace relation; the relation between the trace and its antecedent is apparently unbounded; scrambling obeys island constraints. As far as the last property is concerned, Webelhuth (1989) has shown that scrambling phenomena in German are sensitive to Ross's (1967) island constraints on movement transformations. The following ill-formed sentences from German illustrate the sensitivity of scrambling to island effects such as the Left Branch Condition (3a), the Coordinate Structure Constraint (3b), the PP-island condition (3c) (examples taken from Webelhuth 1989): (3)a.

*..weil meines Bruders gestern [-- Auto] ...because my brother's yesterday car wurde was

gestohlen stolen

4 Norbert Corver and Henk van Riemsdijk

b.

*...weil Hans jemand [-- und Maria] ...because Hans somebody and Mary

angemeldet registered

hat has c.

*...weil ...because gekämpft fought

ihre Freiheit die Leute lange [für--] their freedom the people long for haben have

At the same time, it should be noted that the free word order phenomenon does not always display properties exhibited by movement operations. To mention one example, the phenomenon of split topicalization in German (cf. Van Riemsdijk 1989) as exemplified in (4a), is not found in scrambled structures (cf. Neeleman): (4)a.

Bücher hat Hans nicht [viele --] gekauft Books has Hans not many bought

b.* Hans hat Bücher nicht [viele --] gekauft] It is often less clear whether scrambled structures display the property of apparent unboundedness. Ross (1967) initially formulated the descriptive generalization that scrambling is clause-bound. In contrast with Wh-movement or topicalization, a finite CP may never be crossed by a scrambled constituent in a language like German. (5)

*..weil Hans den Wagen versprochen hat [dass er -...because Hans the car promised has that he reparieren repair

würde] would

It seems, however, that this clause-boundedness restriction does not hold universally. Some languages turn out to exhibit Long Distance scrambling. See, for example, the following contributions in this volume: Mahajan for Hindi, Browning — Karimi for Persian, Müller

Introduction: approaches to and properties of scrambling 5

— Sternefeld for Russian and Japanese, Ueyama for Japanese. German and Dutch have a construction that is somewhere in between Verb Raising (clause union) and extraposition which is sometimes referred to as the "Third Construction" (Den Besten — Rutten 1989). One analysis of the phenomena in question is to assume that it is possible to scramble the object-NP of infinitival zu/te-complements into the matrix-IP. That would be an instance of Long Distance scrambling, (cf. Bayer — Kornfilt for an alternative analysis in terms of base generation.) (6)

...weil Hans den Wagent versprochen hat [PRO i, ...because Hans the car promised has zu reparieren] to repair

If one adopts the view that scrambling is an instance of "Move a", the question arises what kind of movement is involved. Is it similar to NP-movement in passive and raising to subject constructions, or does it resemble Wh-movement in Wh-interrogative and topicalization constructions? Several properties distinguish structures derived by NPmovement from those derived by Wh-movement. These concern the target of movement, the landing site and the trace. In (7), an inventory of the main properties is given: (7) NP-movement

Wh-movement

NP Α-position by substitution NP-position

XP (NP, PP. etc.) A'-position by substitution or adjunction [Spec, CP] or adjoined position

Properties of the antecedent Case Chain

Yes A-chain

No A'-Chain

Properties of the trace Binding theory Theta role CASE

Principle A Yes No

Principle C Yes Yes (when target =NP)

Category of the target Landing site

6 Norbert Corver and Henk van Riemsdijk

Under standard assumptions, the two types of movement processes can also be distinguished on the basis of the following diagnostics. First of all, a Wh-trace (i.e. a variable) as opposed to an NP-trace can license a parasitic gap. This contrast is illustrated in (8). Secondly, Whmovement triggers Weak Crossover effects, as in (9a). Weak crossover is given when an operator at LF, c-commands a bound variable type pronoun and its own trace at the same time. NP-movement, on the other hand, does not yield any WCO-effects (see (9b)). It rather repairs potential WCO-configurations (an anti-WCO-effect). This is exemplified in (9c). Thirdly, a structure derived by Wh-movement exhibits reconstruction or connectivity effects: reflexive binding can be reconstructed, for example, when the anaphor occurs in an A'-position (see (10a)). NP-movement, on the other hand, does not show any reconstruction effects. Rather, movement of an NP creates new binding possibilities (see (10b)). (8)a. b.

Which article, did you file tt [without reading ej? *The articki was filed [without reading ej.

(9)a. b. c.

*Who{ does it seem to hiSi father [that Mary likes tj? John, seems to hiSi father [t( to be intelligent]. WhOi seems to hist father [tt to be intelligent]?

(10)a. [Which picture of himself]) do you think John, likes tj? b. Theyi seem to each othert //, to be intelligent]. On the basis of the abovementioned properties and diagnostics, one could try to find out what movement operation is at the basis of scrambling structures. Do free word order constructions in language X display properties typical of Wh-movement or of NP-movement? This question is elaborately discussed in various articles in this volume. It is characteristic, however, for the present situation that there is no consensus among proponents of the movement approach about the type of movement involved in scrambled structures. Some linguists assume that scrambling behaves more like Wh-movement, and as such involves movement to an A'-position, i.e. a position that is structurally inaccessible to theta-role assignment (See Webelhuth 1989, Müller — Sternefeld (this volume) and Vikner (this volume) for German). In general, this A'-position is analyzed as an adjunction site to VP or IP

Introduction: approaches to and properties of scrambling

7

(cf. Müller — Sternefeld for a brief discussion of parametrization across languages of adjunction sites for scrambling). Other linguists claim that scrambled structures are derived by an NP-movement-like operation and hence involves movement to an Α-position, i.e. a syntactic position which can be assigned a theta role. It is often assumed nowadays by proponents of the "scrambling-to-A-position" that the scrambled NP is moved to the specifier position (an A-position) of some functional head such as AGR-O, in order to receive Case under Spec-head agreement (cf. Mahajan (this volume); Vanden Wyngaerd 1989; Deprez (this volume)). Under such an analysis, scrambling resembles NP-movement in being case-driven movement (i.e. substitution) towards an A-position. Restricting ourselves to German, let us briefly mention some of the argumentation given in support of each of the two movement analyses. The following arguments have been adduced among others to support the Α-movement analysis. First, the fact that a finite clause boundary may not be crossed by a scrambled constituent in German (as in (5)) reminds us of the clause boundedness of NP-movement or anaphoric binding (a Tensed-S-effect) in German, which can be explained in terms of Principle A of the Binding Theory (cf. Fanselow 1990). Secondly, scrambling may give rise to new Α-binding possibilities (cf. Webelhuth 1989; Lee and Santorini). In (1 la), for example, scrambling the direct object die Frauen across the indirect object einander creates a new binding relation. Thirdly, scrambling does not exhibit weak crossover effects (cf. Haider 1988; Fanselow 1990; Lee and Santorini). This is exemplified in (lib). Fourthly, scrambling does not allow for reconstruction. Scrambling of the anaphor in (11c) yields an ill-formed structure, because the moved anaphor will not be properly bound by the antecedent die Frauen since reconstruction does not apply to the scrambled anaphor. (ll)a.

weil wir die Frauen^ einander,· f, vorgestellt haben because we the women each other introduced have

b. weil jedetii seine{Mutter f, mag because everyone-ACC his mother-NOM likes c.

*weil einander die Frauen( nicht mögen because each other-ACC the women-NOM not like

8 Norbert Corver and Henk van Riemsdijk

As the reader will notice, there is not always consensus on the facts either. Some reservations with respect to the robustness of some of the generalizations such as the above are in order. Some of the argumentation in support of the "scrambling as A'movement" approach can be summarized as follows. First, scrambling is blocked in syntactic environments where Wh-movement is blocked (cf. Webelhuth 1989). That is, scrambling exhibits the same island behavior as Wh-movement, as was illustrated in (3). Secondly, scrambling, like Wh-movement, shows anti-crossover effect. This is shown by (12a), which is taken from Webelhuth (1989). Thirdly, scrambling moves NPs as well as non-NPs (cf. Webelhuth 1989; Vikner (this volume)). Fourthly, scrambling appears to license parasitic gaps (cf. Webelhuth 1989; Vikner (this volume)). (12)a.

?...weil [manche der Behauptungen, die HanSi ...because some-of the claims that Hans während der Konferenz gemacht hatte] βη during the conference Made had he zurücknehmen musste back-take had-to

b. ...weil [mit Karl]t er nie ti gesprochen hat ...because with Karl he never spoken has c.

?...weil er den Patienten [ohne PRO vorher e{ ...because he the patient without first zu untersuchen] t( operierte to examine operated

We refer the reader to the various papers in this volume for more elaborate discussion of (the tenability of) each of these arguments. A critical discussion of both types of movement approaches can be found in those papers which defend a base generation approach towards scrambling, viz. Bayer — Kornfilt, and Neeleman. The difficulty of interpreting free word order structures becomes even more clear when we consider scrambled structures which appear

Introduction: approaches to and properties of scrambling

9

to exhibit mixed properties, i.e. both A- and A'-properties. This paradoxical situation is exemplified by the following sentences from German, taken from Webelhuth (1989) (See also Mahajan for Hindi, Browning — Karimi for Persian). (13)a.

?Peter hat Peter has

jeden Gast [ohne e anzuschauen] every guest without to-look-at

seinem his

Nachbarn t vorgestellt neighbor introduced 'Peter introduced every guest to his neighbor without looking at.' b. Peter hat die Gäste,· [ohne e anzuschauen] Peter has the guests without looking-at

einander, t each other

vorgestellt introduced-to 'Peter introduced the guests to each other without looking at them.' The scrambled structure in (13a) manifests both an anti-weak crossover effect (an Α-property) and parasitic gap licensing (an A'property). In (13b), the scrambled NP simultaneously binds an anaphor (an Α-property) and licenses a parasitic gap (an A'-property). This paradoxical situation raises the question whether the standard A/A*-dichotomy (See Chomsky 1981) is sufficient to adequately characterize the array of properties displayed by scrambled structures. According to Webelhuth (1989), the standard partitioning of phrase structure positions into A- and Α-bar positions is too rough. He proposes that at least a third type of phrase structure position should be distinguished, viz. a mixed position exhibiting both A- and A'properties. Scrambling in German is then considered a unitary process in which there is a single derived landing position (a VP or IP adjoined position) for the scrambled phrase which simultaneously exhibits Aand A'-properties. Movement to such a mixed adjoined position creates a chain which has both A- and A'-properties. Webelhuth's paradox and the related question whether the standard typology of A- and A'-positions is adequate or not, are elaborately

10 Norbert Corver and Henk van Riemsdijk

discussed in various papers contained in this volume. Lee — Santorini, for example, also assume that scrambling (in German and Korean) is a unitary process of adjunction to VP (and perhaps IP). However, their approach to the paradox is based on the assumption that the binding and reconstruction properties of a chain are not determined by the structural position of its head, but instead by whether the head has undergone local movement or not. Α-properties (or An(aphoric) properties in their terminology) are characteristic of chains formed by local movement. A scrambled NP in an adjunction position can therefore serve as a binder as long as its position is derived by movement within a local domain. Conversely, only a scrambled phrase in a position derived by non-local movement is eligible for reconstruction. Deprez proposes to redefine the notions of Α-position and A'position so as to allow a different division of properties. She distinguishes [+HR]-positions and [-HR]-position (HR = Head Related). In case the scrambled object displays Α-properties, movement equals substitution for a case-marked specifier position, i.e. a [+HR,+case]position. When the scrambled object displays both A and A'-movement properties, movement equals substitution for a non-case marked specifier, i.e. a [+HR,-case] position. So, Deprez' analysis relates the mixed behavior to the caseless status of the specifier in those languages, and assumes the A and A'-properties are licensed simultaneously. Mahajan's analysis dismisses the paradox as apparent by analyzing scrambling (in German and Hindi) as a non-unitary phenomenon. According to his analysis, the standard A/A' dichotomy is sufficient to adequately characterize the array of properties displayed by scrambled structures. He posits the existence of both an Α-position landing site (a Spec-position) and an A*-position landing site (an adjoined position) for scrambled constituents. Scrambling as Α-movement involves movement to the case-marked specifier of an intermediate functional projection ([Spec,AgrOP]); in the case of A'-movement to a caseless position it is adjoined to a maximal projection (cf. also Vanden Wyngaerd 1989). The combination of A and A'-properties of the scrambled phrase is the result of the successive application of an Α-scrambling operation and an A'-scrambling operation. The first movement yields A-properties, the latter A'-properties. So, a sentence like (13b) is assigned a structural analysis in which the anaphor is bound by an antecedent in

Introduction: approaches to and properties of scrambling

11

an Α-position while the parasitic gap is bound by a distinct antecedent occupying an A*-position. Browning — Karimi also argue that the basic A/A* dichotomy is not sufficient to adequately characterize the array of properties displayed by scrambled structures in a language like modern Persian (an SOVlanguage). They argue that there are three types of scrambling in this language: object shift (cf. Holmberg 1986), clause initial scrambling and long distance scrambling. The object shift process, which moves a specific direct object in the sequence SU - 1 0 - DO - V to a position between the SU and the 10, exhibits mixed properties. It displays both Α-properties (viz. the property of case driven movement and the scrambled specific object can bind a reflexive) and A'-properties (the scrambled object can license a parasitic gap). They propose that the scrambled object occurs in a VP-adjoined position, where case is licensed under government by an inflectional head. Licensing of a parasitic gap is possible under the assumption that variable binding is also permitted from adjoined positions which are case licensed. The Aproperty of reflexive binding is related in their analysis to caselicensing: any position in which case is licensed is a position that can bind reflexives. Clause initial scrambling (adjunction to IP of a (specific or non-specific) argument) and long distance scrambling (adjunction of a specific or non-specific NP, or of a PP to the VP of a higher clause) are not case-driven and behave like A*-movements. Ueyama notes in her article that also in Japanese, scrambling exhibits both Α-movement properties (anaphor binding) and A'movement properties (reconstruction). Given this mixed behavior, she rejects the standard view that scrambling in Japanese involves A'movement (cf. Saito 1985; Hoji 1986) and proposes that the dichotomy of A vs A'-movement should be abandoned. Besides typical Amovements of arguments (i.e. case-driven non-operator movement) and A'-movements of arguments (i.e. non-Case-driven operator movement), scrambling in Japanese is of a third type, viz. non-Case-driven operator movement. So, whereas in Α-movement (as in raising) the landing site is a case position, in scrambling case is assigned to the D-structure position of a scrambling chain. Ueyama assumes that the landing position is an adjoined position. Finer's paper discusses scrambling operations in Selayarese (a VOSlanguage). He claims that there are two preverbal clause-internal positions that elements scrambled out of their postverbal D-structure

12 Norbert Corver and Henk van Riemsdijk

position may ocupy: (i) a focus position, i.e. [Spec,IP] and (ii) a topic position, i.e. a position adjoined to IP. A phrase occupying the focus position exhibits typical A*-binding properties, such as reflexive binding reconstruction and vulnerability to WCO. The focus - gap relation is an A'-chain. The topic position is adjoined to IP and hence occupies an A'-position as well (the A' non-operator). It turns out, however, that a scrambled phrase occupying this topic position exhibits Α-properties (e.g. the topic element does not trigger WCO-effects, nor reconstruction effects with anaphors moved in topic position) besides A'-properties. This mixture of symptoms will follow from its status as a non-operator in A'-positon. So, the crucial distinction between the two positions is not one of argument versus non-argument, but one of operator versus non-operator. Vikner's paper discusses two types of clause-internal objectmovement in the Germanic languages: (i) Object shift found in Germanic SVO-languages (except English), and (ii) Leftward object movement (referred to as "Scrambling") in Germanic SOV-languages. Although the two reordering processes are analyzed as adjunction operations (to VP and, in a language like German, also IP), they differ from each other in their A/A'-properties. Object shift is analyzed as an Α-movement operation, i.e. the moved element ends up in a (VPadjoined) case-assigned position. The Object shift process goes along with V-to-I movement. The finite verb ends up in a position from where it can assign case to the adjacent VP-adjoined object (See also Deprez* contribution). Leftward object movement in SOV-languages, on the other hand, exhibits A'-movement properties: the scrambled element ends up in an adjoined caseless position. Nikanne's paper discusses properties of the [Spec,IP] position in Finnish, which is occupied by the topic of the sentence. A characteristic of Finnish is that it avoids verb initial declarative sentences. The [Spec,IP] position must be filled either by a syntactic movement operation or a PF reordering. It is proposed that arguments originate within VP and that in the unmarked case the topic position is filled at S-structure with the NP highest in the theta hierarchy. In constructions where there is no nominative subject present, there is no element in [Spec,IP] at S-structure. In partitive structures, for example, the [Spec,IP] position need not be filled at S-structure, since the partitive NP receives partitive case from the verb within VP at S-structure. On the assumption that PF does not respect the theta hierarchy, the

Introduction: approaches to and properties of scrambling 13

partitive NP or some other element (e.g. an argument-PP or an adjunctPP) can be moved into the [Spec,IP] position at PF. So far we have seen various approaches towards the generation of free word order structures. Schematically (See also E. Kiss's contribution): (14) Scrambling Syntactic phenomenon Base generation FlaT^Configurational

Stylistic (PF) phenomenon Movement

Τ

A-movement^A'-movement

A/AÄSovement

A question which arises is whether scrambling structures should receive a uniform treatment, both cross-linguistically and languageinternally. E. Kiss notes in her article that in principle all four possibilities (i.e. base generation in a flat structure, base generation in a configurational structure, PF-movement, syntactic movement) may in principle be exemplified across languages. She considers Hungarian to be an instance of a language in which the freedom of word order is the result of base-generation of flat structure of an arbitrary argument order. Hale also notes in his paper that free word order phenomena across languages may be the result of different processes. He argues that whereas a language like Papago achieves word order variation through movement, the variations in overt phrase structure word order in a language like Warlpiri is not the result of movement, but results simply from the fact that (certain or all) overt phrasal expressions are adjuncts, which are linked to the grammatical arguments of the verb represented overtly by agreement morphology in the auxiliary. Whereas some authors in this volume (e.g. Müller — Sternefeld for German) claim that scrambling is a unitary phenomenon intralinguistically, others claim that free word order structures in one and the same language may result from different operations. Bayer — Kornfilt and Neeleman, for example, make a distinction between scrambled structures involving focused elements and "neutral" scrambled structures. The former involve an A'-movement operation, the latter are freely generated in D-structure. Mahajan analyzes the

14 Norbert Corver and Henk van Riemsdijk

scrambling phenomenon in a language like Hindi as a non-unitary phenomenon. He assumes that certain variable word orders result from Α-movement, while others are derived by A*-movement. As will be clear from this introduction and from the papers in this volume, there is no consensus at the moment about what the correct theory of free constituent order is. Competing analyses for scrambling are proposed, even for one and the same language. However, it is clear that this lively, still ongoing debate is gradually bringing us closer to a better understanding of the scrambling phenomenon.

Acknowledgements The present collection of articles grew out of a workshop on Scrambling which took place at Tilburg University in October 1990. We gratefully acknowledge the financial support of The Netherlands Organization of Pure Scientific Research (NWO), Tilburg University and the Beiport Familienstiftung, Vaduz. Anonymous thank-you's are due to our anonymous referees. Finally, our thanks go to Gerrit Rentier and Bart Hollebrandse for their assistance in organizing the workshop, to Conchita Barb6 and Anneke Smits for their assistance in preparing the camera ready copy, to Tim van de Avoird for making the index, and to Ineke van der Craats for help with the proof-reading.

References Chomsky, N. 1981 Lectures on Government and Binding. Dordrecht: Foris.

Den Besten, Η. — J. Rutten 1989 "On Verb Raising, Extraposition and Free Word Order in Dutch", in: D. Jaspers — W. Klooster — Y. Putseys — P. Seuren (eds.), Sentential Complementation and the Lexicon. Studies in Honour of Wim de Geest. Dordrecht: Foris, 41-56. Fanselow, G. 1990 "Scrambling as NP-movement", in: G. Grewendorf — W. Sternefeld (eds.), Scrambling and Barriers. Amsterdam: Benjamins, 113-140.

Introduction: approaches to and properties of scrambling

15

Grewendorf, G. — W. Steraefeld 1990 Scrambling and Barriers. Amsterdam: Benjamins. Haider, H. 1988 "Θ-Tracking Systems - Evidence from German", in: L. Maracz — P. Muysken (eds.), Configurationality. Dordrecht: Foris, 185 - 206. Hoji, H. 1986

Logical Form Constraints and Configurational Structures in Japanese. Doctoral Dissertation, University of Washington.

Holmberg, A. 1986 Word Order and Syntactic Features in the Scandinavian Languages and English. Doctoral Dissertation, University of Stockholm. Riemsdijk, H. van 1989 "Movement and Regeneration", in: P. Beninca (ed.), Dialect Variation and the Theory of Grammar. Dordrecht: Foris, 105-135. Ross, J.R. 1967 Constraints on Variables in Syntax. Doctoral Dissertation, Massachusetts Institute of Technology. Saito, M. 1985

Some Asymmetries in Japanese and Their Theoretical Consequences. Doctoral Dissertation, Massachusetts Institute of Technology.

Webelhuth, G. 1989 Syntactic Saturation Phenomena and the Modern Germanic Languages. Doctoral Dissertation, University of Massachusetts, Amherst. Wyngaerd, G. Vanden 1989 "Object Shift as an Α-Movement Rule", in: P. Branigan — J. Gaulding — Μ. Kubo — Κ. Murasugi (eds.), Proceedings of the Student Conference in Linguistics 1989, ΜΓΓ Working Papers in Linguistics 11, 256-271.

Against scrambling as an instance of Move-alpha* Josef Bayer — Jaklin Kornfilt

1.

Introduction

"Scrambling" is a cover term that arose from Ross (1967), and that roughly means that two constituents can be permuted if they are clause mates. The languages of the world differ greatly as to the extent to which they allow for such permutation. Syntactic research has recently given an increased attention to this phenomenon which has borne a clearer understanding of the data as well as concrete proposals as to how scrambling can be explained by UG and without recourse to ad-hoc means such as "stylistic transformations" or "PF-operations" (see Williams 1984 and others). At the moment there are two major proposals in GB-type generative grammar. According to the first, scrambling is an instance of Chomsky-adjoining an XP (NP, PP, to a more limited extent also AP or ADV) to VP or to IP (perhaps also AP). In other words, it is movement to an A'-position. This view has recently been advocated for German by Webelhuth (1985,1989, 1990) — Webelhuth (1989) actually concludes in the end that scrambling may be halfways between A'- and Α-movement with respect to binding theory ~ , Sternefeld (1990) and Felix (1985); Bennis — Hoekstra (1985) take a similar position for the analysis of Dutch.1 While there is little doubt that certain instances of scrambling can be adequately captured in this way, there are cases which obviously cannot be the result of A*-movement. Relevant observations to which we will turn below have led to the proposal that certain subcases of scrambling are instances of A-movement.2 The goal of this investigation is twofold. We first want to argue that the core cases of scrambling in German can neither be captured adequately by an adjunction operation nor by some movement to an argument position i.e., we will deny a movement account altogether. Second we want to sketch a base-generation account that is not committed to the inadequacies of a non-configurational account of German syntax. It will be shown that this proposal makes use of morphosyntactic and lexical properties that are independently attested in the language. We will first turn to evidence against A'-movement, then to evidence against Α-movement, and finally propose an alternative account that is not committed to syntactic movement at all.

18 Josef Bayer — Jaklin Kornfilt

2.

Scrambling as A'-movement

Assuming a strictly configurational structure for German according to which subject and object are separated by a VP-boundary and the arguments inside a VP are hierarchically organized, it seems natural to derive the scrambled order of arguments by adjunction to either VP or IP. Let us turn to some of the difficulties that emerge from this solution which seems attractive at first glance.

2.1.

Bound pronouns

Mahajan (1988, 1989, 1990) has shown that scrambling in Hindi (unlike Quantifier Raising (QR) or instances of WH-movement that applies only at LF) does not exhibit the weak crossover effects (WCO) that are typical for A'-movement, and that scrambling does not allow for reconstruction. Note that reconstruction is usually possible in instances of A'-movement, thus voiding violations of the binding theory.3 Consider first the following examples from German which involve a coindexation relation between a definite NP and a possessive pronoun: (1)

Adjunction to VP (name-like binder)

a.

Wir wollten ... we wanted [dem Professor/, seinet Sekretärin vorstellen (to)the professor his secretary introduce 'We wanted to introduce to the professor his secretary.'

b.

*seinet Sekretärin [dem Professorin vorstellen.

Against scrambling as an instance of Move-alpha

c.

[Seinei Sekretärin]) his secretary

haben have

wir [dem we (to) the

19

Professor]; ej professor

noch nicht vorgestellt. yet not introduced 'We haven't introduced his secretary to the professor yet.' In (la) the dative NP (which is in canonical position) can bind the possessive pronoun seine, (lb), however, which is a normal case of "object scrambling", does not allow this binding, (lc) shows a clear-cut instance of A'-movement. Here one of the object NPs has been moved to SpecC, which is always an operator position in German. Interestingly, the binding of seine is possible here. The reason which we assume is that NPj can reconstruct into a trace position at the level of LF; thus, the binding NP can c-command NPj, thus allowing seine to be coindexed with the NP dem Professor. If the case of scrambling in (lb) were a case of normal A'-movement, we would wrongly predict that reconstruction is possible here, too.4 The examples in (2) show that the same holds true when the dative-NP is quantified; in this case the possessive pronoun functions as a bound variable: (2)

Adjunction to VP (non-referential binder)

a.

Wir wollten ... we wanted [jedem Professorseine, Sekretärin vorstellen (to) each professor his secretary introduce 'Introduce to each professor his secretary.'

b.

*seinei Sekretärin [jedem ProfessorJi vorstellen.

c.

Seinet his

Sekretärin wollten wir [jedem ProfessorJi secretary wanted we each professor

vorstellen. introduce 'We wanted to introduce his secretary to each professor.'

20 Josef Bayer — Jaklin Komfilt

The pronoun to be bound remains unbound in (2b), but not in the reconstruction case (2c). Consider next adjunction of an object NP to IP. The examples in (3) involve a name as a subject and those in (4) a quantified subject NP i.e., a non-referential expression:5 (3)

Adjunction to IP (name-like binder) a.

daß der Hanst seinei Eltern sehr liebt that the Hans his parents very loves 'that Hans loves his parents very much.'

b.

*daß seinet Eltern der Hansi sehr liebt

(4)

Adjunction to IP (non-referential binder) a.

daß jedert seinei Eltern liebt that everybody his parents loves 'that everybody loves his parents.'

b.

daß seinei Eltern jeder( liebt

The crucial difference concerns (3b) and (4b). The effect is independent of the Case of the scrambled NP. Thus, we would also get it when the scrambled phrase is a dative or a genitive. What is the difference between (3b) and (4b)? Notice that (3b) becomes more acceptable when der Hans receives focal stress or when it is modified by a focusing particle like nur ('only') or sogar ('even'). The proper generalization seems to be that, under scrambling, binding is still possible when the binder is a quantified subject NP. This fact was noted in Webelhuth (1985) and used as an argument for a subject/object asymmetry in German. Notice that it would not suffice to assume a trace into which the scrambled object NP could reconstruct because this possibility would apply to the ill-formed case of (3b) as well. Another problem for the binding-by-reconstruction analysis was to our knowledge first pointed out in Haider (1986) and then independently seen by Anoop Mahajan. If movement to an A'-position were the right account for scrambling, the example in (5) should exhibit a WCO-effect:

Against scrambling as an instance of Move-alpha

(5)

daß jederii seinet Eltern that everybody-ACC his parents-NOM 'that everybody is loved by his parents.'

21

lieben love

Such an effect is, however, absent, and (5) is completely well-formed. Notice that (3b) contrasts with the parallel English case shown in (6). (6)

...that hiSi parents, Johnt really loves

According to what we know about the syntax of English, this must be the result of move-α, in particular, movement to an adjoined operator position. As a consequence, reconstruction should be possible, and pronominal binding should be able to take place. The relative wellformedness of (6) shows that this prediction is indeed correct.

2.2.

Anaphors

Consider next the binding of reflexive and reciprocal elements: (7)a.

daß sicht der Heinrichj e{ haßt that REEL the Heinrich hates 'that Heinrich hates himself.'

b.

daß die Familienmitglieder,· einander nicht mögen that the family-members each other not like 'that the members of the family dislike each other.'

c. d. e.

*daß einander die Familienmitglieder nicht mögen. Einander mögen die Familienmitglieder et bestimmt nicht. that himself ι, John{ really despises e{.

(7a) is well-formed because sich is a clitic that attaches to C and binds a clitic-trace which is c-commanded by the subject NP. Thus no violation of principle A arises. The reciprocal einander in (7b, c), however, is not a clitic. Scrambling it over the subject NP as in (7c) leads to marked awkwardness, and for many speakers to straight ungrammaticality. (7d) shows that reconstruction can, in principle,

22 Josef Bayer — Jaklin Komfilt

rescue such examples. The same seems to hold true for English, as shown by (7e). Thus, (7c) is another indication that object scrambling in German may not be A'-movement.

2.3.

Scrambling as adjunction

Notice as a contrast that PP-adjuncts can freely adjoin to VP or IP as shown in (8). We want to suggest that the preposed PP can be reconstructed into a trace-position as indicated in these representations: (8)a.

daß [in seiner( Wohnung]j Maria den Professor ej that in his apartment Maria the professor schon oft besucht hat already often visited has 'that Maria has often visited the professor in his apartment.'

b.

daß Maria [in seiner, Wohnung]j den Professor e1 schon oft besucht hat.

c.

daß [in seineri Wohnung ]j der Professor,· schon that in his apartment the professor already oft β] von Maria besucht wurde often by Maria visited was

As indicated by the traces, there is always a way of reconstructing the PP into a position where the pronoun seiner is c-commanded by the masculine NP. PPs with argument status, however, behave differently as shown by the grammaticality contrast between examples like daß der Vater die Tochter [auf ihren Stuhl] gesetzt hat ('that father has seated the daughter on her chair') versus *daß der Vater [auf ihren Stuhl] die Tochter gesetzt hat and *daß [auf ihren Stuhl] der Vater die Tochter gesetzt hat. Arguments of the verb can be licensed in scrambling positions, but from there they are obviously unable to reconstruct into a purported D-structure position. On the other hand, many of Webelhuth's examples do fall under the A'-movement generalization e.g., those in (9) which we modify somewhat in order to show a

Against scrambling as an instance of Move-alpha

23

possible co-indexation between the pronoun in a scrambled phrase and a referential subject NP: (9) a.

weil sich [über seine, Frau]s der since REFL about his wife the

HanSi [keinen Hans no

Film ej] anschauen würde film watch would 'since Hans would not watch a movie about his wife.' b.

weil [auf seinei KinderJj der HanSi /e, sehr since of his children the Hans very

stolz] proud

ist is 'since Hans is very proud of his children.' What has escaped the attention of many researchers is, however, the fact that these cases must be quite different from those cases in which an (unfocused) NP (or PP with argument status) is scrambled.

2.4.

Prosody

An obvious surface reflex of this difference is that adjunction to IP creates a prosodic break which is clearly absent in the scrambling of an object-NP. Consider the following examples in (10) where # marks the break: (10)a. daß den Heinrich niemand ausstehen kann that the Heinrich nobody stand can 'that nobody can stand Heinrich.' b. Πdaß den Heinrich # niemand ausstehen kann. c. V.that Henry nobody can stand, d. that Henry # nobody can stand. (10a) is naturally packaged into a single intonation phrase with phrasal stress on ausstehen. A prosodic break after the

24 Josef Bayer — Jaklin

Kornfilt

scrambled NP as in (10b) is rather awkward. In English, the pattern seems to be just the opposite. If there is no prosodic break, scrambling is almost impossible, as the contrast in (10c, d) shows. According to our intuitions, the cases of scrambling which allow for LF-reconstruction in (8) and (9) are most naturally pronounced with a prosodic break after the scrambled PP. This suggests that there is a correlation between types of scrambling and intonational phrasing, according to which only adjoined XPs allow for a prosodic break.

2.5.

Parasitic gaps

Consider next parasitic gap (PG) constructions. PGs have been used by Bennis — Hoekstra (1985), Felix (1985) and others following them as a diagnostic to show that an object NP can be scrambled within IP to an A'-position from where it can license a PG. An informal investigation of native speaker judgements revealed, however, that for those speakers who accept PGs, there are some interesting differences. Consider the data in (11) which summarize the result of our poll: (ll)a.

1 Diesen Manrii hat man [ohne pgt verwarnt zu this man has one without warned to haben] e, ins Gefängnis gesteckt have in-the prison put 'One has put this man into jail without having warned (him).'

b. Man hat ihnt [ohne pgt verwarnt zu haben] e{ ins Gefängnis gesteckt. c.

l*Man hat diesen Μαηηέ [ohne pgt verwarnt zu haben] ei ins Gefängnis gesteckt.

Against scrambling as an instance of Move-alpha 25

d. Da hat ihtii der Polizist [ohne pgt verwarnt there has him the policeman without warned zu haben] e{ ins Gefängnis gesteckt to have in-the prison put "The policeman has put him into jail without having warned (him).' e.

*Da hat diesen Μαηη{ der Polizist [ohne e{ verwarnt zu haben] e{ ins Geßngnis gesteckt.

(1 la, b) are compatible with a movement account because SpecC in (11a) as well as the clitic's position in (lib) head possible reconstruction chains.6 If PGs can be licensed by adjunction to VP,(lc) should be equally acceptable. Most native speakers, however, reject such cases or find them less good than the others. This tendency becomes even clearer when one considers PGs that are licensed from a position to the left of the subject NP. While clitic binders are still well-behaved (see (lid)), full NPs as in (lie) lead to a noticeable ungrammaticality.7 Summing up, we can assume that, contrary to some claims in the literature, PGs do not provide a convincing argument in favor of object scrambling to an A'-position within IP.8 We would like to keep up this claim at least until more thoroughly investigated data are available.

2.6.

Long-distance scrambling

Consider now "Long Distance Scrambling" (LDS), sometimes also referred to as the "Third Construction".9 In German and Dutch it seems possible to scramble the object-NP of infinitival zu-/te-complements into the matrix-IP if the matrix verb belongs to a limited lexical class of control verbs:

26 Josef Bayer — Jaklin Kornfilt

(12)

weil Heinrich den Wagent versprochen hat since Heinrich the car promised has [PRO et zu waschen] to wash 'because Heinrich has promised to wash the car.'10

Consider now a case in which we move an adverb of quantification from its D-structure position in the complement into the matrix clause: (13)a. weil Heinrich versprochen hat [PRO dreimal since Heinrich promised has three-times den Rosenkranz zu beten] the rosary to pray 'because Heinrich has promised to tell his beads three times.' b. [Dreimal]t hat Heinrich (ej versprochen [PRO e{ den Rosenkranz zu beten] (ambiguous) c. weil Heinrich [dreimal], versprochen hat [PRO (*ej den Rosenkranz zu beten] (unambiguous) (13b) is a clear case of A'-movement. Dreimal ('three times') can bind a trace inside the complement, and we thus get the reading of three prayers, which (13b) has in common with (13a). Alternatively, dreimal could also bind a trace inside the matrix IP; we would then get a reading according to which there are three promises. This creates the ambiguity. Assuming now that scrambling is adjunction to VP, the same ambiguity should be available in (13c). However, (13c) is clearly unambiguous, allowing only for the three promises reading. We will return to the phenomenon of LDS below. For the time being it is enough to see that LDS is very unlikely to be a case of A'-movement, as was suggested by den Besten et al. (1988). Our argument is this: Adverbs of quantification can move to an A'-position, with all the consequences of A'-movement as in (13b). If LDS is A'-movement, these adverbs should behave as in A'-movement. However, they do not, as illustrated in (13c). We would therefore be surprised to see that

Against scrambling as an instance of Move-alpha 27

scrambled NPs should behave differently. We will later present evidence that they do not either.

2.7.

Summary

We can conclude from what has been said so far that in German the scrambling of an object NP is clearly not an instance of A'-movement. Evidence from pronominal and reflexive/reciprocal binding, prosody, parasitic gaps and the scope of adverbs has revealed that object scrambling does not give rise to reconstruction effects and that what matters is exclusively the S-structure position of the scrambled item. In these respects, scrambling differs significantly from clitic movement. As the contributions to this volume by Mahajan, Neeleman, Lee — Santorini and others show, there are further arguments against an A'movement analysis of (most cases of NP-)scrambling which we cannot review here. At the same time, we cannot deny that focused adjunct PPs and, presumably, also APs and adverbials do scramble in the simplex clause in the sense of adjunction to IP. For Dutch, this point has been made quite clear in Neeleman (this volume). We will next investigate whether there are reasons to analyze the scrambling of object-NPs in German as an instance of NP-movement.

3.

Scrambling as NP-movement

We start with one immediate reservation which can be held against analyzing the scrambling of NPs as NP-movement, namely that it lacks the functional motivation of NP-movement. NP-movement is classically triggered by the interplay of Theta-theory and Case-theory. Scrambling, however, does not enter this interplay at all, since the scrambled elements are assigned a Case in their purported D-structure position which they retain in their S-structure positions. There is no obvious reason either why Theta-marking of a scrambled position should be impossible. We will show in section 4 that this can even be achieved without giving up the requirement that a Theta-assigner has to mark its sister (see Chomsky 1986). We will first review two proposals in the literature in favor of viewing scrambling as an instance of

28 Josef Bayer — Jak!in Kornfilt

NP-movement, and then point to a problem that an NP-movement account faces in connection with quantifier floating.

3.1.

Adjunction to VP or IP

Fanselow (1990) argues for scrambling as adjunction to XP (VP or IP), but he stipulates that such a newly created position counts as an Aposition. Without extensive justification, however, such a proposal is paradoxical, given that Chomsky-adjunction by definition creates an A'-position; yet, under current assumptions, NP-movement is Α-movement. Therefore, adjunction seems to be the wrong tool to explain the phenomena we have been calling "scrambling".11 Clearly, any theory that can opt for a substitution of an existent A-position would have to be preferred over a theory that has to stipulate Apositions which have to be created by adjunction. As we shall see, however, the matter is not trivial because there are indeed proposals in favor of conventional NP-movement. We will next turn to those, showing that - at least with respect to German - they do not convince us either.

3.2.

Movement to SpecAGR

In analyses positing VP-internal subjects, such subjects are said to move up to the Specl-position to pick up Case. But since scrambling NPs do have Case already, it is hard to see what scrambling NPs would need to pick up, so as to motivate their movement. This objection might be countered by referring to Chomsky (1989), where it is suggested that object Case may not be assigned by the verb itself, but by a functional category, AGR-O, which has its own projection and thus provides a position, say SpecAGR-O, into which the object NP would have to move in order to get Case:

Against scrambling as an instance of Move-alpha 29

(14) AGRP-O NP·

AGR-O' VP e

AGR-0

i

An analysis in this spirit has recently been proposed in Mahajan (1988, 1989, 1990) for Hindi, vanden Wyngaerd (1989) for Dutch and German, and others. It could explain the movement from a VP-internal position to a SpecAGR-0 position. However, we would still have no explanation as to what could drive an NP further up, for instance across the subject's S-structure position. Vanden Wyngaerd (1989) suggests that German cases of subclausal OSV-order are due to "a reversal of Case-assignment patterns that makes the OSV-order possible i.e., accusative is assigned directly to the object by the higher functional head and nominative to the subject by the lower one, AGR. Dutch differs from German in not allowing this Case reversal." In our view, this proposal is nothing but a restatement of the facts, which of course does not explain anything. The AGR-projections enter the stage as some "deus ex machina" in order to provide for a derivation of the surface order. A natural question would be why in German AGR-S and AGR-0 should change their places.12 We are afraid that there is no satisfactory answer to this question, and that the proposal amounts in the end to the claim that AGR-0 can scramble over AGR-S. But this just begs the question. In the following we will give three further arguments against this account of scrambling.

3.2.1.

Long-distance scrambling again

There are further problems that arise with viewing scrambling as NP-movement when we turn back to LDS. Consider, for instance, (12), an example in which the object NP of the embedded infinitival clause has apparently scrambled into a position of the matrix clause which is normally filled by the direct object NP. (12) is reproduced here for convenience:

30 Josef Bayer — Jaklin Kornfilt

(12)

weil Heinrich den Wagen{ versprochen hat [PRO e, zu waschen] 'because Heinrich has promised to wash the car.'

Assuming that we reject an empty proliferation of functional projections, how could the hypothetical movement of NP from its original position to its landing site have taken place here? Specl (i.e. under this proposal, SpecAGR-S) of the embedded clause is already filled with the subject i.e., PRO. If we are right in saying that SpecAGRP-S is the landing site of NP-movement, that position would be unavailable to the embedded object NP, which would then have to move out of the infinitival clause in a way that would violate locality conditions: Assume that for some unknown reason the zu-infinitive lacks an AGR-0 head, but that AGR-0 is available in the matrix clause. Since zu is not a proper governor, it cannot L-mark the VP. According to Chomsky (1986), the VP would then be a "Blocking Category" from which its immediately dominating maximal category, say AGRP-S, would inherit the status of a "Barrier". Thus, even if there is no CP involved - as den Besten et al. (1988) as well as Bayer — Kornfilt (1990) argue - NP-movement into the matrix clause should be impossible. The only way out would be adjunction to VP and then assuming now a CP-complement - movement via SpecC. The illformedness of (15) signals, however, that this analysis cannot be right: (15)

*weil Heinrich because Heinrich

[die the

Zebrasgesagt zebras said

hat has

[et [daß [er dem Kind et zeigen wird]]] that he the child-DAT show will 'because Heinrich has said that he will show the zebras to the child.' Given that the scrambled NP terminates in some quasi A-position of the matrix clause rather than in an A'-position, (15) can be explained. It is then a case of "Improper Movement".13 In their account of scrambling as movement to an adjoined position, Müller — Sternefeld (1990) propose the "Principle of Unambiguous Binding" (PUB) given in (16):

Against scrambling as an instance of Move-alpha

(16)

31

"A variable cannot be (simultaneously) bound by an operator position and by a scrambling position." Müller — Sternefeld (1990: 18)

By stipulating a distinction between scrambling positions and operator positions, they can rule out (15).14 With respect to the NPmovement approach, this shows - in connection with what has been said above - that there will not be an obvious derivation of cases of LDS such as (12), since if (15) is ruled out by the PUB, (12) should be ruled out as well — yet (12) is perfectly grammatical. As we shall discuss in more detail later on, we distinguish between (12) and (15) as follows: The "scrambled" NP in (12) is base-generated in an Aposition which receives its Case marking and its theta role from the embedded infinitive. This is made possible by an operation of Complex Category Formation (CCF), to be elaborated later. CCF cannot apply in (15), since neither the verb sagen nor the complementizer is a trigger of CCF (while versprechen is). Thus, even if the matrix verb were a CCF-trigger, the C-position occupied by the complementizer would prevent CCF. Notice further that not only accusative Case can be assigned under LDS, but any other Case as well, as shown for the dative in (17): (17)

weil Heinrich [dem Kind]( since Heinrich the child-DAT

vergessen hat forgotten has

[PRO ei die Zebras zu zeigen J the zebras to show 'because Heinrich has forgotten to show the child the zebras.' Vergessen is a verb that cannot assign dative Case. How could it then be in the domain of an AGR-0 that could license a dative? We can only conclude that the dative is assigned by zeigen directly. If we are right, LDS cannot be a Case-driven process.

3.2.2.

Floating quantifiers

Sportiche (1988) has suggested a theory of quantifier float (QF) according to which there is no quantifier movement to the right, but

32 Josef Bayer — Jaklin Kornfilt

rather NP-movement to the left. If a quantified NP or DP of the kind [QP Q[dp...]] is generated in a position in which Case cannot be assigned, the DP must move to a Case position, leaving Q behind. A typical case would be [ ^ p The men\ will [VP [QP all [DP e j ] take beer]]. Although Sportiche is not very specific with respect to the nature of the movement involved, it is quite obvious that he considers it "some kind o f ' NP-movement, not A'-movement. Giusti (1990) tries to exploit Sportiche's idea in connection with scrambling in German. QF is said to follow from the fact that German has scrambling anyway. But let us ask the more basic question first: What is scrambling? Although Giusti follows Sportiche (1988), she seems to think that scrambling is adjunction to VP or IP. We have seen, however, that there are serious drawbacks to this proposal. As the following examples show, there is an important difference between English and German, namely that in German, the remnant Q has to be Case marked like a normal DP. Simplifying somewhat, we may say that English pre-DP all and both lack nominal features, while their German counterparts all- and be id- have nominal features, and thus must be Case marked: (18)a. daß that Bier beer b. daß that

die Männer the men-NOM

im Gasthaus beide-e in-the pub both-NOM

bestellen order sie den Männer-n she the men-DAT

gestern beid-en yesterday both-DAT

den the

Ausgang zeigte exit showed c. daß ich mir dies-er Fehler that I REFL these mistakes-GEN

erst jetzt only now

all-er bewußt werde all-GEN aware become 'that I become aware of all these mistakes only now.'

Against scrambling as an instance of Move-alpha

33

Notice that *beide die Manner, *beiden den Männern etc. are illformed QPs which cannot be taken as the D-structure source of (18a,b). By adopting a proposal for explaining so-called "Split Topicalization" that has been made in Fanselow (1988) one could, however, argue that the D-structure Α-position contains two segments, a DP of the form [DP D [NP...]] and a QP of the form [QP Q[DP pro]], where pro should be an empty N-head. Both segments get Case-marked in situ, but the DP has to move to an operator position to the left of QP in order to (Theta)bind its inherent empty element pro. Both the Case assignment procedure and the fact that there is a pro which has to be bound allow only one type of movement: A'-movement. But notice now that examples of QF have all the properties of Α-movement, and that accounts of split topicalization cannot be used automatically for explaining QF, since cases of IP-internal topicalization are ungrammatical (cf. Bayer 1987 — van Riemsdijk 1989). An example is given in (19) which contrasts A'-movement with IP-internal adjunction: (19)a. [cp Socken [zieht [IP der Heinrich im Sommer socks puts the Heinrich in summer keine an]] none on 'Heinrich doesn't put on socks in summer.' b. lldaß [,p Socken [,P der Heinrich im Sommer keine anzieht]] c. *daß der Heinrich [VP Socken [VP im Sommer keine anzieht]] While adjunction of a bare plural DP to IP, as in (19b), seems to be borderline, adjunction to VP as in (19c) is truly offending. If QF were derived by scrambling to an adjoined position, the well-formedness of (18b, c) would be surprising. We conclude - contrary to Giusti's proposal - that if German QF is derived by movement at all, it must be derived by movement to an Α-position. But notice that now the old dilemma reappears: We have direct evidence that the remnant QP is Case-marked.15 Thus, given there is already a Case position in Dstructure, why should the DP have to move to some SpecAGR-position in order to pick up Case?

34 Josef Bayer — Jaklin Kornfilt

Another problem is that there should now be two identical AGRprojections, one for the QP and one for the scrambled DP. This underlines once again the 'deus ex machina' character of the NPmovement account.16 We can safely conclude that in German the QFphenomenon cannot be derived from the need of a DP (inside a QP) to escape the Case Filter. For reasons of space, we cannot attempt to discuss QF in more detail or sketch our own solution. The purpose of this section was rather to show the limits of the NP-movement account with respect to German.

3.2.3.

Obligatoriness and the functional aspect of scrambling

A last point worth mentioning is that - at least given two definite NPs - scrambling in German is always optional.17 Movement of an NP to a (non-thematic) position as dictated by the Case filter is clearly obligatory, as can be seen in the ill-formedness of sentences like * Yesterday was invited Bill or * Yesterday seemed Bill to have been invited. Unscrambled sentences, of course, are fully grammatical. It follows that nothing like the Case filter could possibly force the NP to move. Thus, an analysis of scrambling in terms of NP-movement faces a serious conceptual problem. Note also that this conclusion is corroborated by more traditional accounts, especially those of the Prague school's and other linguistic traditions' "Functional Sentence Perspective". It is general wisdom that scrambling has something to do with the informational structure of the clause according to which "old information" tends to precede information that should be highlighted (see for German Behagel (1929), Lenerz (1977), Lange (1978), Hoberg (1981), Abraham (1984), Eroms (1986) and numerous others). Of course, the grammar has to have ways of providing the prerequisits i.e., it must somehow allow object NPs to surface in exceptional Apositions; but once there are such ways, the question is again whether this kind of NP-movement is obligatory. If it were obligatory, as NPmovement classically is, the reason for scrambling would first and foremost derive from the grammar, and certainly not from some communicative ordering principle. Given that scrambling cannot be driven by both principles, one type of explanation must be on the wrong track. We would be surprised if a substantial part of the linguistic tradition were so completely misguided.

Against scrambling as an instance of Move-alpha

3.3.

35

Summary

Although we acknowledge that unfocused scrambled NPs appear in an Α-position rather than in an A'-position, we have found little reason to explain scrambling in terms of NP-movement to a specifier position of some stipulated functional head such as AGR-O. For methodological reasons, a proliferation of invisible functional projections is undesirable. Serious problems arise in connection with LDS, where NP-movement is likely to cross barriers. German QF clearly shows the properties of movement to an Α-position, but at the same time the QP left behind gets Case as well. NP-movement as a Case-driven process is therefore implausible. Finally we argued that such an approach loses contact with the linguistic tradition in an area where this does not seem to be desirable.

4.

On projecting scrambled order

If scrambling is neither an instance of A'-movement, nor of Α-movement, what is it? We would like to suggest that scrambling of unfocused NPs is not due to movement, but rather is base-generated, in ways we shall elaborate below. Our account is based on the observation that in German, there is no compelling reason to assume that a subject NP, marked for nominative Case, can only be licensed outside VP. This is because INFL in German can be viewed as a morphological category that attaches to V, rather than as a terminal syntactic category which takes VP as its complement. We can then say, in the spirit of Abney (1986), that V is the semantic head, while I is the formal head of the clause. In this way we can capture the insight behind the proposal in Jackendoff (1977) that V is the head of S, without committing ourselves to the view that S is a formal projection of V. Instead, we follow recent developments of X'-theory and propose, for German, that I is the sister of V at morphological structure (with I as the head of the resulting entity), not the sister of VP at syntactic structure, as is the case for languages like English and French. For German, then, V and I are jointly visible at the mother node, while they are kept separate for English until V moves to I from its position in VP:

36 Josef Bayer — Jaklin Kornfilt

The general mechanism which we assume for creating these jointly visible nodes is the following enrichment of the theory of phrase structure: (21)

Complex Category Formation (CCF) In a structure [...X° Y" ...] where X° is a raising category that governs Υ", (0 < η < max), X° will project into the complex category

X*

I Y"

In German, if X° is an inflectional affix I, the V-stem will attach to I in morphosyntax i.e., "before" it heads a VP. In English, however, due to the presence of a designated auxiliary system - cf. the original notion AUX - I is a terminal syntactic node. In German, then, an inflected V is morphologically an X°-category which is however syntactically complex in that it is both V and I \ A rather similar account has been given by Reuland (1990). One difference between Reuland's account and ours is that he takes the inflected verb to be [X7I°], while we take it to be [Χ7Γ]. In other words, we assume that once the (formal) I-head has taken its "lexical complement" it will have ceased to be a syntactic primitive.18 How can I take something as an argument that is not maximal? The status of I as an inflectional morpheme and the lack of a special AUX vocabulary dictates that its complement is a sub-word level category.19 Both V and I assign Case to the left, and license their arguments in the same fashion as under the current assumptions for English phrase structure. The only difference between English and German involves the contrast shown in (20) above. German allows for VP-internal nominatives. This is seen very clearly in the case of ergative verbs, which often show an unmarked constituent order dative-nominative-Verb. Under the more conventional

Against scrambling as an instance of Move-alpha

37

analysis of German phrase structure i.e., the mirror image of (18)a., the nominative Case would have to be assigned into VP. While there have been some suggestions as to how one could achieve this (cf. den Besten 1981; 1985), this situation remains a problem under more parsimonious assumptions about government and Case assignment. In the phrase structure proposed under (18b), this problem does not arise. Another situation where we find nominative subjects in the span of VP arises, of course, in scrambling constructions. It would appear here, at first sight, that the V licenses the subject in the same way in which it licenses its objects. However, this is only apparent. The true reason for the ability of an unergative verb to license a nominative subject in VP is that the verb is, at the same time, an Γ. Thus, the subject-NP will be licensed by a regular Specl-position. This makes it possible to base-generate a scrambled clause such as daß den Postboten-ACC der Hund-NOM beißt ('that the dog bites the postman') as in (22): (22) IP VP NPacc den Postboten

IP V°

NP 1 1 nom der Hund V I beiß

Γ V'

Some clarifications are necessary: First, Case assignment does not take place only under strict string-adjacency. Otherwise, V would be able to assign accusative Case only to its sister NP. In our system, however, the governing force (and hence Case-assigning ability) of V is kept intact in the projection(s) of V. Second, the nominative NP in (22) is governed by V. This does not mean, however, that it is also licensed by V — certainly not in the same way as an argument of V is licensed. If it were the case that V licenses the subject, the marked scrambling order of (20) would be indistinguishable from the comparable unmarked dative- nominative order of ergative or psychverbs. According to standard assumptions, an unergative V cannot take a (definite) subject NP as an argument. The subject NP is (in the active

38 Josef Bayer — Jaklin

Kornfllt

clause) licensed by I for Case and by VP for its theta-role. We adjoin the nominative-NP to V-zero, but this adjoined position must be licensed, since it bears Case and a theta-role. The NP is licensed as the specifier of I. Thus, the only difference between a scrambled IP like (22) and an unscrambled canonical IP is that in the scrambling construction, the VP remains unsaturated until the SpecIP position is licensed. That the nominative NP in scrambling constructions is not licensed by V is shown very clearly by the fact that the nominativeNP/participle sequence cannot be moved together to first position in V2-clauses: (23)

*[Der Hund-NOM the dog

gebissen] hat den Postboten-ACC bitten has the mailman

erst einmal only once intended reading: "The dog has bitten the mailman only once so far.' Since V does not license the nominative NP here, it also does not form a constituent with it. This situation contrasts sharply with corresponding constructions involving ergative verbs: (24)

[Die Luft-NOM the air

ausgegangen] ist dem Taucher-DAT ran-out is the diver

erst einmal only once 'The diver ran out of air only once (so far).' Here, the nominative subject is a genuine argument of the verb, hence is licensed by it and forms a constituent with it.20

4.1.

NP-movement from VP?

It has been suggested by den Besten — Webelhuth (1987) that sentential and VP-modifying adverbs and the negation morpheme nicht are to the left of VP, and that alternative surface orderings are achieved

Against scrambling as an instance of Move-alpha

39

by scrambling the object NP(s) out of VP. As one can expect on the basis of what we have said here so far, the landing site of this movement has the properties of Α-positions, not A*-positions. If we are right in claiming that scrambling is not NP-movement, den Besten — Webelhuth's account is untenable. One reason why it is untenable is that the ordering NEG [VP NP ... V] is ungrammatical unless the constituency changes to [ [ P NEG NP] ... V]. But in this case NEG focuses the object NP, and the we get "term negation".21 Simplifying somewhat, one can say that sentential negation requires NEG to immediately precede the verb. This means that scrambling becomes obligatory here. As we have shown above, however, the scrambling of NPs is always optional. In particular, it cannot be for reasons of the Case filter that NPs have to scramble out of the scope of NEG. Notice that NPs can well remain in the scope of an adverb as long as it does not function as a focusing operator. As Diesing (1988) and Kratzer (1989) have pointed out with respect to indefinite subject NPs and Moltmann (1990) for object NPs, the positioning inside or outside the VP creates a semantic effect. An indefinite NP like Kinder ('children') in (25) is interpreted as existential when it appears inside VP, while it must be interpreted as generic when outside VP. VP

N

(25)a. weil [bestimmt [Kinder auf der Straße spielen]] since certainly children in the street play b. weil Hans [meistens [Kinder verführt]] since Hans mostly children seduces (existential reading on NP, event reading on clause) (26)a. weil Hans Kinder [bestimmt auf der Straße spielen] b. weil Hans Kinder [meistens verführt] (generic reading on NP, state reading on clause) This semantic effect has been connected to scrambling, especially by Moltmann. Diesing's and Kratzer's suggestion is roughly that the VP contains an event argument to which a VP-internal indefinite NP is linked by existential closure. When the indefinite NP scrambles out of the VP, it is bound by a generic operator in the sense of Carlson (1977). This is not the place to discuss this proposal in more detail.

40 Josef Bayer — Jaklin Kornfilt

What we see is a rather stable semantic effect that is also observed in other languages (cf. Kornfilt (1990) and Εης (1991) for Turkish in connection with details due to Case assignment). What we do not see is that this effect provides a strong point in favor of scrambling, and in particular not in favor of NP-movement. We suggest tentatively that (non-focusing) adverbs can be adjoined to any projection of V i.e., also to V° or intermediate projections of V, and that there will ultimately be a semantic explanation of the issue at hand that is based on S-structure scope rather than on purported processes that lead to the S-structure in question.

4.2.

Scope and binding

CCF allows us to generate OSV-order without recourse to syntactic movement. As a consequence, the scrambled object NP will be in a regular Α-position. This yields the results of the NP-movement theory which are desirable with respect to binding and WCO without requiring the functional trigger of NP-movement. Recall now that with respect to a quantified binder we have found a significant difference between apparent adjunction to VP and apparent adjunction to IP. The relevant examples are reproduced below, with traces indicated only for convenience: (2)b. (4)b.

*Wir wollten [VP [seine, Sekretärin]j [jedem Professor],· e, vorstellen ] daß [,p [seine{ Eltern], jeder,· e} liebt]

Imagine now a representation without traces of movement. Thanks to the CCF mechanism the constituent structure of (4b) is obvious. The structure of double object constructions is less clear. Let us hypothetically (and neglecting VP-internal subjects) assume that in the canonical IO > >

[+maximal] [+projected] [-maximal] [+projected] [-maximal] [-projected]

This featural conception of X' theory allows for the existence of an additional categorial make up not represented in more standard conception of X' theory, namely a category with the features [+maximal, -projected]. This, I suggest, may be what characterizes Scandinavian pronouns, and may be more generally on-clitic pronouns: they are categories of the type [+maximal, -projected]. Romance clitics, in contrast, are categories of the type [-maximal, -projected]: (43)

Scandinavian pronouns: [-(-maximal] [-projected] Romance clitics: [-maximal] [-projected]

Note that under this view, Scandinavian pronouns are both heads and maximal projections. If so, we might expect them to share properties with both XP projections since they are [+maximal] and with X° projections since they are [-projected]. This is indeed what we observe, since Scandinavian pronouns appear to have the capacity to occur both in maximal projection positions, and in head positions when they undergo O-M. The pure head status of Romance clitics, on the other hand, may arguably make them unable to occur in maximal projection positions at S-structure, hence forcing movement. A precise elaboration of these ideas would require more space, and is beyond the

Parameters of object movement

125

scope of the present paper. The important point of this discussion is that pronominal O-M is here assumed to be an instance of head movement. In our view, it is this property which accounts for both the Α-properties of the chain they create and the irrelevance of the Case filter. To summarize, I have proposed in this section that, in view of the Case marking capacities of functional heads, the landing site of O-M is a Case marked position in Icelandic but not in Mainland Scandinavian languages or in Faroese. Pronominal O-M in Scandinavian and Faroese is analyzed as a subcase of head movement, a distinction which is supported by its quasi obligatory character which contrasts with the full optionality of NP O-M (Holmberg 1986). Given our definition of HR-positions above, adjunction of a head to an other head is a movement to an HR position, since an adjoined head is a sister to an X°. This explains straightforwardly why pronominal O-M in Scandinavian and NP O-M in Icelandic share Α-chain properties and solves the problem created by the definition (32) which strictly associates Α-chain properties with movement to a Case position. The claim here is that head movement manifest some of the properties which have been standardly assumed to pattern only with standard NP movement chains.18'19

4.3.

O-M to non-Case marked positions

Let us now turn to the second difference described above, namely the additional A*-properties which are manifested by certain cases of O-M but not by others. I will concentrate on German to illustrate this difference. It follows from what we have said so far, that since full NPs can move in German, the landing site of O-M in German should be assumed to be Case marked, just like the landing site of O-M in Icelandic. As noted above however, German is opposed to Icelandic in its additional A'-properties. Recall that, in contrast to Icelandic, German O-M licenses parasitic gaps. How can we then account for this difference? There are in fact two obvious possibilities: First, we could assume that the landing site of both German and Icelandic O-M is Case marked: the additional A'-properties of German O-M can then be taken to result from subsequent movement to an

126 Viviane Deprez

adjoined position as proposed in Vanden Wyngaerd (1989). Under this view, NP objects in German would first move to an HR-position and then to an HR'-position and it is the successivity of these movements which accounts for the combination of A- and A'-properties. For expositional clarity let us call this option: the successivity hypothesis. Note that if we adopt the successivity hypothesis to account for the additional A'properties of German O-M, we still need to restrict additional O-M to an A'-position in Icelandic, so as to account for the absence of parasitic gap licensing in this language. The second possibility is to assume that the landing site of German O-M, as opposed to the landing site of Icelandic O-M, although an HRposition is not a Case marked position. We can then propose that it is the Caseless status of the position which accounts for the additional A'properties. Under this view, A- and A'-properties would be licensed simultaneously from the same Caseless HR-position. Let us call this second option, the simultaneity hypothesis. Given Chomsky's (1986) well formedness condition on Α-chain, if the simultaneity hypothesis is adopted, we need an additional condition to license NP movement to HR-positions which are Caseless positions. It appears thus that under either view, an independent additional condition is needed to limit the movement of NPs to Caseless positions. Either further leftward NP movement to an adjoined position, must be blocked in Icelandic, and for that matter, in other SVO languages which do not permit NP O-M, or movement to a Caseless HR-position must be licensed in German and other SOV languages which manifest additional A'-properties. An obvious property which distinguishes the two sets of languages is their basic word order. Adapting an idea of Bennis (1986), I propose that this additional restriction be formulated as in (44): (44)

Sentence internal NP-movement to a Caseless position must be licensed under directional government

As proposed by Kayne (1983), directionality of government is determined by the basic relation of the verb to its object. It is thus to the right for SVO languages and to the left for SOV languages. Combined with the standard assumption that adjoined positions are not Case marked positions, (44) correctly rules out the possibility of further moving an NP to an HR'-position in Icelandic, as well as in

Parameters of object movement

127

Scandinavian languages and in any other SVO languages such as English and French. Consequently, the fact that parasitic gaps are not licensed by O-M in Icelandic is now accounted for. Note that (44) has further consequences. In particular, it will also rule out leftward sentence internal NP adjunction in languages such as English and French. In other words, (44) rules out (45) under the structure (46): (45)

*John this book read.

(46)

John [ Yp this book [

VP

read t ] 720

(44), however, does not discriminate between the successivity hypothesis and the simultaneity hypothesis for O-M in German and other SOV languages. Indeed both further movement to an HR'position or movement to a Caseless HR-position will be allowed under (44), since (44) only takes into account the Case marked status of a position and not its HR vs HR*-status. The type of data which would force a distinction between these two options is rather clear but somewhat difficult to test. The first option allows for a successive movement and thus for a successive realization of the Α-properties and the A'-properties of O-M. Simultaneity of these properties is excluded. The second option, on the other hand, allows both for a simultaneous realization of the A- and A'-properties and for a successive one. Indeed while the first option limits the A'-properties to adjoined positions, the second permits A'-properties to be licensed in both Caseless HR-positions and Caseless HR'-positions, thus permitting a wider range of options. Although this question is far from being clear, I am at present inclined to adopt the simultaneity hypothesis for two reasons. First, although it has been correctly argued by Vanden Wyngaerd (1989) that Webelhuth's examples of German O-M reproduced in (47), which appear to manifest simultaneous A- and A'-properties could be reanalyzed in terms of successive movement, a similar analysis seems rather difficult for the examples provided in (48).

128 Viviane Deprez

(47) a. Anti WCO effects Peter hat jeden Gast [ ohne e anzuschauen] Peter has each guest [ without looking at]

seinem to his

Nachbarn t vorgestellt neighbour introduced b. Anaphoric binding Peter hat die Gäste [ohne e anzuschauen] Peter has the guests [without looking at] einander vorgestellt to eachother introduced Webelhuth (1989) Consider first (47). (47a) manifests both anti-weak crossover effects and parasitic gap licensing. (47b) manifest both anaphor binding and parasitic gap licensing. But since, in both cases, the position which shows A'-binding effects precedes the position which shows A binding effects the apparent simultaneity of A- and A*-properties could be due to successive movement. First substitution to a position lower than the adjunct clause would license the anaphor and binding of the quantifier. Then a further adjunction above that adjunct clause would license the parasitic gap. But consider now (48): (48)a. Anti WCO-effects i. weil Maria jeden Gast [ohne seinem Partner e vorzustellen ] allein t läßt. ii. weil Maria jede Frau [ohne ihrem Partner e vorzustellen ] allein t läßt. b. Binding effects weil Maria die Gäste [ohne einander e vorzustellen] allein t läßt. (48a) and (48b), like Webelhuth's examples, show both A- and A'properties. (48a) shows anti-crossover effects and parasitic gap licensing, (48b) shows binding and parasitic gap licensing. In these cases however, the Α-properties are manifested within the constituent

Parameters of object movement

129

which also contains the parasitic gap. An analysis in terms of successive movement of the scrambled NP does not help since the trace left by a previous Α-movement could not independently c-command the constituent in which the Α-properties must be licensed. It would also necessarily c-command the position in which the parasitic gap must be licensed. These examples lead us to conclude that it must be the NP c-commanding both bound elements which is responsible for the Aproperties as well as for the A'-properties simultaneously. Whichever solution is adopted, one must then recognize the existence of positions which show simultaneous A- and A'-properties. These examples thus provide an argument in favor of the simultaneity hypothesis.21 A second argument which seem to favor the simultaneity hypothesis comes from the bare quantifier constructions of French given in (49). Since this argument is developed in detail elsewhere, I'll simply summarize it briefly here. I argue, in Deprez (1989), that in sentences like (49) the quantifier tout, which adapting a proposal by Belletti and Rizzi (1981), I assume to have the internal structure [tout PRO],12 occurs in a non Case marked HR-position. (49)a. Jean a tout mange John has everything eaten b. II faut tout manger One must everything eat Part of the argument relies on the fact that bare quantifiers can clearly occur in HR-positions, as shown in (50), where tout is the subject of a passive and a raising construction. In contrast to regular NPs, however, they cannot occur in adjoined or dislocated positions as in (51). (50)a. Tout a ete Everything has been

expedii a temps sent on time

b. Tout semble avoir ete expedii Everything seems to have been sent

ά temps on time

130 Viviane Deprez

(51)a.

*Tout, j'(l') Everything, I

b. Le chocolat, j'(l') Chocolat, I

adore like adore love

This suggests that in (49), the displaced quantifiers occurs in a HRposition and not in an adjoined position. That this HR-position is a Caseless position is shown by the fact that regular NPs cannot occupy it: (52)

*Jean a une pomme mangee John has an apple eaten.

If this is correct, it is interesting to note that this construction manifests exactly the expected properties: as shown in (53), it licenses parasitic gaps and it does not induce WCO violations. (53)a. II faut tout emballer avant d'expedier One must wrap everything before sending b. II faut tout relire avant de One must read everything before (54)

corriger correcting

II a touti renvoye a sotij proprietaire He has everything sent to its owner II a tou^ remballe aprös I'inspection de He has everything wrapped after the inspection of sonj contenu its content

Note moreover that when the same quantifier occurs in a Case marked HR-position, after undergoing passive, as in (55), parasitic gap licensing becomes impossible.

Parameters of object movement 131

(55)

*Ils nous ont affirmi que They us have affirmed that

tout α έΐέ everything was

emballe avant d'expedier wrapped before sending (55) could be assumed to be independently ruled out under the anti-c-command condition on parasitic gaps. But this raises the question as to why the anti-c-command condition does not also apply to (53). Under the view developed here, the contrast between (54) and (55) can be traced back to the Case marking properties of the position. If as I suggest, tout in (49) and (53) is indeed in a Caseless HR-position, then the contrast between (54) and (55) confirms the simultaneity hypothesis, since it shows that parasitic gaps can be licensed from Caseless HR-positions although they cannot be licensed from Case marked HR-positions. Note that this is also precisely the property which under the simultaneity hypothesis distinguishes the landing site of O-M in Icelandic from that of O-M in German and other SOV O-M languages which license parasitic gaps. Recall that it follows from (44) that O-M in Icelandic must be to an HR Case-marked position since leftward movement to a Caseless position is not licensed. This, however, is not necessary in German where movement to a Caseless position is independently licensed by (44). The assumption that O-M in German can be to an HR Caseless position accounts both for its A-properties, since it is an instance of substitution, and for the parasitic gap licensing, since as suggested by the previous discussion, parasitic gaps can be licensed from Caseless HR-positions. Under standard assumptions, parasitic gaps are assumed to be strictly licensed by A'-chains. If we are correct however, both O-M and bare quantifier constructions in French suggest instead that they can be licensed both from adjoined positions and from Caseless positions. If, as proposed by Haik (1985) and Williams (1990), parasitic gaps are manifestations of ATB extractions, then it is not surprising that only Caseless positions (whatever their status as specifiers or as adjunctions) license them. ATB extractions from a Case marked position within the adjunct would necessarily lead to a Case conflict, if the "landing site" of the extraction is itself independently Case marked. These constructions may thus be ruled out on independent grounds. Note that

132 Viviane Deprez

under this view, it is not surprising, that "parasitic gap" licensing is not strictly limited to operator/variable chains. Let us summarize briefly the discussion in this section. I have argued that although in Icelandic, NP O-M is movement to a Case marked position this does not have to be true in other languages. I have suggested that in German and in other SVO languages in which parasitic gaps are licensed, NP O-M can be movement to a Caseless HR-position. Under this view, the difference between O-M to a Case marked HR-position and O-M to a non-Case marked HR-position is that the former manifests strictly Α-properties, while the latter can manifest simultaneous A- and A'-properties. We thus arrive at the conclusion summarized in (56): (56)a. movement to Case-marked HR-positions manifests strictly Aproperties; (Icelandic O-M and standard A-movement) b. movement to Caseless HR-positions manifests ambiguous Aand A'-properties; c. movement to an HR'-position strictly manifest A'-properties.23 Note that it does not follow from (56) that intermediate functional projections in German must necessarily be non Case assigners. If Case marking is generally optional, German can be assumed to independently license O-M to a Case-marked HR-position, like in Icelandic. What is important, however, is that movement to a non-Case marked position HR-position in Icelandic is excluded under (44), while it is is permitted in German. It is this difference which, in our view, accounts for the additional A'-properties of German O-M.24

4.4.

Verb movement and dynamic minimality

Let us finally turn to the last crosslinguistic distinction manifested by certain cases of O-M, namely the overt dependency on verb movement. Recall that as described in section 2, Icelandic and Scandinavian O-M manifest an overt dependency on verb movement. If the verb moves then objects can move. If on the other hand the verb

Parameters of object movement

133

remains in it D-structure position, then objects cannot move. Why should this be? I propose that this necessary dependency is simply the overt manifestation of general constraints on NP-movement: in other words, it simply follows from the ECP. As proposed by Chomsky (1986), A'-movement may escape a barrier by adjunction. For Amovement on the other hand, intermediate adjunction is ruled out under improper movement. Let us assume this to be correct and to translate in our terms, to general substitution (movement to HR-position) and adjunction (movement to HR'-positions) rules. Recall from section 3 that our assumed model includes the VP internal hypothesis. To be more precise, assume that the D-structure position of subjects is the specifier of VP. The D-structure of a transitive verb phrase is then as in (57): (57) VP NP

V' NP

Assume further the provisional definition of Minimality given in (58): (58)

XP is a Minimality Barrier for A iff: A is c-commanded by X, the head of XP

It follows from (58) that VP is a Minimality Barrier for an object. Consequently an object NP cannot move out of VP unless it first adjoins to it. A subject, on the other hand, is not c-commanded by the V head; thus VP is not a Minimality barrier for it. Subjects, contrary to objects, can thus move to HR-positions since they do not need to adjoin to VP. Note that the assumption that VP is not a Minimality barrier for an NP since it is not c-commanded by its head, permits the movement of an object out of VP, if this object first moves into the Spec position of VP as illustrated in (59):

134 Viviane Deprez

(59) VP NP

V'

Due to the theta-criterion, this strategy for O-M is of course only available when the specifier of VP is not a theta-position. In our view, this is how passive or unaccusative constructions are derived. But the cases of O-M we are presently concerned with are all movements out of VPs with theta-marked subjects. The strategy of (59) is thus not available. From what we have said so far, it would seem that O-M to an HR-position should in fact never be possible, unless VP somehow ceases to be a Minimality barrier. As proposed on independent grounds by Baker (1986), there is in fact one situation in which a maximal projection ceases to be a barrier: namely when its head undergoes movement. Adapting somewhat Baker (1986), let us assume (60).25 (60)

XP is a Minimality Barrier if and only if it is headed by a lexical X ° , i.e. but not if it is headed by a trace

It follows from (60) that if the verb moves out of the VP projection, VP is no longer a Minimality Barrier for objects. Under this view, Minimality Barriers are computed dynamically, since they are affected by X°-movement. This has the following effect: when the verb remains within VP, O-M to an HR-position is excluded, since adjunction to VP is necessary and prevents any further substitution. But when the verb moves out of VP, O-M to an HR-position becomes possible, since adjunction is no longer necessary. Given this proposal, the relation between O-M and verb movement now becomes transparent. Whenever the main verb undergoes movement to a functional projection, O-M becomes possible. On the contrary, when the main verb remains inside VP, O-M is excluded by the ECP. As we have seen above, this is exactly the pattern observed in languages such as Icelandic and Scandinavian. When the main verb moves to I (or C), in Icelandic, then O-M is possible. But if an auxiliary or a modal occupies I, thereby blocking V-movement, O-M is impossible. O-M of pronominals in Mainland Scandinavian and

Parameters of object movement 135

Faroese is also subject to the same constraint. Recall that, as we discussed above, pronominal O-M is in our view an instance of head movement. It is thus not surprising that it displays a clear sensitivity to Minimality. Let us now consider some further consequences of this proposal. Since the dependency of O-M on verb movement follows from the ECP, it is expected to be universal. This makes a strict prediction with respect to the overall availability of O-M: (61)

Only languages which allow verb movement will manifest object movement

A corollary of (61) is that a large portion of what is traditionally known as the freedom of word order can now derive straightforwardly from the verb movement parameter, which itself, ultimately, reduces to the properties of functional heads. Consequently, there is no need to assume that languages which manifest free ordering of arguments but which are otherwise configurational have a flat structure. In other words, there is no need for an independent "configurationality" parameter. As it turns out, for most of the scrambling languages we have considered, the existence of verb movement has been argued for on independent grounds. See for instance Mahajan (1989) for Hindi, Choe (1986) for Korean, Fukui (1986) (among others) for Japanese and Vikner (1990) (among many others) for German, Dutch and the Scandinavian languages. Combined with our proposal on the Case marking properties of heads outlined in the previous section, the proposed theory makes predictions both for NP movement and for pronominal movement. The following classification emerges: (62)

1) 2) 3) 4)

Properties of beads

Movements

theta Case V-move NP to Pro O-M transparency HR + + + + + Icelandic, Hindi (German?) + + -/+* + Scandinavian (German?) . . . . English +

To this primary classification, the condition (44) on movement of NPs to Caseless positions (i.e the * in the table for German...), which

136 Viviane Deprez

ultimately is due to properties of heads (left or right headedness) adds a further nuance, which accounts for the variable properties of the chains created by O-M across languages. The prediction that languages with no verb movement will not manifest O-M is straightforwardly verified by the case of modern English where neither object NP nor pronouns can undergo movement. As is known, the rigidity of object order was not consistent throughout the history of English. Examples of the type (63), where pronouns have undergone a movement comparable to that of the Scandinavian pronouns were grammatical in Old and Middle English: (63)

*He loves me not.

As expected, however, the movement of pronouns here again correlates with the movement of the verb. It is well known indeed that Old and Middle English manifested the Verb second phenomenon, a clear sign of verb movement possibilities. Note that given our proposal, it is not necessary to assume that Old and Middle English pronouns had properties distinct from the pronouns of Modern English. Their mobility or absence of mobility simply correlates with that of the verb as a consequence of the ECP. Thus the overt morphological Case manifested by pronouns both diachronically and synchronically is simply irrelevant. The proposed ECP account on the mobility of objects seems, however, to run into some apparent problems when we consider in more detail O-M in languages such as German and Dutch. Recall that, as we noted above, the verb movement/ O-M dependency is not apparent in these languages, since O-M can occur both in main and in embedded clauses as well as in constructions with modals or auxiliaries. I believe however, that contrary to appearances these languages do not constitute counterexamples to our claim. First, it is now quite standard to assume that, in embedded sentences, German and Dutch manifest verb movement to a head final functional projection. If this is correct, the fact that O-M occurs in embedded sentences simply follows. The fact that auxiliary and modal structures do not block O-M may be somewhat more problematic. But it is a well known fact that untensed modals and verbal forms in infinitival constructions in German and Dutch undergo some kind of reanalysis process, the exact nature of which still remains debated. There is a long tradition of interpreting this verbal reanalysis as involving some sort of

Parameters of object movement

137

morphological process (Evers 1975; Den Besten 1983; Van Riemsdijk — Haegeman 1986). Under current assumptions, it is quite plausible to assume that such a process is amenable to an analysis in terms of incorporation (Evers 1975). If this is correct, and if this process extends to modals and auxiliary forms, then again, the fact that O-M can take place in these constructions in German and Dutch simply follows. Plausible support for this view involve restrictions on the distribution of VP adverbs in German. As is well known, no VP adverbs may occur in sentence final position or between an auxiliary or a modal and their dependent verbs, unless the Aux or the modal has overtly moved to C to satisfy the V2 constraint. These constraints are illustrated in (64): *daß sie dieses Buch last that she these books reads

schnell fast

*daß sie dieses Buch gelesen schnell hat that she these books read fast has c. *daß sie dieses Buch gelesen hat schnell As indicated by the English examples in (65), however, VP adverbs as opposed to NPs and PPs do not seem to be directionally restricted in their adjunction possibilities.26 That is, in English as in many other languages adverbs can occur on either side of the VP. (65)a. b. c. d.

She She She She

quickly read the book read the book quickly has quickly read the book has read the book quickly

Note that, if as suggested by the English data, there are no general restrictions on the direction of adverb adjunction, the facts in (64) appear rather mysterious. But if we assume that both the tensed verb and the untensed verbal forms undergo movement to a head final functional projection, then the impossibility of (64) receives an explanation. Under this view, (64) represent structures in which verb movement has failed to apply. (64) can thus be taken to provide plausible evidence for the leftward movement of non-finite verb forms in German. If such an analysis turns out to be feasible, the freedom of

138 Viviane Deprez

argument ordering in German and Dutch will fall under generalization (61).

5. Conclusion I have claimed that O-M involves a movement to the Spec or the head of a functional projection both defined as HR-positions. This, in my view, is what accounts for the crosslinguistic similarities found throughout various cases of O-M. To account for the differences, I have proposed that the Case marking property of heads further determines the properties manifested by the chain created by O-M: Case-marked HR-positions license Α-properties only, Caseless HR-positions license both A- and A'-properties and NP-movement to them is restricted by the directional properties of heads. Finally, I have proposed that O-M be universally constrained by verb movement, and that this constraint follows from the ECP and in particular, from a dynamic definition of Minimality. Under this view, the variable freedom of O-M across languages can be made to follow straightforwardly from independently required parametric properties of functional heads and from general constraints on movement. This proposal suppresses the need for an independent parameter of "non-configurationality", since it subsumes freedom of argument movement under general principles available at various degrees in all languages.27

Notes This paper was presented at the Tilburg Workshop on Scrambling, October 18-20 1990.1 would like to thank the audience of this workshop for challenging remarks and comments. Since this paper is an extension of some parts of my dissertation, I am also deeply indebted to the members of my committee, Noam Chomsky, David Pesetsky, Howard Lasnik as well as to Richard Kayne for extensive and fruitful discussions. Finally, I wish to thank Ken Hale, Arild Hestvik, Vaneeta Srivastav, Christiane Fellbaum and an anonymous reviewer for useful comments and native speaker intuitions. All remaining errors are of course my own. Partial support for this work was provided by a HRRF grant from Rutgers University. 1.

A fourth test often mentioned with regards to the properties of the chain created by O-M is concerned with connectivity or reconstruction effects. Anaphoric reconstruction (principle A) appears to be generally possible although it is limited in some cases (but see Webelhuth (1989), Moltmann (1990) and Santorini (1991) for

Parameters of object movement

139

differing positions on German, and Yoshimura (1989) and Saito (1990) for differing positions on Japanese). Reconstruction effects relevant to other principles of Binding theory (in particular principle C) occur in some languages (as for instance in Japanese (Saito 1991); but see Deprez (1989) for somewhat differing results) but not in others (cf. Mahajan 1989; Moltmann 1990). Since the question of whether reconstruction effects are uniquely associated with A'-dependencies remains controversial, and the data on which various claims have been made shows little consensus, I will leave this test aside pending further research and clarification of the data. Arguments that reconstructions effects are also manifested by Α-dependencies are given in Belletti and Rizzi (1986), Barss (1986) and Deprez (1989). Crucial examples, on which the claim that scrambling do not show reconstruction effects is based, involve, for instance, the following: (i)

*Gestem haben wir sich die Katze im Spiegel gezeigt 'Yesterday we have showed the cat to herself in the mirror.' (Santorini 1991)

(i) shows a structure were both the object NP die Katze and the dative anaphor have been scrambled out of VP. In (i), the anaphor is not directly c-commanded by its antecedent. It is thus unbound at S-structure. LF reconstruction, however, should allow it to move to a position lower than its antecedent and thus to be bound. Since the sentence is ungrammatical, (i) appears to show that scrambling cannot involve a movement to a position which permits reconstruction, i.e an intermediate A'-position. The conclusions drawn from (i), however, are controversial since as shown in (ii), when the anaphor is further moved to the topic position, a position standardly assumed to be an A'-position, reconstruction remains impossible: (ii)

*Sich haben wir die Katze im Spiegel gezeigt.

Given (ii), it is doubtful that the ungrammaticality of (i) should be attributed to limitations on reconstruction. Consequently, the question of whether a scrambled NPs can reconstruct or not remains open. 2. One must be careful when applying the floating quantifier test to eliminate the possibility of a prior Α-chain. Clearly indeed, although (i) manifests the A'-extraction of a subject, a prior Α-chain has been created: (i)

The drug dealers that the Mayor said will all be arrested soon are still alive and well.

3. Cf. Miyagawa (1989) for a discussion of numeral quantifiers in Japanese in relation to standard cases of Α-movement and Deprez (1989) for an extension to cases of scrambling.

140 Viviane Deprez

4. In this and other respects sentence internal object movement is opposed to long distance scrambling, which in languages in which it is possible (Hindi and Japanese), manifests properties of A'-chains: it neither permits new binding, nor repairs WCO violations, nor licenses floating quantifiers. This is illustrated again for Hindi in (i) trough (iii): (i) a.

b.

*apnii bahin-ne socaa [ki raam-ne mohan-ko dekhaa] selfs sister thought that Ram Mohan saw 'Self s sister thought that Ram saw Mohan.' *mohan-koi Mohan,

apnibahin-ne selfs sister

socaa thought

[ki that

raam-ne t{ Ram

sab-ko{ socaa everyone thought

[ki that

raam-ne t{ Ram

dekhaa] saw (ii) a.

nfuskiii bahin-ne His sister, dekhaa] saw

b.

*sab-koi Everyone,

[uskiit his

bahin-ne socaa sister thought

[ki that

raam-ne Ram

dekhaa] saw (iii) a.

b. c.

raam-ne socaa ki phal mohan saare Ram thought that fruits mohan all 'Ram thought that Mohan will eat all the fruits.'

khaa jaae gaa eat will

*phal raam-ne saare socaa ki mohan khaa jaae gaa *phal saare raam-ne socaa ki mohan khaa jaae gaa

See Deprez (1989) for similar facts in Japanese. For speakers who accept long distance scrambling out of tensed sentences the same facts obtain in German. The notion of long distance scrambling needs to be somewhat qualified: as shown by Mahajan (1989), scrambling out of infinitival sentences in Hindi manifest properties similar to those of sentence internal scrambling. This may also be true for German infinitival sentences as argued by Santorini (1991). These findings can nevertheless be taken to be consistent with the Α-chain properties of O-M since sentence external

Parameters of object movement

141

scrambling in these cases appears to show similarities with cases of standard NP raising (cf. Mahajan 1989 for details). 5. Kayne (1984) proposed that properties of chains be determined not by the properties of the landing site, but by the properties of the moved constituent. This proposal accounted for the fact that object clitics although moving to A'-positions do not licence A'-chains. It further accounted for the properties of the French Complex inversion construction which features a moved NP in pre-Comp position with A-chain properties: (i) [CP Mariei est-elle [ tt t partie]] See Deprez (1989) for a reanalysis of this construction within the framework assumed here. See also Taralsen (1986) and Holmberg (1986) for proposals similar in essence to that of Kayne (1984). 6. See also Deprez & Pierce (forthcoming) for rather compelling arguments supporting the VP internal hypothesis based on language acquisition data. 7. It is important to note that this account relies in part on the stipulation that adjunction to CP is impossible (vs May's (1985) condition on adjunction: "only one operator can be adjoined per projection level", ρ 81). Although Chomsky (1986) suggests that the ban on CP adjunction can be in part derived from thematic theory, this proposal has no bearing on matrix wh-movement. Apart from the stipulated ban on CP adjunction, the formulation of the ECP given in Barriers does not, in fact, rule out the adjunction of a second wh-element to a matrix CP, nor does the formulation of X* theory (X" --> Χ"* X' (Chomsky 1986: (lb), 3)) rule out the projection of an unlimited number of specifiers. It thus appears that, ultimately, an account of the uniqueness of wh-movement must rely on the uniqueness enforced by the featural relation that a wh-element entertains with a [+wh] complementizer (Speas 1990), i.e a [+wh] head can only entertain a single feature sharing relation with a single local [+wh] element. If so, however, uniqueness will obtain for a wh-element in Spec CP or an element adjoined to CP independently of structural considerations. See Deprez (1991) for further discussion of this point. 8. Mahajan (1990) follows a proposal by Chomsky (1989) (class lectures) which distinguishes C from all other functional projections on the basis that it does not participate with verbal inflection. This proposal conflicts with a number of analyses of the verb second phenomenon which place tense in C, (Holmberg (1986) among others) and with recent analysis by Rizzi (1990) and Kayne (1990) who propose that in English, an empty [-wh] complementizer is a projection of AGR. 9. For the sake of clarity, I will assume that this functional projection is AGR-O in Chomsky's (1989) terminology, although no particular importance is attached to the labeling of this functional node.

142 Viviane Deprez

10. The fact that pronominal O-M cannot cross a dative in a double object construction shown in (i) might arguably be taken as a further argument for the Α-status of the chain. (i)

*Jag I

gav gave

den it

slutligen Sara t finally Sara

Indeed, if as proposed by Larson (1989), the shifted dative argument is in some sense the subject of smaller VP, then the ungrammaticality of (i) can be attributed to a violation of binding principle A, in parallel to the ungrammaticality of (ii): (ii)

*A bookj was given Sara tf

The fact that O-M across a dative is possible in SOV languages such asHindi and German suggests, on the other hand, that the D-structure dative argument in these languages is an adjunct comparable to the prepositional dative in English, which does not interfere with NP movement. (iii) A book was given to Sara. The fact that a dative argument cannot be the antecedent of a direct object provides some support for this suggestion: (iv)

*Maria

hat

Mary

has

den Gasten ΙΟ to the guests

einander DO eachother

vorgestellt introduced

11. Pushed to its logical conclusion, this proposal also wrongly predicts that movement of a morphologically Case marked wh-element should manifest Α-chains properties: (i) Whom did you think that Mary met. To who do you think that she will talk 12. We expect the parametric variation in the Case marking properties of a given functional projection to have consequences on both synchronic and diachronic data. As a potential candidate for diachronic variation, consider, for instance, the object agreement projection which has been motivated in Kayne (1986) to account for past participle agreement in French. In Modem French, this projection is not a Case marking projection. As expected, O-M is impossible: (i) *Jean a la pomme mangie. Examples like (i), however were acceptable in Old and Middle French, suggesting that at this time, the AGR-O projection was a Case marking projection so that NP O-M was permitted.

Parameters of object movement

143

13. Swedish O-M is further blocked by the presence of a verbal particle: (i)a. *Jag skrev det I wrote it b. Jag skrev upp det.

upp up

(i) like (39) in the text could be assumed to follow from the head movement constraint. There are however some complications, since O-M across a particle appears to be possible in other Scandinavian languages. This is illustrated for Danish in (ii). (ii)

Jeg I

skrev wrote

det this

op up

These two languages exhibit further differences w.r.t particle constructions. Thus while the order V partivle NP is standard in Swedish, it never occurs in Danish. The differences between (i) and (ii) thus appears to be due to factor independent of the Head-Movement constraint. I will leave this problem open for further research. 14. As shown in (i), the presence of a preposition blocks clitic placement: (i)a. Je crois en lui I believe in him b. *Je lui crois en c. Je lui crois beaucoup I to him believe many Ί believe that he has many friends.'

d'amis friends

15. A somewhat surprising consequence of definition (2) is that head substitution as opposed to head adjunction is expected to have A'-properties. Although unusual, this proposal does not seem to create significant problems. It is clear indeed, that independent considerations such as the Head-Movement Constraint, or Minimality, will ensure that head-substitution must be strictly local. Thus long distance movement, one of the characteristic properties of A'-movement is ruled out in a principled way. Furthermore, it is unclear whether any other characteristic A'properties could ever be put to test. To my knowledge the only instance of head substitution proposed in the literature is the movement of I to C. As expected, this has a stricly local character. It is, moreover, quite unclear that this movement could ever be argued to fail typical A'-tests, such as for instance, the parasitic gap test, since given the position of I in the sentence structure, parasitic gaps may simply be ruled out independently by the anti-c-command constraint of Chomsky (1982). 16. A similar abstract lexical division is manifest in other well known languages. Thus for instance, French, which has two lexically distinct pronominal paradigm for stressed pronouns and clitics, has at least one pronominal form which is superficially ambiguous, namely the pronoun lui. Lui can appear in stressed positions as in (ia) or be a clitic as in (ib):

144 Viviane Deprez

(i)a. Lui, je ne veux pas le voir 'Him, I don't want to see." b. Je lui parle tous les jours Ί talk to him every day.' The properties of this pronoun remain, however, somewhat different from those of the Scandinavian pronouns. Indeed, the role of each of these two forms seems to be more clearly distinguished. When the pronoun is a clitic, it cannot remain in D-structure position even if it is stressed. (ii)

*Je parle (a) LUI tous les jours.

17. Further support comes from the fact noted by Holmberg (1986) that a pronoun which is modified cannot undergo O-M. (i) Hon kanner I know (Holmberg 1986)

inte not

[er you

tva]J * Hon kanner [er tva] inte two

In French, where the clitic pronominal series is distinct from the stressed series, only the latter allows modification. (ii) *Ils deux ont repondu Eux deux ont repondu 'They both responded.' Note incidentally that, as noted by Holmberg himself, (i) is a further problem for the morphological Case hypothesis: although the pronoun in (i) bears overt morphological Case, it cannot undergo O-M. 18. Further similarities are noted in Chomsky (1990). Like Α-chains, head-chains are homogenous chains. Consequently, like NP traces, X°+(o) traces cannot undergo LF deletion. 19. The proposal of this section and of the section 3.5 on verb movement has obvious consequences for clitic movement in Romance and other languages: first, it predicts that clitic movement will have the properties of Α-chains, and second, it predicts that object clitics should not exist in languages which do not allow verb movement. I explore these and other consequences in more detail in work in progress. 20. Note that (44) does not by itself suffice to rule out sentences like (45) in a language like English. We further need to rule out NP O-M to a Case marked substitution position, and pronominal movement to a head position. As I discuss in the next section, both restrictions can be traced back to the property of weak agreement in English, which precludes verb movement to I.

Parameters of object movement

145

21. To preserve the successivity hypothesis, it would be possible to claim that the derivation of the examples in (48) is more complex than assumed in the text. In particular, it could be assumed that the adjunct clause itself contains an A-moved parasitic gap which satisfies the Α-binding requirement within the adjunct clause. This more complex derivation, however, predicts that O-M should license derived subject parasitic gaps, a prediction which is clearly incorrect given the utter ungrammaticality of the following sentence: (i)

**Weil Maria because Mary

dieses this

gelesen worden zu sein] being read]

Buch book

[ohne e [without

zurückgeschickt back sent

t

hat has

Similar judgments obtain for the French bare quantifier construction discussed below. (ii)

*Marie Mary

a has

etre being

virifii t] verified]

tout everything

βχρέάίέ sent

[sans e [without

Moreover, positing a more complex derivation leaves unaccounted the double binding example in (iii) which was brought to my attention by Eric Hoekstra (pc). (iii) Ik I aan to

heb have

[elke every

mani z'n auto]i m a n ' s car

[zonder ei without

z'n, his

vrouw wife

terug te geven] back give

kapot gemaakt destroyed

In (ii), binding of the pronoun by the quantifier is possible, even though the parasitic gap does not bear the index of the quantified phrase. This example recalls similar examples in English where a quantifier inside an NP in an Α-position is allowed to bind a pronoun. (iii) Every man's mother thinks that he will become a genius. 22. Support for this analysis comes from the fact that as shown by the contrast in (i), tout cannot occur in object position: (i)a. *J' I

ai have

fait done

tout everything

b. J' I

ai have

tout everything

fait done

146 Viviane Deprez

There are, however, contexts in which tout seems to occur in object position. (ii)

Jean John

fait does

tout everthing

But, as I argue in Deprez (1989), the postverbal position of tout in (ii) is due to the movement of the verb. Argument for this conclusion come from the positioning of tout with respect to VP adverbs. As shown in (iii) the adverb mal must occur to the immediate left of an unmoved verb and the displaced bare quantifier must precede it. (iii) a. Jean α mal fait son travail vs b. *Jean a fait mal son travail 'John has done his work badly.' c. Jean a tout mal fait vs d. "Jean α mal tout fait 'John has done everything badly.' The fact that this ordering constraint must be preserved when the main verb has moved to I indicates that the structure of (iva) must be as in (ivc), not as in (ivd): (iv) a. c.

Jean fait tout mal vs Jean fait tout mal t

b. *Jean fait mal tout d. *Jean fait t tout mal

'John does everything badly.' If as I propose, the internal structure of tout is [tout PRO], the fact that tout is excluded from object position and the fact that, unlike regular NPs, it can occur in Caseless position follow. When occuring in a specifier position, PRO is protected from government by the maximal projection in which it occurs (QP). This maximal projection is a barrier, since it occurs in a position in which it cannot be L-marked. In object position, on the other hand, the containing maximal projection is L-marked so PRO is governed and excluded. Note incidentally, that this proposal will not overgeneralize to permit the occurence an object PRO in the specifier of an intermediate projection. Indeed, unless PRO is within a maximal projection, it will be governed by a head and excluded. 23. Note that under the proposed view, the Case properties of an adjoined i.e HR' position are ultimately irrelevant. If Case theory were to permit (possibly as a parametric option) Case marking to an adjoined position, the proposed theory predicts that such chains should still manifest strictly A'-properties. The case of long distance scrambling in Persian discussed by Browning (1991) in this volume may illustrate this latter possibility.

Parameters of object movement

147

24. The account of German proposed here is in fact quite similar to the one proposed by Webelhuth (1989). There is however, one important difference. Webelhuth proposes that the positions which manifest ambiguous A- and A'-properties are adjoined positions. This raises a serious problem with respect to cases of long distance O-M. As noted in footnote 5, long distance scrambling does not manifest any of the Aproperties characteristic of sentence internal O-M. It is nevertheless most plausible to assume that the landing site of long distance scrambling is an adjoined position. If so, Webelhuth's approach wrongly predicts that long distance scrambling should also manifest Α-properties. Under the view outlined here, on the other hand, Aproperties are assumed to pattern with HR-positions, i.e. substitution positions. But as argued in Deprez (1989) (cf also the discussion of the next section), movement to an HR-position is constrained under the ECP to be strictly local. Moreover, as standardly assumed, movement from an A'-position to an Α-position gives rise to "improper movement" which is ultimately ruled out by the Binding theory (Principle C), while movement from an Α-position to an A'-position is allowed. Adapting these notions to the theory proposed here, it follows that movement may proceed from an HR-position to an HR' -position, but not vice-versa, that is, adjunction may follow substitution but not the reverse. It is then sufficient to assume that the crossing of a sentence boundary by an extracted object implies a necessary adjunction (i.e. in standard terms, a passage through an A'-position) to predict that long distance scrambling cannot involve substitution and consequently can never manifests typical A-properties. The theory of ECP briefly sketched in section 2. above and developed in Deprez (1989) has precisely this result. 25. A more precise definition of Dynamic Minimality is given in Deprez (1989). 26. This absence of restriction may wanant a distinction between D-structure adjunctions and S-structure adjunctions. See for instance Speas (1990) for such a proposal. 27. As has been pointed out by numerous linguists and first and foremost by K. Hale (1983), the property of non-configurational languages is not limited to "freedom" of arguments. I believe, however, that the proposal developed here may be extended to account for languages which display other diagnostic properties of "non-configurationality". Recent proposals for a non-configurationality parameter have considered that languages which manifest clear "flat structure" properties and free null anaphora differ from conflgurational languages in the application of the projection principle (Hale 1983), or in the level of satisfaction of the thematic relations (Jelinek 1984; Speas 1986). For the languages considered in the text, which have all been independently shown to manifest structural properties in spite of their great freedom of word order, I have argued that one property of functional heads which is crucial to the mobility of arguments is their transparency to thematic relations. If functional heads are thematically transparent, verbs can move and arguments can move. It follows from the thematic transparency of functional heads that thematic relations can be satisfied within the functional structure after movement. I have assumed that in the languages studied in the texts, both lexical heads (such as V, Ν, Ρ and A) and functional heads (such as C, I, AGR etc) project in the syntax and are then related through movement) operations. In other words, the mapping

148 Viviane Deprez

between a syntactically projected lexical structure and a syntactically projected functional structure which in part determines grammatical relations, occurs through movement. In languages where V raising is possible, thematic relations can be satisfied either in the lexical structure (i.e within VP, NP etc) or in the functional structure. The apparent freedom of order result from this dual possibility. In languages where raising is not possible, thematic relations can only be satisfied within the lexical structure. Thus, the order is rigid. I suggest that "truly-non-configurational" languages differ from configurational languages in that, they only project a thematically transparent functional structure in the syntax but not a lexical structure. In these languages, the mapping between the lexicon and functional structure occurs not through syntactic projection and movement but through indexation) which can be contextual, i.e discourse related, or local, ie with a phonologically realized but non syntactically projected lexical item. Under this view, the non-constituency of VP or NPs, frequent argument drop, the richness of the Aux/verb compound and the rigidity of the ordering of its component parts (each separate functional heads) which are typical characteristics of non-configurational languages simply follow.

References Baker, M. 1988

Incorporation. Chicago: University of Chicago Press.

Baltin, Μ. 1982

"A Landing Site Theory of Movement Rules", Linguistic Inquiry 13.1.

Barss, A. 1986

Chains and Anaphoric dependency: On reconstruction and Its Implications. Doctoral dissertation, Massachusetts Institute of Technology. Belletti, A. — L. Rizzi 1981 "The syntax of "ne": Some Theoretical Implications", The Linguistic Review, 117-154. Belletti, A. — L. Rizzi 1986 "Psych-verbs and Theta-theoiy", Lexicon Project Working Papers 13, Cambridge: Massachusetts Institute of Technology. Bennis, H. 1986

Gaps and Dummies. Dordrecht: Foris.

Bennis, Η. — T. Hoekstra 1985 "Gaps and Parasitic Gaps", The Linguistic Review 4, 29-87.

Parameters of object movement

149

Chomsky, Ν. 1986a Barriers. Cambridge, Massachusetts: ΜΓΓ Press. Chomsky, N. 1986b Knowledge of Language. New York: Praeger. Chomsky, N. 1989 "Economy of derivations and representations", to appear in: Freidin, R. (ed.), Principles and parameters in comparative grammar. Cambridge, Massachusetts: ΜΓΓ Press. Deprez, V. 1989 On the typology of positions and chains. Doctoral dissertation, Massachusetts Institute of Technology. Deprez, V. 1990 The pattern of Spec-Head agreement in CP. Ms., Rutgers University, presented at the Girona Summer Linguistic Institute, Girona, Spain. Deprez, V. 1991 "Wh-movement; adjunction and substitution", to appear in: Proceedings of WCCFL X. Deprez, V. — A. Pierce 1990 "Negation and functional projections in early grammar", to appear in: Linguistic Inquiry. Cambridge: ΜΓΓ Press. Freidin, R. — R. Sprouse 1990 "Case Licensing", in: R. Freidin (ed.), Principles and parameters comparative grammar. Cambridge, Massachusetts: MIT Press.

in

Fukui, N. — M. Speas 1986 "Specifiers and Projections", MIT working papers in linguistics 6. Hale, K. 1983

Haik, I. 1985

"Walpiri and the Grammar of Nonconfigurational Languages", Natural Language and Linguistics Theory 1, 5-47.

The Syntax of Operators. Doctoral dissertation, Massachusetts Institute of Technology.

Haegeman, L. — H. van Riemsdijk 1986 "Verb projection Raising Scope, and the typology of Rules Affecting Verbs", Linguistic Inquiry 17.3, 417-466.

150 Viviane Deprez

Holmberg, Α. 1986 Word Order and Syntactic Features in the Scandinavian Languages and English, Department of General Linguistics, University of Stockholm. Jelinek, E. 1984 "Case and Configurationality", Natural Language and Linguistic Theory 2, 39-76. Kayne, R. 1984 "Chains, Categories External to S, and French Complex Inversion", Connectedness and Binary Branching. Dordrecht: Foris. Kayne, R. 1986 "Facets of Past Participle Agreement", in: P. Beninca (ed.), Dialect Variation and the theory of Grammar. Dordrecht: Foris. Kayne, R. 1989 "Null Subjects and Clitic Climbing", in: O.Jaeggli — K. Safir (eds.), The Null Subject Parameter. Holland: Kluwer. Kayne, R. 1990 Notes of English Agreement. Ms., New York: CUNY. Koopman, H. — D. Sportiche 1989 Subjects. Ms., Los Angeles: UCLA. Mahajan, A.K. 1989 On the A/A'distinction: Scrambling, Weak Cross Over and Binding in Hindi. Ms., Massachusetts Institute of Technology. Mahajan, A.K. 1990 The A/A' distinction and Movement Theory. Doctoral dissertation, Massachusetts Institute of Technology. May, R. 1985

Logical Form. Cambridge, Massachusetts: ΜΓΓ Press.

Larson, R. 1988 "On the Double Object Construction", Linguistic Inquiry 19. 335-392. Lasnik, H. — M. Saito 1984 "On the Nature of Proper Government", Linguistic Inquiry 15, 235-289. Lasnik, H. — M. Saito 1992 Move a . Cambridge, Massachusetts: ΜΓΓ Press.

Parameters of object movement

151

Moltman, F. 1990 Scrambling in German and the Definitness/Specificity Effect. Ms., Massachusetts Institute of Technology. Myagawa, S. 1989

"Light Verbs and the Ergative Hypothesis", Linguistic Inquiry, 20 659-668.

Pollock, J.Y. 1989 Saito, M. 1989

"Verb movement, Universal Grammar and the structure of IP", Linguistic Inquiry 20, 365-456. "Scrambling as Semantically Vacuous A' Movement", in: M. Baltin — Α.Kroch (eds.), Alternative Conceptions of Phrase structure. University of Chicago Press.

Saito, M. 1990

Long distance Scrambling in Japanese. Ms., University of Connecticut.

Santorini, B. 1991

Scrambling and INFL in German. Ms., University of Pennsylvania.

Sengupta 1990

Scrambling in Bang la. Doctoral dissertation, University of Massachusetts.

Speas, M. 1990

Phrase Structure in Natural Languages. Dordrecht: Kluwer.

Sportiche, D. 1988

"A Theory of Floating Quantifiers and its Corollaries for Constituent Structure", Linguistic Inquiry 19.3, 425-450. Taraldsen , T. 1986 "Som and the Binding Theory", in: L. Hellan — K. Kroch Christensen (eds.), Topics in Scandinavian Syntax. Dordrecht: Reidel, 149-184. Thrainsson, H. "VI, V2, V3 in Icelandic", in: H. Haider — M. Prinzhom M. (eds.), Verb Second Phenomena in Germanic Languages. Dordrecht: Foris. Vikner, S. 1990 Verb movement and the licensing of NP-positions in the Germanic Languages. Doctoral dissertation, Universit6 de Geneve. Webelhuth, G 1989 Syntactic Saturation Phenomena and the Modern Germanic Languages. Doctoral dissertation, University of Massachusetts.

152 Viviane Deprez

Williams, E. 1990 "The ATB Theoiy of Parasitic Gaps", The Linguistic Review 6, 265-279. Wyngaerd, G. van den 1989 "Object Shift as an Α-movement Rule", ΜΓΓ Working Papers in Linguistics, Volume 11.

On the nature of two A'-positions in Selayarese' Daniel L. Finer

Introduction Selayarese1 exhibits all possible permutations of verb, subject, and object in its surface word order, and although there is no case-marking overtly expressed on the NP arguments of the clause, the patterns of deployment of the verbal agreement affixes partially correlate with the placement of arguments within IP (=S), and thus provide interpretive clues that case-marking conveys in other languages. Despite this freedom of word order, it is possible to discern the basic clausal structure of the language, and I argue below that the canonical word order is VOS, with a hierarchical asymmetry obtaining between subject and object, i.e., the language is "configurational"; the verb and object form a constituent which is separate from the subject. Under the analysis assumed here, the subject originates within VP as the Specifier of VP, and V' contains V and its complements. The word order facts which depart from the basic VOS order offer evidence to make the argument that there are two preverbal positions that scrambled arguments may occupy. According to some of the diagnostics discussed below, a phrase in one of these two positions acts as if it were in an argument position (Aposition), and according to another diagnostic, it acts as if it were in a non-argument (A') position. A phrase displaced to the second position acts as if it were in an A'-operator position. Both positions are internal to IP, and the mixed A/A'-position linearly precedes the operator position. In addition, the operator position may be occupied by a whelement, and when a non-wh element occupies it, it receives a focus interpretation. The mixed position peripheral to the focus position appears to be a topic position; the sentence is "about" the element lodged here. Selayarese thus contrasts in this relative placement with more wellknown cases in which operator A'-positions are normally on the periphery of the clause.2 Grammatical theory must therefore recognize another operator position available at S-structure in addition to Spec of CP, the standardly assumed landing site for wh-movement. It would be tempting to conclude on the basis of these word order facts that Chomsky's (1986) analysis of wh-movement to Spec of CP as involving VP-adjunction is overtly supported by the Selayarese data;

154 Daniel L Finer

under such an approach the VP-adjoined site would be the operator position in question and Spec of IP could be analyzed as the mixed position. Data to be discussed in the final sections, however, suggest that these scrambling sites are in fact attached higher in IP. In particular, evidence is offered that Spec of IP is the operator position3 and that elements exhibiting the A/A* mixture are adjoined to IP. This adjunction site is the topic position, and I suggest below that the mixture of symptoms it exhibits will follow from its status as a non-operator in A*-position construed with a null pronoun. The crucial distinction between these two positions is thus not one of argument vs. non-argument but operator vs. non-operator. The discussion to follow first establishes the basic word order and clausal structure of the language, and then the properties of the two scrambling positions are considered in detail.

1. Orientation The agreement affixes mentioned above are given in (l). 4 (1)

Agreement prefixes Agreement suffixes

Is

2fam

ku-a

mu-ko

2hon,lpl. (inc) ri-ki

lpl (ex)

3

to-kang

la-i

The affixes are specified simply for their positional attachment, since while the prefix always cross-references the subject, there are other circumstances described below in which the suffix agrees with the subject. Examples of Selayarese sentences with definite direct objects are shown in (2). Here the verbal prefix encodes the agreement features (cofeatures) of the subject, and the verbal suffix encodes the φ-features of the direct object. These affixes will also license small pro in either position, as shown in (2e), for example.

On the nature of two A'-positions in Selayarese 155

(2)a.

la-?alle-i doe?-ifijo i Boso? 3-take-3 money-the h Baso? 'Baso? took the money.'

b.

ku-alle-i doe?-ihjo ls-take-3 money-the Ί took the money.'

c.

la-keo?-a i Baso? 3-call-Is h Baso? 'Baso? called me.'

d.

ku-keo?-i i Baso? ls-call-3 h Baso? Ί called Baso?'

e.

ku-keo?-ko ls-call-2fam Ί called you.'

Although the claim is that VOS is the basic word order of the language, variation in word-order does occur, as noted above, and any permutation of V,0,S finds a well-formed surface realization, as shown in (3). (3)a.

b. c. d. e. f.

la-alle-i doe injo 3-take-3 money the 'Baso? took the money.'

i Baso? (VOS) h Baso?

i Baso? la-alle-i doe injo. doe injo la-alle(-i) i Baso? i Baso? doe injo la-alle. doe injo i Baso? la-alle-i. la-alle-i i Baso? doe injo.

(SVO) (OVS) (SOV) (OSV) (VSO)

In line with earlier comments, I will take the D-structure of these sentences to be along the lines of (4).5

156 Daniel L Finer

(4)

According to the analysis to be followed, (3a) exemplifies the base order; the S-structure positions of the arguments are unchanged from their D-structure positions (inside VP in (4)), while (3b-e), I will argue in later sections of this paper, are derived by leftward movement and/or left adjunction of the NP-arguments.6 The next section establishes the verb-initial asymmetries and justifies aspects of the structure in (4).

2. Some structure 2.1. Two subject/object asymmetries The argument in favor of configurationality depends on two familiar diagnostics. The first concerns weak crossover. Coreference is possible between Ali and his in the verb-initial example in (5a) (in either the VOS or VSO interpretations), but under wh-movement (5b) or a leftward focussing movement (5c), such coreference possibilities are lost. (5)a.

Ia-janjang-i i Ali ando?-na 3-see-3 Ali mom-3 'HiS; mom saw A V (VOS) 'Alij saw hiSj mom.' (VSO)

On the nature of two A '-positions in Selayarese

b.

inai la-janjang ando?-na Wh 3-see mom-3 'WhOi did hiSj/.j mom see?'

c.

i Ali la-janjang ando?-na h A 3-see mom-3 'HiSj,.; mom saw A L I /

157

The structure formed by movement of the object NP to this preverbal position yields the judgement typical of weak crossover violations, and almost any formulation of the crossover constraint (except the "literal" one) will yield these results.7 In any case, the operator locally binds the relevant pronominal material in the subject as well as the variable at the extraction site, and neither of these structural positions binds the other. If there were c-command between the object position and the subject position, then there would be no weak crossover flavor to the sentences with the preposed arguments. Where c-command between the coindexed arguments in A-position does obtain, as in, for example Inai la-jafijang-ί ando?na? ('Who saw his mom'), the sentence is completely acceptable. A similar line of argument concerns the examples in (6). Here the reflexive anaphor is bound in its domain by the subject in (6a), but (6b) is ungrammatical, indicating that the reflexive is not bound. This follows from the structure we have been assuming — the rightmost NP being in the superior position c-commanding the first.8 (6)a.

b.

la-janjang-i kalen-na 3-saw-3 self-3 'Baso saw himself.'

i Baso h B.

*la-janjang-i i Baso? 3-saw-3 h B. 'Himself saw Baso?'

kalen-na self-3

As an aside, note that both of these diagnostics argue for a purely hierarchical approach to the phenomena of anaphor binding and crossover, since in each case, the linear relations are the opposite of those exhibited by the cases standardly discussed in the literature. That is, the anaphor precedes its antecedent in (6a), and in (5b, c), the

158 Daniel L Finer

pronominal element occurs to the right of the variable created by the extraction, thus illustrating that an approach to crossover based on a "leftness" predicate cannot be universally applicable (cf. Chomsky 1976).

2.2.

[WV....J

Having provisionally established the existence of a hierarchical asymmetry between subject and object in the organization of the Selayarese V-Projection, I now turn to the question of the orientation of that projection, and here I will argue that the VP is head-initial. The argument is largely based on alternations in verbal morphology when the verb takes an indefinite direct object or is intransitive. In the previous data, the verbal prefix agrees with the subject and the suffix agrees with the object. When the object is indefinite, however, this agreement pattern shifts to one where the suffix encodes subject agreement features, and the affix (a)ng/a?, referred to as an "intransitivizer" by Mithun — Basri (1985), appears as a verbal prefix.9 This same pattern of affixation occurs when the verb in question is intransitive, cf. (7d). (7)a.

(a)ng-alle-ko doe? int-take-2fam money 'You took (some) money.'

b.

(a)ng-alle-kang doe? int-take-lpl money 'We took (some) money.'

c.

(a)ng-alle-i doe? i Baso? int-take-3 money h Baso? 'Baso? took (some) money.'

d.

ak-kelong-ko int-sing-2fam 'You sang.'

On the nature of two A '-positions in Selayarese

159

Note that the suffixes in (2) encode the φ-features of the objects while in (7), the suffixes encode the φ-features of the subjects. In each case, the affix is drawn from the same series; that is, -ko, for example, functions as the object agreement marker in (2d) and as the subject agreement marker in (7a) and (7d). For an account of the parallelism between the agreement patterns of the intransitive verbs and the verbs with indefinite direct objects, I will appeal to Baker's (1985) notion of "abstract incorporation". That is, I propose that Selayarese indefinite direct objects are incorporated into the V, rendering it intransitive.10 Such incorporation is termed "abstract" because there is no reordering of the relevant constituents, and the affected NP stays external to the peripheral morphology of the verb. We might further speculate that only definite NP's can be marked with accusative Case,11 and so the indefinite NP incorporates in order that the Case filter may be satisfied (i.e., obviated). The patterns observed above with a postverbal direct object, effects of the incorporation hypothesis, play a large role in the discussion of the implications of some deviations from the basic word order in the following subsection.

2.3.

Subject and object preposing

Selayarese freely allows the subject or object to occur in a preverbal position, and ννΛ-elements occur here as well, as can be observed in the crossover examples. The data in (8) illustrate subject preposing. (8)a.

b.

i Baso? la-alle-i doe?-ifijo h B. 3-take-3 money-the 'Baso took the money.' i Baso? (a)ng-alle(-i) doe? h B. int-take(-3) money 'Baso took (some) money.'

Here, in the definite object construction in (8a), the subject NP simply moves to a lefthand preverbal position. The verbal morphology remains the same. (8b) illustrates a fronted subject in the indefinite object construction, and here there are two possibilities, one with the

160 Daniel L Finer

agreement affix and another without. It will be argued below that the relative position of the fronted element determines the occurrence of this affix. When the preposed element occupies the focus position, the affix is not present, and when the preposed element occupies the topic position, the affix is present. (9b) shows a direct object that has been focussed, and here note also that the verbal agreement suffix drops. (9)a.

b.

la-taro-i 3-put-3 doe?-injo money the

doe?-injo money-the la-taro 3-put

ri lamari in cupboard (*-i) ri (-3) in

i Baso? h B.

lamari i Baso? cupboard h B.

a, b = 'Baso put the money in a cupboard.' A tentative account of the loss of the agreement marker might appeal to an analysis involving the assumption that the object agreement marker prevents proper government of the empty category left by A'-movement (similar in spirit to the prohibition noted in Jaeggli (1982) against extraction out of clitic-doubled constructions), or that it interrupts the binding relation between the A'-binder and the variable resulting in a case of vacuous quantification, but we will not pursue this in detail further here.12 Foreshadowing slightly, however, I will assume that the agreement marker drops as one reflex of an extended form of Spec-head agreement between Spec of IP and I when a focussed argument moves into Spec of IP. If an XP related to the absolutive argument (definite object or intransitive subject) moves into Spec of IP, the affix will not surface. If another element occupies Spec of IP, it will have no effect on the occurence of the affix. Clauses through which A'-movement takes place in Palauan and Chamorro, both Austronesian languages, also show a particular form of agreement (see Georgopoulos 1985; Chung 1983), and the dropping of the verbal suffix can be viewed as the Selayarese instantiation of this Austronesian "wh-agreement".

On the nature of two A '-positions in Selayarese

2.4.

161

Object fronting and incorporation

When the indefinite object is fronted, an unexpected shift in the verbal morphology is exhibited. Compare (10a) and (10b) with the example involving a definite object in (10c): (10)a.

*doe? (a)ng-alle-i i Boso? money int-take-3 h B. 'Baso took (some) money.'

b.

doe? la-alle i Baso? money 3-take h B. 'Baso took (some) money.'

c.

doe?-injo la-alle i Baso? money-the 3-take h B. 'Baso took the money.'

Two things are of note in the above examples. First, the agreement pattern exemplified in (7) is incompatible with a fronted indefinite object, as shown in (10a). (10b) shows that the "intransitivizer" (a)ng/a? is not present, and the subject-verb agreement affix appears as a prefix instead of the suffix shown in (7), even when the object is indefinite. Second, note that this verbal morphology is identical to that found in the object fronting and wh-movement examples in (5b), (9b) and (10c), which in turn, is identical (modulo constraints on the verbal suffix discussed earlier) to that of the examples involving a definite object in situ. The following examples illustrate this perhaps more clearly: (11)a. n-aro-ko int-put-2fam b.

doe? money

ri lamari in cupboard

doe? mu-taro(*-i) ri lamari money 2fam-put(*-3) in cupboard a, b = 'You put money in a cupboard.'

162 Daniel L Finer

c.

mu-taro-i doe?-ifijo ri lamari 2fam-put-3 money-the in cupboard 'You put the money in a cupboard.'

i Boso? hΒ

In Basri — Finer (1987), it was suggested that this agreement flip arises because the empty category left by this A'-movement is a variable, and hence counts as a definite NP for agreement purposes. The definite agreement pattern is thus conditioned, and as above, the object agreement marker does not surface when the direct object is fronted. Adoption of the incorporation analysis suggested above, however, provides another explanation for the shift in agreement morphology if it is assumed, following Baker (1985), that the verb must govern an NP before it can undergo incorporation. In example (lib) immediately above, following the structural assumptions made earlier, there is no government between the verb and the preposed indefinite NP, and so there can be no incorporation. The verb remains transitive, and the appropriate morphology is selected. The conclusion, based on this morphological alternation that correlates with the position of the direct object, is thus that the verb governs the object NP in the V-initial structure as well as in the subject initial structure in both forms of (8a), but not in the examples showing the preposed direct objects. And this structural relation follows if we assume that the verb is initial in all of its projections, as illustrated in (4). Another set of data constituting independent evidence favoring the incorporation analysis will be discussed immediately.

2.5.

PP fronting and incorporation

In the V-initial structures shown in (12a) a PP occurs at the end of the clause. When the PP undergoes fronting, an agreement affix from the series of suffixes in (1) occurs attached to it, as shown in (12b).

On the nature of two A '-positions

(12)a. mu-pallu-i 2fam-cook-3

juko fish

in Selayarese

ri in

kirong pan

ri in

papaluang kitchen

163

ifijo the

ri kirong-i/*?ko in pan-3/2fam mu-pallu juko injo 2fam-cook fish the ri in

papaluang- i/ kitchen-3/-2fam

'You cooked the fish

in in

the the

pan. kitchen.

These facts can be accommodated along the following lines. Assume that a feature of Selayarese grammar is that agreement features are in general always available to XP's which function as predicates, secondary or otherwise. The value of the φ-features is determined by the argument that the PP, for example, is predicated of in a government relation (cf. Williams' 1980 work on predication). That is to say, i.e, 'in the pan' is attached low inside V', in mutual c-command with 'fish'. When a secondary predicate is governed from a theta-position, however, the features do not receive overt expression, but when the PP occurs preverbally, the government relation is broken and the features, specified at an earlier level of representation, are realized phonologically. 'In the kitchen' may of course be attached high or low in the VP, and so either agreement affix is possible once the PP is preposed. If one is cooking the fish in the kitchen, one is in the kitchen, and so is the fish. Of course, 'in the pan' may in principle be attached higher up, but then the pragmatics attached to the sentence become bizarre ('Being in the pan, you cooked the fish'), and this becomes readily apparent once the PP is preposed and the agreement affix is displayed. From the data in (12) then, it looks as though predication relations will determine the distribution of the agreement affixes in these cases of PP preposing.13

164 Daniel L Finer

This approach breaks down, however, when the object is indefinite. Consider (13a-b). ri kirong in pan (13)a. a?-pallu-ko int-cook-2fam

juko fish ri papalluang . in kitchen

ri in

kirong- ko/*i pan-2fam/3

b.

a?-pallu juko int-cook fish ri in

papalluang- ko/*i kitchen in the pan.'

'You cooked fish „ in the kitchen.' The overt expression of agreement features corresponding to the subject ('you') on the PP 'in the pan' is well-formed in this case, but the appearance of features corresponding to the object is ill-formed. This follows directly once we reconsider the incorporation analysis offered above. The internal argument is essentially swallowed up when it is indefinite, and it is thus unable to condition the presence of agreement features, regardless of the predication relations. I assume that the theta-position associated with the indefinite direct object loses its potency to control the agreement features once incorporation occurs. The verb is essentially intransitive at this point, and the subject is the only argument that is licitly able to determine the form of the agreement features on the PP. If the indefinite object were still syntactically viable at the time of PP-fronting, we would expect the PP to show the same agreement as it does when the object is definite (as in (12b)). The relation between the preposed PP and the argument it is construed with is certainly not as transparent in these cases as it is when the direct object is definite; in this case the semantic relation is obscured by independent factors which are syntactically conditioned.

On the nature of two A '-positions in Selayarese

165

Given the form of the agreement that does surface in these cases, I conclude that the indefinite NP is in fact subject to incorporation under government, and the only configuration where the verbal agreement patterns indicate incorporation (and therefore government) is the one where the object NP is postverbal. Hence, I conclude that the VP is Vinitial.

3.

Two preverbal positions

The discussion so far has considered in detail only a subset of the possible word orders of Selayarese, so it will be the job of this section to discuss other possibilities. We have established so far that a VPconstituent (or more accurately, a V'-constituent) plays a role in the prominence relations between subject and object, and it has been argued that the VP is V-initial. What remains is to show that the two preverbal positions mentioned in the introduction exhibit the properties that they were claimed to have. As things stand now, the operator position that we have been alluding to in the discussion of crossover, proposing, etc., could well be Spec of CP, the standard operator position; there has been no evidence yet presented that the landing site for the preposed elements is anything other than Spec of CP. The following discussion first establishes the distinction between the two positions, and then considers their structural positions. Reconsider the data in (3), repeated here as (14), in particular (14d) and (14e). Note that both arguments precede the verb in these examples, unlike the cases discussed in section 2, in which only one argument was fronted. (14)a. la-alle-i doe ifijo i Baso? 3-take-3 money the h Baso? 'Baso? took the money.' b. c. d. e.

ι Baso? la-alle-i doe ifijo. doe ifijo la-alle(-i) i Baso? i Baso? doe ifijo la-alle. doe ifijo i Baso? la-alle-i.

166 Daniel L Finer

A feature of Selayarese which was observed above is that the verbal suffix is absent when an element that it references is moved to the focus position. As has been shown, the focussed XP may be a whphrase or another argument, as the above examples show. Assuming the operator status of this position, this fact can be assimilated to the other cases of Austronesian wh-agreement as noted above. Cases of longer wh-dependencies are shown in (15), and here the agreement suffix cannot be present either. (15)a. Apa mu-kua la-alle(*-i) i Ali Wh 2fam-say 3-take-3 h Ali 'What did you say that Ali took?' b.

Inai mu-kua (a)ng-alle(*-i) doe Wh 2fam-say int-take-3 money 'Who did you say took (some) money?'

As shown in (14e), on the other hand, a topicalized object does not condition such wh-agreement, and the apparent optionality of the agreement suffix shown in (14c) and (8b), repeated here in (16), is thus only apparent. Its occurrence is dependent on the position which the element it corresponds to occupies. (16)a. doe ifijo la-alle(-i) i Baso? (=(14c)) money the 3-take(-3) h B 'Baso? took the money.' b.

i Baso? (a)ng-alle(-i) doe? (= (8b)) h B. int-take(-3) money 'Baso took (some) money.'

A preposed direct object or subject of an intransitive verb must be in either the topic or focus position, and if it occupies focus, there will be no suffix. If it occupies topic, the suffix will be present. This is schematically sketched in (17): (17)a. (XP) NPj V(+*0Ci)

On the nature of two A'-positions in Selayarese

b.

167

NP; (XP) V+oti

Now that the above discussion has established a diagnostic for each of the positions in question, I shall argue in this section that in the examples similar to (14d) and (14e), both preverbal NP's are in A'positions; the rightmost NP (focus) is in an operator position, and the leftmost NP (topic) is in a non-operator position. As before, much of the evidence will come from crossover and reflexive binding. (18a, b) show examples where the focussed element (the rightmost XP of the two preverbal positions) is the direct object, and note that in these cases the possessive pronoun cannot be construed with the focussed element. If, as in (18c), the fronted direct object occupies the leftmost position ("topic"), there is no such violation, and coreference between the relevant elements is possible.14 (18)a. ando?-na inai la-janjang mom-3 Wh 3-see 'Who1 did hiSj,.j mom see?' b.

ando?-na i Ali la-jafijang mom-3 h A 3-see 'HiSj/.; mom saw ALI;.'

c.

i Ali ando?-na la-janjang-i h A mom-3 3-see 'HlSj/j MOM saw Ali;.'

The only difference between the examples in (5b, c), repeated below as (19a, b), and (18a, b) is that the argument corresponding to the subject of the clause is preposed to the external topic position in (18a, b). The interpretive possibilities are identical: no coreference is possible between the relevant items in either set of examples. (19)a.

inai la-janjang ando?-na Wh 3-see mom-3 'Who; did h i s ^ mom see?'

168 Daniel L Finer

b.

i Ali la-janjang ando?-na hΑ 3-see mom-3 'HiSj/., mom saw ALI/

The basic crossover constraints are applicable to the representations underlying (19), but they do not apply so straightforwardly to (18a, b), since at S-structure, the elements in the rightmost proposed position do not c-command the elements in the external position, and so there would be no binding in the relevant sense anyway. This failure of binding, coupled with the judgments given, while it cannot derive a crossover violation, can be seen as a reflex of the operator status of the immediately preverbal position. It is generally assumed that pronouns can be treated as bound variables only if they are c-commanded by an operator. Assuming that Ali is associated with the focus operator, then the third person pronominal possessive cannot be construed with the operator unless it is bound (not merely coindexed with) it. If Ali were not in operator position, then simple coindexing between these two elements should yield an acceptable output. It could also be argued that the element in the focus position undergoes further raising at LF to a clause-peripheral position c-commanding both topic and focus. If so, then the resulting structures would be analyzable as crossover violations of the standard sort. In any case, since no coreferential interpretation is possible here, as would otherwise be expected, I conclude that the immediately preverbal position is an operator position. Consider (18c) in this light. Here, if Ali were in an operator position, then a crossover violation should be detectable since the NP would be the local binder of two coindexed elements, one of which would be a variable. The coreferential interpretation is possible, however, and this indicates that the peripheral XP position is not an operator position. This conclusion is reinforced by the fact that when the wh-form occupies the external position, an echo question interpretation is derived, not one where the wh is in any sense an operator. In such structures, there is no crossover effect.15 Another argument comes from the behavior of anaphors. It is well known that anaphors that have been raised into an operator position may be interpreted as being bound by a lower argument that ccommands a trace of the moved element (see, e.g., van Riemsdijk — Williams 1982; Barss 1986 for discussion and different analyses). Such "reconstruction"16 is not possible, however, when the constituent

On the nature of two A '-positions in Selayarese

containing the anaphor has been displaced to an argument position, when it occupies an Α-position or a non-operator A'-position. Selayarese, reconstruction from one of the preverbal positions possible, as shown in (20a), where kalenna, the direct object, can bound by Ali, even though the former c-commands the latter at structure (assuming that the preposed position is external to VP, argued above). (20)a. kalen-na la-jafijang self-3 3-see 'Ali saw himself.' b. c.

169

or In is be Sas

i Ali h Ali

*kalen-na la-janjang-i i Ali. *kalen-na i Ali la-jafijang-i.

Binding cannot be (re)constructed in (20b, c) where the anaphor has been preposed to topic position (note the presence of the agreement affix), however, and this fact suggests that the anaphor does not occupy an operator position in these structures. If this external position were an argument position, this fact would be readily explicable, since binding proceeds from the new alignments after movement to an Aposition takes place, and these examples could be analyzed as simple violations of Principles A and/or C. The example in (21a), however, offers evidence that the so-called topic occupies an A'-position. Here the anaphor, occupying the focus position, is linked with the VP-internal subject position, and it is construed with the c-commmanding NP Ali in the higher topic position. Regardless of whether the anaphor reconstructs into the postverbal subject Α-position, the sentence is ungrammatical. (21)a.

b.

*i Ali kalen-na la-janjang-i h A self-3 3-see-3 'Himself saw Ali.' i Ali kalen-na la-jafijang h A self-3 3-see 'Ali saw himself.'

170 Daniel L Finer

Even though i Ali c-commands both the S-structure and D-structure (reconstructed) positions of the anaphor, the sentence is ill-formed, indicating that the topic position is not an appropriate binder (i.e., not an Α-position). This binding failure in (21a), plus the failure of reconstruction in (20b, c) point toward the conclusion that the so-called topic position is neither an Α-position nor an operator position, but a non-operator A'-position. (21b), however, contrasts sharply with (21a). Here the focussed anaphor corresponds to the VP-internal direct object, the topic corresponds to the subject, and the sentence is grammatical. This apparent paradox can be eliminated once we consider what the binder of the anaphor is. We have concluded that the topic position is not an Α-position from (21a), but one position that is a possible binder of the anaphor in (21b) is the postverbal subject position, which is linked to the preposed topic. This subject position is uncontroversially an Aposition, and it c-commands the variable linked to the focussed element. The element occupying this Α-position is thus an appropriate binder. The occupants of the Α-positions internal to VP in structures with focussed and topicalized elements is one residual matter to be addressed. The category created by focus movement is a variable; relevant diagnostics for an operator-variable chain are met. Assuming that the topic position is indeed an A' position, as argued above, then, of the inventory remaining elements, pro is the most appropriate. Topicalization structures, by this reasoning, would involve a basegenerated NP in A'-position that binds a null pronoun in A-position. Neither operator-variable nor Α-chain symptoms would be expected.17 Of the two preverbal positions in Selayarese, then, the leftmost is a non-operator A' topic position and the rightmost is an A' operator focus position. The chart in (22) summarizes their respective properties. (22) Focus (XP XP V)

Topic (XP XP V)

anaphor reconstruction

Yes

No

crossover violation

Yes

No

"wh-agreement"

Yes

No

On the nature of two A '-positions in Selayarese

171

In addition, both of these preverbal positions appear to be within IP. Evidence from embedded clauses in the language shows that the complementizer precedes both preposed arguments. As the examples in (23) show, all of the cases discussed so far are embeddable beneath a complementizer in examples with subordinate clauses. (23)

mu-kua 2-say

muko a. la-alle-i doe injo comp 3-take-3 money the

i Ali h Ali

b. ι Ali la-alle-i doe injo. c. doe injo la-alle(-i) i Ali. d. i Ali doe injo la-alle. e. doe injo i Ali la-alle-i. 'You said that Ali took the money.' While it is therefore unlikely that the landing site for wh and focus movement is Spec of CP, the problem now is to determine precisely where the preposed arguments lodge. Given the structure in (4), repeated below as (24), the only available base-generated preverbal position within IP is Spec of IP. (24)

Following usual assumptions, Spec of IP is an Α-position, but neither of the two preverbal positions show Α-position diagnostics. If we wish to exploit Spec of IP in the analysis, therefore, we must claim

172 Daniel L Finer

that it is an A'-position of some sort, either an operator position, or a non-operator position. If we assume that it is the topic position (nonoperator), then we could assume, as mentioned in the introduction, that a VP-adjoined site could host the operator (see May 1985; Chomsky 1986 for discussion). These assumptions, as well as fairly conservative assumptions about clause structure, thus yield (25), where XP; has been moved from a position inside VP and adjoined to VP. (25)

[cpfo YP [VP XPj [VP V...[e],..]]]]

Other data from elsewhere in the language, however, suggest that these two preverbal positions are in fact located higher in the structure. In particular, on the basis of the constructions discussed in the next section, I will suggest that the Topic element is in fact adjoined to IP and that the operator-like focus position is Spec of IP.18

4. Higher attachments Selayarese displays a construction which in some ways is similar to English Exceptional Case Marking (ECM) constructions in that it suggests that a higher verb is able to govern into a subordinate IP. Although this construction differs from ECM in that it does not obviously involve a subordinate infinitive, I will refer to it as "ECMlike" for want of a better phrase. A subset of the verbs which select a full CP show an alternation whereby an agreement suffix agreeing with the NP in focus in the subordinate clause occurs on the higher verb. (26a) shows the full clausal form, and (26b) exemplifies this ECM-like construction. Note that the agreement suffix is missing from the lower verb in both cases, confirming the claim that it is the focussed element which is referenced by the agreement suffix on the upper verb. The subject of a fully transitive verb is focused in the lower clause and is referenced the upper agreement element in (26c).19 (26)a. la-isse?-i lako ng-erang-a loka 3-know-3 comp int-bring-ls banana 'He knows that I brought bananas.'

On the nature of two A '-positions in Selayarese

b.

la-isse?-a ng-erang loka 3-know-ls int-bring banana 'He knows I brought bananas.'

c.

la-isse?-kan to-alle-i bembe-na 3-know-lpl lpl-take-3 goat-3 'He knows we took his goat.'

173

Additional evidence in favor of the claim that it is the focussed element which is referenced by the upper agreement element comes from the data in (27). Here, the direct object bembe/bembena is in focus, while the agreement suffix on the upper verb shows the features of the subordinate subject, which I will assume corresponds to the leftmost of the two preverbal positions (topic) discussed above. bembe (27)

*la-isse?-kan

'

to-?alle

bembe-na goat 3-know-lpl

1 pi-take I goat-3 a goat.'

'He knows we took „ his goat.' . If the agreement suffix could reference the so-called "topic" position, then (27) should be grammatical. Assuming that government between the upper verb and an NP is necessary before the verb can bear the corresponding agreement affix, it follows that the focussed element is governable by the upper verb at some point in the derivation. The VP-adjoined site shown in (25), however, is too far away from the upper verb for government to obtain even if we assume that the CP is transparent; an IP boundary as well as I, another zero-level category, intervene between the adjoined NP and the upper verb. It might be possible to obtain government if the focussed element moves to Spec of CP across Spec of IP (topic), but then, ceteris paribus, we would expect that the "topic" could still

174 Daniel L Finer

occupy its preverbal position (putatively Spec of IP) in this event. This expectation, however, is unfulfilled: (28)

*la-isse?-kan i Baso? la-janjang 3-know-2pl hΒ 3-see 'He knows Baso? called us.'

Movement of this sort should be possible, given the proposed structure in (25). According to Rizzi's (1990) formulation of the ECP, / should be able to properly head-govern the trace left by movement across the preposed NP to Spec of CP (note that wh-movement out of a subordinate clause is possible, as shown in (18)), and even if IP were to count as a barrier, it would be the only one crossed. Another factor which weighs against movement from VP to CP here, followed by government into Spec of CP, is that while whmovement from the focus position through Spec of CP is possible for most if not all verbs which select a CP, only a subset of these verbs participate in this ECM-type construction. Some ad hoc mechanism would therefore be required in order to block examples such as that in (29). (29)

*la-pao-kan to-alle-i bembe-na 3-say-2pl 2pl-take-3 goat-3 'He said we took his goat.'

Another way in which this construction differs from wh-movement constructions is that wh-movement constructions may be of the pattern shown in (30a), where the item remaining within IP is the focussed constituent ((30b) or (30c) is the non-question form of (30a)). (30)a. inai mu-kua tedong injo la-sumbele Wh 2fam-say buff the 3-slaughter 'Who did you say slaughtered the buffalo?' b.

mu-kua muko i Ali tedong injo la-sumbele(*-i) 2fam-say comp h A. buffalo the 3-slaughter 'You said that Ali slaughtered the buffalo.'

On the nature of two A'-positions

c.

in Selayarese

175

mu-kua muko tedong ifijo la-sumbele( *-i) i Ali 2fam-say comp buffalo the 3-slaughter h A. 'You said that Ali slaughtered the buffalo.'

While a non-focussed NP can undergo wh-movement, as in (30a),20 it is not accessible to the construction under discussion, as shown in (27). This ECM-like construction, therefore, is not a special case of movement to Spec of CP (or wh-movement). I suggest instead that verbs such as -isse?- optionally induce CP transparency, and that there is no movement of the focussed item into Spec of CP in the resulting constructions. Since Spec of CP is not implicated in these constructions, it cannot serve as a higher landing site for a focus element adjoined to VP, and therefore the VP-adjoined position cannot be a candidate for the Selayarese focus position, since it must be governable from the higher V-position. I suggest that the focus position in Selayarese is in fact Spec of IP, which, assuming CP transparency, is governable by the upper V.21 The so-called topic position must be to the left of Spec of IP, and since Spec of CP is unavailable (see discussion surrounding the examples in (23)), I conclude that an IP-adjoined position is available. The structure of the cases exhibiting both arguments pre verbally, such as (31a), for example will then be as in (31b). (31c) shows the ECM-type structure with both arguments preposed (cf. (28)). (31)a. i Baso? doe ifijo la-alle hΒ money the 3-take 'Baso took the money.' ν...[β],...ρΐΌ,]]]]

b.

[ C P C [jp NPj[n> N P j I [ V P

c.

W+oci([CpC)[IpNPj[IpNPi I [VP V...[e]i...prOj]]](])

Examples such as (28), the relevant portion of which is schematized in (31c), will be blocked by the failure of government to obtain between the upper V and Spec of IP when an XP is adjoined to IP, assuming a uniqueness condition on head-government such that a head can only govern one XP at a time, and then the closest one (defined in terms of C-command). The adjoined element (NPj) in (31c) thus preempts head government of the Spec of IP (ΝΡ;) by V. When there

176 Daniel L Finer

is no element adjoined to IP, then Spec position is governable by the upper V, under the assumption of CP transparency.22 These same considerations of uniqueness and locality will trigger a failure of head-government if the element in Spec of IP is moved into Spec of CP across the IP-adjoined element. An ill-formed wh-extraction such as (33) will be thereby blocked. Here, the phonologically null C will govern i Ali, not the empty category left by movement of the wh.23 (32)

*tedong te?ean mu-kua i Ali la-sumbele buff wh 2fam-say h A. 3-slaughter 'Which buffalo did you say that Ali slaughtered?'

On the other hand, when nothing is adjoined to IP, government of Spec of IP by a higher head is possible, and no violations ensue.

5.

Conclusion

The foregoing discussion has established that the canonical syntactic structure of Selayarese is verb-initial and configurational, and that there are two preverbal clause-internal attachment sites for elements scrambled out of their postverbal D-structure positions. These two syntactic positions correlate with the discourse notions of topic and focus, and the discussion highlighted their syntactic properties. The focus position clearly exhibited properties normally associated with A'positions, such as binding reconstruction and vulnerability to weak crossover, and while the topic position did not pattern in these ways, it did not reflect pure Α-position properties either. The particular distinctions between the two positions, however, reside in the way that A' properties are manifested in the grammar. Both positions exhibit A'-properties, but only one position, the focus position, displays operator symptoms, and the topic position can be viewed as a non-operator A'-position. For Selayarese, there is thus no A/A' dilemma with respect to scrambling; the distinction is with respect to operators and non-operators. The nature of the Α-positions that these A'-positions are linked to was addressed as well. Only the Α-position bound by the element in the focus position can be a variable, and in fact, given the nature of its

On the nature of two A '-positions in Selayarese

Υ1Ί

binder (an operator), it must be a variable. The Α-position bound by the phrase in topic position cannot be a variable since its binder is not an operator. Further, it cannot be a trace derived by NP-movement, since it is locally bound from an A*-position, and so I conclude that the topic element binds a null pronoun in A-position. Other arguments were directed toward establishing the structural geometry of clauses displaying topic and/or focus. Evidence from the placement of the overt complementizer showed that both positions are within IP, and while the topic position is peripheral to the focus position, the latter is high enough in the clause so that it can have a clause-external governor under certain circumstances. Spec of IP was therefore nominated as a candidate for this operator position, and it was further claimed that the A' non-operator is adjoined to IP. The upshot of this structural argument, of course, is not that Spec of IP qua Spec of IP is the locus of focus, or that the IP adjoined site is where topics must lodge; particular node labels are subject to change, and one could imagine enriching the inventory of functional projections to include Focus Phrases and Topic Phrases, and particular properties could be associated with particular specifiers. What was established was the relative position of the two A'-positions within a hierarchical structure with respect to the complementizer and the upper verb.

Notes *

I would first of all like to express my profound gratitude to Hasan Basri for sharing his knowledge of his language with me, on Long Island and on Selayar Island. My fieldwork in Indonesia was supported by the Stony Brook Research Foundation through a Faculty Development Grant, and it was conducted through the Indonesian Ministry of Sciences (LIPI) under the sponsorship of Tadulako University, Central Sulawesi. Preparation of the manuscript and related research was supported by a Senior Summer Stipend from the National Endowment for the Humanities. Travel to the Scrambling Workshop in Tilburg was made possible by a grants from the Dean of Social and Behavioral Sciences, SUNY Stony Brook, and the United University Professionals. Thanks to Sandra Chung, Carol Georgopoulos, Jane Grimshaw, Luigi Rizzi, Lisa Travis, and an anonymous reviewer for comments on earlier versions and/or related work. Nobody is likely to agree with everything.

1. An Austronesian language of the Makassar group (Grimes and Grimes 1987), from Selayar Island, South Sulawesi Indonesia.

178 Daniel L Finer

2. See, however, Kiss' (1981) discussion of Hungarian, where Topic and Focus are ordered as in Selayarese. 3. The same conclusion, that Spec of IP may act as an A'-position, has been reached by Diesing (1989) for Yiddish and by Goodall (1991a, 1991b) for Spanish. 4. Is = first person singular; 2fam = second person familiar; 2hon = second person honorific; Ipi (inc) = first person plural, inclusive; lpl (ex) = first person plural, exclusive; 3 = third person (singular and plural). In the examples below, i, glossed 'h', is a marker that occurs before [+human] NPs. 5. See e.g., Kuroda (1989), Kitagawa (1986), Sportiche (1988), for the general VPinternal subject hypothesis as well as Guilfoyle — Hung — Travis (1989) for arguments that the VP-intemal subject is a feature of Austronesian syntax. Also, for purposes of the present discussion, I am assuming that Infi is a unitary category, not expanded as, in Pollock (1989). Nor am I assuming an Obj-Agr projection (cf., for example, recent work by Chomsky (1989) and Mahajan (this volume)). 6. I will not consider the VSO order in detail here. This order could be in principle derived in at least two ways, the first via verb-raising out of an SVO-oriented VP (as Guilfoyle, Hung, and Travis (1989) suggest for VSO in Austronesian), or from a VP of the sort shown in (4), via adjunction of the subject to a lower verbal projection (as argued by Chung (1990) for Chamorro). 7. See in particular Koopman and Sportiche (1982), Safir (1984), Sells (1984). 8. (6b) is grammatical under a VSO interpretation, and so this argument is not as strong as it would be if VSO order were impossible. 9. See Mithun and Basri (1985) for a discussion of the different phonological shapes that this affix may take. The geminate [k] in (7d) arises through assimilation of the /?/ of the prefix to the following stem-initial Dd. 10. An anonymous reviewer suggests that the agreement patterns with indefinite objects may instead be a reflex of an antipassive construction. Under such an analysis, the verb would show intransitive marking since the indefinite argument would be oblique under antipassive. There would need to be an extra stipulation in the grammar, however, to prevent the topicalization or focussing of oblique elements in order to prevent the derivation of examples such as (10a). The lack of such examples, on the other hand, follows from the government requirement on incorporation assumed in the text. 11. Perhaps through a set of assumptions involving DP's and case-marking possibilities, but space considerations preclude a full discussion of how Case assignment and realization are carried out in the different constructions discussed here.

Ort the nature of two A '-positions in Selayarese

179

12. Although I will assume that the failure of the affix to drop when a direct object is focussed induces an ECP-related violation, a factor which may militate against wholesale adoption of such a suggestion is that this affix also drops when the subject of an intransitive verb is focussed, as illustrated above. 13. Note that the verbal affix corresponding to the definite direct object is not present in the examples which show a proposed PP controlled by that object. When the subject of a transitive verb controls this PP, the affix occurs. These facts suggest that the presence of the affix on the verb blocks head government of the trace left by PPfronting when the PP is generated within the immediate projection of the V. In the examples where the affix on the verb does occur, the PP is controlled by the subject, and here I assume that the PP is attached higher in the verbal projection and I provides head government for the PP trace. 14. Since the subject occupies focus position in (18c), the object agreement suffix is present, and the presence of the object in the preverbal topic position does not trigger the loss of the suffix. 15. Lasnik and Stowell (1987) and Dobrovie-Sorin (1989) attribute differences in crossover behavior between different A'-elements to differences in the quantificational nature of the A'-elements themselves. It is not clear whether such an approach is compatible with the Selayarese data, since the forms which are variously focussed or topicalized are morphologically identical (in contrast to Dobrovie-Sorin's Romanian examples) and receive their interpretations largely in virtue of the syntactic position they occupy. An intriguing covergence, however, is that the Romanian A'-elements which do not allow clitic-doubling are also implicated in weak crossover violations. 16. I am using the term "reconstruction" in a theory-neutral sense here; as far as I can tell, nothing crucial hangs on whether the anaphor literally moves back into the position of one of its traces at LF. 17. While an argument against an NP-movement analysis of topicalization was never fully mounted, note that any object topicalizations via NP-movement would be immediately suspect, since they would involve movement across a subject, regardless of the status of the landing site. In addition, it is assumed in Finer (1991), for independent reasons, that VP is a barrier, and NP movement would thus be blocked independently. Note that under the analysis where the topic binds pro, (20b, c) and (21a) would consitute violations of both Principles A and Β of the Binding Theory. 18. It is shown in Finer (1991) that this Spec of IP position can be a landing site for whelements in indirect questions and that it is also implicated in cyclic w/i-movement. 19. At this point I have no data which shows a full NP in the focus position in this construction, but other diagnostics are sufficient, I think, to establish the position of the element which corresponds to the agreement affix on the upper verb.

180 Daniel L Finer

20. If the arguments are reversed, such that the subject is focussed and the object extracted by wA-movement, the result is ungrammatical. This "inverse superiority" effect is treated in more detail in Finer (1991). 21. If a more articulated Infi that includes an object agreeement projection is assumed, a promising approach to these data would be to consider movement from the lower Spec of IP up to a higher Spec of O-Agr, thus triggering the occurrence of the agreement affix. Given the A'-properties of Spec of IP, it would follow that Spec of O-Agr would have to be an A'-position, or else constraints on improper movement would be violated. 22. One question that still remains to be addressed of course is why the upper verb's agreement affix can only reference the focussed element, not the IP-adjoined position (cf. (27)). While this question is important to a full understanding of Selayarese grammar, it is not directly relevant to the structural argument I am making here, and so for this reason, as well as space, I will not pursue the issue. Other data that cannot be addressed here (but see Finer 1991) include cases of "long-distance" scrambling, which illustrate two kinds of dependencies. The first bears all of the diagnostics of local movement to focus except that a longer distance is covered (via cyclic movement through Spec of IP and Spec of CP) and overt complementizers may not intervene along the dependency. The second construction shows the properties of topicali/ation (as discussed in the text) along the dependency; there is no weak crossover and the agreement suffix surfaces. Also in this second construction, overt complementizers intervene and Spec of IP can be occupied in intermediate clauses, leading toward the conclusion that long distance movement does not take place in this construction. I suggest that a resumptive pronoun strategy is employed instead. This conclusion nicely matches the conclusion reached for the cases of local topicalization discussed here. 23. Depending on the precise definition of Relativized Minimality with respect to operators vs. non-operators, an antecedent government failure may occur as well in (32) if the A'-status of the topic is sufficient to block antecedent government between the A'-positions in Spec of CP and Spec of IP (even though the topic does not qualify as an operator). If this is so, then perhaps the narrowing of head-government suggested above is unnecessary in this particular case. A similar account could also block the suggested (but abandoned) derivation for (28).

On the nature of two A '-positions in Selayarese

181

References Baker, M. 1985 Incorporation. Chicago University Press. Barss, A. 1986

Chains and Anaphoric Dependence. Doctoral dissertation, Massachusetts Institute of Technology.

Basri, H. — D. Finer 1987 "The Definiteness of Trace", Linguistic Inquiry 18. 1:141-147. Chomsky, N. 1976 "Conditions on Rules of Grammar", Linguistic Analysis 2, 303-351. Chomsky, N. 1986 Barriers. ΜΓΓ Press, Cambridge, Massachusetts. Chomsky, N. 1989 "Some Notes on Economy of Derivation and Representation", MITWPL #10, 43-74, Department of Linguistics and Philosophy, Massachusetts Institute of Technology. Chung, S. 1982 "Unbounded Dependencies in Chamorro Grammar", Linguistic Inquiry 13, 1:39-77. Chung, S. 1990 "VP's and Verb Movement in Chamorro", Natural Language and Unguis tic Theory 8, 4:559-617. Diesing, M. 1990 "Verb Second in Yiddish and the Nature of Subject Position", Natural Language and Linguistic Theory 8, 1:41-79. Dobrovie-Sorin, C. 1990 "Clitic Doubling, Wh-movement, and Quantification in Romanian", Linguistic Inquiry 21, 3:351-397 Finer, D. 1991

C-Agreement, Spec of IP, and Binding Domains in Selayarese. Ms., SUNY at Stony Brook.

Georgopoulos, C. 1985 "Variables in Palauan Syntax", Natural Language and Linguistic Theory 3, 1:59-84.

182 Daniel L Finer

Goodall, G. 1991a Spec of IP and Spec of CP in Spanish wh-Questions, paper presented at the Twenty-First Annual Linguistic Symposium on Romance Languages, UC Santa Barbara. Goodall, G. 1991b "On the Status of Spec of IP", to appear in the proceedings of the Tenth West Coast Conference on Formal Linguistics. Grimes, J. — B. Grimes 1987 Languages of South Sulawesi. Department of Linguistics, Research School of Pacific Studies, Australian National University, Pacific Linguistics, SeriesD. Guilfoyle, E. — H. Hung — L. Travis 1989 Spec of IP and Spec of VP: Two Subjects in Malayo-Polynesian Languages. Ms. McGill and Brandeis. Jaeggli, O. 1982 Topics in Romance Syntax. Dordrecht: Foris. Kitagawa, Y. 1986 Subject in Japanese and English. Unpublished doctoral dissertation, distributed by GLSA Publications, University of Massachusetts/Amherst. Koopman, H. — D. Sportiche 1982 "Variables and the Bijection Principle", The Linguistic Review 2,1:139-160. Kuroda, Y 1988 "Whether We Agree or Not", Linguisticae Investigationes 12, 1:1-47. Lasnik, Η. — T. Stowell 1987 Weakest Crossover. Ms., University of Connecticut and UCLA. Mahajan, A. 1990 "Toward a Unified Theory of Scrambling", this volume. Mithun, Μ — Η. Basri 1986 "The Phonology of Selayarese", Oceanic Linguistics 25, 1-2:210-254. Pollock, J-Y 1989 "Verb Movement, UG, and the Structure of IP', Linguistic Inquiry 20, 3:365-424. Riemsdijk, H. van — E. Williams 1982 "NP-structure", The Linguistic Review, 1:171-217.

On the nature of two A'-positions in Selayarese

183

Rizzi, L. 1990

Relativized Minimality. Cambridge, Massachusetts: ΜΓΓ Press.

Safir, K. 1984

"Multiple Variable Binding", Linguistic Inquiry 15, 4:603-638.

Sells, P. 1984

Syntax and Semantics of Resumptive Pronouns. Doctoral dissertation, University of Massachusetts. Distributed by GLSA publications. University Sportiche, D.of Massachusetts/Amherst. 1988

"A Theory of Floating Quantifiers", Linguistic Inquiry 19, 3:425-449.

Williams, E. 1980 "Predication", Linguistic Inquiry 11, 1:203-238.

Core structures and adjunctions in Warlpiri syntax Kenneth Hale

1. The configurational and non-configurational structures of Warlpiri1 The configurational nature of basic and derived syntactic structures in some free word order languages is quickly detected. In fact, for some languages, among them the Uto-Aztecan language Papago (Hale — Selkirk 1987; Hale, to appear), the configurational nature of syntactic structures is evident at s-structure in virtually every well-formed sentence. In Papago, basic and derived configurational structures are reflected in intonation in an entirely consistent manner and, in addition, they are reflected in the distribution of allomorphic variants of certain determiner elements. The language also shows certain expected effects of extraction processes. The surface form of Warlpiri sentences contrasts with this picture rather sharply. To be sure, in Warlpiri, as in all languages, initial syntactic projections defined by the lexicon exhibit the configurational organization typically reflected in binding processes which are sensitive to c-command asymmetries ~ e.g., anaphora and control. But these aspects of Warlpiri are overtly reflected primarily in word morphology, in the verb, the auxiliary, and the case and agreement systems. The surface representation at which actual words appear in sequence - what I will call the overt phrase structure (OPS) ~ does not directly and consistently reflect, in word order or constituent structure, the configurational organization determined by the projection of argument structure from the lexicon. In the study of Warlpiri overt phrase structure, no truly convincing case has been made for a basic order of constituents, nor has any convincing evidence been forthcoming in favor of a movement analysis to account for the variety of word order arrangements observed.2 The primary focus of the ensuing discussion will be the characteristics of Warlpiri OPS representations. First, however, I will present some elementary data revealing the configurational nature of what I will call the "core argument structure" of Warlpiri clauses ~ that is to say, the structural organization of a predicator and its argument(s) as that is determined in the projection of syntax from the lexicon. The core argument structure is "abstract", so to speak, and its very real

186 Kenneth Hale

existence is discernible primarily through its effects alone. In this respect it contrasts with overt phrase structure. The latter, but not the former, corresponds to an audible string of words. The central issue in much work on so-called non-conflgurational languages is the problem of determining the nature of the relation between these two structures (cf. Hale 1979, 1983; Jelinek 1984; Baker 1990).

1.1. The configurational projection of argument structure In Warlpiri finite clauses, the grammatical arguments of the verb are represented overtly by agreement morphology in the auxiliary. Provisionally, we may consider subject agreement in finite clauses to be regularly pronominal (in the sense of the Binding Theory, cf. Chomsky 1981; but see section 4 below), and it may or may not be associated with a nominal expression in OPS. Object agreement may be pronominal or anaphoric. If pronominal, it may or may not be associated with a nominal in OPS. In any event, a pronominal object is free in its governing category. If the object agreement is anaphoric, it is bound in its governing category, as required by the Binding Theory. And if it is associated directly with an overt nominal in OPS, that nominal is a secondary predicate, not a referential expression. Sentences (1) and (3) exemplify pronominal objects, while (2) and (4) exemplify anaphoric objects. These exemplify precisely the asymmetry expected in the core structural projection of the arguments of a clause. If an object is anaphoric, it is bound by the subject; if it is pronominal, it is free in the domain of the subject. (1)

Ngarrka-jarra-rlu man-DUAL-ERG

ka-pala-jana PRES-3ds-3po

lungkarda-patu bluetongue-PAUCAL

paka-rni strike-NPST 'The (two) men are striking (killing) the bluetongues (skinks).' (2)

Ngarrka-jarra-rlu ka-pala-nyanu paka-rni man-DUAL-ERG PRES-3ds-REFL strike-NPST 'The (two) men are striking themselves/each other.'

Core structures and adjunctions in Walpiri syntax 187

(3)

Karnta-jarra-rlu woman-DUAL-ERG

ka-pala-jana PRES-3ds-3po

miyi yi-nyi food give-NPST

kurdu-patu-ku child-PAUCAL-DAT 'The (two) women are giving the (several) children food.' (4)

Karnta-jarra-rlu ka-pala-nyanu miyi yi-nyi woman-DUAL-ERG PRES-3ds-REFL food give-NPST 'The (two) women are giving each other food.'

If a subject is anaphoric, of course, it is bound from without, i.e, from the matrix, as expected given that it asymmetrically c-commands the other arguments of its own clause. This is the situation in Warlpiri control constructions (cf. Simpson — Bresnan 1983) of the type exemplified by (5): (5)

Kurdu-patu-rlu ka-lu-ngalpa child-PL-ERG PRES-3ps-12po

nya-nyi kuyu see-NPST meat

purra-nja-kurra cook-INF-OBJCOMP "The children see us (plural inclusive) cooking meat.' It is clear from the study of anaphora in Warlpiri that the subject and object are in an asymmetric relation to each other. This would follow, of course, if we assumed (as I will here) that Warlpiri core argument structure contains a constituent, the verb phrase, which includes the verb and its object while excluding the subject. The overt phrase structure of Warlpiri is crucially different from core argument structure in the constituent structure it recognizes, as will become evident in what follows.

1.2. Warlpiri overt phrase structure (OPS) representations It is, of course, not correct to say that Warlpiri OPS representations are "non-configurational", since there is constituent structure, after all. But overt phrase structure simply does not correspond directly and

188 Kenneth Hale

consistently to the universal hierarchical configurations clearly present in the core argument structures defined by predicators ~ in Warlpiri and, by hypothesis, universally. The structural asymmetries of Warlpiri OPS representations are to be observed in the organization of words into syntactic expressions. In certain clear cases, where two or more words belong to a single syntactic expression, this is evident from the surface form of OPS. The syntactic coherence of a Warlpiri linguistic expression is realized in OPS in one of two ways: (i) by solitary functional category inflection (for case, complementizer) at the righthand margin of a string of words forming a syntactic constituent; (ii) by identical functional category inflection of linearly nonadjacent words functioning as a syntactic unit, defining so-called "discontinuous expressions". Only in the former case can we assume that the expression involved is a "constituent" in the familiar sense of a string exhaustively dominated by a common node, a condition further distinguished by the ability of the expression to appear as a unit in pre-AUX position, not otherwise possible for sequences of words. The sentences of (6) through (10) illustrate both continuous and discontinuous expressions in Warlpiri, including determiner (demonstrative) modification expressions, genitive expressions, attribute modification expressions, infinitival clauses, and locatives. (6)

Determiner (demonstrative): Kurdu yalumpu-rlu child that-ERG

ka-jana PRES-3po

maliki-patu dog-PL

jiti-rni tease-NPST 'That child is teasing the dogs.' Kurdu-ngku ka-jana maliki-patu jiti-rni yalumpu-rlu. child-ERG ... Kurdu-ngku ka-jana yalumpu-rlu maliki-patu jiti-rni. Yalumpu-rlu ka-jana maliki-patu jiti-rni kurdu-ngku. *Kurdu ka-jana ... yalumpu-rlu ... . *Yalumpu ka-jana ... kurdu-ngku ... .

Core structures and adjunctions in Walpiri syntax 189

(7)

Genitive: Maliki ngaju-nyangu-ku ka-rna-rla dog me-GEN-DAT PRES-lss-3so kuyu yi-nyi meat give-NPST Ί am giving meat to my dog.' Maliki-ki ka-rna-rla kuyu yi-nyi ngaju-nyangu-ku. dog-DAT ... Maliki-ki ka-rna-rla ngaju-nyangu-ku kuyu yi-nyi. Ngaju-nyangu-ku ka-rna-rla kuyu yi-nyi maliki-ki. *Ngaju-nyangu ka-rna-rla ... maliki-ki... .

(8)

Modifier: Maliki wiri-ngki 0-ji dog big-ERG PERF-lso Ά big dog bit me.' Maliki-rli 0-ji dog-ERG PERF-lso

(9)

yarlku-rnu bite-PST

yarlku-rnu bite-PST

wiri-ngki big-ERG

Infinitival clause: [marna nga-rninja-kurra] ka-rna wawirri [grass eat-INF-OBJCOMP] PRES-lss kangaroo nya-nyi see-NPST 'I see a kangaroo eating grass.'

190 Kenneth Hale

Marna-kurra grass-OBJCOMP

ka-rna wawirri nya-nyi PRES-lss kangaroo see-NPST

nga-rninja-kurra eat-INF-OBJCOMP (10)

Locative expression: Pirli-ngka mountain-LOC

kankarlumparra over

ka ya-ni PRES go-NPST

pintapinta airplane 'The airplane is going over the mountain.' Pintapinta airplane

ka kankarlumparra ya-ni PRES over go-NPST

pirli-ngka mountain-LOC It is noteworthy that the verb phrase, clearly present in the configurational representation of Warlpiri argument structure, is not obviously a constituent in the OPS representation -- consider (11): (11)

Verb and object: *Wawirri nya-nyi kangaroo see-NPST Ί see a kangaroo.' *Nya-nyi see-NPST

ka-rna PRES-lss

wawirri ka-rna kangaroo PRES-lss

An object cannot precede a fronted (pre-AUX) verb in OPS unless it is left-dislocated, a circumstance clearly marked intonationally. Thus, the verb and its overt object do not form a constituent in (12), for example:

Core structures and adjunctions in Walpiri syntax 191

(12)

Left dislocation: Wawirri

nyampu,

pantu-rnu

0-rna

ngajulu-rlu

kangaroo this, spear-PAST PERF-lss I-ERG "This kangaroo, I speared it.' While it is possible to argue that certain strings of words in Warlpiri form constituents of a clause (e.g., certain nominal and infinitival expressions qualify as constituents), the structural relation between such readily identifiable constituents is evidently of a different nature. In the following section, I will consider the question of whether Warlpiri overt phrase structure exhibits asymmetries in command relations among the discernible constituents of a clause, e.g., between the subject and the object.

2. Command relations in overt phrase structure representations A number of phenomena, in a significant number of languages, have been identified by grammarians as being sensitive to command relations, and some of these phenomena are commonly held to be universal. Thus, for example, it is claimed that Warlpiri conforms to the principle that an anaphor is bound by a c-commanding antecedent (cf. (2, 4) and related discussion above). But in Warlpiri this relation obtains in the core argument structure, a level of representation distinct from that which I am temporarily calling overt phrase structure. In the latter, the presence of clear c-command asymmetries among overt phrases linked to the arguments of the same verb, for example, is not obvious. In fact, it is not clear that grammatically relevant asymmetries are present at all. For a variety of reasons, it is somewhat difficult to find relevant tests for command asymmetries among overt constituents (say subject and object). Obvious tests, such as the behavior of anaphoric arguments, are not available in overt phrase structure, since anaphora is realized in Warlpiri by means of bound morphology, and asymmetries are observable only in the core projection of argument structure, a representation whose structure is detectible precisely in the functioning of anaphora and control (as noted in 1.1 above). There are, however, certain dependencies which are observable to some extent in

192 Kenneth Hale

overt phrase structure. Their relevance is not totally clear, but they are suggestive and will be discussed briefly in the following subsections.

2.1. Depictive secondary predication Secondary predication is severely limited by c-command conditions in familiar languages like English (cf. Williams 1980; Rapoport 1990), and correspondingly, the observed positioning of a secondary predicate is restricted in relation to that of its subject. In Warlpiri, by contrast, secondary predication is essentially oblivious to linear order, as illustrated in (13a, c) -- the secondary predicate agrees in case with its subject (i.e., 'dog' and 'tired' agree in (13a), 'kangaroo' and 'tired' in (13b), and so on): (13)a. Maliki-rli ka marlu wajilipi-nyi mata-ngku dog-ERG PRES kangaroo chase-NPST tired-ERG 'The dog, tired, is chasing the kangaroo.' Mata-ngku ka marlu wajilipi-nyi maliki-rli. b. Maliki-rli ka marlu wajilipi-nyi mata dog-ERG PRES kangaroo chase-NPST tired 'The dog is chasing the kangaroo (and the latter is) tired.' Mata ka wajilipi-nyi maliki-rli marlu. c. Karnta-ku 0-rla yu-ngka miyi woman-DAT IMP-3sd give-IMP food wirlinyi-jangka-ku hunting-EL-DAT 'Give the woman some food, as she is back from hunting.' Wirlinyi-jangka-ku 0-rla yu-ngka karnta-ku miyi Although there are stylistic and discourse related preferences among these alternative orderings (and the others that are also possible), there are no strictly grammatical considerations which would preclude any of

Core structures and adjunctions in Walpiri syntax 193

them. On independent grounds, it is clear that command is relevant to secondary predication in Warlpiri, since, for example, a secondary predicate appearing in a matrix clause cannot be construed with a nominal appearing within an infinitival complement: (14)

Kurdu ka-rna mata nya-nyi child PRES-lss tired see-NPST [maliki wajilipi-nja-kurraJ dog chase-INF-OBJCOMP Ί see the child, tired, chasing the dog.'

The only interpretation possible here is that according to which 'tired' is predicated of the nominal 'child', despite the fact that the predicate agrees in (absolutive) case not only with the latter but also with the subject of the infinitival (i.e., with 'dog')3. And the possibilities in secondary predication would not be altered, for example, if the infinitival were reordered in any of the permissible ways relative to the secondary predicate. If secondary predication is sensitive to command relations, then the various orderings in (13) do not differ in relation to the particular command relation which is relevant to secondary predication. If the relevant relation is, say, c-command, so that the secondary predicate must be c-commanded by its subject (i.e., by the nominal of which it is predicated), then there are evidently no relevant c-command asymmetries in the overt phrase structure representations of (13) — the secondary predicate is, observationally, appropriately commanded by its subject in all orderings. I should temper the foregoing remarks with the caveat that a lot of work remains to be done on "secondary predication" in Warlpiri to determine whether such sequences as those represented by (13) are in fact relevant to the present discussion. In particular, a careful study of the intonational phrasing of sentences with nonadjacent nominals "construed together" has yet to be done, particularly in relation to the right-periphery of the clause. Such a study will be essential to a proper understanding of the phrase structures involved and to the question of whether secondary predication is really at issue here, technically speaking. If, for example, some sentences of the general type

194 Kenneth Haie

represented by (13) are not cohesive clauses but, rather, are clauses to which a "tag" (i.e., the final nominal) is appended (as an 'afterthought', 'parenthetical', or 'correction', say), then their relevance to the study of c-command asymmetries is altered considerably. Thus, in any definitive study of these matters, efforts will have to be made to distinguish "parentheticals", and the like, from integral constituents of a clause. And the matter of clausal cohesion is not altogether straightforward in Warlpiri -- except perhaps in what might be called the 'clear cases', of which there are roughly two subcases: (i) pre-AUX position, and (ii) between an AUX and a verb occurring later to the right. Although a parenthetical is possible in the second of these cases, generally the elements appearing in the environments corresponding to (i) or (ii) form integral parts of the clause. By contrast, the status of material following both the verb and the AUX is generally unclear. This caveat must be extended to all claims made here about relations between overt phrase structure and phenomena which may or may not be sensitive to structural asymmetries. I turn now to another phenomenon which, like secondary predication, evidently involves a local dependency.

2.2. Attributive reciprocals Warlpiri nominal expressions formed with the complex suffix -kariyinyanu, informally termed "attributive reciprocals" here, are illustrated in (15) through (20). Attributive nominals are interpreted with reference to a local "antecedent", in roughly the following manner. In a sentence of the form [... NP'... NP-kariyinyanu ...], order irrelevant, where NP and NP' are associated with distinct grammatical functions within the clause (cf. Simpson 1987), neither being restricted to one or the other function, the expression NP kariyinyanu attributes the property denoted by NP to the entity referred to by NP'. The following passage exemplifies this usage:

Core structures and adjunctions in Walpiri syntax 195

(15)

(Wawirri kala mata-jarri-ja), mata-kariyinyanu-lku kala paka-rnu tired-ATRECIP-NOW NARPST strike-PST purlka-ngku old man-ERG '(The kangaroo got tired and), tired like himself, the old man killed it.'

Here, the attributive reciprocal expression mata-kariyinyanu, a secondary predicate of the object (hence in absolutive case), indicates that the predicate mata 'tired' applies not only to the object (represented within the local clause only by agreement, but clearly linked to wawirri 'kangaroo' in the parenthesized clause) but also to the ergative subject purlka-ngku 'old man(-ERG)'. The latter is, so to speak, the "antecedent" of the attributive reciprocal. The antecedent of an attributive reciprocal need not be realized as an NP in the overt phrase structure representation of a sentence; it may, as in (15) for example, be represented solely in the agreement morphology within the auxiliary (and possibly by an associated pro, though this is an issue apart, see below). In general, the antecedent in this relation is "local" — i.e., whether or not it is an overt NP, the antecedent is generally to be found within the same clause as the attributive reciprocal. The effect of this is illustrated in part by the following sentence in which an attributive reciprocal object appears within an infinitival clause. The locality requirement dictates that the reciprocal find its antecedent within its own clause, thus, in this case, it must take the local PRO subject (construed with the main clause object karnta 'woman') as its antecedent and cannot, therefore, take the semantically more appropriate matrix subject. This grammatical circumstance accounts for the semantic ill-formedness of (16):

196 Kenneth Hale

(16)

*Ngarrka-ngku 0 karnta nya-ngu man-ERG PERF woman see-PAST [ngarrka-kariyinyanu paka-minja-kurra ] man-ATRECIP strike-INF-OBJCOMP 'The man saw the woman hit another man (like her).'

This sentence becomes well-formed in all respects, of course, if ngarrka-kariyinuanu (man-ATRECIP) is replaced by karnta-kariyinyanu (woman-ATRECIP). Furthermore, an overt antecedent may not be contained within a constituent which excludes the attributive reciprocal, suggesting that an overt antecedent must c-command the reciprocal: (17)

Kurdu-kurlangu maliki ka-rna child-GEN dog PRES-lss

wajilipi-nyi chase-NPST

mata-kariyinyanu-rlu tired-ATRECIP-ERG Ί , tired like it, am chasing the child's dog.' Here, the antecedent of the reciprocal is the entire possessive expression ('the child's dog'), and not, for example, the genitive ('the child'). Thus, if c-command is really at issue here, then the possessive NP bears that relation to the reciprocal, while the possessor NP does not. To the extent that we can determine this from textual examples, it is evident that the surface ordering of the reciprocal and its antecedent has no effect on well-formedness, suggesting that the overt phrase structure representations do not show asymmetries in the command relation which is relevant to the interpretation of attributive reciprocals. Further examples are given below:

Core structures and adjunctions in Walpiri syntax 197

(18)

Purdapurda-ya-nu liste ning-go-PST

O-rla PERF-3dat

(nyanungu-ju) (he-OI)

purlka-kariyinyanu-ku oldman-ATRECIP-DAT 'He (an old man) went along listening for the other old man.' (19)

Nyanungu-ju-lpa purlka-kariyinyanu-rlu nya-ngu he-OI-IMPERF oldman-ATRECIP-ERG see-PST 'The other old man saw him (an old man).'

(20)

Ngula yika-lu-nyanu ma-ni nyurrpu-kariyinyanu so COMP-3ps-refl take-NPST harmonic-ATRECIP yangka -- karnta-ju ngarrka-ngku-ju thatrEVOC woman-OI man-ERG-OI 'And so men marry women, each of harmonic generation level.'

In each case, the order of the reciprocal NP and its overt "antecedent" NP can be reversed without altering the grammaticality of the sentence.

2.3. Bound kin terms Referentially bound kinship nominale in -nyanu (glossed KINR; cf. the reflexive/reciprocal morphology in AUX, also -nyanu) favor a local antecedent if one is available, as in (21), in which the absolutive subject kurdu nyampu 'this child1 is the antecedent of the (causal) dative argument kirda-nyanu-ku 'self s father', an example of a "bound kin term":4

198 Kenneth Hale

(21)

Kirda-nyanu-ku ka-rla marlaja-nguna father-KINR-DAT PRES-3dat CAUSE-lie(-NPST) — paka-rninja-warnu -- kurdu nyampu strike-INF-RESULT child this "The child is lying prostrate because of his father, from being beaten.'

Although the evidence in this case is not absolutely clear, where a c-command asymmetry is clearly present, the bound nominal prefers a c-commanding antecedent. Thus, in (22), the genitive expression contained within the possessive construction does not readily serve as antecedent for the bound kinship nominal — thus, on the favored reading, the uncle is the child's, not Jakamarra's: (22)

Ngamirni-nyanu-ku ka-rla marlaja-yula MoBro-KINR-DAT PRES-3dat CAUSE-cry(-NPST) Jakamarra-kurlangu kurdu J.-GEN child 'J.'s child is crying because of his/her uncle.'

But here again, the ordering of the relevant words is evidently immaterial, and in fact "cross-binding" is common, in which the subject binds into an object, and vise versa, as in (23) through (25). Thus, linear order of the relevant expressions does not correspond to asymmetries in any command relations relevant to the interpretation of bound kin terms: (23)

Kirda-nyanu-rlu kaji-lpa ka-ngkarla-rni father-KINR-ERG IRR-IMPERF carry-IRR-HITHER ngalapi-nyanu win, ngarrka-yijala offspring-KINR big, male-ALSO 'Were [his father] to bring [his grown son].'

Core structures and adjunctions in Walpiri syntax 199

(24)

Kurdu-nyanu-rlu ka-rna-lu-jana ngarri-rni -child-KINR-ERG PRES-3ps-3po call-NPST purtarirlangu-ju -- ngamirni-nyanu-patu "p." MoBro-KINR-PL '[Their children] (we) call [our uncles] "purtarirlangu".'

(25)

Kurriji-nyanu-kujaku ka malirdi-nyanu-rlu WiMo-KINR-EVIT PRES DaHu-KINR-ERG kurnpariji ngarri-rni fire call-PNST '[Her son-in-law] calls it (i.e., fire) "kurnpariji" in deference to [his mother-in-law].'

Assuming that c-command is relevant to the local binding of these nominals, so that an antecedent must c-command the kin term it binds, then in the overt phrase structure representations of these sentences, there can be no constituent which contains one kin term and excludes the other ~ in the relevant sense.

2.4. Oblique datives Dative arguments in Warlpiri divide into two classes, direct and oblique.5 A pronominal appearing in an oblique dative may be bound by the subject of the same clause, as in (26) and (28) - though the reverse is impossible, of course, by Condition Β of the Binding Theory, as shown by (27) and (29) (from Laughren 1989). Evidently, therefore, the pronoun in an oblique dative expression does not command the subject, in the relevant sense, though the subject evidently does command the pronominal of the oblique, since reversal of the pronominal and the name yields ungrammatically (under coreference):

200 Kenneth Hale

(26)

Jakamarra-rlu ka-nyanu-rla J.-ERG PRES-REFL-3DAT

warri-rni seek-NPST

(kuyu-ku) nyanungu-ku (meat-DAT) him-DAT 'J. is looking for meat for himself.' (27)

*Nyanungu-rlu ka-nyanu-rla warri-rni (kuyu-ku) he-ERG PRES-REFL-3DAT seek-NPST ( kuyu-ku ) Jakamarra-ku (meat-DAT) J.-DAT

(28)

Nyanungu-ku ka-nyanu Jakamarra wangka-mi him-DAT PRES-REFL J. speak-NPST 'Jakamarra is talking to himself.'

(29)

*Jakamarra-ku ka-nyanu nyanungu wangka-mi J.-DAT PRES-REFL he speak-NPST

As these examples show, linear order is evidently immaterial, since the binding relations remain the same for both orders.

2.5. Control and negative adverbial clauses Although Warlpiri cannot have parasitic gaps, strictly speaking, negative adverbial clauses do show parallel "gapping" of the subject and the object in those cases in which the subject and object of the negated infinitival clause are pairwise identical to the subject and object of the matrix. As in control constructions elsewhere in Warlpiri, the linear positioning of the infinitival clause in overt phrase structure is immaterial to the control relation, as illustrated in (30). Sentence (31) illustrates a negated infinitival whose object is not shared in the matrix:

Core structures and adjunctions in Walpiri syntax 201

(30)

J.-rlu J.-ERG

ka wawirri yampi-mi PRES roo left-NPST

[e e luwa-rninja-wangu-rlu] [e e shoot-INF-NEG-ERG ] 'J. is leaving the kangaroo without shooting it.' [e e luwa-rninja-wangu-rlu] ka wawirri yampi-mi J.-rlu. Wawirri ka [e e luwa-rninja-wangu-rlu] yampi-mi J.-rlu. [e e luwa-rninja-wangu-rlu] ka J.-rlu yampi-mi wawirri. (31)

Jakamarra ka nyina-mi J. PRES sit-NPST [e wawirri luwa-rninja-wangu] [ roo shoot-INF-NEG] 'J. is sitting without shooting the/a kangaroo.'

2.6. Summary remarks on command relations The above examples have been cited to illustrate the apparent fact that the surface linear arrangement of constituents in Warlpiri overt phrase structure representations does not correspond to asymmetries in interpretations involving referential dependency. Known structural asymmetries apparently do have the expected effect for the phenomena discussed ~ thus, it matters if one member of a pair of items involved in a dependency relation (binding, predication, etc.) is included in a constituent (e.g., NP, infinitival subordinate clause, etc.) which excludes the other. But the data considered in this section do not seem to reveal any constituent in Warlpiri overt phrase structure representations which, say, includes the object but excludes the subject -- i.e., no verb phrase. This accords with the testimony of auxiliary placement (as exemplified in (11)), of course. Thus, the asymmetry inherent in the subject relation, according to which the subject is external to the subconstituent VP containing the object, clearly relevant in the basic Warlpiri projection of core argument structure (cf., 1.1 above), is not visible in

202 Kenneth Hale

the overt phrase structure representations of Warlpiri clauses. I will turn now to the question of whether there are overt phrase structure asymmetries which can arise through movement.

3. Overt phrase structure and movement of phrasal constituents If the word order variations observed in Warlpiri clauses are due to "scrambling" in the technical sense, i.e., if they are due to movement in syntax (as opposed, say, to PF), then scrambling in overt phrase structure has no "visible" effect in relation to the phenomena discussed in section 2. The possibility exists, of course, that scrambling is not what is responsible for surface word order variations in Warlpiri clauses. In this section I will briefly consider the question of whether movement is involved at all in the derivations of Warlpiri overt phrase structure representations. I begin with a discussion of content questions.

3.1. Content questions in Warlpiri

On the face of it, it would appear that movement must be involved in content question formation, since the question word, with rare exceptions, appears in initial position ~ as illustrated in all of (32) through (38). But apart from the possible local movement into preAUX position, there is little to suggest that the initial position of the question word is actually due to movement (but see section 4 below). Firstly, initial position is possible for any constituent, and we have as yet no clear evidence of movement as the responsible agent of this. Second, as (34) shows, a theoretically possible candidate for the weak crossover effect shows no such effect Moreover, candidates for long-distance movement fail to be credible cases of the phenomenon. Apparent extraction from an infinitival clause typically involves questioning part of a so-called "discontinuous expression", as in (35), where an element marked with the objective complementizer is questioned and construed with an infinitival verb, also marked with the objective complementizer (indicating matrix object controller). Although such discontinuous expressions might be formed by movement, it is not obvious that they are. In any event, they

Core structures and adjunctions in Walpiri syntax 203

are not exclusive to questions ~ they are freely alternative to the corresponding contiguous expressions. That movement is probably not involved here is suggested by constructions like (36), in which the questioned constituent, marked with the objective complementizer, is base generated in the matrix clause and construed, in an appropriate way, with an adjunct finite clause. Again, no evidence for movement. (32)

Ngana-ngku ka karli nyampu who-ERG PRES boomerang this 'Who is trimming this boomerang?'

(33)

Nyiya-ku ka-npala-rla what-DAT PRES-2ds-3sdat

jarnti-rni? trim-NPST

warri-rni seek-NPST

nyumpala-rlu? you: DUAL-ERG 'What are you two looking for?' (34)

Ngana ka nyanungu-nyangu who PRES he-GEN

maliki-rli dog-ERG

wajilipi-nyi? chase-NPST 'Who is his dog chasing?' (35)

Nyiya-kurra what-OBJCOMP

0-npa nya-ngu Jakamarra PERF-2ss see-PST J.

jarnti-rninja-kurra ? trim-INF-OBJCOMP 'What did you see J. trim?' (36)

Nyiya-kurra what-OBJCOMP

0-npa nya-ngu J. PERF-2ss see-PST J.

[kuja-lpa jarntu-rnu]? [COMP-IMPERF trim-PST]

204 Kenneth Hale

Extraction from a finite dependent clause is not possible in Warlpiri, as such clauses are adjuncts. To question or relativize a constituent in an adjunct dependent clause, an in situ strategy must be used. For questions especially, this device is rare in actual usage. When it is used, the requirement that the question word be initial is satisfied by using a "proxy interrogative" (similar in nature to the "pleonastic" interrogative of Hindi; cf. Srivastav 1991) in the main clause, as in (37) and (38): (37)Q: Nyarrpa J. how J.

wangka-ja say-PST

[pirrarni-rli [yesterday-ERG

kuja nyiya luwa-rnu]? COMP what shoot-PST] 'What did Japanangka say he shot yesterday?' A: Ngayi luwa-rnu marlu pirrarni-rli 'He just shot a kangaroo yesterday.' (38)Q: Nyarrpa-rlu 0-ngku ngarru-rnu-rra Japangardi-rli how-ERG AUX-2so tell-PST-CNTF Japangardi-ERG [nyarrpara-kurra kuja ya-nu Japanangka] [where-ALLATIVE COMP go-PST Japanangka] 'Where did Japangardi tell you that Japanangka went?' A: Japangardi-rli 0-ji ngarru-rnu 'Japanangka ya-nu tawunukurra'-waja. 'J. told me 'Japanangka went to town' actually.'

3.2. Relative clauses In relativization, Warlpiri uses the so-called 'internal head', as illustrated in (39) through (42). Though LF movement is undoubtedly involved in the interpretation of these structures, it is not evident that movement is involved in defining the observed overt phrase structure representations associated with them:

Core structures and adjunctions in Walpiri syntax 205

(39)

Karli-ngki kuja-npa yankirri luwa-rnu, boomerang-INST COMP-2ss emu shoot-PST ngulaju rdilyki-ya-nu that broken-go-PST "The boomerang you hit the emu with broke.'

(40)

Yankirri kuja-npa karli-ngki emu COMP-2ss boomerang-INST

luwa-rnu, shoot-PST,

ngulaju pali-ja that die-PST 'The emu you hit with the boomerang died.' (41)

Kurdu yali child that

kuja-ka nyanungu-nyangu maliki-rli COMP-PRES he-GEN dog-ERG

wajilipi-nyi, ngulaju ka yula-mi chase-NPST, that PRES cry-NPST "That child that his dog is chasing is crying.' (42)

Nyanungu-nyangu kurdu-jarra-rlu he-GEN child-DUAL-ERG

kuja-pala-rla COMP-3dus-3odat

miyi yu-ngu ngarrka yangka-ku, food give-PST man that-DAT,

ngula-ngku-ju that-ERG-OI

ka-palangu PRES-3duo

karli-jarra boomerang-DUAL

yi-nyi-lki give-NPST-THEN

kurdu-jarra-ku child-DUAL-DAT 'That man who his two kids gave food to is giving boomerangs to the two kids.' Question formation and relativization, alike, fail to give evidence in favor of the position that movement is responsible for the observable form of overt phrase structure representations in Warlpiri. The same must be said of scrambling, as exemplified in examples already given

206 Kenneth Hede

and in the examples assembled from (43) through (76). A consideration of these sentences, in relation to the Binding Theory and associated effects, provides little in support of a movement theory of Warlpiri free word order.

3.3. Free word order and coreference effects It is possible to make use of well-known diagnostics of movement and constituent structure. A contiguous string of words in overt phrase structure, making up, say, a noun phrase or an infinitival clause, reveals its status as a constituent through its behaviour with respect to Binding Theory conditions, among other things. Thus, for example, in sentences (43) and (44) the R-expression (kurdu wita, Jakamarra) may be coreferential with the genitive pronoun in the possessive NP construction. This is possible by Condition Β of the binding theory. And the fact the sentences are grammatical with coreference indicates that the R-expression is free, as required by Condition C, and hence that it is not c-commanded by the pronoun. None of this is affected by the linear order of subject and object. And, in particular, no ordering, including that of (44), shows the Weak Cross-Over effect. (43)

Nyanungu-nyangu maliki-rli ka kurdu wita (s)he-GEN dog-ERG PRES child small wajilipi-nyi chase-NPST *His dog is chasing the little child.'

(44)

Jakamarra ka nyanungu-nyangu maliki-rli J. PRES he-GEN dog-ERG wajilipi-nyi chase-NPST 'His dog is chasing Jakamarra.'

Surprisingly, (45) and (46) are not possible with coreference, at least not in the dialect of the speakers who have given these sentences their most careful consideration.6 If this is to be explained within the

Core structures and adjunctions in Walpiri syntax 207

Binding Theory, then presumably the reason is that coreference violates Condition C. And this would be expected, of course, if the overt phrase structure representations of Warlpiri lacked the relevant subject-object asymmetry, and if Condition C is in fact operative in the overt phrase structure representations of clauses: (45)

J.-kurlangu maliki-rli ka nyanungu wajilipi-nyi J.-GEN dog-ERG PRES him/her chase-NPST 'J.'s dog is chasing him.'

(46)

Nyanungu ka wajilipi-nyi J.-kurlangu maliki-rli him/her PRES chase-NPST J.-GEN dog-ERG 'J.'s dog is chasing him.'

An alternative conception of (45) and (46) seeks to attribute the lack of coreference there to a property inherent in the Warlpiri pronoun nyanungu. When construed as a direct argument (subject or object) in a root clause, this overt pronoun is marked by comparison with the more neutral non-overt alternative. It is used to focus an entity previously mentioned in discourse. If the overt pronoun in such cases represents a discourse topic, then the binding violation would be at LF, not necessarily at s-structure. Sentence (47) is also not possible with coreference. If this is due to Condition C, then the forward position of the possessive NP does not mitigate this — coreference between the R-expression possessor and the pronoun is still impossible: (47)

J.-kurlangu maliki ka nyanungu-rlu wajilipi-nyi J.-GEN dog PRES he-ERG chase-NPST 'He is chasing J.'s dog.'

Sentences (44) and (47) together show that Warlpiri "scrambling", if it exists, cannot simply be A-movement or A'-movement, as generally understood. If the forward (pre-subject) positioning of an object is by Α-movement (cf. Mahajan 1990), then coreference should be possible in both (44) and (47). Coreference is not possible in the latter, suggesting rather that A'-movement (with "reconstruction") is involved. But if so, then (44) should show the Weak Cross-Over effect, which it does not. In short, the linear ordering is simply irrelevant to

208 Kenneth Hale

the actual coreference relations and, in particular, nothing suggests that movement is involved in deriving the surface ordering in Warlpiri overt phrase structure representations (except, perhaps, for a strictly local movement involved in defining the placement of the auxiliary; see below). Given these observations, the possible coreference relations indicated for (48) and (49) follow from accepted assumptions and from the structural relations involved. In particular, the data follow from the fact that an asymmetrical c-command relation holds between NP and NP' if one of these is contained within a constituent which excludes the other. For example, the possessive constructions of (48-3) include the possessor but exclude the ergative subject. (48)

Nyanungu-nyangu maliki ka wajilipi-nyi he-GEN dog PRES chase-NPST J.-rlu J.-ERG 'J. is chasing his dog.'

(49)

Nyanungu-rlu ka J.-kurlangu maliki wajilipi-nyi he-ERG PRES J.-GEN dog chase-NPST 'He is chasing J.'s dog.'

And in (50-51) the infinitival clause includes the dative object of the infinitive, but it excludes constituents of the matrix, including the ergative subject there. The coreference possibilities follow from standard assumptions, of course: (50)

Marlu-ngku ka J. roo-ERG PRES J.

nya-nyi nyanungu-ku see-NPST it-DAT

wurruka-nja-kurra stalk-INF-OBJCOMP 'The kangaroo sees J. sneaking up on it.' (51)

Nyanungu-rlu ka J. nya-nyi marlu-ku wurruka-nja-kurra. 'It sees J. sneaking up on the kangaroo.'

Core structures and adjunctions in Walpiri syntax 209

3.4.

Summary

The material examined in this and the previous section supports, I believe, the intuition that free word order in Warlpiri is of a different nature from that which has been known, since Ross (1967), as scrambling. Scrambling, properly speaking, involves movement. And some languages which exhibit freedom of word order as great as that of Warlpiri can be shown to achieve this through scrambling in the technical sense - e.g., the Uto-Aztecan language Papago. If so, then Papago is to be classed with German (Webelhuth 1989; but see Bayer — Kornfilt 1990, for an alternative view of scrambling itself) and Hindi (Mahajan 1990), for example. By contrast, Warlpiri is to be classed with the polysynthetic Mohawk (Baker 1990, 1991), and the languages described in Mithun (1987; and see also Farmer 1980, 1984 for an important modular theory of the grammar of free word order). The basic finding here, if it can be assumed to be real, is that variations in overt phrase structure word order - e.g., between subject and object, particularly - do not reveal asymmetries which might be attributed either to (1) a VP or like constituent including one argument and excluding another, or (2) movement to a position asymmetrically commanding both the point of origin and other overt constuents in the clause. In the following section, I will consider a conception of Warlpiri (and other so-called non-configurational) overt phrase structure which has, deservedly, gained acceptance in recent years and which seems to me to be consistent with the observations made here. This is the proposal of Jelinek (1984), as modified recently by Baker (1990).

4. Arguments and adjuncts The nominal expressions appearing in overt phrase structure, according to Jelinek, are not arguments. Rather they are adjuncts, linked to argument positions in the core syntactic projection of a clause. Following Baker, I will assume that the core syntactic projection conforms to the usual configurational type, thereby accounting for the configurational properties discussed in 1.1 above. The core syntactic arguments of a tensed clause are non-overt pronominal or anaphoric elements construed, respectively, with pronominal or anaphoric

210 Kenneth Hale

agreement in the auxiliary. The morphological base of the Warlpiri auxiliary, presumably, is to be identified with the functional category I(nfl). The linking of adjuncts to argument positions is by coindexation ~ this is the manner in which adjunct NPs are licensed (Baker 1990; and cf. Hale 1983, for the relation between overt case and core grammatical function in the linking of adjunct NPs in Warlpiri). Overt nominal expressions, then, are related to the core syntactic projection in a manner similar to the way dislocated NPs in more familiar languages are related to the clauses with which they are associated, i.e., by coindexation with resumptive pronouns. In Warlpiri, however, all overt NPs linked to argument positions (in tensed clauses, at least) are adjuncts, by hypothesis, and all resumptive pronominals are non-overt (small pro, presumably), though construed with agreement (overt, except in the third person singular, which happens to be zero). As an aside, and despite my enthusiasm for this conception of Warlpiri overt phrase structure, I must voice one lingering reservation. Overt NPs in preverbal position do not have the "feel" of dislocated phrases ~ unlike true left dislocated "topics" (as in (12) above, for example), and unlike clear cases of trailing corrective or explanatory tags. I am not sure at this point what to make of this. Perhaps "true dislocation" is adjunction to CP, hence "outside the clause", as suggested by the auxiliary placement in (12). By contrast, perhaps, adjunction of the type under consideration here is adjunction to EP (i.e., S in the abbreviatory notation of Baker (1990)), and accordingly adjuncts of this type are not "excluded" from the clause structurally or intonationally. I will assume for present purposes that IP-adjunction is correct, though I cannot really defend that view at this point. In any event, this is a general problem in the study of non-configurational languages.

4.1.

The obligatory adjunction of overt nominal expressions in Warlpiri tensed clauses7

If the essential characteristics of Warlpiri core and adjunct syntax are as outlined here, then why is this so? What is the fundamental property from which everything follows? Here again, I think that Jelinek (1984) has the correct intuition. Her

Core structures and adjunctions in Walpiri syntax

211

proposal is that the real arguments in a Warlpiri tensed clause are represented by the agreement morphology in the auxiliary. Suppose we interpret this to mean that Warlpiri agreement is "rich", which is to say that it is "pronominal" (or anaphoric). If so, the corresponding argument positions are fully determined in syntax. And as is common in such cases (e.g., Irish, cf. McCloskey — Hale 1984), the argument positions themselves are represented by non-overt elements, pronominal (or anaphoric, as the case may be) by virtue of their identification with pronominal (or anaphoric) agreement morphology. Since the arguments of a (tensed) clause are fully determined, by virtue of the agreement morphology in the manner suggested, no overt NP may appear in a core argument position. Overt NP expressions can therefore only appear as adjuncts, not to the argument positions, of course, for theta-theoretic reasons (cf. Chomsky 1986a), but to some higher position. The fact that no structural asymmetries can be detected suggests that all overt NPs linked to argument functions are (or at least can be) adjoined to the same maximal projection. This must be at least as high as IP. Furthermore, phrases adjoined to the same maximal projection mutually c-command each other, evidently, insofar as the phenomena examined in sections 4 and 5 are concerned. And this follows, presumably, from the fact that all adjunction nodes are segments of one and the same maximal category (cf. Chomsky 1986b) - there is no "complete node" (as opposed to "segment") which differentially counts as the "first branching node" dominating any two co-adjunct NPs.

4.2. On the relation among co-adjuncts In previous sections, I have implied that mutual (c-)command is responsible for certain coreference judgments, under the assumption, for example, that the Binding Theory operates in Warlpiri overt phrase structure in the now well understood manner. But this is as much in the nature of speculation as an opposite assumption would be. I would like briefly to consider and alternative, purely speculative at this point. Imagine that command relations are irrelevant in adjunction structures and, therefore, in Warlpiri overt phrase structure generally. And imagine that this is because the relevant grammatical principles ~ e.g., in particular, those inherent in the Binding Theory - simply do

212 Kenneth Hale

not apply among adjuncts dominated by segments of the same maximal projection. We have seen that subject-object asymmetries are evidently absent in Warlpiri overt phrase structure, and one way in which to understand that is to suppose that adjunct NPs mutually c-command each other, a coherent position if c-command is relevant. The view that it is relevant is supported, to some extent, by the observation that asymmetries do apparently exist, and matter, in overt phrase structure in certain clear cases in which a phrase includes one NP while excluding another — e.g., the possessive construction, or infinitival dependent clauses. And in some such cases, Condition C of the Binding Theory appears to be implicated in accounting for observed coreference possibilities, or impossibilities ~ e.g., in (45) through (47) above). But if the Binding Theory is inoperative among co-adjunct NPs, then Condition C cannot really be what is at work here. In fact, the putative Condition C effect may be apparent only, since an alternative explanation exists which makes reference not to the Binding Theory, as that is normally understood, but to the discourse function of argument-linked overt pronouns. Consider again sentence (45), repeated here as (52): (52)

J.-kurlangu maliki-rli ka nyanungu J.-GEN dog-ERG PRES him/her wajilipi-nyi chase-NPST 'J.'s dog is chasing him.'

If the pronoun nyanungu corresponds to a discourse topic, then the failure of coreference is explained without making direct appeal to the c-command asymmetry which is undoubtedly present. If coreference here is a binding violation of some sort, then it is so because the Rexpression Jakamarra (abbreviated J. in (52)), a name, is bound by a discourse operator, and not because the R-expression is apparently ccommanded by its putative overt "antecedent", an co-adjunct NP. This is not a mere quibble, as it could have empirical correlates. If this alternative conception of coreference in adjunction structures is correct, then a reported difference between Mohawk (Baker 1990) and Warlpiri becomes understandable. In Mohawk, a sentence having much

Core structures and adjunctions in Walpiri syntax

213

the same structure as (52) permits coreference between the possessor NP, a name, and a pronominal argument. Unfortunately, however, comparison here is not perfect, since the Mohawk sentence which Baker cites in relation to this issue has a nonovert pronominal, not an overt one. In Warlpiri, even if the object in (52) were non-overt, coreference would be difficult, probably impossible, suggesting that in cases of this type even a non-overt pronominal corresponds to a discourse topic (cf. Huang 1984). More research is needed, clearly. In any event, in Mohawk, the reasoning would be, pronominals (non-overt ones at least) are not discourse topics, and coreference is possible because the Binding Theory is not directly relevant to adjuncts. In the Mohawk analogue of (52), both the pronoun and the Rexpression, being essentially invisible to one another, behave as if free. The same would be true in Warlpiri, presumably, though the effect is obscured by the special discourse-related property of pronominal adjuncts.

4.3. Adjuncts and the binding theory

Although the binding principles may or may not be operative within the domain of the adjunction structure, this does not mean that the appearance of an argument-linked adjunct NP is entirely without effect in determining the coreference possibilities in a Warlpiri clause. Consider, for example, sentences (26) through (29) above. These show that an overt pronoun or name, linked to a position in the core argument structure, has the effect of "fixing" or "setting" the NP category of that core argument as pronominal or R-expression, respectively. Then, of course, the Binding Theory operates in the normal fashion, within the fully conflgurational core syntactic projection, accounting for the judgments which Laughren reports for sentences of this type (Laughren 1989). Thus, as the judgments indicate, an R-expression subject can bind a pronominal oblique object, but an R-expression oblique object cannot be bound at all.8 I suspect that this category-setting relation, which evidently holds between overt argument-linked NP expressions and the core argument positions with which they are coindexed, is extremely important in the interpretation of Warlpiri sentences. And it is probably this relation

214 Kenneth Haie

which makes it possible to form conventional content questions of the type represented by (32)-(36) above. If adjuncts did not bear the suggested relation to their corresponding core argument positions, then it is difficult to imagine how questions could be formed, since the required operator-variable relation could not arise. But if a content question word "sets" the NP category of a coindexed argument as that of variable, then the appropriate structure is present. I think it is unlikely that this can be a simple matter, however, given the well-known fact that question words cannot, in general, be adjuncts. But Warlpiri might be instructive here. This could be the significance of the general fact that content question words appear in initial position. In particular, they precede the auxiliary, suggesting, perhaps that they actually move to that position. Suppose pre-AUX position is higher than the IP projection to which adjunct NPs are attached (cf Branson 1988, for detailed treatment of AUX and the pre-AUX position). For example, it is possible that pre-AUX constituents are adjoined to CP or in the specifier thereof, the latter possibility being the most likely, given the fact that pre-AUX position is a single position. If this is right, then a fronted question word would bind a variable in the adjunct structure (a possibility, surely). The adjunct variable would be linked to its corresponding argument, fixing its category in the appropriate way, and accounting, among other things, for the impossibility of coreference in variants of (51) and (53) in which ngana-ngku (whoERG) and ngana-ku (who-DAT) replace, respectively, nyanungu-rlu (he-ERG) and Jakamarra-ku (Jakamarra-DAT)). Contrary to what was implied in section 3.1 above, it is suggested here that question formation actually does involve movement in Warlpiri. However, the phrase which undergoes movement in the syntax of content questions is an adjunct, not an argument. It is presumably this fact that accounts for why the putative movement must always be "short", a circumstance which interacts with the general freedom of word order to give the superficial appearance of "no movement at all", i.e., the conclusion reached in 3.1.9 The suggestions just made raise a number of issues which I will not be able to go into here. In particular, however, it suggests that the appearance of an adjunct constituent in pre-AUX position is by virtue of movement. If this is to be classed as Α-bar movement ~ which it surely must be, under accepted assumptions10 - then the adjunct variable left at the foot of the corresponding chain might be expected

Core structures and adjunctions in Walpiri syntax 215

to have some discernable effects, opening up a range of as yet untested possibilities, including the possibility that Warlpiri overt phrase structure actually has configurational properties. Although I doubt it, this might conceivably be the case, and the effects might show up, for example, if variable traces can be shown to occupy different positions correlating with different grammatical functions (subject, object). The fact that it does not seem possible, as yet, at least, to discover such differences is consistent with the conception of Warlpiri overt phrase structure adopted here, following the ideas of Jelinek (1984) and Baker (1990). But the investigation must continue, I feel, since the Warlpiri "evidence" for this position is, in large part, negative. That is to say, evidence is so far lacking for configurational asymmetries in overt phrase structure, but little positive evidence is forthcoming for nonconfigurational structure, except in so far as the negative evidence itself can be interpreted as direct evidence for it.

5. Scrambling and adjunction The term "scrambling" is now associated with a particular theory of free word order, namely, that according to which departures from a "basic" order are effected by means of syntactic movement. And, accordingly, "scrambling" is generally identified with movement and, in fact, means movement. Recent work on a variety of languages which have free word order has given strong support to the scrambling theory, by showing that deviations from a basic order produce effects identical to those associated with established grammatical processes such as NPmovement and Wh-movement, processes whose movement status is unquestioned, given accepted theoretical assumptions. However, Warlpiri cannot be classed unequivocally as a scrambling language. If this is the correct conclusion to draw from the data, then not all free word order languages are scrambling languages. Or, to be more exact, not all free word order can be due to scrambling. It seems to me reasonable that there might exist languages of the type to which Warlpiri is assigned here -- i.e., languages whose free word order results simply from the fact that (certain or all) overt phrasal expressions are adjuncts (resumed, of course, in core argument positions by non-overt pro elements). The existence of such languages

216 Kenneth Hale

is an empirical question, plainly, since no theoretical barrier to it exists. Dislocation structures are commonplace among the languages of the world. Warlpiri (with Mohawk, and a number of others) differs from the commonplace only in that all overt NP expressions (in tensed clauses, at least) are adjuncts - necessarily so, since the core argument positions are fully determined by rich pronominal or anaphoric agreement. The Jelinek-Baker conception of languages of this "nonconfigurationar* type expresses at once their special character and the manner in which they realize universal principles of grammar.

Notes 1. I wish to acknowledge my intellectual debt to a number of people in the Warlpiri community who have been of help to me in my study of the language, either through their writing, through personal contact, or both. These include Mary Napaljarri Laughren, Sam Japangardi Johnson, Dinny Japaljarri Anderson, Darby Jampijinpa, Robin Japanangka Granites, and the late Mick Jupurrula Connell. And I also wish to acknowledge my debt to a number of people whose work on Warlpiri, or other free word order systems, has helped me form some of the ideas set out here. These include Eloise Jelinek, Mark Baker, Ann Farmer, Maria Bittner, Viviane Deprez, Jaklin Kornfilt, Josef Bayer, Henk van Riemsdijk, Gert Webelhuth, Anoop Mahajan, Veneeta Srivastav, and Mamoru Saito. 2. I delay temporarily a discussion of the placement of the auxiliary, a local movement. See section 4 below. 3. Mary Laughren has argued that the bare object in infinitival dependent clauses is not in the absolutive case. Rather, it is unmarked for case, being licensed by adjacency to the verb. It is possible that such objects simply cannot function as subjects of secondary predicates ~ and in fact, it is difficult to get secondary predication of the object of an infinitival. This might account for (14), in which case the command asymmetry is irrelevant there. .4. Mary Laughren points out that here, and in the case of Attribute Reciprocals as well, it is important to bear in mind that while local antecedents are preferred, strongly so in neutral contexts, non-local antecedents are also possible in some contexts. 5. The term "oblique" here corresponds to what is assumed to be a structural relation, yet to be determined, setting the oblique dative apart from the direct dative, there is no intended implication that oblique datives are necessarily less "argumental" in relation to the verb. Adjunct datives, as in (26)-(27) are clearly adjuncts, and accordingly less argument-like, as that term is normally understood. But the dative of (28)-(29) is presumably an argument. See Laughren (1989) for a detailed

Core structures and adjunctions in Walpiri syntax 217

discussion of these datives within a developed theory of Warlpiri case. 6. These are speakers consulted by Maty Laughren in context of the Bilingual Education Program at Yuendumu, Northern Territory. Cf. also Laughren 1989. 7. There is some evidence that an overt object NP in infinitivals is not an adjunct, a matter still under investigation (cf. Laughren 1987). 8. The binding of an oblique pronominal by a subject, as in (26) and (28), is only an apparent violation of Condition B, presumably. Various accounts of this have been offered, e.g., Laughren (1989) and Hale (1981). 9. In this respect, Warlpiri evidently differs from Mohawk (Baker 1990), where question words are not adjuncts but core arguments and, accordingly, can undergo "long" movement, including extraction from dependent clauses, and so forth. 10. A conclusion strengthened somewhat by the observation that coreference in sentences like (45) is not possible; an effect which would follow from reconstruction.

References Baker, M. 1990 Some Subject Object Non-asymmetries in Mohawk and their Theoretical Significance. McGill University manuscript. Baker, M. 1991 On the Absence of Certain Quantifiers in Mohawk. McGill University manuscript. Bayer, J. — J. Kornfilt 1990 "Against scrambling as Move-Alpha", Proceedings of NELS 1990, Amherst: GLSA. Chomsky, N. 1981

Lectures on Government and Binding. Dordrecht: Foris.

Chomsky, N. 1986a

Knowledge of Language. New York: Praeger.

Chomsky, N. 1986b

Barriers. Cambridge, Massachusetts: ΜΓΓ Press.

Farmer, A.K. 1980

On the Interaction of Morphology and Syntax. Doctoral dissertation, Massachesetts Institute of Technology, Cambridge, Massachusetts.

218 Kenneth Hale

Farmer, A.K. 1984 Modularity in Syntax. Cambridge, Massachesetts: ΜΓΓ Press. Hale, K. 1981

Hale, K. 1983

On the Position of Warlpiri in a Typology of the Base. Indiana University Linguistics Club.

"Warlpiri and the Grammar of Non-configurational Languages", Natural Language and Linguistic Theory 1: 3-45.

Hale, K. — S.J. Keyser 1989 On the Syntactic Character of Thematic Structure. Ms., Massachusetts Institute of Technology, Cambridge, Massachusetts. Hale, K. to appear

Basic Word Order in Two 'Free Word Order' Languages. Cambridge, Massachusetts: ΜΓΓ manuscript.

Hale, K. — L. Selkirk 1987 "Government and Tonal Phrasing in Papago", in: C. Ewen — J. Anderson (eds.), Phonology Yearbook. Cambridge University Press, 151-183. Huang, C. — T. James 1984 "On the Distribution and Reference of Empty Pronouns", Linguistic Inquiry 15: 531-574. Jelinek, E. 1984 "Empty categories, case, and configurationality", Natural Language and Linguistic Theory, 2: 39-74. Laughren, M. 1987 "The configurationality parameter and Warlpiri, Northern Territory of Australia Department of Education", (ms), also in: L. Maracz — P. Muysken (eds.), Configurationality. Dordrecht: Foris. Laughren, M. 1989 Some Data on Pronominal Disjoint Reference in Warlpiri. Northern Territory Department of Education manuscript. Mahajan, A. 1990 The A/A-Bar Distinction and Movement Theory. Doctoral dissertation, Massachusetts Institute of Technology. McCloskey, J. — K. Hale 1984 "On the Syntax of Person-Number Inflection in Modern Irish", Natural Language and Linguistic Theory 1: 487-533.

Core structures and adjunctions in Walpiri syntax 219

Mithun, M. 1987 "Is basic word order universal?" in: Russell S. Tomlin (ed.), Coherence and Grounding in Discourse. Amsterdam: John Benjamins, 281-328. Nash, D. 1985

Topics in Warlpiri Grammar. New York: Garland Publishing.

Rapoport, T. 1990 "Secondary Predication and the Lexical Representation of Verbs", Machine Translation 3: 31-53. Ross, J.R. 1967 Constraints on Variables in Syntax. Doctoral dissertation, Massachusetts Institute of Technology. Simpson, J. 1987 -Kariyinyanu — a Bound Class-Marker in Warlpiri, Paper presented at the Australian Linguistics Society, 26 August, 1987. Simpson, J. — J. Bresnan 1983 "Control and Obviation in Warlpiri", Natural Language and Linguistic Theory 1: 49-64. Srivastav, V. 1991 Wh Dependencies in Hindi and the Theory of Grammar. Doctoral dissertation, Cornell University. Webelhuth, G. 1989 Syntactic Saturation Phenomena and the Modem Germanic Languages. Doctoral dissertation, University of Massachusetts, Amherst. Williams, E. 1980 "Predication", Linguistic Inquiry 11: 203-238.

Scrambling as the base-generation of random complement order Katalin E. Kiss

Introduction After a brief survey of various analyses of Scrambling phenomena in section 1, it is hypothesized that the term "Scrambling" covers different types of operations across languages. In section 2 the paper proposes to limit the use of the term "Scrambling" to word order variation that is semantically neutral - thus excluding from the notion of Scrambling e.g. Focusing, Q-raising, or movement creating a predication structure. Section 3 analyzes Scrambling in the V' component of the Hungarian sentence, and claims it to be the result of the base generation of a random complement order.

1. What is scrambling? (Various views) 1.1.

Ross (1967) attributed the freedom of Latin, Russian, Czech, etc. word order to the existence of a stylistic reordering rule in these languages. The rule, called Scrambling, was formulated as follows: c

(1) X -

NP VP Ν < V Adj Adv 2 . 3

r

·>

•

.

NP VP Ν < V Adj Adv 3 . 2

>

J

4 —> 4

OPT

Condition: Sj dominates 2 iff S, dominates 3. Since it seemed problematic to assign a tree structure to the output of this rule, Ross placed it in a component of grammar other than the transformational component, which he proposed to call the stylistic component.

222 Katalin t . Kiss

1.2.

A different approach to free word order was proposed in E. Kiss (1979, 1981) etc. concerning Hungarian, in Farmer (1980) and Hale (1980) concerning Japanese, and in Hale (1981, 1983) concerning Warlpiri. These studies derived the unconstrained order of the major constituents of the Hungarian, Japanese, and Warlpiri sentence, respectively, by base-generating a flat structure in which only the position of the head is fixed, while the rest of the constituents are generated in a random order. The rewriting rule e.g. for Japanese was formulated in Hale (1980) as follows: (2)

X' —> Χ** X

1.3.

Soon, however, evidence was presented - e.g. in Whitman (1982), or Saito (1985) - that Japanese phrase structure cannot be flat but must be of a binary branching type, with the initial argument, whatever its grammatical function, c-commanding the sentence part on its right. The strongest argument for this claim is represented by the following paradigm, quoted e.g. in Whitman (1982): (3)a.

*Karerwa Johnrno he-THEME John-GEN 'Hej loves John's mother.'

haha-o mother-ACC

karerwa aisiteiru loves he*

b.

Johnrno John's

haha-o mother

c.

Johnrno John/s

haha-wa karero mother loves

d.

*Karero Him,

Johnrno John/s

aisiteiru him;

haha-wa aisiteiru mother loves

aisiteiru love

Scrambling as the base-generation of random complement order

223

If Japanese phrase structure were flat, then (3a-d) should all be ruled out by Binding Principle C. Since (3b) and (3c) are grammatical, the pronoun in them cannot c-command the lexical NP coindexed with it. These data have lead to the revision of the Farmer-Hale type analysis of Scrambling - at least for Japanese. (The data in (3) also exclude Ross's analysis, i.e., the placement of Scrambling among stylistic rules, as a Scrambling operation performed in PF could not feed Binding.) Saito — Hoji (1983), then Saito (1985) proposed that - in Japanese Scrambling is an instance of Move alpha: adjunction to S (or, in certain restricted cases, to VP), creating an A'-chain. The analysis of Scrambling as adjunction has also been extended to languages other than Japanese, e.g. to German and Dutch - cf. den Besten — Webelhuth 1987.

1.4.

The possibility that the free word order of the German and Dutch sentence arises from the base-generation of a hierarchical (binary branching) structure in which the arguments and adjuncts are arbitrarily distributed has also been raised in van Riemsdijk (1989).

1.5.

In view of the various approaches to Scrambling, summarized in (4), we may feel tempted to ask which of the interpretations of Scrambling is "the right one". (4)

Scrambling syntactic phenomenon base-generated

stylistic (PF) phenomenon

adjunction

flat str hierarchical str

224 Katalin 6. Kiss

What is actually most likely to turn out is that Scrambling is a cover term for free word order phenomena resulting from different processes. All four possibilities may in principle be exemplified across languages. In this paper I will argue - on the basis of Hungarian material - for the existence of Scrambling as the base-generation of a flat structure of arbitrary argument order.

2. What is not scrambling? Although the term Scrambling is likely to turn out to cover various grammatical processes, it has to be made sure that it is not used simply as a metaphor for any word order variation that appears to be free with respect to grammatical functions. For example, Hungarian would certainly have been classified in Ross (1967) as a language freely applying Scrambling, similar to Latin and Russian. Since then, however, it has become clear that Scrambling is only limited to the postverbal part of the Hungarian sentence - with the preverbal part being the output of specific Move WH type movement rules. Consider the S-structure that e.g. E. Kiss (1987b; 1991a) attributes to the Hungarian sentence: (5) S XPj

VP

The distribution of constituents is independent of their grammatical functions throughout the sentence. Nevertheless, the constituents outside V' have been preposed from their base-generated positions in V' by types of Move alpha other than Scrambling. What distinguishes the

Scrambling as the base-generation of random complement order 225

rules moving material into the preverbal section of structure (5) from Scrambling is that they have well-defined semantic contents. The rule extracting XP; from V* into VP-external position, for example, creates a subject, to which the VP bears a predication relation - not only syntactically in the sense of Williams (1981), but also semantically in the sense of Aristotle — Rothstein (1983). That is, the VP-external constituent foregrounds an entity of the world the existence of which is presupposed, about which the VP will predicate something. For example: (6)a.

[s JänoSi

[VP Marival

beszelte

John-NOM

Mary-with

discussed

a problemdt

meg

PREF

tj]

the problem-ACC 'John discussed the problem with MARY.' b.

[s A

problemdtj

[VP Marival

the

problem-ACC Mary-with

Jänos

tj]

beszelte

discussed

meg

PREF

John-NOM 'The problem was discussed by John with MARY.' c.

[s Marivalj

[VP Jänos

Mary-with

John-NOM discussed

a problemdt

beszelte

meg

PREF

tj]

the problem-ACC 'Mary was discussed the problem with by JOHN.' The semantic import of the "subjectivizing" transformation illustrated in (6a-c) is especially clear if it involves an idiom chunk, which assumes reference in a VP-external position.1 Thus the idiomatic meaning of the Accusative NP in (7a) is lost in (7b);

226 Katalin t . Kiss

(7)a.

[s Jänosnaki John-DAT

[VP kinyilt opened

a bicska the pocket-knife

a zsebeben tj] the pocket-his-in 'John became angry [The pocket-knife opened in John's pocket].' b.

[s Α bicsküi the pocket-knife

[VP

kinyilt opened

Jdnosnak John's

a zsebiben tj] the pocket-in 'The pocket-knife opened in John's pocket.' What distinguishes Hungarian e.g. from English in this respect is that in English "Subjectivization" has not only a semantic motivation but is also triggered by Case considerations. As a result of this, on the one hand, not every NP to which the VP bears a syntactic predication relation also functions as a notional subject. On the other hand, the syntactic subject is restricted with respect to Case. German and Dutch are somewhat closer to Hungarian, also allowing Dative and Accusative subjects functioning as notional subjects in the case of certain verbs (cf. den Besten 1985). In Hungarian, any referential or generic argument of the V can be externalized as the subject of predication with no regard to its Case. That is, the freedom of the selection of the initial constituent of the Hungarian sentence is not due to Scrambling but follows from the lack of Case constraints on subject selection. The situation may very well be similar in Latin and in many other little known, apparently "scrambling" languages, too. The specifier position of VP is filled by Focus movement. Consider (8):

(8)

[s JänoSi John-NOM

[yp Marivalj [r Mary-with

beszelte discussed

meg PREF

a problimdt f, tj]] the problem-ACC 'It was with Mary that John discussed the problem.'

Scrambling as the base-generation of random complement order

227

Although Focus is regarded by many as merely a pragmatic phenomenon (e.g. Saito (1985: 157), Szabolcsi (1981)) clearly demonstrated that the Focusing of a constituent affects the truth conditions of a sentence. That is, Focusing in Hungarian is a Wh-type transformation, which has both a particular landing site and a particular semantic content. Quantifiers are all adjoined to VP by the iterative application of syntactic Q-Raising, triggered by the constraint that operators c-command their scope, which is observed at S-structure in Hungarian. For example: (9)

[s JänoSi John-NOM

[yp minden problemdtj [VP Marivalk every problem-ACC Mary-with

[y beszel meg f, i, tk ]]]] discusses PREF 'John discusses every problem WITH MARY.' Q-raising affects a set of linguistic elements that is considerably larger than quantifiers proper, including e.g. various types of adverbials, often realized as case-marked NPs, or constituents modified by also, too, even, etc. Some of these constituent types have only recently been identified as quantifiers by semantic research. In the Hungarian sentence, the existence of the three different preverbal operator positions is transparent; it can be decided which of them is filled also when there is only a single preverbal constituent present. The three positions can be distinguished partly on the basis of stress (to be indicated by an accent mark in (10)), partly on the basis of the relative order of the verbal prefix with respect to them, and partly on the basis of the particular interpretation they are associated with. For details, see E. Kiss (1987a: ch. 2.5.1.). Consider, for example, the following sentences: (10)a. [s Sok embert [VP Jänos [yhivott meg ]]] many person-ACC John-NOM invited PREF (Subjectivization) 'Many (particular) person were invited BY JOHN.'

228 Katalin

Kiss

b. [s Jänos lyp sok embert [VP meg [yhivott ]]] (Q-Raising) 'John invited many people.' c. [s Jänos [VP sok embert [v. hivott meg ]]] (Focusing) 'It was many people that John invited.' The point that I would like to make is that in lesser-known languages in which the metrical structure of the sentence is not known, distributional criteria are harder to find, and the slight meaning differences of the different word order variants are not clear, Subjectivization (in the above sense), Q-Raising, and Focusing may all be hard to distinguish from Scrambling. The application of the following tentative criterion might help the differentiation of Scrambling and operator movement: (11) Scrambling may not affect the semantic interpretation of the sentence.2 Let us use this criterion to decide whether the order of postverbal constituents in the Hungarian sentence is to be derived by Scrambling or by a movement rule that could be called e.g. VP-internal Topicalization. Although the order of postverbal constituents in V' is free, there are marked and unmarked orders. The preferred order appears to be affected by the same factors that determine the selection of the target of Subjectivization: thus a contextually known, unstressed constituent (e.g. a pronoun) preferably precedes a contextually new, stressed element; [+human] precedes [-human]; [+specific] precedes [-specific]; agent precedes theme. So (12a), with a specific human agent preceding a non-specific non-human theme is somewhat more felicitous then (12b), displaying the opposite order: (12)a.

Väsärolt Jänos bought John-NOM 'John bought a car.'

egy autöt a car-ACC

b. 1 Väsärolt egy autot Jänos

Scrambling as the base-generation of random complement order 229

In spite of its slightly marked character, (12b) does not differ in meaning from (12a): no context can be found in which one of them is false or ungrammatical while the other one is true and grammatical; they have the same consequences, presuppositions, etc. Hence on the basis of criterion (11), the word order variation illustrated in (12a, b) is an instance of Scrambling. The Japanese adjunction rule called "Scrambling" by Saito — Hoji (1983) and Saito (1985), or at least certain types of it, on the other hand, may very well turn out not to qualify as Scrambling on the basis of criterion (11). It appears from the literature on Japanese syntax that the preposing of a constituent in front of the subject creates - or may create - some kind of a focus (cf. e.g. McCawley 1976: 59, fn. 7; Saito 1985: 157, 260, fn. 3, 4, 5; etc.). Long distance Scrambling is actually called Emphatic Fronting by Haig (1976) and Miyara (1982). Unfortunately, however, the systematic analysis of the semantic consequences of Scrambling has not been performed; e.g. Saito's long dissertation on Scrambling (1985) merely contains a few hints concerning the semantic effects of the rule. Saito even prevents his readers from considering these effects, as he admittidly ignores the meaning differences resulting from Scrambling in the English translation of his examples (cf. Saito 1985: 260, fn. 4).3

3. Scrambling in Hungarian 3.0.

In Hungarian, free word order is restricted to the postverbal sentence part. I will argue that this sentence part (more precisely, V') is flat. In the case of a flat structure it makes no sense to derive free word order transformationally; the null hypothesis is that all possible orders are base-generated. Thus, when arguing for the non-configurationality of the sentence part containing the V and its arguments at D-structure, I will also argue for the derivation of Scrambling effects via base-generation. In the course of this, I will briefly review the major areas of Hungarian syntax that are sensitive to hierarchy, and which, therefore, have been used as arguments pro or con in the configurationality debate (involving E. Kiss 1981, 1987a, etc.; Horväth 1981, 1985; Maräcz 1989; Brödy 1991, etc.).

230 Katalin έ. Kiss

In 3.1 and 3.2, I will discuss cases in which the subject-object asymmetry expected in a configurational language is absent from Hungarian (Binding Principle C effects, and Superiority effects, respectively). In 3.3 and 3.4, I will analyze phenomena which do display argument asymmetries (anaphora and pronominal variable binding, as well as incorporation); they, however, will be shown to be thematically, and not structurally, determined. In 3.5,1 will discuss a phenomenon (idiom interpretation) in the case of which the expected distinguished role of the grammatical subject is much weaker in Hungarian than it is e.g. in English, but is not completely absent. It will be shown that what plays a distinguished role in the interpretation of idioms is the notional subject, the coincidence of which with the grammatical subject is almost complete in English, but is merely a weak tendency in Hungarian.

3.1. Disjoint reference What most clearly distinguishes Hungarian from e.g. Japanese with respect to configurationality and the status of Scrambling is that Scrambling in Hungarian does not affect the operation of Binding Principle C. That is, a pronominal argument is in disjoint reference e.g. with the genitive specifier of any of its co-arguments, no matter what their order is. This is illustrated (13): (13)a.

*Miert nem szereti a fiii^ anyja oketj? why not loves the boys' mother them 'Why doesn't the boys;' mother love then^?'

b. *Miert nem szereti oket{ why not loves them

a fiuki the boys'

c. *Μίέη nem szeretik pro/dk^ α fiüki why not love they the boys' anyjdt? mother-ACC 'Why don't they; love the boySj' mother?'

anyja? mother

Scrambling as the base-generation of random complement order

d. *Miert nem szeretik a fiu^ why not love the boys'

231

anyjät mother-ACC

pro/dkj? they The ungrammatically of (13a-d) follows from Binding Principle C if it is interpreted on a flat structure, in which coarguments mutually c-command each other. All speakers consulted, and all sources in the literature find (13a-d) fully ungrammatical, but some speakers find (13c, d) even worse (**) than (13a, b) (*). What is crucial for the analysis of Scrambling is that no speakers find any grammaticality difference between (13a) and (13b), or between (13c) and (13d). For the speakers who find (13c, d) even worse than (13a, b), the disjoint reference effect triggered by the structural binding of an R-expresion is apparently reinforced if the binder is also thematically more prominent than the R-expression. The disjoint reference between a pronoun and the genitive specifier of any of its coarguments is also preserved if the D-structure c-command relation between them is destroyed by the operator movement of either the bound element, or both the bound element and the binder. Confer: (14)a.

*[s A fiukf anyja [VP [v- nem szereti the boys mother not loves 'The boySj* mother does not love them,.'

dket( them]]]

b. *[s Α fiüki anyja [VP dketi [v· nem szereti i, tj]] the boys' mother them not loves 'The boys; mother does not love THEM/ c. *[s A fiuki anyjat [VP[V· nem the boys' mother-ACC not dk/proi r, 777 they "The boysj' mother theyj do not love.'

szeretik love

232 Katalin t . Kiss

d. *[s Α fiüki anyjät [VP ök( [ν· nem the boys' mother-ACC they not szeretik ί, f, ]}] love "The boySj* mother is not loved by ΤΗΕΜ{.' Interestingly, the disjoint reference effect is practically absent between an argument and the genitive specifier of its coargument - or between two coarguments - if both are represented by a lexical NP. For example: (15)a. IMiert nem szereti Jdnost anyja why not loves John's mother-NOM Jdnosti? John-ACC 'Why doesn't John/s mother love John;?' b. IMiert nem szereti J0nost( why not loves John-ACC

Jänos,· John's

anyja? mother-NOM c. IMiert nem szereti Jänos, JdnoSj why not loves John-NOM John's 'Why doesn't John; love John/s mother?' d. IMiert nem szereti why not loves

anyjät? mother-ACC

Jänost anyjät JänosJ John's mother-ACC John-NOM

Some speakers, among them Maräcz (1989) claim that while (15a, b), as well as (16a, b) below are fully grammatical, (15c, d) and (16c, d) are significantly worse (Maräcz even stars (16c, d>). (16)a. IJänos, anyja szereti John's mother-NOM loves 'John's mother loves John/

Jänost, John-ACC

Scrambling as the base-generation of random complement order 233

b. IJdnosti John-ACC

szereti JdnoSi loves John's

anyja mother-NOM

c. TJmoSi szereti Jdnost anyjdt John-NOM loves John's mother-ACC 'John; loves John's mother.' d. IJdnoSi John's

anyjdt mother-ACC

szereti loves

JdnoSi John-NOM

Those who find (15c, d) and (16c, d) significantly worse are speakers of the same dialect in which (13c, d) and (14c, d) are worse than (13a, b) and (14a, b); i.e., in which the disjoint reference effect triggered by Binding Principle C is reinforced if the binder of the R-expression is not only structurally but also thematically more prominent than the R-expression. Notice that the grammaticality difference that speakers of the dialect in question find between the (a, b) and (c, d) sentences in (13), (14), (15), and (16) could not be given a structural explanation. Consider the judgements given by speakers of this dialect: (17)a.

*A fiükj anyja nem szereti the boys' mother-NOM not loves "The boys/ mother does not love them/

b. *Nem szereti not loves

dketi a them the

fiukt boys'

c. **Ökt nem szeretik a fiuk{ they not love the boys' 'Theyj don't love the boys/ mother.' d. **A fiükt the boys'

anyjdt mother-ACC

oketj them

anyja mother-NOM anyjdt mother-ACC

nem szeretik pro/ök, not love they-NOM

(18)a. JdnoSi anyja nem szereti John's mother-NOM not loves 'John's mother does not love John,.'

Jdnostt John-ACC

234 Katalin t . Kiss

b. J mosti John-ACC

nem not

szereti loves

Jdnos( John's

c. llMnoSi nem szereti Jänos, John-NOM not loves John's 'Johnj does not love John's mother.' d. llJänoSi John's

anyjät mother-ACC

anyja mother-NOM anyjdt mother-ACC

nem szereti JdnoSi not loves John-NOM

Let us assume that in this dialect the subject asymmetrically c-commands the object at D-structure (either from outside the VP, as in Horvath (1985), or Maräcz (1989), or from inside the VP, as in Brödy (1991)). Then, if we check Binding Principle C on D-structure, we predict (17c, d) and (18c, d) to be ungrammatical, and (17a, b) and (18a, b) to be grammatical. However, in addition to the fact that it would be rather radical a solution to attribute completely different sentence structures to two dialects of Hungarian which only differ in assigning to sentences (17c, d) and (18c, d) different degrees (but not opposing values) of grammaticality, the ungrammaliticality of (17a, b) would remain unexplained, and the marginal acceptability of (18c, d) would also be unpredicted. These facts could not be derived by means of Scrambling, as the distribution of grammaticality in (17) and (18) is independent of surface order. I see no other (thematically based, etc.) auxiliary principle, either, by means of which the ungrammaticality of (17a, b) could be predicted. That is, while in a non-configurational framework, the unexpected data of the given dialect can be derived by a thematic auxiliary principle, in the configurational framework under consideration the data in (17a, b) and (18c, d) could not be given any principled (whether structural or thematic, etc.) account. What is more, the binding theory of Reinhart (1983) predicts that the disjoint reference effect is more regular in the case of pronoun - lexical NP pairs than in the case of lexical NP - lexical NP pairs, which is another reason to base our description on the ungrammaticality of (17a, b), instead of the acceptability of (18a, b). Reinhart's theory claims that non-coreference is determined by pragmatic strategies rather than by syntactic non-coreference rules. These strategies do tolerate exceptions. "In general, violations of the non-coreference rules which involve two full NPs are much easier to process than violations involving a pronoun

Scrambling as the base-generation

of random complement

order

235

and a full NP. ...it is generally so that the reference of a full NP is more easily recoverable than the reference of a pronoun, so ...it should be easier to identify intended coreference of two identical full NPs than of a pair of a pronoun and full NP which does not allow bound anaphora interpretation [in standard terminology: bound interpretation]." (Reinhart 1983: 170) Summarizing the discussion of Binding Principle C: Binding Principle C makes the expected predictions in Hungarian if it is applied to a non-configurational sentence structure.

3.2. Superiority The Superiority Condition of Chomsky (1976) states that if a rule has two potential targets, it has to apply to the one that is structurally superior. Thus in English multiple questions WH-movement has to prepose into SpecCP position the WH-phrase that is higher up in the tree.4 The landing site of the WH-phrase: SpecVP can host only a single XP in Hungarian, too. The second, third etc. WH-phrases in multiple questions undergo Q-Raising, which is a syntactic rule adjoining QPs to the VP in Hungarian. The interpretation of a WH-phrase undergoing Q-Raising is, in fact, non-distinct from that of a specific universal quantifier - cf. E. Kiss 1990. (In fact, as E. Kiss (1990) shows, the interpretation of an unmoved WH-phrase is non-distinct from that of a specific universal quantifier in English multiple questions, too.) Compare for example: (19)a.

ls[VP

Kii

[vp

mitj

[v•

vdlasztott

tt t} ] ] ] ]

who-NOM what-ACC chose 'What did who choose? [For each person, what did he choose?]' b.

[s [yp Miti

[yp

kij

[y

V0ldSZtOtt

ti tj

]]]]

what-ACC who-NOM chose 'Who chose what? [For each thing, who chose it?]' If Hungarian were configurational, the Superiority Condition would be expected to rule out (19a), in which of an interrogative subject and

236 Katalin t . Kiss

an interrogative object, the object is preposed into WH-position. If, on the other hand, we assume a flat D-structure, the Superiority Condition allows the preposing of either WH-phrase, i.e. the grammaticality of both (19a) and (19b) falls out.

3.3. Anaphora and weak crossover Anaphora, and Weak Crossover, i.e. in the terminology to be used below, pronominal variable binding, do display certain subject-object asymmetries in Hungarian, which is taken in Horväth (1985), Maräcz (1989), and Brödy (1991) as a decisive argument against a flat phrase structure. However, as is argued in E. Kiss (1991b), the facts of asymmetry attested could not be derived from a configurational phrase structure, either. It is claimed there that anaphora and pronominal variable binding are not purely configurationally determined phenomena; their licensing condition also involves thematic prominence and S-structure precedence. The claim that anaphora is - partly or entirely - thematically determined has a long history in generative theory beginning with Jackendoff (1972). The primacy condition of pronominal variable binding in Hungarian was first claimed to combine thematic prominence and S-structure precedence in Maräcz (1986). What is new in the present proposal (following E. Kiss 1991b) is the collapsing of the primacy conditions of anaphora and pronominal variable binding (in the spirit of Reinhart 1983), and the claim that the condition identified is not specific to Hungarian or to non-configurational languages but holds e.g. in English, too. The facts of anaphora to be accounted for are of the following type: (20)a. Nem ismeri Jänos önmagät not knows John-NOM himself-ACC 'John does not know himself.' b. Nem ismeri önmagät Jänos c. Jänos nem ismeri önmagät d. Önmagät nem ismeri Jänos

Scrambling as the base-generation of random complement order

(21)a.

237

*Nem ismeri Jänost önmaga not knows John-ACC himself-NOM '•Himself does not know John.'

b. *Nem ismeri önmaga Jänost c. * Jänost nem ismeri önmaga d. *Önmaga nem ismeri Jänost As is clear from (20)-(21), a subject can bind an object, but not vice versa. Surface order does not change grammaticality, although sentences in which the binder precedes the bindee are more fecilitous. The observed asymmetry also shows up in anaphoric relations involving arguments other than the subject and the object: (22)a.

Gyakran vitatkozom Jänossal önmagärol often argue-I John-with himself-about Ί often argue with John about himself.'

b. *Gyakran vitatkozom Jänosröl önmagäval often argue-I John-about himself-with '*I often argue about John with himself.' Notice that the distribution of grammaticality in the English glosses is the same as in the Hungarian sentences - and the ungrammaticality of the English equivalent of (22b) cannot be derived from phrase structure hierarchy, either. We can assume that in (22a) the preposition preceding John is reanalyzed as part of the V, therefore the antecedent c-commands the anaphor, as required. However, reanalysis should be just as available in the ungrammatical (22b), too, as indicated by the possibility of preposition stranding: (23)

WhOj do you often argue about i, with John?

The problematic English and Hungarian sentences can be accounted for along the same lines: by the assumption of a primacy condition for anaphora that involves, in addition to c-command, a thematic argument hierarchy, too. I assume that the thematic argument hierarchy underlying anaphora is a conceptually motivated hierarchy of arguments in the theta-grid of

238 Katalin t . Kiss

a head, with the argument most actively involved in the action/happening ordered first, and the argument least actively involved in the action/happening ordered last. (Alberti (1991) provides a formal theory with a predictive force of argument hierarchy within the theta-grid.) This hierarchy could most simply be recorded in grammar by the linear order of arguments in the theta-grid. Notice that a thematic argument hierarchy expressed by a so-called lexical structure patterned after configurational phrase structure (as proposed in Hale 1983) would face the same problems that the structural account of anaphora faces e.g. in English. The problem arises, as noted e.g. by Barss — Lasnik (1986), if the binding relation holds between two VP-internal arguments that are of the same category (both are NPs or PPs). In this case a hierarchical structure can either express that both arguments are equally prominent - if it is constructed as (24a, b), or it wrongly predicts the argument that is further away from the V to be the more prominent one - as in (24c). (24)a.

b.

V^PP

PP

c.

VX / " x . PP PP

VP ^ V

pp \ PP

The fact that the structural hierarchy in (24c) is the opposite of the thematic argument hierarchy underlying anaphora is especially clear if one of the arguments involved in the binding relation is an object: (25)

A

sziildk

the parents-NOM

övtdk

a

gyerekeketi

protected

the

children-ACC

egymästöli

each-other-ABL "The parents protected the children; from each other;.'

Scrambling as the base-generation of random complement order 239

S NP-NOM

VP

VP NP-ABL v " NP-ACC α

szüldk

övtäk a gyerekeketj egymdstoli

In the structure assigned to (25), either the anaphor c-commands the antecedent, or the theme/direct object is not a sister of the V, both of which are illegitimate assumptions in GB theory (though for an attempt to legitimitize the latter assumption, see Larson (1988)). The problem can be eliminated if we assign to (25) a structure like (24a), with the two NPs mutually c-commanding each other, and if the primacy condition of anaphora makes reference both to c-command and to the thematic argument hierarchy encoded in the ordering of arguments in the theta-grid, along the following lines: (26) The primacy condition of anaphora (to be revised) An anaphor must be bound by an antecedent that both c-commands it at NP-structure, and precedes it in the thematic argument hierarchy. (NP-structure, i.e., the post-NP movement representation of the sentence (cf. Van Riemsdijk — Williams 1981), is non-disctinct from D-structure in the case of Hungarian.) The claim that c-command interpreted on a flat propositional component plays a role in Hungarian, too, can be illustrated by the following minimal pair: (27)a. Ajdndekot kiildtem [Jdnosnak äs present-ACC sent-I John-DAT and

MarinakJi Mary-DAT

egymässali each-other-INSTR Ί sent presents to John and Mary with each other.'

240 Katalin t . Kiss

b. *Ajdndekot present-ACC

küldtem [Jdnos sent-I John

es and

MariJi Mary

szdmdra egymdssal, for each-other-INSTR Ί sent presents for John and Mary with each other.' If Dative Case and the Ρ szdmdra 'for' encode the same theta role, then the only factor distinguishing the ungrammatical (27b) from the grammatical (27a) is the lack of c-command.5 If we extend the analysis to anaphoric relations between non-coarguments, as well, it becomes clear that S-structure precedence also plays a role in the licencing of anaphora both in Hungarian and in English. Consider the relevant examples: (28)a.

Sokat much

vitatkoztam argued-I

[Jdnossal John-with

is and

Marival]i Mary-with

egymäSi tandrairol each other's teachers-about Ί argued a lot with John and Mary about each other's teachers.' b. Ί Sokat vitatkoztam egymdsi tanärairol [Jdnossal es MarivalJi '*I argued a lot about each other's teachers with John and Mary.' c. Π Sokat vitatkoztam [Jdnosröl es Mariröl]^ egymdst tandraival Ί argued a lot about John and Mary with each other's teachers.' d. *Sokat vitatkoztam egymdsi tandraival [Jdnosrol is Mariroljj '*I argued a lot with each other's teacher about John and Mary.' The grammaticality of the English equivalents in (28a, c and d) coincides with that of the Hungarian sentences. The grammaticality difference between the English and Hungarian versions of (28b) follows from the phrase structure difference between the two sentences: in the

Scrambling as the base-generation of random complement order 241

English sentence the antecedent is subordinated to a Ρ head in a position where the head cannot be reanalyzed as part of the V; therefore, the requirement of c-command is violated. The Hungarian and English sentences in (28) argue for the following reformulation of condition (26): (29)

The primacy condition of anaphoric binding: An antecedent can bind an anaphor if it c-commands the anaphor or its trace at NP-structure, and at least (i) or (ii) holds: (i) it precedes the anaphor in the lexical argument hierarchy; (ii) it precedes the anaphor at S-structure.

(30)

A precedes Β in the lexical argument hierarchy if A is ordered prior to Β or to the constituent containing Β in the theta-grid.

While condition (29) correctly accounts for anaphoric relations between non-coarguments, it is too powerful in licencing anaphoric relations between coarguments: it would also licence e.g. (21a) or (22b), which observe both the c-command and the precedence constraints. What has to be ensured is that the precedence clause of principle (29) only have consequences for binding if there is no direct ordering relation between the binder and the bound element in the theta-grid (i.e., if they are not coarguments). To achieve this goal, I propose the following thematic non-coreference principle: (31)

An anaphor cannot precede its antecedent in the thematic argument hierarchy.

In E. Kiss (1991b) principle (29) is shown to license not only anaphoric binding, but pronominal variable binding, too - i.e., the term "anaphoric binding" can be generalized to "binding", and the term "anaphor" can be generalized to "bound element" in it. For instance, it correctly accounts for the distribution of grammaticality in both the Hungarian and the English versions of (32a-d), which are just as problematic for the standard c-command condition of binding as similar constructions involving anaphora are.

242 Katalin έ. Kiss

(32)a. Mindegyik ferfiva^ sokat each man-with much

beszelgettem talked-I

a the

prOi felesigirdl his wife-about Ί talked a lot with each mai^ about hiSi wife.' b. ?A prOi felesig0rdl mindegyik ßrflvali sokat beszelgettem *?I talked a lot about his; wife with each mai^.' c. ?Mindegyik firfiröli sokat each man-about much

beszelgettem talked-I

a the

prOi felesigivel his wife-with '?I talked a lot about each mar^ with hiSj wife.' d. *A prOi felesigfrel mindegyik fέφτό/, sokat beszelgettem '*! talked a lot with hiSj wife about each marij.' Consider also: (33)a. ΊAbout hist wife, I talked a lot with each man,. b. *With hist wife, I talked a lot about each man. The English variant of (32b) is not ruled out as a violation of the c-command requirement certainly because the feature [+quantifier] can percolate up from the NP to the PP containing it (cf. Harbert 1989). Examples in which the bound pronoun occupies a position other than that of a genitive specifier also confirm the correctness of condition (29): (34)a. Mindegyik länyt felismerte each girl-NOM recognized

azt that

a ferfit, man-ACC

aki benyitott a pro, szobäjäba who entered the her room 'Each girlj recognized the man who entered Ιιεη room.'

Scrambling as the base-generation of random complement order 243

b. lAzt a ferfit, aki benyitott a proi szobäjäba, mindegyik länyt felismerte "?The man who entered he^ room, each girlL recognized.' c. ΊMindegyik each

länyt( girl-ACC

felismerte az a firfi, recognized that man-NOM

aki benyitott a prot szobäjäba who entered the her room '•Each girl;, the man who entered her; room recognized.' d. *Az a firfl, aki benyitott a pro, szobäjäba, mindegyik länyti felismerte '*The man who entered he^ room recognized each girl;.'6 The grammaticality difference between the Hungarian and English versions of (34c) follows from the phrase structure difference between the Hungarian and the English sentence. The English sentence is out because the c-command condition of pronominal variable binding is not satisfied: the object quantifier does not c-command the pronoun embedded in the subject at the level of NP-structure. In the flat NP/D-structure of the Hungarian sentence the c-command relation holds. Anaphora and pronominal variable binding in the context of psych verbs also involve phenomena which would be hard or impossible to derive from a configurational structure. Let us consider the type of data to be accounted for: (35)a. Jänost aggasztja önmaga John-ACC worries himself-NOM '•Himself worries John.' b. Uänos aggasztja John-NOM worries 'John worries himself.' erdekli (36)a. Jänost •Himself interests

önmagät himself-ACC

önmaga John.'

244 Katcdin t . Kiss

b. llJänos '??John (37)a.

ärdekli interests

önmagät himself.'

IJänost meghatotta önmaga •Himself moved John.'

b. *Jänos meghatotta '*John moved

önmagät himself.'

(38)a. Jänosnak tetszik önmaga John-DAT pleases himself-NOM '•Himself pleases John.' b. Jänos tetszik önmagänak7 John-NOM pleases himself-DAT 'John pleases himself.' (39)a. Α länyokatj the girls-ACC

nagyon aggasztjäk egymäSi very-much worry each-other's

pärtfogöi supporters-NOM '??The girlsj, each other's supporters worry a lot.' b. ΊEgymäsi pärtfogöi nagyon aggasztjäk α länyokati '?Each other;'s supporters worry the girlsj a lot.' c. ΊΑ länyoki nagyon aggasztjäk egymäs( the girls-NOM very-much worry each-other's pärtfogöit supporters-ACC 'The girls, worry each other/s supporters a lot.' d. *Egymäst pärtfogoit nagyon aggasztjäk α länyoki '•Each other/s supporters, the girls; worry a lot.'

Scrambling as the base-generation of random complement order 245

(40)a. Minden every köreje

politikusti politician-ACC

bosszantottak annoyed

a the

prot him

seregldk

around flocking-NOM 'Every politician was annoyed by those flocking around h i m / b. ΊΑ pro, the him

köreje around

seregldk flocking-NOM

minden every

politikusti bosszantottak politician-ACC annoyed 'TThose flocking around him; annoyed every politician/ c. Minden

every körej

politikuSi

bosszantotta

a

politician-NOM

annoyed

the him

prot

sereglöket

around flocking-ACC 'Every politician) annoyed those flocking around him;.' d. *A prOi köreje sereglöket minden politikuSj bosszantotta

'•Those flocking around himi( every politician annoyed.' If we attribute to Hungarian sentences containing a psych verb the configurational structure proposed in Belletti — Rizzi (1985), then we cannot predict the grammatically of the (a) sentences in (35)-(38), the English, Italian etc. equivalents of which are all ungrammatical, ruled out as Binding Principle C violations.8 Belletti — Rizzi predict that in case the antecedent and the anaphor mutually c-command each other, Binding Principle C is inapplicable - that is why e.g. Ho affidato a Maria se stessa / Ho affidato Maria a se stessa are both grammatical. They give the following explanation: "If we assume that binding is intrinsically asymmetric (Higginbotham 1983), in cases of symmetric c-command we have to choose the direction of the binding (perhaps linear order gives the favored direction). Then, the antecedent would be free even if it is c-commanded by the anaphor." (Belletti — Rizzi 1985: 29).

246 Katalin t . Kiss

In the framework proposed in this paper, the grammatically of (35a), (36a), (37a), and (38a) falls out straightforwardly - if we accept the general assumption that the experiencer argument of a verb is thematically more prominent than its theme. I assume that the (b) sentences in (35)-(37) are more or less also acceptable because the V's can be attributed - to varying extents - an agentive meaning, as well. This hypothesis seems to be confirmed by the observation that the less the theme can be attributed the conceptual role of an active, intentional cause (cf. meghat 'move* versus aggaszt 'worry'), the worse the theme will sound as a binder. Tetszik remains an exception: its theme can bind its experiencer, in spite of the fact that the verb cannot be attributed an agentive meaning. I suspect that the theta-grid of please simply cannot be identical with that of aggaszt 'worry', erdekel 'interest', or meghat 'move' - or else it would not be mapped on a different case frame/phrase structure across languages (cf. Belletti — Rizzi 1985). Perhaps the Dative argument of please can alternatively be interpreted either as an experiencer or as a goal (cf. Haik 1990). Summarizing the discussion of anaphora and Weak Crossover: I have argued that anaphora and pronominal variable binding are licensed across languages by a primacy condition that involves phrase structure c-command, thematic prominence, and precedence, as well. The differences attested between Hungarian and English can all be derived from the assumption that the c-command condition applies to an asymmetric structure in English and to a flat structure in Hungarian. If we assign to the Hungarian sentence a configurational structure, this explanation is not available - hence the grammaticality of e.g. (28b), (34c), (35a), (36a), (37a), or (38a) remains unaccounted for.

3.4. Incorporation In Hungarian objects, instrumentals, and locatives can be incorporated into the V; subjects, on the other hand, cannot, unless they are theme subjects of ergative verbs. This is a fact observed across languages. Compare for example: (41 )a.

Vendeg jött guest came ['There was guest-coming.']

Scrambling as the base-generation of random complement order 247

b. *Vendäg olvasott guest read ['There was guest-reading.'] These data are evaluated e.g. in Maräcz (1989) as evidence for the configurationality of the Hungarian sentence. In the literature we find two accounts of the impossiblity of non-ergative subject incorporation: (i) Traditionally (cf. Chafe 1970; Mithun 1984) it is assumed that incorporation is thematically licensed, (ii) Baker (1985) claims that it is constrained by the ECP; a VP-external subject cannot be incorporated because it would leave a non-properly governed trace. As is generally acknowledged, subject extraction in Hungarian does not display any ECP effects; consequently, explanation (ii) is not available in Hungarian. What we are left with is the thematic account.

3.5. Idiom interpretation The "idiom argument" for configurationality is based on the assumption that the semantic composition of arguments with the predicate proceedes on phrase structure from bottom to top. It has been observed that in English idioms the non-idiomatic varying element coincides with the subject, and the idiomatic meaning is confined to the VP. Given the above assumption about the order of semantic composition, this falls out if the subject is the highest argument in sentence structure, asymmetrically c-commanding the object. In a "flat", nonconfigurational language the coincidence of the non-idiomatic element with the subject is not expected to be found. The validity of the assumption on which this argument is based, however, has been questioned e.g. by Bresnan (1982). She claims that there is no reason even within the framework of compositional semantics why the order in which the arguments are composed with the predicate should be encoded structurally. If the premises of the idiom-argument were right, we would not find any counter-examples; but Bresnan enumerates a number of them. Haik (1990) accounts for some of the counter-examples (e.g. for What's eating Mary?) by claiming that they involve a psych verb, i.e., they are associated with a D-structure in which the non-idiomatic argument c-commands the

248 Katalin t . Kiss

idiomatic one. Haik's explanation, however, seems to ignore the fact that in the D-structure associated with an idiomatic psych verb the non-idiomatic argument will remain c-commanded by the idiomatic verb; i.e., the alleged requirement that the composition of the idiomatic sentence part precede the addition of the non-idiomatic argument will not be observed. It will not be observed in such examples as God bless x; The devil take jc; The cat's got x's tongue, either. Nevertheless, the question what makes the coincidence of the non-idiomatic varying element of idioms with the grammatical subject a strong tendency in English remains to be answered. In Hungarian, the idiom-collection of e.g. O. Nagy (1966) enumerates scores of idioms in which the varying element is other than the grammatical subject. Consider for example: (42)a. Juliskdnaki [VP leesett [NP az tt άΙΙα]] Julie-GEN fell the jaw-NOM 'Julie's jaw fell [Julie got surprised]' b. Petertf [VP elkapta a gepszij t j Peter-ACC got the driving-belt-NOM "The driving belt has got Peter [Peter has been roped in]' c. Joskdbcij [VP belebüjt az ördög f, ] Joe-into crept the devil 'The devil has crept into Joe.' Although the number of such idioms is large, the number of idioms in which the non-idiomatic element coincides with the nominative argument is even larger in Hungarian, too. The question is why. What English and Hungarian idioms share is that their non-idiomatic element - whether or not it coincides with the nominative NP - is typically represented by the VP-external argument, functioning as the notional subject. The reason for this fact is that the notional subject must be referential (or generic), which follows from its role of selecting and foregrounding an entity of the world. In Hungarian, predication is interpreted at S-structure (there being no LF-representation distinct from S-structure - cf. E. Kiss 1987, 1991a). Consequently, every argument with which the VP is in a syntactic predication relation at S-structure will be interpreted as a notional

Scrambling as the base-generation of random complement order 249

subject, and will involve a referentiality requirement. Recall the discussion of (7), demonstrating that an idiom-chunk in pre-VP position will be attributed reference, losing its idiomatic meaning. In English, predication is interpreted at LF (cf. Rothstein 1983). Since operators are removed from subject position at LF, they are not subject to the referentiality requirement of notional subjects; hence they can retain their idiomatic meaning, as happens e.g. in What's eating him? The externalizability of a non-operator element of an idiom depends on whether or not it can be attributed reference. The bucket in The bucket was kicked by χ cannot be attributed reference; advantage in Advantage was taken ofx, or God in God bless x, on the other hand, can. (Another potential explanation of the acceptability of God bless you and Advantage was taken of him is that God and advantage, bearing the strongest stress of their respectice sentences, are foci, removed from the subject position at LF.) Summarizing this point: the idiom argument supports the non-configurational character of the Hungarian sentence, but for a reason other than what is generally assumed. As I argued, the subject of predication must be referential, therefore, it must be represented by the non-idiomatic, varying element in idiomatic sentences in English and in Hungarian alike. What is restricted in English and free in Hungarian is the selection of the subject of predication from among the arguments of the predicate.

4. Conclusion The facts surveyed in section 3 suggest that Scrambling is to be interpreted in Hungarian as the random base generation of arguments in a flat structure. Let me also add as a methodological conclusion that this claim cannot be refuted by merely pointing out argument asymmetries in Hungarian; it would also have to be shown how exactly these asymmetries fall out from a configurational structure.

Notes 1.

The use of the more traditional term "Topicalization" is avoided because it might be confusing that Topicalization is applied e.g. in German and Dutch

250 Katalin t. Kiss

grammar to a process which is functionally Focusing, unless it proposes the grammatical subject. 2.

It remains a question requiring a cross-linguistic investigation whether the condition in (11) can be strengthened to state that Scrambling may not change the truth conditions of a sentence. In Hungarian the stronger version of (11) could also be maintained. Focusing and Q-Raising undoubtedly change the truth conditions of a sentence (cf. Szabolcsi 1981, and fi. Kiss 1987, 1991a), and Subjectivization may also change them e.g. when it involves an idiom-chunk - cf. the discussion of (7).

3.

This criticism does not apply to the literature on Japanese contained in this volume.

4.

There have been attempts to derive Superiority from the ECP, but as e.g. Pesetsky (1987) showed, they could only account for a part of Superiority facts.

5.

Williams (1987) denies the role of phrase-structure c-command in anaphora altogether, by claiming that (i) is not ungrammatical, merely marginal: (i) ?John talked about herself to Mary.

6.

This example has been adapted from Brödy (1991), where it is intended to prove the existence of Weak Crossover effects in Hungarian. Consider Brödy* s version of the crucial (34c) ((4b) in Brödy (1989)): (i)

*Mindenkiti felismert everybody-ACC recognized

α the

färfi man

aki who

benyitott entered

a prOi szobdjdba the his room ""The man who entered his; room recognized everybody,.'' Binding is, indeed, impossible or very hard in (i), although it becomes easier if the bound reading gets some contextual support - as e.g. in (ii). (Recall that since the thematic precedence condition of binding is not satisfied in (ii), the expected grammaticality is also somewhat marginal.) (ii)

V.Minden eliiiltet, felismert every prisoner-ACC recognized aki benyitott who entered

a prot the his

celldjdba cell.

az that

α the

foglär guard-NOM

Scrambling as the base-generation of random complement order

251

'The quard who entered his, cell recognized every prisonerj.' The acceptability of (43c) and (ii) suggests that the reason for the lack of a bound reading of (i) is not structural. (True cases of disjoint reference remain disjoint also if the situation makes a coreferent reading likely.) My intuition is that what makes (i) wrong is a conflict between the non-specificity of the object, and its distributivity with respect to the NP his room. Mindenki in itself actually allows either a [+specific] or a [-specificity] interpretation. However, when used as an object it requires the indefinite V-conjugation, triggered by non-specific objects (cf. Comrie 1975), which blocks the possibility of a [+specific] reading. Therefore, an object involving minden 'every', unlike a subject, cannot appear in a position requiring a specific NP (for example in an operator position with scope over a specific operator). For example: (iii)

*Mirtden

länyt

mindegyik

every

girl-ACC

each

fid

felktrt

boy-NOM

asked-INDEF

täncolni

to-dance 'Every girl, each boy asked to dance.' Cf. (iv)

Mindegyik

Idnyt

mindegyik

fiii

each

girl-ACC

each

boy-NOM

felkirte

täncolni

asked-DEF to-dance 'Each girl, each boy asked to dance.' (iii) improves if the indefinite verbal suffix, blocking the specific interpretation of the object, is replaced by the definite suffix, although the use of a definite suffix in the context of an object involving minden formally violates the rules of grammar. (v) V.Minden

every

Idnyt

mindegyik

girl-ACC

each

fiii

boy-NOM

felkirte

täncolni

asked-DEF to-dance

The fact that in (i) the determiner azt, indicating the restrictiveness of the relative clause, has been omitted also makes the bound reading harder to obtain, because in lack of azt the relative clause, intended to be restrictive, is formally indistinguishable from a non-restrictive one.

252 Katalin t. Kiss

7.

In the (a) sentences of (35)-(38), the reflexive could not be replaced by a reciprocal pronoun; the reason for this, however, is morphological. My intuition is that egymds 'each other' is not marked for number; whether the V is in singular or in plural, it sounds as if it did not agree with the subject. Cf. (i) *Jdnost John-ACC

is and

Marit Mary-ACC

aggasztja / worry-3.SG /

aggasztjäk worry-3.PL

egymds each-other-NOM 'Each other worry John and Mary.' 8.

A configurational structure assigned e.g. to (35a) would look as follows:

s

(i) NP önmagai

vp ^ ^

VP V aggasztja

NP

Jänost

NP q

'Himself worries John.'

References Alberti, G. 1991 Argument Structure and Thematic Roles. Doctoral dissertation, Linguistic Institute of the Hungarian Academy of Sciences. Baker, M. 1985 Incorporation. Doctoral dissertation, Massachesetts Institute of Technology.

Barss, A. — H. Lasnik 1986 "A Note on Anaphora and Double Objects", Linguistic Inquiry 17, 347-354. Belletti, A. — L. Rizzi 1985 "Psych-Verbs and Th-Theory", Lexicon Project Working Papers 13, Massachesetts Institute of Technology. Besten, Η. den 1985 "The Ergative Hypothesis and Free Word Order in Dutch and German", in: J. Toman (ed.), Studies in German Grammar. Dordrecht: Foris, 23-64.

Scrambling as the base-generation of random complement order 253

Besten, Η. den — G. Webelhuth 1987 Remnant Topicalization and the Constituent Structure of VP in the Germanic SOV Languages. Talk presented at the 1987 Venice GLOW Colloquium. Bresnan, J. 1982 "Control and Complementation", Linguistic Inquiry 13, 343-434. Brödy, M. 1991 "Some Remarks on the Focus Field in Hungarian", in: I. Kenesei (ed.), Approaches to Hungarian III. Szeged: JATE. Chafe, W. 1970 "A Semantically Based Sketch of Onondaga", UAL Memoir 36. Chomsky, N. 1973 "Conditions on Transformations", in: S.R. Anderson — P. Kiparsky (eds.), A Festschrift for Morris Halle. New York: Holt, Rinehart and Winston, 232286. Comrie, B. 1975 "Subjects and Direct Objects", itudes Finno-Ougriennes, tome ΧΠ, 5-17. Farmer, A. 1980 On the Interaction of Morphology and Syntax. Doctoral dissertation, Massachesetts Institute of Technology. Haig, J. 1976

"Shadow Pronoun Deletion in Japanese", Linguistic Inquiry 7, 363-371.

Haik, I. 1990

"Telling 'tell'", GLOW Newsletter 24, 28-29.

Hale, K. 1980

Hale, K. 1981

Hale, K. 1983

"Remarks on Japanese Phrase Structure; Comments on the Papers on Japanese Syntax", in: Yukio Otsu — Anne Farmer (eds.), Theoretical Issues in Japanese Linguistics, ΜΓΓ Working Papers 2, 185-203.

On the Position ofWarlpiri in a Typology of the Base. Bloomington: Indiana University Linguistics Club.

"Warlpiri and the Grammar of Non-Configurational Languages", Natural Language and Linguistic Theory 1, 5-47.

Harbert, W. 1989 Subjects of Prepositions. Ms, Cornell University.

254 Katalm £ Kiss

Honrath, J. 1981 Aspects of Hungarian Syntax and the Theory of Grammar. Doctoral dissertation, UCLA. Horvath, J. 1985 Focus in the Theory of Grammar and the Syntax of Hungarian. Dordrecht: Foris. Jackendoff, R. 1972 Semantic Interpretation in Generative Grammar, Cambridge, Mass.: ΜΓΓ Press. ß . Kiss, K. 1979 Focus and Topic: the Marked Constituents of the Hungarian Sentence. Paper presented at the Pisa GLOW Colloquium. 6 . Kiss, K. 1981 "Structural Relations in Hungarian, a 'Free' Word Order Language", Linguistic Inquiry 12, 381-416. έ . Kiss, K. 1987a Configurationality in Hungarian. Dordrecht: Foris. έ . Kiss, K. 1987b "Is the VP Universal?", in: I. Kenesei (ed.), Approaches to Hungarian II, Szeged: JATE. Ü. Kiss, K. 1990 WH-Movement and Specificity. Ms, Linguistic Institute of the Hungarian Academy of Sciences. έ . Kiss, K. 1991a "Logical Structure in Syntactic Structure", in: James Huang — Robert May (eds.), Logical Structure and Syntactic Structure. Dordrecht: Foris. E. Kiss, K. 1991 b "On the Locality Condition of Anaphora and Pronominal Variable Binding", in: Jan Koster — Eric Reuland (eds.), Long Distance Anaphora. Cambridge: Cambridge University Press. Larson, R. 1988 "On the Double Object Construction", Linguistic Inquiry 19, 335-391. Maräcz, L. 1986 "On Coreferential and Bound Variable Interpretation in Non-Configurational Languages", Theoretical Linguistic Research 2, 85-172.

Scrambling as the base-generation of random complement order 255

Maräcz, L. 1989 Asymmetries in Hungarian. Doctoral dissertation, Groningen University. McCawley, N. A. 1976 "Reflexivization: A Transformational Approach", in: M. Shibatani (ed.), Syntax and Semantics 5, Academic Press. Mithun, M. 1984 "The Evolution of Noun Phrase Incorporation", Language 60, 847-895. Miyara, S. 1982 "Reordering in Japanese", Linguistic Analysis 9.4. O. Nagy, G. 1966 Magyar szöläsok is kozmonddsok. Budapest: Gondolat. Pesetsky, D. 1987 "Wh-in-Situ: Movement and Unselective Binding", in: E. Reuland — A. ter Meulen (eds.), The Representation of(In)definiteness, Cambridge, Mass: ΜΓΓ Press. Reinhart, Τ. 1983 Anaphora and Semantic Interpretation. London: Croom Helm. Riemsdijk, H. van 1989 Talk presented at the Lexicon, Syntax, Interpretation workshop of the Inter-University Center, Dubrovnik. Riemsdijk, H. van — E. Williams 1981 "NP-Structure", The Linguistic Review 1, 171-218. Ross, J. 1967

Constraints on Variables in Syntax. Doctoral dissertation, Massachesetts Institute of Technology.

Rothstein, S. 1983 The Syntactic Forms of Predication. Doctoral dissertation, Massachusetts Institute of Technology. Saito, M. 1985

Some Asymmetries in Japanese and their Theoretical Implications. Doctoral dissertation, Massachusetts Institute of Technology.

Saito, Μ. — H. Hoji 1983 "Weak Crossover and Move Alpha in Japanese", Natural Language and Linguistic Theory 1, 245-259.

256 Katalin t. Kiss

Szabolcsi, A. 1981 "The Semantics of Topic-Focus Articulation", in: J. Groenendijk et al. (eds.), Formal Methods in the Study of Language. Amsterdam: Matematisch Centrum. Whitman, J. 1982

Configurationality Parameters. Ms, Harvard University.

Williams, E. 1980

"Predication", The Linguistic Inquiry 11, 203-238.

Williams, E. 1987

"Implicit Arguments, The Binding Theory, and Control", Natural Language and Linguistic Theory 5, 151-180.

Towards resolving Webelhuth's paradox: evidence from German and Korean Young-Suk Lee — Beatrice Santorini

1. Introduction In the framework of Chomsky (1981), a fundamental theoretical distinction is drawn between Α-positions and Α-bar positions.1 Elements in Α-positions and Α-bar positions have distinct syntactic properties - the former can serve as antecedents for binding, are not generally held to license reconstruction and do not license parasitic gaps, whereas the latter cannot serve as antecedents for binding but license both reconstruction and parasitic gaps. Chomsky (1981: 47) defines an Α-position as "one in which an argument such as a name or a variable may appear in D-structure; it is a potential thetaposition", and subjects occupy Α-positions according to his definition. In recent years, developments in the theory of phrase structure and empirical advances in the study of languages other than English have called into question Chomsky's definition and the A/Abar distinction based upon it. Adopting an idea originally due to Fillmore (1968) and McCawley (1970), a number of syntacticians have proposed the so-called VP-Internal Subject Hypothesis, according to which subjects originate and are assigned a theta-role in a position dominated by a maximal projection of the verb (Fukui 1986; Fukui — Speas 1986; Kitagawa 1986; Koopman — Sportiche 1988; Kuroda 1987; Sportiche 1988). Several different variants of this hypothesis have been proposed, with some authors arguing that subjects originate in the specifier position of VP and others, that they occupy a position adjoined to V-max.2 However, since all variants of the hypothesis have in common that the derived subject position is no longer a potential theta-position, its properties are now exactly the reverse of what is expected given the received definition of Apositions. An even more serious challenge to the traditional A/A-bar distinction is raised by what we will call "Webelhuth's paradox", which is based on the observation that scrambled phrases in German can simultaneously serve as binders and license parasitic gaps (Webelhuth 1989), thus exhibiting supposedly incompatible syntactic properties. In this paper, we will focus on the paradox raised for the A/A-bar distinction by the properties of scrambling in German and Korean

258 Young-Suk Lee — Beatrice Santorini

and present a proposal to resolve it. Given the unclear status of the A/A-bar distinction, we will follow Deprez (1989) in referring to the properties associated in Chomsky (1981) with chains headed by A and Α-bar positions by the neutral labels "anaphoric properties" (Anproperties) and "variable properties" (Vbl-properties), respectively, wherever it is necessary to forestall terminological confusion. In Section 2, we review the An-properties and Vbl-properties of scrambling that have been discussed in the literature (Deprez 1989; Frey 1990; Haider 1989; Mahajan 1990; Miyagawa 1990; Saito 1990; Tada 1990; Webelhuth 1989; Vanden Wyngaerd 1989). As has been widely noted, scrambling generally exhibits An-properties with regard to the weak crossover effect. That is, like raising in English, it creates binding relations between quantified expressions and bound pronouns that do not obtain on the basis of canonical word order and conversely, it can destroy existing binding relations. However, in the special case of scrambling across a subject, scrambling can license reconstruction, a Vbl-property.3 Finally, we show that scrambled constituents can simultaneously act as binders and license parasitic gaps in German. In Sections 3 and 4, we review and reject two previous analyses of these facts. According to the first analysis, proposed by Mahajan (1990) on the basis of Hindi but meant to extend to German and other scrambling languages as well, An-chains and Vbl-chains are distinguished by the structural position of the head of the chain An-chains are headed by elements in specifier positions, whereas Vbl-chains are headed by elements in adjoined positions. Assuming a highly articulated phrase structure along the lines of Pollock (1989) and Chomsky (1989a), Mahajan argues that scrambling is a nonunitary phenomenon - either substitution into the specifier position of various functional heads, or adjunction to the maximal projection of such heads - and that this is what gives rise to the mixed character of scrambling. Despite its apparent conceptual economy, we will show that Mahajan's approach is stipulative and empirically inadequate. According to the second analysis, proposed by Frey (1990) for German, scrambling is a unitary phenomenon, and all instances of it are adjunction to V-max. Frey argues that the reconstruction effects associated with scrambling across the subject are only apparent and that they can be derived without assuming reconstruction from the special status of subjects with regard to

Towards resolving Webelhuth's paradox: evidence from German and Korean

259

subject-verb agreement. We reject Frey's analysis as it stands mainly on comparative grounds - it does not extend naturally to languages like Korean or Japanese, in which subject-verb agreement has been argued to be deficient or lacking entirely (Fukui 1986; Lapointe 1990; Lee 1990 contra Ahn — Yoon 1989; Whitman 1989). Our own proposal, which we present in Section 5, shares a number of important assumptions with Frey (1990). Like him, we analyze scrambling as a formally uniform process of adjunction. Moreover, we adopt his unorthodox position that the properties of chains with regard to binding and reconstruction do not depend on the structural position of the elements heading them. Instead, we assume with Frey that An-properties are characteristic of chains formed by local movement. Elements in adjunction positions can therefore serve as binders, as long as their position is derived by movement within a local domain (and they satisfy other relevant conditions). Conversely, only elements in positions derived by nonlocal movement are eligible for reconstruction. We differ from Frey, however, in distinguishing two types of local domains: a narrow one defined in terms of syntactic argument structure and a broader one defined in terms of the presence of AGR.4 We propose further that parasitic gaps are licensed by elements in structural positions that are neither theta-marked nor case-marked. Finally, Section 6 briefly summarizes our results and outlines future research.

2.

Properties of scrambling in German and Korean

2.1.

Weak crossover

Consider the contrast in English between (1), an instance of weak crossover, and (2).5 (1)

*His mother loves everyone.

(2)a. b.

Everyone loves his mother. Who loves his mother?

This contrast shows that a pronoun must be c-commanded by a quantified expression in order to be interpreted as bound by it.6 As

260 Young-Suk Lee — Beatrice Santorini

is well known, raising a quantified expression to (or through) a ccommanding subject position eliminates weak crossover by creating new binding possibilities that are not available on the basis of underlying word order. This is illustrated by the contrast between (3) and (4). (3)

*It seems to his mother that everyone is charming.

(4) a. b.

Everyone, seems to his mother tL to be charming. WhOi t'i seems to his mother i, to be charming?

Conversely, raising a constituent containing a potential bound pronoun across a quantified expression to a subject position destroys binding possibilities and gives rise to a weak crossover configuration, as shown by the contrast between (5) and (6).7 (5)

It seems to everyone that his mother is charming.

(6)

*[His mother]t seems to everyone tt to be charming.

Now consider the examples of scrambling in (7) and (8), where the subject contains a bound pronoun and the direct object is a potential binder (cf. Webelhuth 1984: 211, 1989: 408).8 The sentences in (7), which reflect canonical word order, exhibit weak crossover.9 (7) a.

*daß seine Kinder that his-NOM children-NOM

jeden everyone-ACC

liebhaben (G) dear-have Intended meaning: 'that everyone is loved by his children.' b.

*pro chinkwu-ka nwukwuna-lul pro-GEN friend-NOM everyone-ACC paypanhayssta (K) betrayed Intended meaning: 'Everyone was betrayed by his friend.'

Towards resolving Webelhuth's paradox: evidence from German and Korean 261

Scrambling the direct object across the subject, as in (8), eliminates weak crossover. The effect of scrambling in (8) is thus parallel to that of raising in (4). (8)a.

daß jeden{ seine Kinder that everyone-ACC his-NOM children-NOM liebhaben (G) dear-have 'that everyone is loved by his children'

b.

nwukwuna-luli pro chinkwu-ka f, everyone-ACC pro-GEN friend-NOM paypanhayssta (K) betrayed 'Everyone was betrayed by his friend.'

Scrambling a direct object binder across an indirect object containing a bound pronoun is parallel to scrambling across a subject. The sentences in (9), with the indirect object preceding the direct object, reflect canonical word order. (9)a.

*daß Maria seinem Nachbarn that Maria his-DAT neighbor-DAT

jeden everyone-ACC

vorgestellt hat (G) introduced has Intended meaning: 'that Maria introduced everyone to his neighbor.' b.

*Kim pancang-i pro iwus-eykey Kim district chair-NOM pro-GEN neighbor-DAT nwukwuna-lul sokayhayssta (K) everyone-ACC introduced Intended meaning: 'The district chair Kim everyone to his neighbor.'

introduced

262 Young-Suk Lee — Beatrice Santorini

Scrambling the direct object across the indirect object eliminates weak crossover, as shown in (10). (10)a. daß that

Maria jeden{ Maria everyone-ACC

seinem Nachbarn tt his-DAT neighbor-DAT

vorgestellt hat (G) introduced has 'that Maria introduced everyone to his neighbor.' b. Kim pancang-i nwukwuna-luli pro Kim district chair-NOM everyone-ACC pro-GEN iwus-eykey i, sokayhayssta (K) neighbor-DAT introduced 'The district chair Kim introduced everyone to his neighbor.' Like raising, scrambling not only creates, but also destroys binding relations. Consider (11), where it is the indirect object that is the binder and the direct object that contains the bound pronoun. (11)a. daß Maria jedem seinen that Maria everyone-DAT his-ACC

Nachbarn neighbor-ACC

vorgestellt hat (G) introduced has 'that Maria introduced everyone to his neighbor.' b. Kim pancang-i nwukwu-eykey-na pro Kim district chair-NOM everyone-DAT-UQ pro-GEN iwus-ul sokayhayssta (K) neighbor-ACC introduced 'The district chair Kim introduced everyone to his neighbor.' Scrambling the direct object across the indirect object, as in (12), gives rise to a weak crossover effect that is not present on the basis of the canonical word order. Thus, the effect of scrambling in (12) is parallel to that of raising in (6).10

Towards resolving Webelhuth's paradox: evidence from German and Korean

(12)a.

*daß Maria [seinen that Maria his-ACC

Nachbarn7, neighbor-ACC

263

jedem everyone-DAT

tt

vorgestellt hat (G) introduced has Intended meaning: 'that Maria introduced everyone to his neighbor.' b.

*Kim pancang-i Kim district chair-NOM

[pro pro-GEN

iwus-ulJi neighbor-ACC

nwukwu-eykey-na r, sokayhayssta (K) everyone-DAT-UQ introduced Intended meaning: 'The district chair everyone to his neighbor.'

2.2.

Kim introduced

Reconstruction

In contrast to raising, topicalization and w/i-movement in English exhibit Vbl-properties. Rather than creating or destroying binding possibilities, they trigger reconstruction of the moved constituent to its underlying position. This is shown by the parallel unacceptability of (13) and (1) vs. the parallel acceptability of (14) and (2). (13)a. b.

*Everyoneit his mother loves r,. *Whoi does his mother love tt?

(14)

[His mother],, everyone loves tt

Scrambling an oblique constituent containing a bound pronoun across a binder that is the subject in German and Korean exhibits the same Vbl-property. Consider (15) and (16), which reflect canonical word order and scrambling, respectively.

264 Young-Suk Lee — Beatrice Santorini

(15)a. daß that

jeder everyone-NOM

seine his-ACC

Kinder children-ACC

liebhat (G) dear-has 'that everyone loves his children.' b. nwukwuna-ka pro uymwu-lul ihaynghayssta (K) everyone-NOM pro-GEN duty-ACC carried-out 'Everyone carried out his duty.' (16)a. daß [seine Kinderjeder f, that his-ACC children-ACC everyone-NOM liebhat (G) dear-has 'that everyone loves his children.' b. [pro uymwu-lul]t nwukwuna-ka tt pro-GEN duty-ACC everyone-NOM ihaynghayssta (K) carried-out 'Everyone carried out his duty.' In contrast to (12), scrambling in (16) does not result in a weak crossover effect (Webelhuth 1984: 211). Rather, the expression containing the bound pronoun undergoes reconstruction to its underlying position in (15).

2.3.

Parasitic gaps

Webelhuth (1989: 41Of.) makes the important observation that scrambled constituents in German can not only act as antecedents for bound pronouns or anaphors, but also simultaneously license parasitic gaps. We illustrate this for the bound pronoun case in (17).11

Towards resolving Webelhuth 's paradox: evidence from German and Korean 265

(17)

daß Maria jeden, ohne e anzuschauen that Maria everyone-ACC without to-look-at seinem Nachbarn ί, vorgestellt hat his-DAT neighbor introduced has 'that Maria introduced everyone to his neighbor without looking at him.'

In a theoretical framework in which An-properties and Vblproperties are held to be incompatible, the acceptability of (17) is paradoxical, and the contrast between the German (17) and the apparently analogous English (18) is unexpected. (18)

2.4.

*[Which article Ji did its author file f,· without reading e?

Summary

Let us recapitulate the properties of scrambling presented in this section. Scrambling within the domain of the subject consistently exhibits An-properties: it creates and destroys binding possibilities in weak crossover environments but does not license reconstruction. Scrambling across a subject in weak crossover environments exhibits either An-properties or Vbl-properties: constituents that have scrambled across a subject can either act as binders or undergo reconstruction. Finally, scrambling in parasitic gap constructions exhibits both An-properties and Vbl-properties: scrambled constituents can simultaneously acts as binders and license parasitic gaps.

3. 3.1.

Against a non-unitary analysis of scrambling Scrambling as a non-unitary phenomenon

Webelhuth (1989: 412f.) proposes to resolve the paradox he raises by analyzing the landing site of scrambling (which for him is adjunction to VP or IP) as a type of position distinct from an argument position or an operator position. According to him,

266 Young-Suk Lee — Beatrice Santorini

constituents in such a third position "should not be restricted in their binding potential, since they are not subject to the restrictions on argument or operator positions" (Webelhuth 1989: 413). What he suggests, then, is that the derived position of scrambling is a "mixed" position that is able to head chains with both An-properties and Vbl-properties (cf. Chomsky 1989b; Deprez 1989; Mahajan 1990). Rejecting mixed positions as conceptually inelegant,12 Mahajan (1990) presents an analysis based mainly on Hindi facts analogous to those presented in Section 2, but meant to extend to German and other scrambling languages as well, according to which scrambling is a non-unitary phenomenon. Assuming a highly articulated phrase structure along the lines of Pollock (1989) and Chomsky (1989a), with various intermediate functional heads between INFL and V, he argues that scrambling can either be substitution into the specifier position of these functional heads or adjunction to their maximal projection. Substitution is Α-movement and exhibits An-properties, whereas adjunction is Α-bar movement and exhibits Vbl-properties. Further, Mahajan (1990: 23) assumes the Weak Crossover Filter, given in (19).13 (19)

Weak Crossover Filter: To be construed as a bound variable, a pronoun must be ccommanded by a binder and its variable (if there is one) at S-structure.

3.2.

Parasitic gaps

Mahajan (1990: 56ff) argues that (17) does not provide conclusive evidence for the existence of a mixed position since it has an alternative derivation not considered by Webelhuth, who assumes that the scrambled constituent jeden 'everyone' moves directly to its derived position. Specifically, Mahajan argues that given a highly articulated phrase structure, jeden is free to first scramble to an intermediate specifier position (= Α-position) immediately preceding the indirect object before moving on to its superficial (= A-bar) position, where it is adjoined to the maximal projection dominating

Towards resolving Webelhuth's paradox! evidence from German and Korean

267

the adjunct clause containing the parasitic gap. The derivation in question is indicated schematically in (20). (20)

daß Maria jeden, ohne et anzuschauen t\ seinem, Nachbarn £,· vorgestellt hat

Given this derivation, Mahajan concludes that Webelhuth's paradox is only apparent, since the bound pronoun and the parasitic gap are each bound from distinct positions: the bound pronoun by the intermediate trace t' in the Α-position and the parasitic gap by jeden in the Α-bar position. Mahajan's analysis leads one to expect a marked contrast between (17) and the word order variant of (17) in which the order of the adjunct clause and the expression containing the bound pronoun is reversed. However, the relevant German word order variant, while not completely acceptable, is on a par with otherwise parallel constructions in which neither the licensing of parasitic gaps nor the availability of a bound pronoun interpretation is at issue, as shown in (21). 14,

15

(21 )a. 7daß Maria jeden that Maria everyone-ACC

seinem his-DAT

Nachbarn neighbor

ohne e anzuschauen vorgestellt hat without to-look-at introduced has 'that Maria introduced everyone to his neighbor without looking at him.' b. Idaß Maria jeden that Maria everyone-ACC

seinem his-DAT

Nachbarn neighbor

ohne zu zögern vorgestellt hat without to hesitate introduced has 'that Maria introduced everyone to his neighbor without hesitating.'

268 Young-Suk Lee — Beatrice Santorini

c.

Idaß Maria jeden that Maria everyone-ACC

ihrer Mutter her-DAT mother

ohne without

e

anzuschauen vorgestellt hat to-look-at introduced has 'that Maria introduced everyone to her mother without looking at him'

The relative acceptability of (21a) is surprising given Mahajan's assumptions. This is because in order to bind the bound pronoun under his assumptions, the scrambled quantified expression jeden must either itself occupy an Α-position or have moved through an Aposition immediately preceding the bound pronoun seinem. But then the scrambled phrase (or its trace) in the Α-position illicitly ccommands the parasitic gap, and the sentence should be as unacceptable as the structurally parallel English examples in (22). (22)a. b.

*[The article], was filed i, without reading e. *[Which article7, t\ was filed i, without reading e?

It might be objected that an alternative derivation exists for (21a), according to which the quantified expression jeden and the NP containing the bound pronoun seinem both scramble directly to Abar positions preceding the adjunct clause that contains the parasitic gap, yielding a representation as in (23).16 (23)

daß Maria jedent [seinemt Nachbarn]) ohne e anzuschauen t\ tj tj vorgestellt hat

In (23), the parasitic gap would be licensed by Α-bar movement at S-structure, and since both scrambled constituents occupy A-bar positions, they could reconstruct to the positions /', and tjy satisfying the Weak Crossover Filter. However, given the derivation in (23), sentences in which the constituent containing the bound pronoun scrambles by itself should be acceptable as well - an expectation that is not borne out, as shown by the unacceptability of (24).17

Towards resolving Webelhulh 's paradox: evidence from German and Korean 269

(24)

*daß Maria [seinen Nachbarn/, ohne e anzuschauen that Maria his-ACC neighbor without to-look-at jedem ί, vorgestellt hat everyone-DAT introduced has Intended meaning: 'that Maria introduced everyone to his neighbor without looking at him (the neighbor)'

In conclusion, then, the relative acceptability of (21a) is problematic for Mahajan's approach regardless of which of the two possible analyses of it that we adopt. If the quantified expression licensing the parasitic gap scrambles to (or through) an A-position from where it binds the bound pronoun, then the derivation violates standard assumptions concerning the licensing of parasitic gaps. On the other hand, if the quantified expression could scramble directly to an Α-bar position, then we would expect scrambling to license reconstruction more generally than it does.

3.3.

Limited availability of reconstruction

The difficulty for Mahajan's analysis just raised in connection with (24) is not limited to parasitic gap constructions. For while the assumption that scrambling can exhibit Vbl-properties accounts for the acceptability of instances of reconstruction like (16) straightforwardly, it simultaneously begs the question why scrambling is ever able to destroy binding relations, as it does in (12). Noting this shortcoming of his analysis himself, (Mahajan 1990: 35, fn. 18) stipulates that (leftward) adjunction to projections lower than IP is ruled out.18 But even given this stipulation, his analysis is empirically inadequate, as we show below on the basis of facts from Korean. Restricting adjunction to IP, as Mahajan proposes, leads one to expect that if a scrambled phrase exhibits the properties of A-bar movement, then any scrambled phrase to its left should exhibit them as well.19 This is because INFL is the highest functional head; there can thus be no specifier position (= Α-position) to the left of a position adjoined to IP. This expectation is not borne out, however.

270 Young-Suk Lee — Beatrice Santorini

To see this, let us begin by considering the Korean example in (25), which reflects canonical word order. (25)

*[e} prOi sangsangha-nun] salam-mataj pro-ACC imagine-REL everyone-UQ caki-eykejj mwues-ul, pwuletuli-ni? self-DAT what-ACC bring in-Q Intended meaning: 'What does everyone who imagines it bring upon himself?'

The verb in (25), pwuletui-ta 'bring in', takes three arguments: a subject, a direct object and an indirect object. The subject is the quantified expression salam-mata 'everyone', which is modified by a relative clause containing the empty bound pronoun pro; the direct object is the w/i-phrase mwues 'what'; and the indirect object is the bound pronoun caki. Since mwues fails to c-command pro, (25) is an instance of weak crossover. Now consider the word order variant of (25) where the ννΛ-phrase and caki have scrambled to the beginning of the clause in that order, as shown in (26). (26)

mwues-ult what-ACC

caki-eykeyj [e} self-DAT

pro{ sangsangha-nun] pro-ACC imagine-REL

salam-mataj (t'J tj f, pwuletuli-ni? everyone-UQ bring in-Q 'What does everyone who imagines it bring upon himself?' According to Mahajan's analysis, (26) should be unacceptable. For if caki occupies an Α-position, then it is not c-commanded by its antecedent salam-mata and does not satisfy the Weak Crossover Filter. On the other hand, if caki occupies an Α-bar position, then the w/j-phrase must be in an Α-bar position as well. But then pro in the relative clause cannot be bound since the ννΛ-phrase is not in an appropriate position: at S-structure, it does not occupy an A-position, whereas after reconstruction, it fails to c-command pro, regardless of whether we take it to undergo reconstruction to the intermediate position indicated by (t') or to its underlying position. Contrary to

Towards resolving Webelhuth's paradox: evidence from German and Korean

271

what Mahajan's analysis leads us to expect, however, (26) is acceptable. From this we conclude that his stipulation concerning adjunction to IP, apart from being conceptually unattractive, is empirically inadequate and that his analysis leaves the facts in Section 2.1 unexplained.

4.

Against binding by I-subjects

Frey (1990: ch. 6), following Haider (1989), proposes an analysis of scrambling in German according to which it is a unitary syntactic phenomenon - formally, adjunction to Vmax - which does not license reconstruction; rather, all bound pronouns must be c-commanded by their antecedents at S-structure.20 The main idea of the analysis is that the exceptional character of scrambling across the subject can be derived from the distinguished status of subjects with regard to subject-verb agreement.21 According to Frey, bound pronouns contained in expressions that have scrambled across a subject are bound not by the lexical subject after reconstruction, but by the inflectional features associated with subject-verb agreement, which ccommand constituents of Vm™ at S-structure.

4.1. 4.1.1.

Background

assumptions

Phrase structure

Following Haider (1986), Frey (1990: 19ff.) assumes that COMP and INFL in verb-second languages like German are not distinct heads but that they merge into a complex head C/I, which is the site of the inflectional features [+/- AGR, +/- finite]. In root clauses, the highest verb in the clause moves to C/I so that the inflectional features may be lexically realized. In formally subordinate clauses, on the other hand, Frey argues that the inflectional features cannot remain on C/I, but must instead be realized on the head of V-max, the complement of C/I. With regard to the structure of Vmax, Frey (1990: 37) assumes that verbs in German project a binary-branching structure and that each projection of the verb is maximal. In contrast to Mahajan, Frey (1990: 204f.) takes scrambling to be a unitary process -namely,

272 Young-Suk Lee — Beatrice Santorini

adjunction to V m ". Thus, the schematic phrase structure of a German clause, whether it involves scrambling or not, is as shown in (27). (27)

4.1.2.

The role of inflection

Frey (1990: ch.4) proposes to treat the complex of inflectional features associated with subject-verb agreement as a full-fledged syntactic argument. In particular, he assumes that a verb's external theta-role is realized on the node bearing the inflectional features i.e. C/I in German. However, the nominative case index associated with the external theta-role, while assigned to C/I, must be realized in German on a lexical NP, which Frey takes to be an argument of C/I. He refers to the formal object consisting of the node bearing the inflectional features and the nominative NP as the "I-subject". Finally, Frey assumes that nominative case can be realized within Vmax in German (Frey 1990: 34; cf. den Besten 1985).

4.1.3.

Locality and reconstruction

In contrast to Mahajan's purely structural definition of Α-positions as specifier positions and Α-bar positions as adjoined positions, Frey (1990: 10), like Chomsky (1981), defines Α-positions in thematic terms - as positions to which a theta-role can be assigned. He assumes that verbs project their argument structure - and that thetaroles are assigned - in a canonical order (which is specific to each particular predicate); hence, the adjunction positions occupied by

Towards resolving Webelhuth's paradox: evidence from German and Korean

273

scrambled constituents are Α-bar positions in his view (Frey 1990:204f.)· Despite its status as Α-bar movement, however, scrambling does not license reconstruction; rather, Frey proposes that reconstruction is licensed by movement out of a local domain.22 He defines the concept of "local domain" as in (28) and adopts the definition of "complete functional complex" in (29) (Frey 1990: 105f.; cf. Chomsky 1985: 169).23 (28)

Definition of local domain: The local domain for an expression A is the minimal complete functional complex containing a licenser for A.

(29)

Complete functional complex (CFC): A CFC for Ε is the minimal maximal projection in which all θ-roles of E's Θ-marker are realized.

4.2.

Deriving the (apparent) reconstruction effect

Given the assumptions just outlined, the local domain of a scrambled constituent is the minimal CP/IP containing it. This is because while the internal theta-roles of the verb that θ-marks the scrambled constituent are assigned within that verb's maximal projection, the external theta-role is assigned to C/I. Within this local domain, Frey argues, pronouns must be c-commanded by a quantified expression in order to be interpreted as bound by it - a condition that is straightforwardly consistent with the facts presented in Section 2.1. According to Frey, constituents in Α-bar positions can act as binders, as long as they are constituents of the same local domain as their bindee (Frey 1990: 114f.). Frey attributes the acceptability of a German sentence like (16a) to the fact that while the bound pronoun is not c-commanded by the nominative NP, it is c-commanded by a link of the I-subject - namely, C/I. The condition on quantifier binding that Frey proposes is given in (30) (Frey 1990: 112f.); he defines "c-command" as in (31).24

274

(30)

Young-SukLee — Beatrice Santorini

Frey's condition on quantifier binding: Let L, be the local domain of A (according to [(28)], and let Q be the chain headed by A, A a quantified expression, C, = < a„ a! >. Let i, i . Let Ρ be a pronoun, Ρ = Β or Β contains P. Let i, i and I c-commands fy. (31)

Definition of c-command: A c-commands Β iff (a) A does not dominate B, and (b) If a node C immediately dominates A, then (i) either C dominates B, or (ii) C is a projection of A, and Β is dominated by a projection of A.

4.3.

Counterarguments

Although we will adopt a number of Frey's assumptions in the analysis that we present below, his analysis as it stands has both conceptual and empirical drawbacks that lead us to reject it. First and foremost, the status of C/I as a theta-role recipient and binder is theoretically anomalous, since heads otherwise assign rather than receive theta-roles and do not otherwise enter into binding relations. Second, since Frey's analysis depends on the presence of inflectional features, it is not clear how to extend it to languages like Korean (or Japanese), which exhibit the same apparent reconstruction effects as

Towards resolving Webelhuth's paradox: evidence front German and Korean 275

German, as we saw in Section 2.2, but which have been argued to lack INFL and its associated features altogether or to exhibit them in a "deficient" form (Lapointe 1990; Lee 1990; cf. Fukui 1986 for Japanese). Finally, Frey's analysis is empirically inadequate on two counts, which we discuss in detail below: his analysis - in particular, his concept of locality - fails to carry over from German to the closely related Dutch, and his c-command requirement on quantifier binding is too strict.25 Frey's assumptions about the locus of the inflectional features in C/I in German, taken together with his definition of local domain as CP/IP, have an interesting consequence: according to his analysis, the binding possibilities of topicalized expressions should parallel those of scrambled ones. That is, in the general case, the binding possibilities in instances of topicalization should be determined on the basis of (strict) c-command. In the special case where an oblique constituent containing a pronoun is topicalized across a quantified subject, the pronoun should be able to be interpreted as being bound by the subject, since C/I, being a head, c-commands any constituent contained in its maximal projection according to the definition in (31). As the examples below show, this expectation is borne out in German (cf. Frey 1990: 95f.). (32)a.

*Seine Kinder his-NOM children-NOM

haben jeden have everyone-ACC

lieb (G) dear Intended meaning: 'Everyone is loved by his children.' b. Jeden everyone-ACC

haben seine Kinder have his-NOM children-NOM

lieb (G) dear 'Everyone is loved by his children.'

276 Young-Suk Lee — Beatrice Santorini

(33)a.

*Seinem his-DAT

Nachbarn neighbor-DAT

vorgestellt (G) introduced Intended meaning: neighbor.'

hat Maria jeden has Maria everyone-ACC

'Maria introduced everyone to his

b. Jeden hat Maria seinem everyone-ACC has Maria his-DAT

Nachbarn neighbor-DAT

vorgestellt (G) introduced. 'Maria introduced everyone to his neighbor.' (34)a. Jedem hat Maria seinen everyone-DAT has Maria his-ACC

Nachbarn neighbor-ACC

vorgestellt (G) introduced. 'Maria introduced everyone to his neighbor.' b. Seinen Nachbarn hat Maria his-ACC neighbor-ACC has Maria

jedem everyone-DAT

vorgestellt (G) introduced. Intended meaning: 'Maria introduced everyone to his neighbor.' (35)a. Jeder hat seine everyone-NOM has his-ACC 'Everyone loves his children.'

Kinder lieb (G) children-ACC dear

b. Seine Kinder hat jeder lieb (G) his-ACC children-ACC has everyone-NOM dear 'that everyone loves his children'

Towards resolving Webelhuth 's paradox: evidence from German and Korean 277

Given Frey's definition of local domain as CP/IP, the effect of topicalization on weak crossover in German is independent of the availability of scrambling in that language. Since Dutch is a verbsecond language like German, Frey's analysis therefore leads one to expect the effect of topicalization on weak crossover to be parallel in both languages, despite the virtual absence of scrambling in Dutch.26 This expectation is not borne out, however. In particular, topicalized quantified objects cannot act as binders in Dutch, in contrast to German (cf. (32b)). This is shown in (36). (36)

Iedereen hebben zijn hinderen lief everyone-ACC have his-NOM children-NOM dear Intended meaning: 'Everyone is loved by his children.'

Not only is the unacceptability of (36) unexpected under Frey's analysis, but there is also no way to relate it to the unavailability of the scrambled word order in (37a) or the unacceptability of the underlying word order in (37b) (vanden Wyngaerd 1989: 260; cf. also Haider 1989: 198f.). (37)a.

*dat iedereen zijn hinderen that everyone-ACC his-NOM children-NOM liefliebben dear-have Intended meaning: 'that everyone is loved by his children.'

b.

*dat that

zijn hinderen iedereen his-NOM children-NOM everyone-ACC

liefliebben dear-have Intended meaning: 'that everyone is loved by his children.' A final shortcoming of Frey's analysis concerns his assumption common in the literature - that bound pronouns must be ccommanded by their antecedents. As the acceptability of the examples in (38) shows, however, c-command is too strict a requirement.

278 Young-Suk Lee — Beatrice Santorini

(38)a.

b.

Wir haben mit jedem sein Lieblingsspiel gespielt we have with everyone his favorite-game played Intended meaning: 'For each person x, we played x's favorite game with x.' Wir haben neben jedem sein Lieblingsspiel we have beside everyone his favorite-game aufgebaut set-up Intended meaning: 'For each person x, we set up x's favorite game beside x.'

On the other hand, the unacceptability of (39) shows that precedence alone is too lax a requirement, at least for German (cf. Frey 1990: 94; vanden Wyngaerd 1989 for Dutch); see Lee (1991) for a detailed discussion of the Korean facts. (39)a.

*Wir haben mit der Mutter we have with the mother

von jedem sein of everyone his

Lieblingsspiel gespielt favorite-game played Intended meaning: 'For each person x, we played x's favorite game with x's mother.' b.

*Wir haben neben der Mutter we have beside the mother

von jedem sein of everyone his

Lieblingsspiel aufgebaut favorite-game set-up Intended meaning: 'For each person x, we set up x's favorite game beside x's mother.' The correct requirement appears to be that the bound pronoun must be preceded and commanded by its antecedent (cf. Langacker 1969; Lasnik 1976: 15; Haider 1989: 199); we give a definition of "command" in Section 5.3.1.

Towards resolving Webelhuth's paradox: evidence from German and Korean 279

5. A resolution of the paradox In this section, we present our own attempt to resolve the paradox raised by the facts in Section 2. We assume the VP-Internal Subject Hypothesis, and we take scrambling to be a formally uniform process of adjunction to VP (and perhaps also to IP). Our analysis is based on the assumption, which we adopt from Frey, that the creation and destruction of binding possibilities on the one hand and reconstruction on the other are not associated with movement to a specific type of position with inherent properties. On the other hand, we assume that constituents that license parasitic gaps do occupy positions with inherent properties - specifically, ηοη-θ-marked, noncase-marked positions.

5.1.

Domains of locality

5.1.1. Argument domains On the basis of the facts presented in Section 2.2, it appears that when an oblique NP precedes a nominative NP, the oblique NP optionally reconstructs. However, the oblique arguments of certain predicates in Korean (which, following Burzio (1981), we will refer to in what follows as "ergative" predicates) do not allow reconstruction, as shown in (40). (40)

*pro emma-eykey nwukwu-na-ka pro-GEN mother-DAT everyone-UQ-NOM choiko-ta the best-DEC Intended meaning: 'Everyone is the greatest to his mother.'

The dative-nominative word order is unmarked for such predicates, and if we assume that it directly reflects their syntactic argument structure, then the proper generalization appears to be that reconstruction is licensed by scrambling across the highest syntactic argument. This correctly accounts for the facts in Section 2.1 and 2.2 as well as those in (40), (41) and (42).

280 Young-Suk Lee — Beatrice Santorini

(41)

nwukwu-na-kdj pro emma-eykey everyone-UQ-NOM pro-GEN mother-DAT choiko-ta the best-DEC 'Everyone is the greatest to his mother.'

(42)a. enu pwumo-eykey-na pro casik-i every parent-DAT-UQ pro-GEN child-NOM choiko-ta the best-DEC 'Every child is the best to his parent.' b. [pro casik-i]ι enu pwumo-eykey-na tt pro-GEN child-NOM every parent-DAT-UQ choiko-ta the best-DEC 'Every child is the best to his parent.' In contrast to Korean, however, the oblique arguments of apparently ergative predicates in German reconstruct, as shown in (43) (cf. Lenerz (1977) and den Besten (1985) for discussion of these predicates and Frey (1990: 114) for the weak crossover facts). (43)

daß seiner Mutter jeder gefällt that his-DAT mother-DAT everyone-NOM pleases 'that everyone's mother likes him.'

Two alternative approaches to the difficulty raised by the contrast between (40) and (43) come to mind. According to the first, which is reminiscent of Frey's approach to the reconstruction facts of Section 2.2, the nominative NP occupies the direct object position at Dstructure and the binding in (43) is only apparently due to reconstruction (Heycock 1991: ch. 3). Rather, Spec(IP) is occupied by an empty expletive,27 which enters into a CHAIN with the nominative NP, as indicated by the co-superscripting in (44), and it

Towards resolving Webelhuth's paradox: evidence from German and Korean

281

is the expletive subject rather than the nominative NP that binds the pronoun. (44)

daß [ip έ [vp seiner Mutter jeder* gefällt]]

While this analysis is consistent with the absence of evidence for empty expletives in Korean (and Japanese - cf. Heycock 1991: ch. 3), it raises the question why expletives cannot license a bound pronoun interpretation in English examples like (45), which are parallel to (44) in the relevant respects (Mahajan 1990: 23, fn. 10). (45)

*There' seems to his mother to be [a manf in the garden.

The second alternative assumes a mismatch in German between the pragmatically unmarked dative - nominative word order associated with the predicates in question and their syntactic argument structure. That is, despite the parallel between German and Korean with regard to S-structure word order (and presumably lexical argument structure as well), the highest argument in German in the syntax is the nominative argument, and the dative nominative word order in German, in contrast to Korean, is the result of scrambling. Under this analysis, the nominative NP in German occupies the underlying subject position at D-structure and German has no true ergative predicates (i.e., ones in which the hierarchy of lexical argument structure is directly reflected in the syntax).29 An apparent difficulty for this second approach comes from the fact that the subjects of the predicates in question behave like direct objects for the so-called was für split (den Besten 1985), as shown by the parallelism between (46) and (47) and the contrast with (48). (46)

Was, hast du deinem Mann für what have you your-DAT husband-DAT for einen Roman geschenkt? a-ACC novel-ACC given 'What kind of a novel did you give your husband?'

282 Young-Suk Lee — Beatrice Santorini

(47)a.

WaSi ist deinem Mann ί, what is your-DAT husband-DAT

Jur ein for a-NOM

Roman geschenkt worden? novel-NOM given been 'What kind of a novel was your husband given?' b. Was( ist deinem Mann ί, what is your-DAT husband-DAT

flir ein for a-NOM

Unfall passiert? accident-NOM happened 'What kind of an accident was your husband involved in?' (48)a.

*Was( ist f, flir ein Roman what is for a-NOM novel-NOM

deinem your-DAT

Mann geschenkt worden? husband-DAT given been Intended meaning: 'What kind of a novel was your husband given?' b. *WaSj ist t{ Jur ein Unfall deinem what is for a-NOM accident-NOM your-DAT Mann passiert? husband-DAT happened Intended meaning: 'What kind of an accident was your husband involved in?' As Heycock (1991: ch. 3) points out, however, the was für split test does not reliably indicate whether a subject originates in direct object position, since it is possible with transitive verbs so long as the subject is adjacent to the verb.30 An example is given in (49).

Towards resolving Webelhulh's paradox: evidence from German and Korean

(49)

Wast haben deinen Mann t{ what have your-DAT husband-ACC

283

für Ameisen for ants-NOM

gebissen? bitten 'what kind of ants bit your husband* We conclude from the above discussion that while superficial word order is sensitive to lexical argument (= thematic) structure in both German and Korean, the two languages differ with regard to the mapping of thematic structure onto syntactic argument structure. In Korean, this mapping is sensitive to a thematic hierarchy in which experiencers outrank themes, whereas in German, the argument that agrees with the verb becomes the highest argument in the syntax regardless of its rank in the thematic hierarchy. We are now in a position to adopt Frey's definition of "Aposition" as a syntactic position to which a θ-role can be assigned, as well as his definition of a local domain, repeated here in simplified form as (50); we will refer to such a local domain as an argument domain. (50)

Definition of argument domain: The argument domain for an expression A is the minimal maximal projection in which all θ-roles associated with A's θ-marker are syntactically realized.

We do not share Frey's assumption that the external θ-role is assigned to INFL. Hence, for us, the argument domain of an argument of a verb is the minimal VP dominating that verb's highest syntactic argument. We assume that movement within an argument domain does not license reconstruction, whereas movement out of an argument domain does so optionally.

5.1.2.

Binding domains

The assumption that movement out of an argument domain optionally licenses reconstruction accounts for the availability of a

284 Young-Suk Lee — Beatrice Santorini

bound pronoun interpretation in cases where a constituent containing a pronoun precedes a quantified expression that is a binder, but is silent about the fact that quantified expressions in Α-bar positions can themselves act as binders. We follow Frey (1990) in making the unorthodox assumption that binders (like bindees) need not occupy Α-positions. In contrast to him, however, we do not take the domain within which an expression can act as a binder (its "binding domain") to be coextensive with its argument domain. Whereas argument domains are defined in terms of the syntactic realization of a predicate's argument structure, we define binding domains in terms of the agreement feature AGR, as in (51); we assume the firstbranching node definition of c-command. In languages without AGR, such as Korean, we simply assume that the binding domain is the root clause by default. (51)

Definition of binding domain: Let C be the chain headed by A, A a quantified expression, C = < a„ a! >. Then the binding domain of A is (i) the minimal maximal projection c-commanded by (a head containing) AGR that dominates a link a; in C, or (ii) the root clause.

In contrast to Frey (1990), we assume that INFL and COMP give rise to separate projections in German, äs they do in English. Following Platzack — Holmberg (1990), we take AGR to be located in COMP in German. Then the binding domain of a quantified expression is the minimal finite IP containing it in German. This is confirmed by the contrast between (52) and (53), where constituents of an extraposed infinitival clause have scrambled into a finite matrix clause (Kvam 1983; den Besten — Rutten 1989).31 (52)a. Heute today

habe ich jedefy seinem,· have I everyone-ACC his-DAT

Nachbarn] versprochen [t\ tj vorzustellen] neighbor-DAT promised to-introduce 'Today, I promised to introduce everyone to his neighbor.'

Towards resolving Webelhuth's paradox: evidence from German and Korean 285

b. ΊHeute habe ich jeden, today have I everyone-ACC

versprochen, promised

[t'i seinemj Nachbarn vorzustellen] his-DAT neighbor-DAT to-introduce 'Today, I promised to introduce everyone to his neighbor.' (53)

*Heute habe ich [seinem Nachbarn/, today have I his-DAT neighbor-DAT

versprochen, promised

[t{ jeden vorzustellen J everyone-ACC to-introduce Intended meaning: 'Today, I promised to introduce everyone to his neighbor.' Following Platzack — Holmberg (1990), we assume that AGR is in INFL in English. The binding domain of a quantified expression in English is therefore the VP of the minimal finite clause dominating it. In instances of raising as in (4) and (6), we assume that the raised subject passes through a position within the ccommand domain of the matrix AGR (either the Spec(VP) of the matrix clause or a position adjoined to the matrix VP), as shown schematically in (54). (54)a. Everyonei seems [vp t\ to his mother [ip r, to be charming]] b. [His mother], seems [vp t\ to everyone [ip f, to be charming]]

5.2.

"Minimality" of reconstruction

In addition to taking reconstruction to be optional, we follow Frey (1990: 100) in taking it to be "minimal" in the sense that while it restores expressions to some more local domain, it does not otherwise cancel properties of a structure prior to reconstruction. The position of a bindee after reconstruction is therefore not necessarily its underlying position but rather the "highest possible" position in which a binder is available.32 Consider a sentence like (55), which reflects canonical word order and where the subject or the indirect

286 Young-Suk Lee — Beatrice Santorini

object can equally easily bind the pronoun contained in the direct object. (55)

Gestern hat [jeder Professor]t [jedem yesterday has every-NOM professor every-DAT Studenten]) seine^ Dissertation gegeben student his-ACC dissertation given 'Yesterday, [every professor] j gave [every student], his^ dissertation.'

Scrambling the direct object across the subject eliminates the indirect object as a binder, as shown in (56). (56)

Gestern hat seine{η Dissertation [jeder yesterday has his-ACC dissertation every-NOM ProfessorJi [jedem Studenten7, gegeben professor every-DAT student given 'Yesterday, [every professor]; gave [every student], hiSj.j dissertation.'

Analogous facts obtain in Korean. Our account of this, following Frey, is as follows. Scrambling the direct object containing the bound pronoun across the highest argument - the subject - licenses optional reconstruction. If reconstruction does not apply, (56) is ruled out since there is no antecedent preceding and commanding the pronoun at S-structure (we give a full statement of our condition on quantifier binding directly). If the direct object does undergo reconstruction, it cannot reconstruct to its underlying position since German allows a word order variant of (56) in which the constituent to be reconstructed precedes the indirect object, just as in (56), but in which it is realized within its argument domain. The scrambled word order in question is shown in (57).

Towards resolving Webelhuth's paradox: evidence from German and Korean 287

(57)

Gestern hat [jeder Professor/, seine^ yesterday has every-NOM professor his-ACC Dissertation [jedem Studenten]· gegeben dissertation every-DAT student given 'Yesterday, [every professor]j gave [every student]; his^ dissertation.'

The direct object therefore reconstructs to the scrambled position in (58), where the indirect object is ruled out as an antecedent because it, unlike the subject, does not precede and command the bound pronoun.

5.3. 5.3.1.

Analysis A condition on quantifier binding

We assume the condition on bound pronoun interpretation in (58): (58)

Condition on bound pronoun interpretation: Let Lj be the binding domain for A, A a quantified expression, and let C, be the chain headed by A, C, = < a„, ..., a, >. Let i, i . Let Ρ be a pronoun, Ρ = Β or Β contains P. Let i, i = j >= i. Then Ρ can be coindexed with A if there is a bj such that ^ precedes and commands bj.

288 Young-Suk Lee — Beatrice Santorini

(59)

Definition of command: A commands Β iff (a) A does not dominate B, and (b) if C, C a major node, dominates A, then C dominates B.

(60)

Definition of major node: A major node is DP or the maximal projection of a head bearing agreement features (CP in German, IP in English, not defined in Korean).

Note that the formulation of (58) makes no reference to a distinct level of representation resulting from reconstruction. Rather, it allows one to determine the availability of a bound pronoun interpretation with reference to S-structure representations. For expository convenience, however, we will continue to speak as if reconstruction involved an optional lowering operation at a level of representation distinct from S-structure. From the discussion in Section 5.1, it should be clear that the facts discussed in Section 2.1 and 2.2 are consistent with the condition in (58). (58) also covers the Korean example in (26), repeated here as (61), which proved problematic for the assumption that reconstruction from Α-bar positions is obligatory. (61)

mwues-uli caki-eykeyj [ej pro( sangsangha-nunJ what-ACC self-DAT pro-ACC imagine-REL salam-matüj (t\) i, pwuletuli-ni? everyone-UQ bring in-Q 'What does everyone who imagines it bring upon himself?'

The binding domain for both the vWi-phrase and the quantified subject is the entire root clause. We assume that of the two constituents that have undergone scrambling across the subject, the ννΛ-phrase does not undergo reconstruction, but the bound pronoun caki does. Then the condition in (58) is satisfied because after reconstruction, the ννΛ-phrase precedes and commands the bound

Towards resolving Webelhulh's paradox: evidence from German and Korean 289

pronoun contained in the relative clause, and the subject precedes and commands caki.

5.3.2.

Comparative considerations

As Frey notes, topicalization and scrambling have parallel effects on weak crossover in German. In our analysis, this follows because the derived Spec(CP) position of topics is outside both their binding domain and their argument domain; hence, if they do not undergo reconstruction, they are available neither as binders nor as bindees. Given that reconstruction is to the highest possible position, the binding properties of a topicalized expression will correspond to its binding properties in a scrambling counterpart if one is available, and otherwise to the expression's properties in the underlying order. Since scrambling is more restricted in Dutch than in German but the two languages are otherwise similar in their syntactic properties, our analysis, in contrast to Frey's, leads us to expect differences between Dutch and German with regard to the effect of topicalization on weak crossover. For us, the unacceptability of (36) is therefore both expected and related to the unacceptability of the sentences in (39). Finally, note that the condition in (58) is not restricted to scrambling languages. In English, for instance, it properly distinguishes the acceptable (5) from the unacceptable (i) in fn. 6. Moreover, in instances of heavy NP shift, it allows us to distinguish between (62a) (cf. Deprez 1989: 271f.) and (62b). (62)a.

*Mary showed his mother without showing his father, Ieverybody who finished on time/,. b. Mary introduced t{ to his neighbor everybody who she knew at the party ]t

If we assume that heavy NP shift moves an expression out of its argument domain, then the quantified expression can optionally reconstruct to its underlying position in both sentences in (62). However, regardless of whether the expression reconstructs or not, it follows rather than precedes the bound pronoun in (62a), the unacceptability of which follows. In (62b), on the other hand, the

290 Young-Suk Lee — Beatrice Santorini

underlying position of the quantified expression precedes the bound pronoun, and (58) is satisfied after reconstruction.

5.3.3.

The licensing of parasitic

gaps

We have argued above that the availability of a bound pronoun interpretation does not depend on the heads of the chains containing the binder and the bindee occupying a position with specific inherent properties. By contrast, we assume that antecedents of parasitic gaps do occupy positions with inherent properties. Specifically, we propose that in order to license a parasitic gap, a constituent must occupy a ηοη-θ-marked, non-case-marked position. Thus, parasitic gaps are licensed by their antecedents in a way that bound pronouns are not, although of course the relation between binders and bindees is structurally constrained. It is interesting to note that our proposal is similar to one advanced by Deprez (1989), though our approach to scrambling differs substantially from hers in other respects. Assuming the same type of articulated phrase structure as Chomsky (1989a) and Mahajan (1990), Deprez (1989) distinguishes head-related ([+HR]) positions from non-head-related ([-HR]) positions: [+HR] positions are substitution positions (i.e. specifiers and complements) and head chains with An-properties, whereas [-HR] positions are adjoined positions and head chains with Vbl-properties. Unlike Mahajan, Deprez does not attempt to eliminate Webelhuth's paradox, but proposes instead to resolve it by further dividing [+HR] positions into ones that are assigned case and ones that are not (Deprez 1989: 266f.). [+HR, +case] positions would then head chains with pure Anproperties, whereas [+HR, -case] positions would head chains with mixed properties. Thus, her proposal and ours agree in taking case assignment to be a determining factor in whether a position can license parasitic gaps or not.

6.

Conclusion

In this paper, we have discussed certain syntactic properties of scrambling in German and Korean, which given the assumptions in

Towards resolving Webelhuth's paradox: evidence from German and Korean

291

Chomsky (1981) are unexpected and paradoxical. After presenting the properties in question, we discussed two previous approaches to resolving the paradox they raise. The first approach, proposed by Mahajan (1990), attempts to dismiss the paradox as apparent by analyzing scrambling as a non-uniform phenomenon. We rejected this approach as stipulative and empirically inadequate. The second approach, proposed by Frey (1990) for German, analyzes certain properties of scrambling -specifically, its ability to license reconstruction - as apparent and relates them to independently motivated properties of German. While rejecting Frey's particular analysis on comparative grounds, the attempt to resolve Webelhuth's paradox that we propose is very much in the spirit of his approach; in particular, it is based on his unorthodox assumption that the binding and reconstruction properties of chains are not determined by the structural position of their heads, but instead by whether the head has undergone local movement or not. In contrast to Frey, however, we find it necessary to distinguish more than one domain of locality: a narrow one, defined in terms of syntactic argument structure, and a broader one, defined with reference to the presence of AGR. In contrast to many recent analyses, our analysis of scrambling does not rely on assuming a multitude of functional categories related to the agreement system, whose specifiers serve as landing sites for scrambled constituents. We regard this as a decided advantage, since the properties of scrambling we are concerned with are strikingly similar regardless of whether a language exhibits both subject and direct object agreement (Hindi), only subject agreement (German), or minimal or no agreement whatsoever (Korean, Japanese). Nevertheless, functional categories are important in our analysis, since our definition of the domain within which a quantified expression can act as a binder makes reference to the functional category AGR. Furthermore, the difference between Korean and German with regard to the mapping of lexical argument structure onto syntactic argument structure appears to be related to the presence of A G R in German vs. its absence in Korean. Finally, the binding domain of a quantified expression in German and Korean appears to be precisely the domain within which scrambling is possible. That is, German (under certain conditions) allows scrambling out of infinitival clauses, but never out of finite clauses,

292 Young-Suk Lee — Beatrice Santorini

whereas Korean freely allows long-distance scrambling out of all clause types. In future research, we plan to continue our comparative investigation of German and Korean, since the striking similarities and equally pronounced differences between these languages promise to allow us to triangulate on a conceptually satisfactory and empirically adequate analysis of the scrambling phenomenon.

Notes 1. This paper is a much revised version of a presentation that we gave at the Tilburg Workshop on Scrambling in October 1990. We have benefited greatly from presenting our work before the audience there as well as at the University of Pennsylvania and CUNY. It is a pleasure to thank Hee Don Ahn, Josef Bayer, Tilman Becker, Hans den Besten, Robert Fiengo, Steve Franks, Tilman Höhle, Jack Hoeksema, Henry Hoenigswald, Soon-hyun Hong, Riny Huybregts, Richard Kayne, Katalin Kiss, Jaklin Kornfilt, Chang-Bong Lee, Anoop Mahajan, Shigeru Miyagawa, Gereon Müller, Owen Rambow, Henk van Riemsdijk, Ramona Römisch, Claudia Schmidt, Wolfgang Sternefeld, Annie Zaenen, Raffaella Zanuttini and an anonymous reviewer for native speaker judgments, helpful discussion, and valuable comments and criticism that have improved successive revisions. A special heartfelt thank-you goes to Robert Frank, Caroline Heycock, Anthony Kroch and Sten Vikner for their moral and intellectual support. All errors and shortcomings that remain are of course our own. 2. In this paper, we will use the two terms VP and V-max to refer to the maximal projection of the verb without intending any significant theoretical distinction. 3. We and and and

focus on the interpretation of bound possessive pronouns rather than reflexive reciprocal pronouns since the binding of reflexives and reciprocals in German Korean is influenced by non-structural considerations such as the theta-role the grammatical case of the binder.

4. Distinguishing between two domains of locality is the counterpart in our analysis to the distinction between "narrowly L-related" and "broadly L-related" positions proposed by Chomsky (1989a). 5. Throughout this paper, we will use italics to represent coreference relations and coindexing to represent antecedent-trace relations. However, in sentences containing more than one relevant coreference relation, we will use coindexing to indicate coreference.

Towards resolving Webelhulh's paradox: evidence from German and Korean 293

6. That precedence alone (apparently) cannot account for the contrast between (1) and (2) is shown by the unacceptability (for most speakers) of sentences as in (i). (i)a. Everyone's wife loves his mother. b. *Whose wife loves his mother? 7. We return below to the fact that the quantified expression in (5) behaves like a bare NP despite being contained in a PP. 8. The abbreviations "G" and "K" stand for "German" and "Korean", respectively. In the glosses, we use the following abbreviations: ACC COMP DECL NOM TOP

accusative complementizer declarative nominative topic

ATT COP GEN Q UQ

attributive CAUS copula DAT genitive INST question REL universal quantifier

causative dative instrumental relativizer

9. There are at least three items which have been identified as bound pronouns in Korean: caki 'self, the overt pronound ku/kukes 'he/it', and the empty pronoun pro (cf. Ahn et al. 1990). For [+human] entities, caki is highly preferred with a subject antecedent, and ku with a dative antecedent. For [-human] entities, kukes is used regardless of the status of the antecedent. The empty pronoun pro can occur in any environment that overt pronouns can occur, though there seem to be subtle differences between overt and empty pronouns with regard to the availability of a bound pronoun interpretation. For instance, in (i), the non-c-commanding antecedent nwukwu 'who' contained in the scrambled object phrase can bind the pronoun contained in the subject phrase only if the pronoun is overt. (i)

[nwukwu-uy who-GEN

cip-ul], house-ACC

ku-uy / *pro he-GEN pro-GEN

sensayngnim-i f , pangmwunhayss-ni? teacher-NOM visited-Q Intended meaning: 'For which person x, x's house was vistied by x's teacher?" Throughout this paper, we use pro as the bound pronoun in Korean unless forced to use an overt pronoun. 10. In examples like (12), some speakers marginally allow reconstruction to the underlying order in (11). The entire discussion below is based on the judgments of those speakers who do not allow reconstruction.

294 Young-Suk Lee — Beatrice Santorini

11. Webelhuth's paradox does Dot arise in Korean, since it can be shown that what appear to be parasitic gaps in Korean are actually null pronouns. See Lee 1991 for details; for the analogous Japanese facts, cf. Saito (1990) and Deprez (1989: 148ff). 12. Mahajan does not note that Webelhuth's proposal is also empirically unsatisfactory: it does not explain why scrambling can ever destroy binding relations, as in (12). For further arguments against Webelhuth's proposal, see Deprez (1989: 269f). 13. Mahajan takes the filter in (19) to apply at S-structure rather than at LF because LF expletive replacement in English does not eliminate weak crossover effects. However, the acceptability of sentences like (16) (and its equivalents in Hindi) shows that the filter must apply at a level of representation reflecting reconstruction that is distinct from both S-structure and LF. In what follows, we will assume this modification. 14. The relevant Hindi sentence is unacceptable, according to Mahajan (1990: 55) (his example (90)); he does not give the acceptability of the Hindi equivalents of control sentences like (21b,c). 15. Deprez (1989: 267f.) raises the same objection to Mahajan's analysis as we do on the basis of German examples like (i) (cf. her example (203)), where the scrambled constituent in the matrix clause binds a parasitic gap and a bound pronoun in the adjunct clause. (i)

Ίάφ that

Maria Maria

[jeden /, everyone-ACC

[ohne without

seinem his-DAT

Partner e vorzustellen] alleingelassen hat partner to-introduce alone-let has 'that Maria left everyone alone without introducing him to his partner' The example is (i) is inconclusive, however, since the parasitic gap might itself have undergone scrambling, as shown in (ii). (ii)

EXPERIENCE!* > THEME ι DAT

I further assume that the nominative and the accusative are ordered, such that the nominative is more prominent than the accusative: (55)

NOM > ACC

The Θ-grid in (54) is now linked onto the case hierarchy in (55), supposedly by a one-to-one linking of cases and unlinked Θ-roles. The result is given in (56).

Scrambling as a D-structure phenomenon

(56)

421

AGENT > EXPERIENCER > THEME I I I NOM DAT ACC

In (56), each case identifies the Θ-grid position of the Θ-role it is attached to. This means that the order of the Θ-roles in German is present in the syntax via the cases the arguments bear. So, ccommand relations do not have to be used to satisfy (51). Consequently, the linking of Θ-roles and structural positions is free. The sentences in (46) are predictably well-formed. Double object constructions in which the indirect object is lexically linked to the accusative (not to the dative) show that even a detailed prediction of the system just sketched is borne out. An example of a verb projecting such a double object construction is lehren (to teach). The case linking for lehren is as in (57). (57)

AGENT > EXPERIENCER > THEME I I I NOM ACC ACC

If German can use morphological case to project the Θ-hierarchy, we predict that lehren does not allow inversion of the direct and the indirect object. The linking in (57) does not express the relation between the Experiencer and the Theme - they are both linked to the accusative. The only way, then, to project this sub-hierarchy is via ccommand relations: like in Dutch the indirect object should ccommand the direct object. And indeed, abstracting away from Focus scrambling, the direct object may not precede the indirect object in clauses headed by lehren: (58)a. Daß der Lehrer die Schüler That the teacher-NOM the pupils-ACC

diese this

Sprache lehrt language-ACC teaches b.

? *Daß der Lehrer diese Sprache die Schüler lehrt.

422 Ad Neeleman

The system proposed here explains the differences between Dutch and German and it shows how in a base generation account scrambling can be restricted. In the movement analyses the relevant facts are hard to explain. We conclude, then, that the contrast in word order freedom between Dutch and German is a final argument for the base generation account.

5. Conclusion In the preceding sections we compared three approaches to scrambling: the A*-movement approach, the Α-movement approach and the base generation approach. We saw that scrambling in neutral contexts does not have the properties one would expect, if it were adjunction to VP or V . Focus scrambling, on the other hand, does. Therefore we concluded that Focus scrambling is adjunction, while neutral scrambling is not. The one argument for the A'-movement analysis, namely the possibility of scrambling licensing parasitic gaps, fell through, as we observed that the premiss that parasitic gaps can only be A'-bound should be rejected on both theoretical and empirical grounds. The analysis of Mahajan (1990), in which neutral scrambling is Α-movement, does not have the shortcomings of the A'-movement analysis. However, it was rejected both on technical and conceptual grounds. This left us with base generation, which we saw in the last section, provides a solid basis for the explanation of restrictions on scrambling. Finally, let us consider the definition of Α-positions in the three approaches. In the adjunction analysis, this definition becomes problematic, given the fact that at least some scrambling positions are Α-positions. Webelhuth (1989) has to assume that there are "mixed" positions: positions that have both an A- and an A'-status. In an analysis like Mahajan's there are Α-positions outside the Θdomain of the verb, which is VP. Again this seems to be an extension of the notion. In the analysis proposed here, the neutral scrambling positions are Α-positions in the original sence: they are Θ-marked. The only Α-position that is not an obligatory Θ-position is the subject position. However, one could say that this position is selected anyway by the inflectional features on the verb: these

Scrambling as a D-structure phenomenon

423

features require specification. So, one could define Α-positions as those positions that are selected by a lexical head (the inflectional features being lexicalized by the verb), a definition that is close to the intuition that Α-positions are projected from lexical information.

Notes *

Thanks to: Werner Abraham, Peter Ackema, Josef Bayer, Hans den Besten, Yael Brenner, Peter Coopmans, Marcel den Dikken, Martin Everaert, Arnold Evers, Ger de Haan, Wolfgang Herrlitz, Erik Hoekstra, Johan Kerstens, Henk van Riemsdijk, Eddy Ruys, Maaike Scboorlemmer, Fred Weerman, Jan-Wouter Zwart and an anonymous reviewer. Earlier versions of this paper were presented at the Tilburg University Workshop on Scrambling, Oktober 18-20, 1990, and at the University of Utrecht, december 7, 1990. I would like to thank the audiences for their useful comments.

1.

Throughout this paper, I will refer to A'-movement to a position following the subject as adjunction to V', and to A'-movement to a position preceding the subject but following COMP, as adjunction to VP. I do so, because I believe that there is no IP in Dutch (and German). However, the argumentation as I present it is not dependent on the absence of IP: if IP is assumed, the adjunctions referred to would be adjunction to VP and to IP respectively.

2.

An anonymous reviewer points out that an explanation in terms of D-structure incorporation of the resultative has to provide an answer to the question why adjuncts cannot also be incorporated. Adjunct incorporation would yield complex predicates which have exactly the order that has to be ruled out, viz. *APAdjunct-V. Complex predicates of this type are excluded by Neeleman & Weerman's (1990) Complexity Constraint, which says that X° predicates cannot have a complex head. This condition is shown to capture a wide range of facts.

3.

Marcel den Dikken has informed me that, contrary to what De Haan assumes, there are reasons to assume that the underlying order in double object construction containing an aan-PP is direct object - indirect object. This does not influence the conclusion drawn here, only we should change our interpretation of the examples. (12b) is the D-structure of (12a), which shows that there is no reconstruction.

4.

The zulke-NPs in (16a) and (16a), which I claim to be focussed, are neccesarily the prosodically most prominent elements. However, this does mean that they do not have focus. One should make a distinction between presence of the feature [+FOCUS] in the syntax and its realization at PF Culicover & Rochemont 1983).

not not the (cf.

424 Ad Neeleman

5.

This is in accordance with Chomsky's (1981) Focus-rule, which moves focussed elements at LF. One could say that the focus movements we find in this section are the S-structure variants of the LF-movements Chomsky proposes.

6.

This movement is known as heavy-NP-shift. It is discussed for Dutch in Weerman (1989). As far as syntax is concerned I hold it to be an instance of focus movement (cf. Rochemont 1983). However, at PF the position the extraposed NP is in makes that it functions as a separate intonational phrase. This probably induces the heaviness effect (cf. Zee — Inkelas 1990).

7.

For some speakers the sentences in (21) are not entirely acceptable. However, even for those speakers, the sentences have exactly the same status as long distance WH-movement in comparable contexts (for instance: Wat zegt Jan zelfs ottder vier ogen niet dat hij gekocht heeft 'What tells John even in private not that he bought has').

8.

The answer to the question why only passive subjects can license parasitic gaps in Dutch is quite simple. In Dutch, parasitic gaps are only found in controlled adjuncts (cf. Bennis — Hoekstra 1984). Except when a fry-phrase is present, the controller will be the S-structure subject. Therefore, identifying the parasitic chain with the subject as well would lead to a variable locally bound by PRO, which violates principle C. This is examplified by the ergative construction in (i). Note that (ia) is wellformed with an anaphoric object instead of a parasitic gap (cf. (ib)). (i)a. *Dat Jant [Ot [zonder PRO, /, helemaal te kennen] ί, sterft That Jan without totally to know dies b. Dat Jani [zonder PROt zichzelf{ helemaal te kennen] f, sterft That Jan without himself totally to know dies

9.

I am grateful to Peter Coopmans for his help in developing this idea.

10. An objection against the use of the COMP-indexing rule here is that in English the presence of a complementizer blocks percolation of the index of [Spec, CP]. This explains the that-t effect of English, as well as the absence of doubly filled COMPs. In Dutch, however, there is no that-t effect and COMP may contain two elements. This suggests that, even if a complementizer is present, the index of [Spec, CP] may be copied. This is why in parasitic adjucts COMP-indexing can take place, even though C is filled by a preposition. 11. Clearly, in Dutch parasitic chains can also be bound from A'-positions (cf. Bennis — Hoekstra 1984). It seems that our analysis predicts that in this language they can only be Α-bound. But note that index percolation is optional (cf. Aoun et al. 1987). If no index is percolated upwards, the parasitic chain is necessarily interpreted an pronominal: without index percolation the chain is no longer governed. Hence, both A- and A'-binding are allowed in Dutch.

Scrambling as a D-structure phenomenon

425

12. This argument against the Α-movement analysis also holds in other frameworks than Relativized Minimality. For instance, Pesetsky's (1982) Path Containment Condition forbids the representation in (36) as well. 13. Note that Weerman's analysis does not entirely forbid Scrambling in VO languages: when specifying elements can also be generated on the verb's righthand side, we might expect free word order effects. A relevant case could be Spanish, which has VOS order. If the position of the subject indicates that there can be specifying elements following the verb in Spanish, Scrambling should be possible. And indeed there are free word order effects in this language. 14. Recall that in section 2.3, we observed that there are Α-bound parasitic chains in Dutch, but not in English. This contrast followed from the claim that adjuncts containing a parasitic chain are in the governing domain of the verb in Dutch, but not in English. This claim, in turn, is confirmed by the movement in (44). The movement places the adjunct in a position that cannot be reached by the verb. Furthermore, the movement is obligatory for phrasal adjuncts, such as the ones that contain parasitic chains. Therefore, English parasitic adjuncts are never governed by V. In Dutch, on the other hand, all adjuncts (may) stay in their base position within VP, so that parasitic adjuncts are in the verb's governing domain. 15. This does not mean that we give up the distinction between internal and external Θ-roles. The external role is assigned to a designated postion, namely [Spec, VP]. This raises a question about objects that precede subjects in German (cf. (46b) and (46c)). If [Spec, VP] is defined via X-bar theory, the preposed objects should be moved, as is suggested by the data presented by Santorini & Lee (1990). However, if [Spec, VP] is defined as that position that agrees with the verb (cf. Hoekstra 1991), the preposed objects could be base generated. This agrees with the analysis of Bayer & Komfilt (1990). Joseph Bayer has informed me that the binding data Santorini & Lee present are not so clear. For me, then, the status of the preposed objects is an empirical matter. In this section I will mainly discuss (45d) and (45d). 16. A possible exception to this observation are NOM-ACC inversion verbs like ärgern 'to annoy' in (i). Here, we find Accusative on an Experiencer-object and Nominative on what could be seen as a Theme-subject. However, the problem with these constructions is that it is not clear that the subject really is a Theme. In fact, it could be analysed as a Cause, given that (i) means something like 'a little thing makes him angry'. Aergern probably has lexically saturated Theme: ärger 'annoyance' (i) Eine Kleinigkeit A little thing-NOM

ärgert ihn anoys him-ACC.

426 Ad Neeleman

References Aoun, J. — Ν. Horstein — D. Sportiche 1981 "Some Aspects of Wide-Scope Quantification", Journal of Linguistic Research 1, 69-95. Aoun, J — N. Hornstein — D. Lightfoot — A. Weinberg 1987 "Two Types of Locality", Linguistic Inquiry 18, 537-577. Bayer, J. — J. Kornfilt 1990 Against scrambling as an Instance of Move a . Talk given at the Tilburg University Workshop on scrambling. Belletti, A. — L. Rizzi 1988 "Psych-Verbs and Θ-Theory", Natural Language and Linguistic Theory 6, 291-352. Bennis, H. 1986 Gaps and Dummies. Dordrecht: Foris. Bennis, H. — T. Hoekstra 1984 "Gaps and Parasitic Gaps", The Linguistic Review 4, 29-87. Brody, M. 1990

Case Theory and Argumenthood. Ms., University College London.

Browning, M. 1987 Null Operator Constructions. Doctoral dissertation, Massachusetts Institute of Technology. Chomsky, N. 1970 "Remarks on Nominalization", in: R. Jacobs — P. Rosenbaum (eds.), Readings in English Trans formational Grammar. Waltham: Ginn & Co. Chomsky, N. 1977 "On Wh Movement", in: P. Culicover — T. Wasow — A. Akmajian (eds.), Formal Syntax. New York: Academic Press. Chomsky, N. 1981

Lectures on Government and Binding. Dordrecht: Foris.

Chomsky, N. 1982

Some Concepts and Consequences of the Theory of Government and Binding. Cambridge, Mass.: ΜΓΓ Press. Chomsky, N. 1986a Barriers. Cambridge, Mass.: ΜΓΓ Press.

Scrambling as a D-structure phenomenon

427

Chomsky, N. 1986b

Knowledge of Language; Its Nature, Origin and Use. New York: Praeger.

Chomsky, N. 1989

"Some Notes on the Economy of Derivation and Representation", ΜΓΓ Working Papers in Linguistics 10, 43-74. Chomsky, N. — H. Lasnik 1977 "Filters and Control", Linguistic Inquiry 8, 425-504. Culicover, P. — M. Rochemont 1983 "Stress and Focus in English", Language 59, 123-165. Grimshaw, J. 1990 Argument Structure. Cambridge, Mass.: ΜΓΓ Press. Haaften, T. van — S. van de Keike — M. Middelkoop — P. Muysken 1986 "Nominalisaties in het Nederlands", GLOT 8, 67-104. Haan, G. de 1979

Conditions on Rules. Dordrecht: Foris.

Haan, G. de 1987

"A Theory-bound Approach to the Aquisition of Verb Placement in Dutch", in: G.J. de Haan — W. Zonneveld (eds.), Formal Parameters of Generative Grammar; OTS Yearbook III. Dordrecht, 15-30. Hoekstra, E. 1991 On the Relation between Arguments and Heads. Talk given at the 1991 GLOW colloquium at the University of Leiden. Hoekstra, T. 1984 Transitivity: Grammatical Relations in Government Binding Dordrecht: Foris.

Theory.

Hoekstra, T. 1986

"Deverbalization and Inheritance", Linguistics 24, 549-584.

Iatridou, S. 1990

"On Agr(P)", Linguistic Inquiry 21, 551-577.

Kerstens, J. 1975

Over afgeleide stnictuur en de interpretatie van zinnen. Ms., University of Amsterdam.

428 Ad Neeleman

Koster, J. 1984

"On Binding and Control", Linguistic Inquiry 15, 417-459.

Mahajan, A. 1990 The A/A-bar Distinction. Doctoral dissertation, Massachusetts Institute of Technology. Neeleman, A. — F. Weerman 1990 Case Theory and the Diachrony of Complex Predicates. Ms., University of Utrecht. Pesetsky, D. 1982 Paths and Categories. Doctoral dissertation, Massachusetts Institute of Technology. Pinker, S. 1984 Language Learnability and Language Development. Cambridge, Mass.: Harvard University Press. Riemsdijk, H. van 1987 "Movement and Regeneration", in: P. Benincä (ed.), Dialect Variation and the Theory of Grammar; Proceedings of the GLOW Workshop in Venice. Dordrecht: Dordrecht. Rizzi, L. 1986

"Null Objects in Italian and the Theory of pro", Linguistic Inquiry 17, 501-557.

Rizzi, L. 1990

Relativized Minimality. Cambridge, Mass.: ΜΓΓ Press.

Rochemont, M. 1978

A Theory of Stylistic Rules in English. Doctoral dissertation, University of Massachusetts. Santorini, B. — Y.-S. Lee 1990 Long Distance scrambling and Anaphor Binding in German. Talk given at the Tilburg University Workshop on scrambling. Stowell, T. 1981 Origins of Phrase Structure. Doctoral dissertation, Massachusetts Institute of Technology. Vikner, S. 1990 Scandinavian Object Shift vs. Continental West Germanic scrambling. Talk given at the Tilburg University Workshop on scrambling.

Scrambling as a D-structure phenomenon

429

Webelhuth, G. 1989 Syntactic Saturation Phenomena and the Modern Germanic Languages. Doctoral dissertation, University of Massachusetts. Weerman, F. 1989 The V2 Conspiracy, A Synchronic and a Diachronic Analysis. Dordrecht: Foris. Wyngaerd, G. vanden 1988 "Object Shift as an Α-movement Rule", MIT Linguistics 11.

Working Papers

in

Zee, D. — S. Inkelas 1990 "Prosodically Constrained Syntax", in: S. Inkelas — D. Zee (eds.), The Phonology-Syntax Connection. Chicago: University of Chicago Press.

Notes on movement to the Spec(IP) position in Finnish1 Urpo Nikanne

I will discuss filling the Spec(IP) position in Finnish. The starting point is the well-known observation that Finnish avoids verb initial affirmative sentences. The Spec(IP) position must be filled but it appears that this is not because of theta-marking or case marking. It seems that on different levels of representation the reasons for the movement to the Spec(IP) position are different. I will focus on a Finnish2 construction in which the NP-argument that normally appears as the subject of the sentence is in the partitive case. I will refer to this structure with the term "partitive structure." This structure seems to be interesting with respect to the case theory and the theta theory. It has been assumed in recent studies within generative grammar (Vainikka 1987, 1989; Nikanne 1988, 1990) that the partitive is a structural case of a complement. Because the argument in question belongs to the argument structure of the verb, it seems that on the relevant level of representation the argument is in the complement position of the verb.3 If so, the sentences are subjectless at the level of representation where the cases are assigned. The subject of the sentence position is filled after that level.

1.

Organization of grammar

I assume that the organization of grammar is roughly as follows (Jackendoff 1983, 1987a, 1987b, 1990):

(1) lexicon

motor output phonological structures

1

syntactic ^ structures

audition

\

conceptual structures

-> vision

auditory, input

Following Jackendoff, I assume that the the human mind consists of separate autonomous components having their own primitives and rules of combination. The components are in interaction with each other

432 Urpo Nikanne

through correspondence rules. The language faculty contains the syntactic component and the phonological component. The lexicon is a part of the correspondence rule system between the conceptual structures and the linguistic structures. The syntactic component is organized as in (2): (2) D-structure S-structure PF

LF

Even though (2) looks like the familiar GB-organization of syntax, one should still know what the labels stand for in the present paper: the D-structure is derived directly from the lexicon. The S-structure is the representation sensitive to the morphological marking such as case marking and agreement. PF is the syntactic representation which interfaces the phonological structure. LF is the syntactic representation in which quantifiers, and possibly pronouns (Pica 1987; Hestvik 1990), are in the places where their scopes are in accord with the syntactic requirements. The syntactic constituent structure in the analyses of this paper is as presented in Chomsky (1986). There are theories of Finnish clause structure in which the IP is decomposed into a set of inflectional phrases in the spirit of Pollock (1989) (e.g. Holmberg et al. 1991; Mitchell 1991),4 but for the present purposes the simple Infi can represent the more articulated structure. Following Nikanne (1990) I assume that the theta-hierarchy is a consequence of the so called zone principle of the conceptual structure. That principle is, though very interesting, still outside of the scope of this paper. However, I assume that the arguments of verbs respect the following theta-hierarchy, which is a part of a larger hierarchy (see e.g. Jackendoff 1990; Grimshaw 1990): (3)

Agent > Theme > Goal/Route/Source/Location

The definitions of the roles can be found in Jackendoff (1983, 1987b, 1990) and Nikanne (1990).

Notes on movement to the Spec(IP) position in Finnish 433

In addition, following Jackendoff (1983,1987b, 1990) and Nikanne (1990), I assume that selection and theta-marking follow from the wellformedness conditions of the conceptual structure: the lexical conceptual structures consist of incomplete parts of the conceptual structure, and the conceptual well-formedness conditions make the words look for each other in order to make a well-formed conceptual structure together. (On the priority of semantic selection over syntactic selection, see also Chomsky (1986a)).

2.

The sentence initial positions

2.1.

Avoidance of verb initial sentences in Finnish

It is well-known that Finnish avoids verb initial declarative sentences (see e.g. Hakulinen — Karlsson 1979; Vainikka 1989; Vilkuna 1989); this is illustrated in (4) through (6). Because Finnish is an SVO language, the possibility of verb initial sentences arises in subjectless structures, such as the partitive structure in (4), the passive construction in (5) and the generic construction in (6). One could claim that in the generic construction the subject is a small pro. (4)5 a.

b. c. d. (5) a.

b.

Partitive structure: Tänään ihmisiä käveli taloon today people+PAR walk+PST(3SG) house+ILL "There were people walking into the house today.' ihmisiä käveli tänään taloon. taloon käveli tänään ihmisiä. *käveli tänään ihmisiä taloon. Passive: kadulla kävel+lä+än taloon street+ADE walk+PASS+AGR house+ILL 'People are walking to the house in the street.' *kävellään taloon kadulla.6

434 Urpo Nikanne

(6) a. c. d.

Generic: kadulla kävelee (3SG). 'one can walk in the street' taloon kävelee (3SG). 'one can walk to the house.' *kävelee taloon/kadulla.

It should be noted that a Finnish sentence can begin with a verb in imperative mood as in (7a), in yes/no questions as in (7b), and if the verb is focused as in (7c). ("Q-CL" stands for "question clitic" and "FOC-CL" stands for "focus clitic.") (7)a.

kävel+kää taloon walk+IMP-2PL house+ILL 'walk into the house!'

b.

kävel+i+fcö ihmisiä taloon walk+PST+Q-CL people 'were there people walking into the house?'

c.

kävel+i(+päs) ihmisiä taloon walk+PST(+FOC-CL) people+PAR house+ILL 'There WERE people walking into the house.'

Let us formulate the requirement preliminarily as follows: (8)

Avoidance of Verb Initiality (AVI) Finnish avoids verb initial declarative sentences.

The AVI is just a descriptive observation, but the following questions arise: is there a specific preverbal position, say Spec(IP), C, Spec(CP) that must be filled or is it any preverbal position that can satisfy the AVI? How can the AVI be explained? At what level of representation does the AVI hold? According to Vainikka (1989) and Vilkuna (1989) the position in the question is Spec(IP). I will not object. Note that word orders like the one in (9) are very natural in Finnish, which supports the idea that wh-movement and the movement "triggered by the AVI" move elements to different positions. Given that the wh-element ends up in the Spec(CP), the most natural place for the "AVI-moved" element is in the Spec(IP):

Notes on movement to the Spec(IP) position in Finnish 435

(9)

mihin kadulla kävellään? where+ILL street+ADE walk+PASS 'Where are people walking in the street?'

Vainikka's and Vilkuna's explanations for the phenomenon are, however, very different. According to Vainikka's (1989) theory, the Spec(IP) position must be filled in order to make the position visible. The element that has to move there can be at least an NP or a PP. However, Vainikka's theory does not tell which factor in each case makes a specific element move to the Spec(IP) when there are different possibilities. Maria Vilkuna (1989, p.c.) assumes that Finnish is a "discourse configurational" language. There are two fixed verb initial positions, as illustrated in (10) (cf. Kiss 1981): next to the verb sits the discourse topic, and in front of that position is a position for a contrasted, or, in my terms, focused element (cf. also Karttunen — Kay 1985). In Vilkuna's theory, the TOPIC-position is reserved for the entity the sentence is about. A nominative subject is a default topic. If we translate Vilkuna's model into current generative terminology, the TOPIC-position corresponds to the Spec(IP) position and the CONTRAST-position corresponds to the Spec(CP) position (Vilkuna p.c., Vainikka 1989). (10) CP CONTRAST

C

C

IP TOPIC

Γ

In the light of Vilkuna's theory, the AVI follows from the discourse fact that a sentence normally has a topic, and the topic very seldom is a verb. Given the analyses of Vainikka and Vilkuna, we can assume two things: (i) the Spec(IP) position must be filled at some level of representation, (ii) the Spec(IP) position functions as the topic position at some level of representation.

436 Urpo Nikanne

2.2.

On the partitive case

Before going into the partitive structure, we must briefly discuss the nature of the partitive. It is often claimed (recently e.g. Toivainen 1985; Larjavaara 1988) that the partitive has an inherent meaning which has to do with partiality, non-entirety, open class or the like, and the partitives in the partitive structure are subjects and not objects. However, the different uses of partitive in the grammar of Finnish can be explained more naturally, if we assume that the partitive does not express any specific meaning but a specific syntactic position, namely complement position of any syntactic category, of the NP to which the ending is attached (Vainikka 1988, 1989, 1991; Nikanne 1988). The partitive does not always express partiality or anything like that. Consider for instance (11). What these examples have in common is that the partitive is assigned to a complement-NP. (ll)a.

[,P[NP Kauko] [r[v.,katsoi] Kauko look-at+PST(3SG) ISP Yrjö+ä]]]] Yrjö+PAR 'Kauko looked at Yrjö.'

b. [,p[Np Matti] [,[v., ei syönytj Matti not(3SG) eat+PTC Ivplv e][NP omena+a]]]] e apple+PAR 'Matti did not eat a/the apple.' c.

[pplp ip kohti] [NP talo+a]]] toward house+PAR 'toward a/the house' päin] [NP puu+ta]]] against tree+PAR 'against a/the tree'

IPPIP

IP

[VP [v

e] e

Notes on movement to the Spec(IP) position in Finnish 437

d.

Matti on [^IA-IA pitempij [NP Ville+ä]]} Matti is taller Ville+PAR 'Matti is taller than Ville.'

e.

[Np[Nls annos] [NP jäätelöä]]] portion ice-cream+PAR 'a portion of ice cream'

In (11a) the partitive is assigned by a verb to its object. As ( l i b ) illustrates, the object of a verb is always in the partitive in the scope of negation. In (11c) the partitive is assigned by a preposition, and in (1 Id) the partitive is assigned by an adjective. In (1 le) the partitive is assigned by a noun. The other objective cases in Finnish, the accusative and the nominative (or the endingless accusative, depending on the theory),7 can be assigned by verbs only, but partitive can be assigned at least by prepositions, adjectives, and nouns, and possibly by numerals (e.g. kolmepoika+a = three boy+PAR 'three boys') and some other quantifiers (e.g. pari poika+a = a-couple-of boy+PAR 'a couple of boys') (see e.g. Vainikka 1989, 1991).

3. 3.1.

The partitive structure The partitive structure and verbs with one NP-argument

Most naturally the verb in the partitive structure has only one NPargument and there is at least one element corresponding to a place or a path in the conceptual structure. However, it is not necessarily the NP-argument of the verb that must be in the verb initial position: a PPargument or a PP-adjunct can satisfy the AVI (we will discuss these adjuncts briefly in section 3.3). (12)a. taloon kävelee ihmisiä house+ILL walks people+PL+PAR 'There are people walking into the house.' b. ihmisiä kävelee taloon. c. *lkävelee ihmisiä taloon. [see, however, (7c)]

438 Urpo Nikanne

(13)a. kadulla kävelee ihmisiä street+ADE walks people+PL+PAR 'There are people walking in the street.' b. ihmisiä kävelee kadulla. c. *kävelee ihmisiä kadulla. [see, however, (7c)] The different acceptable word orders have different statuses in discourse. Vilkuna's theory seems to fit the word order facts: the element in the Spec(IP) position is the topic of the sentence. Sentence (12a) is about the house, (12b) and (13b) are about the people, (13a) is about the street. Assuming that the partitive case is assigned to any complement position at S-structure, it seems to be clear that the partitive is assigned to a complement position here as well. And, because the NP ihmisiä (people+PL+PAR) 'people' is an argument of the verb kävele, the first assumption is, of course, that it is an object of the verb. Note that in Finnish the verb agrees in person and number with the subject in the S-structure: (14)a. [IP minä [r [v., kävele+n ]...]] = I(NOM) walk+lSG b. [lP sinä [r [v., kävele+t]...]] = you(NOM) walk+2SG c. [,p hän/se/koira [r [v_, kävele+e]...]] = (s)he/it/dog(NOM) walk+3SG d. [,p me [r [v., kävele+mme]...]] = we walk+IPL e. [,p te [r [v.j kävele+tte]...]] = you walk+2PL f. [,p he/ne/koirat [,. [v_, kävele+vät]...]] = theyVdogs(NOM) walk+3PL In the partitive structure the verb does not agree with the subject, but it is in the 3SG-form which is the neutral form. In principle, this could be explained by the fact that the verb does not agree with a lexically case marked subject (Maling 1991). But, as the examples above show, the verb kävele 'walk' can have a nominative subject. Thus, the most salient and natural solution is that the partitive is assigned to the Compl(VP) position (see also Wiik 1974; Reime 1986; Vainikka 1989; Nikanne 1988, 1990).

Notes on movement to the Spec(IP) position in Finnish

439

Thus, the S-structure is something like as follows (the continuous line indicates movement and the dashed line stands for caseassignment): (15)7 ν

__ IP I I' V-I Kävele+e (NEUTRAL 3SG)

VP .......—~— ... V' V e

\ -· v

NP ... ihmisi+ä

/' /

^

' '

1

's

>

PAR '

At the PF either ihmisiä or some adjunct moves to the Spec(IP) position.

3.2.

The partitive structure and verbs with two NP-arguments

Thus far we only have had examples of verbs with only one NPargument. Now we will turn into verbs with two NP-arguments. Consider sentences in (17). They contain both an Agent and a Theme or a Goal. Agents are usually considered as external arguments or at least the highest arguments in the theta-hierarchy (see e.g. Williams 1984; Bresnan — Kanerva 1989; Grimshaw 1990; Jackendoff 1990; Nikanne 1990). In the following sentences both the Agent and the Theme or Goal are in the partitive.9 The theta-marking properties of the verbs in the sentence are as follows (the parentheses indicate optionality):10 (16)

pitä 'wear' [Agent, Theme (Location)] äänestä 'vote for' [Agent, Goal] hake 'apply for' [Agent, Theme]

440 Urpo Nikanne

(17)a. Sinne There

ilmestyi appeared

"marttyyreja" martyr+PL+PAR

muistelevia recollect+PTC+PL+PAR

seinälehtiä wall-newspaper+PL+PAR

ja opiskelijoita [Agent] piti mustia and student+PL+PAR wore(3SG) black+PL+PAR käsivarsinauhoja [Theme] kun verilöylystä oli arm-band+PL+PAR when massacre+ELA was kulunut sata päivää. go+PTC hundred day+PAR 'There [on the Peking University campus] appeared wallnewspapers that recollected the "martyrs" and there were students wearing black arm-bands, when it was one hundred days from the massacre.' (SK 38/1989) b.

Valintansa jälkeen presidents Election+GEN+Px3 after president

Urho Urho

Kekkosella oli kaikki polittiset Kekkonen+ADE was(3SG) all political+PL-NOM intressit osoittaa, että myös intrest+PL-NOM show+INF that also oikeistolaisia right-wing+PL+PAR

vatitsijamiehiä [Agent] elector+PL+PAR

äänesti häntä (Goal). vote-for+PST(3SG) he+PAR 'After being elected, it was in president Urho Kekkonen's political interests to show that there were also right-wing electors who voted for him.' (V. Vennamo, HS 9/7/90)

Notes on movement to the Spec(IP) position in Finnish 441

c.

VirkaalTheme] post+PAR

haki apply-for+PST(3SG)

useita several+PAR

henkilötiä (Agent). people+PL+PAR 'The post was applied for by several people.' (Hakulinen — Karl sso η 1979) In the normal case the highest NP-argument in the theta-hierarchy is in the subject position in the S-structure, and it is the only argument that is able to stay there; in other words, the argument occupying the Spec(IP) in the S-structure is interpreted as the highest argument in the theta hierarchy. This is illustrated in (18) (# stands for a pragmatically or semantically strange sentence): (18)a. Ihmiset [Theme] kävele+vät kadulla people(NOM) walk+3PL street+ADE 'People are walking in the street.' b. opiskelijat [Agent] piti+vät students(NOM) wear+PST+3PL käsivarsinauhoja [Theme], arm-band+PL+PAR 'Students are wearing arm bands.' c. tfkäsivarsinauhat [Agent] piti+vät arm-bands(NOM) wear+3PL opiskelijoita [Theme] student+PL+PAR 'Arm bands are wearing students.' d. useat ihmiset [Agent] haki+vat many+PL-NOM people+PL-NOM apply-for+PST+3PL virkaa [Theme] post+PAR 'Many people applied for the post.'

442 Urpo Nikanne

e. #virka [Agent] haki useita post(NOM) apply-for+PST(3SG) many ihmisiä [Theme] people+PL+PAR "The post applied for many people.' f.

vaütsijamiehet [Agent] äänestivät häntä [Goal] elector+PL-NOM vote-for+PST+3PL (s)he+PAR 'The electors voted for him/her.'

g. hän [Agent] äänesti valitsijamiehiä [Goal] (s)he(NOM) vote-for+PST(3SG) elector+PL+PAR '(S)he voted for the electors.' Assuming that the partitive is always an objective case and given the analysis of sentences (12) and (13), it is logical to assume that in the partitive structure both arguments can appear as objects in Finnish. The structure is, thus as follows:11 (19) IP Γ V-I (NEUTRAL 3SG) NP

NP

PAR Again, the partitive is not a lexical case here because the nominative subject is possible as sentences in (18) show. For instance in (18a), the verb pitä 'keep, have, wear' has two arguments. One of them is usually the subject of the sentence and the other one is the object in the Sstructure. In that case the verb agrees with the subject in person and number, as we see in (18b). In the partitive structure there is no subject/verb agreement, and the verb is in the neutral third person singular form.

Notes on movement to the Spec(lP) position in Finnish 443

Because of the Empty Category Principle (e.g. Chomsky 1981, 1986b) one would not assume that the "external argument" has lowered from the subject position of the sentence to the Compl(VP) position, or any other complement position. It is thus natural to conclude that the "external argument" is base-generated in the complement position. Given that the S-structure is the level of representation sensitive to case marking and morphological marking, the movement to the Spec(IP) position has taken place after the S-structure, in the PF. Note that the movement to the subject position at PF does not respect the theta-hierarchy. In (17a) the order is Agent-Verb-Theme, in (17b) the order is Agent-Verb-Goal, and in (17c) the order is ThemeVerb-Agent. It seems that if a verb has two NP-complements, the discourse factors determine which one of them moves to the Spec(IP) position. It is the topic that will occupy the Spec(IP) position. In the partitive structure it is possible to change the order, but if we do that, we change the topic. Finnish seems to have a double object construction, but it differs very much from that in English (on recent views, see Larson 1988, 1990; Jackendoff 1990) or Scandinavian languages (see e.g. Holmberg 1990) in which the "indirect object" is a Recipient (in our terms, Recipient is a Goal in a possessive expression). In Finnish the double object construction has specific semantic properties as well, and we will go into them in section 3.4.

3.3.

On the scope of modal elements

It seems that modal elements can support the analysis that all the elements associated to the situation associated with the verb are VPinternal in the D-structure. Before going into the modals, let us discuss the structure of the VP first. Each syntactic counterpart of the situation associated with the verb seems to be located in the VP. According to Jackendoff (1983, 1987b, 1990) and Nikanne (1990) the selectional and theta-marking properties of the verb kävele 'walk' are as follows: (20)

kävele 'walk' [Theme, PATH]

444 Urpo Nikanne

To say it simply, kävele selects one theta argument and one PP that corresponds to a path. In the sentence Ihmisiä kävelee kadulla taloon (people+PAR walk+3SG street+ADE house+ILL) 'in the street, there are people walking into the house' the verb has two selected complements: ihmisiä 'people' (= Theme) and taloon 'into the house' (= path). Kadulla 'in the street' is an adjunct, in the sense that it is not selected by the verb kävele. As we have seen, one can move words around in Finnish sentences, and it is often difficult to know the Dstructure placement of a specific word. Let us see how the adjunct behaves in an infinitive structure from which it cannot be taken out: (21)

Hallitus antoi... Government let+PST(3SG) a. [ihmisten kävellä kadulla kirkkoon] people+GEN walk+INF street+ADE church+ILL b. [ihmisten kävellä kirkkoon kadulla J c. *[ihmisten kadulla kävellä kirkkoon] d. *[kadulla ihmisten kävellä kirkkoon]

The most unmarked case is (21a) but (21b) is not much worse. It is often claimed that this kind of infinitive structure is a VP (e.g. Nikanne 1988, 1989b, 1990; Vainikka 1989). If this is the case, the only place for an adjunct seems to be the complement position (see (22a)). However, if we assume that the infinitive marker is attached by headto-head movement of the verb to an inflectional head of the infinitive phrase, the whole paradigm looks more natural (see (22b); "InfP" stands for "infinitive phrase"): (22) a. VP ihmisten

V'

V NP PP PP kävellä e kadulla kirkkoon

Notes ort movement to the Spec(IP) position in Finnish 445

InfP

In the latter case kadulla is in the specifier position of the VP. The unmarked case for a specifier is in front of the V*. It is worth noting that such adverbs as nopeasti 'fast' and kauniisti 'beautifully' behave in the same way as kadulla does. These adverbs are more traditionally assumed to be specifiers: (23)

Hallitus antoi... Government let+PST(3SG) a. ihmisten kävellä nopeasti/kauniisti people+GEN walk+INF fast/beautifully kirkkoon church+ILL b. ihmisten kävellä kirkkoon nopeasti/kauniisti. c. *ihmisten nopeasti/kauniisti kävellä kirkkoon. d. *nopeasti/kauniisti kävellä kirkkoon.

Whichever analys is correct, the adjunct is in the VP in the Dstructure. Modal elements are not parts of the Situation but merely elements having scope over the Situation (Possibly John is going to the house; scope over the whole Situation John is going to the house) or some parts of it (Possibly JOHN is going to the house; scope over John only). Note that a sentence initial modal element (e.g. the adverb luultavasti 'probably') in the Spec(IP) position is transparent with respect to AVI. Sentence (24) is strange if nothing is focused, but the sentences in (25) are all fine:

446 Urpo Nikanne

(24)

niuultavasti probably

kävelee ihmisiä kadulla walk+3SG people+PL+PAR street+ADE

(25)a. kadulla luuUavasti kävelee ihmisiä street+ADE probably walk+3SG people+PL+PAR 'Probably there are people walking in the street.' b. ihmisiä luuUavasti kävelee kadulla. '(= a)' c.

tänään ['today'] luuUavasti kadulla kävelee ihmisiä 'Probably there are people walking in the street today.'

d. luuUavasti tänään kadulla kävelee ihmisiä. '(= c)' e.

tänään kadulla luuUavasti kävelee ihmisiä. '(= c)'

f.

(tänään) kadulla kävelee luultavasti ihmisiä '(Today) it is probably people who are walking in the street.'

It seems that in a finite sentence of Finnish, the modal elements with scope over the whole situation are located somewhere in the structure of IP. Thus, they c-command the VP which corresponds to the situation. This can explain why any modal element that precedes the verb seems to have scope over he whole sentence: it has the scope over the expressed situation, including the "external argument." That is, these modal elements have scope even over the NP or PP that has been raised to the Spec(IP) position, supporting the idea that at an earlier level of representation the NP or PP occured in the VP. This explains why the modal element has scope over the raised NP or PP, since at this earlier level the modal element c-commanded the NP or the PP.

3.4.

Quantitative indefiniteness

In Finnish, thus, the highest theta argument of the predicate verb has two possible locations in the S-structure: either it is the subject of the sentence or it is the object of the verb. We can assume that there is a difference between meanings of the partitive structure and the nominative subject structure.

Notes on movement to the Spec(lP) position in Finnish 447

The difference between the nominative subject constructions and partitive constructions is something we could call "quantitative indefiniteness effect." This effect is associated with the partitive construction.12 There is a quantitative indefmeteness associated with the NP that remains in the Compl(VP) position even though it would be able to rise to the Spec(IP) position. TTiose verbs that take two NParguments and those that take only one behave similarly with respect to the quantitative indefiniteness. Consider the sentences in (26): (26)a. Kadulla käveli (nii+tä) street+ADE walk+PST(3SG) (those+PAR) ihmis+i+ä

people+PL+PAR 'There were (some of those) people walking in the street.' b. Kupi+ssa on (si+tä) vet+tä cup+ine is (that+PAR) water+PAR 'There is (some of that) water in the cup.' c. (?Nii+tä) opiskelijo+i+ta piti (TThose+PAR) student+PL+PAR wear+PST(3SG) käsivarsinauho+j+a arm-band+PL+PAR 'There were some (?of those) students wearing arm-bands.' The interpretation is that in (26a) there is some number of people walking in the street. The people can even be from a definite group but the number of them is indefinite. Similarly, in (26b) there is some indefinite amount of water in the cup. The water in the cup can be a part of some larger definite amount of water. In (26c) the number of the students is indefinite, still they perhaps can be members of some definite group of students. However, the definite determiner is strange but not totally ungrammatical in (26c). An alternative for the double object analysis would be to assume that the partitive is assigned by an empty quantifier that causes the quantitative indefiniteness reading:

448 Urpo Nikanne

(27) Q 'some'

ν

- PAR

NP ) x

)

This is not an impossible analysis (see e.g. Emonds 1987), but, to avoid unmotivated abstractions in syntax, before assuming an empty quantifier and an empty head for the NP in the partitive, it is reasonable to check whether it is possible to find an overt head and assume that the indefinite quantity is associated with a specific structural configuration.

4.

Why do we have subjects at all?

In the partitive structure, the "external argument" is not forced to move into the Spec(IP) position for theta-marking. The partitive structure seems to show, as well, that the verb can assign two partitives in Sstructure. If the Spec(IP) position is neither an argument position nor a theta-position in the partitive structure, it is not one in the normal nominative subject construction either. Given the that any operation should be forced by some principle (see e.g. Chomsky 1988), why should an NP-argument rise to the Spec(IP) position ever? We have at least the following possibilities: (i) There is a case hierarchy, according to which NPs favor the nominative before the objective cases (see e.g. Keenan — Comrie 1977). (ii) There is a position hierarchy, according to which the Spec(IP) position is more favorable than the Compl(VP) position (cf. Nikanne 1988). (iii) Infi is "stronger" than the verb in the sense that it can pull one NP out from the VP in the S-structure. Whatever of the possibilities correct, the NP that can be moved to the Spec(IP) position is the highest one in the theta-hierarchy. Infi and V differ from each other in that V can assign a case to two NPs but Infi can assign nominative to no more than one NP. This can be explained functionally: the function of morphological coding, e.g.

Notes on movement to the Spec(IP) position in Finnish 449

case marking and agreement, is to make the derivation from the phonological structure into the syntactic structure effective. The case indicates the status of the carrier of the case in the S-structure. When only one NP can satisfy the thematic requirements of the Spec(IP) position, which, on the other hand, is a more attractive position than the Compl(VP) position where the arguments are base-generated, the thematic status of the NP in the Spec(IP) in the S-structure is unambiguous. There are many different technical possibilities to derive the "external" argument from the the theta-hierarchy and the selectional properties of verbs. Here is one: note that roughly speaking one can passivize every verb in Finnish, and in standard Finnish the passive marker is always counted as an argument having the highest role in the theta-hierarchy (cf. Reime 1986). The only theta-argument the verb naura 'laugh' selects is a Theme, and the theta-arguments selected by the verb maala 'paint' are Agent and Goal. (28)a. kotona naure+tt+i+in home+ESS laugh+PASS+PST+AGR 'people (Theme) were laughing at home.' b.

kotona maala+tt+i+in seiniä home+ESS paint+PASS+PST+AGR wall+PL+PAR 'people (Agent) were painting walls (Goal) at home.'

Note that the place of the passive marker is right after the verb stem before the other inflectional suffixes. It is assumed in Nikanne (1990: 128) that in Finnish the linear order of the arguments is fixed in the Dstructure: first comes the verb, then the arguments in the complement position from left to right respecting the theta-hierarchy.13 Thus, the PFs of the sentences in (28) are as follows:

450 Urpo Nikanne

(29)sl

In the S-stnicture the passive suffix ttA attaches to the verb stem and the verb moves to Infi. And in the PF the adjunct PP moves from the VP to the Spec(IP).

5.

Conclusion

It seems that Vilkuna's theory of discourse configurationality is, to a large extent, on the right track, given that only the PF is sensitive to the discourse factors. Note, however, that syntax overrides the discourse factors: the most unmarked case is that the Spec(IP) position is filled with an NP already in the S-structure when the NP highest in the thetahierarchy will move to the Spec(IP). In that case the topic position is already occupied in the PF, and the subject is interpreted as the topic of the sentence and the discourse factors must take the structure as it is. We found that in Finnish the arguments of verbs have two possible

Notes on movement to the Spec(IP) position in Finnish 451

positions in the S-structure: Spec(IP) and Compl(VP). This makes us to rethink the nature of the notion of "argument position". It is not a position that must be occupied by an argument but merely a position that can be occupied by an argument. Different positions have different requirements for the argument they let in in the S-structure: for instance, the Spec(IP) only allows the NP-argument that is the highest one in the theta-hierarchy. The morphological cases in Finnish seem to be codes that are assigned to NPs in specific positions in the S-structure. There seems to be no requirement for a case assigner to assign the case it is able to assign. For instance, in the partitive structure the Infi does not assign the nominative case. It seems that the different levels of representation have different functional statuses (cf. Nikanne 1989a). In the light of this paper at least the following functional requirements seem to hold:14 the Dstructure is sensitive to the requirements of the lexicon, which guarantees that the words in the sentence are correctly used. In the Sstructure, the movements of arguments purport to make the interpretation of the thematic relations in the sentence as unambiguous as possible. The PF is sensitive to the discourse factors; if it is possible — if the relevant positions are still available ~ the discourse factors can lead to a movement, e.g. to a movement to the empty TOPIC-position (= Spec(IP)) in subjectless structures.

Notes 1. I am extremely grateful to Anders Holmberg, Ray Jackendoff, Joan Mating, Anne Vainikka and Maria Vilkuna for their various insightful comments on earlier versions of this paper. Also I would like to thank the organizers and participants of the Tilburg scrambling workshop. My paper has benefited from the comments but the mistakes, of course, are of my own design. 2. I am a native speaker of Finnish myself. Thus, the judgments of grammaticality or markedness are based on my own competence. 3. I am not assuming the Burzio's (1986) generalization, according to which, "A verb Case-marks its object if only if it theta-marks its subject" (see also Chomsky 1986a: 141).

452 Urpo Nikanne

4. According to Holmberg et al. (1991), the structure of the Finnish finite clause is as follows:

The structure proposed in Mitchell (1991) is basically the same. 5. I use the following abbreviations: + stands for a morpheme boundary; parentheses in glosses indicate that the inflectional category has no overt suffix; a dash indicates a portmanteau morpheme; ABL = ablative case, ALL = allative case, ACC = accusative case, ELA = elative case, ESS = essive case, INF = infinitive, NOM = nominative case, PAR = partitive case, PASS = passive marker, PASS-AGR = passive agreement suffix, PL = plural marker, PTC = participle, Px = possessive suffix (note Px3 stands for the 3rd person Px; in the 3rd person there are no separate Pxs for the plural and the singular), SG = singular. Note that (4d), (5b), and (6d) are acceptable if the verb is focused; we will return to this. 6. The sentence is acceptable if the passive is used as the 1st person singular imperative, or if the verb is focused. 7. In traditional grammars and in many studies within generative grammar, e.g. in Vainikka (1989) and Reime (1991), there are two kinds of accusatives: one has the ending η and one is endingless. Timberlake (1974, 1975), Taraldsen (1986), and Maling (1991) analyze the endingless accusatives as nominative objects. 8. In Finnish he is a personal pronoun whereas ne is a demonstrative pronoun. 9. These structures are not discussed in the literature because sentences like (17) are not accepted by prescriptive grammar books. However, they are rather common in modem Finnish. For instance Belletti (1988) rules these structures out, probably because they cannot be found in traditional grammar books. It's my experience that if you take any daily newspaper you can find a couple of these structures even though the writers are taught to avoid them. (17a) is from a weekly magazine, (17b) is from a daily newspaper, and (17c) is from a Finnish syntax textbook.

Notes on movement to the Spec(IP) position in Finnish 453

10. Note that I am not assuming that the theta-grids are present in the lexical entries, but that they are consequences of the "aspectual stracture" (on the term, see Grimshaw 1990). On theta marking, see Nikanne (1990). 11. Note that I am not assuming that binary branching is necessarily respected. As the examples in (17) show, there seems to be no obvious reason to assume that one of the arguments in the partitive case should be considered syntactically different even though semantically they differ from each other. In the analyses that follow, binary branching is not assumed either. 12. On the quantitative indefiniteness in some predicative sentences, see Larjavaara (1988). 13. According to Nikanne (1990), the arguments have the same order in the D-structure as their conceptual counterparts have in the theta level, which is a tier in the conceptual structure. The order of the arguments in the theta level follows from the zone principle, which we mentioned in section 2. The theta-hierarchy, in turn, follows from the ordering in the theta level. 14. I will not say anything about the LF because it is not discussed in this paper.

References Baker, M.C. 1988 Incorporation: A Theory of Grammatical Function Changing. Chicago: University of Chicago Press. Belleti, A. 1988 "The Case of Unaccusatives", Linguistic Inquiry 19, 1-34.

Bresnan, J. — J. Kanerva 1989 "Locative Inversion in Chichewa: a Case Study of Factorization in Grammar", Linguistic Inquiry 20, 1-50. Burzio, L. 1986

Italian Syntax. Dordrecht: Reidel.

Chomsky, N. 1986a Knowledge of Language. New York: Praeger. Chomsky, N. 1986b Barriers. Cambridge ΜΑ: ΜΓΓ Press.

454 Notes on Movement to the Spec(lP) Position in Finnish

Chomsky, N. 1988 Some Notes on Economy of Derivation and Representation. Ms., Massachusetts Institute of Technology. Emonds, J. 1987

"The Invisible Category Principle", Linguistic Inquiry 18, 613-632.

Grimshaw, J. 1990

Argument Structure. Cambridge ΜΑ: ΜΓΓ Press.

Gruber, J.S. 1965

Studies in Lexical Relations. Doctoral dissertation, Massachusetts Institute of Technology; Indiana Linguistic Club, Bloomington. Hakulinen, A. — F. Karlsson 1979 Nykysuomen lauseoppia. Helsinki: Finnish Literature Society. Hellan, L. 1988

Anaphora in Norwegian and the theory of Grammar. Dordrecht: Foris.

Hestvik, A. 1990

LF-movement of Pronouns and the Computation of Binding Domains. Doctoral dissertation, Brandeis University. Holmberg, A. 1990

On the Scandinavian Double Object Construction. Ms., Uppsala University.

Holmberg, A. forthcom. Case and Other Topics in Finnish Syntax. Dordrecht: Foris. Holmberg, A. — U. Nikanne — I. Oraviita — H. Reime — Τ. Trosterud 1991 "The Structure of INFL and the Finite Clause in Finnish", in: Holmberg — Nikanne (eds.). HS

= Heisingin Sanomat, a daily newspaper.

Jackendoff, R. 1972

Semantic Interpretation in Generative Grammar. Cambridge ΜΑ: ΜΓΓ Press.

Jackendoff, R. 1983

Semantics and Cognition. Cambridge ΜΑ: ΜΓΓ Press.

Jackendoff, R. 1987a Consciousness and Computational Mind. Cambridge ΜΑ: ΜΓΓ Press.

Urpo Nikanne

455

Jackendoff, R. 1987b "The Status of Thematic Relations in the Linguistic Theory", Linguistic Inquiry 18, 369-411. Jackendoff, R. 1990a Semantic Structures. Cambridge ΜΑ: ΜΓΓ Press. Jackendoff, R. 1990b "On Larson's Treatment of the Double Object Construction", Linguistic Inquiry 21, 427-456. Karttunen L. — M. Kay 1985 "Parsing in a Free Word Order Language", in: D. Dowty — L. Karttunen — A. Zwicky (eds.), Natural Language in Parsing. Cambridge: Cambridge University Press. Keenan, E. — B. Comrie 1977 "NP Accessibility and Universal Grammar", Linguistic Inquiry 8, 63-100. Kiss, Katalin E. 1981 "Structural Relations in Hungarian, a 'Free' Word Order Language", Linguistic Inquiry 12, 185-213. Laijavaara, M. 1988 "Kvantitatiivinen spesies", Virittäjä 4/1988, 469-503. Larson, R. 1988 "On the Double Object Construction", Linguistic Inquiry 18, 369-411. Larson, R. 1990 "Double objects Revisited: Reply to Jackendoff', Linguistic Inquiry 21,589632. Maling J. 1991

"The Hierarchical Assignment of Grammatical Cases in Finnish", in: Holmberg — Nikanne (eds.).

Mitchell, E. 1991 "Evidence from Finnish for Pollock's Theory of IP", Linguistic Inquiry 22, 373-379. Nikanne, U. 1988 On Case Marking in Finnish. Ms., University of Helsinki and Academy of Finland.

456

Notes on Movement to the Spec(IP) Position in Finnish

Nikanne, U. 1989a "Kielioppi, syntaksi ja pragmatikka", The 1989 Yearbook of the Linguistic Society of Finland, 177-196. Nikanne, U. 1989b Infinitiivien morfosyntaksia. Ms., University of Helsinki and Academy of Finland. Nikanne, U. 1990 Ζones and Tier: A Study of Thematic Structure. Helsinki: Finnish Literature Society. Nikanne, U. 1991 "On Assigning Semantic Cases in Finnish", in: Holmberg — Nikanne (eds.). Pica, P. 1987

"On the Nature of the Reflexivization Cycle. Ms., Pennsylvania State University, University Park.

Pollock, J.-Y. 1989 "Verb Movement, Universal Grammar, and the Structure of IP', Linguistic Inquiry 20, 365-424. Reime, Η. 1986 On the Notion 'subject of and its Status in the Theory of Grammar. Ms. Reime, Η. 1991 "Accusative Marking in Finnish", in: Holmberg — Nikanne (eds.). Riemsdijk, H. van — E. Williams 1981 "NP Structure", The Linguistic Review 1, 171-217. SK

= Suomen Kwalehti. a weekly magazine.

Taraldsen, T. 1986 "Nominative objects in Finnish", in: Muysken, P. — H. van Riemsdijk (eds.), Features and Projections. Dordrecht: Foris. Timberlake, A. 1974 The Nominative Object in Slavic, Baltic, and West Finnic, Slavistische Beiträge 82. München: Verlag Otto Sagner. Timberlake, A. 1975 "The Nominative Object in Finnish", Lingua 35, 205-230.

Urpo Nikanne 457

Toivainen, J. 1985 Partitiivin syntaksi, Dept. of Finnish Language and General Linguistics, University of Turku, Publications 25. Vainikka, A. 1988 The Finnish partitive: a reply to Belletti. Ms., University of Massachusetts, Amherst. Vainikka, A. 1989 Deriving Syntactic Representations in Finnish. Doctoral dissertation, University of Masachusetts, Amherst. Vainikka, A. 1991

"The Three Structural Cases in Finnish", in: Holmberg — Nikanne (eds.).

Vilkuna, M. 1989 Wiik, K. 1974

Free Word Order in Finnish: Its Syntax and Discourse Functions. Helsinki: Finnish Literature Society. Suomen eksistentiaalilauseiden "subjekti", Dept. of Phonetics, University of Turku, Publications 13.

Williams, E. 1984 "Grammatical Relations", Linguistic Inquiry 15, 639-674.

Against the A/A'-movement dichotomy* Ayumi Ueyama

1. The issue It is often pointed out in recent works that a certain type of movement cannot be classified as either an Α-movement or an A'-movement. The existence of such movements forces us to reconsider the status of the dichotomy of A- and A*-movement. This dichotomy presumes that movement properties should be determined only with respect to landing site. If this presumption is correct, a movement to Α-position should not be expected to share properties with any movement to A'-position. This prediction, however, does not prove to be correct: such movements as Scrambling in Japanese1 have been argued to have properties of both A- and A'-movement. Therefore, the landing site is merely one factor which determines the movement properties. This paper argues that the most significant distinction to be made is the one between operator movements and uniform movements ("uniform movement" is a cover-term for non-operator movements. See section 3.1.). Scrambling is a uniform movement. Since a typical Α-movement is also a uniform movement, Scrambling shares some properties with it. Nevertheless, at the same time, Scrambling differs from a typical Α-movement. I propose that the crucial factor is the Case-position: the Case is assigned to the S-structure position of a typical Α-movement chain, whereas it is assigned to the D-structure position of a Scrambling chain. The overall picture of the typology is as follows: (1)

Typology of movement: operator movement (e.g. Topicalization in English) uniform movement

460 Ayumi Ueyama

argument movement f

I

Case-driven movement I (e.g. Raising) non-Case-driven movement I (e.g. Scrambling) non-argument movement head movement

Thus, a movement which otherwise has had no place in the theory becomes legitimate under this approach. This paper is organized as follows. Section 2 is a brief review to show that scrambling has properties of both A- and A'-movement: the scrambled NP can be a binder of an anaphor, and at the same time, Scrambling can be reconstructed at LF. Section 3 presents the analysis where this apparent contradiction is resolved. Section 4 shows that only when we consider Scrambling as a uniform movement, can a restriction on long distance Scrambling be predicted: because it is a uniform movement, it becomes ungrammatical when one of its intermediate traces violates ECP. This analysis also supports the clause structure proposed in Ueyama (1989). Section 5 presents the concluding remarks.

2.

Properties of scrambling in Japanese

Scrambling in Japanese is one of the movements which shows properties of both A- and A*-movement. Formerly, it has been claimed as in Saito (1985) that Scrambling is an A'-movement because it shares properties with Topicalization in English and also because they are both IP-adjunctions. Recently, however, the status of Scrambling is becoming controversial. For example, Saito (1990) claims as follows: (2) a. Clause internal Scrambling can be an A- or A'-movement, b. Long distance Scrambling can only be an A*-movement. In order to determine the status of Scrambling, it is necessary to know what kind of properties it has. This section will briefly describe these properties.

Against the A/A'-movement dichotomy

2.1.

461

Anaphor binding

To begin with, a Scrambled NP can be a binder of an anaphor, as is pointed out in Saito (1990): (3)a.

7* [Masao-ga] [otagairno sensee -ni] Masao-NOM e.o.-GEN teacher-DAT

[kareraro] they-ACC

syookaisita (koto)1 introduced COMP 'Masao introduced them; to each other's teachers.' b.

?[Kareraro] they-ACC

[Masao-ga] Masao-NOM

[otagairno e.o.-GEN

sensee -ni] teacher-DAT

syookaisita (koto) introduced COMP "Them;, Masao introduced t; to each other;'s teachers.' This example shows that Scrambling prevents a sentence from violating the Binding Condition A. This fact implies that the Scrambled NP is in an Α-position. Therefore, this is undoubtedly an "Α-movement" property. Saito (1990) claims that the situation is different in the case of long distance Scrambling, referring to the following examples: (4)

*[Karera(-o they-ACC

[Masao-ga Masao-NOM

[otagairno e.o.-GEN

sensee]-ni teacher-DAT

[Hanako-ga t( hihansita to] itta] (koto) Hanako-NOM criticized COMP said COMP 'Them,, Masao said to each other/s teachers that Hanako criticized t . '

462 Ayumi Ueyama

(5)

[Kareraro they-ACC

[otagait-no e.o.-GEN

sensee -ga [Hanako-ga f, teacher-NOM Hanako-NOM

hihansita to] omotteiru] towa. criticized COMP think COMP 'Them;, each other's teachers think that Hanako criticized It is true that the acceptability of (4) or (5) does not improve although karera{ c-commands otagaiit but we should not conclude from these examples that the Scrambled NPs are not in an A-position, because both sentences can be excluded for independent reasons. In (4), the existence of a GOAL argument (-ni phrase) makes the long distance Scrambling unacceptable.2 Whatever the reason, the existence of an anaphor is not relevant. (6)

*[Karerai-o they-ACC

[Masao-ga Masao-NOM

sensee -ni teacher-DAT

[Hanako-ga i, Hanako-NOM

hihansita to] itta] towa'. criticized COMP said COMP "TheΓη, Masao said to the teacher that Hanako criticized V In (5), the meanings of karera-o 'them' and otagai 'each other' are incompatible: while the antecedent of otagai 'each other* is required to refer to the whole group, karera-o 'them' in this example is intended to refer to a part of the group. Thus, we can conclude that a long distance Scrambled NP can be a binder of an anaphor, just as a clause-internally Scrambled NP can: Scrambling appears to be an "Α-movement" with respect to anaphor binding.

2.2.

Reconstruction

Let us now turn to the discussion of reconstructability. Reconstruction means moving a phrase back to one of its trace positions at LF. Reconstructability is often regarded as being an "A*-movement" property (cf. Huang 1990a; Mahajan 1990; and so on,) because a

Against the A/A '-movement dichotomy 463

typical Α-movement such as raising is certainly not capable of being reconstructed. (7)a. b.

*It seems to himt that friends of Johnt are honest. Friends of Johnt seem to himj to be honest.

If the raised NP in (7b) is reconstructed at LF, the sentence should be excluded by Binding Condition C just as in (7a), but of course (7b) is a grammatical sentence. Thus, the A/A'-movement dichotomy has compelled reconstructability to be an "A'-movement" property, simply because of its not being an "Α-movement" property. Saito (1990) claims that Scrambling in Japanese can be reconstructed. (8)a.

Masao-ga Masao-NOM

[Hanako-ga Hanako-NOM

tosyokan-kara karidasita library-from checked-out b.

dono hon-o which book-ACC ka] siritagatteiru (koto) Q want-to-know COMP

IDono hon-o Masao-ga which book-ACC Masao-NOM

[Hanako-ga t Hanako-NOM

tosyokan-kara karidasita ka] siritagatteiru (koto) library-from check out Q want to know COMP 'Masao wants to know which book Hanako checked out from the library.' (9)a.

Masao-ga Masao-NOM

[minna-ga [Hanako-ga all-NOM Hanako-NOM

dono which

hon-o tosyokan-kara karidasita toJ book-ACC library-from checked-out COMP ka] siritagatteiru (koto) Q want-to-know COMP

omotteiru think

464 Ayumi fJeyama b.

ll[Hanako-ga Hanako-NOM karidasita checked-out

dono hon-o tosyokan-kara which book-ACC library-from to] COMP

[Masao-ga Masao-NOM

[minna-ga t all-NOM

omotteiru ka] siritagatteiru] (koto) think Q want-to-know COMP 'Masao wants to know which book everyone thinks that Hanako checked out from the library.' Since a wh-phrase dono hon-o 'which book' is out of the [+wh] clause in (8b) and (9b), we expect the example to be excluded by the Proper Binding Condition3 when the phrase is downgraded into the embedded CP-spec at LF. Nevertheless, (8b) and (9b) are not very bad. This apparent inconsistency is explained if we assume that Scrambling can be reconstructed: if the Scrambled phrase is first reconstructed to its D-structure position and then moves into the [+wh] CP-spec, the Proper Binding Condition will not be violated in any way. Thus, if we consider reconstructability to be an A'-movement property, Scrambling appears to be an "A'-movement" in this respect.

2.3.

Summary

This section has shown that Scrambling has the following two properties. (10)a. The scrambled NP can be a binder of an anaphor. b. Scrambling can be reconstructed at LF. These two properties are contradictory, in a sense, since (10a) is an "Α-movement" property and (10b) is assumed to be an "A'-movement" property. Under the A/A' dichotomy, these properties indicate that Scrambling would be neither A-movement nor A'-movement. This is a serious problem in the current framework. If Scrambling is neither Anor A'-movement, its properties cannot be explained by the principles, which is definitely an undesirable situation.

Against the A/A '-movement dichotomy 465

This problem forces us to argue that the dichotomy of A/A'-movement should be abandoned. The movement properties cannot be divided clearly into two groups. Therefore, we must set up other factors so that each property may be independently derived from different factors. The analysis of this approach as shown below allows us to explain the properties of Scrambling in a principled way.

3.

Proposals

3.1.

The characteristics of the scrambling chain

Our argument is based upon the following principle, which Chomsky has proposed in a recent lecture (cited in Huang (1990b)): (11)

With the exception of a minimal {Op, vbl} chain, every legitimate chain must be uniform.

This principle suggests that we should distinguish operator movements from others. In the following, the term "uniform movement" is used as a cover-term for non-operator movements. Using these terms, characteristics of typical A- and A'-movements of an argument can be expressed as follows: (12)

a typical Α-movement of an argument: a. a uniform movement b. Case-driven

(13)

a typical A'-movement of an argument: a. an operator movement b. non-Case-driven

Although the pairs in (12) and (13) are often stipulated as fixed, this assumption need not be correct. Instead, we can consider the third possibility: (14)a. a uniform movement b. non-Case-driven

466 Ayumi Ueyanta

One of the proposals of this paper is to claim that Scrambling in Japanese is of this type. Since it is a uniform movement, the moved NP can be a binder of an anaphor in contrast to being an operator movement. We must address two apparent problems in order to support this proposal: (15)a. Why can Scrambling be reconstructed, while "typical A-movements" cannot? b. How can Scrambling be a uniform movement? These two problems will be discussed in the following two subsections.

3.2.

Reconstructability

Why can Scrambling be reconstructed, while "typical A-movements" cannot? As a matter of fact, reconstructability can be predicted quite straightforwardly from chain characteristics. Consider (16) and (17), which schematically represent a chain formed by raising and Scrambling, respectively: (16)

{ NP;, t\, ... , lt } +Case

(17)

{ NP;, t\, ... , t; } +Case

Reconstruction means moving a phrase back to one of its trace positions at LF. If the head member of a chain is left as a trace after the reconstruction, this trace will violate the Proper Binding Condition3. The head trace of raising cannot be deleted, because this is the position to which Case is assigned, though it is possible in the case of Scrambling. Therefore, Scrambling can be reconstructed, but raising cannot. Given this line of argument, an operator movement does not always have to be reconstructed either. Since intermediate traces of an operator movement are deleted at LF, a chain formed by an operator movement looks like (18):

Against the A/A '-movement dichotomy

(18)

467

{ 0 P i , vblj }

If reconstruction occurs in a chain like (18), the operator-variable relation may be lost at LF. Therefore, except for the cases of pied-piping, we expect that operator movement cannot be reconstructed. Concerning this point, Saito (1990) offers a similar argument. He argues that Scrambling can be reconstructed because it is a movement which does not affect the interpretation, while Topicalization in English cannot be reconstructed because it must receive operator-variable interpretation at LF. We have shown in (8) and (9) above that the Proper Binding Condition violation is avoided by reconstructing the Scrambled phrase. The Topicalization construction in English, however, does show the Proper Binding Condition violation effects, according to Saito (1990). (19)a.

nWho said that [the man that bought what], John knows whether Mary likes t ? b. *Mary thinks that [the man that bought what], John knows who likes t ?

Although neither sentence in (19) is not fully acceptable because of the wh-island violation, (19b) is much worse than (19a). This difference in acceptability can be explained if we assume that the non-reconstructability of Topicalization causes (19b) to violate the Proper Binding Condition. We have distinguished three types of movement of an argument: (i) operator movement, (ii) uniform movement to a Case-position, and (iii) uniform movement to a non-Case-position. Each has distinct chain characteristics, and the reconstructability is determined with respect to them. As a result, only the movement type (iii) can be reconstructed. Thus, it is explained in a principled way that neither Topicalization nor raising can be reconstructed though it is possible in the case of Scrambling.

3.3.

The chain uniformity

Now let's turn to the second problem: how can Scrambling make a uniform chain? To answer this question, it is necessary to consider the

468 Ayumi Ueyama

conditions under which a chain can be uniform. The notion "uniform" means that every member of the chain shares the relevant features in common. Chomsky proposes "L-position" should be one of those features. The complement and specifier positions of Ν, V, A, P, Agr and Tense are L-positions and all the other positions are non-L-positions. However, if the chain uniformity solely depends on the structural position, Scrambling cannot be a uniform movement. Since multiple Scrambling is allowed, this movement cannot be a substitution, hence not a movement to an L-position. Furthermore, Topicalization in English is also often regarded as an IP-adjunction. If the landing site determines the movement properties, Topicalization should show the same properties as Scrambling. Given that Topicalization cannot be reconstructed, however, its properties are crucially different from Scrambling: we have argued that Topicalization is an operator movement and that Scrambling is a uniform movement. Thus, we should assume that the structural position is not the only factor which is relevant to the movement type. This paper proposes that there are at least two kinds of conditions under which a chain can be uniform: (20)

A chain of an NP is uniform if every member of it is in an L-position.

(21)

A chain of an NP with a case-marker is uniform if the case-marker is licensed at every position.

If a case-marker in an IP-adjoined position is licensed, Scrambling forms a uniform chain in contrast to Topicalization in English. I assume (22) to be the licensing condition for case-markers4. (22)

A case-marker is licensed if it is governed by a Case-assigner. (Actual Case-assigning is not necessary in this licensing.)

Furthermore, INFL is assumed to be a Case-assigner, following Takezawa (1987). Thus, since the IP-adjoined position is governed by INFL, a Case-assigner, a case-marker is licensed in this position. Therefore, an argument with a case-marker can form a uniform chain, even if it is not a movement to an L-position.5

Against the A/A '-movement dichotomy 469

4.

Long distance scrambling

This section shows that the analysis of long distance Scrambling provides supporting evidence for the claim made above: Scrambling is not an operator movement but is a uniform movement. A uniform movement is restricted by ECP more strictly than an operator movement is, since in the case of a uniform movement every trace must be properly governed whereas in the case of an operator movement the intermediate traces need not be properly governed. Therefore, given that Scrambling is a uniform movement, the movement is expected to be more local than an operator movement. Although it is often assumed in the literature that Japanese allows long distance Scrambling, I will show below that long distance Scrambling in a certain construction is unacceptable. I will further show that the unacceptability does not come from a subjacency violation. Since an operator movement is still possible in the same construction, it implies in the current framework that the sentence should be excluded by ECP. Given that Scrambling is a uniform movement, we can provide an appropriate explanation of this fact. The definitions of ECP and barrier are listed below: (23) ECP: Α nonpronominal empty category must be (i) properly head-governed (ii) Theta-governed, or antecedent-governed (Rizzi 1990: 74) (24) (i) (ii) (iii) (iv)

Head Government: X head-governs Y iff X is one of {Α, Ν, Ρ, V, Agr, T} X m-commands Y no barrier intervenes Relativized Minimality is respected. (Rizzi 1990: 6)

(25) (i) (ii) (iii) (iv)

Antecedent Government: X antecedent-governs Y iff X and Y are coindexed X m-commands Y no barrier intervenes Relativized Minimality is respected. (Rizzi 1990: 6)

470 Ayumi Ueyama

(26)

4.1.

Α maximal projection (except IP) which is not L-marked is a barrier.6

Unacceptable cases of long distance scrambling

An NP can be Scrambled over a clause boundary, i.e. long distance Scrambling is possible in Japanese: (27)

Sono that

hon-o John-ga book-ACC John-NOM

[Mary-ga t Mary-NOM

katta bought

to] omotteiru towa. COMP think COMP "That book, John thinks that Mary bought t.' However, the acceptability is affected by the choice of the matrix verb. (28)

*lSono that

hon-o John-ga book-ACC John-NOM

[Mary-ga t Mary-NOM

katta bought

to] kakitometeita towa. COMP noted-down COMP 'That book, John noted down that Mary bought t.' (29)

*lSono that

hon-o John-ga [Mary-ga t book-ACC John-NOM Mary-NOM

katta bought

no-o] wasureta towa. COMP forgot COMP 'That book, John forgot that Mary bought t.' (30)

*lSono that

hon-o John-ga book-ACC John-NOM

[Mary-ga t Mary-NOM

koto-o] kakusiteiru towa. COMP hide COMP "That book, John won't tell that Mary bought t.'

katta bought

Against the A/A '-movement dichotomy 471

Thus, verbs are classified into two groups7: one group (31) allows long distance Scrambling of an embedded element, while the other group (32) does not. (31)

omotteiru 'think', itteiru 'say', sinziteiru 'believe', osieteiru 'teach', kantigaisiteiru 'mistake', and so on.

(32)a. kakitometeita 'noted down', kikituketa 'happened to hear', tutaeta 'informed', and so on. (complementizer = to) b. wasureta 'forgot', oboeteiru 'remember', yorokondeiru 'be pleased', and so on. (complementizer = no-o) c. kakusiteiru 'hide', kakiuiinsiteita 'have confirmed', sitteiru 'know', and so on. (complementizer = koto-o)

4.2.

ECP vs. subjacency

Although the unacceptability of sentences such as (28)-(30) has often been attributed to subjacency violation, such analysis turns out to be inadequate: rather, they should be excluded by ECP. More specifically, one of the intermediate traces must be the offending trace, since it is clear that the initial trace does not violate ECP. Consider the following examples, where an NP is moved over a clause boundary to be interpreted as a FOCUS. Let us call this movement FOCUS Shift, in contrast to Scrambling. (33)

ISono hott-o John-wa that book-ACC John-NOM

[Mary-ga t Mary-NOM

katta bought

to] kakitometeita. COMP note-down 'It is that book that John noted down that Mary bought.'

472 Ayumi Ueyama

(34)

ISono hon-o John-wa that book-ACC John-NOM

[Mary-ga t Mary-NOM

katta bought

no-o] wasureta. COMP forgot 'It is that book that John forgot that Mary bought.' (35)

ISono hon-o John-wa that book-ACC John-NOM

[Mary-ga t Mary-NOM

katta bought

koto-o] kakusiteiru. COMP hide 'It is that book that John won't tell that Mary bought.' Because FOCUS interpretation requires an operator- variable relation at LF, FOCUS Shift must be an operator movement as opposed to Scrambling.8 If subjacency violation were responsible for the unacceptability of long distance Scrambling in (28)-(30), it is expected that FOCUS Shift, an operator movement, would also show the same degree of unacceptability. Since the above sentences are not perfect, it is possible that some kind of subjacency is involved. However, the fact that sentences (28)-(30) are much worse than sentences (33)-(35) indicates that the unacceptability of long distance Scrambling should be attributed not only to subjacency but also to ECP.9 The same type of contrast between an operator movement and a uniform movement is further observed in the case of wh-movement. As is shown below, while an extraction of nani 'what' out of the embedded clause of verb type (32) is allowed, an extraction of naze 'why' is blocked: (36)a. John-wa John-top

[Mary-ga Mary-NOM

nani-o katta what-ACC bought

kakitometeita-no ? note-down-Q 'What did John note down that Mary bought?'

to] COMP

Against the A/A '-movement dichotomy 473

b. *John-wa [Mary-ga John-top Mary-NOM

sono hon-o naze katta that book-ACC why bought

to] kakitometeita-no? COMP note-down-Q 'Why did John note down [that Mary bought that book t]?' (37)a. John-wa John-top

[Mary-ga Mary-NOM

nani-o katta what-ACC bought

no-o] COMP

wasureta-no? forgot-Q 'What did John forget that Mary bought?' b. *John-wa John-top

[Mary-ga sono Mary-NOM that

hon-o naze book-ACC why

katta no-o] wasureta-no? bought COMP forgot-Q 'Why did John forget [that Mary bought that book t ]?' (38)a. John-wa John-top

[Mary-ga Mary-NOM

nani-o katta what-ACC bought

koto-o] COMP

kakusiteiru-no? hide-Q 'What won't John tell that Mary bought?' b. *John-wa [Mary-ga John-top Mary-NOM

sono hon-o naze katta that book-ACC why bought

koto-o] kakusiteiru-no? COMP hide-Q 'Why won't John tell [that Mary bought that book t ]?' The problem is not that naze 'why' cannot take the matrix clause as its scope: the same construction is allowed when the matrix verb is of the type (31):

474 Ayumi Ueyama (39)a. John-wa John-top

[Mary-ga Mary-NOM

nani-o katta what-ACC bought

to] COMP

omotteiru-no? think-Q 'What does John think that Mary bought?' b. John-wa John-top

[Mary-ga Mary-NOM

sono hon-o naze katta that book-ACC why bought

to ] omotteiru-no ? COMP think-Q 'Why does John think [that Mary bought that book t ]?' These examples indicate that a uniform movement cannot take place out of an embedded clause of the verb type (32), while it is possible in the case of the verb type (31). The fact that an operator movement out of the same construction is acceptable implies that the unacceptable sentences are not excluded by the subjacency condition: one of the intermediate traces must be in violation of ECP. The following table summarizes the above observations. (40)

4.3.

matrix verb type

(31)

(32)

uniform movement

OK

*

operator movement

OK

OK

WP-analysis

The claim that long distance Scrambling out of a particular clause is excluded by ECP raises an apparent problem: how can an extraction of a complement out of a complement clause violate ECP? With the structure which is currently assumed in the literature, it appears that no trace can violate ECP in such a movement. Consider the tree diagram in (41), where long distance Scrambling forms a chain {NP;, t'j, t j :

Against the A/A '-movement dichotomy 475

(41) CP

ti Since the part of the chain {t\, /J is included in the chain formed by clause internal Scrambling, t, is not expected to be the offending trace in any case of long distance Scrambling. Since t\ is head-governed by I, the required contrast would be explained if t\ is antecedent-governed when the matrix verb is of type (31) and not antecedent-governed when the matrix verb is of type (32). However, t\ is always antecedent-governed by NPS in the structure (41) regardless of the matrix verb type: because the embedded clause is L-marked by the matrix verb, the CP is not a barrier, and there are no barriers between NP, and t\. This apparent problem is resolved if we assume that there is one more category, say WP, above CP:

476 Ayumi Ueyama

(42) WP

Ii Now that the matrix verb L-marks WP, CP is a candidate as a barrier. If CP is a barrier when the matrix verb is of type (32) and is not when the matrix verb is of type (31), the required contrast can be derived. Let us stipulate as follows: (43)

A verb of type (31) selects a W which is an L-marker.

When the embedded W is an L-marker, t\ is antecedent-governed by NP; because the CP governed by the W is not a barrier, and when the embedded W is not an L-marker, t'{ violates ECP, because the CP blocks the antecedent-government from NPj. This analysis also allows us to explain the contrast in (36)-(39): the contrast between an operator movement of nani 'what' and a uniform non-argument movement of naze 'why*. The intermediate trace of naze is expected to be in the embedded CP-spec, which is a normal landing site for a wh-phrase. It is plausible that the embedded W head-governs the intermediate trace when the W is an L-marker. Then when the W is not an L-marker, it does not head-govern the trace and causes an ECP violation. Thus, under the WP-analysis, we can provide a unified

Against the A/A '-movement dichotomy

477

explanation for long distance Scrambling and the extraction of naze 'why'. The remaining problem is the status of the category W: if there is no other independent evidence, it would be too artificial to set up a new category just because it is useful to account for a single fact. As is argued in Ueyama (1989, 1991a, b), however, various sorts of empirical facts suggest the existence of this category. The situation is that the WP-analysis is not motivated by the fact of long distance Scrambling presented above, but that it adds supporting evidence for setting up the category W. (See the Appendix for other facts supporting the WP-analysis.) Note that the category label W stands for nothing: it is a tentative name, because the properties of this category have not been fully studied yet. What can be said about this category at this stage is (i) that the head W carries features relevant to the clause type, (ii) that the WP-spec position has something to do with a topic wa-phrase, and (iii) that a number of facts can be explained if we assume this category.

5. Conclusion I have argued that Scrambling in Japanese is a non-Case-driven uniform movement. This analysis allows us to explain the properties of Scrambling. First, because Scrambling is a uniform movement, it shares such properties as anaphor binding with a typical "A-movement", which is also a uniform movement. Second, because a Scrambling chain is assigned a Case at its D-structure position, it differs from Case-driven uniform movements: for example, whereas Scrambling is freely reconstructed, raising cannot be reconstructed. Of course, because Scrambling is a uniform movement, it shows properties that are different from an operator movement: for example, Topicalization in English cannot be reconstructed either. Based on this characterization of Scrambling, I have shown that the restriction on long distance Scrambling can be explained in a principled way. Long distance Scrambling is unacceptable when the matrix verb is of a particular type, while FOCUS Shift, an operator movement, is acceptable even if the matrix verb is of the same type. This contrast cannot be accounted for unless we regard Scrambling as a uniform movement.

478 Ayumi Ueyama

Two important claims have been made in the course of this argument. First, it was argued that chain uniformity depends upon the form of the head NP: when the head is a bare NP, every member of a uniform chain should be in an L-position; when the head is an NP with a case-marker, every member of a uniform chain should be in a position where a case-marker is licensed. Thus, an IP-adjunction is a uniform movement in Japanese but an operator movement in English. Second, it was suggested that WP-analysis is required to account for the cases of long distance Scrambling. In WP-analysis, every CP is dominated by another category, WP, in a clause: the head of a clause is W, not C. While further considerations are required, it is significant to note that desirable consequences are derived from this analysis.

Appendix: facts suggesting the existence of WP (partly excerpted from Ueyama (1989)) This Appendix lists empirical facts supporting the WP-analysis. 1. Double "COMP" phenomena It is sometimes assumed that to is a [-wh] COMP and that ka is a [+wh] COMP. However, these two "COMPs" can cooccur. As is argued in Fukui (1986), this means that at least one of the two is not a head of CP. (l)a.

Sono the

hito-wa man-top

[syokudoo-wa dining-room-top

doko-desu ka where-cop. Q

to] kikimasita. COMP asked 'The man asked (us) where the dining room is.' b.

Boku-wa I-top

[dare-ga who-NOM

hazime-ni first

kuru ka to] come Q COMP

wakuwakusiteita. was-excited Ί was excited wondering who would come first.'

Against the A/A '-movement dichotomy 479

WP-analysis distinguishes two head positions. Thus, to is W and ka is C in the sentences above. The cooccurence of two "COMPs" is accounted for in WP-analysis without causing any problems. 2. Selection of W form The form of W is selected by the higher verb. Therefore, it is not appropriate to assume such words as to to be just particles. (2)a.

Kare-wa [hahaoya-ga Kyoto-ni sundeiru he-top mother-NOM Kyoto-in live to/*no-o/?*koto-o] itta. COMP said 'He said that his mother lived in Kyoto.'

b.

John-wa John-top

[Mary-ga Mary-NOM

gakusei-dearu *to/no-o/koto-o] student-cop. COMP

wasureteita. forgot 'John did not remember that Mary was a student.' As has sometimes been pointed out, this "COMP" form is correlated with such notions as "factivity" or "assertivity", which are likely to be the features carried by W. 3. Distribution of topic Wa-phrases and major subject Ga-phrases Generally speaking, a topic wa-phrase is excluded in an embedded clause whose W is realized by no or koto, while it can occur in a clause whose W is realized by to. The ungrammatical sentences below are changed into grammatical ones if ga is attached to John instead of wa. (3)a.

b.

[John-wa katta to] John-top won COMP Ί think that John won.'

omou. think

*[John-wa katta no-wa] mezurasii. John-top won COMP rare 'It is rare that John won.'

480 Ayumi Ueyama

c.

*Boku-wa [John-wa katta koto-oj wasureteita. I-top John-top won COMP forgot Ί did not remember that John had won.'

Furthermore, a FOCUS interpretation is obligatory for a major subject ga-phrase in an embedded clause whose W is realized by to, though such an interpretation is optional in a clause where W is realized by no or koto. (4)a.

b.

[New York-ga biru-ga ooi to] omou. New York-nom building-NOM many COMP think Ί think that it is New York where there are many buildings.' [New York-ga biru-ga New York-NOM building-NOM

ooi many

no-wa] COMP

toozen-desu. natural-cop. 'It is natural that there are many buildings in New York.' c.

[New York-ga biru-ga ooi New York-NOM building-NOM

koto-o] many COMP

sitteimasu ka? know Q 'Do you know that there are many buildings in New York?' Thus, there are several phenomena affected by the W form. 4. Pure topic Wa-phrases Most sentences with a topic ννα-phrase can be paraphrased into a sentence without a topic phrase: it is possible to replace wa with some other markers in most cases. However, a certain type of wa-phrase cannot be a non-topic phrase. (5) a. Kore-wa/*-ga/*-o miti-o matigaeta. this-top/-NOM/-ACC way-ACC mistook '(Considered from this situation,) I think I lost the way.'

Against the A/A'-movement dichotomy 481

b.

Sore-wa/*-ga/*-o ukkarisiteita. it-top/-NOM/-ACC careless Ί was careless (as for it).'

c.

Sore-wa/*-ga/*-o kimi-ga it-top/-NOM/-ACC you-NOM '(As for it,) you should go.'

ik-eba ii. go-if good

If it is assumed that this sort of topic-phrase is base-generated in WP-spec, the facts are naturally accounted for: it can neither be a ga-phrase nor an o-phrase because WP-spec is out of the domain where ga-case or o-case is assigned.

Notes *

Early versions of this paper were presented at the Tilburg Workshop on Scrambling in the fall of 1990, and at the meeting of the Kansai Association of Theoretical Linguistics in January 1991. I would like to thank the following people for their helpful comments and discussion at various stages of its preparation: Noam Chomsky, Nobuko Hasegawa, James Huang, Takashi Imai, Yoshihisa Kitagawa, Anoop Mahajan, Shigeru Miyagawa, Heizo Nakajima, Mamoru Saito, Koichi Takezawa, Yukinori Takubo, and Koichi Tateishi. This work was supported in part by Japan Society for the Promotion of Science under Grant No.02951125. The following abbreviations are used in the morpheme-by-morpheme glosses: acc = accusative marker COMP = so-called complementizer, which is to be a W in WP-analysis cop. = copula dat = dative marker e.o. = "each other' gen = genitive marker nom = nominative marker Q = question marker, which is to be a C in WP-analysis top = topic marker

1. See Hoji (1985), Kuroda (1983), Saito (1985,1986,1989,1990), Whitman (1982) for discussions on Japanese Scrambling. Note that "Scrambling" in this paper refers to a construction in which word order is changed without causing any FOCUS interpretation. In contrast, the word order change with FOCUS interpretation is called FOCUS Shift. As is argued in Chomsky (1976), a movement with FOCUS interpretation should be strictly distinguished from a movement without FOCUS interpretation: the former should be regarded as an operator movement while the latter should not. Although it is often assumed that Scrambling is accompanied by FOCUS interpretation, especially in the case of long distance Scrambling, the FOCUS

482 Ayumi Ueyama

interpretation can be detached from Scrambling if we carefully judge the sentence with an appropriate clause type marker. I have used the clause type marker towa for this purpose; because FOCUS interpretation is usually suppressed in a towa clause. Although the typical clause type marker is koto in the literature, koto is not used in this paper except for the quoted examples, because some people tend to permit FOCUS interpretation in a koto clause. Note that the meaning of the clause type marker is not represented in the translation. See Ueyama (1991) for contrasts between Focus Shift and Scrambling. 2. Long distance Scrambling is unacceptable in this sentence, although the verb itta 'say' permits long distance Scrambling in most cases. It appears that the existence of a GOAL argument sensee-ni 'to the teacher' affects the acceptability. The same kind of effect is also observed in the following pair. (i)

Sono that

hon-o John-ga book-ACC John-NOM

[Mary-ga t Mary-NOM

katta bought

to] COMP

itteiru towa. say COMP 'That book, John says that Maiy bought t.' (ii)

? *Sono that

hon-o John-ga book-ACC John-NOM

Bill-ni Bill-DAT

[Mary-ga t Mary-NOM

katta to] itteiru towa bought COMP say COMP 'That book, John says to Bill that Mary bought t.' 3. Proper Binding Condition demands that a trace must be bound by its antecedent at every level of the derivation. For example, this principle is responsible for excluding the following sentences. (Examples are quoted from Saito (1990).) (i)

Mary ordered John to find out [who Bill saw t ]. *Mary ordered t to find out [who Bill saw John],

(ii)

nWhot do you wonder [which picture of ] John likes t? *[Which picture of ] do you wonder whot John likes t?

The principle is also responsible for the trace created at LF. (iii) Mary ordered John to find out [who t saw who], *Mary ordered who to find out [who t saw John]. And the trace created by Scrambling in Japanese is also subject to this principle. (iv)

Hanako-ga Hanako-NOM

Masao-ni Masao-DAT

[[dare-ga who-NOM

kuru] come

ka] Q

osieta (koto). taught COMP

Against the A/A '-movement dichotomy 483

'Hanako told Masao [ Q [who is coming]].' (v)

*Hanako-ga dare-ni [[Masao-ga kuru] Hanako-nom who-DAT Masao-NOM come 'Hanako told who [ Q [Masao is coming]].'

(vi)

*[Hanako-ga tx Hanako-NOM itla]] said

yonda to]j [sono read COMP that

ka] Q

osieta (koto). taught COMP

hon-oj [Taro-ga tj book-ACC Taro-NOM

(koto) COMP

4. See Saito (1985), Kuroda (1988), Miyagawa (1990), and Takezawa (1987) for the discussion on case-marker realization. 5. This approach bears some similarity to the visibility conditions of Mahajan (1990: 69) in that two conditions are distinguished: "an LF visibility condition that requires all NPs (or chains) to bear a structural Case and an S-structure visibility condition that requires all NPs (or chains) to have a Case (inherent or structural)." Saito (1990) proposes another approach which explains the difference in status of IP-adjoined positions in Japanese and English. He suggests a licensing condition of an operator in terms of a new notion called D/D'-position: "A D-position is a position in which an NP can appear at D-structure and can be licensed as uniform. A D'-position is one that is not a D-position (Saito (1990: (41).) According to this approach, an IP-adjoined position in Japanese is a D-position because a major subject is base-generated in this position, while in English it is a D'-position because English lacks a "major subject construction." Although the consequence of this approach is not so different from the analysis claimed here, the notion of "D/D'-position" presents a conceptual problem. As Noam Chomsky (p.c.) points out, the notion "A/A'-position" is being abandoned because it presupposes some unstated notion of structural equivalence among phrase-markers, as is implicit in the phrase "potential theta position." Such phrases as "can appear" or "can be licensed" in the definition of D/D'-position above indicate that this notion also has the same kind of problem as that of A/A'-position. See also Mahajan (1990) for discussions of the notions "A/A'-position" and "L-position." 6. It is assumed that VP is not a barrier, as in Rizzi (1990) and Huang (1990b). 7. Virtually the situation is a little more complex. The possibility of long distance Scrambling is affected not only by the verb type, but also by its argument structure (cf.note 2) and the tense/aspect of the verb. It is for this reason that each verb cited in (33) and (34) is inflected. 8. Because FOCUS Shift is an operator movement, its property is very different from that of Scrambling, which is a uniform movement. Therefore, if the fronted NP is focussed, the judgement reverses. See Ueyama (1991a) for this contrast. Presumably

484 Ayumi Ueyama

the reason why the judgement of Scrambling sometimes differs among people can be attributed to the ambiguity between Scrambling and FOCUS Shift. I am indebted to Yoshihisa Kitagawa for this point. 9. One possible problem of this analysis is that the unacceptability of the sentences are not so serious as typical ECP violations. I cannot account for this point at this time.

References Chomsky, N. 1976 "Conditions on Rules of Grammar", Essays on Forms and Interpretation. Amsterdam: North-Holland. Chomsky, N. 1981

Lectures on Government and Binding. Dordrecht: Foris.

Chomsky, N. 1986a Knowledge of Language, its Nature, Origin, and Use. New York: Praeger. Chomsky, N. 1986b Barriers. Cambridge: the ΜΓΓ Press. Chomsky, N. 1991 "Some Notes on Economy of Derivation and Representation", in: R. Freidin (ed.), Principles and Parameters in Comparative Grammar. Cambridge: the ΜΓΓ Press. Diesing, M. 1990 "Verb Movement and the Subject Position in Yiddish", Natural Language and Linguistic Theory 8-1, 41-80. Fukui, N. 1986

A Theory of Category Projection and Its Applications. Doctoral dissertation, Massachusetts Institute of Technology, Cambridge.

Huang, C.-T.J. 1990a Reconstruction and the Structure of VP: Some Theoretical Consequences. Ms, Cornell University. Huang, C.-T.J. 1990b Current Issues in the Theory of Bounding and Government. A talk presented at Tokyo Linguistic Forum, 8/31/1990.

Against the A/A '-movement dichotomy 485

Kuroda, S.-Y. 1983 "What can Japanese Say about Government and Binding", WCCFL 2. Kuroda, S.-Y. 1988 "Whether We Agree or Not", Linguisticae Investigations

12, 1-47.

Mahajan, A. 1990 The A/A-bar Distinction and Movement Theory. Doctoral dissertation, Massachusetts Institute of Technology, Cambridge. Miyagawa, S. 1990 Functional Category, Case Realization, and Scrambling. A talk presented at Kansai Association of Theoretical Linguistics, July/7/1990, Osaka University. Rizzi, L. 1990 Saito, M. 1985

Saito, M. 1986

Saito, M. 1989

Saito, M. 1990

Relativized Minimality. Cambridge: the ΜΓΓ Press.

Some Asymmetries in Japanese and Their Theoretical Implications. Doctoral dissertation, Massachusetts Institute of Technology, Cambridge.

"Three Notes on Syntactic Movement in Japanese", in: T. Imai — M. Saito (eds.), Issues in Japanese Linguistics. Dordrecht: Foris Publications.

"Scrambling as Semantically vacuous A'-movement", in: M. Baltin — Α. Kroch (eds.), Alternative Conceptions of Phrase Structure. University of Chicago Press, 182-200.

Long Distance Scrambling in Japanese. Ms. University of Connecticut.

Saito, Μ. — H. Hoji 1983 "Weak Crossover and Move alpha in Japanese", Natural Language and Linguistic Theory 1, 245-259. Takezawa, K. 1987 A Configurational Approach to Case-Marking in Japanese. Doctoral dissertion, University of Washington. Ueyama, A. 1989 "FOCUS-no ga to nihongo-no kukoozoo (Ga-phrase with FOCUS interpretation and the phrase structure of Japanese)", The Kansai Linguistic Society, vol.9, 1-15.

486 Ayumi Ueyama

Ueyama, Α. 1991a Scrambling and the FOCUS Interpretation, Rochester Workshop on Japanese Linguistics, Universal Grammar and Their Implications to Language Pedagogy and Human Cognition, University of Rocheste. Ueyama, A. 1991b On the Licensing of Functional Heads: the WP-Analysis of Interrogative Clauses. Ms. Kyoto University of Foreign Studies. Whitman, J. 1986 "Configurationality parameters", in: T.Imai — M. Saito, (eds.), Issues in Japanese Linguistics. Dordrecht: Foris Publication.

Scandinavian object shift and West Germanic scrambling Sten Vikner

1. Introduction This paper will discuss the differences between two types of object movement in the Germanic languages. Both are clause internal, and both are movements from right to left. One kind is found in Danish, Faroese, Icelandic, Norwegian, and Swedish (i.e. the Germanic SVO languages except English), and this (and only this) will be referred to below as "object shift", following Holmberg (1986: 165). A different kind of leftwards object movement is the one found in Afrikaans, Dutch, Flemish, Frisian, (High) German, Swiss German, and Yiddish (i.e. the Germanic SOV languages), and this (and only this) I shall call "scrambling" in what follows.1 Object shift and scrambling are both adjunctions to VP (in some languages, e.g. in German, scrambling may also adjoin to IP, but not in others, e.g. in Dutch, cf. the appendix). Examples may therefore be constructed in which the two processes look identical: (1) Object shift Danish

I gir

Leeste hau [ j

II