Triggers 9783110197433, 9783110181395

Table of contents :
Frontmatter
Contents
The Role of Triggers in Linguistic Theory: Some Introductory Remarks
Snowballing Movement and Generalized Pied-piping
Optionality at the Interface: Triggering Focus in Romanian
How N-words Move: Bipartite Negation and ‘Split-NegP’
The Agreement Parameter
Some Notes on Emphatic Forms and Displacement in Dutch
Scrambling, Optionality and Non-Lexical Triggers
Phonological Content and Syntactic Visibility
On Triggers of Movement and Effects at the Interfaces
Scope Marking Constructions in Dayal-type Indirect Dependency
Hyperbaton and Haplology
On Scrambling as Defocusing in German and West Germanic
Linear Compression as a Trigger for Movement
A Case for Head Movement at PF: SAI in Comparatives
The EPP in Breton: An Unvalued Categorial Feature
Backmatter

Citation preview

Triggers

≥

Studies in Generative Grammar 75

Editors

Henk van Riemsdijk Harry van der Hulst Jan Koster

Mouton de Gruyter Berlin · New York

Triggers

Edited by

Anne Breitbarth Henk van Riemsdijk

Mouton de Gruyter Berlin · New York

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

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

앝 Printed on acid-free paper which falls within the guidelines 앪 of the ANSI to ensure permanence and durability.

ISBN 3-11-018139-8 Bibliographic information published by Die Deutsche Bibliothek Die Deutsche Bibliothek lists this publication in the Deutsche Nationalbibliografie; detailed bibliographic data is available in the Internet at ⬍http://dnb.ddb.de⬎.

쑔 Copyright 2004 by Walter de Gruyter GmbH & Co. KG, 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. Cover design: Christopher Schneider, Berlin. Printed in Germany.

Contents

The Role of Triggers in Linguistic Theory: Some Introductory Remarks Anne Breitbarth and Henk van Riemsdijk

1

Snowballing Movement and Generalized Pied-piping Enoch O. Aboh

15

Optionality at the Interface: Triggering Focus in Romanian Gabriela Alboiu

49

How N-words Move: Bipartite Negation and ‘Split-NegP’ Arthur Bell

77

The Agreement Parameter Chris Collins

115

Some Notes on Emphatic Forms and Displacement in Dutch Norbert Corver

137

Scrambling, Optionality and Non-Lexical Triggers Roland Hinterhölzl

173

Phonological Content and Syntactic Visibility Ruriko Kawashima and Hisatsugu Kitahara

205

On Triggers of Movement and Effects at the Interfaces Mariana Lambova

231

Scope Marking Constructions in Dayal-type Indirect Dependency Anikó Lipták

259

Hyperbaton and Haplology Eric Mathieu

293

vi

Contents

On Scrambling as Defocusing in German and West Germanic László Molnárfi

331

Linear Compression as a Trigger for Movement Andrea Moro

387

A Case for Head Movement at PF: SAI in Comparatives Fumikazu Niinuma and Myung-Kwan Park

431

The EPP in Breton: An Unvalued Categorial Feature Milan Rezac

451

Index

493

The Role of Triggers in Linguistic Theory: Some Introductory Remarks Anne Breitbarth and Henk van Riemsdijk

The present volume deals with the issue of “triggers”, that is, requirements of some sort that cause syntactic effects, most notably displacement.1 1. The Concept of a Trigger Phrase structure and displacement are prominent universal properties of natural language. While some approaches have tried to eliminate transformational operations, displacement continues to play a crucial role in derivational theories such as Minimalism.2 Concentrating on displacement, we can ask ourselves two different questions: first, why does it exist in human language and second, how is it implemented? The latter question has been discussed frequently but not always exhaustively. Chomsky (1986) proposed that movement is governed by a condition of Last Resort: Move-_ is applied only when failure to apply it will lead to a structure that violates general conditions and causes a derivation to crash. For example, the Case feature of a DP would act as a trigger for the movement of the DP, which, unmoved, would violate the Case Filter.3 (1)

Case Filter *NP if NP has phonetic content and has no Case.

The Case Filter can therefore be seen as one of the first ‘triggers’ for displacement in generative theory before the term itself was coined. Assumptions like the triggering effect of Case raise the very general question: what are the possible triggers of displacement? Pursuing the above line of inquiry, Chomsky (1995, 2001a) proposes more generally that formal/morphosyntactic features are required to implement displacement. These features include the  -features (gender, person, number) of Infl and v, the Case features of a DP, and the EPP feature of Comp, Infl, and v, but also ‘peripheral’ features such as [+wh] or [topic]/[focus].

2

Anne Breitbarth and Henk van Riemsdijk

The  -features of Infl and v attract DP subjects to their specifiers, but this attraction would only ‘trigger’ actual, that is overt, displacement in case the the Infl or v head also has an EPP feature. In other words, an EPP feature requires – triggers – the physical/overt displacement of phrasal material to the specifier positions of the heads possessing the feature. The similarity of this general process to the requirement thatpredicational entity must take a subject lead to the name EPP feature, after the older Extended Projection Principle.4 The idea is that such features as  -, [+wh]- or EPP-features can be uninterpretable on certain categories (like Case on DP). Uninterpretability is regarded an unavoidable imperfection of natural language – a ‘perfect’ system being able to generate fully interpretable and pronounceable utterances without uninterpretability and therefore displacement triggered by it. Therefore such features have to be eliminated (‘checked’) before the derivation reaches an interface (PF or LF) at which they would be uninterpretable. This necessity of checking uninterpretable features before the interfaces follows from a general restriction on what counts as legitimate objects at the interfaces – the Principle of Full Interpretation (FI).5 FI basically states that there must be no superfluous symbols (that is, uninterpretable features) in the output. If there are still some unchecked features left on reaching the interface, the derivation will not converge (that is, it will ‘crash’) at the relevant interface. But apart from such cases of displacement that are triggered by the caseagreement and periphery systems (corresponding to the distinction between A-movement and A-bar-movement)6, there are several other types of ‘movement’ that do not seem to be similarly triggered by morphosyntactic ‘imperfections’ (that is, uninterpretable features) in any obvious way. The most obvious such phenomenon is covert displacement. Does it exist or not? It is clear that not all languages show the same (visible) displacements in parallel cases. Think of wh-movement: some languages like Chinese have no (visible) wh-movement at all (wh-in-situ), some have simplex but complete wh-movement like English, others like German or Hungarian allow partial wh-movement and some like Bulgarian even permit multiple wh-movement. Assuming, as is generally accepted, the reality of a universal grammar (UG) with the same general principles of phrase structure building holding for all languages, supplemented by a formal (logical) interpretation algorithm uniform for all human languages, we are forced to assume that in all languages, there should be movement of operators like wh-elements to their scope positions – at least up to the point where structures are being interpreted. This leads to the assumption of covert displacement for some languages.

The Role of Triggers in Linguistic Theory: Some Introductory Remarks

3

Another phenomenon of this type, that is, displacement that is caused neither by morphosyntactic nor by peripheral triggers, is differences in quantifier scope. Assuming that quantifier raising (QR)7 is a core operation in natural language (quantifiers need to take their scope at least at the point when the structure is interpreted), we have to ask how it is triggered. Is it a (non-local) interface property or some local semantic feature which forces covert action at the relevant phase? A very intriguing problem is the question of optionality in displacement as in the case of scrambling. A triggers-centered theory of displacement such as Minimalism (Chomsky 1995, 2001a) has problems expressing optionality as in such a model it is required that all displacement be triggered. In such a theory, as soon as there is a triggering feature for it, displacement is forced to apply, barring optional inertia. Some have therefore claimed that scrambled word orders are base-generated and not generated by Move-_.8 Another case in point are phenomena like stylistic fronting in Scandinavian9 and other cases of inversion which cannot readily be explained in terms of formal features. As some authors in this volume (e.g. Alboiu, Lambova) show, however, some cases of apparent optionality can still be shown to be non-optional, thus confirming the ‘traditional’ conception of triggers. It is an empirical question whether all possible triggers for word order phenomena are morphosyntactic in nature. As just discussed, there are a number of cases for which this is not very likely. Often, extra-syntactic factors such as discourse or information structure (LF interface conditions) or prosodic properties (PF interface conditions) seem to play an important role as well. One possibility of including these factors in the set of triggers is to assign the status of morphosyntactic features to them and to treat them just like traditional features. An example would be the extended left periphery (split CP) as argued for by Rizzi (1997). Some of the papers in this volume (Alboiu, Hinterhölzl, Molnárfi) adopt a similar strategy by putting discourse and information structure requirements into narrow syntax, assimilating them to various extents with ‘regular’ morphosyntactic features. As an alternative to the feature elimination requirement (‘drag chains’), it is possible that output conditions require the presence or absence of a certain element in certain context (‘push chains’).10 There is some diversity in the different approaches regarding the locus of such influences, whether they apply within narrow syntax, at the interfaces, or even beyond. Syntax-internal conditions of this kind could be certain operators requiring local relationships with their operands. Interface conditions might for example require that operations apply in order to make a structure (more easily) pronounceable, without these operations being necessary for morpho-

4

Anne Breitbarth and Henk van Riemsdijk

syntactic reasons. Grammar external conditions could be shaped by general cognitive or communicative principles. Under such a ‘push chain’ approach as well, the problem of determinism vs. optionality would again have to be taken into account. Summing up, the following questions arise: 1. are all movement operations triggered? This question extends in particular to various movements needed to implement the kind of analyses required by Kayne’s (1994) Linear Correspondence Axiom (LCA).11 2. are all triggers for displacement operations of a uniform nature or does UG permit a range of different mechanisms? 3. can prosodic properties of a phrase, requirements of discourse or information structure and quantifier scope or other semantic properties trigger movement?12 4. is external Merge also subject to triggering and if so, how do these triggers differ from the ones found operative in internal Merge, that is, displacement? The fourteen contributions to this volume propose answers to some of these questions and they address a number of the empirical phenomena listed above. In essence, four groups of triggers are isolated: first, syntax-internal triggers, that is, morphosyntactic/formal features as triggers, the ‘traditional’ position in Minimalism; second, syntax-external triggers such as PF- or LFrequirements; third, interface-conditions like discourse information that cause (optional) displacements; and last, requirements or properties of the system itself. In the following sections, ordered according to these four groups, the contributions to this volume will be introduced in brief.

2. Syntax-internal Triggers The role of formal features or functional heads as triggers for displacement is investigated by several authors. Aboh, Lipták, and Mathieu, for example, assume a focus feature operating on a high focus projection to cause displacement. Aboh’s paper is concerned with the triggers for certain focussing displacements especially in Gungbe. Taking issue with Chomsky’s claim that head movement is only phonological and that complement-to-specifier

The Role of Triggers in Linguistic Theory: Some Introductory Remarks

5

movements have to be rejected as a theoretical option, he argues that focus movement of a verb can result in VP-fronting from the complement to the Spec of the focus phrase, particularly in languages with poor verbal morphology,. Such roll-up displacement he calls snowballing movement. Based on empirical data from the D-, I- and C-domains, he shows that both head movement as well as XP-movement are triggered by the same EPP-features resulting in the same interpretive effects. Therefore he argues that both should be within narrow syntax and therefore are just the flip sides of one and the same mechanism, Generalized Pied-Piping. In her paper on Hungarian wh-scope marking constructions (SMCs), Lipták shows that SMCs occur in a variety of guises which can be grouped into sequential vs. subordinated cases. Among the subordinated ones, she distinguishes parenthetical ones from “double question” constructions to which the main body of her paper is dedicated. The trigger for the wh-movement in the embedded clauses (which by definition of the SMC never are selected questions) is that a left-peripheral Focº is able to check not only the [foc]- but also the [wh]-feature on the embedded wh-element. The complementizer hogy is in a higher head of the embedded C-domain. Lipták’s cross-linguistic prediction is that non-standard SMCs (with adverbial or relative clauses) are (only?) possible in languages with a rich left periphery (independently motivated for Hungarian), which can provide a low functional head (Focº) for the licensing of the [wh]-feature. Mathieu’s contribution is concerned with split-XP-structures, traditionally called hyperbaton. The trigger for the movement in question can be either a [wh]- or a [topic] feature, however, only a part of the attracted constituent is moving. Mathieu sets out to explore a hypothesis according to which such XPs split in order to avoid identical (pitch) accents in adjacent positions, a form of haplology. The goal of the paper is to show that haplology is not the driving force behind XP-split (hyperbaton). Focus (and therefore the syntactic correlate of pitch accent), according to Mathieu, is realized syntactically, not prosodically. Therefore, haplology, being a PF phenomenon, cannot be the reason for the XP-split. A bare operator satisfies the checking requirements in the left periphery (wh or Topic). The default option in languages is raising the bare operator only. Movement of the associated nominal along with it occurs not because of the features that are checked by the operator but because of semantic/pragmatic properties. Other internal trigger features are proposed in the contributions by Bell, Corver and Rezac. Bell’s paper presents an account for what triggers the movement of negative indefinites, which he calls N-words, in Afrikaans. In his analysis of

6

Anne Breitbarth and Henk van Riemsdijk

Afrikaans negation, there are two stacked NegPs, headed by nie1 and nie2 , respectively. N-words, according to Bell, move to the specifier of nie1 (the higher Neg°), which entertains an AGREE relation with them. Nie2 , on the other hand, enters into an AGREE relation with v. Because of this, vP is piedpiped to SpecNegP2 , which accounts for the sentence-final positioning of nie2. Material following nie2 can be moved to the Spec of a low FocP above vP (cf. Jayaseelan 2001). Interpretive differences with non-moved clauses/ phrases of the same type (e.g. adjunct clauses) can be explained this way. Bell further shows that and how his analysis extends to other languages with bipartite negation systems like the Afrikaans one, such as Lubukusu, Hausa, or Brazilian Portuguese. Corver argues that emphatic displacement phenomena in Dutch are triggered by a predicate-forming small clause head. Discussing a large variety of Dutch and Frisian constructions featuring a grammatical marker e- associated with emphatic/affective force, he establishes that this element e- is a small clause head linking a predicate to a subject. This head is shown to be able to undergo head movement to a higher head in order to allow for domain extensions (creating equidistance) in cases of Predicate Inversion or for lexicalisation of an operator head in case of Predicate Fronting. This assumption allows for a unified treatment of the “N van een N” (“N of an N”), the “wat voor een N” (“what an N”) and “een A-e XP” constructions. Corver shows that een and –e in these constructions are all instantiations of the same SC head. Rezac, basing himself on novel data from Breton, argues for a feature he calls [X-] as a trigger for all XP-movement (to SpecTP). Because of this triggering property, he says, [X-] is like the EPP-feature. But unlike the EPP-feature, it also acts like a probe, entering an AGREE relation with its goal XP. The reason why most languages have restrictions on what can move to SpecTP, Rezac argues, is the Maximize Matching requirement in combination with Earliness: it is possible to have more than one uninterpretable feature [F-] on a head H, but Match will try topick out the goal that matches most of the other [F-] on H as well, and to do so as soon as possible. Data from Breton illustrate that the morphology of Tº tracks the categorial status of the XP attracted to SpecTP. Rezac takes this to be the overt expression of valued [X-], comparable to  -features, which are uninterpretable on T and valued by moving a DP argument across it to SpecTP. Therefore, [X-]-checking is an instance of A-movement, though without  - or Casechecking, looking for the closest possible XP. If more than one XP can be made equidistant, non-subjects are possible goals as well.

The Role of Triggers in Linguistic Theory: Some Introductory Remarks

7

3. Syntax-external Triggers Niinuma & Park argue for a PF-movement approach to optional head movement in English comparatives (Abby can play more instruments than (can) her father (can)) because, being a focussing operation, it interacts with sentence stress assignment. For their analysis, Niinuma & Park rely on Neeleman & Reinhart’s (1998) analysis of Dutch scrambling. The common property of Dutch scrambling and subject-auxiliary-inversion (SAI) in English comparatives is that both operations seem to be optional and affected by sentence stress assignment. According to Reinhart & Neeleman, an object can undergo scrambling in order to enable unmarked stress assignment to the verb, becausewhen the subject is raised it becomes the most deeply embedded constituent, and thereby liable to receive nuclear stress. By analogy, Niinuma & Park conclude that PF movement of the Aux enables the subject of the comparative clause to receive unmarked stress (if the intention was to focus it). From this analysis, it automatically follows why VP-ellipsis is obligatory in case of SAI in the comparative clause. Assuming that VP-ellipsis, stress assignment, and now also SAI are PF operations, the interaction is expected. Niinuma & Park do not assume all head movement to take place at PF – SAI in wh- and yes/no-questions, not allowing the fronting of multiple heads (as does SAI in comparatives), is analysed as part of narrow syntax.

4. Interface Conditions The contributions by Alboiu, Hinterhölzl, Lambova, Kitahara & Kawashima, and Molnárfi are concerned with the difficult topic of optional displacement phenomena. Under the assumption that all movement is triggered, the existence of such phenomena is difficult to explain theoretically. All authors make use of interface-conditions as causes for (optional) displacements in some way. The empirical phenomena are accounted for by an enrichment of syntax with discourse or other information. The topic of Alboiu’s paper is the triggers of optional movement of contrastively focussed phrases in Romanian. Optional displacement having a number of theoretical and empirical problems, Alboiu argues for a solution according to which in some languages, some features are associated with an OCC/EPP-property13 in the lexicon, while other features obtain the OCCproperty through certain instructions from the semantic component. That is,

8

Anne Breitbarth and Henk van Riemsdijk

they acquire the OCC-property as a result of certain interface requirements during the derivation. Nevertheless, such an insertion of an OCC-feature will have an interpretive effect – moved (preverbal) contrastive foci in Romanian are stylistically more emphatic. In the remainder of her paper, Alboiu presents another piece of evidence, object shift in Romanian, for such an optionally inserted OCC-feature. Hinterhölzl’s paper is concerned with the tricky issue of opionality in German scrambling constructions. He argues that the type of scrambling that affects a destressed constituent is an instance of A-movement, which according to him is triggered by two different features, discourse-related familiarity and relational scope. As such features are not normally assumed to play a role in narrow syntax, he argues for an extension/enrichment of standard minimalism. In this extension, these non-lexical features can be introduced into the derivation by means of functional heads in whose specifiers the scrambling constituents can check them. These functional heads are introduced into the derivation as a result of interface requirements: PF prefers more destressed material to occur to the left and LF requires the correct scope configurations to obtain. Like Hinterhölzl’s, the topic of Molnárfi’s paper is the triggers of Germanic scrambling. What Hinterhölzl calls “familiarity” is called “antifocus” by Molnárfi. Unlike in Hinterhölzl’s paper however, no triggering features are introduced as a consequence of discourse/LF/PF erquirements, they are an element of the syntax from the start. Molnárfi argues that, as opposed to Hungarian, which is a focus-prominent language, German is an antifocus-prominent language. XPs with an antifocus-feature ([+af]) move to the specifier of an antifocus-phrase. This is an A-movement, as in Hinterhölzl’s proposal, because it does not license operator chains. The grammatical effect at Spell-Out of this movement to SpecAfP is anaphoric destressing. The well-known definiteness effects are the result of the fact that [+af] is tied to definiteness by default. Indefinites can only move if they are generic. The reason why scrambling is not always forced (the optionality problem), according to Molnárfi, is that the ± interpretable distinction for features is too narrow and that [af] belongs to a third type of features/triggers which can be interpreted both in syntax (triggering movement) and at Spell-Out (triggering prosodic and/or morphological (Afrikaans vir) effects). Thus, features like [af] give rise to two convergent derivations. Scrambling results from a violable trigger under certain circumstances. The [af]-feature is different from morphosyntactic features (like case) in that its checking is determined by legibility conditions of the prosodic and information structures.

The Role of Triggers in Linguistic Theory: Some Introductory Remarks

9

Basing themselves on data about Scrambling in Japanese, Kawawhima and Kitahara in their article address some problems that arise in phase theory. Pursuing the critique of cyclic transfer in Epstein & Seely (2002), they develop the notion of Strong Cyclic Transfer in a radically derivational manner. The core idea is that feature dissociation, the principle whereby the lower occurrenc of some lexical item is dissociated from the phonological features of that lexical item (that is, is not pronounced), may take place at any point in a derivation but is subject to Last Resort (it will only apply when it has to). The guiding idea behind this approach is that the core principles of syntax and the way they apply are designed so as to maximally reduce operative complexity. Taking the Long Head Movement (LHM) construction (as it used to be called) in Bulgarian as an empirical base, Lambova argues that head movement does belong to narrow syntax as it can have interpretive effects, contrary to Chomsky’s recent claims. While the order auxiliary—participle (Aux—Ptc) in the constructions unter consideration is the more basic one, different factors can be isolated that require the opposite order Ptc—Aux. Lambova singles out a “neutral” and a “non-neutral” case, the neutral one being the one with neutral intonation contours, neutral information structure, and Aux—Ptc order (modulo enclitic behavior of the present tense Aux, a PF phenomenon). The more interesting non-neutral case is the one with a strong high pitch accent on the initial Ptc in the Ptc—Aux order. Lambova argues that this order does indeed arise because of traditional head movement (that is, respecting the HMC), not e.g. remnant VP-movement or even LHM (not respecting the HMC by skipping the Aux position). This head movement is cyclic, successively forming a complex head [T T [Aux Aux [ Ptc ]]]. Lambova is assuming right-adjunction for the formation of this complex head. The high pitch accent on the Ptc is the result of moving the complex head [T T [Aux Aux [ Ptc ]]] to a higher head 6º, the head of 6P, a discourse-related projection checking the focus feature. At Spell-Out, a phonotactic constraint (Bo‰koviç 2001), requiring that the main stress fall on the rightmost element of a prosodic phrase, kicks in. This constraint conflicts with the language-specific requirement that focus be initial in Bulgarian, a focus-fronting language. Lambova suggests that the conflict is resolved by “scattered deletion”, that is, a lower copy of a moved element can be activated under adjacency if the derivation would otherwise crash at PF. As a consequence, Aux is spelt out in the lower copy, while the focussed Ptc is spelt out in 6P ([6 Aux Ptc ] [T Aux Ptc ]). The adjaceny requirement explains why no material such as the subject may intervene between Ptc and Aux in this order.

10

Anne Breitbarth and Henk van Riemsdijk

5. Requirements on / Properties of the System Itself Collins and Moro, the two contributions making up the fourth group, propose that certain properties of the system itself act as ‘triggers’ for certain structures and displacements. In his paper, Collins argues for a parameter to account for the crosslinguistic variation in the relationship between agreement and displacement. This parameter states that if there is an AGREE relation between two categories (a probe and a goal), then internal merger/MOVE is (not) triggered. The paper demonstrates the working of the parameter in the Bantu language Kiswahili where agreement morphology on a predicate invariably triggers movement of the goal to its Spec. If there is no agreement on the predicate for example in expletive constructions), no movement is triggered. The assumption of the Agreement Parameter can account for the correlation between agreement and case checking in Indoeuropean on the one hand and agreement and movement in Bantu languages on the other hand. Moro’s paper approaches the problem of triggers from an exclusively theoretical point of view. Taking Chomsky’s (2000) goal to derive properties of grammar (here: movement) from the structure of organic (here: articulatory) systems and the properties of displacement as his starting point, he attempts to give a unifying approach to both the triggers of movement and the question of why phonological material – lower occurrences of moved material – is deleted. As opposed to standard theories that rely on interpretability of morphosyntactic features, Moro presents a system he calls Dynamic Antisymmetry that makes use of organic/physical circumstances. He assumes that Kayne’s (1994) LCA only holds at Spell-Out when it is needed for linearization of structures. During the derivation itself, symmetric structures are very common. Moro’s main claim is that only movement can create the antisymmetric structures neccessary for well-formed linearization. Therefore, the physical requirement that structures be pronounceable, and therefore linearizable, is the sole trigger for movement.

Acknowledgements This volume grew out of a workshop held from October 24 to October 26, 2002, at the University of Tilburg, Netherlands. We would like to acknowledge an anonymous donation from a Foundation in Liechtenstein that made the workshop possible. This workshop was organized in collaboration with Cornell University. Cornell University also provided financial support,

The Role of Triggers in Linguistic Theory: Some Introductory Remarks

11

especially through the following institutions: Mario Einaudi Center for International Studies, the Institute for European Studies, and the Institute for Cognitive Studies. Thanks to Chris Collins for making this extra support possible. Thanks also to the Linguistic Society of Japan for granting permission to reprint the article by Kawashima & Kitahara.14 We also gratefully acknowledge the help of Nicole van Eijndhoven and Conchita ZerroukBarbé and the work done by all the reviewers for this volume.

Notes 1. This volume is not about linguistic evidence causing the setting of a parameter in one way or another as the term is used in theories on language aquisition. 2. In newer versions of the theory, the displacement is called internal Merge because displacement is remerger of material merged earlier in the derivation. 3. Cf. Chomsky (1981: 49). 4. A short history of the EPP developing from a principle on strucural representations to a ‘feature’ to be checked: In analogy to the obligatoriness of complements (given by the older Projection Principle, cf. Chomsky 1981: 29), by the predicational nature of the VP complement of IP, the Extended Projection Principle stated that while specifiers (like SpecCP) are typically optional, SpecIP is obligatorily filled. (cf. Chomsky 1995: 55) This comes down to saying that clauses must have subjects present at S-structure (cf. Chomsky 1986: 46). In early minimalism (Chomsky 1995), this requirement of the obligatory presence of a SpecIP is expressed by a strong D-feature on I. Consequently, any head having a strong feature and thereby requiring a specifier was said to have this ‘EPP’-property. From there, people went on to assume ‘EPP’-”features” wherever overt filling of specifier positions was obligatory. Thus, EPP has become synonymous with overt displacement. 5. Cf. Chomsky (1995: 27, 194). 6. A-movementstands for displacement to an A(rgument) position asfound in raising or passive constructions (Shei seems to be cold ti; The letteri was written ti). Abar-movement is the displacement to non-argument (or operator) positions, as in wh-movement (Whoi did Mary call ti). 7. Cf. e.g. Heim & Kratzer (1998: ch. 7). Consider a sentence like Everybody on this island knows two languages. It can mean both ‘there are two languages that are such that all inhabintants of this island know these two languages’ and ‘every individual inhabitant of this island is such that s/he knows two languages (but inhabitant x knows languages a and b while inhabitant y knows languages c and d)’. Although the syntactic structure is the same in both cases, the quantifiers two and everybody take different scope, depending on which one of them raises – assuming QR – higher at LF.

12

Anne Breitbarth and Henk van Riemsdijk

8. Cf. Corver & Riemsdijk (1994) for a collection of arguments for both movement and base-generation approaches. 9. Cf. e.g. Holmberg (2000). 10. The terms ‘push chain’ and ‘drag chain’ originate in Martinet’s (1955) work on diachronic phonology. 11. Cf. Müller (2002) for an OT-style approach to triggering such Kaynean displacements. The idea is that there are two constraints Last Resort (LR) (move only to check a strong feature) and Structure Conservation (SC) (the resulting word order must be as in vP). If these constraints are then ordered SC > LR, the desired effect – displacement without checking a strong feature – can be achieved. 12. Nilsen (2003) e.g. argues that the rather rigid order in the placement of adverbs can be derived from their scope taking properties (being positive polarity items, they cannot occur in downwards entailing contexts). This approach is in contrast to such (triggers-based in the ‘traditionally’ minimalist sense) approaches as Cinque (1999) who assumes a strictly ordered hierarchy of functional heads hosting the adverbs in their specifiers. 13. OCC stands for ‘occurrence’ and OCC-features are basically equivalent to EPP-features, cr. Chomsky (2001b). 14. Originally published as: Kawashima, Ruriko & Hisatsugu Kitahara (2003). “Phonological Content and Syntactic Visibility”. Gengo Kenkyu: Journal of the Linguistic Society of Japan No. 123 (March 2003), pp. 137–170.

References Chomsky, Noam 1981 Lectures on Government and Binding. Dordrecht: Foris. 1986 Barriers. Cambridge/Mass.: The MIT Press. 1995 The Minimalist Program. Cambridge/Mass.: The MIT Press. 2000 “Minimalist inquiries: The framework”. In: Roger Martin, David Michaels and Juan Uriagereka (eds.), Step by step: Essays on minimalist syntax in honor of Howard Lasnik. Cambridge/Mass.: The MIT Press. 2001a “Derivation by Phase”. In: Kenstowicz, Michael (ed.), Ken Hale. A Life in Language. Cambridge/Mass.: The MIT Press: 1–52. 2001b “Beyond explanatory adequacy”. MIT Occasional Papers in Linguistics 20, 1–28. Cambridge, Mass.: MITWPL.Cinque, Guglielmo (1999). Adverbs and functional heads – a crosslinguistic perspective. Oxford: Oxford University Press. Corver, Norbert and Henk van Riemsdijk 1994 Studies on Scrambling. Movement and Non-Movement approaches to Free Word-Order Phenomena. Berlin: Mouton de Gruyter.

The Role of Triggers in Linguistic Theory: Some Introductory Remarks

13

Epstein, Samuel D. and Daniel Seely 2002 “Rule applications as cycles in a level-free syntax”. In: Samuel D. Epstein and Daniel Seely (eds.) Derivation and Explanation in the Minimalist Program. Oxford: Blackwell. Holmberg, Anders 2000 “Scandinavian stylistic fronting: How any category can become an expletive”. Linguistic Inquiry 31: 3, 387–444. Kayne, Richard 1994 The Antisymmetry of Syntax. Cambridge/Mass.: The MIT Press. Martinet, André 1955 Économie des changements phonétiques; traité de phonologie diachronique. Berne: A. Francke. Müller, Gereon 2002 “Two Types of Remnant Movement”. In: Artemis Alexiadou, Elena Anagnostopoulou, Sjef Barbiers and Hans-Martin Gärtner (eds.): Dimensions of Movement. Amsterdam: Benjamins, 209–241. Neeleman, Ad and Tanya Reinhart 1998 “Scrambling and the PF-interface”. In: The projection of arguments: lexical and compositional factors, eds. M. Butt and W. Geuder, 309– 353. Chicago: CSLI Publications. Øystein Nilsen 2003 Eliminating Positions. Syntax and Semantics of Sentential Modification. PhD Diss., Utrecht University. (LOT Dissertation Series 73.) Rizzi, Luigi 1997 “The Fine Structure of the Left Periphery”. In: Haegeman, Liane (ed.): Elements of Grammar, Dordrecht: Kluwer, 281–337.

Snowballing Movement and Generalized Pied-piping Enoch Oladé Aboh

1. Introduction In terms of Chomsky (1995), the operation Attract/Move as defined in (1a) may trigger generalized pied-piping (GPP) for convergence, where GPP is described in (1b). (1)

a. K attracts F if F is the closest feature that can enter into a checking relation with a sublabel of K (Chomsky 1995: 297). b. The operation seeks to raise just F. Whatever “extra baggage” is required for convergence involves a king of “generalized piedpiping.” In an optimal theory, […] bare output conditions should determine just what is carried along, if anything, when F is raised (Chomsky 1995: 262)

Under (1), when a formal feature of a head X attracts a matching feature of a category Y, other formal features of Y are automatically moved as “free riders” together with categorical features that are required for PF convergence. As shown in (2), this means that the operation Attract/Move may lead to the formation of three chains: (2a) is required by the operation itself, while the derivative chains (2b) and (2c) could be seen as side effects of GPP. (2)

a. CHF = (F, tF): consists of the feature F and its trace, b. CHFF = (FF[F], tFF[F]): consists of formal feature FF[F] and its trace, c. CHCAT = (_, t_): consists of a category _ carried along by GPP and including the lexical item containing the feature F (Chomsky 1995: 265).

According to Chomsky (1995) the derivative chains (2b-c) are completely free because the computational system is really looking at (2a) even though it can see (2b-c). It is not clear what the status of (2b) is in the system, but

16

Enoch Oladé Aboh

we can reasonably assume that it belongs to narrow syntax. With respect to (2c) however, Chomsky proposes that it is a property of the PF component. As discussed in Chomsky (1995: 263) the mechanism in (2) accounts for the fact that in a situation like (3) – say a partial representation for the sequence [whose book] (Abney 1987; Szabolcsi 1987, 1994; Alexiadou and Wilder 1998) – the whole DP must be pied-piped because (i) the [wh] feature cannot overtly raise alone to check the strong [Q] features under C, (ii) the wh-phrase cannot strand the sequence [D-XP], and (iii) the sequence [wh-D] cannot raise because it is not a syntactic object. (3)

DP 2 who[+wh] DP 2 D XP s’ book

This analysis naturally rules out English wh-questions such as those under (4), and suggests that any of the three choices mentioned will force the derivation to crash at PF. (4)

a. *John bought whose book? b. *Who John bought s’ book? c. *Whose John bought book?

Given the definition of Attract in (1a), the impossibility of choices (i) and (iii) does not seem to be directly related to GPP as defined in (1b). The unavailability of choice (i) suggests that features never extract overtly (i.e., features may not be overtly stripped away from their PF content), in order to check the relevant strong feature acting as probe (Chomsky 1999). Similarly, the computational system excludes choice (iii) because well-formed syntactic objects only are subject to syntactic operations. Accordingly, a specifier cannot be extracted together with its associated head stranding the complement. This, for instance, may explain the inexistence of predicate fronting languages of the type [CP SV [IP …tsv…O… XP]] (Pearson 2000; Rackowski & Travis 2000; Massan 2000, 2001; Paul 2001). However, that choice (ii) is ruled out in the case of (3) and (4b) is not straightforward because it suggests the impossibility of subject extraction, a phenomenon that has been shown to exist cross-linguistically, even though conditioned by appropriate licensing mechanisms of subject traces (Rizzi 1990). For

Snowballing Movement and Generalized Pied-piping

17

instance, Szabolcsi (1994: 204) argues that the Hungarian dative possessor –nak moves obligatorily from an internal position to [spec DP] from where it subsequently extracts out of the possessive construction as schematized in (5a-b). In this framework, cyclical movement through [spec DP] is reminiscent of subject extraction in the clause, in Rizzi’s (1990) work. By moving to [spec DP], the possessor enters in a spec-head agreement that turns the agreeing head D into an appropriate governor that licenses the trace of the possessor in the nominal inflectional domain labeled as (N-I)P, using Szabolcsi’s terminology. (5)

a. [DP [Mari-nak [D° a [(N-I)P tMari-nak kalap-ja]]] b. Mari-nak…[DP [tMari-nak [D° a [(N-I)P tMari-nak kalap-ja]]]

In terms of Rizzi (1990: 60), languages may resolve the tension in choice (ii) (i.e., the impossibility of extracting from a subject position) by adopting repairing strategies “that allow particular systems to circumvent the general ban against subject trace.” Accordingly, that English permits the sequence [whose book] only is a consequence of the repairing strategy that it chooses, namely GPP. In other words, GPP is a last resort strategy that might not be as free as (2c) suggests. I will return to this question in section 4. At this stage of the discussion, it is worth mentioning that the system described in (1) and (2) is meant to account for overt X-raising and XP-raising in the same fashion. But close scrutiny reveals a crucial asymmetry between these two types of movement. For instance, while overt X-raising always results in pied-piping of the head exclusively, XP-raising may allow for pied-piping of ‘extra baggage’, because GPP applies to a category that includes more elements than just the lexical item containing the relevant feature. At first sight, one could think that this asymmetry correlates with other asymmetries that exist between XP-raising and X-raising. For example, Chomsky (1999) observes that: (6)

a. X-raising contrasts with XP-raising in that the former seems to show no surface semantic effect, while the latter systematically does. b. Assuming that D, I, and C have some strong features that need to be checked, the three scenarios under (7) are often taken for granted. Crucially, the scenarios under (8) are not considered even though the system offers no explanation for such choice.

18

Enoch Oladé Aboh

(7)

a.

DP 2 2 D NP N 5 …tN

b.

IP 2 DP 2 I VP V 5 tDP…tV

CP 2 [WH] 2 C IP I 5 tWH…tI

(8)

a.

DP 2 NP 2 D 5 …tNP…

b.

IP 2 VP 2 I 5 …tVP…

c.

c.

CP 2 IP 2 C 5 …tIP…

Chomsky (1999) further proposes that these asymmetries disappear if Xraising is not part of the narrow syntactic computation, but rather a phonological process reflecting the affixal properties of the inflectional categories (Chomsky 1999: 31). This would mean that core syntax implies XP-raising only. The GPP effect that is associated to the operation Attract/Move is therefore a mere PF phenomenon that also applies to what, on the surface, could be described as X-raising. This clearly suggests that in situations, such as, those in (7), the computational system only calculates the chains (DP, tDP) in (7b) and (WH, tWH) in (7c). All the other chains, that is, (N, tN), (V, tV), and (I, tI) that are created by N-to-D, V-to-I and I-to-C movement, respectively, are phonological properties, maybe due to GPP. If this is the right characterization, and given the minimalist notion of the EPP, the question naturally arises what excludes the situations described under (8)? In these cases, the strong features of D, I, and C are checked by XP-movement of the complement to [spec DP], [spec IP], and [spec CP], respectively. This paper discusses these situations and shows that they are instances of GPP. In this framework, GPP is a syntactic phenomenon that is triggered by the need to check the strong (EPP) features under D, I, and C, respectively. Granting that GPP invariably applies to both X-raising and XP-raising, I further conclude that both operations are part of narrow syntax. This would mean that the computational system calculates the three chains under (2). Such conclusion obviously disconfirms Chomsky’s treatment of X-raising as a PF property showing virtually no surface semantic effect (6a). In this regard, section 2 discusses new data on predicate cleft and indicates that, in some languages, verb focusing requires movement of the verb to the left periphery, a clear instance of head movement that has surface semantic

Snowballing Movement and Generalized Pied-piping

19

effects (Koopman 1984; Aboh 1998, 2002, 2004). I therefore conclude that these types of head movement must be treated on a par with XP-raising, as part of narrow syntax. Similarly, the empirical data presented in section 3 suggest that the asymmetry in (6b) is only apparent. In this respect, the discussion of the D-, I-, and C-systems shows that, in some languages (e.g., Gungbe [Kwa], Malagasy [West Austronesian], Niuean [Oceanic]), the operation Attract/Move systematically requires a GPP process that may result in the situations under (8) or else to roll-up structures (i.e. snowballing movement, SBM). Building on this, section 4 argues that GPP is a last resort strategy that is subject to economy conditions on derivations, and should be considered a syntactic phenomenon. Section 5 concludes the paper.

2. X-raising and Surface Semantic Effect Works on predicate cleft (or verb focusing) in Kwa and Kru have shown that such constructions may involve at least four strategies, as indicated in (9) (Koopman 1984, Manfredi 1993). This section discusses the Gungbe situation in (9a), but the reader is referred to section 4.2 and Aboh (2004) where I show that Gungbe manifests strategies (9a) and (9d) in VO and OV contexts, respectively. (9)

[IP …Vi…]]] a. [FocP [Foc° Vi (V-focus leaves a V-copy in IP, e.g., Gungbe) b. [FocP [Nom-V]i [Foc° [IP …Vi…]]] (V-focus is nominalized and leaves a V-copy in IP, e.g., Ewegbe) c. [FocP [YP…[VP…]]i [Foc° [IP …[YP…[VP… ]]i…]]] (VP-focus leaves a VP-copy in IP, e.g., Yoruba) d. [FocP [YP…[VP…]]i [Foc° [IP …ti…]]] (VP-focus leaves a gap in IP, e.g., Gungbe)

2.1. Verb focusing and verb movement in Gungbe The example under (10a) is a Gungbe neutral sentence of the SVO type. The focused counterparts under examples (10b–d) show that argument/adjunct focusing requires movement of the relevant constituent in the focus field. Note that the focused category surfaces in a position left adjacent to the focus marker wE~, and a gap is left in the IP-internal position. In all these

20

Enoch Oladé Aboh

cases, a copy in the IP-internal position leads to ungrammaticality as example (10e) shows. (10) a.

SE!ná
ù blE!
ì lO! Sena eat bread Det ‘Sena ate the bread quickly’

bléblé quickly

b.

BlE!
ì lO! wE~ SE!ná
ù bread Det Foc Sena eat ‘Sena ate THE BREAD quickly’

tblE!
ì lO! bléblé quickly

c.

SE!ná wE~ tSEná
ù Sena Foc eat ‘SENA ate the bread quickly’

blE!
ì bread

lO! Det

bléblé quickly

d.

Bléblé wE~ SE!ná
ù blE!
ì quickly Foc Sena eat bread ‘Sena ate the bread QUICKLY’

lO! Det

tbléblé

e. *[Bléblé] wE~ SE!ná
ù blE!
ì quickly Foc Sena eat bread ‘Sena ate the bread QUICKLY’

lO! Det

[bléblé]

On the other hand, verb focusing in VO contexts requires that the focused verb be moved to the focus field, leaving a copy in the IP-internal position as in (11a).1 The sentence under (11b) indicates that a gap in the IP-internal position leads to ungrammaticality. (11) a.

‹ù (wE~) SE!ná
ù blE!
ì lO! eat Foc Sena eat bread Det ‘Sena ATE the bread quickly’

b. *‹ù eat

SE!ná Sena

t
ù eat

blE!
ì bread

lO! Det

bléblé quickly

bléblé quickly

The examples under (12) further indicate that there is no lexical or semantic restriction on verbal focusing in Gungbe. Verbs that can be focused include: transitive and intransitive verbs (11a) and (12a), double object construction verbs (12c), ergative verbs (12d) in the sense of Burzio (1986) or stative verbs (12e).

Snowballing Movement and Generalized Pied-piping

(12) a.

21

FO!n SE!ná fO!n bléblé stand Sena stand quickly ‘Sena STOOD UP quickly’

b.

Ná SE!ná ná kwE! v˘ı give Sena give money child ‘Sena GAVE the child some money’

lO! Det

d.

Wá SE!ná wá arrive Sena arrive ‘Sena ARRIVED’

e.

NyO!n náwè lO! nyO!n hwànkpE~ Know woman Det know beauty ‘The woman EMBODIES beauty [i.e., she is beautiful]’

Granting the data in (11) and (12), one could suppose that unlike argument/ adjunct focusing where the focused category is moved to the focus field, verb focusing is an instance of base-generation whereby the focused verb is first merged in the focus field. Observe, however, that this analysis cannot explain why the IP-internal V is spelt out in any obvious way.2 In addition, it fails to account for the impossibility of the negative reading in (13), which indicates that V-focusing is sensitive to the intervening negative marker má. (13)

‹ù SE!ná má
ù blE!
ì lO! eat Sena Neg eat bread Det *‘Sena didn’t EAT the bread’ ‘Sena didn’t simply eat the bread, she ATE it greedily’

Similarly, the ungrammatical sentence (14a) shows that, unlike non-verbal focusing (14c), verb focusing is a clause-bound process, because the focused verb cannot move across the intervening complementizer
O~ (14b). (14) a. *‹ù ùn mO~n
O~ SE!ná eat 1sg see that Sena b. Ùn mO~n
O~
ù 1sg see that eat ‘I saw that Sena ATE


ù blE!
ì eat bread

lO! Det

SE!ná
ù blE!
ì Sena eat bread the bread’

lO! Det

c. blE!
ì lO! wE~ ùn mO~n
O~ SE!ná bread Det Foc 1sg see that Sena ‘I saw that Sena ate THE BREAD’


ù eat

22

Enoch Oladé Aboh

I return to the discussion of (13) and (14a) in section 4.2. At this stage of the discussion, it suffices to notice that verb focusing is sensitive to an intervening head (e.g., the negative marker má, and the complementizer
O~). As discussed in the literature (Rizzi 1990; Chomsky 1995; Collins 1996), such case of minimality effect is a typical diagnostic for movement operations. Granting this, I propose that verb focusing in Gungbe VO constructions involves verb movement to the focus head (Foc°) where the verb checks its focus features (Aboh 1998, 2004). Pieces of evidence in support to this analysis come from the fact that the focused verb cannot be pied-piped with its arguments. Accordingly, Gungbe VO constructions disallow VP-fronting. (15) *[Gbá xwé lO! ná Kòfí] build house Det for Kofi ‘Sena BUILT the house for Kofi’

SE!ná Sena

(gbá) build

Similarly, the examples under (16) indicate that verb movement to the focus field is not cyclical. As the ungrammatical examples (16a–b) show, the focused verb cannot cyclically adjoin to the intervening tense and aspect markers on its way to Foc°. Instead, the grammatical example (16c) indicates that the intervening IP-markers must remain in-situ. (16) a. *‹ù-ná-nO~ SE!ná eat-Fut-Hab Sena


ù eat

blE!
ì bread

lO! Det

b. *‹ù-nO~-ná SE!ná eat-Hab-Fut Sena


ù eat

blE!
ì bread

lO! Det

c. ‹ù SE!ná ná nO~
ù eat Sena Fut Hab eat ‘Sena will habitually EAT the bread [i.e., she should not sell it]’

blE!
ì bread

lO! Det

Granting that verb focusing involves V-to-Foc° movement, and assuming Chomsky’s (1995) copy theory of traces, I propose that the V-copy strategy arises as last resort because a gap in the IP-internal position is illegitimate and will violate the Minimal Link Constraint (MLC) or some of its variants (e.g., the Head Movement Constraint).3 In this regard, the Gungbe V-copy strategy is reminiscent to the English do-insertion rule (Chomsky 1995; Aboh 2004). Yet, the presence of the V-copy in the IP-internal position does not dissolve all the minimality effects or chain formation.4 For instance, I argue that the focused verb is sensitive to the intervening negative head because they are both members of quantificational chains that interfere with each

Snowballing Movement and Generalized Pied-piping

23

other. Put differently, focus and negation are quantificational, and V-to-Foc° movement cannot skip an intervening negative head due to minimality effects (Cinque 1990; Rizzi 1990, 1997, 2001; Chomsky 1995; Collins 1996). Similarly, I propose that V-focusing is clause-bound because the verb cannot move across Force°. The latter is always filled by the Gungbe complementizer
O~ ‘that’, which encodes Force and whose subparts (e.g., interrogative force) may interfere with Q-chains. In this regard, that the Gungbe aspect heads do not trigger minimality effect could be linked to the fact that they do not express any [Q] feature or left peripheral feature that may match with those of the verb being focused. The Gungbe verb focus strategy is partially represented in (17). (17) [ForceP [Force°
O~…[FocP [Foc° Vi [FinP … Vi…]]]]] Assuming this is the right characterisation, verb focusing in the Gungbe VO constructions appears to be a counter-example to Chomsky’s observation that head movement manifests virtually no surface semantic effect. Given that Chomsky’s analysis is based on V-to-I movement, contrary to the V-toFoc movement advocated here, a possible alternative would be to propose that head movements that are triggered by inflection have no surface semantic and therefore belong to the PF component, while head movements that are determined by discourse-linked properties (e.g., focus) trigger surface semantic and therefore belong to narrow syntax. Even though this analysis could be interpreted as partial support to Chomsky’s (1999) conclusion that INFL-driven X-raising is a PF process conditioned by the phonetically affixal character of some inflectional category, the facts discussed clearly indicate that that conclusion doesn’t extend to all the relevant cases and cannot be understood as a general statement about head movement. In addition, such a nuanced analysis of head movement in terms of INFL-driven versus discourse-driven head movement fails to account for cases of INFL-driven X-raising (e.g., aspectually determined verb movements or N-raising) that appear to trigger surface semantic effects. Note, for instance, that the French sequences un livre sacré versus un sacré livre are not assigned the same interpretations. For the purpose of this paper, I therefore interpret the Gungbe facts discussed as evidence that both X-, and XP-raising are properties of narrow syntax because they trigger surface semantic effect, even though the degree (or importance) of such surface effect may differ depending on the type of movement involved. The next section addresses Chomsky’s observation under (6b) and shows that languages may vary as to whether they adopt the situations under (7) or (8).

24

3.

Enoch Oladé Aboh

GPP and Predicate Fronting

Under Chomsky’s observation (6b), N-to-D movement, V-to-I movement coupled with DP movement to [spec IP], and I-to-C movement together with wh movement to [spec CP], check the strong features of D, I, and C, respectively. Granting the operation Attract/Move as defined in (1), these various movements are accounted for by proposing that D, I, and C attract the closest matching feature (within their complement) with which they can enter into a checking relation. In this framework, extraction of the goal out of the complement is the preferred option because the system excludes situations under (8) where the goal is embedded in the complement itself. Put differently, the computational system does not seem to allow for situations where the strong features of D, I, and C are checked by movement of their respective complements (i.e., NP, VP, IP) into their specifiers (i.e., [spec DP], [spec IP], [spec CP]). The following section shows that these scenarios arise cross-linguistically and can be regarded as types of predicate fronting. I start with the D-system.

3.1. GPP and Predicate Fronting within D Building on Greenberg’s (1966) typological work, Hawkins (1983) suggests that there are four major sequences with respect to nominal modifiers, as shown in (18). For each sequence, I give the relevant ordering for the present discussion (see Cinque 1996; Shlonsky 2000; Ishane 2003; Aboh 2004). (18) a. 3 modifiers on the left / 0 on the right Dem – Nral – Adj – N (e.g., Mandarin, English, Finnish). b. 2 modifiers on the left / 1 on the right Dem – Nral – N – Adj (e.g., French, Italian). c. 1 modifier on the left / 2 on the right Dem – N – Adj – Nral (e.g., Kabardian, Warao). d. 0 modifier on the left / 3 on the right N – Adj – Nral – Dem (e.g., Gungbe, Yoruba). Hawkins proposes that the sequences in (18b), (18c) and (18d) may originate from the universal base order in sequence (18a), which manifests the hierarchy Demonstrative > Numeral > Adjective > Noun. This order matches

Snowballing Movement and Generalized Pied-piping

25

with the Germanic and Romance languages, where the noun follows the determiner and the nominal modifiers, as shown in (19). (19) a. b. c. d.

Diese These Queste Ces

fünf five cinque cinq

grossen large grandi grandes

Häuser houses case maisons

[German] [English] [Italian] [French]

Sequence (18b), on the other hand, corroborates with the nominal sequencing in languages where the head precedes the adjective. In the Italian and French examples under (20a-b), for instance, the noun surfaces in some intermediate position between the numeral and the adjective. In the Romanian example (20c), however, the noun occurs further left, in a position left adjacent to the determiner. (20) a. Queste cinque case grandi b. Ces cinq maisons merveilleuses c. [ba‡iat]-ul frumos (cel român) boy-the nice (the Rumanian) ‘The nice Rumanian boy’

[Italian] [French] [Romanian]

Granting that nominal modifiers occupy a specifier position within the inflectional domain of the D-system, a fairly standard account for the contrast in the examples (19) and (20) is to assume that the head noun moves past the adjective in (20), but not in (19) where the noun is static. This analysis suggests that the Italian and French examples (20a–b) are cases of partial Nraising (to some intermediate position), similarly to the Romanian example (20c), where the noun left adjoins to the determiner, hence the traditional Nto-D terminology (Grosu 1988; Cinque 1994; Longobardi 1994; Giusti 1994, 1997). Assuming that N-to-D movement is a licensing condition on nominal expressions that applies uniformly at LF, we may explain language surface variations in terms of overt versus covert syntax. This leads to the N-raising parameter, as defined in (21). (21) N raises to D (by substitution) in the Syntax in Romance but not in Germanic (Longobardi 1994: 641). We can therefore propose that N-raising occurs in covert syntax in sequence (18a). On the other hand, sequence (18b) is an instance of overt N-raising

26

Enoch Oladé Aboh

because the noun raises leftward past the adjective. While this analysis is compatible with Chomsky’s characterisation discussed in (6b) and (7a), it fails to account for the sequences (18c) and (18d), which involve GPP (or some of its variant). In these cases, the raised category is a constituent that includes the attracted head N.

3.1.1. Romance versus Germanic With regard to the sequence under (18c), for instance, Bernstein’s (1997) discussion of demonstrative reinforcer constructions suggests the contrast in (22). The modified noun (here in brackets) precedes the demonstrative reinforcer in Romance (22a), but follows in Germanic (22b), see also Ishane (2003) for the discussion. (22) a. Ce [livre jaune] ci this book yellow here ‘this yellow book’ b. Den här [stora huset] this here big house-the ‘this big house’

[French]

[Swedish]

According to Bernstein (1997, 2001a, b) this contrast, can only be accounted for if we assume that, in Romance, the noun raises past the adjective to form the sequence [Noun-Adjective], which is pied-piped to the left of the reinforcer. Similarly, the demonstrative ce must move to [spec DP] to encode deixis, as illustrated in (23a). On the other hand, the Germanic languages do not involve N-raising or GPP, since the noun follows the adjective and the sequence [Adjective-Noun] follows the reinforcer (23b). Note that demonstrative reinforcer constructions differ from normal demonstrative noun sequences (e.g., this book) because they encode some type of emphasis or focus.

Snowballing Movement and Generalized Pied-piping

27

(23) a.

DP b. DP 2 2 DP thisi DP cei 2 2 D FP[Reinforcer] D FP[Reinforcer] 2 2 FP FP : 2 2 # # F F FP[Demonstrative] FP[Demonstrative] # ci 2 here 2 # FP spec FP z-- spec ti 2 ti 2 F NP F NP [yellow book] [livrek jaune tk]

In terms of Berstein, we can interpret GPP in (23a) as the need of the noun to check its focus feature under F. This implies that the focus feature is strong in Romance, but not in Germanic, where no raising occurs, and [spec FP] is presumably filled by a null operator (absent in 23b). Assuming that noun phrases are predicates (Campbell 1996), I further propose that (23a) is the nominal counterpart of predicate fronting. This leads me to conclude that representation (23a) is a counter-example to Chomsky’s observation discussed earlier in (6b), because the Romance demonstrative reinforcement constructions instantiate the situation where checking the strong features of a head leads to GPP of the complement to the specifier of that head. The next section shows that this strategy also occurs in Gungbe.

3.1.2. GPP and Predicate Fronting in the Gungbe D-System The Gungbe example (24) manifests the hierarchy noun > adjective > numeral > demonstrative > determiners, that is, the mirror image of Hawkins’(1983) universal order, determiners > demonstrative > numeral > adjective > noun, which is instantiated by the English translation. (24)

Àgásá
àxó àtO~n crab big Nral ‘These three big crabs’

éhè Dem

lO! Det

lE! Num

28

Enoch Oladé Aboh

For the sake of the discussion I refer to the specificity marker lO!, and the number marker lE! as the Gungbe determiners.5 LO! encodes specificity because it relates to pre-established referents in discourse (Enç 1991; Pesetsky 1987; Ishane & Puskás 2001). Accordingly, a sequence like àgásá lO! should be understood as ‘the aforementioned crab’. LE! on the other hand encodes plurality. In this paper, I assume that the Gungbe determiners are the morphological realizations of the features [±specific] and [±plural], respectively. Granting this, it is obvious from example (24) that we cannot account for the Gungbe sequencing in terms of the N-raising parameter. Put more generally, we cannot derive the ordering in (24) by means of cyclical N-to-D movement, as it has been proposed for Romanian (Grosu 1988; Giusti 1994, 1997). For such analysis to hold, we would have to stipulate that the Gungbe modifiers are heads, and the noun cyclically adjoins to them on its way to D. Yet, there is strong cross-linguistic evidence that nominal modifiers are maximal projections that are located in the specifier of some functional projection within the nominal inflectional domain (Cinque 1994, 1996; Giusti 1994, 1997). In addition, the form and the distribution of the Gungbe determiners suggest that they manifest a different category from other nominal modifiers. Observe, for instance, that the determiners need not be realized in the noun sequence. Accordingly, Gungbe allows some type of bare noun phrases, where the noun may surface with nominal modifiers only (25a–b). Similarly, the sentences under (25c–d) indicate that the determiners need not be realized simultaneously. Finally, example (25e) shows that, in relative clauses, the CP-clause appears between the head noun and the determiners that are set off to the right. In this regard, these markers behave like the Gungbe left peripheral markers, some of which may occur to the right edge as I propose in section 4.2 (see also Aboh 2002, 2004). (25) a. Kòfí xO~ [àgásá] wà Kofi buy crab come ‘Kofi brought a crab home’

xwégbè house

b. Kòfí xO~ [àgásá
àxó éhè] Kofi buy crab big this ‘Kofi brought this big crab home’

wà xwégbè come house

c. Kòfí xO~ [àgásá
àxó éhè lO!] wà xwégbè Kofi buy crab big this Det come house ‘Kofi brought this big crab home [i.e., the aforementioned crab]’

Snowballing Movement and Generalized Pied-piping

29

d. Kòfí xO~ àgásá
àxó éhè lE! wà xwégbè Kofi buy crab big this Num come house ‘Kofi brought these big crabs home’ e. Kòfí wE~ xO~ àgásá
àxó Kofi Foc buy crab big [
e˘ mí wlé] lO! lE! catch Det Num that[Rel] 1pl ‘KOFI bought the big crabs that we caught [i.e., the aforementioned crabs]’ Building on Abney (1987), Szabolcsi (1987, 1994), Ritter (1995) and subsequent works, I account for the Gungbe noun sequences in terms of split-D hypothesis. I propose that the Gungbe specificity and number determiners delimitate the nominal left periphery upward and downward, respectively. DP, headed by lO!, is the highest projection of the nominal left periphery. It represents the interface between the nominal predicate and previous discourse. On the other hand, NumP, headed by lE!, encodes plurality. It manifests the interface between the nominal left periphery and the nominal inflectional domain. In this respect, Num° may encode other features (e.g., ± definite) that agree with those expressed in the nominal inflectional system, that is, the extended projection of N.6 I therefore suggest that the nominal sequence involves the structure in (26), where F is the nominal counterpart of I (or T) in the clause. It has strong [n] features (and maybe a strong EPP-feature that is checked by the possessor in possessive constructions). Similarly, nominal modifiers are like adverbs (i.e., maximal projections) that are merged in the specifier positions of distinct functional categories that compose the nominal inflectional system (Szabolcsi 1987, 1994; Cinque 1996, 1999; Ihsane & Puskás 2001; Aboh 2002, 2004). (26) [DP [D°[±specific] [NumP [Num°[±plural] [FP [Demonstrative] [Numeral] [Adjective] [NP]]]]]] Under Attract/Move, the head noun must raise and substitute into F to check its strong [n] features, as is the case in Romance. But suppose that N-extraction is not possible in Gungbe, presumably due to the extremely poor morphology that this language manifests. As a result, GPP applies through the successive pied-piping of categories that include the head noun leading to SBM. The NP moves past the adjective to form the sequence [noun-adjective]. The latter moves to the left of the numeral to form the sequence [nounadjective-numeral], which moves to the left of the demonstrative giving rise

30

Enoch Oladé Aboh

to the sequence [noun-adjective-numeral-demonstrative], which raises to [spec FP], as in (27a).7 Subsequently, FP moves cyclically to [spec NumP] and [spec DP] in order for the noun to check its number and specificity features against Num° and D°, as illustrated in (27b) (Aboh 2002, 2004). (27) a. [FP [XP [DemP éhè [YP [NralP àtO~n [ZP [AP
àxó [NP àgásá]]]]]]]] : : : : 14243 # # # z---------m # # # # # # 14444244443 # # z-------------m # # # # 1444444424444443 # z----------------m # 1444444442444444443 z--------------m b. [DP[D° lO![NumP[Num°lE![FP[àgásá
àxó àtO~n éhè] [F° [tXP]]]]]]] : 144444424444443 #: z---mz-------------m Under this analysis, SBM is motivated by the need of the noun to check the strong [n] features under F. This operation seeks to raise just the matching feature under N, but the whole NP is pied-piped leading to SBM. I therefore conclude that SBM is a sub-case of GPP, which applies as last resort when the relevant attractee cannot extract. Put differently, SBM arises here because, unlike the Romance noun, the Gungbe noun cannot extract in order to check the strong features of the nominal inflection head F. Notice further that the mechanics of SBM appears parallel to head-to-head movement: both movements result in successive adjunction rule, and SBM does not manifest any clear surface semantic effect in this case. Under Chomsky (1995, 1999) this could be interpreted as further evidence that the Romance N-to-D movement and its Gungbe SBM counterpart are PF properties. This conclusion supports the analysis of GPP as a free PF process. However, we immediately face a contradiction, if we consider this conclusion in the light of Chomsky’s idea that the system calculates XP chains only. In a sense, the computational system calculates the chain that arises from the Gungbe SBM because the latter involves maximal projections, but fails to see the Romance N-to-D counterpart, which is a PF property. As far as I can see, there is no obvious way for the system to handle such asymmetry. Similarly, if we were to maintain that SBM is a PF phenomenon, we would not have any principled way to account for the fact that this process

Snowballing Movement and Generalized Pied-piping

31

displays the mirror image of Hawkin’s (1983) universal order, which could reveal underlying head-complement sequences. Note that under Brody’s (2000: 28) mirror theory “the syntactic head-complement relation expresses the morphological specifier-head relation in inverse topological order.” The question naturally arises why would a PF process care about head-complement relations? This tension dissolves, it seems to me, if we assume that the operation Move/Attract leads to both X-raising and XP-raising as consequences of GPP. Under this analysis, SBM in Gungbe is motivated by the need to check the strong [n] features under F, while cyclical movement of the complement FP to [spec NumP] and [spec DP] is considered an instance of predicate fronting, that is, the nominal counterpart of focus A’-movements (or VP fronting) in the clause. Put more generally, SBM in Gungbe is triggered by the strong INFL-features under F. Fronting of the nominal predicate, however, is determined by the strong EPP features (as defined in Chomsky 1995) that are located in the nominal left periphery. This analysis is compatible with Campbell’s (1996) proposal that, in languages that do not manifest NP-raising (e.g., English), specificity is checked thanks to a null operator in [spec DP], as shown by representation (28). I will not pursue this matter any further here and the reader is referred to Campbell (1996) and Aboh (2002, 2004) for further discussion. (28) [DP Opi [D° the [FP [e]i thief]]] Building on this, I conclude that GPP is determined by the same syntactic operation that triggers XP-raising and X-raising, as part of narrow syntax. This leads me to propose that: (29) a. Attract/Move subsumes GPP b. GPP is not free This generalization appears to be well motivated both on empirical and conceptual grounds. The next section discusses GPP within the clause inflectional domain in Malagasy and Niuean. The discussion shows that the effects of GPP are comparable to both V-to-I movement and DP movement to [spec IP] in typologically different languages (e.g., Romance).

32

Enoch Oladé Aboh

3.2. GPP within INFL: The Case of Malagasy and Niuean Cinque (1999) argues that adverbs fall into distinct classes that are associated with discrete functional projections whose ordering reveals a universal hierarchy partially represented in (30), and illustrated by the Romance and Germanic examples (31a-b). (30) C1 > C2 > C3 > C4 > C5 > C6 generally already anymore always completely well (31) a. Pierre fait toujours bien ses devoirs. b. Peter always does his homework well Under V-to-I movement, it has been proposed that the verb may move past certain middle field adverbs in languages with rich INFL (e.g., French) but not in languages with weak INFL where V-raising is thought to be covert (e.g., English) (Pollock 1989; Vikner 1997; Zwart 1997). But this distinction fails to account for the adverb sequencing in Malagasy, a VO language lacking overt V-movement of the Romance type. As Pearson (2000) shows, the Malagasy adverbs fall in two groups. In this language, preverbal adverbs match with Cinque’s universal hierarchy in (30), unlike postverbal adverbs, which manifest the mirror image of that hierarchy. The hierarchy in (32) partially represents the Malagasy situation, for which the relevant examples are given under (33). (32) C1 > C2 [Verb] < C6 < C5 < C4 < C3 matetika efa [Verb] tsara tanteraka foana intsony generally already well completely always anymore (33) a. Manasa lamba [tsara tanteraka] wash clothes well completely ‘Rakoto completely washes clothes well’

Rakoto Rakoto

b. Manasa lamba [tanteraka foana] Rakoto wash clothes completely always Rakoto ‘Rakoto always washes clothes completely’ c. Tsy manasa lamba [foana intsony] Neg wash clothes always anymore ‘Rakoto doesn’t always wash clothes anymore’

Rakoto Rakoto

Snowballing Movement and Generalized Pied-piping

33

These facts find no obvious answer in a classical theory of V-to-I movement that assumes overt versus covert distinction. In addition, it is not clear how the analysis of head movement as a PF phenomenon would resolve the asymmetry between the Malagasy preverbal versus postverbal adverbs. Under such theory the question naturally arises what properties of the PF component allow the reverse order of the postnominal adverbs, but leave the preverbal ones unaffected. However, this tension disappears if we assume, along the lines of Pearson (2000), that overt (INFL-driven) verb movement past the postnominal adverbs is impossible in Malagasy. As a result, successive GPP applies to the VP, leading to SBM. VP left adjoins to the adverb C6. Then the phrase verbadverb[class 6] moves to the left of the adverb C5. The phrase verb-adverb[class 6]adverb[class 5] further moves to the left of the adverb C4 and the phrase verbadverb[class 6]-adverb[class 5]-adverb[class 4] subsequently moves to the left of the adverb C3, forming the phrase verb-adverb[class 6]-adverb[class 5]-adverb[class 4]adverb[class 3]. The derivation is represented in (34). (34) [FP[FPC3 intsony [FP [FPC4 foana [FP [FPC5 tanteraka [FP [FPC6 tsara [VPmanasa lamba]]]]]]]]]

: : : : 14243 # # # z-----------m # # # # # # 144444424444443 # # z------------m # # # # 144444444424444444443 # z----------------m # 144444444444424444444444443 z------------------m Under this approach, SBM in Malagasy parallels with partial V-to-I movement in Italian, where the verb moves past the lower adverbs (Cinque 1999). However, the two languages differ because, in the Malagasy roll-up structure, the strong features of the attracting INFL head are checked in the specifier of that head by a phrase that contains the goal (i.e., the verb). In Italian, however, the goal itself raises and adjoins to the probe. Here again, Chomsky’s (1999) analysis of X-raising as PF property suggests that the computational system can see the Malagasy case but not the Italian one. This is so even though a DP moves (or merges) in [spec IP] to check the strong EPP feature under I in both languages. I now turn to the issue of predicate fronting in Niuean, an Oceanic VSO language.

34

Enoch Oladé Aboh

Niuean is a V-Particles-S-O-X language that manifests an ergative case system, which marks common nouns and proper nouns (or pronouns) differently. Note, for instance, the absolutive case e– on the object taros in (35a) versus a– on the proper noun object Sefa in (35b). (35) a. Ne kai e Sione e tau talo aki e huki Pst eat Erg Sione Abs Pl taro with Abs fork ‘Sione ate the taros with a fork’ b. Koe tele e Sione a Pres kick Erg Sione Abs ‘Sione is kicking Sefa’

Sefa Sefa (Massam 2001: 155)

In addition, these examples indicate that the initial verb may be preceded by various tense (and aspect) particles, such as, the past and present markers ne and ko, respectively. Similarly, the following example shows that the verb may precede adverbial and emphatic particles. (36) Takafaga tmau n¥ e hunt always Emph Erg ‘He is always fishing’

ia he

e Abs

tau ika Pl fish (Massam 2001: 157)

Granting that Nuean tense is in C (Massam 2000, 2001), one might interpret these data in terms of verb movement to some C-type projection, say Fin° in terms of Rizzi (1997). However, this analysis is not tenable because Niuean also manifests V-O-Particles-S sequences, as the example (37) shows. (37) Takafaga ika tmau hunt fish always ‘He is always fishing’

n¥ ia Emph he (Massam 2001: 157)

Assuming that the direct object merges as the complement of the verb, we might conclude that the sequence in (37) is an instance of VP fronting where the following object is a DP. Close scrutiny suggests that things might not be so clear-cut. Note, for instance, that the object ika in example (37) surfaces without the absolutive case marker e– and the number marker tau. This is unexpected since nominals are obligatorily case-marked in ordinary Niuean sentences and may be followed by various determiners, such as, the number marker and the specificity marker, as shown in (36) and in previous examples.

Snowballing Movement and Generalized Pied-piping

35

In her account for the Niuean data, Massam (2000, 2001) interprets these facts as evidence that the object, in V-O-Particles-S construction, is a bare NP, hence the terminology of ‘pseudo noun incorporation’ (PNI). In this framework, PNI constructions (e.g., 37) are cases where a bare NP merges as the complement of the verb. Unlike DPs, bare NPs fail to move out of the VP in order to check absolutive case, as a consequence of which they are pied-piped together with the verb. This strategy gives rise to the order V-OParticles-S. This implies that, in the Niuean normal transitive sentences, the DP object moves out of the VP to check the strong [D] feature associated with absolutive case before VP fronting takes place. This further suggests that in the V-Particles-S-O-X sequences presented earlier in (35), the fronted category is a VP that includes the trace of the DP object that has moved out, due to case reasons. If this is the right characterization, we may propose that Niuean systematically displays VP or predicate fronting to some intermediate specifier position within the I-system. In this regard, Massam (2001) concludes that, unlike English where a DP must surface in [spec IP], the Niuean sequences V-O-Particles-S and V-Particles-S-O-X can be interpreted as instances of predicate fronting where a predicate (i.e., VP) is moved to [spec IP]. The argument goes as follows. In ordinary transitive sentences, the verb merges with a DP object to form the VP, which in turn merges with the functional projection responsible for absolutive case (say AsbP). The DP object then moves to [spec AbsP] to check the strong [D] features associated with the absolutive case. Then AbsP merges as the complement of little v to project vP, whose spec hosts the external argument checking the ergative case in situ. vP merges as the complement of I to project IP. I is associated with a strong EPP feature in the sense that it requires a specifier (Chomsky 1995). Unlike English, where the EPP requirement is expressed by the [D] features under I that attract the closest DP, the Niuean I lacks the feature [D]. Instead, Niuean I is endowed with the features [Pred] leading to predicate fronting as represented in (38a) for ordinary transitive sentences. Representation (38b) illustrates PNI constructions where the verb and the following bare NP are pied-piped to [spec IP] (Massam 2001: 164–165).

36

Enoch Oladé Aboh

(38) a. IP b. IP 3 2 VP I’ VP I’ 1 2 1 2 vP V NP I AbsP V tDP I 2 2 DP[abs] Abs’ DP[erg] v’ 2 2 v AsbP Abs tVP 2 DP Abs’ 2 Abs tVP The proposed analysis extends to intransitive PNI constructions, as well as predicate nominals and prepositional predicates where a nominal maximal projection is pied-piped in [spec IP]. An interesting typological conclusion here is that languages of the Niuean-type cannot display subject initial sequences (e.g., SVO, SOV) because [spec IP] always hosts a fronted predicate. This further suggests that the Niuean [Pred] and the English [D] features under I are in complementary distribution because they relate to the same principle. Put differently, the English-type SVO order versus the Niuean-type VSO order are two instantiations of a single strong EPP feature under I. According to Massan (2000: 166) this would mean that Niuean predicate fronting “is the VSO realization of EPP.” In other words, surface differences between English-type and Niuean-type languages reduce to the parametrical choice between the features [D] or [Pred] that are associated with I. Cross-linguistic studies therefore suggest that the computational system allows for the situations in (7b) and (8b) described earlier. In a similar vein, the next section focuses on the C-system and shows that, in certain languages, the licensing of the strong features of C may trigger predicate fronting, resulting in situation similar to (8c) repeated here for convenience. (8)

c.

CP 2 IP 2 C 5 …tIP…

Snowballing Movement and Generalized Pied-piping

37

3.3. GPP, SBM and Interrogative Force in Gungbe The sentences under (39a-b) show that topic and focus phrases occur in a space between the complementizer
O~ and the subject, the topic and focus being left adjacent to the topic marker yà and the focus marker wE~, respectively. When topic and focus phrases co-occur, they do so in the fixed order topic-focus as shown in (39c). (39) a. Ùn sè
O~ [àgásá lO!]i yà Kòfí xO~ E~i 1sg hear that crab Det Top Kofi buy 3sg ‘I heard that, as for the crab, Kofi bought it’ b. Ùn sè
O~ [àgásá lO!]i wE~ Kòfí xO~ ti 1sg hear that crab Det Foc Kofi buy ‘I heard that Kofi bought THE CRAB’ c. Ùn sè
O~ [àgásá lO!]i yà [Kòfí]j wE~ tj xO~ E~i 1sg hear that crab Det Top Kofi Foc buy 3sg ‘I heard that, as for the crab, KOFI bought it’ Sentence (40), on the other hand, shows that yes-no question is encoded by a sentence-final low tone as indicated by the additional stroke [`] glossed as Int. (40) Ùn kànbíO
O~ Kófí xO~ àgásá lO! ` 1sg ask that Kofi buy crab Det Int ‘I asked whether Kofi bought the crab?’ But in certain yes-no questions involving the topic and focus markers, the latter occur sentence-finally in the mirror image of the hierarchy in (39c): the focus precedes the topic, which now bears the sentence-final low tone expressing interrogative force. (41) Ùn kànbíO!
O~ Kófí xO~ àgásá lO! wE~ yà ` 1sg ask that Kofi buy crab Det Foc Top Int ‘I asked whether KOFI BOUGHT THE CRAB (as expected)?’ Granting Rizzi’s (1997) split-C hypothesis, I propose that the interrogative head attracts the event head (i.e., the verb inside the IP) but the latter cannot extract in Gungbe. Consequently, SBM applies under GPP. FinP moves to [spec FocP], and FocP moves to [spec TopP], followed by TopP movement to [spec InterP]. The derivation is represented in (42), (Aboh 1998, 2004).

38

Enoch Oladé Aboh

(42) [ForceP [Force°
O~ [InterP [Inter° ` [TopP [Top° yà [FocP [Foc° wE~ [FinP Kófí xO~ àgásá lO!]]]]]]]] : : : 144424443 # # z----------m # # # # 14444244443 # z-----------------m # # 14444444244444443 z-----------------m Under Attract/Move, the discussion suggests that languages vary as to whether checking the strong features of D, I, and C triggers X-raising or XP-raising or a combination of both, as in (7), or systematic predicate fronting, as in (8). In this framework, predicate fronting is regarded as an instantiation of GPP. The latter is a last resort device that applies whenever the targeted head cannot extract overtly. This would mean that GPP is contingent to X-extraction or X-raising. The next question to ask is whether GPP is subject to general economy conditions on a par with X-raising, and XP-raising.

4.

GPP and the General Economy Condition

Under the generalization (29), repeated here for convenience, GPP is part of narrow syntax and should fall under the economy conditions that constrain the operation Attract/Move. (29) a. Attract/Move subsumes GPP b. GPP is not automatic (i.e., free) This section discusses improper movement in English and verb focusing in Gungbe, and shows that certain properties of these constructions (e.g., clause-boundness, sensitivity to negation) can be better accounted for, if we assume that they involve GPP, which obeys the MLC (Rizzi 1990, Chomsky 1995, Collins 1996).

4.1. GPP and Improper Movement Ura (2001) shows that the operation Attract/Move associated with free GPP along the lines of Chomsky (1995) rules in cases of improper movement such as example (43a) where the intended interpretation is represented in (43b).

Snowballing Movement and Generalized Pied-piping

39

(43) a. *Whoi has not yet been decided [CP t’i Q [IP it will be told ti [that Mary has left]]] b.

It has not yet been decided [CP whoi will be told ti that Mary has left]

Granting that the wh-phrase who is a DP, it has a [D] feature in addition to a [wh] feature. If we assume Chomsky’s (1995) definition of GPP in terms of “free riders”, the question arises whether the movement of who involves the pied-piping of a [D] feature as free rider. Like wonder, the verb decide can take an embedded interrogative clause. This means that the embedded C has a strong [Q] feature that attracts the wh-phrase who before spell-out as partially represented in (44). (44) [IP Infl …. been decided [CP whoi Q [IP it will be [VP told ti [that Mary has left]]]] Under GPP, the attraction of the [wh] feature of who also allows the piedpiping of its [D] feature to the intermediate C-domain, as “free rider”. Accordingly, the [D] feature of who should be visible to the matrix INFL, where it is attracted to check the strong EPP feature of I. The derivation will therefore converge as illustrated in (45), contrary to facts. As a result, a mechanism that allows GPP associated with “free riding” cannot exclude the derivation in (45). (45) [IP whoi Infl …. been decided [CP t’i Q [IP it will be [VP told ti [that Mary has left]]]] In this regard, Ura (2001) argues that this inconsistency disappears if we assume that GPP is subject to economy condition and therefore cannot involve anything like “free riding” in Chomsky’s (1995) sense. He therefore defines GPP as in (46). (46) GPP _ carries along ` as a free rider if there is no feature a such that (i) a structurally intervenes between ` and _ and the element that attracts _ and (ii) a matches with ` (Ura 2001: 187). Under this definition, the derivation in (45) is ruled out because the [D] feature of the expletive it structurally intervenes between [spec CP] and the original position of who. Put differently, the [D] feature of who cannot be

40

Enoch Oladé Aboh

pied-piped to the embedded [spec CP] due to the intervening expletive pronoun it. This means that when it reaches the embedded C, the wh-phrase does not have any [D] feature that can be attracted by the matrix INFL. Accordingly, derivation (45) is ruled out as a violation of last resort. The discussion here clearly suggests that the operations GPP and Attract/Move are subject to similar general economy conditions because both are part of narrow syntax. Additional empirical evidence from Gungbe supports this hypothesis.

4.2. Generalized Pied-Piping and Verb Focusing As discussed earlier in section 2, the sentences under (47) show that, in the Gungbe SVO constructions, the focused verb is moved to the focus head Foc°. This construction requires a copy strategy, where the focused verb leaves a copy in the IP-internal position (47b). The ungrammatical sentence (47c) shows that VP-fronting is prohibited in VO constructions. (47) a. Ùn sè
O~ Kòfí ná nO~
ù lE!sì lO! 1sg hear that Kofi Fut Hab eat rice Det ‘I heard that Kofi will habitually eat the rice’ b. …[CP [C°
O~ [FocP [Foc°
ùi [FinP Kòfí ná nO~
ùi lE!sì lO!]]]]] that eat Kofi Fut Hab eat rice Det ‘…that Kofi will habitually EAT the rice [instead of selling it]’ c. *…
O [
ù lE!sì lO!]i Kòfí ná nO~ ti that eat rice Det Kofi Fut Hab In the Gungbe progressive SOV constructions, however, verb focusing requires that the nominalized constituent including the VP (i.e., the OVX sequence) be pied-piped to [spec FocP]. Example (48b) is the focus counterpart of (48a). Note also that movement past the progressive marker in (48b) triggers the agreeing form tè as opposed to tò in (48a). The ungrammatical sentence (48c) shows that, unlike the VO sentences under (47), the V-copy strategy is not allowed in OV constructions.

Snowballing Movement and Generalized Pied-piping

41

(48) a. Ùn sè
O~ Kòfí tò [lE!sì lO!
ù ` ] 1sg hear that Kofi Prog rice Det eat NR ‘I heard that Kofi is eating the rice’ ` ]i [Foc° [FinP Kòfí tè ti]]]]] b. …[CP[C°
O~ [FocP [lE!sì lO!
ù that rice Det eat NR Kofi Prog ‘Kofi is EATING THE RICE’ c. *…
O~
ù i Kòfí tò [lE!sì …that eat Kofi Prog rice

lO! Det


ù i eat

`] NR

What examples (47b) and (48b) have in common is that focus movement seeks to raise just the feature [+Foc] on the verb. But this requirement leads to the two strategies in (49), because the verb is not freely extractable in Gungbe. (49) a. a copy strategy applies in (47b) and V is pied-piped to Focº. b. the nominalized VP is pied-piped to [spec FocP], in (48b) leaving a gap in IP. Granting that both (49a) and (49b) are instances of GPP, I conclude that the constraint that leads to a V-copy in (49a) versus a gap in (49b) is the same as the one operating on the features that are being pied-piped. Put differently, GPP is a constrained strategy that cannot be free. Verbal focusing in Gungbe manifests two other properties that support this hypothesis. 1.

Verbal focusing is local: it is a clause-bound process irrespective of whether GPP involves V or a phrase containing the verb (i.e., VP, or some higher category embedding VP).

[CP
O~ Kòfí ná nO~
ùi lE!sì lO!] (50) a. *‹ùi ùn sè eat 1sg hear that Kofi Fut Hab eat rice Det ‘I heard that Kofi will habitually EAT the specific rice’ b. *[LE!sì lO!
ù ` ]i ùn sè [CP
O~ Kòfí tè ti] Rice Det eat NR 1sg hear that Kofi Prog ‘I heard that Kofi WAS EATING the RICE’ 2.

Unlike focusing of non-verbal elements (51a), verb focusing is sensitive to negation. This is illustrated by the examples under (51bc), see also the discussion in section 2.

42

Enoch Oladé Aboh

(51) a. [DP LE!sì lO!] wE~ Kòfí má
ù rice Det Foc Kofi Neg eat ‘Kofi didn’t eat THE RICE’
ùi lE!sì lO! b. ‹ùi Kòfí má eat Kofi Neg eat rice Det *‘Kofi didn’t EAT the rice’ ‘Kofi didn’t only eat the rice (he swallowed it all)’ c.

*[LE!sì lO!
ù]i Kòfí má rice Det eat Kofi Neg ‘Kofi WASN’T EATING the RICE’

tè ti Prog

These facts find no obvious answer under Chomsky’s theory of GPP in terms of free PF process, because one wonders why should PF care about intervening negative features in just these cases? In terms of Ura’s definition of GPP, however, a possible option would be to assume that, in sentence (47b) the copy strategy is a last resort saving device that allows pied-piping of V to the embedded focus head, where the focus feature is checked. But subsequent V-movement (GPP) to the matrix clause is banned due to the intervening embedded C node realized by the complementizer
O~ in Gungbe, and presumably the intervening I nodes of the matrix clause. In a similar vein, I can explain the agreeing form tè found in focused OV constructions, versus tò in (48a), by assuming that focus movement of the verbal phrase proceeds through the specifier of tò triggering as such a spec-head agreement encoded by the form tè. This mechanism legitimates the gap to the right of the progressive marker and allows GPP to apply to the nominalized VP, which is moved to [spec FocP] to check the focus feature. But here again, subsequent movement to the matrix clause is prohibited due to the intervening components of the embedded C and the matrix I (Aboh 2004). Given that minimality does obtain in the situations under (49) irrespective of whether movement involves X-raising (49a) or XP-raising (49b), the only logical conclusion I reach is that what count here are the features that are being pied-piped under Attract/Move, but not the actual categories they are embedded in. This helps understand the contrast in (51b) and (51c), namely that (51b) is ungrammatical with a negative reading (but acceptable with some emphatic positive reading), while (51c) is ruled out. Following Heageman (1995), Rizzi (2001) and subsequent work, a possibility that comes to mind here is that focus and negative features belong to the class of quantificational features, say [Q]. If true that minimality effects arise within the same class of features, but not across classes, then the contrast in (51) follows

Snowballing Movement and Generalized Pied-piping

43

because the [Q] features of Neg structurally intervenes with that of V or the (nominalized) VP that are being attracted. As a result GPP is blocked due to minimality.

5. Conclusion This paper shows, contra Chomsky (1995, 1999) that GPP of syntactic features is not a free process. Instead, empirical data from the D-, I- and Csystems suggest that, just as Move/Attract, GPP is subject to general economy conditions. Under Chomsky’s minimalist theory, only syntactic operations are calculated by the computational system for human languages (CHL). This therefore implies that GPP is a syntactic operation, not a PF property. In addition, I show that, when applied to Xº-raising, GPP may lead to SBM, a fact that indirectly suggests that head-movement is part of the narrow syntax.

Notes 1. For some speakers the focus marker is optional in verb focusing, but for others, the focused verb and the focus marker are mutually exclusive. From now on, I leave the focus marker out of V focusing, see Aboh (1998, 2004) for the discussion. 2. Alternatively, one could propose that verb focusing results in remnant VPfronting. The focused element expresses a VP that has been emptied of its arguments/adjuncts. Accordingly, the Gungbe situation appears parallel to that of languages like Yoruba that exhibit the strategy in (9c). However, this analysis too faces the problem of why is the IP-internal V spelt out? Note that such copy mechanisms is not allowed in OV constructions where V-focusing fronts a maximal projection, including the VP, leaving a trace in the IP-internal position as strategy (9d) shows. 3. See Chomsky (1995) and Collins (1996) for alternative approaches to the HMC. 4. This supports the idea that V-copy is a last resort strategy to avoid a gap inside IP. 5. See Aboh (2002, 2004) for a detailed discussion of the nominal sequence in Gbe. 6. This fact, it seems to me, may correlate with determiner spreading in languages like Greek (Panagiotidis 2000; Carstens 2000). 7. See also Shlonsky (2000) and Ishane (2003) for a similar analysis for the Semitic noun phrase.

44

Enoch Oladé Aboh

References Abney, S. Paul 1987 The English Noun Phrase in its Sentential Aspect. Ph.D. dissertation, MIT. Aboh, O. Enoch 1998 Focus constructions and the Wh-criterion in Gungbe. Linguistique Africaine 20: 5–50. 2002 La morphosyntaxe de la périphérie gauche nominale. Recherches Linguistiques de Vincennes 31: 9–26. 2004 The Morphosyntax of Complement-Head Sequences. Clause Structure and Word Order Patterns in Kwa. OUP, New York. Alexiadou, Artemis, & Chris Wilder 1998 Adjectival Modification and Multiple Determiners. In A. Alexiadou & C. Wilder, eds., Possessors, Predicates and Movement in the Determiner Phrase. Linguistics Today. Benjamins, Amsterdam. Bernstein, Judy 1997 Demonstratives and reinforcers in Romance and Germanic languages. Lingua 102: 87–113. 2001a Focusing the “right” way in Romance determiner phrases. Probus 13–1: 1–29. 2001b The DP Hypothesis: Identifying Clausal Properties in the Nominal Domain. In M. Baltin & C. Collins (eds.), The handbook of Contemporary Syntactic Theory. 537–561. Blackwell, Oxford. Brody, Micheal 2000 Word Order, Restructuring and Mirror Theory. In Peter Svenonius ed., The Derivation of VO and OV. 27–43. Benjamins, Amsterdam. Burzio, Luigi 1986 Italian Syntax, A Government and Binding Approach. Reidel, Dordrecht. Carstens, Vicky 2000 Concord in Minimalist Theory. Linguistic Inquiry 31: 319–355. Campbell Richard 1996 Specificity Operators in SpecDP. Studia Linguistica 2: 161–188. Chomsky, Noam 1995 The minimalist Program. MIT Press, Cambridge, Mass. 1999 Derivation by Phase. Ms. MIT. Cinque, Guglielmo 1990 Types of A’-Dependencies. MIT Press, Cambridge, Mass. 1994 On the Evidence for Partial N-movement in the Romance DP. In G. Cinque, J. Koster, J.-Y. Pollock, L. Rizzi, R. Zanuttini eds., Paths towards Universal Grammar. 85–110. Georgetown University Press, Washington.

Snowballing Movement and Generalized Pied-piping 1996 1999

45

The “antisymmetric” program: Theoretical and Typological implications. Journal of Linguistics 32: 447–464. Adverbs and Functional Heads, A cross-linguistic perspective. Oxford University Press, London.

Collins, Chris 1996 Local Economy. MIT Press, Cambridge, Mass. Enç, Mürvet 1991 The Semantics of Specificity. Linguistic Inquiry 22: 1–26. Giusti, Giuliana 1994 Heads and Modifiers across Determiners: Evidence from Rumanian. In G. Cinque & G. Giusti eds., Advances in Rumanian. Linguistito Aktuell 10, Benjamins, Amsterdam. Pp.103–125. 1997 The Categorial Status of Determiners. In Liliane Haegeman ed., The New Comparative Syntax. Longman Linguistics Library, London. Greenberg, J. H. 1966 Some Universals of Grammar with Particular Reference to the Order of Meaningful Elements. In J. H. Greenberg ed., Universals of Language (2nd ed.). MIT Press, Cambridge, Mass. Grosu, Alexander 1988 On the distribution of genitive phrases in Rumanian. Linguistics 26: 931–49. Haegeman, Liliane 1995 The Syntax of Negation. CUP, Cambridge. Hawkins, John A. 1983 Word Order Universals. Academic Press, New York, London. Ihsane, Tabea 2003 Demonstratives and reinforcers in Arabic Romance and Germanic. In Jacqueline Lecarme ed. Research in Afroasiatic Grammar II. Benjamins, Amsterdam. Ihsane, Tabea, & Genoveva Puskás 2001 Specific is not Definite. In Generative Grammar in Geneva 2: 39–54 Kinyalolo, Kinsangati K. W. 1995 Licensing in DP in FO~n. Linguistique Africaine 14. 61–92. Koopman, Hilda 1984 The Syntax of Verbs: From Verb Movement Rules in the Kru Languages to Universal Grammar. Foris, Dordrecht. Longobardi, Giuseppe 1994 Reference and proper names: A theory of N-movement in syntax and Logical Form. Linguistic Inquiry 25: 609–665. Manfredi, Victor 1993 Verb Focus in the Typology of Kwa/Kru and Haitian. In Francis Byrne, & Don Winford, eds., Focus and grammatical relations in Creole languages. Benjamins, Amsterdam.

46

Enoch Oladé Aboh

Massam, Diane 2000 VSO and VOS: Aspects of Niuean word order. In Andrew Carnie and Eithne Guilfoyle, eds. The Syntax of Verb Initial Languages. OUP, New York. 2001 Pseudo-noun incorporation in Niuean. Natural Language and Linguistic Theory 19: 153–197. Panagiotidis, Phoevos 2000 Demonstrative determiners and operators: the case of Greek’. Lingua 110: 717–742. Paul, Ileana 2001 Concealed pseudo-clefts. Lingua 111:707–727. Pearson, Matthew 2000 Two Types of VO Languages. In Peter Svenonius ed., The Derivation of VO and OV. 327–363. Benjamins, Amsterdam. Pesetsky, David 1987 Wh-in-situ: Movement and Unselective Binding. In Eric Reuland, & Alice ter Meulen, eds., The Representation of (In)definiteness. MIT Press, Cambridge, Mass. Pollock, Jean-Yves 1989 Verb Movement, Universal Grammar, and the Structure of IP’. Linguistic Inquiry 20: 356–424. Rackowski Andrea & Lisa Travis 2000 V-Initial Languages: X or XP movement and Adverbial Placement. In Andrew Carnie and Eithne Guilfoyle, eds. The Syntax of Verb Initial Languages. OUP, New York. Ritter, Elizabeth 1991 Two Functional Categories in Noun Phrases: Evidence from Modern Hebrew. Perspective on Phrase Structure’. Syntax and Semantics 25. Academic Press, Rothstein, New York. 1995 On syntactic Category of Pronouns and Agreement. Natural Languages and Linguistics Theory 13: 405–443. Rizzi, Luigi 1990 Relativized Minimality. MIT Press, Cambridge, Mass. 1997 The fine structure of the left periphery. In Liliane Haegeman ed., Elements of Grammar. Kluwer Academic Publishers, Dordrecht, Boston, London. 2001 Relativized Minimality Effects. In Mark Baltin & Chris Collins, eds., The Contemporary Handbook of Syntactic Theory. Blackwell, Oxford. Siloni, Tal 1997 Noun Phrases and Nominalizations. Studies in Natural Languages and Linguistic Theory, Kluwer, Dordrecht.

Snowballing Movement and Generalized Pied-piping

47

Shlonsky, Ur 2000 The form of Semitic noun phrases. An antisymmetric, non n-movement account. Ms. Université de Genève. Szabolcsi, Anna 1987 Functional Categories in the Noun Phrase. In Kenesi I ed., Approaches to Hungarian. JATE, Szeged. 1994 The Noun Phrase. In Ferenc Kiefer & Katalin E. Kiss, eds., Syntax and Semantics. The Syntactic Structure of Hungarian 27: 179–274. Ura, Hiroyuki 2001 Local economy and Generalized Pied-Piping. The Linguistic Review 18: 169–191. Vikner, Sten 1997 V-to-I movement and inflection for person in all tenses. In Liliane Haegeman ed., The New Comparative Syntax. Longman Linguistics Library. Longman, London, New York. Zwart, C. Jan Wouter 1997 Morphosyntax of Verb Movement. A Minimalist Approach to the Syntax of Dutch. Studies in Natural Language & Linguistic Theory. Kluwer, Dordrecht.

Optionality at the Interface: Triggering Focus in Romanian Gabriela Alboiu

1. Introduction The present paper focuses on the construal of various focus effects in Romanian. The deeper questions relate to the triggers involved in what appear to be optional dislocation operations in various languages. Given that optionality of movement in one and the same language cannot be reduced to a choice between overt and covert movements, the paper proposes a solution that relies on the division of labour between displacement per se and chain formation/linking for feature valuation. Some of the general questions addressed refer to the role of prosodic properties of phrases and information structure in triggering movement. I show that not all possible triggers for word order phenomena are morphosyntactic in nature and argue that interface factors also play a role. While the main line of inquiry is concerned with contrastive focus operators, rhematic focus constructions are also investigated where appropriate for the general discussion. The paper is organized as follows. Section 2 introduces the puzzle with respect to focus operators and optional dislocation. Section 3 discusses some of the empirical and theoretical problems of optional displacement, while Section 4 focuses on the syntax of contrastive focus. Section 5 provides an analysis of the data from a minimalist perspective, which is strengthened by the discussion of rhematic focus and object shift presented in Section 6. Section 7 is a conclusion.

2. Focusing on Focus Operators Semantically, contrastive focus operators uniquely delimit a member of a presupposed or inferable set for which the predicate phrase can potentially hold. Syntactically, any focus operator that affects the truth conditions of a sentence seems to require special licensing conditions which often include

50

Gabriela Alboiu

movement to an operator position and similar behaviour to wh-phrases in terms of their ability to license operator-variable chains (e.g. Brody 1995; Herburger 2000; Kiss 1995, 1998; Rizzi 1997, in press; Zubizarreta 1998). In some languages, for example Hungarian (see 1), the focus operator is obligatorily associated with a distinct structural position; this position is verb-adjacent and left-peripheral in the clause. (1)

a. Tegnap este MARINAK mutattam be Petert. 1, 2 last night Mary.DAT introduced.1SG AUX Peter.ACC ‘It was to Mary that I introduced Peter last night.’ (not to someone else) b. Tegnap este be mutattam Petert Marinak. last night AUX introduced.1SG Peter.ACC Mary.DAT ‘Last night I introduced Peter to Mary.’ ‘*It was to Mary that I introduced Peter last night.’ (not to someone else) (examples adapted from Kiss 1998)

In (1), the focused phrase MARINAK ‘to Mary’ cannot surface in-situ but has to move to the preverbal operator position; when in-situ, as in (1b), the phrase cannot be interpreted as contrastively focused. The obligatory dislocation of focus operators is not, however, cross-linguistic. In modern Greek and Romance languages, for example, it is well-known that the contrastively focused operator seems to be only optionally associated with a distinct leftperipheral structural position. Interestingly, prosodic marking (heavy stress/ emphasis) on the contrasted phrase is obligatory independent of positioning. Various examples are shown in (2)–(4) below. (2)

Modern Greek (examples from Tsimpli 1995): a. To PETRO simbathi i Maria. the-ACC Petro likes the-NOM Maria ‘It is Petro that Maria likes’ b. Dhanisan to vivlio sto lent-3PL the-ACC book to-the ‘It is to Petro that they lent the book.’

PETRO. Petro

Optionality at the Interface: Triggering Focus in Romanian

(3)

51

Spanish (examples adapted from Zubizarreta 1998): a. LAS PAPAS detesta Juan (y no las espinacas) the potatoes hates Juan (and not the spinach) ‘It is potatoes that Juan hates (and not spinach).’ b. Juan detesta LAS PAPAS (y no las espinacas) Juan hates the potatoes (and not the spinach) ‘It is potatoes that Juan hates (and not spinach).’

(4)

Romanian: a. Pe MIHAIi li-a PE Mihaii CL.3SG.ACC.Mi-AUX.3SG (,nu pe Ion). (not PE Ion) ‘It is Mihai that Victor called, (not Ion).’

strigat Victor called Victor

strigat Victor b. Li-a CL.3SG.ACC.Mi-AUX.3SG called Victor (,nu pe Ion). (not PE Ion) ‘It is Mihai that Victor called, (not Ion).’

ti ti

pe MIHAIi PE Mihaii

The contrastively focused operator may surface in-situ, as in the (2– 4b) examples, but it may also surface in the canonical preverbal verb-adjacent operator position, as in the (2– 4a) examples.3 At this point, it would be worthwhile to ask what triggers displacement of the focused phrase and, furthermore, why optionality of displacement is permitted in some languages. The first question has been discussed extensively in the literature (Chomsky 1971; Kiss 1995, 1998; Rizzi 1997; Rochemont 1986; Zubizarreta 1998 inter alia) and, while there may be disagreement as to the presence or absence of a Focus head in an exploded CP configuration, there is in principle consensus in the current generative literature that the trigger for movement is a formal/morphosyntactic feature [Foc] present in the derivation and responsible for creating an operator-variable chain at some level of representation. However, if morphosyntactic imperfections are the culprits of dislocation, it follows that implementation of movement should be complusory whenever [Foc] is present. Which brings us back to the optionality problem illustrated in (2– 4) above and therefore leaves us with a non-trivial question momentarily unaccounted for. In addition, there is the problem of lack of uniformity among the behaviour of operators in these languages. Consider the Romanian facts in (5a)

52

Gabriela Alboiu

which show the obligatory association of a wh-operator with a well-defined preverbal structural position. This asymmetry with focus operators is even more puzzling given that Romanian is a multiple wh-fronting language in which a wh-phrase is synonymous to displacement, regardless of the number of such phrases, as shown in (5b). (5)

strigat a. Pe cinei li-a PE whoi CL.3SG.ACC.Mi-AUX.3SG called (* pe cinei)? (* PE whoi) ‘Who(m) did Victor call?’

Victor Victor

b. Cuii ce ii-a dat wh-DAT. what CL.3SG.DAT.-AUX.3SG given Mihai (*cui) (*ce)? Mihai (*wh-DAT.) (*what) ‘To whom did Mihai give what?’ In other words, Romanian is a language adamant about dislocating its whoperators but relaxed in terms of positioning its focus operators. Here the questions to ask are why the asymmetry between the two types of operators and what kind of insights does this asymmetry provide into the nature of potential movement triggers.

3. Why Optionality is Inconvenient Until recently, the Chomskian tradition has typically viewed movement as an imperfection.4 Chomsky (1986) suggests that movement is governed by a condition of Last Resort and that Move-alpha will occur only in order to avoid a structure that violates general conditions on well-formedness. Subjects in English, for example, cannot get Nominative Case in-situ and need to undergo movement within the Inflectional domain to Spec,IP, the canonical Nominative position, to avoid a Case Filter violation. In minimalism, this position is redefined as Spec,TP and subject dislocation in English is perceived as a result of feature strength (Chomsky 1995), or the presence of an EPP feature on T (Chomsky 2000, 2001a). The exact status of the ‘EPP’ formal feature is still open to debate but, crucially, in this case it refers to the requirement that subjects (or their expletive associates) have to be in an intimate Spec-Head configuration with T. This is illustrated in (6).

Optionality at the Interface: Triggering Focus in Romanian

(6)

53

a. [TP Mrs. Dallowayi is ti in the kitchen with the flowers]. b. [TP There are always [vP some flowers in the kitchen]]. c. *[TP Are always [vP some flowers in the kitchen]].

No options of in-situ subject and absence of Merge in Spec,TP are permitted for English. Given that movement should be avoided when possible, we do not expect optional displacements in natural languages due to a simple syllogism: if displacement is required to avoid a crash, it should always occur; on the other hand, if lack of movement would not engender a crash, displacement should never occur. Hence the inconvenience of optional displacements. Other possibilities come to mind with respect to optionality of movement. For example, we could assume that languages like Romanian have a choice in checking their [Foc] feature either at LF or pre-LF. This approach has been proposed by a number of authors (e.g. Kiss 1995, 1998; Motapanyane 2000; Tsimpli 1995) who have the merit of being the first to highlight the optionality phenomenon for these constructions. However, among other things, these analyses are faced with the problem of finding a way of explaining why certain languages can optionally avoid Procrastinate given a computational system functioning according to economy principles (see Chomsky 1995 et seq.). In addition, there are other complications with LF feature checking insofar as contrastive focus operators are concerned. On the theoretical side, the first problem that comes to mind is of a general nature. Given that in minimalism narrow syntax is the only computational level available, it follows that all feature-checking/valuation operations should be resolved within this component (see Chomsky’s 2001a claim that all feature-checking is overt). The second theoretical problem is of a more specific kind. Given the Y model assumed in minimalism, and generative grammar more generally, which entails that PF and LF fail to communicate with each other, it follows that feature checking at LF should have no impact on PF; yet regardless of positioning, we have seen that contrastively focused operators require prosodic stress, assigned within the PF component.5 Yet, perhaps the most palpable problem with LF focus valuation comes from empirical data that show contrastive focus reconstructs at LF. Consider the examples in (7): (7)

iubeste a. Pe copilul SAUi îl PE child-the selfi CL.3SG.ACC.M loves ‘It is her own child that any mother loves.’

orice mamai any motheri

t. t.

54

Gabriela Alboiu

b. *Copilul SAUi o iubeste t t child-the selfi CL.3SG.ACC.F loves ‘*It is her own child that loves any mother.’

pe orice mamai. PE any motheri.

In both (7a) and (7b), the anaphor SAU ‘self’ is assigned heavy stress which forces a contrastive focus interpretation of the fronted direct object. The dislocated phrase containing the anaphor is not c-commanded by its appropriate binder in either (7a) or (7b). Note, however, that the trace of the focused phrase is felicitously c-commanded by an appropriate binder in (7a) though not in (7b). Given that (7a) is well-formed, while (7b) is ungrammatical, suggests that at LF – where binding relations hold (Chomsky 2000) – the focused constituent ‘reconstructs’ and the anaphor is thus appropriately licensed. Assuming the copy theory of movement developed in minimalism, the above facts show that for contrastively focused phrases it is the lower copy that is relevant at LF (i.e. the ‘tail’ of the chain in the sense of Hornstein 1995). In conclusion, if for LF purposes, it is the lower copy that counts, covert displacement for feature checking has to be ruled out as it would engender a contradiction at this level of representation.6 At this point, I conclude that focus feature checking at LF has to be ruled out and that either (i) feature checking/valuation is prior to LF or (ii) there is no feature-checking/valuation with contrastive focus operators in Romanian. The absence of a formal feature [Foc] with these constituents would imply the absence of A-bar effects typically associated with operator chains and would also leave unexplained the obligatory verb-adjacency requirement with fronted focused phrases. In the next section, I discuss syntactic properties of contrasted constituents and highlight two major facts: (i) focused constituents show identical syntactic behaviour irrespective of positioning and (ii) focused constituents form operator-variable chains. These results point to the obligatory presence in the derivation of a formal [Foc] feature whenever a contrastive focus semantics is at stake.

4. The Syntax of Contrastive Focus So far, we have established that for Romanian a contrastive focus interpretation obtains regardless of whether the heavily stressed phrase occurs insitu or has moved to the verb-adjacent operator position. In addition, in the previous section I showed empirical evidence that for LF purposes it is the in-situ position that counts, which strengthens the theoretical conclusion that LF feature checking cannot be an option. The next step is to investigate

Optionality at the Interface: Triggering Focus in Romanian

55

syntactic properties of focused phrases in order to determine whether feature checking is involved and, if so, for which cases. Specifically, I discuss evidence from weak crossover and parasitic gap licensing that shows feature checking is operative with contrastive focus in Romanian regardless of surface positioning.

4.1. Weak Crossover Effects The data in (8) show that contrastively focused elements in Romanian induce weak crossover effects whether they surface in-situ, as in (8b), or in the preverbal verb-adjacent position, as in (8c): (8)

dat un mar a. Mama luii ii-a mother-the hisi CL.DATi-AUX.3SG given an apple copiluluii. child-the.DATi ‘Hisi mother gave the childi an apple.’ b. *Mama luii ii-a dat un mar mother-the hisi CL.DATi-AUX.3SG given an apple COPILULUIi. child-the.DATi ‘*It is to the childi that hisi mother gave an apple.’ c. *Mama luii COPILULUIi ii-a dat mother-the hisi child-the.DATi CL.DATi-AUX.3SG given un mar ti. an apple ti ‘*It is to the childi that hisi mother gave sweets.’

Lasnik and Stowell (1991) argue that weak crossover is a distinctive characteristic of quantificational A-bar relations. The ill-formedness of both (8b) and (8c) indicates that the formation of a quantificational A-bar chain is equally involved with both in-situ and preverbal focus. Compare with the grammatical counterpart in (8a) where the indirect object copilului ‘to-thechild’ is left unfocused and, consequently, fails to induce a weak crossover violation as it does not create an operator-variable chain. I conclude that feature checking is involved regardless of surface positioning of the focus operator.7

56

Gabriela Alboiu

4.2. Parasitic Gap Licensing According to Engdahl (1983), parasitic gaps are licensed by a variable which is the result of overt movement. Engdahl’s (1983: 22) examples in (9) show that wh-in-situ does not license parasitic gaps, which has been generalized as an implication that covert movement in general fails to license these empty categories. (9)

a. Which article1 did you [file _1] [without reading _PG] b. *Who [filed which paper] [without reading _PG]

Engdahl’s empirical facts are strengthened by recent theoretical claims vis-àvis the licensing of these gaps. Nissenbaum (2000) has argued that only a modified-predicate configuration, as in (10), can license parasitic gaps. (10) Modified-predicate configuration (Nissenbaum 2000: 117) o vP rp vP DP 3 : vP Adjunct # 5 5 # # …tDP … Oj … ej … # z-----------m The author further proposes that Engdahl’s generalization is explained by a general constraint on movement that forces this modified-predicate configuration to be derived in the overt syntax. Specifically, the DP has to adjoin to the vP prior to insertion of the adjunct or else it should ‘tuck in’ below it (following proposals in Richards 1998); however, due to the fact that merge of the adjunct is an overt operation, it follows that movement of the DP has to take place prior to Spell-Out too (i.e. within narrow syntax and not at LF). Given that variables are the result of A-bar chain formation, licensing of parasitic gaps provides evidence for the construal of operator-variable chains. The Romanian data in (11) shows that parasitic gaps are licensed by both left-peripheral and in-situ contrastive focus which further strengthens the claim made in the previous section that [Foc] feature checking is involved regardless of word order.

Optionality at the Interface: Triggering Focus in Romanian

(11) a. *A mîncat supa [fara sa AUX.3SG eaten soup [without SUBJ ‘*S/he ate soup without heating.’

57

încalzeasca _PG]. heat _PG]

b. Nu, CHIFTELE a mîncat [fara sa no, meatballs AUX.3SG eaten [without SUBJ încalzeasca _PG] (, nu supa)! heat _PG] (, not soup) ‘No, it’s meatballs that s/he ate without heating, not soup!’ c. Nu, a mîncat CHIFTELE [fara sa no, AUX.3SG eaten meatballs [without SUBJ încalzeasca _PG] (, nu supa)! heat _PG] (, not soup) ‘No, it’s meatballs that s/he ate without heating, not soup!’ In (11) only, the presence of a parasitic gap yields well-formed sentences as in (11b) and (11c) which contain a contrastive focus and, consequently, as operator-variable chain. On the other hand, the ungrammaticality of (11a) suggests the absence of an operator-variable chain, expected given the absence of heavy stress on the direct object and any contrastive semantics. Note that I have only taken licensing of parasitic gaps and weak crossover phenomena with contrastive focus to indicate the construal of an A-bar chain due to the presence of an unvalued [Foc] feature that needs checking/ valuation. However, I have said nothing so far about ‘movement’ per se. In the next section, I provide an analysis that explains focus feature valuation solely in terms of the operation Agree (see Chomsky 2000, 2001), without any recourse to movement within the domain of the Probe triggering Agree. However, due to cyclic Spell-Out, I suggest phasal movement is involved with both in-situ and left-peripheral contrastive focus, which is in line with Nissenbaum’s requirement that parasitic gaps are licensed via predicate modification. Last but not least, I propose an account of displacement to the preverbal operator position devoid of optionality construed as feature checking.

5. A Minimalist Analysis of Contrastive Focus The present analysis relies on proposals made in Chomsky (2000, 2001a and 2001b) from which I shall be adopting the following assumptions with respect to feature checking/valuation. Uninterpretable formal features enter

58

Gabriela Alboiu

the derivation without a value and need to be valued in order to avoid a crash at the interface levels. Under this perspective, narrow syntax/the generative engine of language is driven solely by these interface conditions. Valuation (i.e., feature checking) can only occur against a matching feature with a value and is always the result of an Agree operation. The Agree operation is esentially a search and validation mechanism designed to hold between a probe with the unvalued formal feature and a goal with the matching valued feature. Nonetheless, in order for Agree to proceed felicitously, the goal itself has to be ‘active’. Specifically, it has to have some unvalued feature which is valued by a feature within the probe initiating the search if the probe has a value for this feature; if this is not the case, the goal remains active until it establishes an Agree operation with a probe that can value its feature. Crucially, note that feature valuation incumbent on Agree does not in and of itself entail any sort of dislocation mechanism, which is an important departure from earlier minimalist assumptions. However, I do assume that feature valuation via Agree establishes a chain/link with A or A-bar properties depending on the features valued. Displacement to the specifier of the probe is the result of semantic effects due to an EPP/OCC feature present on the probing head. As discussed in section 3, this is in essence a requirement that a phrase must be an occurrence of some probe. Chomsky (2001b) argues that optimally, a head should only have this feature if displacement will yield new scopal or discourserelated properties. However, this is not the case for English subject related EPP, which is compulsory yet lacks any interpretive effect, though it is the case for topicalization phenomena derived by movement for example. In fact, I suggest below that EPP/OCC is only associated with an interpretive effect when optional.

5.1. Contrastive Focus Valuation All current studies on Romanian agree that Romanian is VSO in the sense that Spec,TP is not required to host subjects (see Alboiu 2002; Cornilescu 2000a; Dobrovie-Sorin 1987, 1994; Hill 2002). Specifically, Case is checked via long-distance Agree and there is no subject externalization to Spec,TP in the usual EPP sense. For example, in (12), the Nominative subject ‘the student’ can occur either postverbally (12a) or preverbally (12b), with no requirement of verb-adjacency when dislocated within the left periphery.8 When preverbal, the subject is interpreted as topic and can precede wh-phrases (12c).

Optionality at the Interface: Triggering Focus in Romanian

59

(12) a. A sosit studentul. AUX.3SG arrived student-the ‘The student has arrived.’ b. Studentul (chiar acuma) a sosit (chiar acuma). student-the (just now) AUX.3SG arrived (just now). ‘The student arrived just now.’ c. Studentul cînd a student-the when AUX.3SG ‘When did the student arrive?’

sosit? arrived

Consequently, Spec,TP is assumed to host Romanian sentence-initial operators, such as for example contrastively focused constituents and wh-phrases (Alboiu 2002; Cornilescu 2002; Hill 2002).9 If present in the derivation, the uninterpretable/unvaled [Foc] formal feature parasitically incorporates onto T forming a syncretic category. From within T, it will initiate a search and ‘check off’ in the presence of a valued contrastively focused goal, via the operation Agree. Let us consider example (13) in which ‘car’ is contrastively focused. The subject ‘Victor’ is topicalized and dislocated to the left periphery of the clause, while the contrasted constituent is shown both in the preverbal and postverbal positions. (13) Victor (MASINA) vrea [vP (MASINA)], Victor (car) want.3SG [vP (car)] ‘It’s a car that Victor wants, not a house.’

nu not

casa. house

The lexical array for the derivation in (13) contains, among other features, an uninterpretable focus feature on T, u[Foc], and an uninterpretable focus feature on the lexical item ‘car’, u[Foc], which makes the lexical item active for match.10 u[Foc] on T probes for a matching feature and finds u[Foc] on ‘car’ in its search space. The derivation converges at the interfaces which is synonymous to valuation of all uninterpretable features. However, neither [Foc] feature has a value, so how is convergence ensured given that valuation of an uninterpretable feature can only take place against a matching interpretable one (Chomksy 2001a)? Recall the obligatory heavy stress assigned to contrasted constituents and how in the absence of this stress the semantics and syntax of contrast are absent. Consequently, we need to assume that prosodic stress is the manifestation of the required value for the [Foc] feature on the lexical item. The problem is that the syntactic component

60

Gabriela Alboiu

is blind to phonological features such as ‘stress’, so valuation cannot proceed within narrow syntax; presumably the computational system can recognize formal features and distinguish between valued and unvalued cases but nothing more. In Alboiu (2003), it is suggested that [Foc] valuation has to apply at Spell-Out, specifically, at the intersection between syntax and phonology; this is contra to earlier proposals in Chomsky (2001a: 5) which view Spell-Out occurring “shortly after the uninterpretable features have been assigned values”. However, it is in line with recent proposals which argue that all valuation has to occur at Spell-Out/TRANSFER (i.e., at the point where the derivation is handed over to the interface levels) for various technical reasons (Chomsky 2001b; Epstein and Seely 2002), among which the fact that “shortly” is not only ambiguous but also both too early and too late. Consequently, there is no problem with ‘stress’, a phonological feature, valuing [Foc], a formal feature. Mapping to the phonological and semantic components proceeds cyclically (Chomksy 2001a,b; Epstein and Seely 2002; Uriagereka 1999, among others), where the cycles are phasal (i.e., vP and CP). Chomsky (2001a,b) suggests that the left-edge of the phase need not be spelled out together with the phase; specifically, only the domain of a phasal head has to be spelled out and thus becomes ‘impenetrable’ (PIC). I propose that the lexical item marked u[Foc] has to raise to the left-edge of vP to avoid being spelled out within the vP phase.11 The reasons for this proposal are both empirical and theoretical. On the empirical side, if the lexical item with u[Foc] were not to raise, it would be spelled out within vP and thus incapable of ever being spelled out in Spec,TP (i.e., in the ‘canonical’ verb-adjacent operator position). On the theoretical side, PIC makes v and its sister inaccessible to probing by T, so if focused phrases are to enter Agree with T, they have to raise to the left edge of vP. To sum up, I suggest that lexical items with u[Foc] vP adjoin in their ‘insitu’ instantiation, as in (14). Once Agree is established with T, the relevant A-bar effects ensue regardless of whether displacement to Spec,TP occurs or not, in accord with Nissenbaum’s proposal. (14)

o vP rp DP vP u[Foc] 5 … tDP …

Optionality at the Interface: Triggering Focus in Romanian

61

5.2. Displacement to Spec,TP and OCC I have shown that catering to contrastive focus is satisfied by the operation Agree which matches [Foc] features on probe and goal subsequently valued at the point of Spell-Out. Consequently, the [Foc] feature itself cannot trigger displacement of the focused phrase to Spec,TP. What then is responsible for the ‘optional’ movement operation into the left-periphery of these operators? The obvious candidate is an EPP/OCC feature associated with T. So, for instance, we can assume that T optionally enters the derivation with an OCC feature and in precisely those cases, the focused phrase will dislocate to Spec,TP. Leaving aside for the moment conceptual concerns with respect to the nature of the OCC feature, notice that if OCC is construed as a feature of T, it could in principle trigger movement of any XP valuing features of T. Given that the uninterpretable phi-features of T are valued by the subject, we may ask why the subject itself is incapable of valuing this feature when fronted as in (13) where we see that the focused phrase may also optionally front? One possibility is to assume that, in the presence of OCC on T, all goals of T undergo displacement to Spec,TP. But the availability of in-situ focus in (13) shows that this is not the case. A second possibility is to assume that OCC on T triggers displacement of the first goal, say the subject, but allows for optional displacement of any additional goals, for example focus. This is incorrect at least because of the following: (i) there is no reason to suspect phi-features in T check prior to the [Foc] feature, thereby granting the subject privileged status with respect to OCC on T. In fact, if feature valuation is a property of Spell-Out, tiered checking is ruled out. (ii) the behaviour of focused phrases in the presence of wh-phrases to be discussed below. I suggest that, in fact, the OCC feature is a sub-feature/property construed on features themselves as opposed to a feature of the lexical/functional items. We will see that such an approach helps explain cross-linguistic and intra-language idiosyncrasies, while at the same time conforming to Chomsky’s (2001b) proposal that OCC should be semantic in nature.

5.2.1. OCC and wh-Phrases In section 2, I showed that wh-operators are obligatorily associated with the left-peripheral structural position in Romanian, while focus operators can surface either preverbally or in-situ. Let us suppose that this asymmetry is due to the obligatory presence of OCC with interrogation and its optional occurrence with [Foc] in Romanian. Cross-linguistically, wh-phrases are

62

Gabriela Alboiu

inserted in the Numeration with an interpretable [Q] feature and an uninterpretable [wh] feature (Chomsky 2000). Their uninterpretable feature makes them active for match with a functional head probing for interpretable [Q] (i.e., C or C/T, as in Romanian, with uninterpretable [Q]). The obligatory pronunciation of the upper copy with wh-movement in Romanian can be equated to the obligatory presence of EPP/OCC with [Q], presumably as part of the manner in which the [Q] feature is encoded in the Romanian lexicon.12 Consider next the behaviour of focus operators in derivations containing interrogative phrases in Romanian. The data in (15) show that you cannot have a fronted wh-phrase and a fronted focus simultaneously.13 (15) Cei (*COPILULUI) i-a whati (*child-the.DAT) CL.3SG.DAT.M-AUX.3SG spus el COPILULUI ti (,nu vecinei)? said he child-the.DAT ti (,not neighbour.DAT) ‘What is it that it is to the child that he said, not to the neighbour?’ Despite the impossibility of simultaneous pronunciation in the preverbal field, there is evidence from weak crossover effects that a [Foc] feature is present in the derivation, even in the presence of [Q]. This is shown in in (16): i-a spus (16) a. Cei whati CL.3SG.DAT.M-AUX.3SG said copiluluij ti ? child-the.DATj ti ‘What did hisi mother say to the childi ?’

mama luij mother hisj

b. *Cei i-a spus mama luij whati CL.3SG.DAT.M-AUX.3SG said mother hisj COPILULUIj ti (, nu vecinei)? child-the.DATj ti (, not neighbour.DAT) ‘*What is it that hisi mother said to the childi (, not the neighbour)?’ (16a) is grammatical, given that copilului ‘to the child’, which is coindexed with a pronoun to its left, does not form an A-bar chain and implicitly, does not leave behind a variable engendering a weak crossover effect. On the other hand, (16b) in which the indirect object COPILULUI ‘to the child’ is contrastively focused, is not well-formed. The ungrammaticality of this example shows that a weak crossover effect is triggered in the presence of the in-situ contrastive focus in interrogative clauses. This A-bar effect can

Optionality at the Interface: Triggering Focus in Romanian

63

only be explained if we assume that the focus operator enters an Agree operation with a higher probe, thus acting like a variable which in (16b) is illicitly coindexed with a pronoun to its left. Example (16) highlights the fact that when both a wh-operator and a focusoperator enter a checking/valuation relationship with C/T, only the wh-phrase can and must undergo movement to the verb-adjacent operator position. The fact that Romanian has multiple wh-movement (see 5b) shows it allows for multiple occurrences of the probe and exclusion of fronted focus in (16) cannot be taken to represent a PF ban on multiple specifiers. The empirical facts can be captured under the present proposal that views OCC as an inalienable property of [Q] in Romanian, but an optional property of [Foc].

5.2.2. OCC as a Property of SEM rather than PHON It is not unreasonable to assume that languages differ as to whether the encoding of OCC is part of the lexicon, in which case displacement is obligatory and devoid of interpretive effects (e.g. subject related OCC in English, [Q] related OCC in Romanian14) or OCC is an instruction from the semantic component, in which case it is optionally inserted in the numeration/lexical array and has interpretive impact independent of the feature it encroaches on. This perspective requires a relaxation of Chomsky’s view that OCC should only be present if it yields new scopal or discourse-related properties but it is closer to the empirical facts. At the same time, it maintains the spirit of Chomsky’s proposal by assuming that whenever internal Merge (i.e. displacement to some Spec position) is optional, it should provide new interpretations. In the case of contrastive focus, I assume that displacement to Spec,TP is triggered by an OCC feature encroached on [Foc] in T. In the presence of this feature internal Merge to Spec,TP becomes obligatory and the contrastively focused constituent is spelled out in the left periphery of the clause. What interpretive effect is associated with this displacement but not with the in-situ focus? While both preverbal and in-situ focus operators are interpreted as contrastive phrases, Hill (2002) suggests that in Romanian preverbal focus operators are stylistically more emphatic than their in-situ counterparts. Consequently, OCC on [Foc] serves a stylistic purpose and thus contributes to an outcome at semantics not otherwise expressible. Optionality then is not related to feature valuation but to the interfaces. We could in principle assume the omnipresence of OCC with [Foc]. This would entail obligatory displacement and internal Merge of Spec,TP with

64

Gabriela Alboiu

contrastively focused operators (see discussion in Alboiu 2003). PF would then have two copies to choose from: the vP-adjoined copy and the copy merged as Spec,TP. Bobaljik (2002) and Bo‰koviç (2002) have both argued that PF has access to both copies derived via displacement. Among other things, evidence for this comes from languages with multiple wh-movement like Bulgarian and Romanian which require pronunciation of a lower copy to avoid a violation at the PF interface. Consider (17) from Bo‰koviç (2002: 365) which the author argues to be a result of a PF constraint against consecutive homophonous wh-phrases: (17) a. Ce what

precede precedes

ce? what

b. *Ce ce precede? what what precedes ‘What precedes what?’ However, given the availability of both preverbal and in-situ focus, choice of word order cannot be assumed to follow from a PF violation. Nonetheless, if pragmatics and the stylistic component are perceived as PF properties (see Chomsky 2000; Holmberg 1999; Kidwai 1999, inter alia), the omnipresence of OCC can still be maintained with [Foc]. I suggest this is not the correct avenue to pursue given evidence that pragmatic displacements (i.e., the presence of OCC) can affect binding relationships in Romanian, as discussed in Alboiu (1999) and the following section. OCC as a requirement of the semantic component can account for both interpretive effects and effects of the syntax-pragmatics interface to which PF is blind. It seems to me desirable to maintain a uniform account of OCC and, consequently, displacements more generally. 15

6. Independent Evidence for Optional OCC in Romanian: Object Shift In this section, I discuss evidence for object shift in Romanian and focus on the syntactic and pragmatic effects of constructions with displaced objects to the left edge of vP. I first present the data and next provide an analysis based on the occurrence of an optional OCC feature.

Optionality at the Interface: Triggering Focus in Romanian

65

6.1. Romanian VOS Sequences First let us notice that in Romanian preverbal nominals are constrained semantically to discourse-linking (compare 18a with 18b), while postverbal nominals can be both definite and indefinite (see 18c). In addition, (18c) highlights the availability of two word order sequences for postverbal objects regardless of semantic type; specifically, VOS and VSO. l-a (18) a. Inelul ring-the CL.3SG.ACC.M-AUX.3SG ‘Smeagol found the ring.’ gasit b. *Un inel a a ring AUX.3SG found ‘Smeagol found a ring.’ c. (Smeagol) a gasit (Smeagol) AUX.3SG found inelul / un inel (Smeagol). ring-the / a ring (Smeagol) ‘Smeagol found the ring/a ring.’

gasit found

Smeagol. Smeagol

Smeagol. Smeagol (Smeagol) (Smeagol)

Gierling (1997) and Alboiu (1999, 2002) independently show that, in Romanian, objects can undergo movement to a position outside of the vP domain but below T (i.e., not within the preverbal domain). Alboiu (1999, 2002) argues that VOS constructions are derived from VSO sequences and involve object raising across the subject and left-adjunction to vP for de-rhematization purposes. Let us briefly look at the data in (19). (19) a. Mi-au daruit mereu copiii mei CL.1SG.DAT-AUX.3PL bestowed always children-the my florile din curte. flowers-the from garden ‘My kids have always bestowed upon me the flowers in the garden.’ b. Mi-au daruit florile din curte CL.1SG.DAT-AUX.3PL bestowed flowers-the from garden mereu copiii mei. always children-the my ‘My kids have always bestowed upon me the flowers in the garden.’

66

Gabriela Alboiu

The VSO word order in (19a) is not pragmatically synonymous to the VOS word order in (19b). In VSO word order sequences, both the subject and the object are understood as new information, rhematic focus, while in VOS constructions, the object is de-focused and understood as part of the presupposition (i.e., the theme) together with the verb, while the in-situ subject is maximally rhematic.16 In conclusion, nominals that represent new information remain within their initial Merge position within the vP, while nominals that are part of the presupposition ‘evacuate’ the vP. While vP-adjoined nominals typically denote given information, contrastive focus stress and interpretation are also available (but not required) in this intermediary position. Consider (20), which confirms these facts: (20) a. Îmi daruie FLORI mereu, (nu bomboane). CL.3SG.DAT bestow flowers always, (not sweets). ‘It’s flowers that they are always bestowing upon me, not sweets.’ b. Îmi daruie flori mereu, (* nu bomboane). CL.3SG.DAT bestow flowers always, (* not sweets). ‘They are always bestowing flowers upon me.’ The vP-adverb ‘always’ shows that the object DP in (20) has undergone displacement outside its initial Merge domain. The contrastive focus interpretation obtains in the presence of heavy stress only (i.e., in 20a). The contrast in interpretation between (20a) and (20b) is correlated with a contrast in what triggers displacement. In (20a), displacement si triggered by the presence of an unvalued [Foc] feature on the object DP, which ensures movement of the lexical item outside of the vP phasal domain under cyclic Spell-Out. As discussed in section 5.1, the u[Foc] feature on the contrastively focused lexical item cannot be valued within the vP domain, and consequently, this constituent is not interpretable at the interface levels at this point but has to wait until the next phasal level.17 On the other hand, object raising in (20b) is triggered by the pragmatic requirement that the presupposed object evacuate the vP. Given that this type of movement need not have a contrastive focus correlate, it cannot be viewed as the result of focus feature movement and is not triggered by the presence of unvalued [Foc]. The object raising analysis argued for in Alboiu (1999, 2002) for VOS constructions is theoretically supported by a Kayne (1994) type approach which views linear order construed as hierarchical structure. Empirical support for an object shift/scrambling account is adduced from the reversal of

Optionality at the Interface: Triggering Focus in Romanian

67

binding interactions between the subject and the object DP, condition C effects, and stranded quantifier availability. In addition, effects on Principle A and the absence of weak crossover phenomena point towards properties associated with A-chains as opposed to A-bar chains. Note too that object raising in VOS constructions does not license parasitic gaps (see 21), which again points to an A-type (rather than a quantificational A-bar type) movement account, which is to be kept distinct from contrastive focus displacements. (21) *A mîncat bomboane fata [fara sa desfaca _PG] AUX.3SG eaten sweets girl-the [without SUBJ open _PG] ‘The girl ate sweets without unwrapping them.’

6.2. Object Shift and OCC Holmberg (1999) has recently discussed various object shift constructions from a minimalist perspective. The author extensively argues for movement of non-focused (understood as non-rhematic) constituents into the Diesing (1992) type presupposition domain outside the vP. Given that the properties of these dislocated constituents do not indicate either A-bar or consistent Amovement effects in Scandinavian, Holmberg (1999) concludes, following insights in Chomsky (1995), that object shift is best viewed as a PF-operation within the stylistic component.18 Chomsky (2001a), on the other hand, argues against object shift as a PF phenomenon and proposes that this movement is to be kept within the syntactic component (i.e., narrow syntax) given its interpretive effects. In Romanian, non-focused object shift to the vP edge cannot be viewed as a PF operation due to the effects it has on Principles A and C, both of which are operative at LF. In this language object shift in VOS constructions is best viewed as an instance of EPP/OCC valuation.19 In this sense, our discussion here is not only in line with recent minimalist proposals (Chomsky 2001a,b) in arguing against object shift as purely PF, but also has the merit of clarifying the somewhat erroneous empirical assumption made in Chomsky (2001a) with respect to the general unavailability of object shift in Romance. Not only is object shift available in certain Romance languages but where present it cannot be construed as a PF operation.20 Object shift in Romanian is to be construed as an instantiation of a vrelated OCC feature inserted into the lexical array due to an instruction from SEM. In the presence of this feature, the direct object valuing phi-features

68

Gabriela Alboiu

of v will also value the OCC feature and undergo displacement to the extra edge position of v. The new interpretation resulting from checking of OCC by internal Merge is that of de-rhematization of the object. In the absence of the OCC feature associated with v (or, rather, phi-features of v), the uninterpretable phi-features of v will be valued by the direct object solely via the operation Agree; in this case there is no dislocation and no interpretive effect of presupposition associated with the object. 6.3. In Sum The discussion in the previous two sections has focused on the optional availability of an OCC feature encroached on v in Romanian, availability construed as a requirement of the semantic component rather than PF, in view both of the interpretive effects associated with displacements to the vP edge and resulting syntactic effects reminiscent of A-chains. Given that the OCC feature on v is intimately linked to phi-feature valuation of this head, it is not surprising that displacement effects associated with object shift should show A-movement properties. In a similar vein, contrastively focused constituents can value the [Foc] formal feature in-situ (understood as outside of the initial Merge phase) or in Spec,TP. This dual availability was correlated in Section 5 to the absence versus presence, respectively, of an OCC feature associated with the u[Foc] feature in T. In the absence of this feature, no displacement with internal Merge as Spec,TP is possible, while the presence of this feature triggers obligatory preverbal movement of the contrastively focused operator. However, regardless of this feature, the [Foc] formal feature is valued at SpellOut in the presence of heavy stress, thereby inducing the expected effects of A-bar chains. Semantico-pragmatic displacements are thus triggered by the optional presence in the lexical array of an OCC feature which has to be valued at the point of Spell-Out/TRANSFER on a par with any other feature. Crucially, OCC is responsible for any displacements that are not PF-related and for interpretive effects beyond what is made available by the other formal features valued in that cycle.21 7. Conclusions This paper sets out to investigate the triggers behind optional displacements in natural languages by concentrating on various focus effects in Romanian.

Optionality at the Interface: Triggering Focus in Romanian

69

Much of the discussion centres on properties of focus operators whose semantic and prosodic effects are quite stable in spite of (at least) two linearization possibilities available in this language. Specifically, I show that contrastively focused phrases may but need not dislocate to a preverbal operator position despite evidence that the syntactic component treats both occurrences in the same manner. I argue that A-bar chain/linking effects with both in-situ and dislocated focus operators indicate the presence of an uninterpretable [Foc] feature valued at Spell-Out regardless of word order. I next discuss why optional displacements are problematic in a theory of feature checking which views the syntactic component as nothing more than a generative engine of language whose sole attribute and ability should be to identify unvalued features and to establish linking between probes and goals with the outcome of ensuring valuation and well-formed objects at the two interface levels. I assume the EPP/OCC feature to be the trigger of displacements and suggest that, in fact, there is no optionality of feature checking or of displacements per se but solely optionality of what is inserted into the lexical array. In order to explain cross-linguistic and intra-linguistic asymmetries, I propose a split between features which are parametrically defined as encoding OCC within the lexicon and features which do not encode OCC in their build up. For the former, valuation will of necessity be incumbent on dislocation and consequently, trigger obligatory displacement devoid of interpretive effects, as illustrated by subject related OCC in English and [Q] related OCC in Romanian. For the latter, feature valuation proceeds felicitously regardless of OCC but novel interpretive effects will arise in the presence of OCC (e.g., OCC related to contrastive focus or object shift in Romanian). Specifically, I argue that optional displacements are due to the optional association of an OCC feature with a specific formal feature, which formal feature establishes an A or A-bar link between the probe and goal regardless of the presence of OCC. The presence of an OCC feature simply ensures that the goal becomes an occurrence of the probe via internal Merge and in so doing is interpreted within the probe’s semantico-pragmatic domain. Given that all optional displacements are associated with interpretive effects, I argue that the optional insertion of the OCC feature in derivations is an outcome of interface requirements which cannot be due to PF but have to be related to the semantic component of language. Hence optionality at the interface in triggering various focus readings in Romanian.

70

Gabriela Alboiu

Acknowledgements I’d like to thank Alexandra Cornilescu, Elizabeth Cowper, Rose Marie Déchaine, Jila Ghomeshi, Virginia Hill, István Kenesei, Denis Liakin, Diane Massam, László Molnárfi, Luigi Rizzi and my anonymous reviewer, as well as the audience at the Triggers Workshop for useful comments and discussion. Nonetheless, all errors are mine. This research was partially supported by SSHRC grant # 756-2002-0126, which I gratefully acknowledge.

Notes 1. I use upper case letters to mark contrastively focused elements. 2. The abbreviations used in the example sentences are: AUX: auxiliary, SUBJ: subjunctive, CL: pronominal clitic, SG: singular, PL: plural, NOM: Nominative case, ACC: Accusative case, DAT: Dative case, M: masculine, F: feminine. ‘PE’ is a preposition associated with Romanian direct objects that have an type denotation (see Cornilescu 2000b). 3. That this preverbal position is adjacent to the verbal complex (i.e., verb and clitic cluster) is shown in (i) for Romanian, which is ungrammatical given that the subject phrase intervenes between the focused phrase and the verbal complex. (i) *Pe MIHAIi Victorj li-a strigat tj ti (,nu pe Ion). PE Mihaii Victorj CL.3SG.ACC.Mi-AUX.3SG called tj ti (not PE Ion) ‘It is Mihai that Victor called, (not Ion).’ 4. Chomsky (2001b) is a departure from this view. 5. Kidwai (1999) also questions LF focus feature checking and claims that PF cannot ‘see’ into LF and ‘know’ it has to assign heavy stress. 6. As highlighted by the reviewer, this argument applies in the same way to whmovement, which strengthens the claim that dislocation for feature-checking should be dealt with prior to the LF component. 7. It is beyond the scope of this paper to discuss the quantificational versus nonquantificational nature of all A-bar chains involved with various operators in Romanian – for such a discussion and the requirement that definite direct objects must be clitic doubled in this language, see Alboiu (2002). 8. Given that the lexical verb always raises to T in Romanian (e.g. Dobrovie-Sorin 1987, 1994 et seq.), ‘postverbal’ refers to material below T. 9. Hill (2002) maintains Spec,CP as the locus of wh-movement in Romanian, while Alboiu (2002) and Cornilescu (2002) argue that wh-phrases are hosted in Spec,TP on a par with proposals made for Spanish in Zubizarreta (1998). It is

Optionality at the Interface: Triggering Focus in Romanian

10.

11.

12.

13.

14.

15.

16.

71

not unreasonable to assume that finite T forms a syncretic category with C given the lack of distinctness between the two in Romanian and the otherwise phasal status of finite TP. This in turn might be correlated to the interpretive properties of EPP/OCC in T, unavailable, for example in a language like English, where T and C are distinct heads. Note that the [Foc] feature on the lexical item cannot be assumed to be interpretable, as the information with respect to contrastivity is not part of what is stored in our mental lexicons (see also discussion in Alboiu 2002, 2003). This is no different from what must be the behaviour of other active lexical items with uninterpretable features assuming a cyclic system. In addition, it helps explain why sentences like (i), highlighted by the reviewer, are well-formed and void of any weak crossover effects: (i) PRINTULUIi (ii)-a dat [vP ti [vP prince-the.DATi (CL.3SG.DAT.Mi)-AUX.3SG given [vP ti [vP un inelj [vP mama luii tv [VP ti tV tj ] (nu cersetorului). a ringj [vP mother-the hisi tv [VP ti tV tj ] (not beggar-the.DATi) ‘*It is to the princei that hisi mother gave a ring (, not to the beggar).’ The VP-internal trace coindexed with the pronoun to its left is not a variable, while the vP-adjoined trace (in bold) is. The VP-internal copy of the focused constituent is in a cycle which undergoes Spell-Out prior to [Foc] feature checking and the creation of an A-bar relation. Chomsky (2001b) argues that each language makes a one time selection of a subset [F] of formal features and a one time assembly of [F] as its lexicon. I take the obligatory encoding of OCC on various features to occur at this stage. Rizzi (p.c.) notes the obligatory ‘echo’ reading of this example. Nonetheless, the wh-phrase undergoes movement to the preverbal operator position and, in doing so, obviates optionality of focus pronunciation site. Note that, just as Nominative Case cannot be due to EPP/OCC, neither can the semantics of wh-phrases be due to OCC given the occurrence of wh-in-situ languages. Relegating pragmatics to the PF component has always been suspicious to my mind. Why would the articulatory perceptual level care about linearizations with an interpretive effect? Chomsky (2001a,b) represents a welcome renunciation of this perspective and a more robust proposal with respect to the syntax-pragmatics interface. Clarification of concepts is required at this point. New information/presentational/rhematic focus is to be kept distinct from contrastive focus discussed so far. The former category of focus covers material that represents information newly introduced in the discourse and is the opposite of given/old information, realized by the theme. Contrastive focus, on the other side, is presupposed information, part of what is given and consequently, part of the thematic domain. The distinct semantico-pragmatic properties are parallelled by distinct syntactic properties, as shown in Table 1.

72

Gabriela Alboiu Table 1. Contrastive focus versus rhematic focus

17.

18.

19. 20.

21.

A-bar chain effects

[Foc] formal feature

Prosodic marking

Affects truthfunctional values of S

contrastive focus

+

+

+

+

rhematic focus

–

–

–

–

Languages use various ways of encoding sentence pragmatics (i.e., the themerheme disctinction). For example, English uses intonation to differentiate different theme-rheme partitions, but preserves a constant syntactic structure. Catalan (Vallduví 1995) and Romanian (Alboiu 2002), on the other hand, exploit syntactic structure to package discourse information. Recall that the functional head bearing the feature [Foc] in Romanian is T. In addition, note that (20a) is an instance of contrastive focus ‘in-situ’ given that the focused constituent does not undergo internal Merge to Spec,TP. Holmberg (1999) is careful to note that absence of parasitic gap licensing and lack of Principle A effects is just a manifestation of object shift properties in Scandinavian, while acknowledging that in other languages object shift may manifest other properties. See also proposals in Molnárfi (2002), where an EPP feature is taken to be responsible for triggering object raising in Afrikaans and West Germanic. Note that other Romance languages also seem to allow for vP-adjoined object raising with specific semantico-pragmatic and syntactic properties: for Catalan, see discussion in Vallduví (1995), for Portuguese, see Costa (1999), and for Spanish, see discussion in Ordóñez (1998). It seems then that the v-related OCC feature is available more consistently within Romance. Recall that for v-related OCC ‘other formal feature’ valuation refers to phi-features of v and Case of direct object, while for T-related OCC connected to contrastive focus, it refers to phi-features of T, Case of subject and [Foc] as a formal feature.

References Alboiu, Gabriela 1999 (De)-Focusing and Object Raising in Romanian. Canadian Journal of Linguistics 44 (1): 1–22. Toronto: University of Toronto Press. 2002 The Features of Movement in Romanian. Bucharest: Editura Universitatii Bucuresti.

Optionality at the Interface: Triggering Focus in Romanian 2003

73

Operator Asymmetries in Romanian: Syntax and/or Phonology? In Romance Linguistics: Theory and Acquisition, ed. by Ana Teresa Pérez-Leroux and Yves Roberge. 3–18. Amsterdam and Philadelphia: John Benjamins. Bobaljik, Jonathan David 2002 A-Chains at the PF-Interface: Copies and ‘Covert’ Movement, in Natural Language and Linguistic Theory 20 (2): 197–267. Bo‰koviç, Îeljko 2002 On Multiple Wh-Fronting. Linguistic Inquiry 33(3): 351–185. Brody, M. 1995 Lexico-logical form: a radically minimalist theory. Cambridge: MIT Press. Chomsky, Noam 1971 Studies on Semantics in Generative Grammar. The Hague: Mouton. 1986 Barriers. Cambridge: MIT Press. 1995 The Minimalist Program. Cambridge, Mass: MIT Press. 2000 Minimalist Inquiries: The Framework. Step by Step: Essays on Minimalist Syntax in Honor of Howard Lasnik, ed. By Roger Martin et al, 89–157. Cambridge, Mass: MIT Press. 2001a Derivation by Phase. Ken Hale: A Life in Language, ed. by Michael Kenstowicz, 1–53. Cambridge, Mass: MIT Press. 2001b Beyond Explanatory Adequacy. MIT Occasional Papers in Linguistics 20. MITWPL. Cornilescu, Alexandra 2000a The double subject construction in Romanian. Comparative Studies in Romanian Syntax, ed. by Virginia Motapanyane, 83–134. Dordrecht: Elsevier. 2000b Notes on the Interpretation of the Prepositional Accusative in Romanian. Bucharest Working Papers in Linguistics II(1): 91–107. 2002 Direct Objects at the Left Periphery. Paper presented at the Comparative Romance Linguistics in Antwerp. Costa, Jõao 1999 Word Order Variation. A constraint based approach. Holland Academic Graphics. Diesing, Molly 1992 Indefinites. Cambridge: MIT Press. Dobrovie-Sorin, Carmen 1987 Syntaxe du roumain. Thèse de doctorat d’Etat, Université Paris 7. 1994 The Syntax of Romanian. Berlin: Mouton de Gruyter. Engdahl, Elisabet 1983 Parasitic Gaps. Linguistics and Philosophy 6: 5–34. Epstein, David Samuel and T. Daniel Seely 2002 Rule Applications as Cycles in a Level-Free Syntax. Derivation and Explanation in the Minimalist Program, ed. by Samuel David Epstein and T. Daniel Seely, 65–90. Blackwell Publishing.

74

Gabriela Alboiu

Gierling, Diana 1997 Clitic doubling, Specificity and Focus in Romanian. Clitics, Pronouns, and Movement, ed. by John Black and Virginia Motapanyane, 63–85. Amsterdam and Philadelphia: John Benjamins. Herburger, Elena 2000 What Counts: Focus and Quantification. Cambridge: MIT Press. Hill, Virginia 2002 Adhering Focus. Linguistic Inquiry 33 (1): 164–172. Holmberg, Anders 1999 Remarks on Holmberg’s Generalization. Studia Linguistica 53 (1): 1–39. Hornstein, Norbert 1995 Logical Form. Oxford: Blackwell. Kayne, Richard 1994 The Antisymmetry of Syntax. Cambridge: MIT Press. Kidwai, Ayesha 1999 Word order and Focus Positions in Universal Grammar. The Grammar of Focus, ed. by G. Rebuschi and L. Tuller, 213–245. Amsterdam and Philadelphia: John Benjamins. Kiss, Katalin E. 1995 Introduction. Discourse Configurational Languages, ed. by Katalin E. Kiss, 3–27. Oxford: Oxford University Press. 1998 Identificational Focus Versus Information Focus. Language 74 (2): 245–273. Lasnik, Howard and Tim Stowell 1991 Weakest Crossover. Linguistic Inquiry 22: 687–720. Molnárfi, László 2002 Focus and antifocus in Afrikaans and West Germanic. Linguistics 40/60: 1107–1160. Motapanyane, Virginia 2000 Parameters for focus in English and Romanian. Comparative Studies in Romanian Syntax, ed. by Virginia Motapanyane, 267–296. Dordrecht: Elsevier. Nissenbaum, Jonathan W. 2000 Investigations of Covert Phrase Movement. PhD dissertation. MIT. Ordóñez, F. 1998 Post Verbal Asymmetries in Spanish. Natural Language and Linguistic Theory 16: 313–346. Richards, Norvin 1998 What Moves Where in Which Language. Doctoral dissertation, MIT.

Optionality at the Interface: Triggering Focus in Romanian

75

Rizzi, Luigi 1997 The Fine Structure of the Left Periphery. Elements of Grammar: Handbook of Generative Syntax, ed. by Liliane Haegeman, 281–339. Dordrecht: Kluwer Academic Publishers. 2002 Cartography, Locality, and Asymmetries. Paper presented at the 32nd Linguistic Symposium on Romance Languages, University of Toronto. Rochemont, Michael 1986 Focus in Generative Grammar. Amsterdam: John Benjamins. Tsimpli, I. M. 1995 Focusing in Modern Greek. Discourse Configurational Languages, ed. by K. E. Kiss, 176–207. Oxford: Oxford University Press. Uriagereka, Juan 1999 Multiple Spell-Out. Working Minimalism, ed. by Samuel David Epstein and Norbert Hornstein, 251–283. Cambridge, Mass: MIT Press. Vallduví, E. 1995 Structural Properties of Information Packaging in Catalan. Discourse Configurational Languages, ed. by Katalin E. Kiss, 122–152. Oxford: Oxford University Press. Zubizarreta, Maria 1998 Prosody, Focus and Word Order. Cambridge, Mass: MIT Press.

How N-words Move: Bipartite Negation and ‘Split-NegP’ Arthur Bell

Introduction In this paper I present a syntactic analysis of negation in Afrikaans within the Minimalist framework (Chomsky 1995, 2000, 2001). Afrikaans employs two separate morphemes in most negative sentences. The higher morpheme is either a negative word (N-word, such as nooit ‘never’) or nie. This position is traditionally analyzed as the locus of negative force (Den Besten 1977, 1985; Robbers 1992; Donaldson 1993; Ponelis 1993; Oosthuizen 1998; Bell 2001; Molnárfi 2002). The lower negative morpheme, also spelled-out as nie, cannot be the sole marker of sentential negation. Yet its presence is clearly triggered by the presence of negation, and, in certain cases, its syntactic position appears to perform the important semantic function of disambiguating because-clauses. I argue that displacement of N-words from a vP-internal positions to SPEC-NegP are cases of internal Merge, triggered by the need to eliminate uninterpretable features on a negative head. Furthermore, I propose that the lower negative morpheme in Afrikaans is also a functional head, and triggers vP-movement to its SPEC, pied-piping the contents of vP and accounting for the phrase-final position of the lower negative. Cases of because-clause post-posing are accounted for by positing a left-of-vP functional projection to which a because-clause moves prior to vP remnant movement. Finally, postposed prepositional phrases and complement clauses are argued to be cases of scrambling, analogous to scrambling in non-negative environments. The paper is organized as follows. In §1, I examine the data on Afrikaans negation in detail, and consider the categorial status of the lower negative morpheme nie2. In §2, I briefly present and critique two previous syntactic analyses of Afrikaans negation. Then, in §3 I present my analysis. §4 is an examination the potential applicability of the present analysis to other languages. I offer a summary of the arguments presented in the paper, and some concluding remarks, in §5. To begin, consider the data on Afrikaans negation in §1.

78

Arthur Bell

1.

Negation in Afrikaans: The Data

The data in this paper come from five main sources: Oosthuizen (1998) Donaldson (1993, 2000), and Robbers (1992, 1997). Each of these three authors employs a slightly different scheme in glossing their examples. For the sake of clarity and consistency, I modify their glosses when necessary, in particular for the final negative morpheme nie, which I gloss as nie2. Donaldson (1993) gives only translations, and thus all glosses of his data are my own, and follow the same pattern as my glosses of the other data. 1.1. Negation in Matrix Clauses In this section I discuss negation in matrix clauses, with particular attention paid to the position of nie2. Sentential negation in Afrikaans is normally realized with at least two morphemes: a phrase-internal negator nie and a phrase-final or post-verbal invariant particle nie. Consider the data in (1). (1)

a. Hulle was nie betrokke nie. they were nie1 involved nie2 ‘They were not involved.

AFRIKAANS (Oosthuizen 1998)

b. Sy sluit nooit die deur nie. she locks never the door nie2 ‘She never locks the door.’

(ibid.)

c. Hulle is g’n so arm nie. they are not so poor nie2 ‘They aren’t all that poor.’

(ibid.)

The data in (1) illustrate the general distribution of the two negative morphemes in declarative sentences in Afrikaans. The negator directly follows the tensed verb in matrix clauses, owing to verb movement to second position. The lower negative nie2 follows main verbs (1a), direct objects (1b), and adjectival complements (1c). Next, consider the data in (2). (2)

a. Jy praat nie duidelik nie. you speak nie1 clearly nie2 ‘You’re not speaking clearly.’

(Donaldson 1993)

b. Hy het nie gedink aan die ernstige gevolge nie. he has nie1 think of the serious consequences nie2 ‘He didn’t think of the serious consequences.’

(ibid.)

How N-words Move: Bipartite Negation and ‘Split-NegP’

c. Wie het nie opgedaag nie? Who has nie1 arrived nie2 ‘Who hasn’t arrived?’

79

(Oosthuizen 1998)

In (2) we observe the negator in its normal position immediately following the tensed verb in matrix clauses. As we can see above, nie2 can follow adverbs (2a) and prepositional phrases (2b). The bipartite negation strategy in Afrikaans also occurs in other clause types, such as WH-questions (2c). According to Donaldson (1993), nie2 must always follow direct and indirect nominal object complements. However, he points out that the position of prepositional phrases vis-à-vis nie2 is variable in Afrikaans (Cf. also Robbers 1997: 81–82). Indeed, prepositional phrase complements can occur before or after nie2. While it is more natural for a heavy prepositional phrase to follow nie2, there are no absolute restrictions related to heaviness, and light prepositional phrases can follow nie2 as easily as heavy ones can precede it (Donaldson 1993: 406). Placing the prepositional phrase after nie2 in (3b) and (3d) does not trigger a change in meaning. Consider the data in (3). (3)

a. Sy het niks gesê op die vergadering nie. she has nothing said at the meeing nie2 ‘She said nothing at the meeting.’

(Oosthuizen 1998)

b. Sy het niks gesê nie op die vergadering. she has nothing said nie2 at the meeing ‘She said nothing at the meeting.’

(ibid.)

c. Hoekom word daar nie meer ingegaan op die direkte invloed van how come is there no more delve into the direct influence of die bruin Afrikaanssprekendes nie? the colored Afrikaans-speakers nie2 ‘How come one isn’t delving any more into the influence of coloured speakers of Afrikaans?’ (Donaldson 1993) d. Hoekom word daar nie meer ingegaan nie op die direkte invloed how come is there no more delve nie2 into the direct influence van die bruin Afrikaanssprekendes? of the colored Afrikaans-speakers ‘How come one isn’t delving any more into the influence of coloured speakers of Afrikaans?’ (ibid.)

80

Arthur Bell

Next consider the data in (4), which show that nie2 cannot appear alone as the sole marker of sentential negation. (4)

Sy sluit *(nooit) die deur nie. she locks never the door nie2

(Oosthuizen 1998)

An N-word or the nie1 must be present to give negative meaning to a sentence. There are certain environments where nie2 occurs without a negator. Consider the data in (5). (5)

a. Jy joef my mouliks daarvan te oortuig nie. you have-to me hardly it-of convince nie2 ‘You hardly have to convince me of that.’ (Oosthuizen 1998) b. ?Ek sou dit kwalik sonder sy julp kon regkry nie. I would it hardly without his help could right-get nie2 ‘I would hardly have been able to manage it without his assistance.’

(ibid.)

Observe that in (5), the final particle nie2 occurs without a higher N-word or nie1. Crucially, these sentences are not interpreted as negative. If nie2 carried negative force (had an interpretable negative feature, in the terms I shall adopt below), we would expect the sentences in (5) to be interpreted as negative. They are not, so we must conclude that nie2 has no negative force.1 In certain restricted environments, nie1 or an N-word can stand alone as the sole marker of sentential negation, without nie2. Interestingly, placing nie2 directly following nie1 results in ungrammaticality. Two homophonous morphemes nie cannot appear contiguously, a phenomenon referred to as the “double-nie filter” by Robbers (1997), most likely due to morphological haplology (Cf. Newman 2000: 357 for similar examples of haplology from Hausa). Consider the data in (6). (6)

a. Ek weet nie (*nie). I know nie1 nie2 ‘I don’t know.’

(Donaldson 2000)

b. Ek weet niks (nie). I know nothing nie1 ‘I don’t know anything.’

(Donaldson 1993)

How N-words Move: Bipartite Negation and ‘Split-NegP’

81

The presence of nie2 immediately following the negator nie in (6a) renders the sentences ungrammatical. In (6b), I argue, contra Oosthuizen (1998) and Donaldson (1993), that the optional nie morpheme is nie1, not nie2. This is based on two observations. First, nie1 is optional (for some speakers) in exactly the same position following N-words in transitive clauses, as I show in (7) below.2 Second, it appears to be a general rule that haplology deletes nie2, that is, the morpheme without any interpretable features.3 If nie2 is deleted across the board by haplology, it would be surprising to find that it is optional in this one context. Consider the data in (7). (7)

Ek sal jou nooit (nie) vergeet nie. I will you never nie1 forget nie2 ‘I will never forget you.’

(Oosthuizen 1998)

In the next section, I present data on negation in the context of embedded and relative clauses.

1.2. Negation in the Context of Embedded and Relative Clauses Negation of an embedded clause is normally accomplished with two particles, a negator and nie2. The negator in embedded clauses appears to the left of the verb, as there are no verb-second effects in embedded clauses in Afrikaans (De Kleine 1997; Robbers 1997). Consider the data in (8). (8)

Ek het geweet dat hy nie sou kom nie. I AUX know that he nie1 would come nie2 ‘I knew that he wouldn’t come.’

(Donaldson 1993)

A negative relative inside an affirmative clause is negated with nie2 immediately following the relative, not at the end of the sentence. Consider (9). (9)

Mense wat nie rook nie lewe langer. people who nie1 smoke nie2 live longer ‘People who don’t smoke live longer.

(ibid.)

In cases where the matrix clause is negated, but the embedded or relative clause is not, the position of nie2 in Afrikaans is variable. Consider the data in (10).

82

Arthur Bell

(10) a. Ek het nie geweet dat hy sou kom nie. I AUX nie1 know that he would come nie2 ‘I didn’t know that he would be coming.’ (Donaldson 1993) b. Ek het nie geweet nie dat hy sou kom. I AUX nie1 know nie2 that he would come ‘I didn’t know that he would be coming.’

(ibid.)

c. Ons het nooit die mense geken wat in hierdie huis gebly het nie. we AUX never the people know who in this house live AUX nie2 ‘We never knew the people who lived in this house.’ (ibid.) d. Ons het nooit die mense geken nie wat in hierdie huis gebly het. we AUX never the people know nie2 who in this house live AUX ‘ We never knew the people who lived in this house.’ (ibid.) As we see in (10), the nie2 particle associated with a negative matrix clause can appear either at the end of the sentence, following a non-negative embedded clause (10a), or immediately following the matrix clause (10b). Similar facts hold for relative clauses, as I show in (10c–d). According to Donaldson (1993), the sentences in (10a) and (10c), where nie2 is sentencefinal, illustrate the unmarked position for nie2, although (10b) and (10d) are perfectly acceptable. Any theory of negation in Afrikaans must account for both of these possible positions for nie2. When both the matrix and the embedded clause are negative, there are of course two negators, one associated with the matrix clause, and one with the embedded clause. The position of nie2 is again variable. Consider the data in (11). (11) a. Ek het nie geweet dat hy nie sou kom nie. I AUX nie1 know that he nie1 would come nie2 ‘I didn’t know that he wouldn’t come.’ (Donaldson 1993) b. Ek het nie geweet nie dat hy nie sou kom nie. I AUX nie1 know nie2 that he nie1 would come nie2 ‘I didn’t know that he wouldn’t come.’ c. *Ek het nie geweet dat hy nie sou kom nie nie. I AUX nie1 know that he nie1 would come nie2 nie2 d. *Ek het nie geweet nie dat hy nie sou kom. I AUX nie1 know nie2 that he nie1 would come

(ibid.)

How N-words Move: Bipartite Negation and ‘Split-NegP’

83

In cases where both the matrix and the embedded clause are negative, we observe either two separate instantiations of nie2 (11b), or a single nie2 at the end of the sentence (11a). Native speaker judgments confirm that both (11c), with a double nie2 in sentence-final position, and (11d), where the embedded clause lacks nie2, are ungrammatical. As with nie2 variation around prepositional phrases, native speakers report no change in meaning when nie2 associated with a negative matrix clause appears either before or after an affirmative embedded clause. However, there is a specific context in which the position of nie2 triggers a change in meaning. I consider these data in subsection 1.3.

1.3. Negation in Because-Clauses The particle nie2 in Afrikaans has a privileged role in the disambiguation of because-clauses in negative environments. Consider the sentence in English shown below in (12). Depending on intonation and stress, (12) can have either the reading shown in (a), in which Peter did not leave, or the reading in (b), in which Peter did leave, but not due to the fact that it was raining. (12) Peter didn’t leave because it was raining. a. ¬ leave(Peter) b. leave(Peter) ¬ because it was raining Now consider the data in (13a-d) below. In Afrikaans, the reading in (12a), in which the because-clause is outside the scope of negation, is available when nie2 appears to the left of the because-clause, as in (13a) and (13c). The reading in (12b), where the because-clause is inside the scope of negation, is available when nie2 appears to the right of the because-clause, as in (13b) and (13d). (13) a. Hy het dit nie gedoen nie omdat hy betaal is. he have it nie1 done nie2 because he paid is ‘He did not do it, because he was paid.’ (Robbers 1997: 40) b. Hy het dit nie gedoen omdat hy betaal is nie. he have it nie1 done because he paid is nie2 ‘He did not do it because he was paid (but for another reason)’ (ibid.)

84

Arthur Bell

c. Hy kom nie vandag nie omdat hy siek is. he come nie1 today nie2 because he sick is ‘He’s not coming today because he’s sick.’

(Donaldson 1993)

d. Hy kom nie vandag omdat hy siek is nie. he come nie1 today because he sick is nie2 ‘He’s not coming today because he’s sick.’

(ibid.)

The position of nie2 is variable in embedded and relative clauses, but affects no change in meaning, as we saw in §1.2 above. However, the position of nie2 in because-clauses the determines the scope of negation over the because-clause. As Donaldson (1993) points out, this property is unique to because-clauses with omdat ‘because.’ Consider the sentence in (14). (14)

Hy kan nie kom nie want hy is siek. he can nie1 come nie2 because he is sick ‘He won’t be coming because he’s sick.’

(Donaldson 1993: 407)

Unlike the sentences in (13), the only available position for nie2 in (14) is to the left of the because-clause. Note that (14) has want ‘since, because’ rather than omdat ‘because.’ This fact is comparable to the well-known behavior of since in English, where a similar restriction on holds. Consider for example “Peter is not going since he is sick,” where the only possible interpretation is that Peter is not going. Next, I will consider in greater detail the data on negative words in Afrikaans.

1.4. Negative Words and the Negator Nie1 As I illustrated in §1.2 above, the locus of sentential negation in Afrikaans is the higher negative, which is normally realized either as an N-word, or as nie1. Donaldson (1993) calls nie1 the negator, a term that I have adopted here. N-words in Afrikaans include niks ‘nothing,’ nooit ‘never,’ nêrens ‘nowhere,’ and niemand ‘nobody,’ and are adverbials or nominals that have their own inherent negative force. They can, for example, be used as single-word negative replies to questions. We have seen above that the canonical position of an N-word or nie1 is immediately following the tensed verb in matrix clauses, and preceding the tensed verb in embedded clauses. However, N-words, as lexical categories, can also occur as subjects, and undergo topicalization and leftward displacement (focus movement). Consider the data in (15).

How N-words Move: Bipartite Negation and ‘Split-NegP’

(15) a. Geen mens weet nie. no person knows nie2 ‘No-one knows.’ b. Niemand kom nie. nobody come nie2 ‘Nobody is coming.’

85

(Donaldson 1993)

(ibid.)

As we would expect, negative quantifiers in Afrikaans can appear in subject position. This confirms the general intuition that they are lexical items and can undergo movement to A-positions. Note that in all cases the presence of negation in subject position licenses the presence of nie2. Now consider the data in (16). (16) a. Ons voel nêrens veilig nie. we feel nowhere safe nie2 ‘We don’t feel safe anywhere.’

(Oosthuizen 1998)

b. Nêrens voel sy veilig nie. nowhere feels she safe nie2 ‘Nowhere does she feel safe.’

(ibid.)

We observed in (15) that N-words can appear in subject position. However, the sentences in (16a) and (16b) already contain a subject, namely sy ‘she’. In (16b), nêrens ‘nowhere’ undergoes displacement, most probably to a focus or topic position in the left periphery. Note that nêrens in this position triggers verb-second. As we have seen above, in simple sentential negation Afrikaans employs nie in the position of the negator. One important issue in an analysis of Afrikaans negation, to which I shall return presently in §2, is the categorial status of the negator nie. In some ways the negator nie patterns with other N-words. Consider the data in (17). (17) a. Hy het nie gekom nie. he has nie1 come nie2 ‘He didn’t come/hasn’t come.’

(Donaldson 1993)

b. Nie hy het gekom nie. nie1 he has come nie2 ‘He is not coming.’

(Molnárfi 2002, my gloss)

c. *Nie het hy gekom nie. nie1 has he come nie2

(ibid.)

86

Arthur Bell

The example in (17a) illustrates simple sentential negation in Afrikaans. As in (16b) above, the negator nie can be fronted, as in (17b). The result, according to Donaldson (1993), is a contrastive focus interpretation on hy ‘he,’ as suggested by the boldface type in the translation. First, note that, while fronting nêrens focuses nêrens, fronting nie1 focuses the subject. Second, note that fronting nie in (17b) does not trigger verb-second, unlike fronting nêrens above. Indeed, employing verb-second with fronted nie1 results in ungrammaticality (17c), as pointed out by Molnárfi (2002). This suggests that (17b) and (16c) have rather different syntactic structures, as I discuss below. Finally, I consider the categorial status of nie2 in §1.5.

1.5. Categorial Status of Nie2 Unlike N-words, nie2 is an invariant particle. Its phonological shape never changes, and it can appear in a very limited number of positions. It cannot be fronted, nor can it appear as a topic or in a focus position. Also, as Molnárfi (2002) points out, it is phonologically reduced. Based on these facts, and following I will argue that nie2 is a functional head, following Oosthuizen (1998). One clear piece of evidence that nie2 is of a different category that the negator comes from adverbial modification facts. Consider the data in (18). (18) a. Dit blyk dat sy absoluut niks (nie) kan onthou nie. it seems that she absolutely nothing nie1 can remember nie2 ‘It seems that she can remember absolutely nothing.’ (Oosthuizen 1998) b. *Dit blyk dat sy niks (nie) kan onthou absoluut nie.

(ibid.)

c. *Dit blyk dat sy niks absoluut (nie) kan onthou nie.

(ibid.)

The adverb absoluut ‘absolutely’ can modify the N-word niks ‘nothing’ in (18a). However, absoluut cannot modify nie2 in (18b). The resulting sentence is ungrammatical. Were nie2 a lexical item along the lines of niks, we might expect it to be modifiable by adverbs. This is contrary to fact. Note in addition that the higher, optional nie1 in (18c) cannot be modified in this context. I argue below that this is due to the absence of a structural position between the N-word and nie1. Note that nie1 can be modified by adverbs when it is the sole negator. Consider the sentences in (19).

How N-words Move: Bipartite Negation and ‘Split-NegP’

(19) a. Hulle was nie betrokke nie. they were nie1 involved nie2 ‘They weren’t involved.’ b. Hulle was glad nie betrokke nie. they were entirely nie1 involved nie2 ‘They weren’t involved at all.’

87

(Oosthuizen 1998)

(ibid.)

In the remainder of the paper, I will provide a comprehensive analysis of the syntax of negation in Afrikaans. I will focus on four main issues: i) the categorial status and syntactic function of the negative morpheme nie2; ii) the syntactic structure, features and operations that capture the position and behavior of nie2; iii) the specific mechanisms by which N-words undergo displacement in Afrikaans; iv) the categorial status and syntactic function of the negator nie1. First, I briefly consider two previous analysis of negation in Afrikaans, Oosthuizen (1998) and Molnárfi (2002).

2. Previous Analyses of Nie2 In this section I critique two recent analyses of negation in Afrikaans, both of which make specific proposals with regards to the syntax of the final particle nie2. I look first at the analysis of Oosthuizen (1998), in §2.1, and then examine the analysis of Molnárfi (2002) in §2.2. First, consider the analysis of Oosthuizen (1998).

2.1. Oosthuizen (1998) Oosthuizen (1998) proposes an analysis of nie2 in the Minimalist framework of Chomsky (1995). He argues that nie2 heads a Polarity Phrase, PolP, projected in the left periphery, above CP. In his account, feature-checking drives movement of the entire CP to [SPEC,PolP], pied-piping the whole clause and accounting for the phrase-final position of nie2. This account avoids the problems inherent in a NEG Criterion analysis, which would be forced to rely on a right-branching phrase to house nie2. Consider the derivation in (20), abstracted from Oosthuizen’s (1998) analysis.

88

Arthur Bell

(20)

PolP 3 : Pol’ # 3 z------- CP nie2 6 Sy het niks gesê op die vergadering ‘She said nothing at the meeting.’

Oosthuizen’s (1998) analysis makes the strong prediction that nie2 is a radically final particle in Afrikaans. Because PolP dominates CP, and the entire CP moves to [SPEC,PolP], it follows necessarily that nie2 must occur sentence-finally. As we have seen above, this is contradicted by the data. While nie2 can appear in sentence-final position, it very often occurs to the left of prepositional phrases and embedded clauses. Oosthuizen (1998: 88) notes that post-posed prepositional phrases are “clearly problematical” for his analysis of nie2, but proposes no solution. In addition, we have seen that nie2 can appear to the left of because-clauses, and that this positional difference triggers a change in meaning. Again, the analysis proposed by Oosthuizen (1998) cannot account for these data. We could propose a modified version of Oosthuizen’s (1998) analysis in an attempt to overcome some of these problems. Under this account, which I will call “modified-Oosthuizen,” nie2 is the head of PolP, as above. To accommodate post-posed PPs, assume a position intermediate between PolP and CP. PolP dominates this position XP, which in turn dominates CP. This account can easily explain the variable position of nie2 vis-à-vis PPs. Unfortunately, Oosthuizen’s (1998) analysis faces further difficulties. First, CP is generally assumed to be a barrier to movement. As pointed out by Chris Collins (personal communication), Oosthuizen’s (1998) account would require prepositional phrases to cross a CP-boundary to land in [SPEC,XP]. This predicts, assuming WH-island effects in Afrikaans (De Kleine 1997; Robbers 1997), that questions with post-posed prepositional phrases should be ungrammatical in Afrikaans. As we have already seen above in (3d), this is contrary to fact. Post-posed PPs in negative questions are grammatical. This problem could be overcome by assuming that IP, not CP, moves to [SPEC,PolP], in a further modification of “modified-Oosthuizen.” Second, there are problems with Oosthuizen’s (1998) analysis of a more theoretical nature. In two recent papers, Déchaine and Wiltschko (2001, 2002) argue that sentential negation can be broadly categorized as being located either in the domain of CP, or the domain of IP. CP-domain negation,

How N-words Move: Bipartite Negation and ‘Split-NegP’

89

found in Salish and Algonquian languages, correlates with certain morphosyntactic and semantic properties, while IP-domain negation, found in Romance and Germanic languages for example, does not correlate with these same properties. Déchaine and Wiltschko (2002) illustrate that in languages CP-domain negation, the simplex negative can be used as a reply to a yes/ no question, and can be used to form negative predicates. Oosthuizen (1998) places his Polarity Phrase in the CP-domain, and argues that nie2 is associated with negative features that attract and check the (uninterpretable) negative feature associated with the verb. This would seem to suggest that Afrikaans negation resides in the CP-domain. Yet Afrikaans patterns not with CPdomain negation languages, but with IP-domain negation languages. Therefore, the projection housing nie2 is most likely not part of the left-periphery structure, but is rather projected somewhere in the domain of IP. I will pursue an analysis in which Afrikaans negation is in the domain of IP in §3 below. 2.2. Molnárfi (2002) Molnárfi (2002) argues that the lower negative morpheme nie2 is a phonologically reduced copy of the negator nie1, and an overt manifestation of the end of negative scope. The notion of copy employed by Molnárfi (2002) should be distinguished from the minimalist “copy theory of movement” (Chomsky 1995; Fox 2000). A deleted copy in the minimalist sense corresponds roughly to a trace in older principles-and-parameters work. A copy in Molnárfi’s (2002) sense is a phonologically-reduced homophonous particle, but does not relate to displacement. Rather, the copy of nie is due to a process of “percolation” of NEG throughout the VP, and “the post-verbal negator is in this respect only a phonetically reduced copy of nie1,” and not the head of its own projection (Molnárfi 2002: 224).4 The result of negative “percolation” is a negative doubling (Negationsklammer) effect, i.e. a copy of nie which surfaces at the end of the domain of negative scope. I will argue against Molnárfi’s account on both empirical and theoretic grounds. Molnárfi’s (2002) theory predicts that the copy of the negator nie, nie2, should be phonologically reduced. This prediction is borne out in the data, since nie2 cannot receive contrastive stress. Following this chain of reasoning, I argue that Molnárfi’s (2002) theory makes a further prediction, namely that nie2 should not have a separate syntactic function as a negative head, since it is merely a copy of the negator nie. This prediction is straightforwardly disconfirmed by the data on because-clauses, which show that nie2, as a separate syntactic head, plays a very important role in displacement of because-clauses.

90

Arthur Bell

In general, it is difficult to reconcile Molnárfi’s (2002) account with the theoretical assumptions adopted in this paper. For example, Molnárfi posits the availability of a mechanism allowing a copy of a syntactic head to be spelled out to the right of its base position. Translating this into principles-andparameters or minimalist terms, we would be forced to appeal to rightward head movement, or a right-branching structure, allowing nie1 to move to some lower position. Yet such movement operations and such structures are universally ruled-out since Kayne (1994); thus, adopting a minimalist version of Molnárfi’s (2002) theory would require some radical shifts in assumptions about the nature of displacement and syntactic structure. Even assuming we could adapt Molnárfi’s account to minimalist syntax, there are further empirical arguments against his theory of a cross-linguistic nature. We see in Afrikaans that the higher morpheme nie1 and the lower morpheme nie2 are homophonous. This is a key component of Molnárfi’s theory, since we would not expect a phonologically-reduced, syntactically inactive copy of a syntactic head to differ in phonological shape. However, as I show in §4 three below, there are several languages whose negation strategies parallel the negation strategy of Afrikaans in which the higher and lower negative morphemes are not homophonous. Furthermore, it appears to be a general fact about these languages that the lower negative morpheme is phonologically reduced. Rather than isolate Afrikaans as a separate case governed by a unique process of copying, I argue that a unified syntactic account can explain how negation functions in all of these languages. Finally, as I argue in Bell (2003), the fact that the nie morphemes are homophonous (as are the não morphemes in Vernacular Brazilian Portuguese, the no morphemes in contact varieties of Spanish, etc.) has much more to do with the particular historical development of Afrikaans than with its syntax.

3. Analysis I now propose an alternative to the analyses given by Oosthuizen (1998) and Molnárfi (2002), both of which, I have argued, have important shortcomings. In my analysis, I take into account not only the general position of nie2, but also certain syntactic properties of negation in Afrikaans, such as the behavior of because-clauses. Furthermore, I propose an account of the triggers for displacement of N-words and of the phrase-level categories that move to the left of nie2. I begin with an analysis of nie2 in § 3.1. I discuss prepositional phrases and embedded clauses in § 3.2, and because-clauses in §3.3. Finally, I develop an analysis of N-word displacement in § 3.4.

How N-words Move: Bipartite Negation and ‘Split-NegP’

91

3.1. The Position of Nie2 In this section I offer a syntactic analysis of the position of nie2. I propose that nie2 heads a functional projection NegP2 that is projected in the domain of IP, left of vP. I assume first that nie2 has an uninterpretable feature [uNeg] (Oosthuizen 1998) and an OCC feature (Chomsky 2001). Furthermore, following Oosthuizen (1998), I stipulate that v in sentential negation environments has an interpretable [+neg] feature. There are many languages which are argued to have so-called negative verbs. These include Samoyedic, Permic and Mordvin languages (Hajdu 1963), as well as Turkish, Finnish, Hungarian and Korean (Dahl 1979; Payne 1985). The existence of such languages adds support to the proposal that v can possess a feature [+neg] in sentential negation environments. The [+neg] feature on v would be overtly realized in such languages as either a negative verb or a negative auxiliary.5 Further evidence for a [+neg] feature on v may come from verb-movement to Negº in Nweh (see §4 below and Bell, 2004). For the remainder of the paper, I assume an interpretable feature [+neg] associated with v.6 This feature values the [uNeg] on nie2. An Agree relation is established between nie2, acting as a probe, and v. The OCC feature on nie2 requires some phrasal element XP to merge with [SPEC,NegP2] via Merge. This phrasal element is vP. vP merges with [SPEC,NegP2]. In other words, the head of NegP2, nie2, selects for a [+neg] v.7 I provide the steps in the derivation in (21) and a partial derivation in (22). (21) a. Agree(nie2, v) b. OCC(nie2, vP) c. Merge(vP, NegP2) (22) vP-movement to [SPEC,NegP2]

b. [NegP2

: # z-----------m

a. [NegP2

nie2 [uNeg][OCC]

Agree [vP S v O [+neg][UF]

nie2 [vP S [uNeg][OCC] : Merge # # # z-----------m

O

V

]]]

V

]]]

92

Arthur Bell

I assume that derivations are cyclic and obey the extension condition. The [uNeg] feature on nie2 is eliminated under the Agree relation (Chomsky 2000). An EPP-feature OCC on nie2 requires some XP-level category to move to [SPEC,NegP2], licensing the operation Merge (21b). vP merges with [SPEC,NegP2] (21c), The derivation in (22) is represented in the following partial structure in (23b). (23) a. Sy het niks gesê op die vergadering nie. she has nothing said at the meeing nie2 ‘She said nothing at the meeting.’

(Oosthuizen 1998)

b.

NegP2 qp a--> vP Neg’ ! 6 2 ! sy het niks gesê op nie2 ! die vergadering # z------------------m In the next two subsections, I examine the data on post-posed prepositional phrases, complement clauses, and because-clauses, in light of this analysis.

3.1.1. Prepositional Phrases and Nie2 As discussed above, one serious shortcoming of the account offered by Oosthuizen (1998) is its inability to accommodate post-posed PPs and CPs which follow nie2. I have argued above that vP moves to [SPEC,NegP2], pied-piping everything contained in it to this position as well. Yet certain constituents and clauses can be left behind, stranded below nie2. To account for this, I propose a further functional projection to the left of vP. Scrambling of post-verbal constituents is a well-known phenomenon in both Dutch (Zwart 1997) and Afrikaans (Robbers 1997). Therefore, it is highly probably, assuming that there is no identifiable particular semantic difference between piedpiped PPs and CPs on the one hand, and post-posed PPs and CPs on other, that the position to which scrambled elements move in negation is analogous to the position they move to in affirmative clauses. Assume this position is an XP, and that XP selects vP. The projection XP intervenes between NegP2 and vP. This requires a modification to our analysis above, as it must necessarily

How N-words Move: Bipartite Negation and ‘Split-NegP’

93

be the case that NegP2 selects XP. The expanded left-of-vP structure now appears as in (24). (24)

NegP2 3 Neg’ 3 XP nie2 3 X’ 3 Xº vP

Given this structure, we can now develop an analysis to account for postposing of prepositional phrases and CP-complements after nie2, in a framework where rightward movement is disallowed. To begin, consider the example of a post-posed PP in (25). (25)

Sy het niks gesê nie op die vergadering. she has nothing said at the meeing nie2 ‘She said nothing at the meeting.’ (Oosthuizen 1998)

The left-of-vP structure proposed in (24) accommodates the data in (25) as follows. First, the prepositional phrase merges with [SPEC,XP], as in (26a) below. Second, the vP-remnant merges with [SPEC,NegP2] in (26b), via mechanisms described in §3.1.1 above. nie2 [XP [vP S [VP O V [PP op die vergadering]]]]] : # z-----------m b. ... [NegP2 [vP [ nie2 [XP [PP op die vergadering []]]]]] : # z-------------------m

(26) a. ... [NegP2

Neg

Here, the PP has moved out of vP to [SPEC,XP] prior to vP-remnant movement to [SPEC,NegP2]. I assume that a similar operation can account for post-posed complement clauses, including the data on embedded and relative clauses given above. In addition, clauses that must follow nie2, such as adverbial conjuncts headed by tog and al, now have a position to which they can move. A remaining question concerns the nature of the category XP suggested above. I will explore this question presently. First, let us look once more at the because-clause data.

94

Arthur Bell

3.1.2. Because-Clauses and Nie2 Consider again the very interesting case of because-clauses in Afrikaans. As we saw in §2.1 above, because-clauses can either precede or follow nie2. Unlike the position of prepositional phrases and CP-complements, however, the position of the because-clause relative to nie2 results in a change in meaning. If the because-clause occurs to the right of nie2, negation takes scope only over the matrix clause. If, on the other hand, the because-clause occurs to the left of nie2, negation appears to take wide scope over the whole clause. I repeat the relevant data in (27). (27) a. Hy het dit nie gedoen nie omdat hy betaal is. he have it nie1 done nie2 because he paid is ‘He did not do it, because he was paid.’ (Robbers 1997: 40) b. Hy het dit nie gedoen omdat hy betaal is nie. he have it nie1 done because he paid is nie2 ‘He did not do it because he was paid (but for another reason)’ (ibid.) Note that the sentence in (27a) can also be rendered in English by fronting the entire because-clause, yielding, “because he was paid, he didn’t do it.” This is the correct interpretation of the Afrikaans sentence in (27a). In Afrikaans, a potential ambiguity (either, “he didn’t do it, the reason being because he was paid” (27a), or, “he did do it, but not because he was paid” (27b) is resolved by the position of nie2 relative to the because-clause. Essentially, what is achieved in English by fronting of the because-clause, or intonation (a pause before the because-clause), is achieved via syntactic means in Afrikaans.8 In what follows, I offer an analysis of because-clauses in negative environments in Afrikaans which is consistent with my analysis in sections 3.1.1 and 3.1.2 above, and which can clearly explain the different interpretations of (27). The core of my proposal involves the notion that nie2 is a scope-delimiting particle. In (27a), the because-clause has moved outside the scope of negation. I argue that this is overt syntactic movement to a position in the left-of-vP periphery. I propose that this movement is to a specifier position of a second XP which, given semantic concerns, is clearly distinct from the XP proposed for PP post-posing. Again, this corresponds to the general effort in this paper to define the complex functional structure in the left-ofvP periphery (Cf. Jayaseelan 2001 on a left-of-vP focus phrase).

How N-words Move: Bipartite Negation and ‘Split-NegP’

95

When nie2 occurs to the right of the because-clause, the because-clause is inside the scope of negation, and vP to [SPEC,NegP2] accounts for the position of nie2. In the case of sentences such as (27a), the derivation is slightly more complex. First, I propose that an uninterpretable feature [uF] on the head of the because-clause phrase XP, acts as a probe whose goal is the CP because-clause. This analysis assumes a difference in feature-structure between two homophonous lexical items omdat ‘because.’ One contains an interpretable feature valued by the [uF] on the head of XP, the other does not.9 This proposal immediately explains the difference between omdat ‘because’ and want ‘since.’ Only the former shows the alternation described in this section. If omdat contains the relevant interpretable feature, an Agree relation is established between the XP head and omdat. Subsequently an OCC feature licenses internal merge of the CP omdat-clause and [SPEC,XP]. I provide the relations in (28), an a partial derivation in (29). (28) a. Agree(Xº,CP) b. OCC(Xº,CP) c. Merge(CP, SPEC-XP) (29) a. ... [NegP2

nie2 [XP [vP S [VP O V [CP omdat hy betaal is]]]]] : # z-----------m nie2 [XP CP [vP [VP O V V’ ]]]]

Neg

b. ... [NegP2 : # z-------------m

The partial derivation in (29) illustrates CP-movement to XP, and subsequent remnant movement of vP to [SPEC,NegP2]. An alternative analysis could claim that the proposed scope phrase is actually some type of focus phrase, along the lines of the left-of-vP focus phrase proposed by Jayaseelan (2001) for English and Malayalam. However, a focus-phrase analysis has an overgeneration problem. If a because-clause can move to [SPEC,FocusP], why don’t other constituents and phrases move to this position? The account offered here is sufficiently restrictive. It applies only to the special case of because-clauses, owing to a feature distinction between two homophonous omdat LIs.

96

Arthur Bell

3.2. Movement of N-words In this section I propose an account of N-word displacement that shows the NEG Criterion can be reduced to internal merge. To begin, consider the sentence in (30). (30) Sy sluit nooit die deur nie. she locks never the door nie2 ‘She never locks the door.’

(Oosthuizen 1998)

First, I propose a negative projection below T, which I will call NegP1 to distinguish it from the left-of-vP NegP2 discussed above. Second, I propose, following standard assumptions and the specific analyses of Pollock (1989), Haegeman (1995), and Zanuttini (2000) that N-words such as nooit ‘never’, nêrens ‘nowhere’ and niemand ‘nobody’ are initially externally-merged into a vP internal or vP adjunct position below [SPEC-NegP1] and subsequently moved to [SPEC-NegP1] before Spell-Out. The head of NegP1 has an uninterpretable feature [uNeg] that must be eliminated before Spell-Out. The N-word has an interpretable [+neg] feature that values the unvalued [uNeg] on nie1, the functional head.10 The elimination of uninterpretable features is achieved via Agree (Chomsky 2000: 122), which is established between a probe (P) and a goal (G); in this case Negº (P) and N-word (G). In addition, I argue that the head of NegP1, Negº, has an OCC feature that must be eliminated. Internal merge is triggered by the need to eliminate the OCC feature on Negº. I give the operations in (31) and the partial derivation in (32). (31) a. Agree(Negº, N-word) b. OCC(Negº, N-word) c. Merge(N-word, NegP1) : # z-----------m

(32) a. [NegP1

Negº … [AdvP nooit]] [+Neg][UF] [uNeg][OCC]

b. [NegP1

Negº … [AdvP nooit]] : [uNeg][OCC] # z---------------m

This analysis results in a SPEC-head relation between the N-word nooit and the negative head Negº – a key component of the standard principles-and-

How N-words Move: Bipartite Negation and ‘Split-NegP’

97

parameters analysis of negation, the NEG Criterion (Haegeman 1995). Here, the SPEC-head relation is achieved by feature-elimination and internal merge. We have in effect derived part of the NEG Criterion within Minimalism. Now consider (33), in which I illustrate the complete derivation of the sentence in (30). The derivation proceeds as follows: 1) vP merges with NegP2; 2) the N-word merges with NegP1; 3) the verb merges with T; and, 4) the subject merges with TP. (33)

TP 3 T T’ ! 3 sy Tº NegP1 g 3 : AP Neg’ # sluit g 3 # : NegP2 # # nooitk Neg g ri # : # -l Negº vP v’ # # # 3 VP z--------m v 3 V DP sema Quote “say”

The Agreement Parameter

121

A plausible alternative to right adjunction in (14) which would also be consistent with the Agreement Parameter would be leftward movement of IP (see Dobashi 2003, Buell 2000). Possibly, Juma would be in Spec TopP, and the IP would move to Spec ForceP. 3 The main evidence that QI in Kiswahili is different structurally from QI in English involves the Transitivity Constraint (see Collins 1997, 2003a), illustrated for English below: (15) a. “Can I be excused?” The children asked the teacher b. … *asked the children the teacher c. … *asked the teacher the children (16) a. “I am here!” b. … c. …

John told the chilren *told John the children *told the children John

The Transitivity Constraint prohibits QI with distransitive verbs, such as ask, tell, warn, advise, etc. Significantly, there is no such constraint in Kiswahili: (17) a. “Tu-ondok-e?” wa-toto 1pl-leave-subj 2-child “Should we leave?” the children

wa-ka-mw-uliza mw-alimu 3pl-subs-3sg-ask 1-teacher asked the teacher

b. …

*wa-ka-mw-uliza 3pl-subs-3sg-ask

wa-toto mw-alimu 2-child 1-teacher

c. …

wa-ka-mw-uliza 3pl-subs-3sg-ask

mw-alimu wa-toto 1-teacher 2-child

(18) a. “Ni-ko!” 1sg-here “I am here!”

Juma a-li-w-ambia Juma 3sg-past-3pl-tell Juma told the children

wa-toto 2-child

b. …

*a-li-w-ambia 3sg-past-3pl-tell

Juma Juma

wa-toto 2-child

c. …

a-li-w-ambia 3sg-past-3pl-tell

wa-toto 2-child

Juma Juma

Sentences (17c) and (18c) show that inversion is possible with ditransitive verbs in Kiswahili. The unacceptability of (18b) (and similarly (17b)) follows from the assumption that Juma is right adjoined to IP (and therefore, must follow the complements of the verb).

122

Chris Collins

In the analysis of QI given above, I suggested that the subject can appear right adjoined to IP (coindexed with pro in Spec IP). The following sentence from Keach (1995: 100) suggests that right adjunction is an option in declarative (non-quotative) clauses as well. (19) a. Wa-tu w-a Kenya wa-na-wa-penda wa-toto 2-people 2Agr-of Kenya 3pl-pres-3pl-like 2-child “the people of Kenya like children” b. Wa-na-wa-penda wa-toto wa-tu w-a Kenya 3pl-pres-3pl-like 2-child 2-people 2Agr-of Kenya “They like children, the people of Kenya”

4. Locative Inversion In both Collins (1997) and Collins (2003a), I speculate that Quotative Inversion and Locative Inversion in English are derived syntactically in essentially the same way in English. Adopting this conclusion, it follows that Locative Inversion in English involves Agree without internal Merge. This is illustrated in the following example: (20) a. Into the house ran John b. Into the house runs John c. Into the house run the children In these examples, the relation Agree(Infl, John) values the phi-features of Infl. Furthermore, the post-verbal subject is in Spec VP (in locative inversion with unaccusative verbs), or Spec vP (in locative inversion with unergative verbs). The PP is in Spec IP, where it satisfies the OCC feature of Infl. Assuming that internal Merge requires Agree (see Chomsky 2001: 13), it follows that movement of the PP into Spec IP requires some form of agreement. I will assume, without argument, that Agree(Infl, PP) is defective in English, involving [uPerson] only (see Hiraiwa (2001) on the general theory of Multiple Agree). The prediction of the Agreement Parameter is that the syntax of Locative Inversion in Kiswahili should be quite different from that found in English. Given this background consider the following sentence in Kiswahili:

The Agreement Parameter

(21)

Chumba-ni a-li-ingia John room-in 3sg-past-enter John “Into the house, he entered, John”

(22)

nyumba-ni wa-li-fika wa-tu house-in 16Agr-past-arrive 2-people “at home, they arrived, the people”

123

The prediction of the Agreement Parameter is that John cannot be in-situ in (21), and watu “people” cannot be in-situ in (22), since Agree would force internal Merge into Spec IP. An analysis of (21) consistent with the Agreement Parameter would be that John is right adjoined to IP, with pro in Spec IP (similar to the analysis of Quotative Inversion in Kiswahili in (14)). Some evidence for this analysis comes from the fact that John cannot be replaced by a universal quantifier, as the following sentences show: (23) a. *Chumba-ni a-li-ingia kila m-ume room-in 3sg-past-enter every 1-man b. kila m-ume a-li-ingia chumba-ni every 1-man 3sg-past-enter room-in (24) a. *Nyumba-ni a-li-fika kila m-geni house-in 3sg-past-arrive every 1-guest b. kila m-geni a-li-fika nyumba-ni every 1-guest 3sg-past-arrive house-in “Every guest arrived at the house” I can explain these facts using Rizzi’s condition on A’-binding of a pronoun: (25) A pronoun cannot be locally [A-bar] bound by a quantifer (Rizzi 1986; Baker 1996) Since the quantifier in (23a) would be locally A’-binding the pro in Spec IP, the structure is ruled out. It is also possible that (23a) should be ruled out as an instance of Weak Crossover. The crucial assumption of my account of (23a,b) is that the QP in (23b) is in Spec IP. If the QP is actually left adjoined to IP, then ruling out (23a) in terms of Weak Crossover would be more accurate empirically. I leave this issue to further research. Crucially, English locative inversion has no such constraint:

124

Chris Collins

(26) Into the house entered every man The reason why (26) does not violate (25) is because there is no pro or overt pronoun in Spec IP. Rather, the uninterpretable phi-features of Infl are valued by every man in the relation Agree (Infl, every man). If the locative agrees with Infl, then a post-verbal universal quantifier is OK, as shown below: (27) a. Chumba-ni m-li-ingia room-in 18Agr-past-enter “Into the house entered John” b. Chumba-ni room-in

m-li-ingia 18Agr-past-enter

John John kila m-ume every 1-man

(28) a. Nyumba-ni pa-li-fika wa-tu house-in 16Agr-past-arrive 2-people “People arrived at the house” b. Nyumba-ni house-in

pa-li-fika kila m-geni 16Agr-past-arrive every 1-guest

In the examples in (27b) and (28b), there is no pro or overt pronoun in Spec IP, so there is no violation of (25). I put aside the issue of how the Case feature of the post-verbal DP is valued/checked in (27, 28). One possibility is that the post-verbal DP in locative inversion does not have a Case feature. Another possibility is that some sort of defective Agree relation is established between Infl and the post-verbal DP allowing Infl to value the DP’s Case feature (e.g., Agree([uPerson], DP)). The later analysis would entail a restatement of the Agreement Parameter if no internal Merge is involved (see the related discussion of na “with” at the end of section 2). It is unclear how to decide between these alternatives for now. For more on the syntax of Locative Inversion in Bantu, see Bresnan and Kanerva (1989).

5. DP Structure The Agreement Parameter makes the following prediction about word order internal to the DP: DP internal agreement between a noun and X should only be possible if X follows NP. The reason for this is that Agree(X, NP) will give rise to internal Merge, thus placing NP in a position preceding X at

The Agreement Parameter

125

Spell-Out. Surprisingly, this prediction is born out (see Carstens 1991 for a systematic discussion of Kiswahili DPs). Consider the following example involving a NP and a possessor. (29) a. ki-tabu ch-a Juma 7-book 7Agr-Assoc Juma “Juma’s book” b. *ch-a Juma 7Agr-Assoc Juma

ki-tabu 7-book

The associative head precedes the possessor and follows the modified NP. Modifying Nkemnji (1995) (see also Carstens 2001), I analyze this in terms of NP movement to Spec AssocP. (29b) is unacceptable since there the relation Agree(Assoc, kitabu) is established, but there is no internal Merge. (30)

AssocP AssocP 3 3 Assoc PossP NP Assoc’ Agr-a ki-tabu 3 3 : Assoc : DP Poss’ PossP # Agr-a 3 # Juma 3 # Poss NP # DP Poss’ # ki-tabu # Juma 3 # z--------------m Poss

# Agree z----------------m internal Merge

This analysis of DP structure supports the Agreement Parameter, in that Agree(Assoc, NP) gives rise to internal Merge, and the NP precedes the Assoc. I am assuming that Assoc must be associated with uninterpretable phi-features as a lexical property (of Assoc). In other words, the order AssocPoss-NP with no agreement on Assoc is not possible. Note the absence of the word order in (29b) entails that there is no right adjunction in the DP (unlike what was found in (14)). A technical question comes up with respect to the structure in (30). Given Relativized Minimality, or the MLC (Minimal Link Condition), why doesn’t Juma in Spec PossP block movement of the NP to Spec AssocP. The easiest solution to this problem is to adopt the stipulation of Chomsky (1995: 356) (see also Collins (1997)), and assume Juma and the NP are equidistant from

126

Chris Collins

Assoc. On this assumption, Juma does not block Agree(Assoc, NP), nor movement of NP to Spec Assoc. An alternative, given in Collins (2003b) is to assume that when Assoc checks the Case feature of Juma 4, the relation Agree(Assoc, NP) is licensed by Multiple Match (see Baker and Collins (1993) for another possible way of handling apparent MLC violations). Agreement in the DP argues against the Spec-Head agreement hypothesis. If the Spec-Head relation were the fundamental structural relation for agreement, it is unclear why Juma in (30) does not agree with Poss. On the other hand, since Agree requires c-command between a probe and a goal, it is impossible for Poss to agree with a DP in its specifier. A similar ordering restriction holds for constructions involving adjectives: (31) a. ki-tabu ki-zuri 7-book 7-good “a good book” b. *ki-zuri ki-tabu 7-good 7-book I give an account of this word order restriction that is similar to the account of the word order restriction in (29). Let us assume that adjectival constructions involve two heads AgrAdj (an agreement projection associated with adjectives) and FAdj (functional projection associated with adjectival modification). The adjective itself appears in the specifier of FAdj, and the NP moves to the specifier of AgrAdj. This structure is illustrated as follows: (32)

AgrAdjP 3 NP Agr’Adj kitabu 3 FAdjP AgrAdj : ki3 # AdjP F’Adj # zuri 3 #

FAdj # z---------------------m

Now consider how the above analysis handles a more complicated paradigm involving adjectival agreement.

The Agreement Parameter

(33)

a. ki-tabu 7-book

127

ki-zuri ch-a Juma 7-good 7-Assoc Juma

b. *ki-zuri ki-tabu ch-a

Juma

c.

ki-tabu

ch-a

Juma

ki-zuri

d. *ki-tabu

ch-a

ki-zuri Juma

I am assuming that the phrase structure is as follows: Assoc Poss AgrAdj FAdj NP (Assoc c-commands Poss, etc.). The word order in (33a) involves Agree (Assoc, NP), with pied-piping of AgrAdjP into Spec Assoc. (33b) is unacceptable for the same reason that (31b) is. The word order in (33c) also involves Agree(Assoc, NP) but with no pied-piping of AgrAdjP. The derivation of (33c) suggests that the notion of equidistance is not relevant for the derivation illustrated in (30), since the NP and possessor are not equidistant. Lastly, (33d) is unacceptable, since there is no position between Assoc and Juma where AgrAdjP could have landed. The above analysis of adjectives requires an Agr head, which is composed purely of features uninterpretable at the LF interface. In this sense the Agr head is similar to the linker head studied in Collins (2003b) and Baker and Collins (2003). There are a number of cases (mainly borrowed adjectives) where a postnominal adjective does not agree with the noun: watu tofauti (people different) “different people”, mtu safi (person clean) “a clean person”, chakula tayari (food ready) “ready food” (see Hinnebusch and Mirza 1998: 109). I assume that these cases also involve agreement (hence internal Merge), but the agreement is not overtly realized for reasons that are as yet unclear. The Agreement Parameter also gives us insight into the syntax of quantifiers such as kila “every” and ote “all” in Kiswahili. Examples involving these elements are given below5: (34) a. kila ki-tabu every 7-book “every book” b. kila m-tu every 1-person “every person”

128

Chris Collins

(35) a. vi-tabu vy-ote 8-book 8-all “all the books” b. wa-tu w-ote 2-person 2-all “all the people” There are two related differences between kila “every” and ote “all” in Kiswahili. First, kila “every” precedes the NP it quantifiers over, whereas ote “all” follows. Second, kila “every” does not agree with the NP it quantifiers over, whereas ote “all” does. These two facts are correlated by the Agreement Parameter, since Agree gives rise to internal Merge. When there is agreement between the NP and the quantifier, the NP must precede the quantifier. The analysis is illustrated below: (36)

Agr-Q NP NP Agr-Q




(internal Merge)

Curiously, ote “all” can appear preceding the NP in some sentences: (37) a. vi-tabu vy-ote vi-me-potea 8-book 8Agr-all 8Agr-perf-lost “all the books are lost” b. vy-ote vi-tabu vi-me-potea 8Agr-all 8-book 8Agr-perf-lost “all the books are lost” The sentence (37b) is used for emphasis, when there is surprise that all the books are lost. This sentence is an apparent counter-example to the Agreement Parameter, since vy-ote “all” agrees with vi-tabu “book”, but vi-tabu “books” follows vy-ote “all”. I will speculate that the interpretation of the sentence is related to a FocP, into which vy-ote “all” raises, after agreeing with the NP. Thus the derivation of this construction is a type of remnant movement. The derivation is sketched below6: (38) a. b. c. d.

[QP Agr-Q Agr-Q [QP NP [DP D [QP NP Agr-Q [DP NP D [QP

NP ] ] ] Agr-Q

]







internal Merge Merge D internal Merge Merge Foc

The Agreement Parameter

e. [FocP Foc [DP NP D f. [FocP [QP Agr-Q

129

[QP Agr-Q ] Foc

]
internal Merge [DP NP D ]

Lastly, consider the question of why the NP instead of the DP possessor raises to Spec AssocP. This is illustrated below: (39) Juma w-a kitabu Juma 1Agr-assoc book a. *“Juma’s book” b. “the book’s Juma”

(Juma is the main character of a book)

The only possible gloss of the example in (39) is that kitabu “book” is the possessor, not Juma. Nothing would seem to block Agree(Assoc, Juma), giving rise to internal Merge. I would like to suggest two alternatives. First, note that there is never person agreement on the Assoc head: (40) a. mimi ni w-a Juma 1sg cop 1Agr-assoc Juma “I am Juma’s” (an object could say this in a fairy tale) b. *mimi ni ni-a 1sg cop 1sg-assoc

Juma Juma

In example (40a), there is no agreement in person between the subject and the Assoc head, rather the Assoc head bears Class 1 agreement (since mimi is singular and animate). Now suppose that in a structure such as [Assoc DP Poss NP], Agree(Assoc, DP) and Agree(Assoc, NP) are both possible in principle. I propose that Agree(Assoc, NP) wins out, because an NP (as opposed to a DP) has only number and gender features, thereby maximizing agreement with the probe (which does not have person features, as (40b) shows). An alternative analysis, suggested to me by Yoshi Dobashi is that the agreement morpheme and associative morpheme are two independent heads. Given this analysis, the structure of (39a), would be as follows: (41) Agr

Assoc

[PossP

Juma Poss NP ]

130

Chris Collins

Now suppose that Assoc values/deletes the Case feature of Juma. Then Juma would be inaccessible to further Agree relations (see Chomsky 2000). Therefore, it would be impossible for Juma to move to Spec Assoc (assuming internal Merge requires Agree, see the introductory remarks of this paper). I will not attempt to establish the correct alternative in this paper.

6. The Agreement Parameter and Case Theory There is another difference between Bantu and Indo-European languages involving the presence of multiple phi-complete agreement in constructions involving auxiliary verbs. The basic facts are illustrated below: (42) a. ni-li-kuwa ni-ki-soma 1sg-past-be 1sg-cont-study “I was studying” b. ni-ta-kuwa ni-ki-soma 1sg-fut-be 1sg-cont-study “I will be studying” c. ni-li-kuwa 1sg-past-be “I had studied”

ni-me-soma 1sg-perf-study

d. ni-ta-kuwa ni-me-soma 1sg-fut-be 1sg-perf-study “I will have studied” (Hinnebusch and Mirza 1998:232) (43) a. Juma a-li-kuwa a-me-pika ch-akula Juma 3sg-past-be 3sg-perf-cook 7-food “Juma had cooked food” b. (Mimi) ni-li-kuwa (1sg-pron) 1sg-past-be “I was still working”

ni-ngali 1sg-still

ni-ki-fanya kazi 1sg-cont-do 9-work

c. ch-akula ki-ta-kuwa ki-ki-pik-wa 7-food 7Agr-fut-be 7Agr-prog-cook-pass “The food will be cooked in the kitchen” d. ku-li-kuwa ku-me-nyesha m-vua 17-past-be 17Agr-perf-rain 9-rain “It had rained”

jiko-ni 5kitchen-loc

The Agreement Parameter

e. m-tindi u-li-kuwa 3-brew 3Agr-past-be “Abunuasi was drunk”

131

u-me-m-vaa Abunuasi 3Agr-perf-3sg-Iar Abunuasi (Carstens 2001: 150)

The analysis of these constructions given in Kinyalolo (1991), Carstens (2001) and Demuth and Gruber (1994) (who discuss Sesotho and Setswana) is that the DP undergoes successive cyclic A-movement into the functional projections dominating the VP, giving rise to multiple agreement. A similar case of successive cyclic A-movement giving rise to multiple agreement is the agreeing complementizer found in Lubukusu (see Washburn 2003). Note that the lack of a subject agreement morpheme sandwiched between the perfect prefix and the verb argues against the Spec-Head agreement hypothesis (e.g., me and pika in (43a)), on the assumption that the subject starts out in Spec vP. If there were Spec-Head agreement in general, then why wouldn’t the DP in Spec vP be able to agree with the light verb v? In the Minimalist framework, Agree requires c-command between the probe and the goal. Therefore, there can be no agreement between v and its specifier. Under the analysis of Kinyalolo (1991) and Carstens (2001), the structure of (43a) is given below: (44)

IP 3 DP Infl’ 3 : Infl VP # past 3 # V AspP # # be 3 z-------- Asp’ 3 : Asp vP # perf 3 #

v’ # 3 z-------m v VP 3 V DP cook food

132

Chris Collins

Carstens (2001) makes the important point that these examples argue that Case deletion does not correlate with phi-completeness (contrary to Chomsky 2000, 2001), since the multiple occurrences of agreement are phi-complete (person, number and gender) in each example. Carstens (2001) also makes the point that gender is always excluded with subject verb agreement in the Indo-European languages. This suggests that agreement of person and number together deletes Case in the Indo-European languages (see Carstens 2001: 149 for discussion). I would like to relate the existence of multiple phi-complete agreement with Kiswahili auxiliary verbs to the Agreement Parameter. Basically, I would like to account for the following correlation: (45) If Agree requires internal Merge (e.g., in Kiswahili), then Agree(X, DP) (phi-complete) does not necessarily check Case. If this correlation holds, it explains why phi-complete agreement does not delete a Case feature in Kiswahili and other Bantu languages. How can we explain this correlation? In the Minimalist framework, there are three types of unintepretable features, listed below (for other types of movement, other uninterpretable features may be involved, such as [uWH]): (46) Three Types of Uninterpretable Features: a. [uPhi] of Infl, v b. OCC on probe c. Case of DP In English and other Indo-European languages, person-number agreement is associated with Case deletion as Chomsky (2001) observes. So in English, valuing [uPhi] (where [uPhi] is a combination of [uPerson] and [uNumber]) gives rise to [uCase] being valued. Now note that in Kiswahili, valuing [uPhi] always gives rise to internal Merge. In other words, valuing [uPhi] is necessarily combined with satisfying an OCC feature. Let us suppose that languages choose one possible grouping: either [uPhi] is grouped together with OCC (giving rise to Kiswahili) or [uPhi] is grouped together with [uCase] giving rise to Case checking of the Indo-European type (see Baker 2002 for a related idea).

The Agreement Parameter

133

7. Conclusion I have shown in this paper how wide ranging differences between Kiswahili and English arise because of the setting of the Agreement Parameter. To the extent that such differences are adequately explained in terms of the Agreement Parameter, I have provided an argument against the Spec-Head relation as the fundamental structural relation of agreement (see Kinyalolo 1991). If agreement always depended on a Spec-Head relation, it would be impossible to account for the differences between English and Kiswahili. A further argument against the Spec-Head agreement hypothesis is that there are several instances of a Spec-Head configuration where there is no agreement (between DP and v, and between DP and Poss). There are a number of cases where a subject precedes an element with which it does not agree overtly. These include the habitual (Keach 1995: 111), the copula ni, and some cases of subjects preceding infinitives 7. Assuming that internal Merge is involved in the positioning of the subject in these constructions, it appears that internal Merge can take place without the probe of Agree having overtly realized agreement features. I put aside these important cases for further research. I have not commented on how the parameter is set by a child learning English or Kiswahili. There are at least three easily observable properties of agreement in Kiswahili that might be relevant to setting the Agreement Parameter. First, Kiswahili and all Bantu languages are pro-drop languages. This suggests that the Agreement Parameter is related to richness of agreement. Second, in Kiswahili, the tense and agreement are independent morphologically in the sense that the same agreement markers appear in all tenses and aspects, unlike English where past tense verbs do not have overt agreement morphology. Third, in Kiswahili, there are no independent morphemes for number and gender agreement, rather a single portmanteau morpheme signals both number and gender information (see Carstens 1991, 1993). These are properties of the agreement system which the child may attempt to exploit in setting the Agreement Parameter. I am not claiming that there is a direct implicational relationship between any of these three properties and a particular setting of the Agreement Parameter. How to derive the Agreement Parameter from these easily observable properties is a topic for further research.

134

Chris Collins

Acknowledgements I would like to thank Aggrey Wasike for his Kiswahili judgements. I would like to thank Aggrey Wasike, Yoshi Dobashi, Mark Baker and three anonymous Triggers reviewers for comments on an earlier version of this paper. Carstens (2002) has independently come to some of the same conclusions found in this paper.

Notes 1. I leave the relationship between sentences such as Ki-tabu ki-po hapa “the book is right here” (Hinnebusch and Mirza, pg. 95) and the Agreement Parameter for future work (thanks to an anonymous reviewer for bringing this issue to my attention). 2. The feature person is represented on Kiswahili verbs. For example, the subject markers are the following: ni (1sg), u (2sg), a (3sg), tu (1pl), m (2pl), wa (3pl). If the uninterpretable feature of na is [uPerson], I must crucially assume that na does not have a [uNumber] feature as well, since person-number combinations correspond to overt morphemes in Kiswahili. 3. Given the agreement parameter, I assume that agreement with post-verbal subjects in relative clause constructions would have to be analyzed in a way similar to that presented in Buell (2002) (see also Keach (1985) and Spence (1997) on relative clauses). 4. Examples such as kabla ya ku-pata “before receiving” (Hinnebusch and Mirza, pg. 117) pose a problem for the hypothesis that associative -a always checks a Case feature of a following DP, since ku-pata is an infinitive. Thanks to an anonymous reviewer for bringing this issue to my attention. 5. Thanks to Mark Baker for pointing out to me the relevance of the Kinande equivalent of the Kiswahili kila “every” to the Agreement Parameter. 6. It is unclear whether Agree(D, NP) is involved in the derivation in (38). If so, the lack of overt agreement would have to be similar to the lack of overt agreement found with subjects in the habitual. I assume that a derivation similar to (38) holds for prenominal demonstratives as well (see Hinnebusch and Mirz, pg. 148). 7. Thanks to an anonymous reviewer for pointing out these later two cases to me.

The Agreement Parameter

135

References Baker, Mark 1996 The Polysynthesis Parameter. Oxford University Press, Oxford. 2002 Agreement, Dislocation and Partial Configurationality. Ms., Rutgers University. Baker, Mark and Chris Collins 2003 Linkers and Verb Phrase Structure in African languages. Ms., Rutgers University and Cornell University. Bresnan, Joan and Jonni M. Kanerva 1989 Locative Inversion in Chichewa: A Case Study of Factorization in Grammar. Linguistic Inquiry 20:1–50. Buell, Leston 2000 Swahili Relative Clauses. Ms., University of California, Los Angeles. Carstens, Vicki 1991 The Morphology and Syntax of Determiner Phrases in Kiswahili. Doctoral Dissertation, UCLA. 1993 On Nominal Morphology and DP Structure. In Sam Mchombo (ed.), Theoretical Aspects of Bantu Grammar. CSLI, Stanford. 2001 Multiple Agreement and Case Deletion: Against Phi-(In)Completeness. Syntax 4.3: 147–163. 2002 Against Spec-Head Agreement: inversion, intervention and EPP. Ms., University of Missouri-Columbia. Chomsky, Noam 1995 The Minimalist Program. MIT Press, Cambridge. 2000 Minimalist Inquires. Step by Step, eds. R. Martin et. al, 89–155. Cambridge, Mass.: MIT Press. 2001 Beyond Explanatory Adequacy. MIT Occasional Papers in Linguistics 20. MIT Working Papers in Linguistics. Collins, Chris 1997 Local Economy. MIT Press, Cambridge. 2003a The Distribution of Particles in Quotative Inversion. Ms., Cornell University. 2003b The Internal Structure of vP in Ju|’hoan and =Hoan. Studia Linguistica 57, 1: 1–25. Demuth, Katherine and Jeffrey Gruber 1995 Constraining XP-Sequences. In Manfredi, V. and K. Reynolds, NigerCongo Syntax and Semantics. Vol. 6: 3–30. Boston Uninversity African Studies Center. Dobashi, Yoshi 2003 Phonological Phrasing and Syntactic Derivation. Doctoral dissertation, Cornell University. Hinnebusch, Thomas J. and Sarah M. Mirza 1998 Kiswahili, 2nd Edition. University Press of America, Lanham.

136

Chris Collins

Hiraiwa, Ken 2001 Multiple Agree and the Defective Intervention Constraint in Japanese. MIT Working Papers in Linguistics 40: 67-80. Keach, Camillia 1985 The Syntax and Interpretation of the Relative Clause Construction in Swahili. Garland Publishing, Inc., New York. 1995 Subject and Object Markers as Agreement and Pronoun Incorporation in Swahili. In Akinlabi Akinbiyi (ed.), Theoretical Approaches to African Linguistics. African World Press, Trenton. Kinyalolo, Kasangati 1991 Syntactic Dependencies and the Spec-Head Agreement Hypothesis in KiLega. Doctoral Dissertation, UCLA. Ndayiragije, Juvenal 1999 Checking Economy. Linguistic Inquiry 30.3: 399–444. Nkemnji, Michael 1995 Heavy Pied-Piping in Nweh. Doctoral Disseration, UCLA. Rizzi, Luigi 1986 On the Status of Subject Clitics In Romance. In Studies in Roamance Linguistics, ed. Oswaldo Jaeggli and Carmen Silva-Corvalan: 391– 420. Foris, Dordrecht. Spence, Justin 1997 Feature Checking and Subject Position in Kiswahili Tensed Relative Clauses. Masters Thesis, Cornell University. Washburn, Paul 2003 Agreement Complementizers and Remnant Movement in Bukusu. Talk given at the 4th World Congress of African Linguistics, June 22.

Some Notes on Emphatic Forms and Displacement in Dutch Norbert Corver

1. Introduction The starting point of this paper are the following pairs from (variants of) Dutch: 1 (1)

a. ’n kist törref a box peat ‘a box with peat’

(1’)

b. ’n hóóp wáeter a heap water ‘a lot of water’ (2)

(3)

a. b. c. d.

a. ’n kiste törref a box-e peat ‘a box with peat’

(dialect of Katwijk)

b. ’n hóópe wáeter a heap-e water ‘a lot of water’

verdomd aardig damned nice

(2’)

ik (I) dat (that) dit (this) wat? (what?)

(3’)

verdomd-e aardig damned-e nice

(colloquial Dutch)

a. ikke (I-e) b. datte (that-e) c. ditte (this-e) d. watte? (what-e?)

The linguistic expressions in (1)–(3) differ from those in (1’)–(3’) in the presence of the element –e, i.e. the sound ‘schwa’. In the former expressions we have a bare category, whereas the latter have a schwa attached to them. In (1’), the schwa combines with the quantity denoting noun of a pseudopartitive construction. In (2’), schwa is attached to a degree adverb contained within an adjectival phrase. In (3’), finally, the schwa is added to different types of pronouns: the first person singular personal pronoun ik, the singular demonstrative pronouns dat and dit, and the neuter interrogative form wat. In traditional reference grammars of Dutch (cf. Overdiep 1937), the schwa-patterns in (1’)–(3’) are often referred to as ‘heavy’ forms and they

138

Norbert Corver

are said to be associated with emphatic or affective force. More specifically, this emphatic force is associated with –e, which is traditionally called an emphatic suffix. The question arises whether these linguistic expressions have other grammatical properties in common besides the surface appearance of -e and the meaning-related property of emphasis/affect. In this article, I will argue that in all of the above constructions a predicative relationship is involved between two elements (a subject and a predicate). I will further propose that the predicate undergoes a leftward predicate displacement process. The element to which –e is (right-)attached is the displaced category which has been moved leftward across the subject over which it predicates. If movement is involved in the derivation of these expressions, a number of questions arise, such as: What triggers displacement of the predicative category? What is the nature of the grammatical marker -e and what operations of the computational system apply to it? And in what way is predicate reordering involved in the derivation of pronominal expressions like ikke, ditte, datte and watte? At the surface, these pronominal expressions seem to be simplex expressions; that is, they do not seem to consist of two subparts (i.e. a subject-like and a predicate-like part). After having given an analysis of the linguistic expressions in (1’)-(3’), I will explore the internal syntax of a number of other grammatical constructions which also feature the grammatical marker –e and which arguably should also be treated in terms of predicate displacement.

2. Predication and Predicate Movement in Nominal Phrases Before turning to a detailed discussion of the linguistic expressions in (1’)(3’), let me first give some background information about the phenomenon of predicate displacement in the nominal domain, since this process will figure prominently in this article. In recent generative studies, a number of nominal construction types have been (re)analyzed in terms of predicate displacement, most notably the so-called N of/van N-construction (cf. 4a,b). Kayne (1994: 106) proposes an analysis according to which a noun like idioot (English: idiot) originates as a clause-internal predicate and is preposed across the subject dokter (English: doctor) in Spec,IP to the specifier position of a clause headed by a prepositional determiner van (English: of). The derived representation is given in (5). (4)

a. die idioot van’n dokter b. that idiot of a doctor

(Dutch)

Some Notes on Emphatic Forms and Displacement in Dutch

(5)

139

die [D/PP [NP idiootj] [van [IP ’n dokter Io [e]j…

An alternative implementation of the DP-internal predicate displacement analysis is given by Den Dikken (1995, 1998). He proposes that in constructions like (4a,b) the displaced predicate originates in a DP-internal Small Clause configuration, i.e. XP. This XP is asymmetrically constituted such that the argument of the predicate, located in the complement position of the Small Clause head X, occupies the specifier position of XP, which – following Bowers (1993) – may als be referred to as Pr(edicate)P(hrase).2,3 Predicate displacement involves movement of the predicate (i.e. the complement of the small clause head) to a higher functional head FP. Schematically: (6)

[DP die [FP idiootj [F’ F (= van)+Xi (= een) [XP dokter [X’ ti tj]]]]]

According to Den Dikken, predicate movement as found in (6) is taken to be an A-movement operation. What characterizes this movement operation is that the inverted nominal predicate skips an intermediate A-position, viz. that of the small clause subject (i.e. Spec,XP). Hence, the movement of the nominal predicate appears to be a non-local A-movement. As Den Dikken points out, however, the predicate movement is local if one adopts Chomsky’s (1993) locality theory in terms of equidistance. Under this theoretical proposal, the moved predicate can cross the subject as long as the two nominals are technically equally far away from the predicate’s extraction site (i.e. tj). Under Chomsky’s assumptions, this situation is obtained by the application of a domain-extending head movement operation that creates a minimal domain that contains both the raised predicate and the small clause subject. Den Dikken argues that in the case of DP-internal predicate inversion, the requisite domain extending head-movement operation consists of raising of the functional head (X) of the small clause to a higher functional head (labeled here as ‘F’). As indicated in (6), I will take X to be the indefinite article een. As noted in Bennis et al. (1998), een is spurious in this nominal environment in the sense that it does not seem to ‘belong to’ the noun that follows it, nor in fact to the noun that precedes it. Normally, the indefinite article is compatible with singular NPs only (see 7a, b). Furthermore, it does not cooccur with proper names and mass nouns (see 7c, d).

140

(7)

Norbert Corver

a.

Ik heb een boek gelezen I have a book read

b. *Ik heb een boeken gelezen I have a books read c. *Ik heb een Westertoren gezien I have a Westertoren seen d. *Ik heb een spinazie gegeten I have a spinach eaten As illustrated in (8), the second noun of the N van een N construction may be plural (8a); een also precedes proper names (8b) and mass nouns (8c). That een does not belong either to the preceding noun (i.e. the displaced predicate) is shown by the existence of examples like (8d), in which the first noun (and also the second one) is plural (data drawn from Bennis et al 1998).4 (8)

a.

?

die ramp van een getalscongruentiefeiten that disaster of a number agreement facts

b.

die pracht van een Westertoren that beauty of a Westertoren

c.

een pracht van een spinazie a beauty of a spinach

d.

die schatten van een kinderen those darlings of a children

Consider, finally, the prepositional element van in (6). With Den Dikken (1995), I will assume that van in the N van een N construction is a nominal copula, which surfaces at PF in the functional head position F. In fact, this nominal copula is considered to be the (nominal) equivalent of the verbal copula (to be). As shown in (9a), the infinitival copula can be freely omitted in copular sentences with a straight subject-predicate order. It cannot be left out, however, in the Predicate Inversion counterpart of (9a) given in (9b) (see Moro 1991; Heycock 1991 for discussion): (9)

a. I consider John (to be) the best candidate b. I consider the best candidate (*to be) John

In Bennis et al. (1998), it is argued that next to predicate displacement of the A-type (henceforth: Predicate Inversion), there is predicate displacement of

Some Notes on Emphatic Forms and Displacement in Dutch

141

the A-bar type (so-called: Predicate Fronting) in the nominal domain. They give the following examples of the latter type: (10) a.

b.

wat een boek(en)! what a book(s) ‘what books!’ wat voor een boek(en)? what for a book(s)’ ‘what kind of books?’

It is assumed that in both the exclamative construction (10a) and the interrogative construction (10b), there is a predication relationship between the noun boek(en) – the subject – and the predicate wat. The surface pattern in (10a) is derived by fronting of the predicate wat across the Small Clause subject boek(en) to a left edge (i.e. A-bar) position within the extended nominal projection (i.e. Spec,DP). It is proposed that the exclamative interpretation of wat is the result of its being moved into the Spec of a functional projection whose head (D) is specified for the illocutionary feature [+EXCL]. Thus, this [+EXCL] operator head, just like the C-head in clausal constructions, triggers overt raising of the wh-form wat to its specifier. By raising to Spec,DP, wat ends up in a position to the left of its subject, as required in the light of the word-order facts. The derived structure then looks as follows: (11) [DP watj [D’ [D[+EXCL] [X een]i] [XP boek(en) [X’ ti tj]]]] As indicated in (11), Bennis et al. assume that movement of wat to [Spec,DP] requires the [+EXCL] operator head (i.e. D) to be lexical in exclamative DPs. They interpret this head movement to D as being parallel to the Verb Second effect in (main) clauses. The way to provide D with content is to raise een, the X-head internal to the Small Clause, to D. Notice that een is the spurious indefinite article: it does not belong to the noun that follows it, as is clear from the fact that a plural noun (boeken) can follow it. Nor does it belong to the wh-form wat (i.e. the fronted predicate). Let us next consider the derivation of the interrogative noun phrase in (10b). Following Bennis et al., I will assume that voor is the lexicalization of the [+WH] operator D-head present in the structure.5 It is this prepositional D-head which defines interrogative force on the nominal expression.

142

Norbert Corver

With voor filling the D-head, the spurious indefinite article is not required to lexicalize D. However, it can still be made use of for the process of Predicate Inversion. Head raising of een to F allows the predicate wat to first undergo predicate inversion to [Spec,FP] before it subsequently undergoes the process of predicate fronting to [Spec,DP]. Schematically, then, the derived structure of the wat voor ’n N-expression looks as follows:6 (12) [DP watj [D’ voor [FP t’j [F’ [F [X een]i] [XP boeken [X’ ti tj]]]]]] Summarizing, two types of predicate displacement operations can be distinguished within the (Dutch) nominal domain: (i) Predicate Inversion and (ii) Predicate Fronting.7 Whenever Predicate Inversion takes place, the functional head F is necessarily included in the syntactic structure for purely structural reasons, having to do with the minimalist theory of locality (equidistance). The presence of nominal copula van can be taken to be a reflex of structurally driven functional head movement of X to F. Predicate fronting moves the predicate to the left edge of the extended projection (i.e. DP). It is driven by the need to associate the predicate with illocutionary properties such as exclamation and interrogation. It should further be emphasized that the motivation for the insertion of spurious een in the exclamative construction is different from the motivation of een in the N van een N construction. In the exclamative noun phrase, spurious een is inserted and moved in order to lexicalize the head of the exclamative operator projection (DP). In the N van een N construction een is inserted and moved in order to allow A-movement to take place without violating strict locality. Having provided the reader with some background of the phenomenon of DP-internal predicate displacement, I will now turn to a discussion of the emphatic expressions in (1’)–(3’) featuring the grammatical marker –e. I will start my discussion with the pseudopartitive pattern in (1’).

3. Emphatic -e in Pseudopartitive Contexts In Corver (1998), it is argued that the derivation of pseudopartitive constructions like those in (13) involve an underlying subject-predicate relationship and feature predicate movement. Of course, from a superficial point of view, the parallelism between the N of N-construction and the pseudopartitive construction is already quite telling: both constructions feature a meaning-less preposition-like element of, which links two nouns.

Some Notes on Emphatic Forms and Displacement in Dutch

143

(13) a. John drank [a bottle of wine] b. John bought [a bunch of flowers] Adopting a predicate displacement analysis (along the lines of Den Dikken 1995 and Bennis et al 1998) of the linguistic expression a bottle of wine, we start with an underlying representation like (14a) and end up with a derived structure like (14b): (14) a. [XP wine [X’ [X Ø] bottle]] b. [DP a [FP bottlej [F’ F (= of)+Xi (= Ø) [XP wine [X’ ti tj]]]]] In (14a), wine – the quantified element – is the subject of the predication relationship, and bottle – the quantity-designating nominal – acts as the predicate. Empirical support for the predicative status of the quantity-designating noun comes from constructions like (15), in which the quantitative noun is the pro-element so. This coordinate pattern yields an approximative reading.8,9 (15) a. a [bottle or so] of wine b. a [liter or so] of water What is important is that so generally functions as a pro-predicate, as is also clear from the examples in (16): (16) a. Mary is very busy and she expects to be so for the next four hours at least b. Mary became a doctor and remained so for the rest of her life The occurrence of so as a right conjunct in expressions like bottle or so is completely in line with an analysis of the quantity-designating nominal as a predicate. That is, the requirement that two conjoined elements be of the same semantic type (e.g. argument & argument; predicate & predicate) is satisfied under a predicate displacement analysis: both the quantitative noun and so are treated as predicates. The conjunction of the two predicates undergoes Predicate Inversion in structures like (15). Further evidence for the parallelism between the N of N-construction and the pseudo-partitive construction comes from the dialect of Katwijk, as described in Overdiep (1940: 139). Consider the following constructions:

144

Norbert Corver

(17) a. D’r waere [’n vrácht van ∂n vaertuige] There were a load of a vessels ‘There were a lot of vessels!’ b. D’r lagge dan [∂n mácht van ∂n tonne] op tie dam There lay then a power of a barrels on that dam ‘There lay a lot of barrels on that dam.’ The bracketed expressions in (17) are quite similar to the N van een N constructions in (4a) and (8). They feature the element van, which we analyze as the nominal copula which shows up in the head of FP as a result of the application of Predicate Inversion. Furthermore, the spurious indefinite article ∂n is present in (17). Note that this article cannot possibly belong to the following noun (vaertuige, tonne), given the plural status of this noun. All this leads to the following derived structure of the linguistic expressions in (17): (18) [DP ’n [FP vráchtj [F’ F (= van)+Xi (= ∂n) [XP vaertuige [X’ ti tj]]]]] The quantity-designating predicate nominal (vrácht/ mácht) has undergone predicate inversion to [Spec,FP], after the small clause head ∂n (X) has raised to the higher functional head F (this for reasons of equidistance). Besides the pseudopartitive pattern featuring the elements ∂n and van, the dialect of Katwijk, as described in Overdiep (1940), also has a pattern in which the two nouns are simply juxtaposed (cf. 19). This juxtaposition pattern for pseudopartitive constructions is also the one found in standard Dutch (cf. 20). (19) a.

Toe krége we [’n hóóp wáeter] en toe riep de skípper… Then got we a heap water and then shouted the sailor…

b. [’n Kist törr∂f] is chaauw óp a packing-case peat is quickly gone (20) a. b.

een hoop water a heap water een kist turf a packing-case peat

In Corver (1998), I argue that these juxtaposed pseudopartitive patterns involve head-movement (i.e. N-raising) of the quantity-designating predicate to the F+X-complex, which results from the domain extending X-to-F raising

Some Notes on Emphatic Forms and Displacement in Dutch

145

operation (see also Corver 2003). From the absence of the nominal copula van in these juxtaposed patterns, one might conclude that the raised predicatenominal head has been substituted for F. Since the F-position is lexically filled in syntax by the raised nominal head, the appearance (i.e. phonological spell out) of the copula van at PF, is blocked. The derived structure looks as follows.10,11 (21) [DP ’n [FP [hóópj+Xi] [XP wáeter [X’ ti tj]]]] Notice that, as opposed to the pseudopartitive variant in (17), there is no spurious ∂n present in (21). In the spirit of Bennis et al. (1998), I will assume that X can be radically featureless (i.e. ø), meaning that it does not possess any morphosyntactic or phonological features. It is a mediating syntactic head which establishes a predication relationship between a predicate and a subject (i.e. external argument).12 What is interesting is that besides the pattern in (19a,b) the dialect of Katwijk also allows the following pattern: 13 (22) a. Toe krège we-n-om ’n uur of drìe toch [’n hóope wáeter], man! Then got we -n- around an hour or three but a heap water, man ‘Oh man, around three o’clock, our boat made so much water!’ b. We hadde-n-[∂n kiste törref] staen We had -n- a packing-case peat stand The expressions ’n hóope wáeter and ∂n kiste törref minimally differ from those in (19), namely in the appearance of the element –e, which is pronounced as ‘schwa’. Importantly, Overdiep (1936, 1937, 1940) remarks that the presence of –e has an interpretative effect: it has an intensifying meaning and signifies strong affect or strong emphasis on the part of the speaker. Thus, while in the neutral-declarative sentences in (19) this intensifying element –e is absent (i.e. the amount is neutrally described), it must be present in constructions where the amount is presented as being amazingly or surprisingly great. Thus, hóope in (22a) designates a surprisingly great amount of water.14 Some additional examples of this intensifying pseudopartitive pattern are given in (23) (Examples are drawn from Overdiep 1936, 1937, 1940): (23) a. Toe vinge we daer in ’n week [∂n skuite haering]! Then captured we there in a week a barge herring ‘We then caught such an enormous amount of herring in a week!’

146

Norbert Corver

b. De boere, die reen altijd nae Lâaje toe mit [∂n vrachte visch]! The farmers, they drove always to Leiden PRT with a load fish ‘Those farmers always drove to Leiden with such an enormous amount of fish!’ c. Hael effe-n-[∂n nemmere waeter] Get just-n- a bucket water ‘Get me a large bucket with water!’ d. Die vent die ving [’n berge haering] That man that caught a mountain herring ‘That man caught an enormous amount of herring!’ e. We zette-n-[∂n potte brokken] op (’t vuur) We put-n- a pot lumps on (the fire) ‘We put a really large pot full of lumps on the fire!’ The question, of course, arises how to analyze these emphatic pseudopartitive constructions. Let us first consider the status of the element –e. Overdiep refers to it as an inflectional ending. Importantly, it should be distinguished from the inflectional ending –e that appears on attributive APs that enter into an agreement relationship with a noun, as in ’n ròòij-e ochend (a red morning; ‘a morning with the rising sun’) and ’n skoon-e gront (a clean soil). This inflectional ending –e does not appear on attributive APs modifying singular neuter nouns that are preceded by an (overt or covert) indefinite article, as in blâeuw waeter (blue water; i.e. transparent water) and ’n skoon overhemd (a clean shirt-blouse). Given the absence of -e on APs that modify singular neuter nouns, it can be concluded that the element –e on nemmer in (23c), which is followed by the neuter noun waeter, is not the same type of inflectional element that appears on attributive adjectives.15 When we look at what can intervene between two nominal elements in contexts of DP-internal predicate displacement, we come to two possible candidates for the analysis of –e. First, it could be analyzed as an instance of the nominal copula; i.e. it would be a bound-morphemic equivalent of van and as such occupy the F-position. Secondly, it could be treated as an instance of the Small Clause head X. Under this second interpretation, it would be treated on a par with the spurious indefinite article een. The former analysis is excluded if one adopts the above-mentioned proposal that the raised predicate nominal (i.e. N) substitutes for F, so that PF-spell-out of this position by means of a nominal copula is blocked. This brings us to the second option: -e as an instance of the Small Clause head X, i.e. the head position which is also the locus of the spurious indefinite article. This analysis

Some Notes on Emphatic Forms and Displacement in Dutch

147

would give us the following representation of the string ’n hóope wáeter in (22a): (24) [XP waeter [X’ –e [hóop]]] (25) [DP ’n [F’ [F hóopj]+Xi (= -e) [XP waeter [X’ ti tj]]]]] In this structure, the Small Clause head X (= –e) raises to F for reasons of domain extension: after X has raised to F, the predicative head hóop can move across ti into the [F+X]-complex, where it substitutes for F. I will assume here that the Small Clause head –e carries emphatic meaning.16 Suppose further that the quantity designating predicate receives an emphatic/affective reading (i.e. “a surprisingly great heap” in (25)) when it gets attached to –e. In a way, the emphatic reading of the displaced predicate is structurally determined.

4. Emphatic –e in Adjectival Contexts As noted in Verdenius (1939b), adverbs of degree that precede a predicative adjective may optionally get attached the grammatical marker –e in Dutch.17 In present-day colloquial Dutch, the occurrence of this grammatical marker is rather limited. In the dialect of Groningen (a Low Saxon dialect), this phenomenon seems to be more common. In (West) Frisian, the emphatic marker –e obligatorily occurs on intensifying degree adverbs that modify a predicative adjective (cf. also Hoekstra 1997). (26) a. Die auto is [verdomd-e duur] That car is damned expensive

(colloquial Dutch)

b. Dat meisje is [verrekt-e aardig] That girl is damned nice (27) a. ’t Was [gloepend-e mooi] it was very-E beautiful b. ’t Was [donderz-e kòld] it was deuced-E cold c. ’t Was [bitterliek-e kòld] it was bitter-E cold

(dialect of Groningen, Ter Laan 1953)

148

Norbert Corver

(28) a. dat skilderij is [wakker-e moai] ([West] Frisian, Hoekema 1996) that painting is very-E beautiful b. hy wie [stjerrend-e benaud] He was dead-E scared c. hy rôp [ôfgryslik-e lû] he shouted terrible-E loud Verdenius (1939b: 363) notes that only those degree adverbs can carry –e which have a strongly affective or emphatic force. More neutral degree adverbs (e.g. matig ‘moderately’) generally do not permit attachment of –e. The question arises how to analyze these adjectival expressions featuring the grammatical marker –e. Could –e, for example, be interpreted as a grammatical marker that attaches to a displaced predicate, just like in the pseudopartitive constructions of the previous section? Even though the standard analysis of the sequence ‘degree adverb + adjective’ is one in which the two constituents are simply combined via (external) merge, there are phenomena in certain languages which hint at the application of predicate displacement in adjectival constructions featuring a degree adverb. A very suggestive case is provided by Romanian (cf. Corver 2000, 2001). Romanian has constructions like those in (29) (examples drawn from Mallinson 1986): (29) a. Ion e [extrem de înalt] Ion is extrem of tall-ms ‘Ion is extremely tall’ b. Maria e [enorm de fericita] ‘Maria is enormously happy’ c. Alearga [enorm de repede] dar înoata cam încet run-3s enormous of quickly but swim-3 rather slowly He runs enormously quickly but swims rather slowly’ d. un rol [deosebit de important in aceasta action] […] a-m role particular of important-ms in this-fs action ‘a particularly important role in this action […]’ What makes these examples very interesting is the fact that the degree adverb (e.g. extrem) and the gradable adjective (e.g. înalt) are separated from each other by an intervening preposition-like element de.18 As noted in

Some Notes on Emphatic Forms and Displacement in Dutch

149

Corver (2000, 2001), the N of N construction and the A de A construction display similar syntactic behavior. As shown in (30) and (31), the first lexical element cannot be moved away from the sequence of N/de A. Furthermore, the latter sequence cannot be fronted to the beginning of the clause (cf. (32)). (30) a. Fools of policemen, they certainly are – ! b. *Fools, they certainly are [– of policemen]! (31) a. Extrem de destept te mai crezi tu! Extremely of clever yourself still believe you ‘You fancy yourself extremely clever!’

(Grosu 1974)

b. *Extrem te mai crezi tu [– de destept]! (32) a. *Of policemen they certainly are fools b. *De destept te mai crezi tu [extrem –]! Of clever yourself still believe you extremely ‘You fancy yourself extremely clever!’ Whatever the exact analysis of these facts, they are suggestive of a parallel treatment of the two construction types. If one adopts a predicate displacement analysis for the N of N construction, one should also do so for the Romanian A de A construction. Such an analysis leads to the following derivation of a linguistic expression like extrem de înalt in (29a): In the ‘underlying’ structure, there is a predicative relationship between the gradable adjective înalt – the ‘subject’ of the predication – and the degree adverb extrem – the predicate of the relationship. Under the assumption that this predication relationship is configurationally defined in terms of a Small Clause configuration, we end up with an underlying structure like in (33). I will assume that predication involves association (say: coindexation) of the external argument (i.e. 1) of extrem with the G(rade)-argument of the gradable adjective înalt. This association leads to the interpretation: ‘tall to degree D, where D is extreme’. (33) [XP înalt [X’ X extrem]] Application of the Predicate Inversion process leads to the derived representation in (34). The linking element de can now be interpreted as the nominal

150

Norbert Corver

copula, which shows up as a surface reflex of the application of A-type predicate displacement. (34) [FP extremj [F’ F (= de)+Xi [XP înalt [X’ ti [AP t]j]]]] As shown by (35) it is impossible to remove the degree adverb out of the adjectival projection. Pied piping is required. (35) a. *Extrem te mau crezi tu [– de destept]! Extreme yourself still believe you – of clever ‘You fancy yourself extremely clever!’ b. [Extrem de destept] te mau crezi tu! When the degree word can be shifted further to the left edge of the adjectival projection (i.e. an A-bar position) – for example, when the degree word carries an interrogative feature – it can leave the extended adjectival projection: (36) a. Cît de frumoasa e Maria! How-much of beautiful is Maria ‘How beautiful Maria is!’

(Grosu 1974)

b. Cîti e Maria [ti de frumoasa]! I will assume that the string cît de frumoasa has the derived representation in (37). The wh-word cît has first undergone Predicate Inversion, triggering the presence of the nominal copula de. The wh-word is subsequently moved to Spec,DegP via Predicate Fronting. (37) [DegP cîtj [Deg’ Deg[+WH] [FP t’j [F’ F (=de)+Xi [XP [AP frumoasa] [X’ ti [AP tj]]]]]]] Grosu (1974) also observes that there are patterns which lack the intervening linking element de. The absence of the nominal copula and the possibility of subextracting the degree adverb, as shown in (38), suggest that the degree adverb – adjective order results from the application of direct Predicate Fronting of the adverb from its base position to the A-bar position Spec,DegP (cf. 39). (38) a. Tare destept te mai crezi tu! extremely clever yourself still believe you `You fancy yourself real clever!’ b. Tare te mai crezi tu [– destept]!

(Grosu 1974)

Some Notes on Emphatic Forms and Displacement in Dutch

151

(39) [DegP tarej [Deg’ Deg [XP [AP destept] [X’ X [AP tj]]]]]]] (Predicate Fronting) Having provided some evidence for the existence of predicate displacement within the adjectival system on the basis of Romanian, let us see whether there is evidence for predicate displacement within the Dutch adjectival system. Given Chomsky’s (2001) Uniformity Principle, which states that “In the absence of compelling evidence to the contrary, assume languages to be uniform, with variety restricted to easily detectable properties of utterances”, one would expect there to be instances of predicate movement in the Dutch adjectival projection. Even though I know of no examples in which the preposition-like element van intervenes between the degree word and the gradable adjective, there are examples in which a linking element –s appears in between the two constituents. In Royen (1948: 342), the following examples are given from late 19th and early 20th century Dutch. (40) a. …alhoewel dat [dekselkaters lastig] aan me vallen zal …although that cover+tom-cat-s difficult to me fall will ‘…although that will be deucedly difficult for me’ b. Ben jij die jongen die [zoo bliksems mooi] kan teekenen Are you that boy who so lightning-s beautifully can draw ‘Are you that boy who can draw so bloody/devilsihly well?’ c. Ja die Marianne weet [weerga’s handig] het vuurtje te stoken Yes that Marianne knows lightning-s handy the fire to make Yes, Marianne knows devilishly well how to make a fire’ d. Het was [verdraaid(s) lekker] It was distorted(s) tasty ‘It was deucedly tasty’ But also in present-day Dutch we run into adjectival expressions like: sterven-s benauwd (die-s sultry; ‘very sultry’), dood-s bang (death-s afraid, ‘very afraid’), hond-s brutaal (dog-s impudent; ‘very impudent’), bliksem-s goed (thunder-s good; ‘very well’), mieter-s lastig (damned-s difficult; ‘very difficult’), deksel-s mooi (deuced-s beautiful; ‘very beautiful’), drommel-s heet (deuced-s hot; ‘very hot’), duivel-s aardig (devil-s kind; ‘very kind’), hel-s koud (hell-s cold; ‘very cold’).19 In all of these expressions, the degree denoting element and the gradable adjective are separated from each other by an intervening ‘linking’ element –s, which, in view of the parallelism with the Romanian adjectival construction in (29), I analyze as a nominal copula, i.e. the surface reflex of the

152

Norbert Corver

presence of F in predicate inversion environments. The derived structure then looks as follows:20 (41) [FP duivelj [F’ F (= –s) +Xi [XP aardig [X’ ti [AP t]j]]]] devil –s kind (‘very kind’) Let us now turn back to the examples in (26)–(28), which feature the element –e rather than –s. I will assume that –s is the bound-morphemic copula that surfaces in contexts of predicate inversion. The grammatical marker –e should then be a different type of grammatical marker. I propose that the emphatic marker –e is the Small Clause head X, just like in the emphatic pseudo-partitive constructions of the previous section. This analysis leads us to the following representation of an example like wakkere moai in (28a). (42) [DegP wakkerj [Deg’ [Deg –ei] [XP moai [X’ ti [AP t]j]]]] As indicated in (42), the grammatical element –e (i.e. X) raises to Deg (the operator head position of the extended adjectival projection). I will assume that quite along the lines of the exclamative wat-construction (cf. section 2), the X head has to raise to Deg in order to lexicalize the operator head position, into whose Spec the emphatic phrase has landed. Let me close off this section with the expression donderz-e kòld in (27b). This example from the dialect of Groningen is interesting for the following reason: it exemplifies the application of Predicate Inversion plus subsequent Predicate Fronting to the left edge of the extended adjectival projection.21 Close investigation of the internal make-up of the degree designating expression donderze leads to the identification of two bound-morphemic elements: –s (orthographically realized as –z here) and –e. The former we analyzed as the nominal copula, the latter as the emphatic grammatical marker (i.e. the Small Clause head X). This then leads to the following derived structure of donderz-e kòld. (43) [DegP donderj [Deg’ [–s+-ei]k [FP t’j [F’ tk [XP [AP kòld] [X’ ti [AP tj]]]]]]]

5. Emphatic –e in Pronominal Contexts In this section, I will consider the internal syntax of the pronominal forms in (44’), which are all emphatic forms of the pronominal elements in (44).22

Some Notes on Emphatic Forms and Displacement in Dutch

(44) a. b. c. d.

ik dat dit wat?

(I) (that) (this) (what?)

(44’) a. b. c. d.

ikke datte ditte watte?

(I+e) (that-e) (this-e) (what-e?)

153

(Dutch)

The emphatic forms in (44’) carry the grammatical marker –e. In this respect, they are similar to the emphatic categories that we have analyzed thus far. The question then arises, of course, whether these pronominal forms have the same internal syntax as the emphatic expressions that were previously discussed. More in particular, the question arises, whether these pronominal forms feature a predicative relationship underlyingly, and whether predicate displacement might be involved. At first sight, such an analysis seems highly unlikely given the fact that these emphatic forms appear to exist of a single lexical item; since predication involves a relationship between two elements, there seems to be no such relationship involved in these emphatic pronominal expressions.23 A very interesting observation, however, has been made by Sassen (1953: 118) in his book on the dialect of Ruinen (a dialect of Low Saxon). Sassen makes the observation that in this dialect the emphatic form ikke never cooccurs with a finite verb. Sassen’s generalization, which extends to standard Dutch, may alternatively be formulated as follows: the emphatic pronominal ikke cannot function as a (subject) argument of a verb. In a sense, it displays the behavior of a (nominal) root expression. This restricted distributional behavior of ikke is illustrated in (45) with some examples from Dutch. Notice that the pronominal form ik, its near-surface-equivalent, cán function as an argument within the clause. (45) A: Wie is er ziek? B: Ik/Ikke. (46) a. Jan dacht dat ik/?*ikke ziek was John thought that I/*I-e ill were b. Die jongen ken ik/?*ikke niet That boy know I/*I-e not This restricted distribution is also found with the other emphatic pronominal forms.

154

Norbert Corver

(47) a. Dat/*Datte heeft mij aangevallen That has me attacked b. Jan heeft dat/*datte uitgespuugd Jan has that/*that-e out-spitted (48) a. Dit/*Ditte zat in mijn tas This/*This-e was in my bag b. Ik heb dit/*ditte vandaag gevonden I have this/*this-e today found (49) a. Wat/*Watte zit er in die tas? What/*What-e is there in that bag b. Wat/*Watte heb je gekocht? What/*What-e have you bought c. Ik vraag me af [wat/*watte hij gekocht heeft] I wonder REFL PRT wat/*what-e he bought has d. Jan heeft WAT/*WATTE gekocht? John has WHAT/WHAT-E bought In all these examples, the emphatic pronominals cannot occur as arguments of the verb. Notice now that the emphatic forms are permitted when they occur in isolation, for example as answers to questions (50)–(52) or as a reaction to incomprehensible speech – cf. (53). (50) A: Wat heeft Jan uitgespuugd? A: What has John out-spitted

B: Datte B: That-e

(51) A: Wat zat er in Maries tas? A: What was there in Mary’s bag?

B: Ditte B: This-e

(52) A: Jan heeft iets leuks gekocht? A: Jan has something nice bought?

B: O ja? Watte? B: Oh yes? What-e?

(53) A: xxxxx (murmured speech)

B: Watte?

One might interpret this restricted distributional behavior of the emphatic pronominals as evidence that they do not constitute regular nominal phrases (say, DPs). If they were DPs, why wouldn’t they be able to occur in positions where other (pronominal) DPs are permitted. One way to go would be to say that the emphatic forms in in (44’) are ellipted constituents, more in

Some Notes on Emphatic Forms and Displacement in Dutch

155

particular ellipted small clausal constituents. This would imply that one element of the subject-predicate relationship of the Small Clause configuration is missing. I would like to propose that the missing element is the subject. More specifically, I propose that the empty subject (let’s assume a ‘pro’) represents information that is pragmatically presupposed (i.e. familiar on the basis of information previously provided in the discourse). In (45), for example, there is the presupposed (i.e. background) information that ‘there is a person X who is ill’, when person B answers person A’s question. And in (50), to use another illustration, there is the presupposed information ‘there is a thing X which John has spitted out’, when B gives his answer ‘Datte’. The emphatic pronominal provides the new, focused information. In (45), it identifies the person X who is ill; and in (50), it identifies the thing X which John has spitted out. Given this interpretation of the emphatic pronominals, we arrive at the following underlying structure: (54) [XP pro (i.e. presupposed information) [X’ [X –e] ik/dat/dit/wat]] In line with our interpretation of the grammatical marker –e thus far, I will assume it to be the head of a Small Clause structure. The surface structure will be derived by fronting of the pronoun to an A-bar position to the left of the small clause subject pro. For the sake of explicitnesss, I have termed the projection containing this A-bar position CP. The marker –e raises to the functional head C, in order to lexicalize this functional head (a phenomenon which is reminiscent of the Verb Second effect in contexts of operator movement within a clause). The derived structure then looks as follows:24 (55) [CP ik/dat/dit/wati [Z’ –ej [XP pro [X’ tj ti]]]] At this point it is interesting to also mention the following patterns, in which the emphatic pronoun ikke combines with the negative adverb niet or the positive adverb wel (cf. Sassen 1953). (56) A: Houd jij van voetbal? Like you of soccer B: Ikke niet / Ikke wel I-e not (‘I don’t’) / I-e certainly (‘I do’) Under the assumption that the adverbs niet/wel occupy the Spec-position of some functional head within the (small) clausal projection, we end up with the following representation:

156

Norbert Corver

(57) [CP iki [C’ –ej [YP niet/wel [Y’ t’j [XP pro [X’ tj ti]]]]]] The Small Clause head –e raises to the functional head Y and moves on to C. The pronominal ik undergoes A-bar movement to the left edge of the small clause structure. Consider also the following structural context, noted by Sassen (1953), in which the emphatic form ikke may appear: (58) Wij, mijn vrouw en ikke, gaan vandaag naar de stad We, my wife and I-e, go today to the city At first sight, the appositional phrase mijn vrouw en ikke – a coordinate structure – seems problematic for my analysis, that is, if one assumes an analysis in which mijn vrouw is coordinated with ikke: [[mijn vrouw] en [ikke]]. If my analysis according to which ikke is a small clause-like structure is correct, then the coordinate pattern (a noun phrase together with a small clause structure) is rather strange, so it seems. Notice, however, that an alternative analysis is available, which makes perfect sense under the proposal given above. This analysis is the following: (59) wij, [CP [mijn vrouw en ik]i [C’ –ej [XP pro [X’ tj ti]]]], … In this structure, the coordinated structure is mijn vrouw en ik. It is this coordinate phrase which is preposed to the left edge of the Small Clause. The emphatic marker –e raises to C, the operator head, yielding the surface sequence: mijn vrouw en ik-e. The empty subject pro (i.e. the presupposed information) arguably corresponds to the referent of wij.25 Let us, finally, turn back to Sassen’s generalization that, for example, the pronominal form ikke does not co-occur with a finite verb.26 The reason for this distributional restriction may relate to the fact that small clauses typically do not occur as subject arguments of verbs. To give an example, even though it is possible to have a small clausal phrase as a left dislocated expression (i.e. external to the clause) (cf. 60), it is impossible to have it as an argumental expression in the subject position of the clause: (60) a. [Jan een boek lezen], dat lijkt me onwaarschijnlijk. John a book read, that seems me unlikely b. *[Jan een boek lezen] lijkt me onwaarschijnlijk John a book read seems me unlikely

Some Notes on Emphatic Forms and Displacement in Dutch

157

6. Other Environments Featuring –e Thus far we have come across three types of linguistic expressions which, at first sight, do not seem to have much in common besides the appearance of the grammatical marker –e and the fact that they share the property of having an emphatic/affective meaning. I have argued that these expressions also share a syntactic property: they all feature the phenomenon of predicate displacement (and the phenomenon of head movement of X to a higher functional head). The question, of course, arises whether there are more grammatical constructions that display this set of properties. Even though an elaborate discussions of various constructions is beyond the scope of this article, I would like to point out a few expressions which, at a more abstract level, can be considered to be instances of the same type of syntactic construct. A first construction I would like to mention here is the Dutch vocativelike expression in (61): (61) arm-e ik! poor-e I ‘poor me!’ This expression has the remarkable property that what appears to be an attributive adjective precedes a pronominal element. Normally, however, attributive adjectives do not modify pronouns. Notice also that when the expression functions as an argument within a clause, it becomes less acceptable: (62) ??Jan dacht dat [arme ik] ziek was John thought that poor-e I ill was Given this special behavior of this expression, one might wonder whether arme in arme ik! is a truly attributive adjectival phrase (i.e. a phrase as we find it in: een arme jongen; a poor-e boy). Also when we look at equivalent expressions in other languages, it becomes clear that arme ik! has a ‘special’ syntax. Consider, for example, the following expressions from French, Portuguese and Spanish: (63) a.

pauvre de moi! poor of me ‘poor me!’

(French)

158

Norbert Corver

b. pauvres de nous! poor-pl of us ‘poor us!’ (64) a. pobre de mim! poor of me

(Portuguese)

b. pobres de nós! poor of us (65)

pobrecitos de nosotros poor-pl of us

(Spanish; cf. Jespersen 1977)

Of course, what is interesting about these examples is the presence of the linking element de. Crucially, the sequence de+pronoun is not a PP-complement of the preceding adjective. Rather, there is a predicative relationship between the pronoun and the adjective: an example like pauvre de moi! has the interpretation that the property ‘poor’ is predicated over ‘me’ (i.e. the speaker). As the predicate precedes the subject, an analysis seems likely in which the predicate is fronted via predicate displacement to a position preceding the subject. The appearance of the nominal copula de suggests that predicate displacement is of the A-movement type here (i.e. movement to Spec,FP). Schematically:27 (66) [FP pauvrej [F’ F (= de)+Xi [XP moi [X’ ti [AP t]j]]]] As shown in (65) for French, it is also possible to leave out the linking element de: (67) a. Pauvre moi! Poor me b. Pauvre lui! Poor him If de shows up in contexts of Predicate Inversion (i.e. A-type predicate displacement), one might want to analyze the pattern in (67) as an instance of Predicate Fronting (i.e. A-bar-type predicate displacement); compare the contrast between extrem de destept and tare destept in Romanian adjectival expressions (cf. section 4). If Predicate Fronting is involved in (67), the derived structure looks as follows:

Some Notes on Emphatic Forms and Displacement in Dutch

159

(68) [CP pauvrej [C’ C [XP [DP moi] [X’ X [AP tj]]]]] Suppose now that in Dutch, as opposed to French, there is the ‘Verb Secondlike’ requirement that the operator head position must be lexically filled when some phrase has been moved into the A-bar position (see my discussion of the wat-exclamative discussion in section 2). The grammatical marker –e, which I assume to be the Small Clause head, raises to C in order to satisfy this lexicalization requirement. The resulting derived structure of the string arme ik! is then the following: (69) [CP armj [C’ -ei [XP [DP ik] [X’ ti [AP tj]]]]] The analysis given above of vocative expressions like arme ik, pauvre (de) moi! et cetera can arguably be extended to phrases in which the noun following the (strictly speaking predicative) adjective is a non-pronominal element (e.g. a proper name). (70) arm-e Jan! poor-e John Normally, proper names cannot be preceded by an attributive adjective either: (71) a.

Ik heb Jan gezien I have John seen

b. ??Ik heb arm-e Jan gezien I have poor-e John seen The well-formedness of the expression in (70) suggests that it is derived along the same lines as arme ik!. That is: (72) [CP armj [C’ -ei [XP [DP Jan] [X’ ti [AP tj]]]]] And in French, which does not require the lexicalization of C when the A-bar position is occupied by some operator-like (e.g. focalized) phrase: (73) [CP pauvrej [C’ C [XP [DP Jean] [X’ X [AP tj]]]]]

(i.e. Pauvre Jean!)

An analysis of strings like arme Jan/pauvre Jean along the lines depicted in (72)/(73) paves the way for the analysis of ‘phrases of addressing’. Dutch

160

Norbert Corver

phrases like Beste Jan (Dear-e John), Lieve Marie (Dear-e Mary), i.e. phrases that you use to address people in letters or speeches, plausibly receive an analysis like (74): (74) [CP bestj [C’ -ei [XP [DP Jan] [X’ ti [AP tj]]]]] And in a language like French:28 (75) [CP cherj [C’ C [XP [DP Jean] [X’ X [AP tj]]]]]

(i.e. Cher Jean)

Consider, finally, the vocative expressions from Dutch in (76): (76) a. Hemeltje lief! Heaven-DIM dear ‘Good heavens!’ b. Lieve hemel! Dear-e heaven ‘Good heavens!’ What makes these examples interesting is that, even though they are fixed expressions – you cannot say, for example, Lief Hemeltje! – they show the characteristics that would be expected under our analysis: in the vocative expression (76a) we have the basic small clause pattern with the straight word order (i.e. subject-predicate: ‘heaven (is) dear/good’). In (76b), we have the pattern featuring predicate displacement. The structural representations of the two expressions is given in (77): (77) a. [XP hemeltje [X’ X lief]] b. [CP liefi [C’ -ej [XP [NP hemel] [X’ ti [AP tj]]]]]]]

7. Concluding Remarks In this article, I have examined a variety of construction types (from Dutch, (West) Frisian, the dialect of Katwijk, the dialect of Ruinen) that feature the grammatical marker –e, an element which has often been said to carry emphatic/affective force in the traditional literature. I have argued that the element (i.e. a quantity designating noun, a degree adverb and a restricted set of pronouns) to which this so-called marker of emphasis is attached constitutes

Some Notes on Emphatic Forms and Displacement in Dutch

161

a displaced predicate. I have furthermore proposed that –e should be treated as the mediating Small Clause head, which takes the predicate as its complement and the ‘subject’ as its specifier. This mediating head can raise to higher (functional) head positions. With Bennis et al. (1998), I argued that head raising of X is either motivated by purely structural reasons of locality (i.e. equidistance) or by reasons of lexicalization of the operator head position of a phrase whose Spec contains a fronted constituent. I have argued that –e is a small clause head whose movement enables structural domain extensions (for reasons of locality; i.e. X-to-F movement) or lexicalization of operator-heads. As such, it fulfills the same function as the spurious indefinite article een that appears in a variety of nominal environments: N van een N, pseudopartitve constructions (cf. section 3), wat voor een N-constructions and wat-exclamative phrases. In view of the fact that the emphatic marker –e is superficially similar to the adjectival inflection –e, one might wonder whether the former could be an instance of a spurious –e, i.e. an inflectional element that does not seem to belong to what follows and what precedes (cf. the discussion in endnote 15). One might want to say that just like the spurious indefinite article een, the spurious inflection –e is featurally un(der)specified and that this accounts for its unexpected distribution: e.g. combination with a following pronoun (arm-e ik) and a proper name (arm-e Jan) or attachment to a preceding noun and combination with a following indefinite neuter noun (cf. ∂n nemmere waeter). Although a full investigation of the existence of a spurious inflection –e is beyond the scope of this paper, I would like to point out that if een and –e are instances of the mediating Small Clause head (X) and if both elements can have a ‘spurious nature’, one might expect to find the two elements in the same types of structural environments. Now consider the following pairs: (78) a. Hij heeft [een mirakel van een bibliotheek] He has a miracle of a library b. Hij heeft [een mirakel-e biliotheek] He has a miracle-e library ‘He has a miracle of a library’

(Overdiep 1937: 166)

(79) a. wat een grote vis! What a big fish

(standard Dutch)

b. Wát-∂ gróote viss! What-∂ big-e fish! ‘How big a fish!’

(dialect of Katwijk; Overdiep 1940: 91)

162

Norbert Corver

(80) a. wat voor een jongens what for a boys ‘what kind of boys’ b. Wa ff r- jòang s Wat-for-e boys ‘what kind of boys’

(standard Dutch)

(the dialect of West-Voorne; Van Weel 1904)

(81) a. Ik haar zoo’n ontzettend-en dörst I have such a great-a thirst ‘I am so extremely thirsty’ b. I heb zo’n ontzettend-e dorst I have such a great-e thirst

(dialect of Groningen; Ter Laan 1953) (standard Dutch)

In the a-examples, we find instances of the spurious indefinite article. Notice, in particular, pattern (81a) from the dialect of Groningen. As noted in Ter Laan (1953: 37–38), the adjective can be followed by the indefinite article en when the adjective carries an emphatic meaning.29 As shown by the b-examples, the inflectional element –e can occur in those structural environments where we also find the spurious indefinite article. This distributional parallelism may further strengthen the idea that both the indefinite article een and the (adjectival) inflectional element –e have spurious counterparts in syntax.30 Being instances of the Small Clause head X, both elements arguably undergo movement for the same reasons: either to enable domain extension in contexts of Predicate Inversion or to enable lexicalization of an operator head in contexts of Predicate Fronting.

Acknowledgements I would like to thank two anonymous reviewers for helpful and insightful comments. I am also grateful to Joost Zwarts for discussion of certain parts of this paper.

Some Notes on Emphatic Forms and Displacement in Dutch

163

Notes 1. In Dutch orthography, there are two ways of spelling the singular indefinite article: een and ’n. In the examples from the dialect of Katwijk, we will also encounter the form ∂n for the indefinite article. This is a phonetically based spelling given in Overdiep’s (1936) study on this dialect. In this article, the spelling used for the examples from the Katwijk dialect are those given in Overdiep’s study. 2. For the idea that predication (in the clausal system) is mediated through a functional head, see among others Hornstein and Lightfoot (1987), Moro (1988). 3. Along the lines of Stowell’s (1981) original insight, I will assume that the predicative category belongs to the set of lexical categories (but see section 5) and that, semantically, the predication relationship consists of saturating or completing a structure that is inherently unsaturated or incomplete (i.e. the “Fregean view”). In the small clause configurational structure adopted in this paper (i.e. XP), it is the complement to the small clause head (i.e. X), which functions as the unsaturated predicate. The argument in [Spec,XP] saturates the structure. Following Chierchia (1984), I will assume that the lexical phrase (e.g. NP) in the complement position of X denotes a simple property and that the small clause head X defines a predication relationship which informally reads as: “the property associated with the complement of X holds of the subject”. See Bowers 1993 and Adger and Ramchand 2003 for further discussion. 4. As pointed out in Bennis et al. (1998), there is some variation among speakers of Dutch with respect to their judgments of some of the examples in (8). This may be related to the speaker variation found in the N van een N constructions with differences in gender between the two noun phrases (cf. Everaert 1992 for relevant discussion). The judgments given in (8) are those given in Bennis et al. (1998). 5. The idea that voor is a [+WH] operator head (i.e. a prepositional element with an interrogative force) is plausible in light of the fact that many southern varieties of Dutch and substandard Dutch in fact use voor as the infinitival complementizer in constructions that feature operator movement to Spec,CP, as the example of an infinitival relative in (i) shows (cf. Bennis et al. for discussion). (i) een boek [OP voor in te kijken] a book for into to look ‘a book to look into’ 6. Bennis et al. (1998) discuss another pattern of the wat voor-construction, viz. one in which there is direct movement of the predicate (wat) to Spec,DP. In those constructions, the small clause head is said to be empty as regards its semantic and morphosyntactic feature specification. 7. In section 3, I will discuss a third type of predicate movement, viz predicate displacement of the head-movement type (N-raising).

164

Norbert Corver

8. Interestingly, the approximative pattern in (15) is also found with numerals: [twenty or so] minutes. See Corver (2001) for a discussion of this pattern and for further arguments in support of the predicative status of numerals. 9. Not all pseudopartitives allow the pattern in (15): e.g. ??a bunch or so of flowers. The intuition seems to be that so only appears with measures, while bunch is a collective. Another issue raised by the pattern in (15) is the fact that besides a pattern like two liters or so of water, English also permits: two liters of water, or so. For those cases, one might explore an analysis according to which not the entire quantity designating phrase has undergone DP-internal predicate displacement, but only the left conjunct of the coordinated phrase (leaving behind the sequence or so). Schematically: (i) [FP [two liters]i [F’ of [XP water [ConjP ti [Conj’ [Conj or] so]]]]] See Corver (1998) for a discussion of this split pattern in Dutch and Modern Hebrew. 10. The structure in (21) predicts that N1 (hóóp) cannot be phrasal. As pointed out by an anonymous reviewer, this may lead one to expect that it cannot be modified by an attributive adjective, as in a phrase of the type ‘a large heap of water’. Unfortunately, I haven’t been able to find such examples in the studies by Overdiep. But is should also be noted that attributive modifiers arguably occupy a position external to the N (of) N sequence. As is clear from the one-pronominalization test, an adjective like small in a small bottle of beer does not form a constituent with bottle. That is the representation is not: [a [[small bottle] of beer]]. If this were the right representation, one would incorrectly predict that the string bottle of beer could not be substituted for by the pro-form so: 11.

12.

13.

14.

(i) John drank [a small bottle of beer] and Mary [a large one] One might alternatively want to argue that the quantity designating noun (hoop) first undergoes head movement to X and that the complex head [N+X] subsequently raises to F. This sequence of head movement steps is compatible with the Head Movement Constraint. The analysis presented in the main text can be made compatible with the Head Movement Constraint if it is assumed that after X-to-F raising, the empty X in XP is invisible for head raising of the nominal head of the predicate. Bennis et al. assume this featureless small clause head to be present in wat voor-constructions that lack the spurious indefinite article, like in: wat voor boeken. In Standard Dutch, the effects of the emphatic suffix –e in the Katwijk dialect seem to be ‘mimiced’ by stress. Compare the neutral stress pattern een berg HAring (a mountain herring; ‘a lot of herring’) with the pattern featuring emphasis on the quantity designating element: een BERG HAring (a mountain herring; ‘a surprisingly large amount of herring’). This emphatic marker in pseudopartitive constructions was also found in 17th century Dutch. E.g. ien hoop-e stof (a heap-e dust; ‘a lot of dust’); cf. Weijnen (1952).

Some Notes on Emphatic Forms and Displacement in Dutch

165

15. However, one might hypothesize that it is somehow related or close to this attributive-adjectival inflection –e, maybe in a way the regular indefinite article and the spurious indefinite article are related to each other. And on analogy with the notion of ‘spurious indefinite article’ one might want to talk about a ‘spurious inflection’, i.e. an inflection which does not seem to ‘belong’ to the elements in its near surroundings. Just like the spurious indefinite article does not seem to belong to the following noun (nor to the preceding noun, for that matter), the ‘emphatic’ inflection –e does not seem to belong to the following noun either (cf. (23c), where –e precedes a singular neuter noun). Nor does it seem to ‘belong to’ the preceding noun: the regular adjectival inflection –e, for example, normally does not combine with nouns in Dutch. I will briefly return to the relationship between the adjectival –e and the emphatic –e in section 7. 16. Alternatively, one might want to argue that the grammatical marker –e does not carry the emphatic meaning as a lexical property, but that the property of emphasis (or maybe Focus) is associated with the functional head position to which it is moved. Arguably, the spurious indefinite article een in wat-exclamative noun phrases (cf. (10)) also receives its exclamative meaning by occupying the operator head position D (i.e. the D-position that is specified for the exclamative force feature [+EXCL]). 17. The phenomenon that the emphatic marker –e appears on a degree adverb that modifies a predicative adjective should not be confused with the –e that proleptically appears on degree adverbs modifying attributive adjectival phrases (cf. also Corver 1997). For example: (i) een [erg-e dur-e] fiets a very-e expensive-e bike ‘a very expensive bike’ For one thing: the adverb erg in (i) cannot carry a –e in predicative adjectival environments: (ii) Deze fiets is [erg(*-e) duur] (Compare: verdomd(-e) duur) This bike is very(-e) expensive damned(-e) expensive For discussion of the phenomenon of proleptic agreement, see Corver (forthcoming). 18. As illustrated in (i), the modifying degree element never displays agreement with the subject (and the modified adjective). It is only the gradable adjective that agrees in gender and number with the subject of the clause (cf. (29b)): (i) *Fata e [enorma de fericita] Girl-det is enormous-F.Sg of happy-F.Sg ‘The girl is enormously happy’ 19. Elements such as bliksems (lightning-s), duivels (devil-s), hels (hell-s) also occur as attributive adjectives in nominal contexts: (i) a. die helse pijnen those hell-s-e pains

166

Norbert Corver b. die duivel-s-e kerel that devil-s-e man c. die bliksem-s-e jongens those lightning-s-e boys Traditionally, the element –s is interpreted here as a derivational suffix that turns a noun into an adjective. The –e (schwa) that follows the sequence N+s is the inflection that shows up on attributive adjectives in Dutch. In the spirit of the DP-internal predicate displacement analysis as defended here, I would like to propose that in those constructions –s is a nominal copula as well. Drawing a parallel with the Dutch N van N-construction (cf. (iii-a)), this would lead us to a derived structure like (iii-b): (iii) a. [DP die [FP duivelj [F’ F (= van)+Xi (= ’n) [XP jongen [X’ ti tj]]]]] that devil of a boy b. [DP die [FP duiveli [F’ F (= -s) +Xj (= -e) [XP jongen [X’ tj ti ]]]]]

As indicated in structure (iii-b), I tentatively assume here that the Dutch inflectional element –e, which appears on attributively used adjectives, is the surface reflex of the raised small clause head X. In its base position, X stands in a Spechead relation with the subject of the small clause (i.e. the [-neuter], [+singular] nominal phrase jongen in (iii-b)). Let’s assume that the grammatical features on the subject noun phrase value the small clause head, and that X surfaces as the attributive inflectional morpheme –e. 20. Interestingly, Den Dikken (1998) takes the ‘possessive marker’ –s (like in Jans auto; Jan-s car) to be an instance of the nominal copula. Presumably, the –s in the ‘Npossessor –s Npossessee’ construction and the –s in the ‘degree word –s A’ are tokens that are of the same type, viz. the nominal copula. 21. See also Frisian: bliksem-s-e goed (lightning-s-e well; ‘very well’) and duvel-s-e kald (devil-s-e cold; ‘very cold’) 22. The pronominal forms watte, datte and ditte are typically found in informal (Standard Dutch) speech; see Haeseryn et al. (1997: 309, 323). These forms are also quite common in Dutch child language. 23. From the semantic point of view, one may find it rather hard to believe that pronouns can be analyzed as predicates. Pronouns have generally been analyzed as non-lexical categories of the type DP (i.e. being referential) and they do not seem to be unsaturated. As shown by the Dutch examples in (i), though, there are structural contexts in which pronominal elements can clearly function as nominal predicates. This suggests that pronouns are not necessarily incompatible with a predicative status. (i) a. A: Wat is dat? B: Dat is een auto (Dutch) What is that That is a car b. Jan is trots op zijn moeder en Peter is dat ook John is proud of his mother and Peter is that too

Some Notes on Emphatic Forms and Displacement in Dutch

167

c.

Ik wil dit later ook worden I want this later also become ‘I also want to become (like) this’ The conclusion that pronominals can sometimes be predicative has also been drawn in a recent paper by Adger and Ramchand (2003) on predication structures in Scottish Gaelic. They follow Zamparelli (2000) in decomposing the DP into different layers of functional projection, namely (a) SDP (i.e. Strong Determiner Phrase), which is the only truly referential part of the nominal projection, (b) PDP (i.e. Predicative Determiner Phrase), which is the site of numerals and certain types of adjectives, and (c) KDP (i.e. Kind Determiner Phrase), which is the phrase that designates an atomic property, or a kind. Adger and Ramchand argue that pronominal elements in Scottisch Gaelic may be bare KDPs in positions where they are not arguments (i.e. where they function as predicates designating properties). Along these lines, one might argue that the emphatic pronominals are not DPs (i.e. SDs), but that they rather instantiate a ‘lower’ phrase in the extended nominal projection. 24. In Frisian, next to ikke, ditte, datte and watte, one finds the adverbs hjirre (hjir; ‘here’), dzre (dzr; ‘there’), wzre (wzr; ‘where’). I will assume that in line with the analysis in (55), these emphatic forms receive the following derived structure: (i) [CP hjir/dz r/wz ri [C’ –ej [XP pro [X’ tj ti]]]] The element pro stands for the presupposed information that there is some location X where, say, some event takes place. The emphatic adverb identifies this location and provides new, focused material. 25. The analysis in (59), in which –e is assumed to apply to the coordinate structure, appears to make the incorrect prediction that a string like (i) is also permitted. The derived structure would be the one in (ii). (i) Wij, ik en mijn vrouw(*-e), gaan vandaag naar de stad We, I and my wife-e, go today to the city (ii) wij, [CP [ik en mijn vrouw]i [C’ –ej [XP pro [X’ tj ti]]]], … The impossibility to attach –e to the noun vrouw may be related to certain categorical restrictions on the phrasal host to which –e morpho-phonologically attaches; i.e. it needs a pronominal element as its host. 26. In Zwart (2000), it is noted that Sassen’s generalization does not extend to all dialects of Dutch. There are Dutch dialects which seem to permit the co-occurrence of ikke and a finite verb. He gives the following example from SouthernEast-Flemish: (i) £ke zal dat doen I-e will that do ‘I will do that’ There are two possible lines one could explore here: First of all, one might want to analyze this pattern as a hidden left dislocation construction, ‘hidden’ in the

168

27. 28.

29.

30.

Norbert Corver sense that a demonstrative element (in Spec,CP) is missing. Under such a left dislocation analysis, ikke would not be the true subject-argument of the verb. Secondly, one might try to set up a cleft-like analysis, roughly along the following lines: ‘it is ik (who) will do that’. The copular part ‘it is ik’ would underlyingly be represented as: [pro –e ik]. Ik will then be fronted to the left edge of the small clause, yielding [CP ikj[C’ ei [XP pro ti tj]]] (+ (who) will do that). Needless to say, that for a complete answer these constructions need further investigation. Possibly, pauvre moves on to the left edge (i.e. A-bar position) of the small clausal configuration. Not unexpectedly, we also find cher toi (dear you) in French. This phrase receives the following structure under the approach adopted here: (i) [CP cherj [C’ C [XP [DP toi] [X’ X [AP tj]]]]] Also in Frisian we find patterns in which an attributive adjective has the ending –en attached to it. As noted in Hoekema (1996: 12), those adjectives always have an emphatic meaning. Arguably this –en is a weakened form of the indefinite article in. (i) Ha se ús gjin ôfgryslik-en fertriet besoarge? Has she us no terrible-en pain delivered ‘Hasn’t she caused great trouble?’ See als Van der Meer (1987) and Verdenius (1939) for discussion. In Lilley (2001), interesting examples from Northern Swedish dialects are given of what he calls dummy indefinite articles (i.e. spurious indefinite articles in our terminology). (i) a. Vi fått [fint e var] We have got fine a weather ‘We got fine weather.’ b. Dänna var he [stor a husa] Over-there were there big a houses ‘There were big houses over there.’ c. [en stor en ful en kar] a big an ugly a man ‘a big ugly man’ In (ia) and (ib), the spurious indefinite article is followed by an uncountable noun and a plural noun, respectively. The example in (ic) shows that this spurious indefinite article also shows up in recursive environments. It is interesting to see that the spurious indefinite article functions as a mediating head in structural contexts where a noun is modified by an attributive adjective. If spurious articles typically show up in contexts of predicate displacement (see e.g. the wat voorconstruction, the wat-exclamative construction, the N van N-construction), then these facts from Northern Swedish may be suggestive for the application of predicate displacement in contexts of DP-internal modification by (prenominal) attributive adjectives.

Some Notes on Emphatic Forms and Displacement in Dutch

169

References Adger, D. and G. Ramchand 2003 Predication and equation, Linguistic Inquiry 34: 325–359. Bennis, H., Corver, N and den Dikken, M. 1998 Predication in nominal phrases. The Journal of Comparative Germanic Linguistics 1: 85–117. Bowers, J. 1993 The syntax of predication. Linguistic Inquiry 24: 591–656. Chierchia, G. 1984 Topics in the syntax and semantics of infinitives and gerunds. Doctoral Dissertation, University of Massachusetts, Amherst. Chomsky, N. 1993 A Minimalist program for linguistics theory. In K. Hale and S.-J. Keyser (eds.), The view from building 20: Essays in linguistics in honor of Sylvain Bromberger: 1–52. MIT Press, Cambridge, Mass. 2001 Derivation by phase. In M. Kenstowicz, ed., Ken Hale: a life in language, MIT Press, Cambridge. Corver, N. 1997 The internal syntax of the Dutch extended adjectival projection. Natural Language and Linguistic Theory 15: 289–368. 1998 Predicate movement in pseudopartitive constructions. In A. Alexiadou and C. Wilder (eds.), Possessors, predicates and movement in the Determiner Phrase: 215–257. John Benjamins, Amsterdam/Philadelphia. 2000 Degree adverbs as displaced predicates. Italian Journal of Linguistics 12.1.: 155–191. 2001 On predicate numerals. In T. van der Wouden and Hans Broekhuis, eds., Linguistics in the Netherlands 2001: 65–76. John Benjamins, Amsterdam/Philadelphia. 2001 Predicate displacement within the adjectival system. Evidence from degree modfiication in Rumanian In: R. Bok-Bennema, B. de Jonge, B. Kampers-Manhe and A. Molendijk (eds.), Adverbial modification, Rodopi, Amsterdam. 2003 On three types of movement within the Dutch nominal domain. In. M. Coene and Y. d’Hulst (eds.), From NP to DP. Volume I: The syntax and semantics of noun phrases: 297–328. John Benjamins, Amsterdam/Philadelphia. forthc. Proleptic agreement as a good design property. In J. Costa et al. (eds.), Proceedings of the Lisboa Workshop on Agreement. Dikken, M. den 1995 Copulas. Paper presented at GLOW 18, Tromsø, Unpublished manuscript, Vrije Universiteit Amsterdam/HIL. 1998 Predicate Inversion in DP. In A. Alexiadou and C. Wilder, (eds.), Possessors, predicates and movement in the Determiner Phrase: 177–214. John Benjamins, Amsterdam/Philadelphia.

170

Norbert Corver

Grosu, A. 1974

On the nature of the Left Branch Condition, Linguistic Inquiry 5: 308–319. Haeseryn, W., K. Romijn, G. Geerts, J. de Rooij en M.C. van den Toorn 1997 Algemene Nederlandse Spraakkunst, Band 1, Tweede, geheel herziene druk, Martinus Nijhoff Uitgevers, Groningen. Heycock, C. 1991 Layers of predication: The non-lexical syntax of clauses. Doctoral Dissertation, Upenn. (published 1994, New York: Garland.) Hoekema, T. 1996 Beknopte Friese vormleer. Afûk, Leeuwarden/Ljouwert. Hoeksema, J. 2000 Ikke as default-nominatief, TABU 30: 27–46. Hoekstra, J. 1997 Absolutely intensifying adjectives and the emphatic ending –e in North Frisian (Fering-Öömrang), West-Frisian and other West-Germanic languages. Us Wurk 46: 90–117. Hornstein, N. and D. Lightfoot 1987 Predication and PRO. Language 63: 23–52. Jespersen, O. 1977 (first published in 1924) The philosophy of grammar. George Allen & Unwin, London. Laan, K. ter 1953 Proeve van een Groninger spraakkunst. Van der Veen, Winschoten. Kayne, R. 1994 The antisymmetry of syntax. Cambridge, Mass., MIT Press. Lilley, J. 2001 The syntax of Germanic post-adjectival articles. Talk at the 16th Workshop on Comparative Germanic Syntax (CGSW 16), McGill University, Montréal, Canada. Mallinson 1986 Rumanian Croom Helm, London. Meer, G. van der 1987 It is in dregen baas (de ‘b¥gings’ –(e)n bij eigenskipswurden), Us Wurk 36: 97–111. Moro, A. 1988 Per una teoria unificata delle frasi copulari. Rivista di Grammatica Generativa 13: 81–110. 1991 The raising of predicates: Copula, expletives and existence. In L. Cheng and H. Demirdache, eds., MIT Working Papers in Linguistics 15: More papers on Wh-movement, 193–218. Department of Linguistics and Philosophy, MIT.

Some Notes on Emphatic Forms and Displacement in Dutch

171

Overdiep, G. S. 1936 Zinsvormen en woordvormen, Onze Taaltuin IV: 362–375. 1937 Stilistische grammatica van het moderne Nederlandsch, W.E.J. Tjeenk Willink, Zwolle. 1940 De volkstaal van Katwijk aan Zee, Standaard-Boekhandel, Antwerpen. Royen, G. 1953 Buigingsverschijnselen in het Nederlands. Amsterdam. Sassen, A. 1953 Het Drents van Ruinen. Van Gorcum/Hak & Prakke, Assen. Stowell, T. 1981 Origins of phrase structure. Doctoral Dissertation, MIT, Cambridge, Mass. Verdenius, A. 1939a Over de vormen van het adnominale adjectief en het lidwoord van bepaaldheid in de 17de eeuwse Amsetrdamse volkstaal. Nieuwe Taalgids 33: 97–108. 1939b Adverbia van graad op –e. Nieuwe Taalgids 33: 361–368. Weel, M. van 1904 Het dialect van West-Voorne. E.J. Brill, Leiden. Weijnen, A. 1952 Zeventiende-eeuwse Taal. Thieme & Cie, Zutphen. Zwart, J.-W. 2000 Ikke en de default-naamval: een reactie op Hoeksema, TABU 30: 175–182.

Scrambling, Optionality and Non-Lexical Triggers Roland Hinterhölzl

1. Introduction This paper provides a comprehensive discussion of both the empirical nature and the theoretical implications of scrambling. I argue that the phenomenon of scrambling can only be done justice to if it is recognized that scrambling is subject to both PF- and LF-interface conditions. In particular, I address the claim by Haider&Rosengren (1998), henceforth H&R, that trigger accounts are essentially inappropriate for the phenomenon of scrambling on account of its optional nature. Contrary to H&R, I argue that a trigger account is indeed feasible in a copy theory of movement in which both LF- and PFconditions determine which copy is to be spelled out. Furthermore, I argue that one type of trigger for scrambling involves scopal features of arguments. Scopal features are relational syntactic properties of scope taking elements and as such they are inherently non-lexical. I propose an extension of the minimalist framework that allows for the introduction of non-lexical features in the course of the derivation to account for this aspect of scrambling. The paper is organized in the following way. Section 2 provides a discussion of the properties of scrambling in German. On the basis of these data, I argue in favour of an A-movement analysis of the phenomenon, discarding arguments having been put forth in favour of an A’-movement analysis. The section concludes with a discussion of how the pertinent A/A’distinction can be drawn independently of the notion of L-relatedness. In Section 3, I discuss the trigger problem relating to scrambling. In Section 4, I present a solution to the trigger problem. Outlining the distributional properties of definite and indefinite, specific and non-specific, focused and non-focused DPs with respect to the negative marker in German, I argue that scrambling is not optional and show how apparently irregular word order patterns can be explained by the interaction of output conditions that determine the Spell-out of movement copies. In Section 5, I address the issue of how scopal features representing relational properties of constituents can be handled in a simple trigger account.

174

Roland Hinterhölzl

I propose an extension of the standard minimalist framework and show that the computational complexity of the generalized transformation proposed is sufficiently reduced by the standard economy conditions as to render it an attractive alternative to assuming optionality in syntactic computation. Section 6 briefly summarizes the chapter.

2. The Properties of Scrambling in German Since it became clear that scrambling cannot be considered a stylistic operation, as was assumed by Ross (1967) who coined the term scrambling, a vivid debate arose about the issue of how the syntactic operation of scrambling can be fitted into the A/A’-dichotomy of movement operations. Since scrambling does not fit either movement type completely, the question of what type of movement operation scrambling is and whether scrambling should be decribed as the result of movement at all (rather than the result of base-generation), was far from being subject to general consensus for a long time. I cannot go into the details of this historic debate and refer the reader to two volumes that deal extensively with the question sketched above (cf. Grewendorf &Sternefeld 1990, Corver & van Riemsdijk 1994), instead I will provide a discussion of the descriptive properties of scrambling in German and explain why I consider the debate as having been decided in favor of A-movement. Scrambling has been used as a cover term for different operations that affect word order in the German middlefield. It is important to distinguish two such operations in order to detect the constitutive properties of the operation of word order change that we are interested in here. Depending on whether the moved item bears stress or not, the movement operation displays quite significantly different properties. The operation of scrambling in which the moved item is stressed shows clear properties of A’-movement: it can affect arguments as well as predicates and is not clause-bound. This operation moves contrastive topics and comes with a special intonation, the so-called hat contour comprising a fall-rise tone on the moved topic and a fall tone on the (constrastive) focus-element in the remainder of the clause. This operation has been called focus scrambling by Neeleman (1994), I-topicalization by Jacobs (1997) and T-scrambling by Haider & Rosengren (1998). Because the moved item bears stress, we shall use the term S-scrambling. This movement operation is illustrated in (1) and (2), in which letters of stressed syllables are given in small capitals. In Dutch, contrary to German, an object cannot scramble over the subject.

Scrambling, Optionality and Non-Lexical Triggers

175

However, if the operation is accompanied with the special intonation of a contrastive topic, scrambling of the object across the subject is fine (cf. (1bc)). (2) is an example of long distance S-scrambling in German: the embedded object has been moved into the middlefield of the matrix clause. (1)

a. dat Jan de boeken niet koopt that Jan the books not buys b. *dat de boeken Jan niet koopt that the books Jan not buys c. dat ZULke boeken selfs JAN niet koopt that such books even Jan not buys

(2)

Noch gestern haben DIEse Frau alle geglaubt, dass Still yesterday have this women-AKK all believed that NIEmand einladen wird nobody invite will ‘Of this woman everyone believed still yesterday that no one would invite her’

Once the distinction is made between S-scrambling and scrambling operations in which the moved element bears no special stress, it can be shown that scrambling proper is restricted to arguments (3) and strictly clausebound as well (4). In (3), the infinitival predicate has been scrambled across the adverb leading to ungrammaticality. In (4), long distance scrambling of the embedded object results in ungrammaticality. (3)

a. weil jeder oft gewinnen möchte since everyone often win wants b. *weil jeder gewinnen oft möchte since everyone win often wants since everyone often wants to win’

(4)

a. Gestern hat niemand geglaubt, dass er die Maria einladen wird yesterday has nobody believed that he the Maria invite will b. *Gestern hat niemand die Maria geglaubt, dass er einladen wird yesterday has nobody the Maria believed that he invite will ‘yesterday nobody believed that he will invite Maria’

176

Roland Hinterhölzl

Scrambling can create new binding possibilities. In (5a), the quantifier cannot A-bind the pronoun simply for lack of c-command. However, if the direct object quantifier is scrambled across the subject containing the pronoun, the latter can be bound with no WCO-effect being noticible. (5)

jedeni liebt a. * weil seinei Mutter since his mother-NOM everyone-AKK loves seinei Mutter liebt b. weil jedeni since everyone-AKK his mother-NOM loves

Scrambling CAN lead to scope ambiguities. Stress in the former statement is on the modal can, since there seem to be two groups of speakers. For speakers of the first group, scope is solely a matter of surface relations. For these speakers, including myself, the scrambled structure in (6b) is as unambiguous as the base structure in (6a), though the scope relations have been inverted by scrambling. For speakeres of the second group, the base structure is unambiguous as well, but the scrambled structure is ambiguous. In (6b), these speakers not only get the reading that results from surface scope (as speakers of the first group do), but also get the reading that results from reconstructing the scrambled object into its base-position. (6)

a. weil [mindestens eine Frau] [fast jeden Mann] liebt since at least one woman-NOM almost every man-AKK loves b. weil [fast jeden Mann] [mindestens eine Frau] t liebt one woman-NOM loves since almost every man-AKK at least

It is important to note that also for the second group binding relations are strictly read off from surface relations. In German, scrambling may not only create new binding possibilities, it may also destroy binding possibilities, as the example adopted from Haider & Rosengren (1998) shows. Even for speakers of the second group, the unbound anaphor in (7b) cannot be saved by reconstructing it into its base-position. (7)

a. dass der Mann die Bilder einander anglich that he the pictures-AKK each other-DAT made-alike b. *dass der Mann einander die Bilder t anglich that he each other-DAT the pictures-AKK made-alike c. einander hat der Mann die Bilder angeglichen the pictures-AKK made-alike each other-DAT has he

Scrambling, Optionality and Non-Lexical Triggers

177

To sum up, the fact that scrambling is a) restricted to arguments, b) clausebound and c) can create new binding possibilities, speaks in favor of an analysis in terms of A-movement. Moreover, the fact that scrambling cannot be reconstructed for reasons of binding speaks against an analysis in terms of A’-movement, since as is shown in (7c), an anaphor that has been A’moved into [Spec,CP] can be bound via reconstruction. Thus, there is little reason to assume that scrambling in German should not be A-movement.

2.1. Discarding Arguments for A’-movement One argument that played an important role in the debate on the nature of scrambling was the observation that scrambling can license parasitic gaps. If parasitic gaps can only be licensed by A’-movement, as seems to be the case in English, then scrambling ought to be A’-movement, so the argument went. Neeleman (1994) provides an excellent discussion of this argument and convincingly argues that the evidence that is given by examples of the type of (8a) has been overrated. It is sufficient here to present his strongest argument, namely the observation that the so-called parasitic gap in (8a) can also be licensed by A-movement. In (8b), the presumed parasitic gap is licensed by a passive subject. This observation also holds for German (cf.(8c) and (8d)). Given Neeleman’s observation, the argument coming from the licensing of (so-called) parasitic gaps is mute. (8)

a. dat Jan het boek [zonder pg in te kijken] aanprijst that Jan this book [without at to look] offers ‘that Jan offers this book without looking at it’ b. dat [de boeken] door Jan [zonder in te kijken] angeprijst worden that the books by Jan without at to look offered were ‘that the books were offered by Jan without looking at them’ c. dass Hans die Buecher ohne zu lesen weitergegeben hat that Hans the books without to read on-passed has d. dass die Buecher ohne zu lesen weitergegeben wurden that the books without to read on-passed were

In an influential paper, Grewendorf & Sabel (1999) investigate scrambling in German and Japanese and argue that clause-internal scrambling in German, contrary to clause internal scrambling in Japanese, must be A’-movement.

178

Roland Hinterhölzl

Their claim is decisively based on two arguments/observations. First, scrambling of a potential binder in German, contrary to scrambling in Japanese, cannot license an anaphor contained in a DP. Secondly, they argue that the lack of WCO-effects with scrambling in German should not be taken as evidence for an A-movement analysis, since clear instances of A’-movement in German can be found which do not give rise to a WCO-effect either. I will take up this issue in the following subsection. To illustrate their first point, let us look at the data in (9). In (9a), the anaphor contained in the direct object is licensed by the c-commanding subject. In (9b), the anaphor contained in the subject cannot be licensed by the direct object, a potential antecedent, for lack of c-command. If scrambling were an instance of A-movement in German, so they argue, then the direct object in (9c) should be able to license the anaphor, contrary to fact. Since in the parallel case in Japanese the anaphor is licensed (9df), clause internal scrambling in German must be an instance of A’-scrambling. (9)

a.

einen Freund von sich i eingeladen hat weil Peter i friend of himself-ACC invited has since Peter-NOM a

b. *weil ein Freund von sich den Peter eingeladen hat has since a friend of himself the Peter-ACC invited c. ??weil den Peter ein Freund von sich eingeladen hat since the Peter-ACC a friend of himself-NOM einvited has d. ?*[IP otaga i i-no sensei-ga [VP karera i–o hihansita ]] each other-GEN teacher-NOM them-ACC criticized

(koto)

sensei-ga [VP hihansita ]]] f. ?[IP karera i–o [IP otaga i i-no they-ACC each other-GEN teacher-NOM criticized

(koto)

Note, however, that the postnominal anaphor in German behaves like a long distance anaphor in being strongly subject-oriented, as is shown in (10). In (10), only the subject qualifies as an antecedent for the anaphor within the complex noun phrase. Thus, I would like to contend that (9c) is ungrammatical not because scrambling is not an instance of A-movement in German but because an object does not qualify as an antecedent for the subject-oriented anaphor in German. den Hansj zu einem Freund von sich i/??j (10) weil der Peter i since the Peter-Nom den Hans-ACC to a friend of himself geschickt hat sent has

Scrambling, Optionality and Non-Lexical Triggers

179

Since the binding of anaphors is the only empirical argument they advance in favour of an A’-movement analysis and since clause-internal scrambling in German in their account would stand out compared to clause internal scrambling in Hindi (which has been shown by Mahajan (1991) to be a clear instance of A-movement) and Japanese (which they themselves argue to be an instance of A-movement as well), I feel justified to conclude that the observation in (10) weakens their argument to a degree that it seems illadvised to base such a strong claim on the sole data of (9).

2.2. Vikner’s Argument Vikner (1994) puts forward an important comparative argument in favor of an A’-movement analysis of scrambling in German. Vikner points out that scrambling in Dutch and object shift in Scandinavian, contrary to scrambling in German, may not permute arguments. For instance, in Icelandic, a definite direct object cannot move across the indirect object, as is shown in (11). (11) a. Petur keypti bokina ekki Peter bought book-the not b. *Petur syndi bokina oft Mariu Peter showed book-the often Mary-DAT c. Peter zeigte das Buch oft der Maria Peter showed the book often the Maria-DAT If object-shift is analyzed as A-movement into AgrO, then the contrast in (11) can be explained as a violation of (relatized) minimality (cf. Rizzi 1991): the intervening indirect argument in an A-position would block object-shift of the direct object in (11b). If scrambling in German is A-movement as well, so Vikner argues, then it is surprising that no minimality effect can be observed in this language (11c). Vikner concludes that free permutation of arguments in German calls for an account in terms of A’-movement: not only would this account explain the lack of minimality effects, adjunction would also be the most elegant way to derive the great number of possible word-orders in German. (12) shows for the case of three arguments that all possible permutations, namely six, are grammatical. The permutations of the base order in (12a) are most easily derivable by (multiple) adjunction to VP and IP.

180

Roland Hinterhölzl

(12) a. weil Peter der Maria das Buch zeigte since Peter the Maria-DAT the book-AKK showed b. weil Peter das Buch der Maria zeigte c. weil das Buch Peter der Maria zeigte d. weil der Maria Peter das Buch zeigte e. weil der Maria das Buch Peter zeigte f. weil das Buch der Maria Peter zeigte ‘since Peter showed the book to Mary’ Vikner then goes on to challenge the arguments that favor an analysis of scrambling in terms of A-movement. He notes that almost all the arguments are based on the lack of weak cross-over effects in German scrambling. Given that wh-movement does trigger weak cross-over violations (13a), it is argued that scrambling and wh-movement cannot be of the same kind. However, so Vikner argues, it is not possible to have this kind of weak cross-over violation at all in German and concludes the crucial difference would not seem to be between scrambling and wh-movement but between German and English. As is shown in (13b) not even (local) wh-movement does trigger a weak cross-over violation in German. (13) a. *Whoi does hisi mother love? liebt seinei Mutter nicht? b. Weni Who-AKK loves his mother-NOM not ‘who is such that his own mother does not love him’ c. Weni liebt [ti [ seine Mutter nicht ti]] The observation is correct, but the above conclusion is invalid. It is true that no weak cross-over effects are observable in a single clause in German. However, when it comes to long distance movement, we can detect an interesting contrast (cf. Frey 1990). Long distance wh-movement of the embedded object does give rise to a WCO-effect if the pronoun is contained in the matrix subject, but does not if the pronoun is contained in the embedded subject, as is illustrated in (14). If A’-movement in German, as Vikner argues, were exceptionally not subject to WCO, then the contrast in (14) remains unexplainable. glaubt Peter dass seinei Mutter t nicht liebt (14) a. Weni mother not loves Who-AKK believes Peter that his ‘who does Peter believe is such that his own mother does not love him’

Scrambling, Optionality and Non-Lexical Triggers

b.

181

*Weni glaubt seinei Mutter daß der Peter t nicht liebt Who-AKK believes his mother that the Peter not loves ‘who does his own mother believe is such that Peter does not love him’

However, if we assume that the WCO-effect can be circumvented by prior scrambling of the wh-word (cf. 13c), then the contrast in (14) follows simply from the fact that scrambling is clause-bound. Given this assumption, scrambling can provide an A-binder for the pronoun in the embedded subject but not for the pronoun in the matrix subject. Consequently, wh-movement in (14b) only leads to an A’-bound pronoun in the matrix clause and therefore does give rise to a WCO-violation in as much as wh-movement does in English. On the other hand, the (well-motivated) assumption that German scrambling is A-movement gives us for free the property that scrambling, contrary to S-scrambling, is clause-bound. Were scrambling indeed an operation of A-bar movement then it would be hard to explain why long distance scrambling, for instance in (4b) above, cannot make use of [Spec,CP] of the embedded clause as a licit escape hatch. If scrambling is treated as A-movement, then movement via the embedded [Spec,CP] falls prey to the uniformity condition on chains, an option that we may assume is open for the A-bar movement operation of S-scrambling. At this point, a question concerning the A/A’-dichotomy arises, namely, how the distinction between A- and A’-positions is to be made. It seems illadviced to me to base the distinction on the L-or non-L-relatedness of the respective head. Due to V2, Spec,CP should count as L-related in German, nevertheless movement to this position clearly and unambiguously has the properties of A’-movement. Furthermore, there are languages in which movement into a designated focus position, an operation that is standardly analysed as A’-movement, is accompanied by verb movement. Thus, I would like to propose to make the nature of movement dependent on the type of feature that is checked in the target head, irrespective of L-relatedness. If the feature checked is an operator feature, [wh], [foc] or [neg] to name a few, then the movement operation will have the properties of A’-movement. If a non-operator feature is checked, for instance, [Specificity] or [Topicality] of an argument, then the movement operation will have properties of A-movement. That is to say, that A-movement will be extended to check other features than just Case. S-scrambling involves movement of a contrastive topic. A contrastive topic presupposes a set of alternatives and by introducing a variable that ranges over the members of the presupposed set, will involve, despite of its topicality, an operator feature. Therefore, S-scrambling counts

182

Roland Hinterhölzl

as A’-movement. In a parallel fashion, the features checked by scrambling (proper) will thus have to be of the non-operator type. To conclude, Vikner’s refutation of the A-movement approach to German scrambling fails.The comparative issue that he raises is important but will not be addressed in this chapter (cf. Hinterhölzl (2000) for a solution in terms of minimality).

2.3. Weak Pronouns and Scrambled DPs Having argued that scrambling is A-movement, we have to address the question of what the landing positions and the triggers of this type of Amovement are. In this paper, I argue that scrambling in German does not have a unified trigger but is initiated by two types of triggers, namely Specificity in the sense of Enc (1991)1 and Scope. In the following, I want to propose that scrambling of specific DPs, is movement into the Specifier-position of heads licensing clitics. In German object clitics can be licensed in two different positions, one below the subject and one above it, as is shown in (15). Subject clitics are licensed in a position above the position for object clitics, as is shown in (16ab). We thus arrive at the representation in (17).2 (15) a. weil der Hans ihn gestern t getroffen hat since the Hans him yesterday met has b. weil ihn der Hans gestern t getroffen hat since him the Hans yesterday met has (16) a. weil er ihn gestern getroffen hat since he him yesterday met has b. * weil ihn er gestern getroffen hat since him he yesterday met has (17) [C [Cl-S [Cl-O1[ Su [Cl-O2 [ … ]]]]] Given the structure in (17), it can be shown that all the orders in (12) above can be derived without adjunction by assuming that scrambled DPs move into the Specifiers of heads licensing clitics. (12a) corresponds to the base order. In (12b), the direct object has moved into the Specifier of the lower clitic-object position, whereas the order in (12c) results from movement of

Scrambling, Optionality and Non-Lexical Triggers

183

the direct object into the higher clitic-object position. (12d) involves movement of the indirect object into the higher clitic-object position. Finally, in (12ef) both clitic positions above the subject are filled with the two objects in alternating orders.3 Summing up, the scrambling orders in (12) can be derived without adjunction, simply by using positions which are needed independently for the licensing of weak pronouns. One argument that could be launched against this account of scrambling is the observation that weak pronouns are subject to conditions that are not observed by scrambled DPs and which are specified in (18). (18) a. weak pronouns appear in the order NOM >ACC > DAT b. weak subject pronouns precede scrambled DPs Weak pronouns reach their licensing positions by either XP-movement only or by XP-movement (if necessary) and subsequent head movement. In the latter case, they form a cluster and I would like to propose that condition (18a) pertains to clitic clusters, which are licensed in the highest clitic head position. Since scrambled DPs do not form clusters, they are not subject to condition (18a). Furthermore, if weak pronouns ‘cliticize’ by XP-movement only, they can appear in various orders and different positions, very much like scrambled full DPs. This later option is available in Westflemish (cf. Haegeman 1994) and various German dialects (cf. Gärtner & Steinbach 2000). To account for condition (18b), I will assume that weak subject pronouns always have to move to the highest clitic position and that a clitic-licensing head cannot simultaneously license an element in the head position and another element in its Specifier. The proposal that one type of scrambling is movement into the Specifiers of heads licensing clitics is supported by cross-linguistic considerations. Alexiadou & Anagnostopoulou (1997) argue that there are some striking resemblances between scrambling in Germanic and clitic doubling constructions in Greek and Romance. It is interesting to note that in those languages (or dialects) that allow for clitic doubling as in River Plate Spanish and Romanian a typical WCO-violation can be circumvented by the introduction of a clitic (data taken from Suñer (1988) and Dobrovie-Sorin (1990), respectively), as is indicated in (19) and (20). In River Plate Spanish (19), the WCO-effect induced by QR at LF is obviated if the quantifier phrase is doubled. Likewise, a WCO-effect induced by the overt operation of Wh-movement in Romanian (20) is obviated by clitic-doubling of the wh-phrase.

184

Roland Hinterhölzl

(19) a. * Sui madre quire a todosi their mother likes a everyone b. Sui madre los quire a todosi their mother them likes a everyone no aguanta ni su i madre (20) a. * [A cuales de ellos] i a which ones of them not stand even their mother b. [A cuales de ellos] i no los aguanta ni su i madre a which ones of them not them stand even their mother We have seen above that overt scrambling in German can obviate WCOeffects induced by QR and proposed that the lack of WCO effects in local wh-movement can be explained by assuming that the wh-object first scrambles across the subject containing the pronoun, providing an A-binder for the latter before wh-movement moves the wh-object into SpecCP. Assuming that scrambling is movement into the Specifiers of clitic positions, will allow us to provide a uniform explanation of the parallel effects of Scrambling in German and Clitic Doubling in Romance (and Greek) with respect to WCO, if we make the assumption, which seems natural, that clitic doubling involves movement of the doubled phrase into the Spec of a functional projection licensing clitics but spells-out the lower copy, which is then only prosodically marked as being part of the background of the sentence (while in German, this is indicated by the overt position of a discourse-anaphoric element via Spell-out of the higher copy). The data in (19) and (20) provide indirect evidence against Vikner’s and Grewendorf & Sabel’s (1999) position that WCO-effects are not a good criterion for deciding whether scrambling is A- or A’-movement, since they clearly show that a WCO-effect that is induced by an uncontroversial instance of A’-movement is circumvented by the introduction of clitics which are assumed to be base-generated in or related to an uncontroversial instance of an A-position.

3. The Trigger Problem H&R argue that scrambling may not be considered as being triggered by a feature that needs to be checked in a designated position. They point out that trigger accounts are often circular in the sense that they postulate features which apparently are only introduced into the theory to just trigger

Scrambling, Optionality and Non-Lexical Triggers

185

scrambling. Moreover, they argue that accounts which introduce substantive, independently motivated features prove to be either too weak, too strong or both. A trigger account is too weak if the trigger just involves DP-type features, like Case or a strong [D]-feature, since it does not cover scrambling of PPs and CPs in German. What is needed are substantive features that are independent of the syntactic category of an argument. A number of features that relate to the semantic or discourse properties of an argument have been proposed in the literature. Such accounts, so H&R argue, are too strong by necessity since they entail that if the respective feature is present scrambling is obligatory, and then they go on to show that scrambling is optional. Firstly, they argue that the semantic and/or pragmatic effects induced by scrambling cannot be taken to be the triggering factor of scrambling, since the interpretation effect that is induced by scrambling is found in unscrambled structures as well. Scrambling seems to reduce, but not to replace, the interpretation potential of a phrase. As evidence they provide examples in which a generic interpretation (21a), a specific interpretation of an indefinite (21b) and a specific definite interpretation (21c) are applicable to DPs in situ. For instance, in (21a), the definite DP object follows an indefinite pronoun subject. So the generically interpretable DP is likely to be in situ. Analogous considerations apply to (21bc). In (21b), the indefinite is interpreted as specific, given that its reference is picked up by Maria, though it seems to occur in its base-position following an indefinite subject. The same holds for the definite DP her dress in (21c), which – though occurring in its base position – can have a specific interpretation referring to Maria’s dress. (21) a. dass wer die Pockenviren ausrotten sollte that someone the pockvirus exterminate should b. wenn wer eine rothaarige Frau sucht dann ist das Maria if someone a red-haired woman seeks then is it Maria c. dass er wem ihr Kleid gezeigt hat, hat Maria nicht gefallen that he someone her dress shown has, has Maria not pleased Note that indefinites may not occur in the domain of negation in German (22ab). However, H&R argue that this case of alleged obligatory scrambling rests on a controversial premise, namely that negation universally ccommands the whole VP. They argue that the relevant condition for German and Dutch is that negation only needs to c-command the finite verb in its

186

Roland Hinterhölzl

base position. This condition is fullfilled if negation is adjoined to V below the base position of arguments rendering scrambling of indefinite w-pronouns superfluous in order to derive the grammatical order in (22b). (22) a. *dass jemand nicht wen jagte that someone not somebody chased b. dass jemand wen nicht jagte that someone somebody not chased Note that if this idiosyncratic assumption about the base-position of negation in German is not taken, the data in (21) are not less problematic. Even if we assume that indefinite w-pronouns can scramble in principle, it seems hard to come up with a motivation that has the indefinite argument move across the generically interpreted definite argument. Other triggers that have been proposed involve semantic-driven movement – like movement of weak DPs into strong-DP positions as in De Hoop (1992) – or pragmatic features like familiarity as in the account of Delfitto & Corver (1997). But again, so H&R argue, these accounts prove to be too strong. To illustrate their argument against semantic driven scrambling, consider (23). If indeed scrambling was triggered by a specific interface feature, (23ab) should differ to the extent defined by the absence or presence of that feature. However, (23a) can have the same interpretation as (23b), namely that Max in general admires primaballerinas. Therefore, so H&R argue, the claim that a strong “generic” feature triggers scrambling cannot be correct. (23) a. dass ja Max Primaballerinas bewundert that PRT Max primaballerinas admires b. dass ja Primaballerinas Max bewundert that PRT primaballerinas Max admires In this case, I find the evidence less convincing. In my judgment, I get an existential reading of the bare plural if the direct object in (23a) receives nuclear stress and a generic reading if the verb is assigned nuclear stress. This latter fact would indicate that the direct object in this reading of (23a) does not occupy its base-position. In this case, it also seems plausible that the definite subject DP Max has moved to a high position above the generically interpreted object. This analysis of (23a) presupposes that the modal particle ja, which is generally assumed to mark the VP-boundary (cf. Diesing 1992) can occupy a position high up in the clausal domain. Such an

Scrambling, Optionality and Non-Lexical Triggers

187

analysis, however, is not available to H&R since they exclude string vacuous scrambling in general. With the reservations indicated above, I agree with H&R’s observations. Also I think that these observations are rather problematic for simple minded trigger-accounts. To summarize the discussion above, we have seen that definite and indefinite specific DPs obligatorily scramble across negation while in other contexts, generics, definites and specific indefinites seem to be allowed to stay in their base position. So the picture we arrive at is rather puzzling. There is an instance of scrambling that is uncontroversally obligatory. This operation is scrambling for reasons of scope-taking. In German, a quantified object has to scramble across a quantified subject to take scope over it, as is shown in (24). (24) a. weil jeder mindestens zwei Bücher gelesen hat two books-ACC read has since everyone-NOM at least (SU > OB only) b. weil mindestens zwei Bücher jeder gelesen hat has since at least two books-ACC everyone-NOM read (OB > SU preferred) Note, however, that the property of obligatoriness does not make it easier to provide a satisfactory trigger account of at least this type of scrambling. The difficulty arises since it is quite inappropriate to assign the respective scopal feature to any specific F-head in the clause. Scopal features – if we introduce the features [w], [n], [i] (for wide, narrow and intermediate scope) for the sake of concreteness – by their very nature are not absolute properties. Scopal properties are relative properties: a DP has wide scope only in relation to another DP. Thus a checking account in terms of privative features seems inadequate for scopal phenomena in principle. In the account of H&R, the problem does not arise, since scrambling is not considered as movement into a designated position to check the relevant feature. In their account scrambling is the result of the syntactic mechanism of chain formation, applying blindly in the identification domain of the selecting head, which can be exploited at the interface. In this approach, the presumed scopal features [w],[i] and [n] could be treated as interpretable semantic features of scopal elements which do not require syntactic checking and which have to obey the following interface condition at LF which will filter out all derivations/representations that do not conform to it.

188

Roland Hinterhölzl

(25) Scopal Wellformedness: a) a phrase with the feature [w] must c-command a phrase with the feature [n] b) a phrase with the feature [i] must c-command a phrase with the feature [n] and must be c-commanded by a phrase with the feature [w] Note that this conception of grammar is very much in the spirit of minimalism. It is based on a maximally narrow syntax in which only morphological and a small selected set of formal features drive the derivation, with semantic and pragmatic properties being utilized in the workings of sophisticated interface conditions. In short, the syntactic derivation is not geared by features that relate to semantic or pragmatic properties of constituents. This analysis seems to be a relatively simple and maximally elegant solution to the problem of relational features but it comes with the prize of allowing for optional operations in the syntax.

4. Coping with Optionality Taking H&R’s observations at face value, we are confronted with two problems. A) There are data which require elements like wh-indefinites to scramble, though the accounts given sofar would have them as non-scramblable elements. B) There are data which suggest that elements which should scramble, like specific DPs, occupy their base position. We will tackle the first problem by proposing that next to [Specificity] there is another factor that triggers scrambling, namely Scope. H&R assume that wh-indefinites do not scramble. That this assumption is wrong and that wh-indefinites can indeed scramble in German is shown in (26). In (26a), the wh-indefinite is in its expected position following the indefinite (negative) subject. (26a) is unambiguous, meaning “that nobody met anyone does not surprise me.” In (26b), the wh-indefinite has scrambled across the subject yielding the interpretation “that there is someone that nobody met does not surprise me.”

Scrambling, Optionality and Non-Lexical Triggers

189

(26) a. dass keiner wen getroffen hat, überrascht mich nicht surprises me not that noone-NOM someone-ACC met has, b. dass wen keiner getroffen hat, überrascht mich has, surprises me that someone-ACC noone-NOM met To utter, (26b) the speaker does not need to have a specific individual in mind, it suffices that he has some evidence that there is a person with the relevant property. Thus, the wh-indefinite is only specific in the sense that it is the DP with widest scope, but it is not specific in the sense that it is presupposed, known to the speaker or in any other way anaphorically anchored in the context. This shows that the scopal properties of arguments furnish as a trigger of scrambling and it confirms our former observation, that scrambling in this case is obligatory since the reading of (26b) is only available if the object is scrambled across the subject. Assuming that scope is a trigger for scrambling, of course, requires explaining how scopal requirements can be dealt with in a feature checking mechanism. This issue will be taken up in Section 5. In the following, I want to tackle the second problem, namely the issue that definite DPs and the like seem to scramble optionally. That is to address the observation that sometimes they appear to have moved and sometimes they appear to be in their base-position. My answer to the optionality issue is that DPs that have the triggering property always undergo scrambling but that there are intervening factors which prevent the Spell-out of the scrambled DP in its checking position. To show that the distribution of scramblable arguments is not optional but subject to specific conditions, we will investigate the distribution of arguments with respect to the negative marker. More specifically, we will address the assumption by H&R that the negative marker in German occupies a very low position in clause. This assumption allows them to refrain from assuming obligatory scrambling of specific DPs to account for the unmarked word order in (27). That the negative marker in German occupies a much higher position in the clause is shown in the following section. (27) weil der Hans die Maria nicht liebt since the Hans the Maria not loves

190

Roland Hinterhölzl

4.1. The Syntax of the Negative Marker in German The negative marker nicht (not) obligatorily precedes manner adverbs as is shown in (28). Assuming as is standardly done that manner adverbs are adjoined to the VP, I conclude that the negative marker in German cannot occur VP-internally – adjoined to the verb as is assumed by H&R – but occupies a functional head position in the I-domain of the clause, as is the standard assumption for the syntactic representation of negation in various languages (cf. Haegeman & Zanuttini 1994). (28) a. weil der Hans das Buch nicht sorgfältig gelesen hat since the Hans the book not carefully read has “since Hans has not read the book carefully” b. ??weil der Hans das Buch sorgfältig nicht gelesen hat since the Hans the book carefully not read has Definite nominal arguments generally precede the negative marker (29). Definites may only then follow the negative marker if they are contrastively focussed (cf. 29b vs. 29c). If an indefinite NP precedes the negative marker it is interpreted as specific (30a). If an indefinite NP follows the negative marker, the reading one gets most easily is the one in which the negative marker is interpreted as negating only the nominal argument that follows it, which receives a (negative) contrastive interpretation. Depending on whether it is the determiner or the noun that receives the non-neutral (contrastive) stress, the negative marker negates the (cardinality of the) determiner or the descriptive content of the NP in (30b). (29) a. weil der Hans das Buch nicht gelesen hat since the Hans the book not read has “since Hans did not read the book” b.??weil der Hans nicht das Buch gelesen hat since the Hans not the book read has c. weil der Hans nicht das BUCH gelesen hat (sondern das Heft)

Scrambling, Optionality and Non-Lexical Triggers

191

(30) a. weil Hans ein Buch nicht gelesen hat (only specific interpretation) since Hans a book not read has “there is a book that Hans did not read” b. weil Hans nicht ein Buch gelesen hat (sondern zwei; sondern ein Journal) since Hans not a book read has (but two; but a journal) “Hans did not read ONE book, he read TWO books; Hans did not read a BOOK, he read a JOURNAL” It has been argued that in these cases, the negative marker acts as “constituent negation”. Because the negative marker can be topicalized together with an argument of the verb, it is assumed that negation in German can simply be adjoined to an XP which it narrowly or exclusively negates (31ab). (31) a. [?P Nicht der Hans] hat das Buch gelesen not the Hans has the book read “It was not Hans that read the book” b. [DP nicht [DP der Hans]] The nonspecific (existential) interpretation of an indefinite NP in a negated German sentence is expressed with the determiner kein (32a). In this case, the negative marker is non-overt or fused4 (32b). Thus, we cannot determine in (32) whether a nonspecific indefinite NP has to move across the negative marker. That a nonspecific indefinite NP must at least move up to the negative marker is indicated by the behavior of negative existentials in Upper Austrian. This dialect exhibits, like other Bavarian dialects, the phenomenon of negative concord which allows for the Spell-out of the negative marker even in the presence of negative constituents. As (33) shows, the negative existential NP kein Buch (no book) has to precede the negative marker net (not). In the following, I will assume that a negative existential NP with a kein-determiner occupies [Spec,NegP]. (32) a. weil Hans kein Buch gelesen hat since Hans “kein” book read has “since its not the case that Hans read a book” b. weil Hans (*nicht) kein Buch (*nicht) gelesen hat since Hans not “kein” book not read has

192

Roland Hinterhölzl

(33) a. woei da Hans ka Buach net glesn hot since the Hans “kein” book not read has “since its not the case that John read a book” b. *woei da Hans net ka Buach glesn hot since the Hans not “kein” book read has That nonspecific indefinite NPs may not only follow what may be analysed as constituent negation, but may also follow sentential negation can only be shown with more than one indefinite NP. If a sentence contains more than one indefinite NP, the highest non-specific argument is spelled out with the kein-determiner, that is, checks the negative marker in [Spec,NegP]. This is shown in (34). (34a) is a sentence that contains three indefinites. If all three DPs are interpreted nonspecifically, the negation of (34a) must be (34b). In (34c), where the indirect object is spelled out with a kein-determiner, the higher subject must be interpreted as specific, if we exclude focus-affected and quantificational readings. (34b) in conjunction with (33) and (34c) shows – if we analyze the keinphrase as occupying [Spec,NegP] – that nonspecific indefinites may follow negation. Given these conclusions, (34b) is analyzed as shown in (34d). (34) a. weil ein Mann einer Frau eine Blume schenkte since a man a woman-Dat a flower gave eine Blume schenkte b. weil kein Mann einer Frau since “kein” man a woman-Dat a flower gave “since it is not the case that some man gave some flower to some woman” c. weil ein Mann keiner Frau eine Blume schenkte since a man “keiner” Frau a flower gave “since a (certain) man did not give some flower to some woman” d. [CP weil [TP [NegP kein Mann [einer Frau eine Blume [VP schenkte]]]]] Also quantified NPs may follow the negative marker without giving rise to a “constituent negation”-interpretation (35ab). In (35ab), the negative marker can act as sentence negation. As (35c) shows, the negative marker can be construed narrowly with the higher subject NP, which is a typical property of sentence negation, but is unexpected of constituent negation. Hence, it follows that quantified NPs need not move across negation.

Scrambling, Optionality and Non-Lexical Triggers

193

(35) a. weil der Hans nicht viele Bücher liest since the Hans not many books reads “since it is not the case that Hans reads many books” b. weil der Hans nicht jede Frau anbetet since the Hans not every woman adores “since it is not the case that Hans adores every woman” c. Der HANS hat nicht viele Bücher gelesen, der PETER hat viele the Hans has not many books read, the Peter has many gelesen read “it was not Hans but Peter who read many books” Let us summarize what we observed so far. Specific NPs obligatorily move across the negative marker while nonspecific indefinite NPs, unless they move into [Spec,NegP] to check sentential negation, remain below the negative marker. Definite NPs may only then remain below the negative marker if they receive a contrastive interpretation. Along the same lines, nonspecific indefinite NPs may defy movement into [Spec,NegP] only if they receive a contrastive interpretation as illustrated in (30b) above. Finally, quantified NPs depending on their scope may stay below or move across the negative marker. A QP below the negative marker may have a specific or non-specific interpretation. (36a) is an example of a specific, that is, partitive QP that occurs below the negative marker that acts as sentence negation. But if a QP scrambles higher than the negative marker, then it can, like indefinites, only have a specific interpretation as is shown in (36b). (36) a. Der HANS hat nicht viele der Bücher gelesen, der PETER hat the Hans has not many of the books read, the Peter has viele davon gelesen many there of read b. weil der Hans viele Bücher nicht gelesen hat (only specific interpretation) since the Hans many books not read has “since for many of the books it holds that Hans did not read them”

194

Roland Hinterhölzl

Following the proposal in 2.3, I assume that movement across negation is triggerd by specificity. To get rid of constituent negation, I will take up a proposal by Richard Kayne (p.c.) and assume that there is a Focusphrase just below negation into which contrastively focussed elements move. Sofar I have said nothing about where and how arguments are Case-licensed in German. I will assume without discussion that they are licensed outside of the VP in Case-agreement projections as is indicated below. The resultant structure is given in (37) (recall that according to (17) scrambled (specific) DPs can be licensed below or above the subject). (37) [CP [IP Specifics (Su) T Specifics [ Neg [ Focus AgrC [ V ]]]]] Given (37), the regularities discussed above can be described in the following way. It seems that a strong Focus-feature blocks the movement of definite NPs into the licensing positions of specifics as well as the movement of the highest nonspecific indefinite into [Spec,NegP]. A specific QP may stay below negation if it is to be read with narrow scope, while a definite NP must (in the absence of any focus-feature) check its specificity feature. It is not evident how to properly express these regularities in a system of feature checking. One possibility is to assume that something like (38) holds. (38) Once Case is checked, only the feature of a DP with the closest licensing head is checked overtly (38) may be okay as a descriptive generalization but it is unsatisfactory as a statement of grammar. Why should it be that the possibility of checking a certain feature is dependent on the presence or absence of certain other features? So, (38) cannot be correct. However, (38) has the virtue of showing that the distribution of arguments and of definite DPs in particular is not optional at all, as claimed by H&R, but subject to specific restrictions. In the following section, I will provide an account in terms of conditions on the Spell out of copies that allows us to get rid of the generalization in (38) and to solve the problem of optionality.

4.2. Conditions on Spell-out The solution to the problem posed by (38), namely, the fact that certain features can only be checked in the absence of other features (conditional checking) is to assume that feature checking is unconditional but to refrain

Scrambling, Optionality and Non-Lexical Triggers

195

from positing that the checked category is unconditionally spelled out in the position of the highest feature checked. Let us assume as above that specificity is the relevant feature that triggers movement of arguments across negation. Then, we may assume a) that a specific DP (independently of other features) always moves to check its feature in a position above the negative marker and b) that its Spell-out is determined by the conditions in (39). (39) a. contrastive DPs are spelled-out in the focus position b. quantified DPs are spelled-out in their scope position c. de-accented DPs are spelled-out before accented DPs Condition (39a) captures the facts in (29) and (30) above, that is, the fact that a definite DP unless constrastively stressed appears before negation. Condition (39b) captures the facts in (35) and (36) above, namely the fact that specific QPs are spelled out above negation, unless they are to be interpreted with narrow scope with respect to negation. The notion “scope position” that is used in (39b) will be defined in the following section.While the conditions in (39ab) are related to the LF interface, condition (39c) is a condition related to the PF-interface. While the LF-related conditions are unviolable- we may assume that the heads licensing contrastive focus und scope have a phonological EPP-feature - the PF-related condition is soft. This is illustrated in (40). For the working of condition (39c) I assume that backgrounded and discourse-anaphoric DPs are deaccented. The question in (40) can either be answered with (40a) or with (40b). While (40a) is completely unmarked and the preferred option, (40b) is slightly marked but completely grammatical. (40) Q: Wem hat Otto das Buch gegeben? Who has Otto the book given? A: a. Otto hat das Buch dem PETER gegeben b. Otto hat dem PETER das Buch gebeben Otto has (the book) to Peter (the book) given The marked/unmarked status of the examples of (40) can be ascribed to the workings of an interface condition that determines the mapping between information-structure and prosodic structure. I assume that in both answers the direct object has scrambled (across the indirect object) with the difference following from Spelling-out either the higher or the lower copy.

196

Roland Hinterhölzl

(41) Interface Condition: the phonological phrase containing the focus (main accent) must be rightmost within its intonational phrase. (cf. Chierchia 1986; Hayes & Lahiri 1991; Frascarelli 2000) As is illustrated in (42), the different status of the answers follows from the prosodic condition in (41). (42) shows the prosodic structure of both answers, where round brackets indicate phonological phrases and iP indicates an intonational phrase. We see that (42a) optimally fullfills the prosodic conditon in (41), while (42b) violates this prosodic condition. I propose that this is the reason why (42a) is preferred over (42b). (42b) is repaired by being assigned a stronger pitch accent, while in (42a) the assignment of normal sentence accent suffices to mark the focussed constituent. Thus (42b) is prosodically more marked than (42a), but speakers are free to use the more marked forms for their communicative purposes, whatever they are. (42) a. [iP (Otto hat) (das Buch) (dem PEter gegeben) ] b. [iP (Otto hat) (dem PEter) (das Buch gegeben )] In sum, (39c) is a statistical consequence of the workings of the Interface Condition in (41). Note that we managed to restrict optionality to the workings of condition (41) only. By this manoeuvre, optionality is confined to a PF-interface condition that specifies prosodic requirements on the linearization of phonological material. The syntactic computation, however, including the branch leading to the LF-interface is deterministic throughout. Thus, contrary to optimality theoretic syntax, I assume that surface constraints have no place in the grammar itself. Alternatively, I assume that the grammar specifies a limited set of options (one being the spell-out of copies) that are fixed by the child which has access to (next to general syntactic principles) vobabulary and conditions operative at the interface levels only, which are necessarily ‘surfacy’. In short, ‘surface constraints’ are only relevant for the Spell-out of copies but not for the internal working of syntax that creates the copies.

Scrambling, Optionality and Non-Lexical Triggers

197

5. Feature Checking and Scope In Section 4, we have seen that scrambling for reasons of scope taking, though being non-optional, still poses a problem for trigger accounts because of the relational nature of scope. Also since I showed that scrambling proper is an A-movement operation, we need a flexible mechanism to replace adjunction. That is, we need to devise a mechanism which allows us to check one and the same feature in different positions in different occasions. In this section, I want to sketch a possible account of scope in terms of feature checking and explore its implications for the theory of grammar. For the sake of discussion, let us assume that the above introduced features [w], [i] and [n] are sufficient to account for scope phenomena and furthermore that these features drive the derivation. One question that arises is whether these features are formal or purely semantic in nature. For sure, these features are non-lexical. Take note of the fact that these features are not in any sense part of the lexical properties of a quantifier. As such these are essentially different from the scopal features that Stowell & Beghelli (1994) abstract from specific (types of) quantifiers to account for their scopal properties. The scopal features we are assuming are essentially independent of the properties of lexical items and are purely relational (or syntactic) in nature. Thus it seems appropriate to assume that they cannot be assigned to a specific lexical or functional head in the numeration. I will assume that non-lexical features can be assigned to any head in the course of the derivation. To restrict scrambling to the middle field, I assume that scopal features can be assigned to extended projections of the verb. The enrichment of an existing structure with a non-lexical feature is defined as given in (43). (43) a) assign the feature to an existing structure (the head at the root) in the course of the derivation b) assign the feature to (a copy of) a bare functional head and merge the head with the existing structure Economy of derivation guarantees that operation b) which is more complex than a) will only apply in case the derivation using only a) does not converge. Furthermore, fewest steps ensures that these features are not assigned repeatedly to a structure and shortest step requires that only the smallest extension that guarantees a convergent derivation will be taken. This is illustrated in (44).

198

Roland Hinterhölzl

(44) a. [CP C [ QP-Su[w] AgrNom[w] [QP-Ob[n] AgrAcc[n] [VP V]]] b. [CP C [ F[w] [ QP-Su[n] AgrN[n] [QP-Ob[w] [VP V]]] c. [CP C [F[w]…[ F[w] [ QP-Su[n] AgrN[n] [QP-Ob[w] AgrA [VP V]]]]]] d. [CP C [ T …[ F[w] [ QP-Su AgrN[n] [QP-Ob AgrA [VP V]]]]] e. [CP C [F[w]…[ T [ QP-Su AgrN[n] [QP-Ob AgrA [VP V]]]]]] In (44 a), the scopal features are assigned by operation a) to the respective Case-agreement heads. Assuming that the subject is to be read with wide scope with respect to the object, no extension of the derivation, i.e., no scrambling, is necessary. The scopal features can be checked directly in the Case-positions (I assume that DPs come equipped with the respective scopal features matching the propositional intentions of the speaker). In (44be), we assume that the object is to be read with wide scope with respect to the subject. The structures in (44b-e) show possible extensions of the derivation before the direct object moves to check its scope feature. In this case, the derivation (44b) wins out over the derivation in (44c) due to fewest steps. In the same vein, the derivation in (44d) will be selected over the derivation in (44e), since the movement of the object in (44d) to check its scope feature will be shorter than its movement in (44e) (shortest step). In short, the assignment of scopal features has to obey the regular economy conditions. Furthermore, it has to meet the interface condition in (45). (45) Scopal Filter a) a head assigned the feature [w] must c-command a head assigned the feature [n] b) a head assigned the feature [i] must c-comamnd a head assigned the feature [n] and be c-commanded by a head assigned the feature [w] For the sake of better illustration, let us discuss the derivation of the sentences in (46). First, the arguments are merged in their thematic position in the VP. Then the Case-checking heads are merged and the arguments move into their Case-licensing positions in a parallel fashion. (46) a. weil jeder mindestens eine Frau liebt since everyone-NOM at least one woman-ACC loves b. weil mindestens eine Frau jeder liebt since at least one woman-ACC everyone-NOM loves

Scrambling, Optionality and Non-Lexical Triggers

199

In (46a), where the Subject is to be read with wide scope with respect to the object, the scopal features can be directly assigned to and checked in the Case-positions, with AgrAcc being assigned and checking [n] and AgrNom being assigned and checking [w]. In (46b), where the object is to be read with wide scope with respect to the subject, only one scopal feature can be assigned and checked in the Case position, since the assignment of [w] to AgrAcc and [n] to AgrNom would violate the scopal filter in (45). Thus, the assignment of [n] to AgrNom, which according to (43) is more economic than not using any Case-position for the checking of scopal features, forces the assignment of the remaining scopal feature [w] to a higher head. Since the direct object in (46) is non-specific – if it were specific the scopal features could be assigned to the head licensing weak pronouns (cf. Section 2.2) – and since the object does not have any other features to check in the I-Domain, the scopal feature is assigned to a bare functional head which is merged with the existing structure. Then, the direct object moves across the subject to check its scopal feature. Finally, the complementizer is merged to complete the derivation of the clause. The assignment of a scopal feature to a functional head defines the scope position of an argument. Since according to (39b) quantified DPs have to be spelled-out in their Scope positions, the copies of the two arguments can only be spelled-out in the order given in (46), with the wide scope phrase preceding the narrow scope phrase. Along the same lines it follows that a quantified DP if it is to be read with narrow scope with respect to negation (cf. (35) and (36) above) must be spelled out below negation even if it has a specific (partitive) interpretation. A DP of this type will check its scopal feature in its Case-position below negation, move across negation to check its [Specificity] feature but be spelled out in its scope position below negation. The same considerations apply to contrastively focussed definite and indefinite specifics (cf. (29) and (30) above). The derivation of (29c) is given in (47). (47) a. weil Hans nicht das BUCH gelesen hat since Hans not the BOOK read has b. [weil Hans [ Spec das Buch [ nicht [Foc das Buch [Case das Buch gelesen hat]]]]] The direct object in (47b) moves to check first its Case, its focus and then its [Specificity] feature and is spelled out in its Focus-position below negation. Let us now address the question of whether these scopal features are to be considered as formal or as semantic features. Remember that there are

200

Roland Hinterhölzl

two different dialects concerning the interpretation of quantified expressions. For speakers of the first dialect scrambled structures are unambiguous. That is to say, the scrambled phrase is interpreted in its surface position. For speakers of the second dialect scrambled structures are ambiguous. That is to say, the scrambled phrase can be interpreted in its surface position or in its base position (I am leaving open the question whether it is the Theta- or the Case-position that is relevant here). Note that the existence of ambiguity in the second dialect is problematic for the assumption that scopal features are semantic features. A phrase that is assigned the feature [w] for wide scope can be interpreted with narrow scope in dialect 2. The scopal features that we assumed here do not determine the semantic interpretation of scopal elements. Rather, they provide a syntactic limit to the interpretation of a scopal element. This property is more akin to the nature of a formal feature (that gears but does not (directly) determine the interpretation of the element it is assigned to). Given the notion of scope, the interpretation of these elements in the two dialects can be characterized rather simply as given in (48). (48) Dialect 1: A scopal element is interpreted in its scope position Dialect 2: A scopal element can be interpreted in its scope or in its base position Again, this characterization of the properties of scrambling in the two dialects highlights the fact that the scopal features that I introduced are rather formal in nature than being purely semantic. Rather than being interpreted directly, they drive the derivation that provides the input for more general principles of interpretation. To summarize, if we want to refrain from positing an optional mechanism in the syntax like H&R’s mechanism of freely creating scrambling chains that can be exploited at the interfaces, we have to extend the computational system and allow for the introduction of non-lexical features in the course of the derivation. The enrichment mechanism I propose obeys cyclicity and is conservative in that it involves either the assignment of non-lexical (=relational) features to pre-existing structure or the introduction of a bare functional head with the feature in question which represents the smallest extension.

Scrambling, Optionality and Non-Lexical Triggers

201

6. Conclusions I have argued that one type of scrambling, namely the one that affects an unstressed or destressed constituent is an instance of A-movement. Furthermore, I showed that there are two types of triggers that drive this movement, namely the discourse related feature [Specificity] and a relational scope feature. I have sketched an account of this type of scrambling – abstaining from the use of adjunction – that has these features checked in the Specifiers of the respective functional projections. Finally I addressed the issue of optionality connected with scrambling. I have argued that optionality is not a property of the syntactic computation itself, but can be relegated to various options in the Spell-out component that are determined by interface conditions at LF and PF.

Notes 1. A DP is specific if it denotes a member of a set of individuals introduced in the previous discourse. It has been pointed out that names and generic expressions can scramble even in the absence of a discourse antecedent. Thus, the feature [familiarity] has been proposed which encompasses discourse-antecedency and membership in the common ground (cf. Delfitto & Corver 1997). I will leave the empirical question open, whether one type of trigger of scrambling is to be characterized with the notion [familiarity] rather than [specificity]. 2. That there are two licensing positions for object clitics (one below and one above the subject) is a relatively conservative assumption. Instead, one could assume that there is only one position for licensing object clitics, which is above the subject and that the subject itself has moved into a higher position in (15a). Since this alternative proposal is neutral with respect to the main argument defended in this section, I will not pursue this issue any further. 3. Whether all scrambling orders, including those with several adverbs present, can be derived in this manner, is subject to empirical investigation. More specifically, it remains to be seen whether this clitic-licensing heads occupy fixed positions in the tree or whether they can be introduced at various points in the course of the derivation. For how this latter idea can be implemented – albeit for the purposes of checking scopal properties – see section 5. 4. The determiner kein has been analyzed as created by fusing a determiner with existential force with negation (cf. Kratzer 1989)

202

Roland Hinterhölzl

References Alexiadou & Anagnostopoulou 1997 Towards a uniform account of scrambling and clitic doubling. In W. Abraham & E. van Gelderen (eds.) German: Syntactic ProblemsProblematic Syntax: 142–161. Max Niemeyer Verlag. Chierchia, Gennaro 1986 Length, Syllabification and the Phonological Cycle in Italian. Journal of Italian Linguistics 8: 5–34. Corver & Riemsdijk 1994 Studies on Scrambling. Berlin/New York: Mouton de Gruyter. Delfitto & Corver 1997 Feature Primitives and the Syntax of Specificity. Ms. Tilburg University. Diesing, Molly 1992 Indefinites. Linguistic Inquiry Monographs. MIT Press. Cambridge, Massachusetts. Dobrovie-Sorin 1990 Clitic Doubling, Wh-movement and Quantification in Romanian. Linguistic Inquiry 21: 351–397 Enc, Mürvet 1991 The Semantics of Specificity. Linguistic Inquiry 22: 1–25. Frascarelli, Mara 2000 The Syntax-Phonology Interface in Focus and Topic Constructions in Italian, Dordrecht: Kluwer Academic Publishers Frey, Werner 1993 Syntaktische Bedingungen für die Interpretation. Berlin: Studia Grammatika 35. Gärtner & Steinbach 2000 What do reduced pronominals reveal about the Syntax of Dutch and German. Linguistics and Philosophy 9: 7–62. Grewendorf & Sabel 1999 Scrambling in German and Japanese. Natural Language and Linguistic Theory 17: 1–65. Dordrecht: Academic Publishers. Grewendorf & Sternefeld 1990 Scrambling and Barriers. Amsterdam: John Benjamins. Haegeman, Liliane 1994 The Typology of Syntactic Positions: L-Relatedness and the A/A’Distinction. Groninger Arbeiten zur germanistischen Linguistik 37: 115–157. Haegeman & Zanuttini 1994 Negative Heads and the Neg Criterion. The Linguistic Review 8, 233–252.

Scrambling, Optionality and Non-Lexical Triggers

203

Hale & Keyser 1993 On Argument Structure and the Lexical Expression of Syntactic Relations. In The View from Building 20. Haider & Rosengren 1998 Scrambling. Sprache & Pragmatik 49. Germanistisches Institut der Universität Lund. Hayes & Lahiri 1991 Bengali Intonational Phonology. Natural Language and Linguistic Theory 9: 47–96. Hinterhölzl, Roland 2000 Scrambling. Ms. Humboldt-Universität zu Berlin. Hoop, Helen de 1992 Case Configuration and Noun Phrase Interpretation. Ph.D. diss., University of Groningen. Jacobs,Joachim 1997 I-Topikalisierung. Linguistische Berichte 168: 91–133. Mahajan, Anoop 1994 Toward a Unified Theory of Scrambling. In Corver and v. Riemsdijk (eds.): Studies on Scrambling: 301–330. Berlin: Mouton de Gruyter. Neeleman, Ad 1994 Complex Predicates. Ph.D. diss., Utrecht University Suñer, Margharita 1988 The role of AGR(eement) in Clitic Doubled Constructions. Natural Language and Linguistic Theory 6: 391–434. Vikner, Sten 1994 Scandinavian Object Shift and West Germanic Scrambling. In Corver & Riemsdijk (eds.): Studies on Scrambling. Berlin: Mouton de Gruyter.

Phonological Content and Syntactic Visibility Ruriko Kawashima and Hisatsugu Kitahara

A language L (more accurately, an I-language in the sense of Chomsky 1986) includes a cognitive system that stores information about sound, meaning, and structural organization; and performance systems (sensorimotor systems and systems of thought) access that information. Given this much, we are (in principle) ready to address the question of how well L feeds into the performance systems. For those who decide to deal with this question, the Minimalist Program (MP) provides a number of guidelines that (in part) determine the condition of minimalist adequacy for L. Such guidelines include what we might call the “minimum design” guideline (seeking to minimize the class of descriptive devices) and the “simple computation” guideline (seeking to reduce operative complexity for computation). In this paper, assigning the latter guideline slightly more prominence, we assume that the reduction of operative complexity is a crucial factor in determining the minimalist design of language. With this assumption, we first review some of the specific proposals advanced in MP (section 1), in particular, Chomsky’s (2001a) analysis appealing to the presence of phonological content (section 2). We then extend the current approach to Japanese scrambling, specifying the mechanisms for this type of movement (section 3) and elaborating the relation between phonological content and syntactic visibility (section 4). Finally, following Epstein and Seely (2002), we reformulate the relation between phonological content and syntactic visibility in strongly derivational terms (section 5). 1. Minimalist Assumptions In this section, with special attention to the reduction of operative complexity, we review some of the specific assumptions, adopted in MP. 1.1. Lexical Array and Derivation Assume that L is some form of recursive procedure which involves the universal set F of features (linguistic properties) and the computational system

206

Ruriko Kawashima and Hisatsugu Kitahara

CHL for human language (operating on F). Further assume that L makes a one-time selection of a subset [F] of F and a one-time formation of a lexicon LEX by assembling features of [F]. Postulation of [F] and LEX (dispensing with further access to F and [F], respectively) reduces operative complexity in some natural sense. Similarly, if CHL makes a one-time selection of a lexical array LA from LEX for each derivation, this step greatly simplifies computation. Chomsky illustrates this point with the following example: Suppose automobiles lacked fuel storage, so that each one had to carry along a petroleum-processing plant. That would add only bounded “complexity,” but would be considered rather poor design. Something similar might well be true for language … If the derivation accesses the lexicon at every point, it must carry along this huge beast, rather like cars that constantly have to replenish their fuel supply [note 29 omitted, RK/HK]. (Chomsky 2000: 99–101)

Given the lack of LA as a design flaw, we proceed, assuming LA to be part of minimalist design. With this assumption, let us ask how CHL generates a derivation, operating on lexical items available in LA. First consider the component of CHL which employs three operations: Merge, Agree, and Move. Merge takes two syntactic objects (_, `) and forms K(_, `) from them. Agree establishes a relation between a lexical item _ and a feature F. Move establishes a relation between _ and F and merges P(F) to _P (where P(F) is a phrase determined by F, and _P is a projection headed by _). Given these formulations of Merge, Agree, and Move, Chomsky articulates the complexity of Move as follows: Plainly Move is more complex than its subcomponents Merge and Agree, or even the combination of the two, since it involves the extra step of determining P(F) (generalized “pied-piping”). (Chomsky 2000: 101)

Preference of simpler operations over more complex ones capture the “last resort” character of Move and derives much of the empirical motivation for Procrastinate.1 Let us take concrete cases and see how the choice between Merge and Move is determined. Consider “standard” (1a–b) and “problematic” (2a–b): (1)

a. *there is likely [_ someone to be t in the room] b. there is likely [_ t to be someone in the room]

Phonological Content and Syntactic Visibility

(2)

207

a. I expected [_ someone to be t in the room] b. *I expected [_ t to be someone in the room]

Each derivation involves the stage (3) where T_ requires that something occupy Spec-T_: (3)

[T_ [be someone in the room]]

Suppose, for each pair, two options were available: merge there or move someone for (1a–b) and merge I or move someone for (2a–b). Then, given preference of Merge over Move, CHL would select Merge over Move in each pair, a wrong result. Chomsky (1995) takes the contrast in (2 a–b) to mean that an argument merged in non-e-position becomes an illegitimate object, violating the principle of Full Interpretation (FI) (thereby causing the derivation to crash). He suggests that at the stage (3), CHL “looks ahead” the crashing fate of the first choice Merge, thus allowing the second choice Move to take place. This analysis captures the contrast in (2a–b) but at the cost of increasing the degree of operative complexity. Chomsky (2000) suggests an alternative analysis that does not increase the degree of operative complexity. He takes the contrast in (2a–b) to mean that Merge of an argument in non-e-position is not available for CHL in the first place. Chomsky (2000: 103) proposes the following condition (where pure Merge is Merge that is not part of Move): (4)

Pure Merge in e-position is required of (and restricted to) arguments.

This e-theoretic condition is understood to be part of the definition of Merge. Consequently, Merge of an argument in Spec-T_ is not an option for CHL at the stage (3); (2b) is simply not derivable. Assuming that operative complexity matters for minimalist design, this alternative analysis is preferable to the preceding one with the “look ahead” procedure. Merge of an expletive in Spec-T_ is still permitted under (4), and the choice between Merge and Move in (1a–b) is locally determined by preference of Merge over Move. It is important to note that the choice depends on whether or not an expletive is available in LA. Thus, postulation of LA receives not only conceptual but also empirical support.

208

Ruriko Kawashima and Hisatsugu Kitahara

1.2. Lexical Subarray and Phase As demonstrated above, the initial choice of LA plays a crucial role, but this cannot be the whole story. Consider the structure (5): (5)

there is a possibility [_ that someone will be in the room]

This structure is derived from LA containing an expletive, but Move (of someone to Spec-T) is selected over Merge (of there in Spec-T) in the embedded phrase _. A question arises: why is preference of Merge over Move not applicable when constructing _? Chomsky (2000) presents a straightforward answer to this question, which further reduces operative complexity. He proposes that CHL extracts a subarray LAi from LA and places it in active memory. As LAi is exhausted, CHL may proceed if possible, or it may return to LA and extracts LAj, proceeding as before. Under this proposal, CHL accesses only part of LA at each stage of the derivation, thus contributing to the reduction of operative complexity. Now, returning to the derivation of (5), if the phrase _ is constructed from the subarray containing no expletive, then Move is the only option for CHL at the corresponding stage of the derivation. Thus, postulation of lexical subarray is motivated on both conceptual and empirical grounds. Suppose that postulation of lexical subarray meets the condition of minimalist adequacy. Then, the next task is to characterize the subarray LAi that can be selected for active memory. Chomsky (2000) assumes that LAi should determine a natural syntactic object SO, an object that is relatively independent in terms of interface properties. With this assumption, he suggests that SO is the closest syntactic counterpart to a proposition, which may undergo operations such as movement. Given these considerations, SO is assumed to be CP or vP (but not TP). LAi can then be selected straightforwardly. LAi contains exactly one C or v, determining clause or verb phrase. SO derived from LAi is called a phase of a derivation; and in a “phase by phase” fashion, CHL generates a derivation.2 1.3. Uninterpretable Features and Movement Assuming CHL to be a “phase by phase” recursive procedure, let us examine the specific mechanisms of CHL, in particular, those “triggering” movement. In the early stage of minimalist theorizing, there was a very important breakthrough, namely, the recognition of uninterpretable features as a serious

Phonological Content and Syntactic Visibility

209

problem, and why they exist became one of the core minimalist questions. To answer this question, MP has been pursuing an intuition that uninterpretable features are implemented for movement (where movement is, by hypothesis, required to satisfy FI), and this pursuit has substantiated the view that movement is “carried out” in the process of deleting those (otherwise useless) uninterpretable features. To be concrete, let us consider the structure (6): (6)

[_ an unpopular candidate] T-was elected t_

This structure involves three kinds of uninterpretable features: (i) the q-features of T that identify T as a target of movement, (ii) the EPP-feature of T that requires that something be merged with the projection headed by T, and (iii) the Case-feature of candidate that identifies _ as a candidate for such merger (hence movement). Given these uninterpretable features, Chomsky (2000) proposes a probe-goal system as an optimal device that yields the structure (6) by deleting such uninterpretable features. Let us then review the mechanisms of the probe-goal system. Suppose that CHL has constructed the structure (7), having merged T with the copular-headed phrase: (7)

T-be elected [_ an unpopular candidate]

In this structure, T has uninterpretable features of two types: the selectional feature EPP and the set of q-features. Like other selectional features, EPP seeks an XP to merge with the projection headed by T; and the q-set is taken to be a probe that seeks a goal, namely, matching features that establish agreement. For the q-set of T in (7), there is only one choice of matching features, namely, the q-set of candidate. Locating this goal, the probe erases under matching. Taking structural Case to be a reflex of an uninterpretable q-set, the Case-feature of the goal too erases under matching with the probe. The erasure of uninterpretable features of probe and goal is the operation called Agree. As for EPP of T, it is satisfied by pied-piping of a phrase P(G) (determined by the goal of T’s probe), which merges with the structure (7), becoming Spec-T. Thus, the combination of selection of P(G), merger of P(G), and feature-deletion under agreement is the content of Move, and this composite operation dislocates _, eliminating all uninterpretable features. MP further elaborates the relation between probe P and goal G. Since not every matching pair of P and G induces the operations Agree and Move, there must be some other pre-requisites for these operations. First note that G must

210

Ruriko Kawashima and Hisatsugu Kitahara

be in the complement domain of P. Also note that P and G must be active, meaning that each bears some uninterpretable features. Given this much, it follows that P(G) gets “frozen in place” when structural Case of G gets deleted, since caseless G is no longer active (bearing no uninterpretable feature).3 The notion matching too has to be clarified. Note that manifestation of structural Case of goal G depends on probe P: Nominative for T and Accusative for v. Given this fact, structural Case may be understood as a single undifferentiated feature, assigned a value under agreement. The same would be expected for the uninterpretable q-set of P. Its manifestation depends on the interpretable q-set of G, so that the q-set of P too is undifferentiated as to the value of the individual features. On these assumptions, matching should be identity of the choice of feature, not of value. For both P and G, the value of uninterpretable features is determined by Agree in the course of a derivation. Note that neither structural Case nor EPP alone induces any movement if matching is pre-requisite for the operations Agree and Move. As discussed above, matching is a necessary condition for agreement, but it cannot be a sufficient condition. Consider the structure (8) (where > is c-command, ` and a match the probe _, but ` is inactive): (8)

_>`>a

In this structure, _ cannot agree with a because _ matches with inactive ` which is closer to _ than matching a is, which means that the goal must be the first matching category of the probe, in order to yield agreement. To be concrete, let us consider the structure (9) (a case of superraising): (9)

[_ T]-seem that [` it] was told [a friends] CP

In this structure, the q-set of matrix T cannot agree with the q-set of friends because the q-set of matrix T matches with the q-set of inactive it which is closer to matrix T than matching friends is. Note that such defective intervention effect is induced, whether or not ` and a are identical in featurevalue. Also note that this locality requirement restricts the search domain of _, thereby contributing to the reduction of operative complexity. Returning to the structure (7), let us consider the case that (7) is constructed from LAi containing an expletive. In this case, the composite operation Move is pre-empted by Merge and Agree. Thus, CHL merges the expletive with the projection headed by T, yielding the structure (10): (10) there [T-be elected [_ an unpopular candidate]]

Phonological Content and Syntactic Visibility

211

In this structure, Merge of there and TP satisfies the EPP of T, and Agree of T and candidate deletes the q-set of T and the Case of candidate. Note that Agree can establish a Case-agreement relation between probe and goal, in principle, without any movement.4

1.4. Phase-Impenetrability Condition and Cyclic Transfer In MP, it has been assumed that uninterpretable features must be removed from the narrow syntax (to satisfy FI at LF) but left available for the PF component (to yield language-variant PF-manifestation). This assumption poses a problem for the current approach if Spell-Out applies at a single point of a derivation. Chomsky states the contradictory aspect of featuredeletion as follows: pre-Spell-Out, the probe must delete when checked yet remain until SpellOut [note 99 omitted, RK/HK] (Chomsky 2000: 131)

To resolve this problem, he suggests that deleted features are literally erased, but only after they are sent to the phonological component. In the subsequent work, however, Chomsky (2001a) elaborates this point, advancing the following proposal: the uninterpretable features, and only these, enter the derivation without values, and are distinguished from interpretable features by virtue of this property. (Chomsky 2001a: 5)

Note that the notions “interpretable and uninterpretable” were problematic since it was not clear how CHL detects the uninterpretable status of features (in particular, q-features) in the course of a derivation. Under this new proposal, they are replaced by the notions “valued and unvalued,” and CHL distinguishes the q-set of probe P from the q-set of goal G by lack of specification of value. This system, unlike the previous one, predicts that, after valuation (assigning values to the q-set of P), the q-set of P and the q-set of G are indistinguishable, which led Chomsky (2001a: 5) to propose cyclic Spell-Out. Specifically, he suggests that Spell-Out applies “shortly after” the uninterpretable features have been assigned values, where the notion “shortly after” is determined (in part) by the Phase-Impenetrability Condition (PIC).5 Let us first review the recent developments concerning

212

Ruriko Kawashima and Hisatsugu Kitahara

the notion cyclicity, and then examine how PIC determines the cyclic application of Spell-Out. Chomsky (2000: 107) presents the following cyclicity condition (as a condition on the head of a phase): (11) The head of a phase is “inert” after the phase is completed, triggering no further operations. This cyclicity condition prohibits a phase head from probing its complement domain in a later phase. Thus, the empirical content of the extension condition follows. But, like the extension condition, it allows a higher head to probe into the phase after it is completed. To restrict such probing from higher heads into the phase, Chomsky (2000: 108) proposes a much stronger cyclicity condition, namely, PIC (where the edge is understood as the residue of H-bar, either specifiers or elements adjoined to _): (12) In phase _ with head H, the domain of H is not accessible to operations outside _, only H and its edge are accessible to such operations. Upon the completion of _, PIC prohibits any head outside _ to probe into the complement domain of H. But notice, unlike the cyclicity condition (11), PIC allows H to probe into the complement domain of H even after _ is completed. Such countercyclic probing, however, is blocked by the following condition (Chomsky 2000: 132): (13) Properties of the probe/selector _ must be satisfied before new elements of the lexical subarray are accessed to drive further operations. Further notice, the effects of (13) arguably follow if simple computation matters for minimalist design. Chomsky (2000: 134) makes the following suggestion, for a probe F in _: (14) F has to be readily detectable, hence optimally in the label L(_) of _, its sole designated element. Given (14), it follows that CHL must satisfy the probe F while F is located in L(_). If F is further embedded and no longer located in the label of the structure, then CHL cannot detect F for the application of Agree, meaning that countercyclic probing in question is simply not an option for CHL.6

Phonological Content and Syntactic Visibility

213

With (14), PIC ensures that, once _ completes, the elements inside the complement domain of H become “inert,” undergoing no further syntactic operation. Such “syntactically inert” domain (in turn) becomes a target of cyclic Spell-Out, more accurately, cyclic Transfer in the sense of Chomsky 2001b. Let us review the operation Transfer. Chomsky (2001b: 4) assumes that a language L has the following three components: (15) a. narrow syntax (NS) maps LA to a derivation DNS b. the phonological component \ maps DNS to PHON c. the semantic component Y maps DNS to SEM Given (15), Chomsky (2001b: 5) formulates the operation Transfer as follows (see also Epstein et al. 1998; Epstein 1999; Uriagereka 1999): (16) Transfer hands DNS over to \ and to Y. Note that the mapping of DNS to \ is what has been called Spell-Out. Now suppose that NS, \, and Y are allowed to “forget” transferred materials in subsequent stages of a derivation. Then, operative complexity is greatly reduced. Given this possibility, we assume that Transfer applies to _ in full if _ is the root clause; otherwise, it applies to the “syntactically inert” domain of _, determined (in part) by PIC; and to ensure the reduction of operative complexity, we expect that Transfer applies to a “syntactically inert” domain as soon as it is determined. Thus, under PIC formulated as (12), Transfer applies to the complement domain of H at the completion of _. In the subsequent work, Chomsky (2001a) slightly modifies the formulation of PIC. He first presents the following general principle (where Ph1 is a strong phase, and Ph2 is the next highest strong phase): (17) Ph1 is interpreted/evaluated at Ph2. Assuming that PIC is subject to this general principle, Chomsky (2001a: 14) reformulates PIC as (18), for the structure (19) (where HP is a strong phase, and ZP is the smallest strong phase): (18) The domain of H is not accessible to operations at ZP; only H and its edge are accessible to such operations. (19) [ZP Z … [HP _ [H YP]]]

214

Ruriko Kawashima and Hisatsugu Kitahara

Under PIC formulated as (18), Transfer applies to YP of HP as soon as HP is embedded within ZP. Note that this revision slightly increases the size of active memory of CHL since CHL must “remember” the complement domain of the lower phase until the derivation reaches the next higher phase. Thus, on conceptual ground (concerning operative complexity), PIC formulated as (18) is less preferable. On empirical ground, however, Chomsky (2001a: 14) mentions Icelandic (20), given in glosses, as a potential support for (18): (20) Jon(DAT) like (pl) the-books(NOM,pl) In this structure, T agrees with quirky Nominative object occupying its base position. If the structural analysis of (20) is correct, then the complement domain of v cannot be “syntactically inert” at the completion of vP, thereby constituting an empirical argument for PIC formulated as (18). The choice between (12) and (18) does not affect any of our arguments to be presented below. For expository purposes, we proceed our discussion with PIC formulated as (18) (hereon, PIC).

2. Phonological Content and Intervention Effect In this section, adopting the general framework reviewed in section 1, we first examine the case that has led Chomsky (2001a) to revise the locality requirement on the probe-goal relation. We then present an analysis of West Ulster English as a potential support for this revision.

2.1. A Problem Given PIC and cyclic Transfer, Chomsky (2001a: 26) points out that there is no other way to derive the structure (21): (21) (guess) whatObj [JohnSubj T [vP t’Obj [tSubj read tObj]]] PIC requires that whatObj move to Spec-v for subsequent wh-movement; consequently, the q-set of T (seeking to match the q-set of JohnSubj) probes into vP, crossing over whatObj (occupying Spec-v). This probing of T to JohnSubj faces a problem if the first matching q-set of whatObj (occupying

Phonological Content and Syntactic Visibility

215

Spec-v) cancels the further search of T. Recall the structure (8), repeated in (22) (where > is c-command, ` and a match the probe _, but ` is inactive): (22) _ > ` > a In this structure, _ cannot agree with a because _ matches with inactive ` which is closer to _ than matching a is, which means that the goal must be the first matching category of the probe, in order to yield agreement. Given this locality requirement, the first matching whatObj would interfere with Agree of T and JohnSubj, and the structure (21) could not be derived. The current formulation of the locality requirement is thus too strong.

2.2. A Proposal To resolve this problem, Chomsky (2001a) suggests that the q-set of T can “bypass” the first matching whatObj (occupying Spec-v). He implements this “bypass” analysis as follows. He first assumes that the subsequent movement of whatObj from Spec-v to Spec-C, in effect, licenses the probe-goal relation between T and JohnSubj. With this assumption, he proposes the following condition, for the structure (22): (23) The first matching ` prevents Match of _ and a only if ` has phonological content. Chomsky further assumes that the probe-goal relation is evaluated for the condition (23) at the next strong phase level (subjecting to the general principle (17)). Under these assumptions, JohnSubj is allowed to move to Spec-T over whatObj (occupying Spec-v), and the probe-goal relation between T and JohnSubj is evaluated for the condition (23) after it is known whether the intervening position of whatObj, namely Spec-v, has become a trace (losing its phonological content). Returning to the derivation of (21), the intervening position of whatObj has no phonological content at the relevant phase level, namely, CP; hence, the first matching whatObj does not interfere with Agree of T and JohnSubj.7 This revision arguably finds some empirical support in West Ulster English, where wh-movement can leave items such as all in Spec-C but not in Spec-v (McCloskey 2000). Consider (24a–e):8

216

Ruriko Kawashima and Hisatsugu Kitahara

(24) a. [CP what all [TP he [T’ T [vP told him [CP (that) he [vP wanted t]]]]]] b. *[CP what [TP he [T’ T [vP all told him [CP (that) he [vP wanted t]]]]]] c. [CP what [TP he [T’ T [vP told him [CP all (that) he [vP wanted t]]]]]] d. *[CP what [TP he [T’ T [vP told him [CP (that) he [vP all wanted t]]]]]] e. [CP what [TP he [T’ T [vP told him [CP (that) he [vP wanted t all]]]]]] The fact that all cannot remain inside Spec-v (as in 24b and 24d) follows if what and all forms a single category DP (see among others, Sportiche 1988). Under the single constituent analysis, in each deviant case, the first matching DP occupying Spec-v is partially overt; it contains a category with phonological content (namely, all), thereby inducing an intervention effect, under the condition (23).

3. Specifying the Mechanisms for Scrambling In this section, adopting the probe-goal analysis with the revision made in section 2, we specify the mechanisms for Japanese scrambling and seek to identify the relevant parametric properties that permit this type of movement.

3.1. A Problem As is well known, Japanese allows Object (Obj) to undergo scrambling to sentence-initial position. Consider (25a–b) (see among others, Saito 1985, 1989, 1992): (25) a. [Taroo-ga sono hon-o katta] Taroo-NOM that book-ACC bought “Taro bought that book”

(koto) (fact)

b. [sono hon-o [Taroo-ga t katta]] that book-ACC Taroo-NOM bought

(koto) (fact)

Assuming scrambling to be an instance of movement, Saito (1992) argues that clause-internal scrambling can form an A-chain. Consider (26a–b):

Phonological Content and Syntactic Visibility

217

(26) a. ?*[[otagai-no sensei]-ga karera-o hihansita] (koto) e.o.-GEN teacher-NOM they-ACC criticized (fact) “Each other’s teachers criticized them” b. ?[karera-o [otagai-no sensei]-ga t hihansita]] (koto) they-ACC e.o.-GEN teacher-NOM criticized (fact) In the deviant (26a), an anaphor otagai “each other” inside Subject (Subj) is unbound; hence, Binding Condition A is violated. In the well-formed (26b), Obj is scrambled to sentence-initial position, where Obj binds the anaphor inside Subj. The observed contrast suggests that clause-internal scrambling can form an A-chain (whose head binds the anaphor inside Subj, serving as its antecedent). Given Saito’s movement approach to scrambling, we ask how A-type instances of Japanese scrambling (such as the one in (26b)) take place under current assumptions.9 To seek an answer to this question, we first extend Chomsky’s (2001a) analysis of (21) (discussed in section 2) to the derivation of (26b) involving A-type instance of scrambling (hereon, scrambling), and see exactly where and how such extension fails. Suppose that Obj moves to Spec-v for subsequent movement to sentence-initial position.10 Then, the q-set of T (seeking to match the q-set of Subj) probes into vP, crossing over Obj (occupying Spec-v). Here notice, prior to the completion of CP, Obj moves out of Spec-v; consequently, the intervening position of Obj, namely, Spec-v, has no phonological content when the probe-goal relation between T and Subj is evaluated for the condition (23). Thus, the first matching Obj does not interfere with Agree of T and Subj. Note that this conclusion follows only if Obj has undergone scrambling (moving out of Spec-v). The next task is to determine whether scrambling of Obj (from Spec-v to sentence-initial position) can take place under current assumptions. The relevant uninterpretable feature is EPP of T. Like other selectional features, EPP of T seeks XP to merge with TP. In addition, Chomsky (1995: 354) suggests the following parametric property of EPP: (27) EPP has an option of remaining active even after merger. If EPP of T exercises this option once, then CHL merges two XPs with TP, yielding multiple specifiers of T. Given that this option is available in languages like Japanese, we ask whether CHL can derive the structure (28): (28) [TP Obj-Acc [Subj-Nom [[vP t’Obj [tSubj [[VP tObj V] v]]] T]]]

218

Ruriko Kawashima and Hisatsugu Kitahara

Suppose that the EPP of T exercises the option (27) once. Then, we expect that CHL merges Subj and Obj with TP, yielding multiple specifiers of T (as in (28)). Under the probe-goal system, Agree of T and Subj identifies Subj as a candidate for such merger. But what about Obj? The probe-goal system provides no (obvious) means to select Obj for such merger. Notice that Match of T and Obj does not induce Agree of T and Obj because Obj (occupying Spec-v) has no unvalued feature (meaning that Obj is inactive). Further notice that EPP alone is not sufficient to induce movement (since EPP is not a matching feature). Thus, merger of Obj with TP – scrambling of Obj to sentence-initial position – cannot take place; CHL fails to derive the structure (28), contrary to fact.

3.2. A Proposal At this point, there are several ways to proceed. We would like to pursue the possibility that the probe-goal system provides a crucial distinction which gives CHL a means to select Obj for this type of merger/movement, namely scrambling. Let us first examine A-movement and scrambling under current assumptions. Within the probe-goal system, A-movement may be characterized as follows: (29) A-movement is an instance of Agree-driven movement which values q-features and satisfies EPP. One of the guiding intuitions behind (29) is that agreement induces movement. Now notice, under the probe-goal system, there is a pre-requisite for agreement, namely, matching. Given these two distinct but closely related conceptions, Agree and Match, scrambling may be characterized as follows: (30) Scrambling is an instance of Match-driven movement (supplementing Agree) which satisfies only EPP. Given (29) and (30), we suggest that both Match-driven and Agree-driven movement are permitted in languages like Japanese.11 Note, however, Match-driven movement (even if permitted) is restricted to a supplement to the operation Agree, because an “extra” matching relation is necessarily a byproduct of an agreement relation. In this sense, the proposal (30) is in accord with the guiding intuition: agreement induces movement. Returning

Phonological Content and Syntactic Visibility

219

to the structure (28), it is Agree of T and Subj that uses its supplementary relation, namely, Match of T and Obj, to select Obj for the satisfaction of the EPP of T. More specifically, in the single mapping, Agree of T and Subj selects Subj for first merger with TP, and Match of T and Obj selects Obj for second merger with TP. The precedence of Agree-driven movement over Match-driven movement arguably follows from the supplementary nature of the latter. Now notice, this proposal, assuming scrambling to be an instance of Match-driven movement, makes the following prediction. Consider the continuation from (31), but this time EPP of T does not exercise the option (27): (31) [TP [vP Obj-Acc [Subj-Nom [[VP tObj V] v]]] T] Recall that, to license the probe-goal relation between T and Subj, Obj must move out of the intervening position, namely, Spec-v. To implement such movement, Agree of T and Subj may use its supplementary relation, namely, Match of T and Obj, to select Obj for the satisfaction of EPP of T.12 Given these considerations, CHL constructs the structure (32) by merging Obj with TP: (32) [TP Obj-Acc [[vP t’Obj [Subj-Nom [[VP tObj V] v]]] T]] In this structure (unlike the structure (28)), Subj remains inside vP, and Agree of T and Subj occurs without movement of Subj. Suppose that the analysis illustrated above is (more or less) correct. Then, scrambling of Obj to sentence-initial position results two distinct structures (namely, (28) and (32)), depending on whether or not the option (27) is exercised. This prediction is indeed borne out. Miyagawa (2001) presents data such as (33a–b), where scrambling feeds into the scope interpretation: (33) a. [zen’in-ga sono tesuto-o uke-nakat-ta] (to omou) all-Nom that test-Acc take-Neg-Past (Comp think) “(I think that) all did not take the test” Subj(™) necessarily takes scope over Neg b. [sono tesuto-o [zen’in-ga t uke-nakat-ta]] (to omou) that test-Acc all-Nom take-Neg-Past (Comp think) “(I think that) all did not take the test” Either Subj(™) or Neg may take scope over the other

220

Ruriko Kawashima and Hisatsugu Kitahara

In (33a), Subj(™) moves to Spec-T (leaving Obj in situ), and it exhibits no scope ambiguity: Subj(™) necessarily takes scope over Neg. In (33b), Obj undergoes scrambling to sentence-initial position, and it exhibits scope ambiguity: either Subj(™) or Neg may take scope over the other. Under current assumptions, CHL assigns the following two structures to (33b), from which the observed scope ambiguity naturally follows: (34) a. [TP Obj-Acc [Subj(™)-Nom [[NegP [vP t’Obj [tSubj [[VP tObj V] v]]] Neg] T]]] b. [TP Obj-Acc [[NegP [vP t’Obj [Subj(™)-Nom [[VP tObj V] v]]] Neg] T]] In (34a), the EPP of T exercises the option (27) once; hence, CHL merges both Subj(™) and Obj with TP, placing Subj(™) above Neg; hence, Subj(™) takes scope over Neg. In (34b), the EPP of T does not exercise the option (27), hence, CHL merges just Obj with TP, leaving Subj(™) below Neg; hence, Neg takes scope over Subj(™) (yielding partial negation). To the extent that the proposed analysis of (33b) (with the structures (34a-b)) is tenable, the mechanisms of scrambling specified in this section receive additional support.

4. Phonological Content and Scrambling In this section, we examine Takahashi’s (2001) data concerning the (in)applicability of scrambling. We then identify a deeper condition governing the relation between phonological content and syntactic visibility.

4.1. A Problem Takahashi (2001) shows that scrambling cannot apply to phonologically null categories such as pro. Consider (35) and (36): (35) Q: John to Mary-wa doo sita no John & Mary-Top how did Q “What happened to John and Mary?” A: ?*[TP [otagai-no sensei]-ga karera-o hihansita] e.o.-Gen teacher-Nom they-Acc criticized “Each other’s teachers criticized them”

Phonological Content and Syntactic Visibility

221

A’: ?[TP karera-o [[otagai-no sensei]-ga t hihansita]] they-Acc e.o.-Gen teacher-Nom criticized (36) Q: John to Mary-wa doo sita no John & Mary-Top how did Q “What happened to John and Mary?” A: ?*[TP [otagai-no sensei]-ga pro hihansita] e.o.-Gen teacher-Nom them criticized “Each other’s teachers criticized them” The deviant (36A) suggests that pro (unlike its overt counterpart) cannot undergo scrambling to sentence-initial position. The following structure (where pro is scrambled to sentence-initial position) won’t be generable: [TP pro [Subj-Nom t V]]; otherwise, scrambling would circumvent a violation of Binding Condition A, as in (35A’). The inapplicability of scrambling follows if the absence of phonological content renders pro syntactically invisible, preventing pro from undergoing scrambling.

4.2. A Proposal Suppose that the absence of phonological content renders pro syntactically invisible. A deeper question arises: why does the presence (or absence) of phonological content matter for the (in)visibility of syntactic categories? As an answer to this question, we propose the following condition: (37) goal G is visible for probe P only if G bears LF-uninterpretable features (e.g. unvalued features, phonological features) This condition accounts for the absence of scrambling in (36A), as follows. Upon the valuation of Case, pro is no longer syntactically visible, bearing no LF-uninterpretable feature; hence, pro cannot undergo scrambling to sentence-initial position.13 The condition (37) further captures the effects of the condition (23), repeated in (38): (38) The first matching ` prevents Match of _ and a only if ` has phonological content.

222

Ruriko Kawashima and Hisatsugu Kitahara

Recall the structure (22), repeated in (39) (where > is c-command, ` and a match the probe _, but ` is inactive): (39) _ > ` > a Now notice, if inactive ` becomes a trace (losing its phonological content), then ` is no longer syntactically visible (bearing no LF-uninterpretable feature); hence, ` does not interfere with Agree of _ and a. In short, under the condition (37), if goal G is a phonologically null category with no unvalued feature, then G, being syntactically invisible, neither induces intervention effect nor undergoes Match-driven movement.

5. Subjecting Cyclic Transfer to a Minimalist Critique In this section, we examine Epstein and Seely’s (2002) argument against the current implementation of cyclic Transfer. Given their argument, we seek to reformulate the analysis of syntactic visibility in strongly derivational terms.

5.1. A Problem Epstein and Seely (2002) argue that the current implementation of cyclic Transfer has a serious problem. Referring to Chomsky’s (2001a: 5) statement “after application of Agree, the distinction is lost,” they point out that, if the distinction between interpretable and uninterpretable features is lost after valuation, then the distinction is necessarily lost “shortly after” valuation as well. In other words, the current implementation of cyclic Transfer (applying “shortly after” valuation) is problematic if Chomsky’s statement (“after application of Agree, the distinction is lost”) is correct. Note that, if CHL has the capacity to detect the distinction after valuation (contrary to the above statement), then the application of Transfer, in principle, need not be cyclic. In effect, Chomsky (2001a, b) assumes that CHL can detect uninterpretable features after valuation up to the next strong phase level, though how CHL carries out this task is not clear. Presumably, CHL requires some kind of memory buffer to “remember” the initial status of unvalued features up to the next strong phase level. But notice, the size of such memory buffer matters for minimalist design because the reduction of memory load simplifies computation; and what Epstein and Seely suggest is that, under a minimalist perspective, the ideal size of memory buffer should be minimum.

Phonological Content and Syntactic Visibility

223

5.2. A Proposal Departing from cyclic Transfer, Epstein and Seely (2002) explore the possibility that each syntactic operation transfers the relevant aspects of its output to the phonological component and to the semantic component, which we might call strongly cyclic Transfer (SCT). Note that SCT occurs in every transformational mapping, removing uninterpretable features in the process of valuation. This approach raises a number of questions.14 In what follows, we limit ourselves to the question of whether the analysis of syntactic visibility (developed in the preceding section) is reformulable in strongly derivational terms. Recall the condition (37), repeated in (40): (40) goal G is visible for probe P only if G bears LF-uninterpretable features (e.g. unvalued features, phonological features) Under SCT, the probe-goal relation is evaluated for the condition (40) at every derivational point; hence, CHL must know whether or not G is visible for P when the search of P takes place. To be concrete, let us consider the structure (41) (where Obj and Subj match the probe T, but Obj is inactive): (41) T [vP Obj [Subj v [VP V tObj]]] In this structure, the q-set of T (seeking to match the q-set of Subj) probes into vP, crossing Obj (occupying Spec-v). Under the condition (40), this probing of T to Subj requires that Obj be syntactically invisible; and under SCT, this requirement must be met when the probing of T to Subj takes place. But how can CHL determine the invisible status of Obj at this point of a derivation? To answer this question, we first appeal to the notion occurrence of lexical item. Chomsky (1995) proposes that an occurrence of lexical item _ is a sister of _, and that the two occurrences (H, L) of _ are formed if _ is merged first with L and later with H. Now suppose _ has phonological content PC. Then, unless more assumptions are added, PC of _ is associated with (H, L), but PC receives interpretation only at higher occurrence H. This fact alone motivates the dissociation of PC of _ from lower occurrence L, and we assume that such feature-dissociation takes place as CHL merges _ with H. The intuition behind this assumption is that such derivational feature-dissociation reduces operative complexity by minimizing the role of lower occurrence.15

224

Ruriko Kawashima and Hisatsugu Kitahara

Now, advancing this derivational analysis, we first make the following general assumption: (42) feature-dissociation may take place at any point of a derivation We then assume that feature-dissociation takes place when necessary (subjecting to the principle of Last Resort). With these assumptions, let us return to the structure (41). Recall that, to allow the probing of T to Subj, PC of Obj must be dissociated from its intervening occurrence at this point of a derivation. Here notice, under the assumption (42), during the search of T, such feature-dissociation may take place (otherwise, the derivation terminates here). Given this application of feature-dissociation, the intervening occurrence of Obj, being syntactically invisible, does not block the probing of T to Subj. But notice, this feature-dissociation analysis makes the following two predictions: (43) if CHL establishes no new occurrence of Obj after this application of feature-dissociation, then there won’t be any place for PC of Obj to receive interpretation (44) if CHL forms a new occurrence of Obj after this application of featuredissociation, then PC of Obj can receive interpretation at this “newly formed” occurrence of Obj Thus, it follows that, in the derivation of (41), Obj (bearing PC) must undergo further movement; otherwise, PC of Obj would receive no interpretation (inducing a violation of FI).16 In short, under current assumptions, CHL dissociates PC of Obj from the intervening occurrence, in order to allow the probing of T to Subj; and subsequent movement of Obj, in effect, allows the derivation to converge. Note that the current approach achieves this result with the minimum size of memory buffer, namely, the single transformational mapping.

5.3. An Analysis To attain a better understanding of the feature-dissociation analysis, let us examine how it captures the cases discussed in the preceding sections. Recall Chomsky’s (2001a) case (concerning the absence of intervention effect), given in (45):

Phonological Content and Syntactic Visibility

225

(45) (guess) whatObj [JohnSubj T [vP t’Obj [tSubj read tObj]]] This structure involves object-shift, subject-raising, and wh-movement.17 Each movement satisfies EPP and values uninterpretable features; and under SCT, such “newly valued” features are removed as they get valued. Now notice, after object-shift, the q-set of T (seeking to match the q-set of JohnSubj) probes into vP, crossing over whatObj (occupying Spec-v). Under the condition (40), this probing of T to JohnSubj is allowed, given that PC of whatObj was dissociated from the relevant intervening occurrence, and that Case was removed from whatObj. As the derivation continues, wh-movement forms a new occurrence of whatObj; and PC of whatObj (associated with this “newly formed” occurrence) receives interpretation there. Recall Saito’s (1992) case (yielding multiple specifiers of T), given in (46): (46) [TP Obj-Acc [Subj-Nom [[vP t’Obj [tSubj [[VP tObj V] v]]] T]]] This structure involves object-shift, subject-raising, and scrambling. Each movement satisfies EPP and values uninterpretable features; and under SCT, such “newly valued” features are removed as they get valued. Now notice, after object-shift, the q-set of T (seeking to match the q-set of Subj) probes into vP, crossing over Obj (occupying Spec-v). Under the condition (40), this probing of T to Subj is allowed, given that PC of Obj was dissociated from the relevant intervening occurrence, and that Case was removed from Obj. Subsequently, scrambling (induced by Agree of T and Subj) forms a new occurrence of Obj, and PC of Obj (associated with this “newly formed” occurrence) receives interpretation there. Recall Miyagawa’s (2001) case (yielding a single specifier of T), given in (47): (47) [TP Obj-Acc [[vP t’Obj [Subj-Nom [[VP tObj V] v]]] T]] This structure involves object-shift and scrambling. Each movement satisfies EPP and values uninterpretable features; and under SCT, such “newly valued” features are removed as they get valued. Now notice, after objectshift, the q-set of T (seeking to match the q-set of Subj) probes into vP, crossing over Obj (occupying Spec-v). Under the condition (40), this probing of T to Subj is allowed, given that PC of Obj was dissociated from the relevant intervening occurrence, and that Case was removed from Obj. Subsequently, scrambling (induced by Agree of T and Subj) forms a new

226

Ruriko Kawashima and Hisatsugu Kitahara

occurrence of Obj, and PC of Obj (associated with this “newly formed” occurrence) receives interpretation there. Recall Takahashi’s (2001) case (concerning the inapplicability of scrambling to pro), given in (48): (48) [TP Subj-Nom [[vP pro [tSubj [[VP tObj V] v]]] T]] This structure involves object-shift and subject-raising. Each movement satisfies EPP and values uninterpretable features; and under SCT, such “newly valued” features are removed as they get valued. Now notice, after objectshift, the q-set of T (seeking to match the q-set of Subj) probes into vP, crossing over pro (occupying Spec-v). Under the condition (40), this probing of T to Subj is allowed, given that Case was removed from pro. Note that the dissociation of PC is irrelevant since pro has no PC in the first place. Thus, it follows that, upon the valuation of Case, pro is no longer syntactically visible; the structure [TP pro [Subj-Nom t V]] (where pro is scrambled to sentenceinitial position) won’t be generable.18 As demonstrated above, the feature-dissociation analysis incorporating SCT, which executes the condition (40) derivationally, captures the four cases in question, invoking neither “look ahead” nor “back tracking” procedure.

6. Concluding Remarks As we have seen above, all the cases discussed in this paper were shown to follow from the condition (40) within the feature-dissociation analysis. Although further research is necessary to conduct a more comprehensive study of the specific proposals made in this paper (in particular, the analysis of syntactic visibility), to the extent that they are (more or less) tenable, the strongly derivational system incorporating SCT receives not only conceptual but also empirical support.

Phonological Content and Syntactic Visibility

227

Acknowledgements We would like to thank Noam Chomsky, Chris Collins, Masuyo Ito, Yasuhiko Kato, Yoichi Miyamoto, Masayuki Oishi, Mamoru Saito, Shigeo Tonoike, and especially Samuel D. Epstein and T. Daniel Seely for valuable comments and helpful discussion. Part of the material in this paper was presented at Tilburg University (the Workshop on Triggers) and Aoyama Gakuin University (the 20th National Conference of the English Linguistic Society of Japan). We are grateful to the participants for clarifying remarks. Also, special gratitude goes to the Department of Linguistics and Philosophy at Massachusetts Institute of Technology, where this research was conducted. The research reported here was supported in part by the J. William Fulbright Foreign Scholarship (research grant given to Hisatsugu Kitahara, 2000-2002).

Notes 1. See Kitahara 1994, 1997 for related discussion. 2. Verbal configurations lacking external arguments are called weak phases, and they are distinguished from other verb phrases and clauses that are called strong phases. 3. This derivational analysis can be seen as an elaboration of Epstein 1992, where he derives the “last resort” property of NP-movement from the assumption that NP becomes an LF object upon the satisfaction of its Case. For related discussion, see Kitahara 2000, 2002. 4. We leave aside the problem concerning the valuation of an expletive. See Kitahara 2001 for relevant discussion. 5. Epstein and Seely (2002) find this suggestion problematic. They point out that, if the relevant distinction is lost after valuation, then it is necessarily lost “shortly after” valuation as well. We will discuss this issue in section 5. 6. See Collins 2002 for further discussion. 7. The “bypass” analysis does not extend to the cases where an intervening category induces agreement. We continue to assume that the search of probe P terminates when P finds goal G that agrees with P. Recall that this assumption restricts the search domain of P, thereby contributing to the reduction of operative complexity. In the derivation of (21), therefore, the q-set of whatObj becomes inactive upon the valuation of Case, despite the presence of other uninterpretable features rendering subsequent wh-movement possible. 8. For recent discussion concerning data such as (24a–e), see Pesetsky and Richards 2002. 9. For recent discussion concerning scrambling and interpretability, see Kitahara 2002, Saito 2001, and Takahashi 2001.

228

Ruriko Kawashima and Hisatsugu Kitahara

10. Recall that the search of T terminates when T agrees with Subj. If T involves Atype instance of scrambling, then Obj must move to Spec-v (above Subj) to be an element in the search domain of T. 11. To some extent, the distinction between Agree-driven and Match-driven movement can be seen as the revival of the distinction between forced and unforced q-agreement (in the sense of Fukui 1986 and Kuroda 1988). 12. We assume that Agree of T and Subj may induce Agree-driven movement or Match-driven movement, or both if the option (27) is exercised. 13. Note that, for the valuation of Case, pro can move to Spec-v, from which pro ccommands in-situ Subj, but no relevant binding relation can be established between pro and (any part of) in-situ Subj. We tentatively assume that only syntactically visible categories enter into a binding relation. Given this assumption, upon the valuation of Case, pro cannot enter into a binding relation with (any part of) in-situ Subj. 14. For recent discussion concerning the strongly derivational approach, see Epstein and Seely 2002, forthcoming. 15. See Saito 2001 for related discussion. 16. Note that the dissociation of PC of Obj from the intervening occurrence renders the q-set of Obj invisible for the probe T (see note 7). But this does not mean that PC is removed from Obj. Thus, if CHL affects other accessible features of Obj for subsequent movement (say, wh-movement), then Obj (including PC) will be merged with the relevant higher occurrence. 17. We continue to assume PIC (motivated on both conceptual and empirical grounds), though the relation between PIC and Transfer (reviewed in section 1) is no longer tenable. 18. Given the condition (40), one might expect that, after the removal of Case, pro and its overt counterpart whose PC was dissociated should behave alike. Indeed, both do not induce intervention effect. But notice, unlike pro, its overt counterpart whose PC was dissociated can undergo scrambling. This asymmetry arguably follows, given that CHL executes feature-dissociation after having established the relevant matching relation (required for scrambling). To be concrete, let us consider the structure (i) (where Obj and Subj match the probe T, but Obj is inactive): (i) T [vP Obj [Subj v [VP V tObj]]] In this structure, the q-set of T (seeking to match the q-set of Subj) probes into vP, crossing Obj (occupying Spec-v). Now notice, the q-set of T first matches the q-set of Obj. To allow the further search of T, CHL dissociates PC of Obj from the relevant intervening occurrence at this point of a derivation. The q-set of T then matches the q-set of Subj. In this way, the probing of T to Subj can establish an “extra” matching relation: Match of T and Obj. Note that the q-set of T has no way to match the q-set of Obj (occupying Spec-v) if Obj is pro, because pro bears no PC in the first place; hence, inactive pro is not visible for the probe T, under the condition (40).

Phonological Content and Syntactic Visibility

229

References Chomsky, Noam 1986 Knowledge of Language: Its Nature, Origin, and Use. New York: Praeger. 1995 The Minimalist Program. Cambridge: MIT Press. 2000 Minimalist inquiries: The framework. In Step by step: Essays on minimalist syntax in honor of Howard Lasnik, Roger Martin, David Michaels, and Juan Uriagereka (eds.), 89–155. Cambridge: MIT Press. 2001a Derivation by phase. In Ken Hale: A life in language, Michael Kenstowicz (ed.), 1–52. Cambridge: MIT Press. 2001b Beyond explanatory adequacy. MIT Occasional Papers in Linguistics 20, Cambridge, Mass. Collins, Chris 2002 Eliminating labels. In Derivation and explanation in the minimalist program, Samuel D. Epstein and T. Daniel Seely (eds.), 42–64. Oxford: Blackwell Publishing. Epstein, Samuel D. 1992 Derivational constraints on A’-chain formation. Linguistic Inquiry 23: 235–259. 1999 Un-principled syntax and the derivation of syntactic relations. In Working Minimalism, Samuel D. Epstein and Norbert Hornstein (eds.), 317–345. Cambridge: MIT Press. Epstein, Samuel D., Erich M. Groat, Ruriko Kawashima, and Hisatsugu Kitahara 1998 A derivational approach to syntactic relations. New York: Oxford University Press. Epstein, Samuel D. and T. Daniel Seely 2002 Rule applications as cycles in a level-free syntax. In Derivation and explanation in the minimalist program, Samuel D. Epstein and T. Daniel Seely (eds.), 65–89. Oxford: Blackwell Publishing. Epstein, Samuel D. and T. Daniel Seely forthc. Transformation and derivations. Cambridge: Cambridge University Press. Fukui, Naoki 1986 A theory of category projection and its application. Doctoral diss., MIT, Cambridge, Mass. Kitahara, Hisatsugu 1994 Target _: A unified theory of movement and structure-building. Doctoral diss., Harvard University, Cambridge, Mass. 1997 Elementary operations and optimal derivations. Cambridge: MIT Press. 2000 Two (or more) syntactic categories vs. multiple occurrences of one. Syntax 3: 151–158.

230

Ruriko Kawashima and Hisatsugu Kitahara

2001

Topics in the minimalist program. (Paper presented at the LOT Winter School 2001, Amsterdam, The Netherlands.) 2002 Scrambling, case, and interpretability. In Derivation and explanation in the minimalist program, Samuel D. Epstein and T. Daniel Seely (eds.), 167–183. Oxford: Blackwell Publishing. Kuroda, Shige-Yuki 1988 Whether we agree or not: A comparative syntax of English and Japanese. Linguisticae Investigationes 12: 1–47. McCloskey, James 2000 Quantifier float and wh-movement in an Irish English. Linguistic Inquiry 31: 57–84. Miyagawa, Shigeru 2001 The EPP, scrambling, and wh-in-situ. In Ken Hale: A life in language, Michael Kenstowicz (ed.), 293–338. Cambridge: MIT Press. Pesetsky, David and Norvin Richards 2002 Wh-movement: Probes and Goals. (Paper presented at MayFest 2002: The Minimalist Fact, University of Maryland, College Park.) Saito, Mamoru 1985 Some asymmetries in Japanese and their theoretical implications. Doctoral dissertation, MIT, Cambridge, MA. 1989 Scrambling as semantically vacuous A’-movement. In Alternative conceptions of phrase structure, Mark Baltin and Anthony Kroch (eds.) 182–200. Chicago: University of Chicago Press. 1992 Long distance scrambling in Japanese. Journal of East Asian Linguistics 1: 69–118. 2001 Toward the Unification of Japanese Scrambling. MIT Working Papers in Linguistics 41: 287–307. Cambridge, MA. Sportiche, Dominique 1988 A theory of floating quantifiers and its corollaries for constituent structure. Linguistic Inquiry 19: 425–450. Takahashi, Daiko 2001 Scrambling and empty categories. MIT Working Papers in Linguistics 41: 47–58. Cambridge, MA. Uriagereka, Juan 1999 Multiple spell-out. In Working Minimalism, Samuel D. Epstein and Norbert Hornstein (eds.), 251–282. Cambridge: MIT Press.

On Triggers of Movement and Effects at the Interfaces Mariana Lambova

1. Introduction I argue that head movement belongs to syntax proper and conditions and/or effects holding at the interfaces cannot be triggers for movement for conceptual reasons. Empirically, the discussion focuses on contrastive focalization of predicates in Bulgarian (BG). I show that heads undergo overt focus fronting, an A-bar operation, on a par with phrases. Given that head movement can have semantic import, I reject recent suggestions that it is a PF operation. I also argue that apparent instances of PF reordering should be resolved via activation of lower copies of movement. The phenomenon of interest is the Part(iciple) Aux(iliary) construction where the main verb is fronted with respect to the finite Aux with a null subject: (1)

Gledali bjaxa filma. seen Aux.PT.3P.PL movie-the “They had SEEN the movie.” cf. Bjaxa gledali filma. Aux.PT.3P.PL seen movie-the

The exemplified verbal form of the Past Perfect tense (minalo predvaritelno vreme) is complex: it consists of a finite past Aux and the Part of a main verb, and I will refer to it as the periphrastic predicate. Descriptive grammars note that the past Aux alternates freely with Part of the main verb in two contexts: clause initially, as in (2), and after a pause, indicated by # in (3).The alternation, however, is blocked when the periphrastic predicate is preceded by another “word” (Stojanov 1980: 377):1 (2)

a. Bjaxme privârs
ili rabotata, kogato zavalja. Aux.PT.3P.PL finished work-the when rain-PT.3P.SG

232

(3)

(4)

Mariana Lambova

b.

Privârs
ili bjaxme rabotata, kogato zavalja. finished Aux.PT.3P.PL work-the when rain-PT.3P.SG

a.

Kogato zavalja, # bjaxme privârs
ili rabotata, when rain-PT.3P.SG Aux.PT.3P.PL finished work-the

b.

Kogato zavalja, # privârs
ili bjaxme rabotata, when rain-PT.3P.SG finished Aux.PT.3P.PL work-the

a.

Az bjax pristignal naskoro. I Aux.PT.3P.PL arrived not-long-ago

b. *Az pristignal bjax naskoro. I arrived Aux.PT.3P.PL not-long-ago Embick and Izvorski (1995) are the first to note that the Part Aux order, as in (1a), is perceived as “marked” (but see Wilder and Cavar 1994 for a similar observation concerning Romance). The effect appears to relate to the appropriateness of the utterance in discourse. I will argue that the main verb is contrastively focused (see the accompanying translation), and will derive the restriction on the occurrence of overt material with the reordered periphrastic predicate from an interaction of certain constraints on stress assignment and intonation. The paper is organized as follows: I will first show that the Part Aux construction in question involves two different cases but only one displays a semantic effect. In spite of different motivations, the reordering will be shown to arise in a similar manner in PF. After I sketch an analysis, I will address the issues surrounding the claim to be defended here that interface effects cannot be triggers for movement.

2. The Part Aux Construction The noted reordering within the periphrastic predicate is possible with the present Aux as well, in the Perfect tense (minalo neopredeleno vreme).The subject is again null: (5)

Gledali sa filma. seen Aux.PRES.3P.PL movie-the “They have seen the movie.”

On Triggers of Movement and Effects at the Interfaces

233

The fact has received a fair amount of interest in the literature, and prompted the controversial L(ong) H(ead) M(ovement) hypothesis (Lema and Rivero 1989, Rivero 1991, and others) according to which Part moves directly to Cº across Aux: (6)

a. [CP [C’ gedalii [IP [I’ sa [VP ti filma ]]]]]

= (5)

b. [CP [C’ gedalii [IP [I’ bjaxa [VP ti filma ]]]]]

= (1a)

However, as observed by Embick and Izvorski as well as Bo‰koviç (1997), for the past Aux (7a) the presence or absence of a lexical subject has no effect on the “normal” Aux Part order. This is not the case with the present Aux; see (7b) and (5). They all agree that the difference is attributed to the enclitic nature of the present Aux. More precisely, the example in (7b) is unacceptable when the subject is null since the present Aux needs a host on the left: (7)

a. (Decata) bjaxa gledali filma. kids-the Aux.PT.3P.PL seen movie-the b. *(Decata) sa gledali filma. kids-the Aux.PRES.3P.PL seen movie-the

Given that the contrast in (7) arises because the Part Aux order with the present Aux is conditioned by a phonological requirement of the clitic Aux and is therefore unavoidable, Embick and Izvorski label the case in (5) “obligatory” and that in (1) “optional”: (8)

a. *Sa gledali filma. Aux.PRES.3P.PL seen movie-the b. Gledali sa filma. seen Aux.PRES.3P.PL movie-the

(9)

= (5)

a. Bjaxa gledali filma. Aux.PT.3P.PL seen movie-the b. Gledali bjaxa filma. seen Aux.PT.3P.PL movie-the

= (1a)

The distinction is important because the “optional” reordering has semantic import, unlike the “obligatory” one. However, these terms are misleading for several reasons.

234

Mariana Lambova

The noted contrast obtains only in the root context with a null subject, as in the examples above. There the present Aux is clause initial and its dependency obvious, so the reordering is phonological in nature. Importantly, it is crucially different in motivation from the case with the past Aux.2 The proposed terms thus link motivation to contextual distribution. However, in the embedded context the reordering with the present Aux is “optional” just as with past Aux (see below). This necessitates a reexamination of the distributional facts. Furthermore, on Minimalist assumptions all operations apply as a “last resort”, hence there should be no “optional” operations. Indeed, I argue that no syntactic optionality is involved in the construction under consideration. Recall that the “optional” Part Aux order is perceived as “marked”. Such an utterence is felicitous in a situation in which the speaker is presenting the activity under discussion as an alternative. For example, (1a) can be uttered when the discourse contains either an explicit or implied reference to the movie being in possession, i.e. rented or owned. A possible paraphrase will be “They have only seen the movie.” In other words, the main verb is contrastively focused. There is a further correlation with intonation. As observed by Penchev (1978), contrastive elements in BG attract a high tone. I suggest that the perceived effect associated with the “marked” Part Aux order is due to a specific tonal contour: (10) a. Bjaxa gledali filma. Aux.PT.3P.PL seen movie-the b. Gledali bjaxa filma. seen Aux.PT.3P.PL movie-the

gp gp g

Thus, while the “normal” Aux Part order in (10a) is pronounced on a medium pitch gradual fall, the “marked” Part Aux order in (10b) is pronounced on a high pitch gradual fall. The two contours differ only in pitch level. Gradual fall is characteristic of both neutral/SVO statements and those with a contrastively focused fronted element, but the latter require higher pitch.3 In Lambova 2003, I argue that contrastive focalization in BG, which is associated with the high pitch fall, involves overt focus movement, an operation that is not discourse neutral. For this reason, I take the medium fall to be neutral and the high fall to be non-neutral. Accordingly, I consider the Aux Part order in (10a), which is pronounced on the medium fall, neutral; the Part Aux order in (10b), which is pronounced on the high fall, is non-neutral.

On Triggers of Movement and Effects at the Interfaces

235

Next, observe that the phonologically conditioned Part Aux order is also pronounced on medium pitch or the neutral fall. I other words, it patterns with the “normal” Aux Part: (11) Gledali sa filma. seen Aux.PRES.3P.PL movie-the

gp

Therefore, the “obligatory” Part Aux order is discourse neutral. On the other hand, the “optional” Part Aux order is discourse non-neutral which correlates with the perceived effect. The neutral/non-neutral distinction is not tied to contextual distribution. This is all the more better since, as Embick and Izvorski show, “optional” reordering is possible for the present Aux in the embedded context. The parallel is significant: there is no difference between the two Aux’s. (12) Decata kazvat, … kids-the say-PRES.3P.PL a. âe bjaxa gledali filma. that Aux.PT.3P.PL seen movie-the “The kids say that they had seen the movie.” b. âe gledali bjaxa filma. that seen Aux.PT.3P.PL movie-the “The kids say that they had SEEN the movie.” (13) Decata kazvat, … kids-the say-PRES.3P.PL a. âe sa gledali filma. that Aux.PRES.3P.PL seen movie-the “The kids say that they have seen the movie.” b. âe gledali sa filma. that seen Aux.PRES.3P.PL movie-the “The kids say that they have SEEN the movie.” Note that the complementizer provides sufficient support for the clitic Aux when the subject is null. That is, the Aux Part order in (13a) is acceptable, unlike in (8a). Therefore, there is no need for an “obligatory” or phonologically conditioned reordering in (13b). As a consequence, the Part Aux order in (13b) is in free variation with the Aux Part order in (13a), just as for the past Aux in (8) and (12).

236

Mariana Lambova

If the Part Aux order is equally available to both Aux’s in the embedded context, then it must be that there is no syntactic difference between the Part Aux constructions in (12b) and (13b). Simply stipulating the contexts in which the non-neutral Part Aux orders can occur is undesirable. It must be shown if and why a particular construction is disallowed in a certain context. I conclude that the non-neutral Part Aux order is possible with the present Aux in root contexts as well. In other words, in addition to the phonologically conditioned reordering producing a neutral Part Aux order there must be an alternative derivation for (8b), which is obscured by the properties of the present clitic Aux. Importantly, on closer inspection there turns out to be ambiguity associated with a matrix context Part Aux order with present Aux. Observe that the Part Aux order in (14) can be associated with the high pitch. That is, it correlates with either the neutral or the non-neutral tonal contour. This fact has not been noticed before. I will show that in (14b), just as in (9b), the verb is contrastively focused, the higher pitch reflecting the prominence on Part, easy to miss with the difference between (14a) and (14b) being only in pitch level: (14) a. Gledali sa filma seen Aux.PRES.3P.PL movie-the “They have seen the movie.”

gp

b. Gledali sa filma. seen Aux.PRES.3P.PL movie-the “They have SEEN the movie.”

ggp

The high tone which contributes to the prominence of a contrastively focused element is associated with the beginning of an utterance. For XPs (given here in small capitals), it is uncontroversial that they undergo fronting in both root and embedded context (15). I will show that heads front too. (15)

ste prodali ti. (Ivan kaza, Ëe) KOLATAi Ivan say-PT.3P.SG that car-the (foc) Aux.PRES.2P.PL sold “(Ivan said that) it was the car that you sold.”

To summarize so far, the neutral word order is Aux Part, which for the present clitic Aux is only possible with clause-mate overt lexical material preceding it. No such restriction applies to the non-clitic past Aux. There are two kinds

On Triggers of Movement and Effects at the Interfaces

237

of Part Aux orders with the present Aux – one neutral and another nonneutral, the former restricted to the root context with a null subject. For the past Aux, however, the Part Aux order is always non-neutral.

3. An Analysis Previous accounts appeal to LHM or adjunction.4 The proponents of the LHM hypothesis (cf. Rivero 1991; Roberts 1992; Wilder and Cavar 1994) propose that Part invariably moves across Aux in the syntax. Bo‰koviç (1997) argues that the relevant operation involves adjunction below C, not LHM. Interestingly, none of these analyses mentions semantic differences between the two kinds of Part Aux orders. On the other hand, Embick and Izvorski (1995: 113), whose starting point is that the “optional” Part Aux order is perceived as marked, advocate a non-uniform analysis. However, they end on a cautionary note that the reordering in both cases may result from one operation but on different motivations. Their speculation will turn out to be very relevant to the particular analysis I develop below.

3.1. Why the LHM Hypothesis is Untenable? Lema and Rivero’s (1989) account is the earliest generative attempt to capture the reordering with periphrastic predicates. Assuming that Part moves in the syntax to provide the clitic Aux with a host, they propose that the movement occurs in one step skipping the position occupied by Aux; hence LHM since it violates the head movement constraint: (16) [CP Parti [IP Aux [VP ti (NP)]]] Important evidence that the Part Aux order indeed involves head movement rather than remnant VP fronting (with the direct object moving out of VP prior to the VP fronting) is the impossibility of uncontroversial VP fronting where the object remains within VP (Rivero 1991: 322–323): (17) a. *[Proãel knigata] (Petâr) e (Petâr). read book-the Petâr Aux.PRES.3P.SG Petâr b.

Proãel e (Petâr) knigata (Petâr). read Aux.PRES.3P.SG Petâr book-the Petâr

238

Mariana Lambova

There are quite a few empirical facts arguing against the LHM hypothesis. First, recall that the (non-neutral) Part Aux order is possible in the embedded context. This suggests that Part cannot be in C which is filled by the declarative complementizer (on ruling out CP recursion, see Embick and Izvorski 1995). Second, as Bo‰koviç (1997) argues convincingly for a similar construction in Serbian/Croatian (SC), Part cannot be as high as C. Under the standard analysis, Aux in SC is invariably in C (cf. Wilder and Cavar 1994), but this cannot be true, given the contrast: (18) a. Jovan je nesumnjivo istukao Petra. [SC] Jovan.NOM Aux.PRES.3P.SG undoubtedly beaten Peter.ACC “Jovan undoubtedly beat Peter.” b. *Istukao je nesumnjivo Petra. beaten Aux.PRES.3P.SG undoubtedly Peter.ACC “He undoubtedly beat Peter.” The data show that, unlike Aux, Part cannot move across sentential adverbs. Importantly, for BG, that is true for the neutral Part Aux order,5 i.e. the case for which Lema and Rivero posit movement of Part to C: (19) *Gledali sa nesâmneno filma. seen Aux.PRES.3P. PL undoubtedly movie-the “They have undoubtedly seen the movie.”

[BG]

Assuming with Watanabe (1993, see also Bo‰koviç 1997) that sentential adverbs are universally TP adjoined, the ungrammatical examples above show that the Part does not raise above TP. Yet, it does not necessarily follow that Part remains within VP, as the behavior of ambiguous adverbs shows: (20) a. Ivan pravilno e otgovoril na vâprosa. [BG] Ivan correctly Aux.PRES.3P.SG answered to question-the “Ivan gave a correct answer to the question.” “Ivan did the right thing in answering the question.” b. Ivan e otgovorili pravilno ti na vâprosa. Ivan Aux.PRES.3P.SG answered correctly to question-the “Ivan gave a correct answer to the question.” “*Ivan did the right thing in answering the question.”

On Triggers of Movement and Effects at the Interfaces

239

If such adverbs have different attachments, the loss of the sentential meaning indicates that while Part moves out of VP (and above VP adjoined manner adverbs), it does not raise very high in the tree, certainly not all the way to C. The strongest argument against LHM, however, comes from the restriction on the overt subject. Embick and Izvorski (1995: 111) point out that if Part were indeed moving across Aux to Cº, nothing should, in principle, ban a lexical subject from occurring between Part and Aux, contrary to fact: (21) a. *Gledali decata bjaxa filma. seen kids-the Aux.PT.3P.PL movie-the “The kids had SEEN the movie.” b.

[CP [C’ gledalii [IP subject [I’ bjaxa [VP ti filma ]]]]]

Related issues are where and why LHM applies. If LHM is phonologically motivated in the case of the clitic Aux, can syntax see as far as PF? Or, why it should apply “optionally” in the case of the past non-clitic Aux? On the standard assumption of barring lookahead, the phonologically motivated neutral reordering cannot be anticipated in syntax. Given the associated semantic effect, the non-neutral reordering must be syntactic. If it is triggered by feature checking, there shouldn’t be any optionality. To sum up, in the face of these difficulties I conclude that LHM does not capture the phenomenon of predicate fronting even descriptively. The alternative Bo‰koviç (1997) pursues for SC is adjunction of Part to Aux. Extending his analysis to BG has some merit since no lexical material is allowed into the periphrastic verbal form, even with a non-clitic Aux, regardless of reordering:6 (22) a. *Ivan be‰e togava oti‰âl na riba. Ivan Aux.PT.3P.SG then gone at fishing b. *Ivan be‰e na riba oti‰âl (togava). Ivan Aux.PT.3P.SG at fishing gone then (23) a. *Oti‰âl togava be‰e na riba. gone then Aux.PT.3P.SG at fishing b. *Oti‰âl na riba be‰e (togava). gone at fishing Aux.PT.3P.SG then

240

Mariana Lambova

Bo‰koviç proposes that the direction of adjunction in SC is free, i.e. Part can left-adjoin to Aux. This would produce a Part Aux complex.7 However, the issues that remain in BG are the restriction on the overt subject as well as the perceived semantic effect in the non-neutral case.

3.2. The Neutral Case In light of the fact that nothing can intervene between Aux and Part, I propose that Part adjoins to Aux in BG. Presumably, the operation happens as soon as Aux is merged, and the verbal complex Aux+Part undergoes further raising since it surfaces to the left of VP-adjoined adverbs (24). Following Bo‰koviç 1997 (who in turn follows Wilder and åavar 1994), I assume the motivation is checking an [+Aux] feature.8 (24) Ivan e (*napâlno) zabravil (napâlno) Ivan Aux.PRES.3P.SG completely forgotten completely tezi podrobnosti (napâlno). these details completely Assuming that movement creates copies, syntax provides at least two copies of the verbal complex, and thus the possibility for pronouncing a lower copy of Aux if necessary: (25) a. syntax: b. phonology:

[sa gledali [sa gledali ]] … [gledali … sa]

Following Franks 1998 (see also Bo‰koviç 2001), I assume that it is heads of non-trivial chains that are normally pronounced but a lower copy can be activated instead to avoid a PF violation. Here, it is the absence of a host for the clitic Aux. Recall that when the subject is null (or nothing else precedes the verbal complex) the clitic Aux surfaces clause initial. However, pronouncing a lower copy of Aux, which I show in bold, will meet the requirements of the clitic: (26) a. *Sa gledali filma. Aux.PRES.3P.PL seen movie-the b. [TP [T’ sa gledali [VP sa gledali filma]]]

= (8b)

On Triggers of Movement and Effects at the Interfaces

241

On the other hand, activation of lower copies of movement is not free. As a last resort strategy, it will be banned if not necessary, e.g. when there is overt material to the left of the clitic Aux to support it (27)–(28). Similarly, in the case of the past Aux which does not need a host (29).This is why the absence of an overt subject with the past Aux makes no difference, i.e. there is no neutral reordering with the past Aux: (27) a. Decata sa gledali filma. kids-the Aux.PRES.3P.PL seen movie-the b. *[TP decata [T’ sa gledali [VP sa gledali filma]]] (28) a. Filma sa gledali. movie-the (top) Aux.PRES.3P.PL seen b. *[TopP filma [T’ sa gledali [VP sa gledali]]] (29) a. (Decata) bjaxa gledali filma. kids-the Aux.PT.3P.PL seen movie-the

= (7a)

b. *[TP (decata) [T’ bjaxa gledali [VP bjaxa gledali filma]]] For concreteness, I assume that Aux in BG is in “little” vP (in the sense of Chomsky 1995, or possibly the predication projection of Bowers 1993). Nothing will change if Aux is base generated in the highest VP shell and subsequently moves to “little” v. First Part right-adjoins to Aux and then the complex head undergoes further movement to T°. (30) a.

vP g v’ 2 v VP g g Aux tj 1 Aux Partj

b.

TP g T’ ro T vP 2 # T vk # g v’ Aux 2 VP 1 tk g Aux Partj g tj

242

Mariana Lambova

As to the phonetically null subject, it raises as mere formal features, presumably adjoining first to Auxº and then moving as part of the verbal complex (which I will ignore in the derivation below). Then the derivation of example (5) is: (31) a. [v’ sa [VP pro gledali filma]] b. [vP [v’ sa+gledali [VP gledali filma]]] c. [TP [T’ sa+gledali [vP [v’ sa+gledali [VP gledali filma]] The output in (31c) will crash at PF if the highest copy of sa is pronounced because there is no appropriate host for the clause initial clitic Aux. Adopting Bo‰koviç’s (1997, 2001) “weak phonology approach to the syntax-phonology interface”, I assume that PF filters the syntactic output passively. Specifically, copy deletion is a left to right evaluation procedure: pronouncing the highest copy of sa is not acceptable but pronouncing its lower copy is allowed. (32) [sa+gledali [ sa+gledali]]


*[sa gledali] [gledali … sa]

Note that the pronounced Part and Aux in (32) belong to two different (but adjacent copies) of the complex verbal head. This is an instance of “scattered deletion”. As discussed in Bo‰koviç 2001 and Nunes 1995, “scattered deletion” is a disfavored option – it arises only when deletion cannot yield a legitimate PF output, which is the case in the construction under consideration. “Scattered deletion”, which Bo‰koviç discusses based on pronominal clitics in Slavic, targets partially two separate copies. All his examples involve PF adjacency of the two copies but he does not impose it as an explicit condition on the process. I suggest that adjacency is crucial to the syntaxphonology mapping under “scattered deletion”. This analysis differs from Embick and Izvorski (1997) who have PF movement in the neutral reordering (regardless of its instantiation as morphological merger or prosodic inversion). Given Bo‰koviç’s (2001) book-long conceptual and empirical justification of the copy theory of movement and the activation of lower copies, there is no place for PF movement in the grammar. The gist of the argument is that if apparent instances of PF movement can be reanalysed without appealing to such movement, it is simpler not to have it. Finally, it is crucial for the analysis proposed here that adjunction is to the right.9 Recall that Aux cannot precede a sentential adverb in BG (19) which I take as evidence that Aux cannot move out of the verbal complex. If

On Triggers of Movement and Effects at the Interfaces

243

so, in the presence of an appropriate host leftward adjunction (as well as free direction of adjunction) will result in an unacceptable Part Aux sequence (see fn. 7). Therefore, the direction of adjunction in BG must be to the right. 3.3. The NonNeutral Case The non-neutral Part Aux order, and in particular its most puzzling property – its co-occurrence with an obligatorily null subject – is of more interest. Recall Embick and Izvorski’s observation that the construction is perceived as “marked”. I made that intuition more precise by correlating the reordering with the high pitch fall, the characteristic tonal contour of contrastively focused elements. Taking the shared intonation as an indication of the same underlying process, I propose that the main verb in the non-neutral Part Aux construction is contrastively focused (see the accompanying translations).10 If the verb is focused, the perceived semantic effect follows. It is wellknown that contrastive focalization has semantic import. Furthermore, there is nothing optional about the operation. If the verb is taken out of the lexicon with a focus feature, the feature must be checked. In BG, a contrastively focused phrase has a strong [+Foc] feature, and therefore undergoes overt focus fronting. In Lambova 2003, I argue that focus is licensed in 6P, a discourse-related projection above TP. The null hypothesis is that a contrastively focused head should also front. Recall from the previous section that Part adjoins to Aux and the complex head moves to Tº. What fronts to 6º is Aux+Part: 6P ei 6 TP 2 g T’ 6 Ti 1 2 vP Ti vk ti g g Aux g 1 v’ Aux Partj 2 VP tk g tj

(33) a.

244

Mariana Lambova

On these assumptions, the syntactic output in (33) is an Aux Part sequence. However, the linearized string is Part Aux. I claim that the reordering is due to the interaction of certain constraints on stress assignment and intonation of focused elements. Crucially, although Part bears the strong [+Foc] feature, it cannot excorporate out of the verbal complex. If it did, an overt subject should be able to surface between it and Aux. Let me show how the reordering arises. I suggest that, given the two adjacent copies of the verbal complex in 6º and Tº, Part is pronounced in 6º and Aux is pronounced in Tº, i.e. the Part Aux sequence results from “scattered deletion” which targets partially 6P and TP: (34) syntax: phonology:

[6P [6’ bjaxa+gledali [TP [T’ bjaxa+ gledali [vP bjaxa+gledali … ]]]]]] [gledali … bjaxa]

Now suppose that although Part and Aux are in different projections, they must map onto a single prosodic constituent. As Howard Lasnik (personal communication) reminds me, this is a rather standard assumption about affixation in English. Therefore, I propose that adjunction structures in general must map onto a single prosodic constituent (cf. Lambova 2003).11 Assuming this is correct, when there is a subject in TP, it must be obligatorily null, hence given in outline font (35). A lexical subject, which presumably maps onto a prosodic constituent on its own, would have to intervene into the prosodic constituent of the complex verb, and this is not allowed. Thus I derive the restriction on the overt subject in the non-neutral Part Aux construction: (35)

[6P [6’ bjaxa+gledali [TP subject [T’ bjaxa+gledali]]]]




*[ gledali [subject]t bjaxa ]t Lower copies of movement can only be activated if necessary to avoid a PF violation. In order to show how the particular violation responsible for the “scattered deletion”arises, I will have to make a digression on stress assignment and intonation of contrastively focused elements in BG. Recall that contrastive focalization in BG involves overt displacement. (I will discuss only root cases but everything carries over to the embedded context.12) The fronting is obligatory (36), i.e. a contrastively focused phrase cannot remain in situ. It must move to a clause initial position to be pronounced on a high tone, i.e. the characteristic contour is high fall:

On Triggers of Movement and Effects at the Interfaces

245

(36) a. *Vidjax ZENA. see-PT.1P.SG woman-the (foc) b. ZENA vidjax. woman-the (foc) see-PT.1P.SG “I saw a WOMAN.”

gp g

The high tone is the pitch accent or phrasal tone and it is associated with the clausal edge. It is perceived as higher pitch on the contrastive element, now in the left periphery, making it more prominent. In other words, fronting and the higher pitch are jointly responsible for singling out the contrastively focused element. Note that a corresponding non-focused phrase, as in (37), does not require fronting, and is pronounced on a medium pitch gradual fall: (37) Vidjax zena. see-PT.1P.SG woman-the “I saw a woman.”

gp

This marking of contrastive focalization is not a universal strategy. Compare, in this respect, a non-focus-fronting (or focus in situ) language, such as English. There is no overt displacement of a contrastively focused phrase (given in small capitals and bold) but nuclear stress “shifts” onto that element: tu (38) a. I bought flowers for MARY. gpt tp b. I bought FLOWERS for Mary. gut In (37a), the prepositional object is contrastively focused: it remains in situ but is pronounced on a perceptibly higher pitch associated with contrastive stress.13 In (38 b), it is the direct object that is contrastively focused. The word order has not changed but now the direct object receives contrastive stress, i.e. the nuclear stress “shifts” onto it. I suggest that contrastive stress is invariably perceived as higher pitch: both focus fronting and focus in situ languages utilize pitch to make the element in question more prominent. However, languages differ in how they encode focus in syntax: the high tone in BG has a fixed domain at the clausal edge, and overt fronting puts the respective phrase in that domain. In English the high tone is not associated with a specific domain but is realized

246

Mariana Lambova

on the phrase(s) in question resulting in “relocation” of pitch accent.14 (Interestingly, while BG allows only one contrastively focused phrase per clause, there is no such restriction in English.) Assuming that an operator variable chain is necessary for the interpretation of contrastive focus, English may be employing covert displacement.15 Now while the Aux+Part complex head undergoes fronting in BG, the string surfaces as a Part Aux sequence. Recall that the high tone is aligned with the left edge of the clause. The contrastively focused element is Part but it is not at the left egde. Furthermore, it gets the phrasal stress. I assume a theory of prosodic phonology according to which a hierarchical layered representation structures a string of phonological segments into syllables, the syllables into feet, and the feet into phonological words; then come the phonological and intonational phrases and the utterance (cf. Selkirk 1986; Nespor and Vogel 1986). Some of the prosodic constituents – namely the phonological word, the phonological phrase, and the intonational phrase show systematic relations to constituent structure but they are often argued not to be isomorphic to the respective syntactic constituents (cf. Nespor and Vogel 1986: 37; Truckenbrodt 1999: 220). I follow Nespor and Vogel (1986) who propose that right-branching languages, such as BG, stress the rightmost element in the phonological phrase, i.e. the element on the left is weak (indicated by a subscripted ‘w’) and the element on the right is strong (indicated by a subscripted ‘s’). Thus within the prosodic constituent which corresponds to the complex head in 6º, stress falls onto Part given below in small capitals: (39) syntax: phonology:

[6P bjaxa+gledali] [ bjaxaw gledalis]




[bjaxa GLEDALI] t

Suppose contrastive stress implies alignment of the stressed element with the high tone at the left edge. I further suggest that stress cannot “relocate” the leftmost Aux in BG. Thus Part faces conflicting requirements in the non-neutral construction: it is assigned stress which cannot shift, and it must be at the left edge of the clause to be aligned with the high tone marking contrast but it is linearly preceded by Aux by virtue of being rightmost in the prosodic constituent. I suggest that this conflict gets resolved through “scattered deletion”. Adjunction underlies the verbal complex and requires it that it maps onto a single prosodic constituent. This will be true for the Aux Part and the Part Aux sequences but only in the latter is Part at the left edge. Furthermore,

On Triggers of Movement and Effects at the Interfaces

247

activating a lower copy of Aux by “scattered deletion” is possible because there is an adjacent copy of the verbal complex in T°: (40) [6P bjaxa+gledali [ZP bjaxa+gledali [vP bjaxa+gledali filma]
[bjaxa GLEDALI]t [bjaxa gledali]t …
[bjaxa GLEDALI]t [bjaxa gledali]t … gp g [GLEDALI … bjaxa]t Compare how intonation and stress assignment interact in the neutral case with the non-clitic past Aux. Stress will fall on Part as before but there is no requirement that Part be at the left edge to be aligned with the high tone. Accordingly, there is no reordering: I assume that the medium pitch gradual fall is realized as a low high tone, with the low being aligned with the initial Aux and the high with the stressed Part. Now recall that the Part Aux order with the present clitic Aux in matrix contexts is ambiguous (14), the sole difference being in the range of pitch. The difference is between neutral stress and contrastive stress, the latter arising via alignment with the high tone. “Scattered deletion” now applies for two reasons: to meet the phonological requirements of the enclitic Aux but also to put Part at the left edge for alignment. Presumably, meeting the phonological needs of Aux happens early in the linearization process with stress assignment following so that Part is already in place when it has to be aligned with the high tone. Two observations are in order. First, what rules out (41) is that stress cannot relocate from Part to Aux for the purpose of tone alignment (on the relevant reading):16 (41) a. *BJAXA gledali filma. Aux.PT.3P.PL seen movie-the “They have SEEN the movie.” b. [6P bjaxa GLEDALI [TP bjaxa gledali]] *[BJAXA gledali]t




Neither is it possible to place contrastive stress on both Aux and Part: (42)

*BJAXA GLEDALI filma. Aux.PT.3P.PL (foc) seen (foc) movie-the “They HAVE SEEN the movie.”

248

Mariana Lambova

In this respect, note that for reasons that are not entirely clear multiple contrastive clause-mate foci are extremely rare in languages with overt focalizaltion (see Kiss 1995). This restriction pertains to both heads, as in the predicate fronting case, and phrases in BG (with or without fronting): (43) a. *DECATA FILMA gledaxa. kids-the (foc) movie-the (foc) see-PT.3P.PL b. *FILMA DECATA gledaxa. movie-the (foc) kids-the (foc) see-PT.3P.PL c. *DECATA gledaxa FILMA. kids-the (foc) see-PT.3P.PL movie-the (foc) d. *FILMA gledaxa DECATA. movie-the (foc) see-PT.3P.PL kids-the (foc) “The KIDS saw the MOVIE.” More interestingly, the restriction holds not only of two independent XP’s but also of a noun and its modifier within a single NP: (44) a. *VISOKA ZENA vidjax. tall (foc) woman (foc) see-PT.1P.SG “I saw a TALL WOMAN.” b. cf.VISOKA zena vidjax. tall (foc) woman see-PT.1P.SG “I saw a TALL woman.” Therefore, I conclude that it should not be possible to contrastively focus both Aux and Part in (42) either.

4. On the Nature of Head Movement In this section, I first discuss some predictions and consequences of the proposed analysis. Of particular interest is the material that can follow the nonneutral Part Aux sequence. It shows that the complex head has indeed undergone focus fronting but also constitutes important independent evidence for distinguishing between neutral and non-neutral Part Aux orders, which I have largely motivated based on intonational differences. I then argue that if focus fronting can have semantic import, head movement cannot be a postsyntactic phenomenon.

On Triggers of Movement and Effects at the Interfaces

249

4.1. More on the Non-Neutral Part Aux Order There is further evidence for the “scattered deletion” analysis. The constraints on stress assignment and intonation in the non-neutral Part Aux construction are also observable on phrases. Thus, BG allows only one contrastively focused element and it must be clause initial as in (44b). When the leftmost constituent of a clause initial NP cannot be contrastively focused independently, as in (45), the reordering is obligatory. It is only possible to have the head noun contrastively focused (46): while the whole NP is fronted, the noun is not at the clausal edge. (45)

*NJAKAKVA zena vleze v stajata. some (foc) woman enter-PT.3P.SG in room-the “*SOME woman entered the room.”

(46) a. *Njakakva ZENA vleze v stajata. some woman (foc) enter-PT.3P.SG in room-the b.

cf.ZENA njakakva vleze “Some WOMAN entered the room.”

v

stajata.

I assume that “scattered deletion” also applies in (47) on the reading in which the noun is contrastively focused:17 (47) a. *Visoka ZENA vleze v stajata. tall woman (foc) enter-PT.3P.SG in room-the b.

cf.ZENA visoka vleze v stajata. woman (foc) tall enter-PT.3P.SG in room-the “A tall WOMAN entered the room.”

The proposed analysis also makes a testable prediction. Recall that the subject must be obligatorily null in the non-neutral Part Aux construction (21). If “scattered deletion” bans the subject from intervening in the verbal complex, it should be possible for the subject to surface below it. The prediction is fully borne out:18 (48) a.

b.

Gledali (*decata) bjaxa (decata) filma. seen kids-the Aux.PT.3P.PL kids-the movie-the “The kids had SEEN the movie.” [6P [6’ bjaxa+gledali [TP decata [T’ bjaxa+gledali [vP decata bjaxa+gledali [VP decata gledali filma]]]]]

250

Mariana Lambova

This is yet another instance of activating lower copies of movement. Although the subject cannot be realized in SpecTP, nothing prevents pronouncing the copy in its base-generated VP-internal position. More importantly, this fact (not noticed by traditional grammarians) provides independent support for the non-neutral Part Aux order, and thus for the present analysis. Further evidence for the different syntax of the two kinds of orders comes from the possibility for a sentential adverb to follow a non-neutral Part Aux sequence:19 (49)

Gledali bjaxa verojatno filma. seen Aux.PT.3P.PL probably movie-the “They had probably SEEN the movie.”

This is in contrast with (19) where the verb has not moved beyond TP, and accordingly cannot precede such an adverb. Not surprisingly (50) with the present clitic Aux, which allows only one reading, patterns with (49) in the relevant respect. This fact confirms that the verbal complex is syntactically higher on the contrastive focus reading. Thus semantic, intonational, and syntactic evidence converges to differentiate the two kinds of orders. (50)

Gledali sa verojatno filma. seen Aux.PRES.3P.PL probably movie-the “They have probably SEEN the movie.” “*They have probably seen the movie.”

Finally, recall that positing LHM was motivated by the unavailabilty of uncontroversial VP fronting (17). Therefore, the non-neutral Part Aux order could not involve remnant VP fronting, as in German below: (51) a. b.

[VP Das Buch gestohlen ] hat Fritz sicher. the book stolen has Fritz certainly [VP Gestohlen] hat Fritz das Buch. stolen has Fritz the book

On my analysis, the lack of VP fronting in BG simply follows from the assumption that after adjoining to Aux in vP, Part moves out of it. There is, then, no phrasal constituent that contains Part but not Aux which fronting could target.

On Triggers of Movement and Effects at the Interfaces

251

4.2. Head Movement is (Purely) Syntactic I have shown that the reordering with the periphrastic predicate consisting of an aspectual Aux and the Part of a main verb in BG can involve contrastive focalization of the main verb, and when this is so the construction is perceived as “marked”.20 In that, there is no difference between the present and the past Aux , which happen to have different phonological properties, since both allow the non-neutral reading. Crucially, I argue that the complex verbal head undergoes overt focus fronting in these cases. The proposed analysis has implications for the nature of head movement. Within the Principles and Parameters framework head movement is standardly considered to be syntactic, on a par with phrasal movement. The idea that it is best viewed as a PF phenomenon is more recent and largely motivated by theory-internal considerations (cf. Chomsky 2000). One reason for adopting this idea is that head adjunction, if taken as an operation of the syntactic component, is a well-known exception to the Extension Condition (Chomsky 1993), which requires every operation to expand the tree. However, the reasoning is valid only if the Extension Condition is assumed to hold. As Bo‰koviç and Lasnik (1999) argue, the Extension Condition should be eliminated from the grammar since its empirically important effects are captured by the strength-based deduction of the cycle (Chomsky 1995).21 If the Extension Condition is redundant, head movement does not need to be treated as special in this respect. Of more interest is Chomsky’s observation that, unlike phrasal movement, head movement does not have a semantic effect (as cited in Boeckx and Stjepanoviç 2001). My goal has been to show that head movement can and, in fact, does have semantic import. BG is a focus fronting language. Apriori, if a contrastively focused phrase fronts obligatorily in a language, it should also be possible for a head to undergo focus fronting in that language. This is the null hypothesis, and I have argued for it in BG. I began by observing that the Part Aux construction in BG can be perceived as “marked”. Having shown that the responsible process is contrastive focalization which involves head movement, I am now in a position to explain the noted intuition as a semantic effect normally attributed to discourserelated operations. The LF effect can only arise if discourse is encoded in syntax. It is less controversial that focus fronting is a syntactic A-bar operation in the case of phrasal displacement. Since focus fronting of the complex verbal head underlies the perceived effect in the non-neutral Part Aux construction I conclude that head movement cannot be a post-syntactic/PF operation.

252

Mariana Lambova

Furthermore, this reasoning, based on the possibility for semantic import, entails extending the A/A-bar distinction to head movement. Phrasal A-bar movement, such as topicalization or focus fronting, is typically associated with a particular interpretation. It has also distinct locality conditions compared to A-movement. I have explicitly argued that the reordering in the nonneutral Part Aux construction in BG is not due to a “long” head movement of the contrastively focused Part.22 The issue is not trivial since locality has been the defining characteristics behind the A/A-bar distinction.23 It has been crucial to the analysis proposed here to recognize that two different processes underlie the Part Aux orders in BG. Furthermore, I correlate the noted semantic effect in the non-neutral case with verb movement beyond T°. Adverb placement clearly shows that the focused verb is higher in the tree. That is an important result with respect to establishing a parallel between phrasal and head movement. In both cases discourse-related displacement targets the functional layer above TP, i.e. the domain beyond the layer(s) responsible for licensing Nominative Case and tense.

5. Distinguishing between Triggers and Effects at the Interfaces I have also addressed the question of what triggers movement. The proposed analysis of the Part Aux construction in BG (and predicate reordering in general) motivates a complex interaction between syntax and phonology. While I show that the syntax of the neutral and the non-neutral case is different, I argue that there is no PF movement in the construction under consideration, contrary to what has been claimed before. In particular, adopting Bo‰koviç’s (2001) “weak phonology approach to the syntax phonology interface”, I make a case for PF filtering passively the syntactic output. I have argued that the Aux+Part output of syntax gets linearized as Part Aux orders via “scattered deletion” in PF but on two different motivations. In other words, I show that predicate reordering is derived through activation of lower copies of movement driven by PF constraints in both cases but the exact PF reason does not have to be the same in each case. This way I keep the neutral and the nonneutral cases apart, and still appeal to extra movement in the latter. Adopting standard Minimalist assumptions, movement is driven by syntactic feature checking. Adjunction of Part to Aux appears to be a language particular property motivated by a strong [+Aux] feature (see section 3.2). BG verbs rise overtly, as shown by adverb placement tests, a parameter usually attributed to a strong feature of T°. Here I blame overt focus fronting on a

On Triggers of Movement and Effects at the Interfaces

253

strong [+Foc] feature of the moved element but in Lambova 2003, I argue that the target, 6º, has also a a strong feature to check. Head movement in both verb raising and focus fronting is thus feature driven, and hence syntactic in nature. My analysis is consistent with Bo‰koviç’s (2001) system which allows PF to affect word order but without actual movement. In particular, word order changes arise through the mechanism of activating lower copies of movement motivated by PF considerations. It stands in contrast with accounts appealing to movement in the PF component (e.g. Marantz’s 1988 morphological merger or Zubizaretta’s 1998 prosodic movement) or in a specially designated module after syntax (e.g. Rivero’s 2000 recasting of LHM as stylistic movement) as well as those positing syntactic movement to meet certain requirements at the PF or LF interfaces (e.g. Rivero’s 1991 LHM or Diesing’s 1996 object shift). It is thus important not to confuse observable effects with their triggers. A note is order: I appeal to strength as motivation for syntactic movement whereby strength entails interpretability at the intereface(s). This imposes a constraint on a feature to be checked in the syntactic component or it will be uninterpretable in PF or LF and the derivation crashes (cf. Lasnik 1999). It is different from inducing syntactic movement in anticipation of satisfying certain properties at the interfaces. The latter motivation introduces globality in the system. On Minimalist assumptions, it is conceptually undesirable to invoke lookahead from syntax to PF or LF as this introduces redundancy or complexity in the system. This is all the more obvious given that there is a way to resolve conflicts at the interface(s). The present analysis also endorses the copy theory of movement which makes it possible to assume that a lower copy of a moved constituent is pronounced in certain well defined contexts. This allows us to maintain that movement belongs to syntax proper and avoid the potential lookahead problem of syntactic movement driven by PF considerations when accounting for more or less well-known rebracketing effects in the process of mapping from syntax to PF.

6. Conclusion I have shown that there are two types of Part Aux orders in BG, neutral and non-neutral. Importantly, the non-neutral one is available in both root and embedded clauses but the clitic nature of the present Aux obscures this fact for the matrix context. I have argued for a non-uniform analysis of the two cases, which is correlated with their associated different interpretations. Having shown that

254

Mariana Lambova

adjunction underlies verb raising in BG, I have proposed that the nonneutral Part Aux order is perceived as “marked” because the complex verb head undergoes focus fronting. As a result, I contend that head movement can have semantic import, and therefore should not be necessarily best viewed as a PF phenomenon.

Acknowledgements I acknowledge gratefully the guidance of my advisors, Îeljko Bo‰koviç and Howard Lasnik as well as the provocative but constructive criticism of an anonymous reviewer. The organizers of the Tilburg Workshop on Triggers deserve special thanks for providing the opportunity to share my ideas and for the financial support. The research reported here was first presented at HUMIT 2001. Thanks also to the audiences at these forums for insightful comments. The usual disclaimer applies.

Notes 1. The generalization is not quite correct. As I will show, it is not just any category (or “word”) but only the subject that cannot precede a Part Aux complex. Thus, while (i) is ungrammatical, the example improves to perfection when the subject is dropped: (i) *Toj tâkmo poluãil be‰e izvestieto. He just-then received Aux.PT.3P.SG letter-the “He had just RECEIVED the letter.” 2. Embick and Izvorski (1995) assume that the “optional” Part Aux order has semantic import suggesting that the fronting of Part is syntactic. Actually, the reordering will be shown to arise in PF in both cases but for a different reason. 3. This is a gross oversimplification of the intonation facts in BG. In particular, while I rely on the early and insightful observations of Penchev 1978, I am not attributing a specific meaning to any intonational contour. What is important is that a high tone at the beginning of the utterance contributes to the prominence of a contrastively focused element in BG. I am describing it in terms of higher pitch for ease of exposition disregarding the well-known fact that pitch fluctuations can signal emotional or other non-linguistic variables. I am not aware of any formal study of BG intonation though it has just been been brought to my attention that some work is being done (Bistra Andreeva, personnal communication).

On Triggers of Movement and Effects at the Interfaces

255

4. I cannot discuss these in detail here for reasons of space but I refer the interested reader to chapter 5 of Lambova 2003. 5. I will discuss the non-neutral reading later. 6. Claims in the literature that the split is possible with the past Aux (Izvorski 1993) or only with short adverbs such as veãe “already” (Krapova 1997, 1999) miss the prominence on Aux in these cases which I assume is an instance of verum focus. 7. On the free direction of adjunction Part can either left or right-adjoin to Aux. Left-ward adjunction produces a Part Aux order which can be problematic since Aux surfaces obligatorily to the left of Part with an overt subject: *SUBJ Part Aux. However, there is no problem for SC because Bo‰koviç shows that Aux can move out of the verbal complex (i). Assuming that sentential adverbs are TP-adjoined, Aux excorporates to AgrSo alone. (i) Jovan je nesumnjivo istukao Petra. [SC] Jovan.NOM Aux.PRES.3P.SG undoubtedly beaten Peter.ACC In BG, Aux and Part form an inseparable verbal complex. Specifically, a sentential adverb cannot intervene between Aux and Part (ii). Therefore, free direction of adjunctioon is a problem. (ii) *Ivan e nesâmneno nabil Petâr. Ivan Aux.PRES.3P.SG undoubtedly beaten Petâr

8. 9.

10.

11.

12. 13.

I will not try to argue either for or against movement from T° to AgrS° in BG. For ease of exposition, I assume with Chomsky (1998, 1999) impoverished clausal structure without Agr-projections. I leave it open whether the feature in question is located in the target or in the moved element. In Slavic, it is a standard assumption that heads right-adjoin (e.g. Rudin 1997 on Slavic clitcs). Chomsky (1994) leaves room for such a possibility unlike Kayne 1994. In addition, if adjunction is parametrizable, we would expect any given language to pick one option. As Rudin 1988 argues, rightward adjunction underlies the order of multiple fronted wh-phrases in BG. Embick and Izvorski 1995 imply that the fronted Part is discourse non-neutral (either backgrounded or focused), presumably as a result of a syntactic operation, but offer neither justification nor an analysis. I leave open the exact identity of the prosodic constituent in question. Below I will use “phonological word” (indicated as t) in a non-technical sense for concreteness. The relevant contour is actually a fall-rise-fall with the rise marking the high tone of the embedded contrastive element. In a neutral sentence, the nuclear pitch accent – a fall – is associated with the most deeply embedded element, the prepositional object, which is bearing the simple new information focus (shown in bold):

256

Mariana Lambova (i) I bought flowers for Mary.

14.

15.

16. 17.

Thus the tonal contour is similar in shape to that of (38a) but the pitch is less high. This description is a gross oversimplification in view of the extensive literature on English intonation. I am interested here in establishing a correlation between displacement and the domain of the tonal marking of contrast. Apparently, syntactic fronting contributes to prominence what relocation of the pitch accent does, and I will assume that neither of these means can be used redundantly. This is not to suggest that syntactic movement is driven by tonal requirements. Rather, I assume, as is standard, that contrastive focus is encoded in syntax, interpreted in LF where it can affect truth conditions and being realized in PF. Aux can receive contrastive stress but the verum focus construction is different. It is possible to have prominence on the adjective in contrast to (45): (i) VISOKA zena vleze v stajata. tall (foc) woman enter-PT.3P.SG in room-the “A TALL woman entered the room.”

18. This fact was noted in Wilder and Cavar 1994 in support of the claim that the verbal complex is in C°. 19. If sentential adverbs are TP-adjoined, as proposed in Watanabe 1993, (49) should involve more structure. In particular, I need an extra projection between 6P and TP, to which the adverb is adjoined: [6P Aux+Part [XP Aux+Part [TP adverb [TP Aux+ Part…]]]], where XP is AgrS P. Thus the two highest copies of Aux are above the sentential adverb. This is essentially evidence for split INFL in BG but I have ignored AgrSP for ease of exposition; see fn. 7.. 20. This predicate reordering is certainly not a construction specific phenomenon in BG. I refer the interested reader to chapter 6 of Lambova 2003. 21. They argue for this version of the cycle. In particular, they show that late insertion, which is possible in certain well-defined contexts, is correctly let in only on the strength-based definition(s). 22. In Lambova 2003, I show conclusive evidence that head movement has stricter locality than phrasal movement, i.e. even A-bar operations are not allowed to skip heads. 23. Compare Lasnik’s (2000) claim that A-movement does not leave traces. His argument is based on the unavailability of certain readings concerning scopal ambiguity. In other words, interpretational differences is what underlies the A/A-bar distinction. In this respect, recall that I appeal to the activation of A-movement trace/ copy in the case when a lexical subject surfaces below the verbal complex in the non-neutral Part Aux construction. Thus it cannot be that A-movement leaves no traces; rather, these traces are not significant for the LF.

On Triggers of Movement and Effects at the Interfaces

257

References Boeckx, Cédric and Sandra Stjepanoviç 2001 Heading toward PF. Linguistic Inquiry 32: 245–355. Bo‰koviç, Îeljko 1997 The Syntax of Non-Finite Complementation: An Economy Approach. Cambridge, MA: MIT Press. 2001 On the Nature of the Syntax Phonology Interface: Cliticization and Related Phenomena. London: Elsevier. Bo‰koviç, Îeljko and Howard Lasnik 1999 How strict is the cycle? Linguistic Inquiry 30: 691–703. Bowers, John 1993 The syntax of predication. Linguistic Inquiry 24: 591–656. Chomsky, Noam 1995 The Minimalist Program. Cambridge, MA: MIT Press. 2000 Minimalist Inquiries: The Framework. In Roger Martin et al. (eds.). Step by Step: Essays on Minimalism in Honor of Howard Lasnik, 89–155. Cambridge, MA: MIT Press. Diesing, Molly 1996 Semantic variables and object shift. In Höskuldur Thráinsson et al. (eds.) Studies in Comparative Germanic Synatx, vol. II, 66–84. Dordrecht: Kluwer. Embick, David and Roumyana Izvorski 1997 Participle auxiliary orders in Slavic. Formal Approaches to Slavic Linguistics 4: 210–239. Franks, Steven 1998 Clitics in Slavic. Position paper. Workshop on Comparative Slavic Morphosynatx, 5–7 June 1998, Indiana University. Downloadable at www.indiana.edu/~slavconf/linguistics/index.html). Halpern, Aaron 1995 On the Placement and Morphology of Clitics. Stanford, CA: CSLI Publications. Kiss, Katalin É. (ed.) 1995 Discourse-Configurational Languages. New York: OUP. Lambova, Mariana 2001 On A-bar movements in Bulgarian and their interaction. The Linguistic Review 18: 327–374. 2002 Head movement is syntactic: Evidence from Bulgarian. HUMIT Student Conference in Language Research 2001: 89–102. 2003 On information structure and clausal architecture: Evidence from Bulgarian. Ph.D. diss.: University of Connecticut.

258

Mariana Lambova

Lasnik, Howard 1999 On feature strength: Three Minimalist approaches to feature movement. Linguistic Inquiry, 30: 197–217. 2000 Chains of arguments. In Working Minimalism, Norbert Hornstein and Samuel Epstein (eds.), 189–216. Cambridge, MA: MIT Press. Marantz, Alec 1988 Clitics, morphological merger, and the mapping to phonological structure. In Theoretical Morphology: Approaches in modern linguistics, Michael Hammond and Michael Noonan (eds.), 253–270. San Diego, CA: Academic Press. Penchev, Jordan 1978 Osnovni intonacionni konturi v bulgarskoto izrechenie. Bulgarski Ezik 28: 293–302. Rivero, Maria-Luisa 1991 Long head movement and negation: Serbo-Croatian vs. Slovak and Czech. The Linguistic Review 8: 319–351. 2000 Finiteness and second position in long verb movement languages: Breton and Slavic. Syntax and Semantics 32: 295–323. Roberts, Ian 1991 Excorporation and minimality. Linguistic Inquiry 22: 209–218. Wilder, Chris and Dimitar Cavar 1994 Long head movement? Verb movement and cliticization in Croatian. Lingua 93: 1–58. Zubizaretta, Maria-Luisa 1998 Word Order, Prosody, and Focus. Cambridge, MA: MIT Press.

Scope Marking Constructions in Dayal-type Indirect Dependency Anikó Lipták

Abstract 1 The purpose of this paper is to present new data from the realm of scope marking constructions in Hungarian and other languages, in order to argue that these provide primary evidence for Dayal’s (1994, 2000) indirect dependency analysis. This analysis in turn sheds light on the nature of the overt wh-movement step that we find in the embedded clause of these constructions: it argues for treating that as an instance of run-of-the-mill whfronting, similar to what we find in matrix interrogatives. The empirical novelty to this effect comes from constructions involving embedded adjunct clauses: relative and noun-associate clauses, which, similarly to well-studied cases of argumental embedded clauses in languages with scope marking, can license embedded wh-items with matrix interpretation. It will be shown that unlike argumental embedded clauses, which in principle can lend themselves to various analyses, the newly discovered adjunct embedded clauses can only be analysed along the lines of Dayal’s proposal. This will have repercussions for the analysis of wh-movement inside the embedded clause of scope marking constructions: since the embedded clause according to this analysis is a full question, it contains an ordinary feature on a functional head that triggers the movement of the wh-element in overt movement languages. An exception to this is relative clauses, which will be shown to behave differently for independent reasons. The article is structured in the following way. Section 1 introduces scope marking constructions from a bird’s eye view and lists the characteristic properties of these constructions. Section 2 presents the standard scope marking data from Hungarian. The new data will be presented in section 3, while the theoretical impact as well as the subsequent analysis of these data will be handled in section 4. It will be shown that the current account of Hungarian scope marking (Horvath 1995, 1997, 1998, 2000) would face serious difficulties if it was extended to these data. Instead, a Dayal-type

260

Anikó Lipták

analysis is adopted. Section 5 is dedicated to the question what triggers the movement of the embedded wh-item. I will argue that just as in ordinary main clause interrogatives, the presence of a triggering feature is to be held responsible for overt fronting of the embedded wh-item. Evidence for this will come from Hungarian and crosslinguistic facts at the same time, involving Frisian and Slavic languages.

1.

Scope Marking Phenomena: Properties and Explananda

1.1. Properties of scope marking Since the early 1980’s, scope marking (or partial wh-movement) has always been on the generative research agenda for languages like German (van Riemsdijk 1983), Romani (McDaniel 1989), Hindi (Mahajan 1990), Hungarian (Marácz 1990, Horvath 1995), and many more. Consider a run-of-the-mill example of a scope marking question and an answer to it from German: (1)

Was1 denkt sie [wen1 Fritz t1 eingeladen hat ]? what thinks she whom Fritz invited has ‘Who does she think Fritz invited?’

(1A) Anna. ‘Anna.’ As (1) illustrates, scope marking involves a bi-clausal structure, with one wh-item in each clause. The wh-item in the matrix clause is referred to as the scope marker, and the one in the embedded clause as the contentful whphrase. A question like (1) is at first sight equivalent to a question with long whextraction (as the translation also indicates), which shows that in the particular example in (1), the matrix wh-item (was) is a placeholder element, while the embedded wh-item (wen) is what the question is about.2 Looking at scope marking constructions crosslinguistically, the following properties characterize them: (2)

(i) (ii)

there is a scope marker wh-item in the superordinate clause any wh-item can occur in the embedded wh-position (who, what, when, where, why, etc)

Scope Marking Constructions in Dayal-type Indirect Dependency

261

(iii) the answer given to a scope marking question specifies the embedded wh-item (cf. ex. (1A)) (iv) scope marking is unbounded; scope markers are usually spelled out in every intermediate clause, as illustrated in (3): (3)

Was denkt sie [was Hans gesagt hat [wen Fritz eingeladen hat]]? what thinks she what Hans said has whom Fritz invited has ‘Who does she think Hans said Fritz has invited?’ (v) the embedded clause hosting the contentful wh-item cannot be a selected question (matrix predicates like ask are not allowed), cf. (4):

(4)

*Was fragt sie [ wen Fritz eingeladen hat]? what asks she whom Fritz invited has ‘(lit.) ‘Who does she ask Fritz invited?’

Properties (i)-(v) will come handy in section 3, where new instances of scope marking constructions will be identified with the help of these. Other properties that characterize scope marking constructions, which I will have little to say about in this paper, are subject to variation across languages. In German or Hungarian, for example, the scope marker wh-item is overtly fronted, while in Hindi, it can also stay in-situ. Factive verbs can be matrix predicates in Hindi, but not in German. Similarly, yes/no questions are fine in the embedded clause in Hindi, but not in German or Hungarian.

1.2. Explananda Scope marking phenomena present theoretically interesting puzzles that are not easy to explain. One puzzle concerns the syntactic and interpretive relation between the scope marker and the embedded question word. Under the general assumption that only wh-items with matrix scope get answered, the fact that the embedded wh-item in scope marking constructions is filled in by the answer suggests that the embedded wh-item has matrix scope. However, its overt position does not reflect this: it is found in the embedded clause. Various solutions have been proposed to resolve this issue, arguing either for LF-raising of the embedded wh-item or the whole embedded clause (via expletive replacement) or for an underlying semantic mechanism that ensures matrix scope for the embedded question. In this paper I

262

Anikó Lipták

am going to argue that the data I discuss from Hungarian can be given a uniform analysis along the lines of the latter. Another syntactic puzzle (noted, among others by Simpson 1995; Fanselow and Mahajan 1996; Müller and Sternefeld 1996) concerns the motivation for the obligatory fronting of the embedded wh-item (in languages with overt fronting): what triggers this movement? Again, under standard assumptions wh-movement is triggered by a feature on interrogative complementizers – i.e. complementizers that head question clauses, over which the wh-item takes scope. In scope marking constructions, the embedded whitem ends up in the specifier position of a CP that is not a question (the embedded clause is not a -clause, witnessed by (3) above).3 A number of proposals have been suggested concerning this problem, with arguments to the effect that the attracting feature is in fact independent of question-semantics (Fanselow and Mahajan 1996); or that the condition that movement is attraction has to be relaxed (Müller 1998). Section 5 below will show that neither of these unorthodox approaches is necessary to account for the data: the embedded clause in scope marking constructions is a fully specified interrogative clause in the syntax (as well as the semantics): its interpretive properties and syntactic behaviour are exactly like that of matrix interrogative clauses. Before we can elaborate on these issues in more detail, an introduction to the data, both standard and new, is necessary. I turn to these in section 2 and 3 respectively.

2. Hungarian scope Marking In this paper I am concerned with the type of scope marking constructions that have been discussed in Horvath (1995) and subsequent work (Horvath 1997, 1998, 2000). I review the core data found in these articles in the present section, and I add new pieces of data to these in section 3.4

2.1. Sequential versus subordinated scope marking Initially, Hungarian scope marking constructions fall into two basic types: subordinated and non-subordinated scope marking constructions. Non-subordinated, or sequential, scope marking (the term coined by Dayal 1994) involves two juxtaposed, prosodically and syntactically autonomous clauses (É. Kiss 1987), like those in (5):

Scope Marking Constructions in Dayal-type Indirect Dependency

(5)

263

Mit gondolsz? Ki nyeri a versenyt? what-ACC think-2SG who win-3SG the competition-ACC ‘What do you think? Who will win the competition?’

(5A) Péter. ‘Péter.’ The order of these sentences is freely reversible. Yes/no questions are allowed in them: (6)

Mit gondolsz? Péter nyeri-e a versenyt? what-ACC think-2SG Péter win-3SG-Q the competition-ACC ‘What do you think? Will Péter win the competition?’

The “matrix” predicate cannot be negated: (7) *Mit nem javasolsz? Kit vegyünk fel? what-ACC not suggest-2SG whom hire-SUBJ-1PL PV ‘What don’t you suggest? Whom should we hire?’ The most frequent predicates to occur in these constructions are: gondol “think”, tud “know”, hall “hear”, mond “say”, szeretne “would like”, akar “want”, számít “count on”, ajánl “recommend”, javasol “advise”, jósol “predict”. Sequential scope marking is the most frequently occurring type among native speakers. 25% of the consulted speakers prefer these constructions to any other types reviewed below. Subordinated scope marking differs from non-subordinated constructions in that it clearly involves syntactic subordination. In Hungarian embedded argumental clauses subordination is indicated by the presence of hogy “that”, a finite complementizer (available both in indicative and interrogative clauses). An example of subordinated scope marking is given in (8), with its characteristic intonation pattern in (8’): (8)

MitŒl fél Mari, hogy ki lesz az igazgató? what-FROM fear-3SG Mari that who be-FUT.3SG the director (lit.) ‘What does Mari fear that who will be the director?’

(8A) Attól, hogy Péter. that-FROM that Péter-NOM ‘(Mari fears that it will be) Péter.’

264

Anikó Lipták

(8’) | 'MitŒl fél Mari |
hogy `ki lesz az igazgató? | 5 Unlike in sequential scope marking, the order of the clauses is not reversible (9), and yes/no questions are not allowed (10): (9) *Hogy ki lesz az igazgató, mitŒl fél Mari? that who be-FUT.3SG the director what-FROM fear-3SG Mari (10) *MitŒl fél Mari, hogy Péter lesz-e az igazgató? what-FROM fear-3SG Mari that Péter be-FUT.3SG the director (lit.) ‘What does Mari fear whether Péter will be the director?’ The matrix clause can be negated to some extent; subject to individual variation and choice of the predicate: (11) *MitŒl nem fél Mari, hogy ki lesz az igazgató? what-FROM not fear-3SG Mari that who be-FUT.3SG the director (lit.) ‘What does Mari not fear that who will be the director?’ Subordinated scope marking can occur in many environments. Both response-stance and non-stance predicates can take part in this pattern: elfelejt “forget”, emlékezik “remember”, észrevesz “notice”, rájön “find out”, megbán “regret”, említ “mention”, fél “fear”, megesküszik “swear”, megakadályoz “block”, (meg)jósol “predict”, kihirdet “make public”. Similarly, predicates taking subject clauses: zavar “bother”, kiderül “turn out” occur with this pattern. For completeness’ sake it has to be mentioned that subordinated scope marking actually comes in two flavours. Apart from the pattern described in (8)–(11), some matrix predicates also allow for what we could call a parenthetical subordinated scope marking construction, that is, where the matrix clause functions as a parenthetical.6 The parenthetical nature of this clause can be seen from the fact that the matrix clause has reduced prosodic and syntactic autonomy in these clauses: the left periphery of these clauses cannot contain a stressed element like focus or negation, as illustrated in (12); and the embedded complementizer can cliticize to the matrix verb, as shown in (13) (all these sentences have to be read by the intonation pattern in (13b)):

Scope Marking Constructions in Dayal-type Indirect Dependency

265

(12) a. *Mit gondolt MARI, hogy ki lesz az igazgató? what-ACC thought-3SG Mari that who be-FUT.3SG the director (lit.) ‘What did MARI think that who will be the director?’ b. *Mit nem gondolsz, hogy ki lesz az igazgató? what-ACC not think-2SG that who be-FUT.3SG the director (lit.) ‘What didn’t you think that who will be the director?’ (13) a. Mit gondolsz, hogy ki lesz az igazgató? what-ACC think-2SG that who be-FUT.3SG the director (lit.) ‘What do you think that who will be the director?’ b. | 'mit gondolsz hogy `ki lesz az igazgató |? Parenthetical scope marking also differs from standard cases of scope marking (i.e. (8)–(11) above) in that it accepts short answers, involving a single constituent only, while standard cases of scope marking usually trigger a full clausal answer (compare (8A)): (14) Mit gondolsz, hogy ki lesz az igazgató? what-ACC think-2SG that who be-FUT.3SG the director (lit.) ‘What do you think that who will be the director?’ (14A) Péter. ‘Péter.’ In the remainder of this paper I put cases of parenthetical scope marking aside, and concentrate on non-parenthetical ones only. The term subordinated scope marking will uniquely refer to these.

2.2. Argumental versus adjunct scope marking The previous section has introduced Hungarian scope marking constructions that occur with argumental embedded clauses. The examples above contained embedded clauses that function as internal arguments of the matrix verb. Next to these, one can find subject clauses as well, as (15) illustrates (Horvath 1995, 1997, 1998, 2000):

266

Anikó Lipták

(15)

Mi zavarta Marit [hogy kinek telefonáltál]? what bothered-3SG Mari-ACC that who-DAT phoned-2SG (lit.) ‘What bothered Mari that whom did you phone?’

(15A)

Az [hogy Péternek (telefonáltam)]. that that Péter-DAT phoned-1SG (lit.) ‘That I phoned Péter.’

Argumental clauses are not the only clauses, however, that can occur in scope marking. As Horvath (1995) has already shown, adverbial clauses can also be found: (16)

(16A)

Miért vagy dühös [mert kivel találkoztál]? what-FOR be-2SG angry because who-WITH met-2SG (lit.) ‘Why are you angry because you met whom?’ Azért [mert Péterrel találkoztam]. that-FOR because Péter-WITH met-1SG ‘Because I met Péter.’

The common property characterizing both argumental and adverbial embedded clauses in scope marking constructions is that both combine with a pronominal associate in the main clause. This pronominal associate shows up as a case-marked mi “what” in interrogative contexts and a case-marked az “that” in declarative contexts (e.g. mitŒl – attól in (8), (8A); mi – az in (15), (15A); miért – azért in (16), (16A)). The wh-variant of this sentential pronominal, a suitabl case-marked mi “what” is what plays the role of the scope marker in scope marking constructions. This, however, is not an absolute requirement: pronominal mi is not the only element that can be a scope marker in Hungarian. The next section will focus on new instances of scope marking constructions that involve full, lexical NP/DP phrasal scope markers. The examples will contain other types of adjunct embedded clauses.

Scope Marking Constructions in Dayal-type Indirect Dependency

267

3. New Ccases of Scope Marking: Adjunct Clauses Embedded under NP/DPs The previous section concerned itself with the various types of scope marking constructions that have been mentioned in the previous literature. The present section shows that subordinate scope marking has a much wider empirical base than previously recognized: it occurs with noun-associate clauses and relative clauses as well. These will be introduced in sections 3.1. and 3.2. in turn. The existence of these latter types is of high theoretical significance because they present a challenge to extant analyses of Hungarian scope marking (Horvath 1995, 1997, 1998, 2000), as will be shown in section 4. While Horvath’s account is irreconcilable with these facts of Hungarian, these constructions can easily be accounted for by Dayal’s (1994, 2000) analysis.

3.1. Scope marking with relative clauses Relative clauses in Hungarian can be headed relatives or free relatives. The type of relative clauses that are important for purposes of illustrating scope marking data are the headed restrictive relatives, which can be either headed by a pronominal (17) or a full NP/DP (18) (in bold): (17) Az megy át a vizsgán [aki 20 pontot szerez ]. that go-3SG PV the exam-ON who-REL 20 point-ACC score-3SG ‘The person who scores 20 points passes the exam.’ (18) Az a diák megy át a vizsgán [aki 20 pontot that the student go-3SG PV the exam-ON who-REL 20 point-ACC szerez ]. score-3SG ‘The student who scores 20 points passes the exam.’ Under standard assumptions about relativization (not adopting Kayne (1994)), (17)–(18) conform to the following schematic structure: (17’/18’) [DP az (a diák) [CP aki 20 pontot ér el ]] (17’/18’) underlies scope marking constructions as well. In these cases, we find two wh-elements: the embedded relative clause contains a wh-item and the head of the relative clause must be or must contain a wh-phrase, as well:

268

Anikó Lipták

(19) Kii megy át a vizsgán [akii hány pontot who go-3SG PV the exam-ON who-REL how many point-ACC szerez ]? score-3SG (lit.) ‘Whoi, whoi scores how many points, passes the exam?’ (intended) ‘How many points does one have to score to pass the exam?’ [akii hány (20) Melyik diáki megy át a vizsgán which student go-3SG PV the exam-ON who-REL how many pontot szerez]? point-ACC score-3SG (lit.) ‘Which studenti, whoi scores how many points, passes the exam?’ (intended) ‘How many points does a student have to score to pass the exam?’ In these examples we are dealing with two questions: the matrix question ranges over individuals (ki “who” or melyik diák “which student”) and the embedded question ranges over the number of points (hány pontot “how many points-ACC”). The interpretation of these questions is clearly reflected by the particular answers they trigger: 20 pontot szerez ]. / *Mari. (19A) Azi [akii that who-REL 20 point-ACC score-3SG / Mari ‘Who(ever) scores 20.’ / ‘Mari.’ 20 pontot szerez ]. / *Mari. (20A) Az (a diák)i [akii that the student who-REL 20 point-ACC score-3SG / Mari ‘The student who scores 20 points.’ / ‘Mari.’ As we can see in these examples, the answer necessarily has to specify the embedded question, i.e. the number of points that need to be scored for passing the exam. An answer naming particular individuals who pass the exam is not satisfactory. The intonation contour of these complex constructions is parallel to that of argumental subordinated scope marking constructions, as was illustrated in (8’) above: (20’) | 'Melyik diák megy át a vizsgán, |
aki `hány pontot szerez? |

Scope Marking Constructions in Dayal-type Indirect Dependency

269

The constructions in (19)–(20) comply with all criteria we identified in section 1 as defining properties of scope marking. There is a scope marker in them (property (i)); the choice of the embedded wh-phrase is free (property (ii)); the question is answered by providing a value for the embedded whitem (property (iii), see (19A), (20A)). The relation is unbounded, it can involve multiple layers of embedding (property iv): (21) Melyik diák megy át a vizsgán, [aki which student go-3SG PV the exam-ON who-REL milyen könyvbŒl tanul [amit ki írt ]]? what book-FROM study-3SG what-REL.ACC who wrote-3SG (lit.) ‘Which studenti, whoi studies from what kind of bookj, thatj who wrote, passes the exam?’ The ban on selected interrogative subclauses (property (v)) is satisfied vacuously, since relative clauses are never selected to be interrogative. In fact, not only are they never selected, they cannot ever contain a wh-item in any other constructions but the constructions under investigation here. If the matrix clause was not an interrogative clause, the relative clause would fail to license a question: (22) *Az megy át a vizsgán [aki hány pontot szerez ]? that go-3SG PV the exam who-REL how many point-ACC score-3SG (lit.) ‘Who(ever) scores how many points, passes the exam.’ The matrix interrogative clause has to comply with one requirement: the whphrase in it has to either correspond to the head of the embedded relative clause or ask for a property that is also spelled out in the relative clause. The following two examples illustrate these points: (23) *Hány diáki megy át a vizsgán [akii how many student go-3SG PV the exam-ON who-REL hány pontot szerez ]? how many point-ACC get-3SG (lit.) ‘How many studentsi, whoi score how many points, pass the exam?’

270

Anikó Lipták

(24) Kineki a diákja megy át a vizsgán, [akii who-DAT the student-POSS.3SG go-3SG PV the exam-ON who-REL hány pontot szerez ]? how many point-ACC get-3SG (lit.)‘Whosei studentj, whoi/*j scores how many points, passes the exam?’ (intended) ‘How many points does a teacher have to score to pass a student?’/ ‘*How many points does a student have to score to pass the exam?’ In (23) we see that although the matrix and the embedded wh-phrases are identical in meaning (hány “how many”), the sentence fails to be interpretable, because the relative clause is not construed as a numeral modifier of students. In (24), the relative clause has to be interpreted as a modifier over the smallest wh-phrase, kinek “who-DAT”, and not the larger phrase kinek a diákja “whose student-NOM”, even though the resulting meaning is pragmatically unlikely. This shows that in case the matrix wh-phrase can be found in a referentially independent larger NP/DP, the relative clause in scope marking has to associate with the smallest wh-phrase possible, as a scope marker. To summarize, this section has shown beyond doubt that the constructions in (19)-(20) instantiate an example of scope marking, namely scope marking with an adjunct embedded clause. The semantic and intonational properties of these clauses are exactly parallel to well-established cases of scope marking with argumental embedded clauses. The scope marker is (or is found within) the head of relativization, and the embedded clause is contained inside the relative clause. The answer necessarily has to fill in the embedded wh-variable.

3.2.

Scope marking with noun-associate clauses

The behaviour of relative clauses in scope marking is fully paralleled by adjunct-type noun-associate clauses in Hungarian.

3.2.1. Noun-associate Clauses in Hungarian Hungarian noun-embedded clauses have been argued to be of two kinds: arguments or adjuncts (Kenesei 1992). As far as their internal structure is concerned, they both look the same: they are run-of-the-mill finite embedded clauses, introduced by the finite complementizer hogy “that”.

Scope Marking Constructions in Dayal-type Indirect Dependency

271

The difference between the argumental and the adjunct type can be seen from the kind of DP-structure these clauses occur in. Argument clauses, which are selected by a derived event/process nominal, need case. Given that they cannot bear case (Stowell 1981), they have to be linked to a casemarked expletive. This expletive, a demonstrative pronominal, occupies the only available case position of a possessive DP, the dative case position Sp,DP (Szabolcsi 1992): (25) a. annak belátása [hogy…] that-DAT realization-POSS.3SG that ‘the realization that…’

argumental N-clause

b. [DP annaki [D0 a [NP belátása [CP hogy …]i]] Due to this structural requirement, nouns with an argumental CP cannot have other possessors: (25) c. *Péternek a belátása [hogy…] Péter-DAT the realization-POSS.3SG that ‘Péter’s realization that…’ Adjunct noun-embedded clauses, on the other hand, do not have to comply with such a restriction: in this case the embedded CP is not a selected argument, but an adjunct that is associated with the lexical-semantic frame of the (simplex or result) nominal. These clauses can occur in NP/DPs with overt possessors (26): (26) a. az az indok [hogy …] adjunct N-clause that the argument that ‘the argument, that…’ b. [DP az [NP az indok [CP hogy …]]] c. Péternek az az indoka [hogy…] Péter-DAT that the argument-POSS.3SG that ‘Péter’s argument, that…’

3.2.2. Scope Marking with Adjunct Noun-associate Clauses Scope marking with adjunct-type noun-associate clauses are grammatical for all speakers of Hungarian, while argumental-type embedded clauses

272

Anikó Lipták

show some variation: many informants found them just as good as adjuncttype embedded clauses; several of them, however, found them degraded or ungrammatical. Therefore, in the following I illustrate the patterns with adjunct-type noun-associate clauses only. Scope marking with noun-associated adjunct clauses is illustrated in (27): (27)

kapott Péter [hogy hova kell mennie]i? Milyen üzeneteti what message-ACC got-3SG Péter that where need go-INF-3SG (lit.) ‘What message, where does he have to go, did Péter get?’

kapta [hogy a rendŒrségre (27A) Péter azt az üzeneteti Péter that the message-ACC got-3SG that the police-TO kell mennie]i need go-INF-3SG ‘Péter got a message that he has to go to the police force.’ Just like with relative clauses, the matrix wh-phrase is a “what (kind of)” question that asks for the same kind of property that is also expressed by the embedded clause. As far as intonation is concerned, these sentences are most frequently pronounced with the same intonation contour as argumental or relative clauses above: (27’) |'Milyen üzenetet kapott Péter,

|
hogy `hova kell mennie? |

(27) also complies with all criteria for scope marking established in section 1, namely the presence of a scope marker (milyen üzenetet “what messageACC”); the choice of the embedded wh-phrase; the required answer pattern (property (i), (ii), (iii)). The unbounded nature of the constructions is illustrated in (28): (28)

Milyen üzenetet kaptál, [hogy melyik állítást what message-ACC got-2SG that which claim-ACC ellenŒrizzük [hogy melyik üzem nyereséges ]]? check-IMP-1PL that which factory profitable (lit.) ‘What message, that we should check which claim, that which factory is profitable, did you get?’

The nominal with which the embedded clauses are associated has to be a “what kind” wh-phrase in each clause. The ban on selected -clauses

Scope Marking Constructions in Dayal-type Indirect Dependency

273

is complied with as well. If the embedding noun requires a question, like the noun kérdés ‘question’, scope marking is unavailable: (29) *Milyen kérdéssel foglalkoztak [CP+whhogy mire kell a pénz]? that what-ON need the money what question-WITH dealt-3PL (lit.) ‘What question, that do they need the money for what did they treat?’ This section has shown that just like relative clauses, adjunct noun-associate clauses are capable of hosting a wh-phrase with matrix interpretation as long as the nominal they are associated with is a “what kind” wh-expression. These constructions show the same properties as standard cases of scope marking, and therefore should be considered as such.

3.3. Empirical summary: the patterns of Hungarian subordinate scope marking On the basis of the discussion in the previous sections it can be concluded that Hungarian subordinated scope marking occurs in Hungarian across the following constructions types, both argumental (a) and adjunct (b),(c),(d) ones: (a) (b) (c) (d)

argumental embedded clauses associated with az/mi adverbial clauses associated with az/mi relative clauses associated with an NP/DP adjunct noun-associate clauses with an NP/DP

– ex. (8), (15) – ex. (16) – ex. (19), (20) – ex. (27)

Of these constructions, the literature (Horvath 1995, 1997, 1998, 2000) has only discussed type (a) and mentioned in passing type (b). These are the ones that occur with a uniform scope marker mi pronominal. (c) and (d), as was shown above, associate with a full NP/DP. The correct empirical generalization therefore seems to be that Hungarian subordinated scope marking occurs with any embedded clause that is associated with a (pro)nominal constituent, whose meaning it specifies. The schematic representation of scope marking constructions is the structure in (30): (30) [CP

wh-(pro)nouni

[CP … wh … ]i ]

274

Anikó Lipták

The next section reviews the literature on the analysis of scope marking, and puts forward the claim that relative clauses and adjunct noun-associate clauses (type (c) and (d)) are to be analysed along the lines of Dayal’s indirect dependency approach, which, apart from being the only analysis capable of dealing with these data, can be easily extended to cases of argumental scope marking and adverb scope marking as well (type (a) and (b)).

4. The Analysis of Adjunct Scope Marking Scope marking constructions have been analysed in various ways in the literature. These can be classified into two general types of approaches: the direct and the indirect dependency approaches. The two approaches differ in the kind of relationship they ascribe to the embedded wh-item and the matrix scope marker. In the direct dependency, the embedded wh-item directly replaces the scope marker at LF. The indirect dependency approaches, which have two subtypes as well, argue either that there is a syntactic link between the scope marker and the embedded clause, or a semantic mechanism that links these two. In this section I briefly sketch each approach and show whether or not it suits the newly discovered cases of Hungarian scope marking. As it turns out, only one type of approach can account for these: the semantic indirect dependency account.

4.1. Direct dependency approach According to the advocates of the direct dependency approach (van Riemsdijk 1983, McDaniel 1989, Cheng 2000, among others) the embedded wh-item is directly linked to the matrix wh-item in the syntax and semantics, via LF-expletive replacement of the sort well-known from there-expletive constructions. The scope marker is an expletival placeholder for the embedded wh-item in the main clause: (31) S-str [CP+wh was LF [CP+wh wh-phrase

[CP–wh wh-phrase [CP–wh ti

[IP … ti … ]]] [IP … ti … ]]]

Expletive replacement at LF gives rise to a structure that is parallel to cases of long extraction, and is forced by Full Interpretation and the Wh-criterion, which apply at LF: at the level of interpretation the embedded clause cannot

Scope Marking Constructions in Dayal-type Indirect Dependency

275

contain any wh-element, since it is not an embedded question. What triggers the partial movement of the contentful wh-item to the embedded Sp,CP position receives various answers in the different versions of the direct dependency approaches. The general unavailability of this approach to the cases of Hungarian scope marking under discussion can easily be seen from the fact that these constructions constitute islands for extraction (CNPC): megy át a vizsgán [aki ti szerez ]? (32) *Hány pontoti how many points-ACC go-3SG PV the exam-ON who-REL score-3SG (intended) ‘How many points does one have to score to pass the exam?’ The same has been noticed about subject clauses and adverbial clauses as well (Horvath 1995): scope marking, unlike long extraction, is possible across islands (CED-effects). This militates against an analysis in terms of LF-long extraction.

4.2. The syntactic indirect dependency approach In contrast to the direct dependency approach, the indirect dependency approaches posit an indirect relationship between the wh-items: the scope marker is directly linked to the whole embedded clause. There are two types of ideas about what provides the link between the scope marker and the embedded clause: in some analysis the link is syntactic, in others it is semantic in nature. In this section I review the syntactic accounts (which were called mixed approaches in Lutz, Müller and von Stechow 2000). Apart from Mahajan (1990) and Fanselow and Mahajan (2000), the extant analysis of Hungarian, Horvath (1995, 1997, 1998, 2000), belongs to this type of approach as well. In my discussion I focus on Horvath’s analysis only.

4.2.1. Horvath (1995, 2000) Account of Hungarian Scope Marking In Horvath’s analysis, the scope marker is a (wh-)pronominal anticipatory pronoun, generated in A-position (AgrP in Horvath 1997); associated with the embedded CP proposition, bearing the case that is assigned to the CP and which the CP cannot carry due to the case resistance principle (Stowell

276

Anikó Lipták

1981). In scope marking constructions, just as in any case of clausal subordination, the subordinated CP needs to “meet” its case before the end of the derivation (to satisfy Full Interpretation). To achieve this, the CP has to adjoin the sentential pronominal at LF: (33) [CP [FocP mi+case [AgrP tj [CP [FocP wh-phrasei [IP… ti … ]]] ]]] 144444424444443 :: ! #z-------m !LF z---------------------m

The LF movement step of clausal pied-piping can only be successful, according to Horvath, if the CP and the sentential expletive match in wh-features (cf. non-distinctness, Chomsky and Lasnik 1991). The scope marker is a item, which then requires the embedded clause to have a matching feature as well. This feature will have to come from the embedded wh-item, since in scope marking constructions the embedded clause is never selected to be a question, and consequently it does not possess any inherent -feature. In other words, in order for proper matching to take place, there must be a free, transmittable -feature available in the embedded clause, that can percolate up onto the embedded CP. This requires the presence of a whphrase in the embedded Sp,CP (Horvath glosses over the fact that the overt position of wh-items is lower than CP in Hungarian). After -feature transmission, the wh-item looses its wh-hood, and its operator nature. As a “disarmed” wh-item, it does not cause any violation of the Wh-criterion. This mechanism explains why scope marking is only possible if the WhCriterion does not otherwise require a wh-operator in Sp,CP, i.e. only with verbs which do not select questions (cf. ex. (4)): az igazgató]? (34) *Mit kérdeztél [CP+wh hogy ki lesz that who be-FUT.3SG the director what-ACC asked-2SG (lit.) ‘What did you ask that who will be the director?’ The right interpretation of scope marking constructions (i.e. a meaning similar to long wh-questions) comes about in the following way. As a result of case-driven expletive replacement by the whole embedded CP, the embedded wh-item acquires matrix scope at LF due to its structural position: it will ccommand out of the specifier of a specifier (Kayne 1994), thereby taking scope over everything in the matrix clause. (35) [CP+wh [CP+wh whi+wh [C’ C–wh [IP … ti …]]]-mi [AgrP tj …] ]

Scope Marking Constructions in Dayal-type Indirect Dependency

277

Given that the embedded CP moves as a whole, Horvath’s analysis predicts that island violations do not occur, which is borne out by the facts. 4.2.2. Adjunct Scope Marking: A Problem for Horvath’s Account Section 3 has shown that the empirical base of subordinate scope marking is much larger than previously thought. It occurs with relative and noun-associate adjunct embedded clauses as well. It is easy to see that these newly discovered cases of scope marking do not lend themselves to an analysis that was sketched in the previous section. As we have seen, Horvath’s account is crucially based on expletive replacement, and the analysis of scope markers as expletives. While this is certainly an a priori possible stand for the analysis of embedded clauses that combine with a uniform pronoun mi “what”, the same analysis cannot be carried over to relative and nounassociate clauses for the simple fact that these are never associated with expletival elements. The scope markers in these constructions are not (wh-) expletives, but full-blown argument NP/DPs, with a lexical meaning of their own. Therefore, an analysis in terms of expletive replacement by the embedded CP at LF is not tenable: (36) [CP [FocPmelyik diáki [DP ti [CP-wh aki [FocPhány pontoti [IP… ti … ]]] ]]] 1444444424444443 :: ! # z-------m !LF z------------=-----------m

Note that this is true even if expletive replacement is taken to be adjunction of the embedded CP to the matrix pronominal, as argued by Horvath, with reference to case requirements of the embedded clause. Such an adjunction step would be totally unmotivated in the case of relative and noun-associated embedded clauses, as these clauses, being adjuncts, are not in need of case. In the next section I turn to the only account that can handle the newly found Hungarian facts of scope marking: Dayal’s (1994, 2000) indirect dependency.

4.3.

Semantic Indirect Dependency Approach (Dayal 1994, 2000)

4.3.1. Outline of the Analysis The other, semantic type of indirect dependency approach, represented by works of Dayal (1994, 2000), argues for an underlying semantic link between

278

Anikó Lipták

the scope marker and the embedded clause. The scope marker in this account is a standard argumental wh-phrase, which quantifies over propositions. The embedded clause, a full-blown question, restricts the domain of propositions that the scope marker quantifies over. For a semantic representation, Dayal follows Hamblin (1973) in taking questions to denote the set of possible answers to them. Wh-expressions are existential quantifiers whose restriction is either implicit or provided by some overt restriction. In this view, the main clause interrogative in (8), repeated from above (8)

MitŒl fél Mari, hogy ki lesz az igazgató? what-FROM fear-3SG Mari that who be-FUT.3SG the director (lit.) ‘What does Mari fear that who will be the director?’

has the following logical representation: hpšq[p a proposition & p=^fear (Mari,q)]. Dayal assumes furthermore that quantification is always restricted in natural languages, thus also with quantification over propositions; the overt or covert restrictor of the matrix propositional quantifier can be represented by a variable T: hpšq[T(q) & p= ^fear(Mari,q)]. The meaning of the embedded clause is hpšx [p= ^will-be-director (x)], which can be made the restrictor T in the interpretation of the matrix question. The end result is: hpšq[šx [q=^will-be-director (x)] & p=^fear(Mari,q)]. In informal rephrasing, (8) denotes the following question: “what proposition p, such that p is a possible answer to ‘who will be the director?’ is such that Mari fears p?” Possible answers to the question “who will be the director” are propositions like Péter will be the director; Anna will be the director; Hugo will be the director. From this set of propositions, (8) will ask for the one that Mari fears. The embedded clause under this account crucially has to be interpreted as a question, because only questions can function as restrictions over a propositional variable scope marker in the matrix clause. This semantic mechanism underlies all scope marking constructions according to Dayal. The syntax of scope marking constructions can be different from language to language. The relation between the matrix wh-item and the embedded clause can range from a loose juxtaposition to a real syntactic dependency. Crucial to this analysis is the treatment of sentential pronominals as full arguments, which follows the spirit of a considerable amount of syntactic proposals (Rosenbaum 1967; Bennis 1986; É. Kiss 1987; Torrego and Uriagereka 1989; Müller 1995; Moro 1997; Stepanov 2000) and the analysis of the embedded clause as a syntactic adjunct, a semantic restrictor over the matrix argument nominal.

Scope Marking Constructions in Dayal-type Indirect Dependency

279

4.3.2. Adjunct Scope Marking in Dayal’s Account The above sketched analysis suits Hungarian double question scope marking like a glove: as we have seen, in this language scope marking does not only occur with standard sentential subordination, where the expletive-replacement analysis is in principle available, but also with other types of embedding, where expletive replacement is completely out of the question. Relative and noun-associate clauses do not combine with expletives, rather, they appear with full NP/DPs. On the other hand, their role is exactly as described by Dayal’s account: to provide a restriction over the matrix (pro)nominal constituent, the NP/DP they modify. The particular interpretation of adjunct-type scope marking constructions then must also be due to a Dayal-kind semantics: the scope marker is restricted by the embedded clause, which explains why it has to be answered in terms of the restrictive embedded clause. If one takes a noun phrase and provides it with a full clausal restriction, it is expected that in case the clausal restriction contains a wh-question, the noun phrase as a whole contains a variable, i.e. is questioned as well.

4.3.2.1. Relative Clauses The most obvious clausal restrictions in languages are restrictive relative clauses. They form a restriction on the denotation of the noun head, and they are syntactically subordinated to their nominal head which takes scope over them. If they contain a question, that question restricts the denotation of the nominal head as well, which makes it necessary that this nominal be quantificational (and not referential). This explains why the presence of a wh-item inside the relative forces the nominal head to appear as a wh-item as well (cf. 22, repeated here from above): (22) *Az megy át a vizsgán [aki hány pontot szerez ]? that go-3SG PV the exam who-REL how many point-ACC score-3SG (lit.) ‘Who(ever) scores how many points, passes the exam.’ Adopting Dayal’s semantics for a question like (20), and the semantics for ordinary restrictive relative clauses, we end up with the following picture. The matrix wh-phrase melyik “which” has two restrictions: the NP diák “student” and the restrictive relative clause (37).

280

Anikó Lipták

(20) Melyik diáki megy át a vizsgán [akii which student go-3SG PV the exam-ON who-REL hány pontot szerez]? how many point-ACC score-3SG (lit.) ‘Which studenti, whoi scores how many points, passes the exam?’ (intended) ‘How many points does a student have to score to pass the exam?’ (37) Melyik diáki megy át a vizsgán

aki hány pontot szerez?

i : z-------------------------m

restriction The interpretation corresponds to (38): (38) the set of propositions p, such that for some individual x such that x is a student and x is in the set of individuals who have the property related to a proposition, namely the proposition who scored n-many points, p = x passed I refer the exact logical formalisation of this question to further research, as it is much more complex than the formalisation of argumental scope marking. Due to the fact that a question (i.e. a set of propositions) cannot directly restrict an individual variable, the formalisation might involve some extra semantic tools like choice functions (Veneeta Dayal, p.c.)

4.3.2.2. Adjunct Noun-associate Clauses I propose that the analysis of relative clauses in the previous section carries over in all relevant respects to adjunct noun-associate clauses. As these instances of embedded clauses have been argued to be clausal adjuncts (Stowell 1981; Grimshaw 1990; Kenesei 1992, 1994), these can be treated in the same way as relative clauses for our purposes. That is, the embedded clause functions as an adjunct modifier that restricts the matrix nominal expression, which is a full argument: (27) Milyen üzenetet kapott Péter [hogy hova kell mennie]? what message-ACC got-3SG Péter that where need go-INF-3SG (lit.) ‘What message, where does he have to go, did Péter get?’

Scope Marking Constructions in Dayal-type Indirect Dependency

(39) Milyen üzenetet kapott Péter

281

hogy hova kell mennie?

: z----------------------m

restriction (40) the set of propositions p, such that for some individual x, such that x is a message and x is in the set of individuals who have the property related to a proposition in T, namely where Péter has to go, p = Péter got x.

4.4. The analysis of argument scope marking This section has shown that the newly identified cases of scope marking constructions receive a straightforward analysis in Dayal’s framework. They refute any other analysis, in terms of LF expletive replacement, due to the fact that they do not contain any expletives. The question arises whether one would be justified in treating all instances of subordinated scope marking in Hungarian under the semantic indirect dependency approach. The strongest (and therefore the most interesting) claim would be to say that all constructions are alike, and, generalizing to the “worst case”, they all involve adjunct embedded clauses, which must receive an analysis along the lines of Dayal’s framework. This would argue for treating sentential pronominals as arguments of the matrix predicate and the embedded clauses as adjuncts, modifying the argumental pronominals (É.Kiss 1987), as in Dayal’s original proposal. As recent developments have shown, however, (Lutz, Müller and von Stechow 2000), it is clear that scope marking constructions, and their analysis, might differ across languages: properties of Hindi scope marking, for example, are easier to explain with the indirect dependency approach, while those of German with the direct or the syntactic indirect approach. Moreover, different types of scope marking can exist even within one and the same language: a case in point is Passamaquoddy, which has two distinct scope marking constructions, with different scope markers and different verbal agreement patterns (Bruening 2001, to appear). Space limitations block me to enter a discussion here about whether a uniform analysis for all Hungarian facts is desirable or not. Pending futher, more detailed comparison of argumental and adjunct scope marking constructions and their detailed properties, I refer this topic to future research. It is clear, however, already at this point, that in case such a uniform analysis is indeed warranted, no theory to date except Dayal’s account could cover all data.

282

Anikó Lipták

5. The Trigger of Embedded Overt Movement After spelling out the analysis of the Hungarian facts in the previous section, the present section returns to the problem of the triggering factor, as introduced at the outset of this paper.

5.1. The application of Dayal’s approach As already mentioned, scope marking constructions involve an overt whmovement step in the embedded clause (except in languages that allow for wh-in-situ in general). For illustration, recall (1) from German. The embedded wh-item obligatorily has to undergo fronting, it cannot stay in situ: (1)

Was1 denkt sie [wen1 Fritz t1 eingeladen hat ]? what thinks she whom Fritz invited has ‘Who does she think Fritz invited?’

(41) *Was1 denkt sie [Fritz wen eingeladen hat ]? what thinks she Fritz whom invited has Hungarian facts are fully parallel to the German ones, they involve obligatory fronting in standard argumental scope marking (15)/(42): (15)

Mi zavarta Marit [hogy kinek telefonáltál]? what bothered-3SG Mari-ACC that who-DAT phoned-2SG (lit.) ‘What bothered Mari that whom did you phone?’

(42) *Mi zavarta Marit [hogy telefonáltál kinek ]? what bothered-3SG Mari-ACC that phoned-2SG who-DAT These Hungarian facts, however, do not instantiate wh-fronting for feature checking/licensing, due to the fact that in this language, overt fronting of wh-items happens for focusing reasons. Wh-items, including those in echo questions, behave like contrastive focus constituents and they occupy Sp,FocP, a relatively low functional projection in the left periphery, embedded under CP, TopP(s) and QP(s) (Horvath 1981, É.Kiss 1987): (43)

[CP

[TopP* [QP* [FocP Focº [… ]]]]]

Scope Marking Constructions in Dayal-type Indirect Dependency

283

Movement to Sp,FocP is motivated by a feature on Focº and on the wh-phrases. Next to a focus feature, wh-items also possess a feature that needs checking in the syntax, as was shown by Horvath (1981) and Lipták (2001), this however, does not motivate overt movement distinct from focusing. For this reason, the Hungarian facts above are not indicative of the presence or absence of any particular wh-syntax of scope marking constructions. I take it, however, that even though there is no overt whmovement involved in Hungarian questions, covert wh-movement (feature movement) does happen to the interrogative complementizer Cº (Lipták 2001), and thus, at LF, the Hungarian facts in (15)/(42) are parallel to the German facts in (1)/(41), where overt wh-movement is visible. How to explain German (1)/(41) then? Under the assumption that the embedded wh-item is only interpreted in the matrix clause, the embedded overt movement step remains unexplained: wh-movement is usually triggered by a feature on interrogative complementizers within which the wh-item takes scope. If we were to assume, with the direct or the syntactic indirect approach, that the embedded wh-item does not take scope in the embedded clause, but gains matrix scope at LF (via expletive replacement), obligatory fronting remains mysterious: there is no reason for the wh-item to leave its base-position its overt syntax. The embedded Cº head in German does presumably not possess a feature that could trigger the movement of the embedded wh-items to Sp,CP. With the indirect dependency analysis of scope marking constructions in place, it is not difficult, however, to identify the trigger for the movement of the embedded wh-item. The wh-item has to front in the embedded clause for the same reason why it has to in matrix interrogative clauses: to check a feature on a high left peripheral head. The embedded clause in standard scope marking constructions for all intends and purposes behaves like a full-fledged matrix question: it is interpreted as a normal question, so it is expected that its syntax is also that of a normal question. It differs from matrix questions in one respect only: it has subordinated syntax (presence of a complementizer, embedded word order, etc). If my analysis is on the right track, these constructions qualify as an embedded root phenomenon in this particular respect: they contain a matrix interrogative that occurs with subordinated syntax. Following Dayal (1994, 2000) I therefore put forward the claim that in scope marking constructions of the type examined in this paper, the embedded question is a fully specified question-clause. It contains a attracting (or unvalued) feature on Cº or another functional head that triggers overt movement of the embedded wh-phrase.

284

Anikó Lipták

Note that this idea is by no means in conflict with the fact that scope marking does not occur with embedded clauses that are selected questions (as shown above in (4), (34)). This is because the morphosyntactic feature that triggers the movement of wh-phrases in the syntax is never the selectional feature, in any language, in any construction. This can be seen, among other things, from the fact that selectional features are not always present in all cases of embedded interrogatives that feature wh-movement. Two illustrative examples are given in (44)-(45), which show that embedded interrogativity can be licensed by mood or negation in the matrix clause, also with matrix predicates that otherwise do not select interrogative clauses ((45) is taken from Adger and Quer (2001)): (44) a. *I am aware who killed Mary. b. I am not aware who killed Mary. (45) a. *Julie admitted/heard/said if the bartender was happy. b. Julie didn’t admit/hear/say if the bartender was happy. This shows that the feature that triggers syntactic movement is not a selectional one, but a morphosyntactic feature that can be independent of selectional properties. Its presence cannot be deduced or derived from selectional features in a one-to-one manner. I believe that scope marking constructions under Dayal’s analysis are similar to cases like (44)–(45), in that they contain a attracting feature on a functional head in an unselected embedded clause: (46) Was denkt sie [CP wen Fritz eingeladen hat ]? Although the embedded feature on Cº is not related to selectional properties, it is not an “out of the blue” feature: its presence on embedded C is linked to the presence of a feature on the matrix “scope marker”, a wh-constituent. This is possible due to feature-sharing between these two constituents, that take part in an underlying subject-predicate (small-clause) relation. (47) [denkt sie …

[SC [was][CP wen Fritz eingeladen hat ]]

This configuration, in which the scope marker and the embedded clause form one constituent at some level of represenation, was proposed by, among others, in É. Kiss (1987) for Hungarian sentential complementation

Scope Marking Constructions in Dayal-type Indirect Dependency

285

and in Herburger (1994) for German. In this configuration wh-feature sharing is possible, similarly to other types of feature-sharing processes (agreement, case). As a result, the embedded CP can inherit the feature from the matrix wh-phrase. This in turn explains why we find overt embedded movement in these clauses. As we can see, Dayal’s approach, coupled with regular assumptions about subject-predicate relationships, successfully takes care of the problem of triggers in a rather straightforward manner and without unorthodox auxiliary assumptions about wh-licensing. In the next section I turn to the discussion of the triggering factor in adjunct embedded clauses.

5.2. The wh-syntax of adjunct embedded clauses in scope marking The situation in adjunct-type scope marking is very similar to the one found in argumental scope marking, illustrated and explained in the previous section. Hungarian is again lame about the wh-syntax of the embedded clause: although all constructions discussed in this paper feature overt wh-movement, these always take place for focusing reasons: (27)

Milyen üzenetet kapott Péter [hogy hovai kell mennie ti]? what message-ACC got-3SG Péter that where need go-INF-3SG (lit.) ‘What message, where does he have to go, did Péter get?’

(48) *Milyen üzenetet kapott Péter [hogy kell mennie hova ]? what message-ACC got-3SG Péter that need go-INF-3SG where (19)

szerez ti]? Ki megy át a vizsgán [aki hány pontoti who go-3SG PV the exam who-REL how many point-ACC score-3SG (lit.) ‘Whoi, whoi scores how many points, passes the exam?’

(49) *Ki megy át a vizsgán [aki szerez hány pontot ]? who go-3SG PV the exam who-REL score-3SG how many point-ACC Looking at other languages with parallel facts, however, gives clear indication about the wh-syntax of these clauses. Scope marking constructions of the type discussed in this paper, namely with noun-associate and relative clauses, are not restricted to Hungarian only.

286

Anikó Lipták

According to my survey on 18 languages 8, adjunct-type scope marking constructions occur in a subset of the languages that have standard argumental scope marking constructions, and do not occur in languages where argumental scope marking cannot be found. Specifically, I found that Frisian and some Slavic languages (Serbian and Slovenian), which are known to have subordinate scope marking (cf. Hiemstra 1986, Golden 1995, Stepanov 2000) have constructions parallel to the Hungarian facts discussed in this paper. Frisian and Slovenian are extremely instructive to look at when discussing the embedded wh-movement step in scope marking. In these languages wh-movement in embedded questions targets Sp,CP, which can be seen from the fact that the wh-phrase is placed before the embedded complementizer (Frisian and Slovenian allow for a doubly-filled comp). In nounassociate clauses with scope marking, the embedded wh-phrase lands in Sp,CP as well, as the data from Frisian (50) and Slovenian (51) illustrates: 9 (50) Wat boadskip hast krigen, wêr’tst hinne moatst? where-that-2SG to must what message have-2SG got (lit.) ‘What message, where do you have to appear, did you get?’ (51) Kakšno sporo ilo si dobil, kam da mora‰ iti jutri? what message aux get-PTC where that must go tomorrow (lit.) ‘What message, where do you have to appear, did you get?’ In scope marking with relative clauses, embedded wh-movement does not occur, wh-phrases stay in situ ((52) illustrates Frisian, and (53) Serbian10): (52) ?Hokker studint komt dertroch, dy’t hoefolle punten hat? which student comes through REL-that how-many points has (lit.) ‘Which student, who scores how many points, passes the exam?’ (53) Koji student prolazi ispit, koji dobije koliko poena? which student passes exam which gets how many points? (lit.) ‘Which student, who scores how many points, passes the exam?’ Without further discussion, I take it that Hungarian (27) is in all respects similar to (50), (51), and (19) is similar to (52), (53). The difference is that while wh-movement to Cº is overt in Frisian and Slovenian noun-complement clauses, it is covert in Hungarian. How to account for these data then? Following the analysis of argumental scope marking in the previous section, we can extend the proposal to cases

Scope Marking Constructions in Dayal-type Indirect Dependency

287

of adjunct scope marking as well: the embedded question in these constructions, too, behaves as a fully specified interrogative clause, and as such, it becomes available as a propositional restrictor to the matrix wh-quantifier. In the case of noun-associate questions, the embedded wh-phrase fronts to Sp,CP to check the feature on C0: (54) Wat boadskip hast krigen… what message have-2SG got [ hinne moatst ti?]]? … [CP wêri’tst to must where-that-2SG The availability of the feature on Cº is the result of a feature sharing process, that was introduced in the previous section. The matrix NP and the embedded clause share a wh-feature: (55) [hast krigen…[SC [wat boadskip][CP wêr’tst

hinne moatst]]

The situation with relative clauses is different, as the empirical evidence in (52), (53) above shows. In these contexts we do not find overt wh-movement in the embedded clause. This is due to independent syntactic reasons. Although the same feature-sharing process as in (55) is available here, too, (the head of the relative clause and the relative clause itself form a constituent in the base), the relative complementizer cannot carry any interrogative feature due to the fact that the CP domain of relative clauses is fully specified for relativization. The relative Sp,CP position hosts a relative operator (overt or covert) and the relative Cº complementizer’s feature content is incompatible with features. For this reason, overt wh-movement does not occur inside relative clauses in languages where wh-movement does otherwise move wh-items into the CP domain. Without an attracting feature, the embedded wh-item has to stay in-situ. The exact licensing conditions of this wh-item are unclear to me. Lacking insight, I leave this for further investigation, briefly mentioning only that what we are dealing here is not unknown in the syntax of English, either. Where English differs from Hungarian, Frisian or Slavic languages is that relative clauses with questions are only grammatical under special circumstances, namely in the context of quizquestions 11: (56) Which actor, who was nominated for Oscar in which film in 1965, died in 1980?

288

Anikó Lipták

The availability of adjunct-type scope marking shows differences crosslinguistically: it is restricted to quiz-contexts in English-type languages, while it is available in “ordinary” scope marking questions in some languages that allow for argumental scope marking as well. In this section I provided crosslinguistic evidence for the empirical observation that embedded wh-movement takes place in the overt syntax in adjunct scope marking constructions and I argued that it is possible to characterize the embedded wh-movement step in these as an instance of matrix whmovement, if a Dayal-type approach to these constructions is adopted.

6. Summary This paper introduced hitherto unidentified scope marking constructions from Hungarian, Frisian, and Slavic languages, and showed that these involve complex questions embedding adjunct clauses (noun-associate or relative clauses). Their existence is of great theoretical importance because they provide primary evidence for a Dayal-type indirect dependency analysis. Such an analysis in turn argues for a conventional picture concerning the movement triggers in the embedded clause of scope marking: wh-movement to a high licensing position has to happen for interpretive reasons.

Notes 1. My research on scope marking constructions was financed by NWO (Netherlands Organisation for Scientific Research), whose help I hereby thank. I owe gratitude to Crit Cremers, Veneeta Dayal, and István Kenesei for their comments on my ideas; and two anonymous reviewers for their invaluable comments on the previous version of this paper, both when it comes to content and presentation. All remaining errors are mine. 2. More detailed investigation (Herburger 1994; Lahiri 2002) shows that the parallel with long extraction is not absolute. 3. Note that the overt position of wh-items in Hungarian is not Sp,CP, but a lower focus projection, Sp,FocP (Horvath 1981, É. Kiss 1987). 4. The Hungarian data in this paper were collected in the form of written questionnaires and oral consultations in Hungary, during the years 2001 and 2002. The material represented here is based on the judgments of the following people: Péter Antonyi, Huba Bartos, Péter Boross, Judit Gervain, Beáta Gyuris, Ildikó

Scope Marking Constructions in Dayal-type Indirect Dependency

5. 6.

7.

8.

9.

10. 11.

289

Kasza, István Kenesei, Attila B. Kis, Katalin É. Kiss, Márta Maleczki, EnikŒ T. Németh, Lászlóné Sipos, Balázs Surányi, Gabriella Tóth, Ildikó Tóth, Péter Vajda, László Varga. Symbols are taken from Varga (2002): | = edge of intonational phrase;  = pause; ' = full fall major stress; ' = half-fall major stress The prosodic characteristics of parenthetical scope marking constructions are close to those that Reis (2000) calls Integrated parenthetical (IP) constructions in German. They are not the same, though. German IP can be characterized by the following properties, not present in Hungarian: (a) With IP in German, the embedded clause has V2, signalling that it is a root clause; (b) IP allows for yes/no questions; (c) preference predicates (wünschen “would like”) cannot occur in IP. These effects can be illustrated by (i) and (ii): (i) a. *Kineki zavarta Marit [hogy ti telefonáltál]? who-DAT bothered-3SG Mari-ACC that phoned-2SG b. Mi zavarta Marit [hogy kinek telefonáltál]? what bothered-3SG Mari-ACC that who-DAT phoned-2SG (lit.) ‘What bothered Mari, that you phoned whom?’ (ii) a. *Hogy voltál szomorú [mert viszonyultak hozzád]? how were-2SG sad because related-3PL 2SG-TO b. Miért voltál szomorú [mert hogy viszonyultak hozzád]? what-FOR were-2SG sad because how related-3PL 2SG-TO (lit.) ‘Why were you sad, because they related to you in which manner?’ These were: Moroccan Arabic, Bavarian, Mandarin Chinese, Danish, Dutch, English, Finnish, Flemish, Frisian, German, Greek, Hindi, Hungarian, Italian, Japanese, Serbian, Slovenian, Spanish. The Frisian data are based on the judgements of Henk Wolf, Siebren Dijk and Willem Visser (p.c.), the Slovenian ones on judgements by Franc Maru‰iã, Tatjana Marvin, Rok Îaucer (p.c.). The Serbian data are from Boban Arsenijeviã and Radoslava Trnavac. Although these sentences are treated as ordinary questions in Kempson et al (2001).

References Adger, D. and J. Quer 2001 The syntax and semantics of unselected embedded questions. Language 77. Beck, S. and S. Berman 2000 Wh-Scope Marking: Direct vs. Indirect Dependency. In Wh-Scope Marking, U. Lutz, G. Müller and A. v. Stechow (eds.).

290

Anikó Lipták

Bennis, H. 1986 Brody, M. 1995 Bruening, B. 2001 forthc. Cheng, L. 2000

Gaps and dummies. Dordrecht: Foris. Lexico-Logical Form. Cambridge, Mass. MIT Press. Syntax at the Edge. Diss. MIT. Two Types of Wh-Scope Marking in Passamaquoddy. Natural Language and Linguistic Theory.

Moving just the feature. In Wh-Scope Marking, U. Lutz, G. Müller and A. v. Stechow (eds.). Chomsky, N. and H. Lasnik 1991 Principles and parameters theory. In Syntax: An International Handbook of Contemporary Research, J. Jacobs, A. von Stechow, W. Sternefeld and T. Vennemann (eds.). Walter de Gruyter, Berlin. McDaniel, D. 1989 Partial and multiple Wh-Movement. Natural Language and Linguistic Theory 7. Dayal, V. 1994 Scope Marking as Indirect Wh Dependency. Natural Language Semantics 2. 1996 Locality in Wh Quantification: Questions and Relative Clauses in Hindi. Dordrecht Kluwer. 2000 Scope Marking: Cross Linguistic Variation in Indirect Dependency. In Wh-Scope Marking, U. Lutz, G. Müller and A. v. Stechow (eds.). Fanselow, G. and A. Mahajan 1996 Partial movement and successive cyclicity. In Wh-Scope Marking, U. Lutz, G. Müller and A. v. Stechow (eds.). 2000 Towards a Minimalist Theory of Wh-expletives, Wh-Copying and Successive Cyclicity. In Wh-Scope Marking, U. Lutz, G. Müller and A. v. Stechow (eds.). Golden, M. 1995 Interrogative Wh-movement in Slovene and English. Acta Analytica 14. Grimshaw, J. 1990 Argument structure. Cambridge, MA, MIT Press. Hamblin, C. L. 1973 Questions in montague English. Foundations of Language 10. Herburger, E. 1994 A semantic difference between full and partial Wh-Movement in German. Paper presented at SLA, Boston. Horvath, J. 1981 Aspects of Hungarian Syntax and the Theory of Grammar. Diss. Los Angeles: UCLA.

Scope Marking Constructions in Dayal-type Indirect Dependency 1995 1997 1998 2000

Höhle, T. 1990

291

Partial Wh-Movement and the Wh “Scope-Markers”. In Kenesei, I (ed.) Approaches to Hungarian 5. Szeged, Jate. The status of ‘wh-expletives’ and the partial movement construction in Hungarian. Natural Language and Linguistic Theory 15. Multiple WH-phrases and the WH-Scope-Marker strategy in Hungarian Interrogatives. Acta Lingustica Hungarica 45. On the Syntax of “Wh-Scope Marker” Constructions: Some Comparative Evidence. In Wh-Scope Marking, U. Lutz, G. Müller and A. v. Stechow (eds.). Die w…w-Konstruktion im Deutschen. Talk presented at DGfS, Saarbrücken.

Kayne, R. 1994 The antisymmetry of syntax. Cambridge: MIT Press. Kempson, R., W. Meyer-Viol, D. Gabbay 2001 Dynamic Syntax: The Flow of Language Understanding. Oxford: Blackwell. Kenesei, I. 1992 Az alárendelés. [Subordination.] In Kiefer, F. (ed.), Strukturális magyar nyelvtan I, Mondattan. [Structural Hungarian Grammar I. Syntax.] Budapest: Akadémiai Kiadó. 1994 Subordinate clauses. In The syntactic Structure of Hungarian, K. Kiefer and K. É. Kiss (eds.). Syntax and Semantics 27. San Diego: Academic Press. Kiss, K. É. 1987 Configurationality in Hungarian. Budapest: Akadémiai Kiadó. 1991 An argument for movement. In Representation and Derivation in the Theory of Grammar, H. Haider and K. Netter (eds.). Dordrecht, Kluwer. Lahiri, U. 2002 On the Proper Treatment of “Expletive Wh” in Hindi. Linqua 112. Lipták, A. 2001 On the syntax of Hungarian wh-items. Diss, LOT. Lutz, U., G. Müller and A.v. Stechow (eds.) 2000 Wh-Scope Marking. Amsterdam: John Benjamins. Mahajan, A. 1990 The A/A-bar Distinction and Movement Theory. Diss. MIT. Marácz, L. 1990 Asymmetries in Hungarian. Diss. University of Groningen. Müller, G. 1995 A-bar syntax. The study of movement types. New York, Mouton. 1998 Incomplete category fronting. Dordrecht: Kluwer. Müller, G. and W. Sternefeld 1996 A-bar chain formation and economy of derivation. Linguistic Inquiry 27.

292

Anikó Lipták

Moro, A. 1997

The raising of predicates. Predicative noun phrases and the theory of clause structure. Cambridge, Cambridge University Press.

Reis, M. 2000

On the Parenthetical Features of German Was… W-Constructions and How to Account for Them. In Wh-Scope Marking, U. Lutz, G. Müller and A. v. Stechow (eds.). Riemsdijk, H. van 1983 Correspondence Effects and the Empty Category Principle. Tilburg Papers in Language and Literature. Also in: Studies in Generative Grammar and Language Acquisition, edited by Y. Otsu et al. (1983), Tokio: ICU. Rosenbaum, P. 1967 The grammar of English predicate complement constructions. Cambridge, MA. Rothstein S. D. 1995 Pleonastics and the interpretation of pronouns. Linguistic Inquiry 26. Simpson, A. 1995 Wh-movement and the Locality of Feature-Checking. Amsterdam: John Benjamins. Stechow, von A. 1996 Against LF Pied-Piping. Natural Language Semantics 4. Sternefeld, W. 1999 Wh-Expletives and Partial Movement: Two Non-existing concepts? Ms. Universität Tübingen. Stepanov, A. 2000 Wh-Scope Marking in Slavic. Studia Linguistica 54. Stowell, T. 1981 Origins of Phrase Structure. Cambridge, MIT Press. Surányi, B. 2003 Multiple operator movements in Hungarian. Diss, LOT. Szabolcsi, A. 1992 The noun phrase. In Kiefer and É. Kiss, The Structure of Hungarian. Syntax and Semantics 28. Academic Press, New York. Torrego, E. and J. Uriagereka 1989 Indicative dependence. Ms. University of Massachussetts, University of Maryland, Amherst. Varga, L. 2002 Intonation and Stress. Evidence from Hungarian. New York, Palgrave Macmillan.

Hyperbaton and Haplology Eric Mathieu

1. Introduction The present paper highlights an apparent parallelism between the phenomena of hyperbaton and haplology that can be drawn as the consequence of previous work by Fanselow & Cavar (2001) – F&C, henceforth. Although the connection is at first rather appealing, after a detailed analysis of the properties of hyperbaton, the parallelism is dismissed. Other phonological accounts of hyperbaton are also examined, including F&C’s Distributed Deletion analysis. The hypothesis defended in this paper is that hyperbaton is not due to phonological properties of the two split elements themselves, but simply to a visibility requirement associated with the left periphery of the clause. The conclusion will be that one cannot steer away from the movement analysis of the bare operator (or its base-generation in Spec-CP). Minimally, a bare operator satisfies the relevant morpho-syntactic features in the left field and therefore bare operator movement is the default option cross-linguistically. As an answer to the question as to why full pied-piping is possible when phonological material has already been pied-piped, it will be argued that full movement (of the bare operator and the remnant) is motivated, not by morpho-syntactic features, but by semantico-pragmatic properties. This contribution is part of a larger work on split-XPs that began with Mathieu (2002). Here I concentrate on the phonological properties of splitXPs, a topic that I have not discussed before. As a defence for my case against the phonological account, I will rely on Butler and Mathieu (2004a, b), and Mathieu and Sitaridou (2004a, b). Section 2 introduces examples of hyperbaton. Then, three phonological accounts of the phenomenon are introduced: section 3 looks at the clitic approach; section 4 introduces F&C’s Distributed Deletion account; section 5 concentrates on F&C’s alternative suggestion about DP-splitting and develops an apparent parallelism between hyperbaton and haplology based on that suggestion. After all three phonological approaches are dismissed, section 6 offers an alternative account of hyperbaton, and gives a rationale

294

Eric Mathieu

for why at least one element must moved to the left periphery in hyperbaton configurations, and also why full pied-piping is also allowed in (apparent) blatant violation of economy. The conclusion can be found in section 7.

2. Hyperbaton The term ‘hyperbaton’ comes from a combination of hyper and the aorist participle of baino ‘to come’ and means ‘stepped over’ or ‘passed over’. It is a rhetorical device with the discourse function to highlight certain constituents. It is frequently found in poetry and has been claimed to also be present in colloquial/spoken registers. This is certainly the situation found today in languages that allow rearrangement of basic DP word order. For example, in Polish and Russian, this process is possible both in literary and colloquial registers (see Sekerina 1997). The most basic form of hyperbaton involves movement of a noun out of a prepositional phrase (sometimes called phrasal wrapping, also found in Latin). The extracted element is stressed and receives a strong or contrastive interpretation. To illustrate, in (1) the emphatic genitive dinameos precedes the word which governs it (i.e. ekino). (1)

a. d’ is ekino dinameos monon vlepo. see-1SG and to this force-GEN only ‘I only see power in this.’ b. DINAMEOSi d’is ekino ti monon vlepo. ‘I only see POWER in this.’

(Denniston 1997: 47)

(2) illustrates extraction of an emphatic quantifier: (2)

a. Echei elpidas pollas. had-3SG hopes-ACC.FEM.PL many-ACC.FEM.PL ‘He had many hopes.’ b. POLLASi echei ti elpidas. ‘He had MANY hopes (e.g. not a few).’

Example (3) shows extraction of a demonstrative:

(Herodotus 5.36)

Hyperbaton and Haplology

(3)

295

a. Tous chrômenous tê those-ACC.MASC.PL use-PART.ACC.MASC.PL the-DAT.FEM.SG tautê ergasia. this-DAT.FEM.SG profession-DAT.FEM.SG ‘Those that exercise this profession.’ b. Tous TAUTEi chrômenous tê ti ergasia. ‘Those that exercise THIS profession (e.g. not that one).’ (Aeschines 1.119)

Next, in (4b) extraction of a negative quantifier is exhibited: (4)

a. Hymas pothein akousai oudemian prophasin. you-ACC.PL desire-INF hear-INF.PAST no-ACC.FEM.SG excuse-ACC.FEM.SG ‘That you desire to hear no excuse.’ b. OUDEMIANi hymas pothein akousai ti prophasin. ‘That you desire to hear NO excuse.’ (Lysias 14.1)

Adjectives can split too. It is clear from the context in which they appear that such discontinuous structures in Classical Greek involved a set of alternates, i.e. involved strong focus. This is well-documented in Devine and Stephens (2000). In (5b) the interpretation is: ‘private, not public’. (5)

a. Kateskeuakasin oikias tas idias. build-PLUPERFECT.3PL house-ACC.FEM.PL the-ACC.FEM.PL private-ACC.FEM.PL b. TAS IDIASi kateskeuakasin ti oikias. ‘They have built their private homes (e.g. not their public ones).’ (Demosthenes 23.208)

Finally, (6) involves an interrogative (here the extracted element receives informational rather than contrastive focus): (6)

a. Tína dynamini what-ACC.FEM.SG power-ACC.FEM.SG b. Tínai echei ti what-SG.FEM.ACC has-3SG ‘What power does it have?’

echei ti? has-3SG

(Plato Laws 643a)

dynamin? power-ACC.FEM.SG (Plato Republic 358b)

296

Eric Mathieu

Turning now to Modern Greek, the same patterns can be found, except for (6b). (7b) demonstrates extraction of a quantifier. (7)

a. Ixe poles elpides. have-PAST.3SG many-ACC.FEM.PL hope-ACC.FEM.PL ‘He had many hopes.’ b. POLESi ixe ti elpides. ‘He had MANY hopes (e.g. not a few).’

(8b) shows extraction of a demonstrative: (8)

a. Ida afto to forema. see-PAST.1SG this-ACC.NEUT.SG the-ACC.NEUT.SG dress-ACC.NEUT.SG ‘I saw this dress.’ b. AFTOi ida ti to forema. ‘I saw THIS dress (e.g. not that one).’

(9b) is an illustration of negative extraction: (9)

a. Den thelo na akuso kamia dikeologia. not want-1SG PRT-SUBJ listen-1SG no-ACC.FEM.SG excuse-ACC.FEM.SG ‘I want to hear no excuse.’ b. KAMIAi den thelo na akuso ti dikeologia. ‘I want to hear NO excuse.’

Next, whereas in (10a) the adjective kokino ‘red’ is adjacent to the noun it modifies, i.e. forema ‘dress’, in (10b) the adjective has been extracted on its own. If the condition that the extracted element must receive stress in order to be interpreted contrastively is not met, the sentence is ungrammatical, as illustrated by (10c). (10) a. Agorase to buy-PAST.3SG the-ACC.NEUT.SG ‘She bought the red dress.’

kokino red-ACC.NEUT.SG

forema. dress-ACC.NEUT.SG

b. To KOKINOi agorase ti forema. c. *To kokinoi agorase ti forema. ‘She bought the RED dress (e.g. not the blue one).’

Hyperbaton and Haplology

297

The type of example illustrated in (10b) was first introduced by Androutsopoulou (1997: 2). A variant of (10b), originally discussed by Horrocks and Stavrou (1987: 91), is introduced in (11). It is identical to (10b) apart from the additional determiner present to the left of the stranded nominal; a process dubbed Determiner Spreading (the choice to have an additional determiner appears to be dialectal/a matter of register): forema. (11) To KOKINOi agorase ti to the-ACC.NEUT.SG red-ACC.NEUT.SG buy-PAST.3SG the-ACC.NEUT.SG dress-ACC.NEUT.SG ‘She bought the RED dress (e.g. not the blue one).’ In Classical Greek, the situation was more straightforward, since in the canonical word-order (Adjective-Noun, as in the modern form of the language), a doubled determiner never appeared in this kind of constructions (cf. Devine and Stephens 2000 and Manolessou 2000).1 Finally, it must noted that WH-elements like ti can no longer undergo splitting in Modern Greek (note also that ti is no longer marked for case, gender or number). The reason behind this diachronic change need not concern us in this paper. I simply add that there are two exceptions to the rule against WH-splitting in Modern Greek: for some (but not all speakers) both tinos ‘whose’, a genitive remnant of Classical Greek (13b) and pianu ‘whose’, its modern counterpart (14b) can split. In Mathieu and Sitaridou (2004b), an account of this contrast is given, together with an explanation for the general loss of WH-splitting in Modern Greek, as well as for the dialectal variation. (12) a. Ti what

dinami power-ACC.FEM.SG

exi? have-3SG

b. *Tii exi ti dinami? ‘What power does it have?’ (13) a. Tinos to vivlio eferes? whose-GEN.SG the-ACC.NEUT.SG book-ACC.NEUT.SG bring-PAST.2SG b. Tinosi eferes ti to vivlio? ‘Whose book did you bring?’

(Horrocks and Stavrou 1987:89)

(14) a. Pianu to vivlio eferes? whose-GEN.MASC.SG the-ACC.NEUT.SG book-ACC.NEUT.SG bring-PAST.2SG b. Pianui eferes ti to vivlio? ‘Whose book did you bring?’

298

Eric Mathieu

Hyperbaton is found in a wide range of languages. The following constructions from French (15b), German (16b) and Dutch (17b) can be grouped together with the split constructions introduced above: (15) a. Combien how-many

de of

livresi books

as-tu have-you

lus ti? read-MAS.PL

b. Combieni as-tu lu ti de livres? ‘How many books have you read?’ (Obenauer 1976, 1983, 1994) (16) a. Was what

für for

ein a

Werkzeugi tool

sucht looks

er ti? he

b. Wasi sucht er ti für ein Werkzeug? ‘What kind of tool is he looking for?’ (17) a. Wat what

voor for

boekeni books

heb have

jij you

(de Swart, 1992: 389)

gelezen ti? read

b. Wati heb jij ti voor boeken gelezen? ‘What kind of books have you read?’

(de Swart 1992: 389)

Languages are not wholly consistent with respect to whether it allows splitDPs. Modern Greek is a good example. Some DPs can split (tinos, pianu, adjectives, indefinites, and negative elements), while others cannot (ti, pios). Thus, there cannot be such thing as a split- versus a non split-DP language. Cross-linguistically, different combinations arise. For example, bare combien extraction is possible in French, as shown by (15b), whereas the equivalent construction is not possible in Modern Greek (18b). On the other hand, whereas bare demonstrative extraction is available in Modern Greek (8b), it is impossible in French (19b). (18) a. Posa how-MANY.ACC.NEUT.PL

vivlia book-ACC.NEUT.PL

diavases? read-PAST.2SG

b. *Posai diavases ti vivlia? ‘How many books did you read?’ (19) a. J’ai acheté cette robe. I.have bought this-FEM.SG dress-FEM.SG ‘I have bought this dress.’

Hyperbaton and Haplology

299

b. *CETTE j’ai acheté de robe.2 this-FEM.SG I-have bought of dress-FEM.SG ‘I have bought THIS dress (e.g. not that one).’ For a parameterization account, see Mathieu and Sitaridou (2004a, b).

3. Phonology Account I: The Clitic Approach A well-known analysis of hyperbaton is that of the so-called trompe-l’oeil approach. The clitic approach, as it is also called, argues that splitting is an illusion. Rather than an XP splitting into two, rightward movement of an element originally to the left is what creates variation in word order. For example, in (20b), although one might be tempted to think that the nominal has raised above the preposition, the proposal is that, in fact, it is the preposition that has moved to a second position as illustrated by (21): (20) a. Apo from

melainaon the.black

neon. ships

b. Melainaoni apo ti the.black from ‘From the black ships.’ (21)

neon. ships (Il 16.304)

apoi neon. ti melainaoni the.black from ships ‘From the black ships.’

The idea behind this proposal is that prepositions are reduced clitics (reduced in accentuation) and therefore need something to host them. The prepositional head therefore moves for prosodic reasons. Since phrase initial clitics can in fact be either in first or second position, it is expected that prepositions will also be in first position from time to time. Proponents of this theory for Greek are for example: Golston (1988). A similar analysis for Serbo-Croatian split-XPs has been proposed independently. As discussed in Fanselow and Cavar (2002), Zec and Inkelas (1990) argue that syntactic constituents are split by clitics in Serbo-Croatian. A clitic cluster may appear after a complex DP (22a) or inside the complex DP (22b) – the clitic cluster is in italics:

300

Eric Mathieu

(22) a. Taj this

c ovjek man

joj her

ga it

je poklonio. be-3SG present-PTC

b. Taji joj ga je ti c ovjek poklonio. present-PTC this her it be-3SG man ‘This man presented it to her.’ (Fanselow and Cavar 2002: 80) As shown by Devine and Stephens (2000), the main problem for this approach is that prepositions need not be in first or second position in Classical Greek: material can also intervene between the preposition and the nominal. The conclusion thus must be that they are not clitics: (23)

basilikon domon huper. the-royal palace for ‘For the royal palace.’

(Phoen 1326)

On a similar note, if one adheres to the clitic theory, then it is no longer possible to explain the availability of hyperbaton with interrogatives, since in this case the intervening material cannot be considered a clitic (for example, verbs were not clitics in Classical Greek, cf. Devine and Stephens, 2000): (24) Tina boais logon? what shout-PRES.2SG words ‘What words are you shouting?’

(Hipp 571)

Fanselow and Cavar (2002) independently make more or less the same point with regard to Serbo-Croatian. In that language too, material other than clitics can intervene between the two parts of the split construction (see Browne 1976): (25)

Kakav je Ivan kupio auto? what-kind-of be-3SG Ivan buy-PTC car ‘What kind of car has Ivan bought?’ (Fanselow and Cavar 2002: 80)

In fact, as shown by Cavar (1999) and Fanselow and Cavar (2002), the same type of syntactic discontinuity is possible with the constructions discussed by Zec and Inkelas (1990). In (26) a demonstrative is topicalized and is separated from the head noun of the complex DP by the subject Ivan, and not simply by clitics:

Hyperbaton and Haplology

(26) Taj je Ivan kupio auto. this be-3SG Ivan buy-PTC car ‘Ivan bought this car.’

301

(Fanselow and Cavar 2002: 80)

To conclude, the clitic solution to XP-splitting cannot be right.

4. Phonological Account II: Distributed Deletion In this section F&C’s (2001, 2002) Distributed Deletion account and their tentative solution to the problem as to what motivates splitting are introduced. These authors argue that split-XP constructions are best analysed via distributed deletion of phonological copies at LF. Assuming Chomsky’s approach to movement (Chomsky 1995), where raising of elements in the phrase-marker leaves copies behind, which are subsequently deleted, the authors argue that the deletion operation may affect both copies and that it is the best way to account for split-XPs (it is not clear that this idea is in fact consistent with Minimalism, since scattering of phonological features is not expected, but I will leave this point aside). The theory aims to offer a unified account of both PP-extraction as in (27b) and noun extraction as in (28b). F&C call (27b) a pull split (the order of the split elements is preserved A … B) and (28b) an inverted split (the order of the split elements is not preserved B … A): (27) a. Na on

jaki what-kind

dach roof

Marek Marek

kocił? jumped

b. Na jaki Marek dach kocił? ‘On what kind of roof did Marek jump?’ (28) a. Keine Bücher habe none books have

ich I

gelesen. read

b. Bücher habe ich keine gelesen. ‘As for books, I haven’t read any.’ The second part of the proposal is that in split constructions, a DP or a PP is split up in case its phonetic material is linked to at least two different pragmatic features. On the assumption that focus is checked in a specific position, they argue that the two spread elements split up because the feature

302

Eric Mathieu

that they each bear cannot both be checked in the same position. One bears a +WH or a +TOP feature and is checked in a higher interrogative position and the other carries a Focus feature that is checked in a lower focus position. (29) and (30) illustrate the proposal. (29)

Marek [F+Foc dach+Foc [F+WH Na jaki+WH on what-kind Marek roof ‘On what kind of roof did Marek jump?’

kocił]]? jumped

(30)

[F+TOP Bücher+TOP habe ich [F+Foc keine+Foc books have I none ‘As for books, I haven’t read any.’

gelesen]]. read

In the case where everything raises to the left, only one feature is involved, namely the +WH feature for the case of (31) and the +TOP feature in the case of (32): (31)

dach]+WH Marek kocił]]? [F+WH [Na jaki on what-kind roof Marek jumped ‘On what kind of roof did Marek jump?’

(32) [F+TOP [Keine Bücher]+TOP none books ‘No books, I have read.’

habe ich have I

gelesen]. read

(33) and (34) give the derivations for (31) and (32) respectively: (33) Marek na jaki dach kocił? A movement + copying Na jaki dach Marek na jaki dach? A distributed deletion Na jaki dach Marek na jaki dach kocił? (34) Ich habe keine Bücher gelesen. A movement + copying Keine Bücher habe ich keine Bücher gelesen. A deletion Keine Bücher habe ich keine Bücher gelesen. For sake of completeness, the derivations for full movement are represented in (35) and (36): (35) Marek na jaki dach kocił? A movement + copying Na jaki dach Marek na jaki dach? A distributed deletion Na jaki dach Marek na jaki dach kocił?

Hyperbaton and Haplology

303

(36) Ich habe keine Bücher gelesen. A movemen = copyingt Keine Bücher habe ich keine Bücher gelesen. A deletion Keine Bücher habe ich keine Bücher gelesen. The two main motivations for this proposal are (i) instances where both parts of the copy are pronounced, and (ii) constraints and anti-constraints on movement. Fanselow and Cavar take the case of the ‘Copy-construction’ (Hiemstra 1986; Höhle 1990; Fanselow and Mahajan 2000; Nunes, 2001) in German as evidence that it is possible for two copies to be pronounced. (37)

Wer denkst du denn wer who think you that who ‘Who do you think you are?’

du you

bist? are

(Fanselow and Cavar 2002: 83) Moreover, they argue that certain resumptive pronouns (e.g. Hebrew (38)) reflect the failure of copies of movement to delete completely (on resumptive pronouns, see Doron 1982; McCloskey 1990; Demirdache 1992, 1997; Shlonsky 1992). (38)

Ha-‘iš ‰e paga‰ti ‘oto. the-man that I-met him ‘the man that I met.’

Next, F&C want to show that simple movement analyses of a bare operator is problematic and that a Distributed Deletion account is preferable (they also argue against a treatment according to which both split parts are basegenerated in situ, but for lack of space I will not review their critique of such approaches). Movement theories posit extraction of an element x out of a constituent y. Although problematic for pre-Abney (1987) theories of NPs, extraction of Bücher in (39) is no longer problematic now we have a better understanding of the internal architecture of noun phrases (Bücher is not a submaximal projection). (39) [Bücher]i habe ich [keine ti] gelesen. As pointed out by F&C, the case of PPs is more complicated. Since (or rather on the assumption that, see below) the extracted preposition and the determiner do not in fact form a constituent, like others F&C assume remnant

304

Eric Mathieu

movement (in the sense of den Besten and Webelhuth 1990; Müller 1998, etc.). First, the nominal is extracted from the PP, then the remnant [na kakav t] raises to Spec-CP. (40)

je Ivan [krov]i skoãio ti. [Na kakav ti] on what-kind has Ivan roof jumped ‘On what kind of roof has Ivan jumped?’

In other words, this case reduces to the previous one. We have extraction of a nominal. The only difference between (39) and (40) is that in (40) (further) movement is required (but not in (39)). The arguments that F&C put forward against the traditional movement account are as follows. First, they notice that the two split parts can contain more phonetic material than fits into a single constituent. In (41a) two prepositions surface, while in (42a) two determiners appear. Full topicalization is not possible when the decuplicated element is present: preposition doubling is not possible within one consituent (41b) and the indefinite article and the negative quantifier kein compete for the same position in German, therefore they cannot appear together (42b): (41) a. In Schlössern habe ich noch in keinen gewohnt. in castles have I yet in no lived ‘As for castles, so far I have not lived in any.’ b. *In keinen in Schlössern have ich gewohnt. (42) a. Einen an keinen no

amerikanischen Wagen kann ich mir American car can I me neuen leisten. new afford

b. *Keinen neuen einen amerikanischen Wagen. (Fanselow and Cavar 2002: 71) The second problem concerns extraction out of PPs. These are normally islands for WH-extraction (43b), topicalization (43c), and scrambling (43d). (43) a. Ivan se popeo [PP na veliko Ivan self climbed on big ‘Ivan climbed on a big tree.’

drvo]. tree

Hyperbaton and Haplology

b. *Štoi what

305

se self

Ivan Ivan

popeo [PP na climbed on

veliko ti]. big

c. *Drvoi se tree self

Ivan Ivan

popeo [PP na climbed on

veliko ti]. big

d. *Ivan Ivan

drvoi popeo [PP na veliko ti]. tree climbed on big (Fanselow and Cavar 2002: 72)

se self

These examples are taken to suggest that in (44) the nominal is not extracted out of the PP: (44) Na veliko se Ivan drvo popeo. on big self Ivan tree climbed The same kind of problem arises in the case of split-topics in German. For instance, these do not respect three types of islands (I discuss only one of them, i.e. the subject island. I refer the reader to F&C (2002) for the remaining two). (45) shows that subjects (of non-unaccusative verbs) are islands for the extraction of PPs. Fanselow (1988, 1993) argues that subjects can nevertheless be split up (46). (45) a. *[An Maria]i haben mir [keine Briefe ti] gefallen. to Mary have me no letters pleased. b. *[An Maria]i hat mich [kein Brief ti] erschreckt. to Mary has me no letter frightened ‘No letter to Mary has frightened me.’ (Fanselow and Cavar 2002: 72) (46) a. Briefe an Maria gefallen mir keine. letters to Mary please me no ‘As for letters to Mary, they do not please me.’ b. Briefe an Maria haben mich keine erschreckt. Letters to Mary have me no frightened ‘As for letters to Mary, they have not frightened me.’ (Fanselow and Cavar 2002: 72) F&C demonstrate that, although split-topics do not exhibit strong islands effects of the subject condition kind, they respect the complex noun phrase kind (47).

306

Eric Mathieu

(47) *Bücher habe ich [eine Geschichte dass sie keine liest] gehört. book have I a story that she no reads heard (Fanselow and Cavar 2002: 78) According to F&C (2002), the apparent paradox (evidence for movement and evidence against movement) is solved in the following way. Movement is involved in split constructions (which is reflected by the availability of copies), but movement is not out of a constituent. “A movement barrier Y does not block the formation of a split XP if and only if Y itself is the XP to be split up” (Fanselow and Cavar 2002: 13). Although sympathetic to the idea that no extraction out of an element is ever involved in split constructions (for independent reasons, see discussion below), I do not actually think that there is a paradox in the first place. Many native speakers of German that I have consulted do not accept sentences such as (46). The status of these sentences is thus not clear. In French split constructions, extraction from a subject position is clearly unavailable, as shown by (48). (48) *Combieni tu penses que ti de personnes ont éternué? of persons have sneezed how.many you think-2SG that ‘How many people do you think sneezed?’ Second, it is not actually clear that examples in (41a) and (42a) can receive an account on the DD approach, because the two elements are certainly not two identical copies created by movement. In fact, the example in (41a) is reminiscent of the case in (11), repeated here. In this example, two determiners surface. (11) To KOKINO agorase to forema. the-ACC.NEUT.SG red-ACC.NEUT.SG buy-PAST.3SG the-ACC.NEUT.SG dress-ACC.NEUT.SG ‘She bought the RED dress (e.g. not the blue one).’ These examples in fact seem to suggest that two separate constituents are involved, each one headed by a different element: a determiner in the case of (11) and a preposition in the case of (41a). More precisely, the configuration before movement in (41a) for instance must be [In Schlössern] [in keinen] side by side. On this view, movement involves, not extraction of an element out of a constituent, but a whole phrase. We develop this analysis in Section 6. The same account can be given for the case in (38). I would like to argue that this kind of example is not really good evidence for the claim that DD is a possible operation in the grammar. The moved element and the resumptive

Hyperbaton and Haplology

307

pronoun do not appear to be part of the same copy. For example, (49) is not possible. (49) *Ha-‘iš ‘oto ‰e the-man him that ‘The man that I met.’

paga‰ti . I-met

Instead, we can think of the extracted element and the resumptive pronoun as forming a phrasal complex in the base (from which the nominal is extracted along the lines of Kayne 2002 and Boeckx 2003), or alternatively as involving movement of a null operator (Demirdache 1992, 1997). Third, the DD account overgenerates. The fact that some PPs can be split, but not others remains a mystery under such a view. In (50 b) extraction of à combien is possible, but in (51b) extraction of en combien is not. (50) a. A combien de personnesi as-tu to how-many of people have-you

écrit ti? written

b. A combieni as-tu écrit ti de personnes? ‘To how many people have you written?’ fini ta thèse ti? (51) a. En combien d’annéesi as-tu in how-many of-years have-you finished your thesis b. *En combieni as-tu fini ta thèse ti d’années? ‘How many years have you taken to finish your thesis?’ The difference between (50b) and (51b) has to do with differences in thematic relations. What raises in (51b) is an adjunct, whereas an argument raises in (50b) (for other contrasts of this sort in split combien constructions and an explanation of these facts, see Mathieu 2004). One final problem is that in the case of split WH-constructions and pull splits in general, it is not clear that we are dealing with two different pragmatic functions. Devine & Stephens (2000) have shown that a pull split construction A… B that leaves the order of A and B intact may arise whenever A bears a pragmatic function, while an inverted hyperbaton structure is possible only if A and B have a special informational value. Interestingly, F&C do not appear content with the idea that there should be specific focus positions involved in the case of pull splits. In a footnote, they consider the idea that XPs are split because of their suboptimal phonological properties. Following a personal communication made to them by Caroline Féry, they speculate that two prominent accents should not be

308

Eric Mathieu

adjacent in a string and XP-splitting follows from that. If an NP has two independent, but otherwise similar, foci, it must realize two prominent accents. The splitting of the phrase avoids a situation according to which these two accents would be too close to each other, since they are otherwise identical. Although not spelled in any detail, the idea behind F&C’s remark about adjacency of prominent accent must be the process of haplology wellmotivated in the domain of morphology, and which consists of the avoidance of repeated morphemes. That haplology should be extended to the domain of syntax has recently been given full empirical support in Neeleman and van de Koot (2002). In the case at hand, the hypothesis is that haplology interacts with other phonological properties, e.g. prosodic structure and thus that it sometimes depends on the suprasegmental properties of the complex word in question, such as syllabic structure and stress. This idea is very interesting and although it might seem unfair to pronounce a judgement on a theory that is not developed yet, their proposal has prompted me to make a possible link between hyperbaton and haplology. I believe that the suggestion is worth pursuing a priori. However, it will turn out to be incorrect for reasons that will be discussed in the section after next, where an alternative phonological proposal suggested by reviewer #2 is also assessed.

5. Phonological Account III: Hyperbaton and Haplology In this section, I explore the relation between hyperbaton and haplology. What needs to be establish is whether XP-splitting is due to the avoidance of two identical prosodic forms. The answer is negative. There are mainly two arguments against the view that haplology is involved in hyperbaton: (i) in an interrogative where at least one element has raised to Spec-CP, the two split-XPs do not in fact receive identical accents; (ii) isomorphy in stress patterns is neither sufficient nor necessary for XP-splitting. First a word about haplology. Haplology is a deletion process best described as the avoidance of identical phonetic or phonological material in morphologically complex words. It occurs in almost any language with enough morphology to create phonetically identical sequences which requires adjacent phonemes to be contrastive. The principle is reminiscent of the Obligatory Contour Principle in phonology (Leben 1973; Goldsmith 1979, and much subsequent work). For example, the existence of Greek amphoreús beside Homeric amphiphoreús ‘two-handled pitcher or jar’ has been explained by haplology. In (52) examples from English can be found:

Hyperbaton and Haplology

(52) a. b. c.

*femininize *minimumize *metathesisize

309

feminize minimize metathesize

Haplology appears not to be totally universal in that it is not only languagespecific but construction specific. More precisely, it is affix-specific. This means that there are many exceptions. For example, in German adjectival inflection sequences such as eigen-en ‘own.pl’ are possible (Plag 1999). At the morpho-syntactic level, well-known examples include the sentences like (53) and (54). In Mandarin Chinese (53) the perfective marker le and the particle le, which marks a ‘new situation’, trigger deletion when adjacent (Yip 1998 and references cited there). Grimshaw (1997) points out that in some dialects of Spanish a sequence of a reflexive and an impersonal clitic, expected to surface as se se, in fact surfaces as a single clitic (54). (53) Bing dou hua le (*le). ice all melt PERF CHANGE-OF-STATE ‘The ice melted’. (54) Se (*se) lava. one oneself washes ‘One washes oneself.’ Like in the case of affixes, the process of haplology at the morpho-syntactic level is not completely rule-governed (see Neeleman and Van de Koot 2001 for details). Pushing the idea that there is a parallelism between hyperbaton and haplology would lead us to the following two schemas. (55) for split WH-constructions and (56) for split topics. (55)

(56)





FOCUS

FOCUS

na jaki+WH

dach+Foc





TOPIC

FOCUS

Bücher+TOP

keine+Foc

The idea is that because the two adjacent elements both receive a prominent accent, they cannot remain side by side. Here one hidden assumption behind

310

Eric Mathieu

F&C’s suggestion is that, although a topic, the bare extracted element receives a prominent accent. However, as pointed by reviewer #1, avoiding identical tones is not the only (and perhaps not even the most plausible) way of spelling out a phonological account). Rather, topicality and focus may be linked to specific tones (on this view the extracted bare topic and the stranded focused element bear distinctive tones), and splitting up noun phrases may be a way of making the specific tone melody easier to pronounce. In other words, the pragmatic roles which the parts of the Noun Phrase bear may imply a certain contour of prosodic phrasing, and the splitting of noun phrases may serve the need of balancing the prosodic phrases in a clause. According to this approach, haplology is thus irrelevant. I shall argue, however, that this alternative phonological account cannot be on the right track. But before we come to that, let me point out that a problem arises in connection with (55) if we are to follow F&C’s conjecture. In a split question, the raised WH-element does not appear to involve a prominent accent; only the in-situ element does. (57)

lu ti de LIVRES? Combieni as-tu how.many have-you read of books ‘How many books have you read?’

Let me say a little more about focus to illustrate the point more clearly. The notion of focus presented here stems from Chomsky (1971, 1976), Jackendoff (1972) and Zubizarreta (1998) – see also Reinhart (1995) and Neeleman and Reinhart (1998). In particular, I adopt Zubizarreta’s recent thesis about focus. According to her analysis, in Germanic and Romance, the most embedded word in the sentence receives default (weak) focus. This is made to follow from the Nuclear Stress Rule (NSR):3 (58) Nuclear Stress Rule (Romance)4 Given two sisters Ci and Cj, the one lower in the asymmetric c-command ordering is more prominent. (Zubizarreta 1998: 150) Take French as an example. To a question like (59a), an answer such as (59b) is appropriate.

Hyperbaton and Haplology

311

(59) a. ‘What did you do?’ b. Nous avons rendu le livre we have returned the book ‘We returned the book to MARIE.’

[à MARIE]F. to Marie

The symmetry between syntactic ordering and phrasal prominence may be broken because of an independent requirement that a focused constituent must contain the most prominent element in the sentence. If le livre receives focus, the PP à Marie is defocalized and thus becomes ‘metrically invisible’. Then, le livre is more prominent (the defocalized element is in italics): (60) a. ‘What did you return to Marie?’ b. Nous avons rendu [le LIVRE]F à Marie. we have returned the book to Marie ‘We returned the BOOK to Marie.’ This is made to follow from the Focus Prominence Rule (FPR): (61) Focus Prominence Rule Given two sister nodes Ci (marked [+F]) and Cj (marked [-F]), Ci is more prominent than Cj. (Zubizarreta 1998: 150) In languages like Spanish, all phonologically specified material is metrically visible. Spanish has recourse to a different mechanism to resolve cases of conflict between the FPR and the NSR: (62) a. Volvimos el libro [a MARIA]F. returned the book to Maria ‘We returned the book to MARIA.’ b. Volvimos a Mariai [el LIBRO]F ti. ‘We returned the BOOK to Maria.’ The PP a Maria undergoes so-called P-movement which is prosodically motivated. As Zubizarreta shows, the case of French is in fact more complicated, since it is also possible in French for a defocalised element to undergo Pmovement: 5

312

(63)

Eric Mathieu

Nous avons rendu à Mariei [le LIVRE]F ti. we have returned to Marie the book ‘We returned the BOOK to Marie.’

Importantly, Germanic and Romance focus is licensed prosodically in statements, but syntactically in questions. A fronted WH phrase is licensed in virtue of occupying the specifier position of a functional category bearing the feature [+WH] (i.e., via the feature-checking mechanism) (Zubizarreta 1998: 92). According to Zubizarreta, whereas in statements, the nuclear stress is contained within the focused constituent, in questions it is contained within the presupposed part of the sentence. I assume that even in the case of split-WH constructions. French single-WH-in-situ constructions, focus is licensed syntactically. So, it is not the case that the bare WH operator is licensed prosodically. The FCP is revised accordingly: (64) Focus Prominence Rule Given two sister nodes Ci (marked [+F]) and Cj (marked [-F]), Ci is more prominent than Cj, unless Ci is a WH phrase and is syntactically licensed by the WH head of Cj. In summary, in an interrogative the raised element that targets the Spec-CP position is not sensitive to prosodic structure. Since it is not licensed prosodically, but syntactically, the Focus Prominence Rule is irrelevant in the case at hand. This indicates that at least in the case of split-WH constructions, haplology has nothing to do with the availability of splitting. It also turns out that splitting can occur even when two adjacent elements are contrastive and therefore should be fine together. For the haplology account, this means that the notion of tonal identity as relevant for splitting is not obligatory. For example, the in-situ equivalent of (11) is equally possible in Modern Greek. The element KOKINO receives contrastive focus while forema receives a informational focus interpretation. Splitting should not be possible, but it is (cf. (11)). (65)

Agorase to KOKINO to forema. buy.PAST.3SG the-ACC.NEUT.SG red-ACC.NEUT.SG the.ACC.NEUT.SG dress.ACC.NEUT.SG ‘She bought the RED dress (e.g. not the blue one).’

Examples from other languages illustrate more or less the same point, but this time they provide counter-arguments to the specific tones account (from reviewer #1). (66) is a special type of NP-construction in Sardinian

Hyperbaton and Haplology

313

where the head noun is introduced by de and placed to the right of a modifier or complement which, in other constructions, would normally follow the head noun (cf. Jones 1993):6 (66) a. Préstami sa tua de pinna. lend-me the your of pen ‘Lend me YOUR pen (i.e. not his).’ b. Keljo comporare sa ruja de mákkina. ‘I want to buy the RED car (e.g. not the blue one).’ (Jones 1993: 76) This construction is fairly common in colloquial Sardinian and is used primarily to emphasize the modifier (sa tua and ruja are contrastively focused and receive heavy stress) at the expense of the head noun (which is always unstressed).7 The crucial observation is that the modifier complex can be detached from the NP despite the fact that sa manna is stressed and de ampulla distressed. These examples invalidate the specific tones phonological account. (67) a. Appo postu sa manna de ampulla in mesa. the big of bottle on table have-1SG put b. Sa manna appo postu de ampulla in mesa. ‘I have put the BIG bottle on the table (e.g. not the small one).’ (Jones 1993: 78) Further evidence against phonological accounts is that a split construction has very different properties from its full movement counterpart. To mention only one of them, split constructions exhibit intervention effects (Obenauer 1976, 1983, 1994; Rizzi 1990; de Swart 1992; Honcoop 1998; Butler & Mathieu 2004a), whereas the full movement alternative does not. tu (68) a. Combien de livresi n’as how-many of books NE-have you

pas not

lus ti? read-MAS.PL

b. Combieni n’as-tu pas lu ti de livres? ‘How many books have you not read?’ If splitting is motivated by prosodic structure, then it becomes a mystery why intervention effects should start appearing in split constructions. It is generally accepted that island and intervention effects are core (narrow)

314

Eric Mathieu

syntactic or for others semantic phenomena that have nothing to do with PF constraints. To summarise, phonological accounts have difficulties explaining most properties associated with split constructions. The trigger for splitting does not appear to have anything to do with prosody or tonal properties of the moved elements. It thus appears to be the case that a bare operator movement account is unavoidable.

6. An Alternative Account In this section I wish to address the following puzzle: if we cannot steer away from the fact that split XP-constructions arise via movement (or basegeneration) of the bare operator (or adjective or noun) involved in hyperbaton, then it becomes a mystery why pied-piping of both the split element and the nominal or other element with which it is associated should ever raise. Presumably, the element that has undergone raising has checked the relevant features associated with C. For example, in (27b) na jaki has checked the strong D(WH) feature of C. Once features have checked and deleted, they are no longer accessible to the computation (cf. Chomsky, 1995). I follow essentially Mathieu & Sitaridou (2004a, b) for the question as to why splitting is possible in the first place, and Butler & Mathieu (2004b) for the triggers behind movement of the bare operator and for why full pied-piping is possible.

6.1. The two DP hypothesis Following and adapting a series of unrelated proposals (Fanselow 1988; Androutsopoulou 1994; Van Geenhoven 1998; and Devine and Stephens 2000), Mathieu & Sitaridou (2004a, b) – M&S, henceforth – put forward the hypothesis that two XPs are involved in hyperbaton: an NP and a DP (thus incidentally the Left Branch Condition is not violated).8 Whereas on Devine and Stephens’ view the two DPs are in apposition, M&S argue that the two XPs are part of a hierarchical structure (on their view, Classical Greek is thus a configurational language). The structure that they propose for split-DPs has the advantage to conform to the well-accepted view that phrase structure is binary, and secondly it allows q-feature matching between the first and the second XP. The DP consists of a modifier (i.e. an adjective) and an empty nominal. Following Devine and Stephens (2000), the modifier

Hyperbaton and Haplology

315

is referred to as ‘null head modifier’, which definition can be given as follows: a modifier that can stand by itself in place of a noun phrase without the support of a noun or an overt pronoun. The idea that it is a whole XP that moves rather than a part of a constituent is most obvious in the case of bare adjectival extraction, since not only the adjective raises, but so does the determiner (cf. (10b) and (11)). M&S adopt Androutsopoulou’s (1997) basic insight that DP-splitting is possible in Modern Greek because null head adjectives/DP fragments are available independently (M&S add to this generalisation the connected fact that Determiner Spreading is also possible because of the availability of noun ellipsis). In English very few null head modifiers/DP fragments are possible, which explains why DP-splitting is virtually impossible. Whereas (69a) is possible in Modern Greek, (69b) is not possible in English (one support is needed). (69)

Which dress did you buy? a. to kokino the.acc.neut.sg red.acc.neut.sg b. *the red.

Speaker A Speaker B

No agreement marking is present on English red, which explains why noun ellipsis is not possible (the empty nominal cannot be licensed). The second XP is an NP; it contains a bare noun that behaves like a predicate rather than an argument (for full details about the predicative nature of the stranded nominal, see Mathieu 2002, 2004; Mathieu and Sitaridou 2004a). Furthermore, M&S argue that the modifier is base-generated in the predicate position within a Small Clause, and that it then undergoes predicate inversion. Motivation for such a structure comes from the fact that splitting is possible only with predicative adjectives. Non-predicative adjectives cannot be split, as shown by (70). ti dolofono. (70) *TON IPOTITHEMENOi ida see.PAST.1SG murderer-ACC.MASC.SG the.ACC.MASC.SG alleged-ACC.MASC.SG ‘I saw the ALLEGED murderer (e.g. not the real one).’ Besides, only predicative adjectives can be used as fragments, as (71a) shows.

316

(71)

Eric Mathieu

Which criminal did you arrest? a. *ton ipotithemeno. the-ACC.MASC.SG alleged-ACC.NEUT.SG

Speaker A Speaker B

b. *the alleged. The technical implementation of their proposal is as follows. In (72a) the subject of the Small Clause is forema ‘dress’ while kokino ‘red’ is the predicate. The adjective kokino is part of a DP that contains a determiner and an empty nominal standing for forema. The double DP analysis does not follow from den Dikken’s (1998) original analysis of DP internal predicate inversion, but is motivated by the facts about DP-splitting (namely, that what raises is not an adjective, but a DP). In those dialects/registers where to does not surface they assume that the head of the Small Clause is simply not spelled out. The nominal copula is not spelt out either in Greek. In languages like French it surfaces as de (den Dikken 1998). They also give the structure for the Classical Greek example (6b) in (72b). In this case, tina is the predicate while dynamin is the subject of the Small Clause (e stands for the null head; coindexation shows binding of the null head by the nominal).9 (72) a. [DP ’ [SC [NP foremai] X to [DP to kokino ei]]]. b. [DP ’ [SC [NP dynamini] X [DP ’ tina ei]]]. The predicate undergoes inversion as shown in (73a) and (73b). In this case, they assume it raises to the specifier of an agreement projection, which following den Dikken they call FP. Adjectival agreement in Modern Greek is rich and therefore the movement is triggered by an EPP feature on F. a---------------------------l ?

(73) a. [DP ’ [FP [DP to kokino e]i [F ’ +X toj [SC [NP forema] [tj ti ]]]]] a------------------------l ?

b. [DP ’ [FP [DP tina e]i [F ’ +Xj [SC [NP dynamin] [tj ti ]]]]] This is a variant of what Alexiadou and Wilder (1998) have proposed for predicative adjectives. The difference with M&S’s proposal is that they assume that no CP is projected and that the second to is the spell-out of the head of the Small Clause. Determiner Spreading is thus a misnomer. Next, in order to account for the focus reading of split-constructions, M&S assume that the predicate futher raises to Spec-DP, which, like many others, they take to be an A’-position in Modern Greek (for DP-internal

Hyperbaton and Haplology

317

focus and topics in Greek, see Horrocks and Stavrou 1987, and for other languages Cinque 1980, Szabolcsi 1983, Stowell 1991, and Giorgi and Longobardi 1991). M&S assume that movement of the adjectival complex is forced by an EPP feature present on D. This EPP feature is correlated to the richness of the adjective involved in the empty nominal complex. Finally, since the predicate raises as high as Spec-DP, further raising of the predicate is possible, in which case we get a split-construction. The last step is not feature driven, see Mathieu and Sitaridou (2002) for details. a----------l a----l a------------------l ? ? ?

(74) [CP [DP to KOKINO e] i…[Spec-DP t i D’[FP t i’ F ’+X toj [SC [NP forema] t j t i ]]]] a--------------l a-----l a------------------l ? ? ?

(75) [CP [DP ’ tina e] i echei [Spec-DP ti D ’[FP t i’ F ’ +X j [SC [NP dynamin] t j t i ]]]] To recapitulate, the steps of a split-DP derivation are the following: (i) the extracted element must be an adjective, and not a determiner; (ii) rich agreement on the adjective is compulsory so that the empty noun can be licensed; (iii) the adjective undergoes inversion and raises to Spec-FP because of an EPP feature on F; (iv) the predicate raises to Spec-DP because of an EPP feature on D; (v) the predicate further raises to the C area. It is interesting to note that in their 2002 paper F&C report that the stranded element in a German split-topic construction must often bear rich inflection and that this correlates with noun ellipsis (in fact, as they point out, for German the observation is an old one, cf. Haider 1985). (76) a. Geld money

hat has

er he

keines/*kein. no

b. Er hat keines/*kein aus Deutschland. he has none from Germany (Fanselow & Cavar 2002: 93) From these data, they conclude that the split element takes the shape of a well-formed complete independent noun phrase. However, if I understand their suggestion correctly, they relegate these inflectional properties to a post copying and deletion level, thus leaving aside the possibility that it is these properties (i.e. agreement, independent phrase) that are responsible for the existence of split constructions. The fact that the two parts of the noun phrase in a split construction behave like two independent XPs motivate an account like the one presented in this section, and if it is on the right track, the DD account simply becomes superfluous.

318

Eric Mathieu

6.2. Triggers and semantics In this final section I wish to concentrate on the motivation behind rising of (at least) one element in Spec-CP in the case of WH-elements. Here, I essentially follow Butler and Mathieu’s (2004b) analysis. Checking of the EPP feature associated with Q/WH appears to be universal (cf. Watanabe 1992; Chomsky 1995). However, the way such checking is achieved can vary from one language to another. It can be via Merge: insitu languages like Chinese have Q particles that can be base generated in the C domain; or via Move: in multiple WH questions at least one question word needs to raise to Spec-CP, while in single WH questions raising of the single WH element is necessary. Since French does not have question particles, movement of material susceptible to check the EPP feature in C is standardly required for a question interpretation. That is, minimally, a WH element should raise. The case of split combien constructions is a good illustration of this. However, it is also possible for no morphological element to be present in Spec-CP, as (77a) illustrates. In such cases a special intonation is required, which wclaim satisfies the visibility requirement for the interrogative. That is, the special intonation pattern signals the presence of a null operator that is able to check the EPP feature associated with Q/WH (cf. Cheng and Rooryck 2000). It follows that a WH in-situ question like (77a) gives rise to a split configuration, much as a split combien construction would (77 b). Thus, visibility is not only achieved morphologically, but also phonologically. This shows that phonology is not completely irrelevant for the cases at hand. However, it must be stressed that the visibility requirement is encoded, and does not simply follow from phonology. (77) a. [Op Elle a vu qui]? she has seen who ‘Who has she seen?’ b. Tu as lu combien de livres? you have read how-many of books ‘How many books have you read?’ That a special intonation is required for signaling interrogation is independently motivated. In French, alongside the possibility of inserting a Q particle (est-ce que) (78a) and inverting the subject and the verb (78b), there is a third option for asking a yes-no question, that of using a rising tone (78c):

Hyperbaton and Haplology

319

(78) a. Voulez-vous un café? want-you a coffee b. Est-ce que vous voulez un café? Q particle you want a coffee c. Vous voulez un café? you want a coffee ‘Do you want a coffee?’ In Japanese ga is a marker that indicates that the subject represents new information. On the other hand, wa is used for old information (Kuno 1972). In (79a), the generic subject counts as old information. In (79b) John is treated anaphorically, thus it represents old information. (79c) is odd, unless it is contrastive, e.g. ‘rain is falling, not snow.’ (79) a. Kuzira wa honnyuu-doobutu desu. whale WA mammal is ‘A whale is a mammal.’ b. John wa watakusi no John WA I ’s ‘John is my friend.’

tomodati desu. friend is

c. #Ame wa hutte imasu. rain WA falling is ‘It is raining.’ In (80), ga is for descriptive use or represents exhaustive interpretation/new information. (80)

John ga kimesita. John GA came ‘John came.’

Another illustration comes from Berber. The item ay is used as a focus marker in that language (Ouhalla 1990). (81) a. MOHAND ay-ushi-gh i. Mohand AY-saw.1SG ‘I saw MOHAND.’ b. Nni-gh-ak qa TIN’ASHIN ay-ushi-gh Mohand. said-SG-you that money AY-gave.1SG to Mohand ‘I told you that I gave Mohand MONEY.’

320

Eric Mathieu

Hungarian is a language where focus is encoded syntactically. In that language, focused phrases move to the Comp domain, as shown by (82a), and cannot remain in-situ, as illustrated by (82b). In addition, the focused phrase must be adjacent to the verb as shown by (82c). (82) a. ATTILÁT szereti Emöke. Attila-ACC love-PRES.3SG Emöke-NOM b. Emöke szereti ATTILÁT. c. ATTILÁT Emöke szereti. ‘Attila loves Emöke.’

(Puskàs 2000: 2)

In English, focus is encoded phonologically. It involves special stress patterns. In (83b), Mary receives default main stress, whereas in (83b), Mary is heavily accented. (83) a. Who does John like? b. John likes MARY. (84) a. John likes Mary and Lucy, right? b. Well, he likes MARY. In sum, asking a question is like focus, it can be achieved morphologically, syntactically or phonologically. The question that arises is why there should be a special marking for questions (or focus generally). The hypothesis is that the visibility requirement associated with the Q/WH feature is exactly this: the speaker has to make explicit that this criterion is met. That is, asking an interrogative, just like focus, cannot be carried out covertly. Information/ communication is necessarily overt. In the case of split constructions, the bare extracted element checks the relevant feature: the visibility requirement has been met and the competent addressee has the relevant cue in order to interpret the sentence (see also Platzack 1998). I now turn to the question as to why pied-piping is required when an intervener is present? Here I follow Butler and Mathieu (2004a) – B&M, henceforth. Like Honcoop (1998), B&M take the dynamic semantics stance on interpretation as their starting point. This is basically the observation that, for a discourse to be interpretable, information pertaining to the discourse itself is needed. They call this usage information. An archetypal example is the contrast between (85) and (86) (due to B. Partee). (85) I dropped ten marbles and found all of them, except for one. It is probably under the sofa.

Hyperbaton and Haplology

321

(86) I dropped ten marbles and found only nine of them. ??It is probably under the sofa. The first sentences in (85) and (86) are truth-conditionally equivalent: they provide the same information about the world. But unlike (85), the first sentence of (86) does not give sufficient usage information to allow for the resolution of the anaphoric link. B&M adopt a perspective on usage information due to Dekker (2002) that sees it as providing information as to the possibly intended referents that support an utterance (the ideas of Dekker 2002 are in turn based on ideas from Kamp 1990; Van Rooy 1997; Stalnaker 1998 and Zimmermann 1999). Dekker shows how adding such general pragmatic information to a classical, referentially based notion of meaning has the effect of deriving a dynamic semantics (and this without any shift in the notion of meaning). For example, the occurrence of one in the first sentence of (85) comes with an š usage instruction and so is assumed to make available an ‘intentionally present’ individual (i.e. the lost marble) that the pronoun in the second sentence can take as its referent. In contrast, (86) is bad because there is no such intentionally present individual. Note that the referential intentions associated with the use of indefinites like one in (84) are to be attributed to the speaker, who is supposed to be able to support what she says. The hearer has no such requirements, and so can generally take an š occurrence to introduce a new subject. Now consider the speaker’s support for (87). (87) There isn’t a person in sight. This can be characterized as a ban on an update of her state with (88). (88) A person is in sight. Consequently, the requirement of referential intentions is replaced by the requirement to have evidence that she, as a hearer, bans any update with (88), no matter who, with whatever intentions, would try to attempt to bring about such an update. This situation arises because the š usage instruction a person makes available is assumed to only be visible from within the scope of the negation. From outside negation’s scope, š is opaque, accounting for the absence of referential intentions, which in turn accounts for the absence of any potential anaphoric pick up. B&M argue that there is a related visibility requirement on the C domain for interrogative sentences to meet. They propose that this arises because

322

Eric Mathieu

cross-linguistically interrogatives come with a bare interrogative operator Q base generated as the highest element. Thus, all interrogatives give rise to split configurations. To have values under question to impart to variables, Q is taken to rely on the presence of ‘wh’ usage instructions. Whenever WHphrases carry wh, they must be in a visible relation with the C domain, in the sense of not falling underneath a scopal operator, for Q to function as a binder of WH variables. This explains why (89) is well-formed, but (90) is not. (89)

pas Op Combien de livresi n’as-tu Q wh of books NE-have-you scopal-op Q x[y[wh x = |y|] books(y)] Neg you_read(y)

lus ti? read-MAS.PL

pas lu ti de livres? (90) *Op Combieni n’as-tu NE-have-you scopal-op read of books Q wh Q Neg (you_read(y) [y[wh x = |y|books(y)]) (89) is bad, with Q left to ask a question without the support required to bind any variables. That is, (89) is incoherent. On the one hand, because of the intervening scopal operator, it fails to show any wh usage information, and thus no values under question are introduced (hence the plain Q in the interpretable representation), and on the other hand, it really does have a WHphrase primed to receive values under question (the free x). The approach generalizes as in (91) and can be seen as a rationalization of Pesetsky’s (2000) Intervention Effect Condition and other related proposals. (91) Op [Matrix … (*scopal_op)[ … [Restriction … usage-instruction … ] … ] … ] Stated thus, the approach is close to a reworking of Honcoop’s (1998) analysis, which can be pictured as in (92). (92) Op x[Matrix … (*scopal_op)[ … [Restriction …•y … ] … ] … and x equals dynamically bound y] Under this perspective, a usage instruction equates approximately to the function of a dynamic existential quantifier. That is, it ensures that a new value will be present in the evaluation. Cases of split-topics in German are to be treated differently, however. This is because what is stranded is an indefinite. The intervention facts that such constructions display easily fall out of systems of dynamic interpretation à la Honcoop (1998). On the other hand, the case of PP fronting in

Hyperbaton and Haplology

323

Serbo-Croatian and in other languages are amenable to the usage conditions analysis described above. With these considerations in mind, let us now turn to one last question (another that was posed at the outset): why raising of the nominal in French is ever possible if bare operator movement can satisfy the visibility requirement? B&M have established that intervention effects arise out of the interpretation procedure crashing, and so syntax itself cannot be sensitive to intervention effects. It follows that the optionality syntax displays follows from it blindly applying movement to meet coverage demands. For example, the full DP movement in (89) is necessary, else the speaker would lack the ability to ask the question, there being no grammatical split alternative ((90) is ruled out since it brings about an intervention effect). In contrast, for a question without an intervener, full movement is not required, as the speaker can ask the same question with a split-DP. It follows that syntax allows potentially vacuous movement ‘just in case’ the worst happens and an intervener is present in the structure.

7. Conclusion This paper was an attempt to link hyperbaton with haplology. However, I have reviewed a number of arguments that indicate that hyperbaton has nothing to do with haplology. It was argued that with respect to interrogatives, the split version is the default alternative. The trigger for movement is a standard strong +WH/Q feature in the C domain. If this assumption is correct, then it becomes a mystery why the nominal with which the bare operator is associated should ever raise, since economy is satisfied. Once a feature has been checked it is not longer accessible to the computation. Thus, raising of the stranded nominal could not be motivated. It was argued that raising of both the operator and the nominal is motivated by semantic considerations.

Acknowledgements I wish to thank Alastair Butler, Sophie Heyd, Ioanna Sitaridou, and two anonymous reviewers. All errors remain my own.

324

Eric Mathieu

Notes 1. In Classical Greek, it appears that Determiner Spreading was only possible when an adjective followed a noun, but the process was optional, and not obligatory like the case of Modern Greek. 2. The presence of de is usually required in many split constructions in French. 3. I concentrate here on Romance and ignore the differences between Germanic and Romance with regard to the NSR. 4. On the Nuclear Stress Rule, see also Reinhart (1995) and Selkirk (1995). 5. The fact that French allows both metrical invisibility and P-movement is seen a reflex of the fact that French is in a transitional stage of language change with respect to certain aspects of its prosodic properties. The metrical invisibility option belongs to one grammar, while the P-movement alternative belongs to another (on dual grammars, see Kroch 1989). 6. Reviewer #1 points out that these constructions are also possible in Italian. 7. The use of the singular suggests that these constructions are not analogous to partitive structures such as una de sas pinnas meas (‘one of my pens’). 8. The underlying structure M&S propose for split-DP constructions is closer to Fanselow’s earlier work on split-DPs. However, there are several crucial differences between his and M&S’s proposal. To account for split topic constructions, Fanselow assumes that, in German, two NPs are generated in the so-called Middlefield, one of which contains the empty pronoun pro. The other NP is nonreferential, denoting a property, and is coindexed with pro. For the pro NP to satisfy Principle C of the Binding Theory (which excludes that referential expression are A-bound), the non-referential NP raises to the sentence-initial nonargumental position. Although there are two XPs, Fanselow argues that the Case Filter is not violated because the nominal is adjoined to V° in order to get accusative case from the verb (it does not move there but is base-generated in that position). Thus, both NPs receive case. The stranded XP via government; the topicalized nominal via adjunction to V. As for the e-Criterion, the claim is that it is not violated because it applies only to term-denoting argument NPs. Non-argument NPs without a e-role are possible. Fanselow’s account faces several problem. As argued by Kuhn (1999), the reasons for case matching between the raised topicalized element and the remnant remain vague in Fanselow’s analysis. Although the two discontinuous elements form two separate units in split-topics, it has been observed that the two NPs have to agree in number, case and gender. The alternative solution proposed by M&S captures the agreement problem as well as the case problem in a straightforward manner, since the two XPs interact in a Spec-Head configuration where agreement and case properties can be satisfied. 9. In so-called N of N constructions, the nominal copula de surfaces in French. However, the head of the Small Clause is not spelt out (i). In English, both are available (translation of (i) and (ii)). On the other hand, as shown by (ii) in

Hyperbaton and Haplology

325

Modern Greek although the head of the SC is spelt out (as o, i.e. before jatros), the nominal copula is not morphologically marked (there is no element between vlakas and the second occurrence of to). According to den Dikken (1998) and Moro (1997), a copula is obligatory when predicate inversion arises. However, this does not appear to be the case in Greek. We discuss facts about the nominal and the clausal copula and its non obligatoriness in Greek in Mathieu and Sitaridou (2004b). (i) Cet idiot de docteur! that idiot of doctor (ii)

Aftos o vlakas o jatros! that a idiot a doctor ‘That idiot of a doctor!’ French is consistent in that respect in that the nominal copula also surfaces in bare adjectival constructions. (iii) Le grand, je veux de café latte! the tall I want of café latte ‘I want the TALL café latte (i.e. not the small one).’

References Abney, Stephen 1987 The English Noun Phrase in its Sentential Aspect. Doctoral diss., MIT. Alexiadou, Artemis and Chris Wilder 1998 Adjectival Modification and Multiple Determiners. In Possessors, Predicates and Movement in the DP, Artemis Alexiadou and Chris Wilder (eds.), 303–332. Amsterdam: Benjamins. Androutsopoulou, Antonia 1994 The Distribution of the Definite Determiner and the Syntax of Greek DPs. Proceedings of Chicago Linguistic Society 30: 16–19. den Besten, Hans and Gert Webelhuth 1990 Stranding. In Scrambling and Barriers, Günter Grewendorf and Wolfgang Sternefeld (eds.), 77–92. Amsterdam: Benjamins. Boeckx, Cédric 2003 Islands and Chains: Resumption as Stranding. Amsterdam: Benjamins. Butler, Alastair and Eric Mathieu 2004a The Syntax and Semantics of Split Constructions: A Comparative Study. Basingstoke/New York: Palgrave-MacMillan. 2004b Split-DPs, Generalized EPP, and Visibility. To appear in MIT Working Papers in Linguistics.

326

Eric Mathieu

Browne, Wayles 1976 Two WH-fronting Rules in Serbo-Croatian. Juznoslovenski filolog 32: 194–204. Cavar, Damir 1999 Aspects of the Syntax-Phonology Interface. Doctoral dissertation. University of Potsdam. Cheng, Lisa and Johann Rooryck 2000 Licensing WH-in-situ. Syntax 3: 1–19. Chomsky, Noam 1971 Deep Structure, Surface Structure and Semantic Interpretation. In Semantics: an Interdisciplinary Reader in Philosophy, Linguistics and Psychology, Daniel Steinberg and Leon Jakobovits (eds.), 183–216. Cambridge: Cambridge University Press. 1976 Conditions on Rules of Grammar. Linguistic Analysis 2: 303–351. 1995 The Minimalist Program. Cambridge, MA: MIT Press. Cinque, Guglielmo 1980 On extraction from NP in Italian. Journal of Italian Linguistics 5: 47–99. Demirdache, Hamida 1992 Resumptive Chains in Restrictive Relatives, Appositives and Dislocation Structures. Doctoral dissertation, MIT. 1997 Dislocation, Resumption and Weakest Crossover. In Materials on Left Dislocation, Elena Anagnostopoulou, Henk van Riemsdijk and Frans Zwarts (eds.), 193–231. Amsterdam, John Benjamins. Denniston, J. D. 1997 Greek Prose Style. Bristol: Bristol Classical Press. Dekker, Paul 2002 Meaning and use of indefinite expressions. Journal of Logic, Language and Information 11: 141–194. Devine, A.M. and Laurence D. Stephens 2000 Discontinuous Syntax: Hyperbaton in Greek. Oxford: Oxford University Press. Dikken, den Marcel 1998 Predicate Inversion in DP. In Possessives, predicates and movement in the DP, Artemis Alexiadou and Chris Wilder (eds.), 177–214. Amsterdam: John Benjamins. Doron, Edit 1982 On the Syntax and Semantics of Resumptive Pronouns. In Texas Linguistics Forum 19, Robert Bley-Vroman (ed.), 1–48. Austin, Texas. Fanselow, Gisbert 1988 Aufspaltung von NP und das Problem der ‘freien’ Wortstellung. Linguistische Berichte 114: 91–113.

Hyperbaton and Haplology

327

Fanselow, Gisbert and Damir Cavar 2001 Remarks on the Economy of Pronunciation. In Competition in Syntax, Gereon Müller and Wolfgang Sternefeld (eds.), 107–150. Mouton de Gruyter, Berlin. 2002 Distributed deletion. In Theoretical Approaches to Universals, Artemis Alexiadou (ed.), 66–107. Amsterdam: John Benjamins. Fanselow, Gisbert and Anoop Mahajan 2000 Towards a Minimalist Theory of WH-expletives, WH-copying, and Successive Cyclicity. In Wh-Scope Marking, Uri Lutz, Gereon Müller and Arnim von Stechow (eds.), 195–230. Amsterdam: John Benjamins. Giorgi, Alessandra and Giuseppe Longobardi 1991 The Syntax of Noun Phrases: Configuration, Parameters and Empty Categories. Cambridge: Cambridge University Press. Goldsmith, John 1979 Autosegmental phonology. New York: Garland. Golston, Chris 1988 Phrasal Morphology in Homeric Greek. MA dissertation. University of California, Los Angeles. Grimshaw, Jane 1997 The Best Clitic: Constraint conflict in morpho-syntax. In Elements of grammar, Liliane Haegeman (ed.), 169–199. Kluwer Academic Publishers. Hiemstra, Inge 1986 Some Aspects of WH-questions in Frisian. Western European Language Evolution 8: 97–110. Höhle, Tilman 1990 Die w-w-Konstruktion im Deutschen. Universität Tübingen. Honcoop, Martin 1998 Dynamic Excursions on Weak Islands. The Hague: Holland Academic Graphics. Horrocks Geoffrey and Melita Stavrou 1987 Bounding Theory and Greek Syntax: Evidence for wh-movement in NP. Journal of Linguistics 23: 79–108. Jackendoff, Ray 1972 Semantic Interpretation in Generative Grammar. Cambridge, MA: MIT Press. Jones, Michael 1993 Sardinian Syntax. Oxford: Routledge. Kamp, Hans 1990 Prolegomena to a structural theory of belief and other attitudes. In Propositional Attitudes: The Role of Content in Logic, Language and Mind, C. Anthony Anderson and Joseph Owens (eds.), 27–90. Stanford: CSLI.

328

Eric Mathieu

Kayne, Richard 2002 Pronouns and Their Antecedents. In Derivation and Explanation in the Minimalist Program, Samuel Epstein and David Seely (eds.), 133–166. Oxford: Blackwell. Kroch, Anthony 1989 Reflexes of Grammar in Patterns of Language Change. Journal of Language Variation and Change 1: 199–244. Kuno, Susumu 1972 Functional sentence perspective: A case study from Japanese and English. Linguistic Inquiry 3, 269–320. Kuhn, Jonas 1999 The Syntax and Semantics of Split NPs in LFG. In Empirical Issues in Formal Syntax and Semantics 2, Thésus, Francis Corblin, Carmen Dobrovie-Sorin and Jean-Marie Marandin (eds.), 145–166. The Hague. Leben, William 1973 Suprasegmental phonology. Doctoral dissertation, MIT. Manolessou, Io 2000 Greek Noun Phrase Structure: A Study in Syntactic Evolution. Doctoral dissertation, University of Cambridge. Mathieu, Eric 1999 French WH in situ and the Intervention Effect. UCL Working Papers in Linguistics 11: 441–472. 2002 The Syntax of Non-Canonical Quantification: A Comparative Study. Doctoral dissertation, University College London. 2004 The Mapping of Form and Interpretation: the case of optional WHmovement in French. In Lingua as part of a special issue ‘Variation in Form versus Variation in Meaning’. Guest editors: Marco Haverkort, Helen de Hoop, Maurits van den Noort. Volume 114/9–10: 1090–1132. Mathieu, Eric and Ioanna Sitaridou 2002 Split DP syntax and defocalised nominals: a view on the syntax-discourse interface. Talk presented at GLOW 2002, Workshop on ‘The Syntax-Discourse Interface: Linguistic and Psycholinguistic Issues.’ April 5–7, Utrecht. 2004a Noun Ellipsis, Determiner Spreading and Split-DPs. Ms., University College London / University of Hamburg. 2004b Split-WH constructions in Classical and Modern Greek: A Diachronic Perspective’. In Grammaticalization and Parametric Change, Montserrat Batllori and Francesc Roca (eds.). Oxford University Press. McCloskey, Jim 1990 Resumptive Pronouns, A’-Binding and Levels of Representation in Irish. In The Syntax and semantics of the Modern Celtic Languages, Syntax and Semantics 23. New York, Academic Press.

Hyperbaton and Haplology

329

Moro, Andrea 1997 The Raising of Predicates: Predicative Noun Phrases and the Theory of Clause Structure. Cambridge: Cambridge University Press. Müller, Gereon 1998 Incomplete Category Fronting. Dordrecht: Kluwer. Neeleman, Ad and Tania Reinhart 1998 Scrambling and the PF-interface. In The Projection of Arguments: Lexical and Compositional Factors, Miriam Butt and Wilhelm Geuder (eds.), 309–353. Chicago: CSLI Publications. Neeleman, Ad and Hans van de Koot 2002 Syntactic Haplology. Ms. University College London, to appear in SynCom Project. Nunes, Jairo 2001 Sidewards Movement. Linguistic Inquiry 32: 303–344. Obenauer, Hans-Georg 1976 Etudes de Syntaxe Interrogative du Français. Tübingen: Niemeyer. 1983 Une Quantification Non-Canonique: La Quantification à Distance. Langue Française 58: 66–88. 1994 Aspects de la Syntaxe A-Barre. Thèse de Doctorat d’Etat. Université de Paris VIII, Paris. Ouhalla, Jamal 1990 Focusing in Berber and Circassian and the V2 Phenomenon. Talk presented at University College London. Pesetsky, David 2000 Phrasal Movement and Its Kin. Cambridge, MA: MIT Press. Plag, Ingo 1999 Morphological Productivity. Structural Constraints in English Derivation. Berlin/New York: Mouton de Gruyter. Platzack, Christer 1998 A visibility Condition for the C domain. Working Papers in Scandinavian Syntax 61: 53–99. Puskás, Genoveva 2000 Word Order in Hungarian: The Syntax of A-bar positions. Amsterdam: John Benjamins. Reinhart, Tanya 1995 Interface Strategies. OTS Working Papers. Utrecht University. Rizzi, Luigi 1990 Relativized Minimality. Cambridge, MA: MIT Press. Sekerina, Irina 1997 The Syntax and Processing of Scrambling Constructions in Russian. Doctoral dissertation, City University, New-York. Selkirk, Eloise 1995 Sentence Prosody: Intonation, Stress, and Phrasing. In John Goldsmith (ed.), The Handbook of Phonological Theory. Oxford: Blackwell, pp. 550–569.

330

Eric Mathieu

Shlonsky, Ur 1992 Resumptive Pronouns as Last Resort. Linguistic Inquiry 12: 679–682. Stalnaker, Robert 1998 On the representation of context. Journal of Logic, Language and Information 7: 3–19. Stowell, Tim 1991 Determiners in NP and DP. In Views on Phrase Structure, Kathleen Leffel and Denis Bouchard (eds.), 37–56. Dordrecht: Kluwer Academics. Szabolcsi, Anna 1983 The possessor that ran away from home. The Linguistic Review 3: 89–102. de Swart, Henriëtte 1992 Intervention effects, Monotonicity and Scope. Proceedings of SALT, Vol. 2: 387–406. Columbius, Ohio: Ohio State University Press. Van Geenhoven, Veerle 1998 Semantic Incorporation and Indefinite Descriptions: Semantic and Syntactic Aspects of Noun Incorporation in West Greenlandic. Stanford: CSLI. Van Rooy, Robert 1997 Attitudes and Changing Contexts. Doctoral dissertation, IMS, Stuttgart. Watanabe, Akira 1992 Subjacency and Structure movement of WH-in-situ. Journal of East Asian Linguistics 1: 255–291. Yip, Moira 1998 Identity avoidance in phonology and morphology. In Morphology and its relation to Phonology and Syntax, Steven LaPointe, Diane Brentari and Patrick Farrell (eds.), 216–246. Stanford: CSLI Publications. Zec, Drage and Sharon Inkelas 1990 Prosodically Constrained Syntax. In The Phonology-Syntax Connection, Sharon Inkelas and Draga Zec (eds.). Chicago: The University of Chicago Press. Zimmermann, Ede 1999 Remarks on the epistemic role of discourse referents. In Logic, Language and Computation II, Lawrence Moss, Jonathan Ginzburg, and Maarten de Rijke (eds.), 513–540. Dordrecht: Kluwer Academic Publishers. Zubizarreta, Maria-Luisa 1998 Prosody, Focus, and Word Order. Cambridge, MA: MIT Press.

On Scrambling as Defocusing in German and West Germanic László Molnárfi

Abstract The present paper investigates the problem of the extent to which the discourse functional and prosodic dimension of word order variation can be integrated into a formal clausal syntax of (West) German(ic). In contradistinction to previous theories, the problem of discourse-syntax mapping will be looked at from a different angle: from the perspective of defocusing. Supported by morphological evidence from modern spoken Afrikaans, I will argue for the existence of designated destressing positions in West Germanic, called antifocus. This novel clausal relation will be shown to represent a a violable defocusing license for anaphoric elements, triggering optional movement of pronouns and definites to the left. Postulating a prosodic trigger for scrambling allows one to derive the discourse functional correlates of word order variation as the interaction of the licensing requirements of the antifocus feature and Cinque’s (1993) stress asignment rules. Under this view, languages may employ different information packaging strategies in terms of word order disctinctions, leading to a Hungarian type of focus prominence, or the West Germanic type of antifocus prominence.

1. Introduction Movement approaches to word order variation in West Germanic regard scrambling as reordering of elements proceeding from an underlying, base word order, towards a derived word order. The tacit premise of derivational theories is that the base word order (which is heuristically identifiable in terms of wide focus projection, cf. Höhle 1982) represents the contextual maximally unmarked order, while reordering always has some semantic or pragmatic effect on the output, destroying contextual unmarkedness and yielding refocusing effects (cf. among others, Abraham 1995a, 1997, 1999, 2003; Haider & Rosengren 1998; or Doherty 2001). If so, the effect achieved

332

László Molnárfi

by the reordering will identify the common grammatical property shared by all scrambled elements and as such constitute a possible trigger for word order variation (cf. Meinunger 2000; Diesing/Jelinek 1995; Jäger 1995; or Frey 2000). Recently, important conceptual and empirical objections have been put forward against such accounts from a minimalist point of view, arguing for a base generation approach to srcambling (cf. Zwart 1997a; Reinhart/Neeleman 1998 or Fanselow 2003). The objections raised concern, (i) the assumption that derived word orders always have to be equated with marked word orders (cf. Molnárfi 2002 and Fanselow 2003), (ii) the status of triggered scrambling movement, which, if exists, has been claimed to have to involve altruistic movement (Fanselow 2003), and (iii) the theoretical status of optional movement as implied by derivational approaches in the light of principles of derivational economy (cf. Zwart 1997a). The present paper defends a derivational approach to word order variation, extending the views unfolded in Abraham (1997), Molnárfi (2002, 2003) and Abraham & Molnárfi (2002). Investigating empirical data from German (with side-glances to Dutch and Afrikaans), the main argument presented here will be that, in order to maintain a derivational approach to word order variaton, one needs to rethink some basic assumptions concerning the relation between base and derived word orders. First, following (Molnárfi 2002), I will argue that reordering operations such as scrambling do not lead necessarily to the emergence of more marked word orders. Given that the domain of wide focus must not contain elements belonging to the presupposition structure, base word orders, being a function of lexical mapping only, do not always optimally satisfy discourse licensing conditions imposed on the surface order. If driven prosodically, derived word orders involving referential definites may in fact satisfy the criterion of contextual unmarkedness better than base word orders do. Second, I will argue that the common property all scrambled elements share is not a pragmatic feature such as familiarity or topichood, but simply their lack of focus. The function of scrambling is to enable the proper computation of focus by evacuating presupposed definites from the focus domain. Following Molnárfi (2002, 2003, 2004) and Abraham/Molnárfi (2002), the corresponding trigger will be called antifocus, representing a novel clausal relation for destressing of referential definites. While antifocus movement obviously affects the computation of the focus domain, such movement is never altruistic, in the sense that focalization of any element not participating in the defocusing chain follows independently from the assignment conditions of nuclear stress.

On Scrambling as Defocusing in German and West Germanic

333

Third, I will argue that there is no optionality of scrambling as with respect to chain formation. Apparent optionality arises as, under certain circumstances, it is not only the head position, but also the base position of a non-trivial derivational chain that may be spelled out (cf. Molnárfi 2003). The spell-out rules for antifocus licensing are determined by external eligibility conditions, which are directly relevant for the discourse interpretation of the elements involved in chain formation. While the spell-out of the higher copy of the scrambling chain licenses destressing and discourse linking of the involved element, the phonological realization of the base copy can be exploited for deictic or contrastive interpretation. In a typological perspective, the question arises whether there exist crosslinguistic correlates that uniquely identify a language as a scrambling-language. Among others, two such criteria have been proposed in the literature: strong (rich) case morphology (cf. Haider & Rosengren 1998 or Weiss 2001) and SOV-characteristics (cf. Abraham 1997 et passim and Haider & Rosengren 1998). I will investigate both, arguing that only the latter typological criterion is relevant for word order variation. At least for West Germanic, the existence of a verbal bracket alone suffices to license word order variation in the sense investigated here, rich case morphology being only a necessary condition of scrambling in more complex predicates. This paper is structured as follows. In section 2, I will investigate the empirical range of middle field scrambling, clarifying some terminological issues. In section 3, I will review some of the basic assumptions of derivational and non-derivational approaches to word order variation. Sections 4, 5 and 6 deal with critical aspects of such strategies, scrutinizing issues concerning the empirical identification of base word orders, the possible trigger of word order variation and the pragmatic correlates of scrambling. In section 7, I will investigate the relation between movement and focus assignment in the middle field, while section 8 will identify and establish a novel clausal relation as the common property shared by all scrambled elements: antifocus. Section 9 investigates the typological correlates of word order variation. Section 10 gives the conclusions.

2. The Empirical Range of Scrambling A distinct feature of West Germanic word order is the relative freedom of argument linearization in the sentence, a phenomenon that, after Ross (1967), has become known as scrambling in the literature. In recent years, the empirical range of scrambling has considerably varied to exclude, or include,

334

László Molnárfi

operations such as pronominal fronting, “Dutch scrambling” (Adv-DO vs. DO-Adv-patterns), or Scandinavian object shift. In order to find an adequate tertium comparationis of word order variation (cf. Haider & Rosengren 1998 for an insightful discussion), some terminological clarification seems advisable. As (1a–b) show, the structural space between C and the base position of the lexical head V in SOV-Germanic, i.e. the middle field in topological terms, allows for the existence of alternative word order patterns. In (1a–b) such variation is schematized as a permutation between the elements X and Y: (1)

a. C° [ b. C° [

XY YX

V°] V°]

While the domain of alternative linearization of the elements X and Y has been shown to be clause-bound (cf. Haider & Rosengren 1998: 2f.),1 scrambling may apply across different syntactic categories of elements as far as they represent arguments. In this spirit, Y may be a DP, a PP or a CP as demonstrated in (2a–f) ((2f) is after Haider & Rosengren 1998: 10). The relevant scrambled constituents in (2) are put in italics: (2)

a. dass [Peter [das Buch] gelesen hat] that Peter [the book] read has b. dass [[das Buch] Peter gelesen hat] c. dass [Peter [mit der Frau] gesprochen hat] that Peter [with the woman] talked has d. dass [[mit der Frau] Peter gesprochen hat] c. dass [ja niemand [dass die Erde rund ist] bezweifeln würde] that PRT nobody [that the earth round is] doubt would d. dass [ja [dass die Erde rund ist] niemand bezweifeln würde]

The syntactic status of X in (1a–b) is more controversial. In a broader view on scrambling, X may also be a left-edge marker, i.e. a modal particle or adverbial element, which is traditionally taken to occupy a fixed (and perhaps hierarchically organized) position in the tree (Pollock 1989, and recently Cinque 1999). Such a view has a long tradition in Dutch grammar writing (cf. Corver & van Riemsdijk 1994, Neeleman & Reinhart 1998 or de Hoop 1997, 2000a, 2000b).

On Scrambling as Defocusing in German and West Germanic

335

In “Dutch” scrambling, the permutation of X and Y is indicated by the relative distribution of Y with respect to the fixed position of the left edge marker X in (1a) and (1b). Witness the following sentences from Dutch, where X (in italics) corresponds to a sentence adverb, while Y to a definite direct object: (3)

a. dat ik [gisteren de man heb gezien] that I yesterday the man have seen ‘that I saw the man yesterday’ b. dat ik de man [gisteren heb gezien ] that I the man yesterday have seen

In (3a–b), word order variation occurs with respect to the assumedly fixed position of the sentence adverb gisteren, allowing the direct object to preceed or to follow the left-edge marker. Similar distributional distinctions have been observed to hold for personal pronouns (and, to a lesser degree, for definite full nouns) in Scandinavian languages, where the permutation of X and Y correlates with overt verb movement and is referred to as (Scandinavian) object shift (cf. Holmberg 1998, 1999). In a more restricted theory of scrambling, word order variation is confined to argumental elements only (Haider & Rosengren 1998: 23). That is, (1b) represents an instantiation of scrambling only if X itself is an argument. This disqualifies (3b) and Scandinavian object shift as a scrambling operation. In the present paper, I will adopt the broader view on scrambling. While there is no compelling theoretical argument in favor of the assumption that the permutation between X and Y should involve argumental elements only, methodological considerations in the spirit of Occham’s Razor seem to favor a unified treatment of (2a–f) and (3a–b). The assessment of the two models hinges on the theoretical assessment of the status of adverbs and particles as left-edge markers. On the one hand, a restricted view on scrambling seems to give up the main methodological insight that adverbial material may be used to uniquely identify syntactic positions and as such to detect movement of elements across maximal projections (cf. especially Haider & Rosengren 1998: 61f. and Zwart 1997a: 91f.). This has far-reaching consequences not only for word order variation, but also for the reassessment of traditional analyses of overt verbal movement in general (cf. Pollock 1989, Chomsky 1993, Holmberg 1999). On the other hand, even restricted scrambling theories are bound to rely on the shibboleth-function of left-edge markers such as

336

László Molnárfi

modal particles (cf. Haider & Rosengren 1998: 10f.) to some extent. Otherwise, the domain of scrambling chain-formation could not be restricted in any meaningful way within the middle field. In this spirit, Haider & Rosengren (1998: 10f.) use modal particles to demonstrate that the chain formation in (2a–f) takes place within the domain of the lexical head, that is, within VP. It is thus by far not clear whether and to what extent modern syntax theory can do without the empirical foothold of left-edge markers.2 This is the methodological argument. The empirical argument for a broader view on scrambling comes from the observation that word order variation in the spirit of (3a–b) may correlate with regular morphological patterns in West Germanic. In Afrikaans, definite direct objects to the right of sentence adverbs are marked by the particle vir, while such definites to the left of sentence adverbs surface without any morphological marking (cf. Molnárfi 1997, 1999: 78f. and 2002: 1127f. for a detailed discussion): (4)

a. dat ek [VP gister vir die man/ vir Jan/ vir hom gaan sien het] that I yesterday for the man/ for Jan/ for him gone seen have ‘that I went to see the man/Jan/him yesterday’ b. dat ek die posbode/Jan/hom [VP gister gaan sien het]

Note that the use of vir is obligatory with non-scrambled personal pronouns and proper names, while it is strongly preferred with full definite nouns if to the right of sentence adverbs (cf. Ponelis 1979: 202f.). The necessary (but not sufficient) grammatic corrrelate for the licensing of vir is lexically or category-determined definiteness. Irrespectively of their distributional or semantic status, indefinite NPs can never be marked morphologically in Afrikaans (cf. Molnárfi 2002: 1130f.). Given that vir, if used in two-place predicates, does not exhibit the characteristic properties of prepositional elements (cf. Molnárfi 1999 for discussion), but, in many respects, behaves like a discourse sensitive morpheme, it is more than plausible to relate its distribution to the different derivational positions the direct object may take up with respect to the sentence adverb in (4a) and (4b). The alternative is to relate the morphological pattern in (4a–b) to the higher or lower attachment of the sentence adverb (cf. Zwart 1997a: 91, Haider & Rosengren 1998: 61f.). However, it is not clear how free adverbial attachment could possibly trigger morphological effects on the definite object. Note that, whatever model is chosen, the notion of scrambling should not make direct reference to adverbial material. Word order variation is also

On Scrambling as Defocusing in German and West Germanic

337

allowed in more complex predicates, independently of the presence or absence of left-edge markers (cf. also 2 a–f): (5)

a. dass ich dem Studenten (gestern) das Buch gegeben habe that I the studentDAT yesterday the bookACC given have b. dass ich das Buch (gestern) dem Studenten gegeben habe that I the bookACC yesterday the studentDAT given have

That is, the presence of certain types of adverbs is, at best, an important shibboleth, but never a structural perequisite of word order variation. While word order variation as exemplified in (3a–b) is attestable in all West Germanic (and, with some additional restrictions, in Scandinavian too), scrambling in more complex predicates as in (5a–b) has only been observed in German. Contrary to Haider & Rosengren (1998: 2f.), however, I will not infer from this fact that (3a–b) and (4a–b) involve fundamentally different kinds of constructions. Rather, given the similar discourse-functional and prosodic correlates of (3a–b) and (5a–b), I will regard word order variation as in (5a–b) to be a natural extension of (3a–b) in more complex predicates, licensed only in languages where the SOV-characteristic correlates with rich (distinctive) case morphology. Among West Germanic, only German fulfills such typological criteria. That is, word order variation in more complex predicates is excluded by independent reasons in other West Germanic, and (5b) does not substantiate the only empirically relevant form of scrambling. Clearly, such an assumption is methodologically desirable. A theory of word order variation in the spirit of (3a–b) will have to say something about (5a–b) as well, and, if we are on the right track, no separate account for word order variation in two-place and three-place predicates is needed. The relevant contrasts between (3a–b) and (5a–b) will fall out as reflexes of independent factors. In the sense used here, scrambling refers to every instance of word order variation where relative reordering of arguments (including reordering with respect to sentence adverbs) in the middle field occurs. This excludes optional topicalization movement of maximal projections to SpecCP as a scrambling operation, but not pronominal fronting, which, following Lenerz (1993) and Molnárfi (2002, 2004), I will take to be an instance of word order variation in the sense discussed here.

338

László Molnárfi

3. Basic Strategies of Dealing with Word Order Variation The major question is of how the permutation in (1a) and (1b) is to be assessed in a theory of grammar. The null hypothesis is that scramblig is not rooted in core grammar. If so, by representing a post-grammatically determined stylistic variation, (1a) and (1b) are not interesting for syntactic theorizing in any relevant sense. This is the view taken by Ross (1967), de Hoop (1997, 2000a, 2000b) and also by Chomsky (1995, 1998). However, scrambling has been shown to systematically interact with structurally relevant conditions on scoping and binding (cf. Haider & Rosengren 1998: 4f.) as well as with conditions on the assignment of nuclear stress and focus-background structure (cf. among others, Abraham 1997 and Molnárfi 2002). Clearly, by placing word order variation into the unsystematic domain of pragmatics such important correlations would be lost for grammar theory. This calls for a syntactic ac of word order variation. For a long time, the prevailing syntactic account of scrambling has been that (3b) should be derived from (3a) by optional movement of the XP to the left, the position of the adverb between the verb and direct object in (3b) indicating that the latter is not in its base position. This is what I will call the derivational hypothesis: (6)

a. dass ich den [VP gestern [VP den Mann gesehen habe]] b. dass ich den Manni [VP gestern[VP ti gesehen habe]]

If gestern is taken to mark the VP-boundary (cf. Pollock 1989), the serialization patterns in (6a–b) indicate optional movement of the direct object across the sentence adverb to the VP-external domain. Analyses in a similar spirit have been proposed for the theoretical modelling of object shift in the Scandinavian languages (cf. Chomsky 1993; Holmberg 1999). Note that the tacit premise of an optional object shift analysis is threefold: (i)

There is a base word order identifiable on empirical grounds (cf. Höhle 1982).

(ii)

The verb and its objects form a core constituent that excludes all adverbial material (cf. Chomsky 1965).

(iii) Certain types of adverbs have a fixed attachment site in the tree (cf. Pollock 1989). In (5a) and (5b) this is the maximal projection of the verb.

On Scrambling as Defocusing in German and West Germanic

339

Taking this for granted, the discussion has mainly centred around the question of what kind of position the scrambling of the XP in (6b) targets (an A or an A’-position) and what the optional object shift is triggered by (cf. among others Bennis & Hoekstra 1984; Vanden Wyngaerd 1989; Lenerz 1993; Abraham 1997; de Hoop 1992, 1997; Diesing 1992; Heim 1982; Corver & van Riemsdijk 1994; Reinhart 1996 and also Neeleman & Reinhart 1998 for an overview of the relevant literature). Given that there are clear markedness contrasts with respect to the discourse functional use of the alternative word orders (cf. Abraham 1997 or Neeleman & Reinhart 1998), scrambling has often been related to the licensing of some pragmatic feature. In this spirit, word order variation can be taken to be a function of the pragmatic property the scrambled element is associated with in the lexicon, typically involving checking of features such as [+topic], [+familiar] or [+presupposed] (cf. Diesing & Jelinek 1995; Jäger 1995; Adger 1993; Reinhart 1996; or Frey 2000 and Meinunger 2000). Whatever trigger for the rearrangement is identified, one important theoretical consequence of the derivational view is that syntax has to allow overt movement to be optional. If the deviating word order [Y X] in (1a) is derived from the underlying basic order [X Y] in (1b) by movement of Y across X [Yi X ti], it is clear, that, contrary to movement driven by case or EPP, scrambling of Y cannot be a forced operation. The existence of such free transformations is at odds with basic assumptions of the Minimalist Program (cf. Chomsky 1995, 1998). More precisely, the derivational link traditional approaches predict with respect to (6a) and (6b) is in conflict with the spirit of derivational economy, the methodological point of departure for most Chomskyan syntacticians. In its most radical form, such methodology demands the full exclusion of optionality from syntax (see Chomsky 1995, Zwart 1993, 1997a/b or Neeleman & Reinhart 1998). Movement is seen here as a costly operation, which is either forced in morphological terms of feature checking (cf. Last Resort) or does not take place overtly at all (cf. Procrastinate). There is thus no optionality in terms of feature checking, but only with respect to the question of whether feature checking takes place covertly, i.e. after Spell-out, or overtly, pied-piping phonological material in syntax. In this spirit, there cannot be any optional XP-movement to the VP-external domain. Consequently, a derivational relation between (6a) and (6b) is impossible.3 This is what I will call the non-derivational hypothesis. Such minimalist methodology leaves two options for a corresponding theoretical model of word order variation. Either one has to face the problem of explaining non-forced overt movement in a derivational framework, or the idea of an underlying base word order has to be given up and any

340

László Molnárfi

derivational (movement) relation between the alternative word orders in (1a) and (1b) must be excluded. In the former case, only technical solutions have been put forward so far, assuming the optional assigment of a vacuous scr-feature (or, recently, EPP), which triggers or prevents movement in accordance to its presence or absence in the numeration (cf. Müller 1996, Grewendorf & Sabel 1999 or Grewendorf 2002: 197f.). In the latter case, the serialization patterns can be taken to be freely base generated by the computational system and the resulting prosodic/discourse functional distinctions are read off as interface effects of the alternative merging-structures (Neeleman & Reinhart 1998 or Fanselow 2003). Derivational and non-derivational approaches are also incompatible in the sense that they take a fundamentally different theoretical position with respect to the question of how sentence structure is shaped. While minimalistically inspired solutions see the clausal architecture as a function of the licensing requirements of morpho-syntactic features attracted by matching functional projections, derivational approaches to word order variaton allow a (certain degre of) discourse configurationally determination of sentence structure. In a broader perspective, this relates to the question of to which extent peripheral features such as focus and topic may enter the core computational system. In what follows, I will review some of the arguments base generation theories have raised recently against derivational approaches. Such objections concern (a) the identifiability of an underlying base order as the inputorder of derivation, (b) the existence of a grammatical trigger for scrambling and (c) the issue of optionality. I will argue that a derivational view can be maintained, if one is willing to rethink some basic assumptions regarding the relation of base and derived word orders.

4. On the Identification of Base Word Order as the most Unmarked Order Movement approaches to scrambling hinge on the assumption that one of the possible permutations constitutes the more basic, underlying order, from which the scrambling orders are derived by movement. In the spirit of Bierwisch (1988), Haider (1993) or Wunderlich (1997) such base word orders have been argued to be a direct function of lexical mapping, translating hierarchical relations holding between semantic primitives (formalized in terms of h-operators) into c-command hierarchies holding between arguments. The syntactic discharge of arguments is thus lexically (and eventually

On Scrambling as Defocusing in German and West Germanic

341

semantically) determined, something which can be supported by the close link of possible base word orders to different semantic verb classes in German (cf. Czepluch 1988, Haider 1993 or Molnárfi 1998).4 Obviously, one needs a clearly defined heuristic criterion to identify the base word orders as produced by lexical mapping. Such criterion has been proposed in terms of wide focus projection (cf. Höhle 1982). It is held that wide focus, allowing for the largest set of possible contextual links and triggering the smallest set of pressupositions, can only be associated with the base word order. Deviation from the base order (scrambling) inevitably blocks focus projection and leads to contextual markedness: derived word orders are pragmatically charged. In this spirit, base word orders can be equated with the pragmatically maximally unmarked word orders (cf. Abraham 1997). The crucial grammatical factor for identifying wide focus is the assignability of sentential accent (SA) in the spirit of Cinque (1993). Sentential accent is a default stress pattern, assigned blindly to the most deeply embedded complement in the tree, or, in the absence of such constituent, to the most deeply embedded lexical head. That is, unless overridden by pragmatically assigned marked stress patterns such as contrastive stress (CA), (neutral) stresss assignment is a function of syntactic structure. The element identified by sentence stress will allow focus projection, of which the domain may be larger than that of the actual focus exponent. Any deviating stress pattern will narrow the focus domain maximally, attracting contrastive focus and blocking focus projection (Haider & Rosengren 1998: 15f.). Given that the assignment of sentential stress is determined by depth of embededdnes, prosodic correlates allow direct clues about the derivational history of a given syntactic structure. The marked status of scrambled word orders can be derived by the interaction of assignment of nuclear stress and trace intervening effects (Haider & Rosengren 1998: 15f.). Witness the following sentences from German (SA is marked by italics, CA is marked by small capitals): (7)

a. dass [Peter dem Studenten das Buch vorgelesen hat] that Peter the studentDAT the bookACC out-read has ‘that Peter has read the book to the student’

(SA)

b. dass [Peter das Buchi dem STUDENTEN ti vorgelesen hat]

(CA)

c. dass [Peter den Studenten der Operation unterzogen hat] (SA) that Peter the studentACC the operationDAT subjected has d. dass [Peter der Operationi den STUDENTEN ti unterzogen hat] (CA)

342

László Molnárfi

The prosodic patterns associated with (7a–b) identify the word order N-D-A in (7a) as the basic word oder for the verb vorlesen, from which the N-A-Dorder in (7b) is derived by movement. This follows as sentential stress may only be assigned in (7a), but not in (7b). This in turn indicates that the definite object in (7a) is still in its base position, while in (7b) it has been removed from the position designated for stress assignment. As the trace movement has left behind in (7b) cannot be accented, the assignment of SA is blocked. This leads to the emergence of a pragmatically marked structure in (7b), in which, due to the intervening trace, the indirect object must be contrastively focused. The opposite accent distinctions hold in (7c) and (7d), indicating that N-A-D is the base word order for the verb unterziehen in German, while N-D-A in (6d) is a derived (contextually marked) permutation (cf. among others Czepluch 1988). However, and this is rarely noted in the literature, scrambling of definites may also preserve contextual unmarkedness (cf. Molnárfi 2002: 1113f. and Fanselow 2003: 198f. for discussion). Compare the following sentences (the relevant focus domains are marked by +F): (8)

a. Ich habe [VP gestern +F[das MÄDCHEN] geküsst] (*und sie I have yesterday the girl kissed and her danach +F[nach Hause begleitet]). then to house escorted ‘I kissed the girl yesterday and then escorted her home.’

(CA)

b. Ich habe das Mädchen +F[VP gestern geküsst] (und sie danach (SA) +F[nach Hause begleitet]). c. Ich habe [VPgestern +F[SIE] geküsst] (*und danach +F[nach I have yesterday her kisssed and then to Hause begleitet]). house escorted

(CA)

d. Ich habe sie +F[VP gestern geküsst] (und danach +F[nach Hause begleitet]). (SA) e. Ich habe +F[VPgestern ein Mädchen geküsst] (und sie danach (CA) +F[nach Hause begleitet]). f. ?Ich habe ein +F[MÄDCHEN] [VPgestern geküsst] (*und sie danach [nach Hause begleitet]). (CA) +F Taking gestern to be a marker of the VP-boundary, (8a–f) indicate that sentential stress cannot be assigned to all elements in the base position. In par-

On Scrambling as Defocusing in German and West Germanic

343

ticular, the selection of a focus exponent for wide focus projection seems to be sensitive with respect to the semantic property of definiteness (Molnárfi 2002: 1118f.). Other than indefinites, referential definites (pronouns or full nouns) are not able to take up neutral stress in the VP-internal base position of two-place predicates (cf. Abraham 1997). Such definites will attract contrastive stress if to the right of sentence adverbs, destroying wide focus projection (cf. the contrast between 8a and 8b). Contextual unmarkedness is restored once such elements have been scrambled out of the focus domain, enabling assignment of sentential stress to the past particple and projection of wide focus. That (8a) and (8b) indeed license different focus domains is indicated by their different conjoinability with a contextual embedding that allows projecting focus. Indefinite objects, on the other hand, behave exactly like predicted by Cinque’s Null Hypothesis, taking up sentential stress in the most deeply complement position (cf. 8e). Scrambling of such indefinites cannot preserve contextual unmarkedness, (8f) triggering a contrastive interpretation, which destroys focus projection. The data indicate a different discourse functional base positon for definites and indefinites in West Germanic (cf. Abraham 1997 and et passim). For at least some elements, therefore, the position determined by lexical mapping within VP does not correspond to the position in which the contextually most unmarked interpretation is achieved. Full definite nouns and personal pronouns, which seemingly exhibit some discourse functional value on their own, strive to the left of sentence adverbs, realizing such unmarkedness in terms of destressing. This indicates the existence of a mechanism that reorders the base word order under discourse functional and/or prosodic weight (Abraham 1997, Abraham & Molnárfi 2002). As (8a–f) shows, such reordering also may serve to preserve, rather than to destroy maximal contextual unmarkedness (cf. also Fanselow 2003). The conclusion that we draw in line with Fanselow (2003) from this is that pragmatic markedness need not tell us anything about the derivational history of a given word order. The some goes for the logical link between between maximal contextual unmarkedness and base word orders which is just as little justified. Base word orders, being a function of lexical mapping, do not always satisfy discourse licensing conditions imposed on the surface order. This follows as the computational domain of wide focus must not contain elements belonging to the presupposition structure, which, among others, includes referential definites. As markedness or unmarkedness of a certain word order is heuristically determined by the projection domain of focus (Höhle 1982), base word orders involving referential definites may be marked as with respect to the focus-background structure of the sentence.

344

László Molnárfi

Derived word orders thus may in fact satisfy the criterion of contextual unmarkedness better than base word orders do. Pragmatic unmarkedness identifies base word orders in the case of unaccented indefinites only, where the discourse functional factor is controlled for (cf. Meinunger 2000: 48f).

5. Scrambling as Movement: The Search for the Trigger As Fanselow (2003) points out, the above observations strongly indicate that word order variation cannot be movement driven by pragmatic features encoded in the lexicon. Per definitionem, normal word orders, allowing wide focus projection, are pragmatically unmarked. Therefore, if “normal” word orders involving wide focus projection are derived transformationally from the base word order, such movement can never involve pragmatic licensing. This excludes that scrambling of the definite object in (7b) and (8d) is triggered by some common discourse functional feature such as [+topic] or [+presupposed], for which the scrambled element is marked in the lexicon (contra Meinunger 2000 and in line with Abraham 2003). Similarly, the base orders in (7a) and (7d) indicate that pragmatic markedness of definite arguments can very well occur in-situ, without involving any functional licensing. That is, against the logic of derivational economy, it is sometimes more costly to block scrambling, than to allow it (cf. Molnárfi 2002 and 2003 for further discussion).5 Such observations considerably weaken the case for scrambling theories that see word order variation as a function of pragmatic licensing (cf. Meinunger 2000, Diesing & Jelinek 1993, Jäger 1995, Adger 1993 or Frey 2000). One alternative to the concept of a pragmatically driven scrambling is an interpretative chain formation mechanism as proposed by Abraham (1995a, 1997, 1999). Such mechanism is based on the idea that the middle field in German allows for a structural split between a thematic (VP-external) and a rhematic (VP-internal) domain. Within these domains, discourse functions are identified by the structurally governed distribution of clausal accent (Cinque 1993) and contrastive stress. Definites and indefinites are identified in different discourse functional base positions inside and outside VP, and any movement out of the basic discourse chain inevitably triggers refocusing effects. In this scenario, the discourse interpretation of scrambled elements does not follow from lexical encoding, but solely from the conditions of stress assignment, which associate accent distinctions with the designated base positions. If the base position is outside VP, as in the case of definites, the scrambled word order will yield pragmatic unmarkedness, while contrastive

On Scrambling as Defocusing in German and West Germanic

345

stress in (8a) and (8c) indicates that the “in-situ”-definites are not in their basic discourse functional domain (see Abraham & Molnárfi 2002 for details). While such refocusing mechanism represents a reliable diagnostics for the discourse interpretation of definite and indefinite arguments in the alternative permutations, it is not clear whether the concept of refocusing itself could (or should) be formalized in terms of triggered movement. Abraham himself (1997) takes movement out of base positions to be adjunction which targets a discourse functional position and is driven by discourse optimization only (Abraham & Conradie 2001: 102f). If so, such movement does not fall under the category of triggered, feature driven displacement and cannot be formalized in the minimalist framework. The question arises of whether, in the light of the markedness-repairing character of scrambling, it is possible at all to regard word order variation as triggered movement. Fanselow (2003: 209f.) argues against such an assumption. Following Rosengren (1993), he points out that scrambling seems to serve to allow the licensing of a grammatical property of a term that itself does not participate in chain formation (cf. also Alboiu 2000 for Romanian). In this spirit, scrambling in (7b) would allow the indirect object to be in focus, while in (8b) and (8d) word order variation would enable the assignment of sentential stress to the past participle. However, according to general assumptions (cf. Chomsky 1993, 1995), movement of _ can be triggered by the morphosyntactic properties of _ only. That is, benefiting of others terms, not involved in the transformation, is excluded for reasons of derivational economy. If so, the altruistic character of word order variation indicates that the alternative orders are not a result of a derivational transformation, but produced by free base-generation. In the following section, I will address this issue, showing that word order variation, while affecting the computation of focus domain, in fact does not have to involve altruistic movement.

5.1. Does scrambling involve altruistic movement? A non-derivational view on scrambling seems to be corroborated by the observation that, if a derivational approach is adopted with respect to the permutation of X and Y, altruistic movement of Y cannot be excluded. Following (Rosengren 1993), Fanselow (2003) claims that in some cases, movement of Y does not seem to be driven by the grammatical properties of Y (i.e. by Greed), but such movement seems to benefit the grammatical properties of X, an element that itself does not participate in the chain formation. However, operations licensing grammatical properties should only

346

László Molnárfi

include terms participating in the operation, indicating that the permutation of X and Y cannot be due to movement. Witness the following sentence taken from Fanselow (2003: 211): (9)

dass man niemanden nach MITTERNACHT anrufen sollte. that one nobody after midnight call should

(CA)

As inherently negative quantifiers are not referential, there seems to be no obvious candidate for a pragmatic feature that could trigger scrambling in (9). Rather, the negatively quantified indefinite seems to move in order to allow the temporal adverb to be in focus (cf. also Alboiu 2000 for a similar analysis of word order variation in Romanian). This is, however, a clear instance of altruistic movement, which is better analyzed in terms of base generation. Notice two things. First, (9) involves scrambling of indefinites, something, which has been analyzed along the lines of semantically driven movement, depending on the availability of quantificational type shift and semantic incorporation for the indefinite (cf. van der Does and de Hoop 1998: 402f.). That is, the availability of the object shift in (9) does not have to tell us anything about the nature (or the lack) of a possible trigger for scrambling of referential definites. Second, shifting of the indefinite in (9) is not vacuous, but is related to the semantic property of genericity, as it might become clear from the following slightly modified sentences: (10) a. dass man einen Fisch im WASSER fangen sollte. that one a fish in water catch should b. dass man im Wasser einen FISCH fangen sollte.

(CA) (CA)

There is a clear semantic shift in the interpretation of the indefinite in the scrambled position (cf. 10a). While (10a) is interpreted as a generic statement about fishes, the in-situ position of the indefinite in (10b) is interpreted existentially (along the lines of Heim 1982 and Diesing 1992). As niemand is prototypically interpreted generically, the negatively quantified indefinite in (9) can be argued to move in order to escape the domain of existential closure. Such movement, however, has nothing to do with the focus properties of the temporal adverb. Similarly, by moving presupposed elements out of VP, scrambling of definites restricts the focus domain and this obviously has prosodic consequences for the elements remaining in-situ. Given that the assignment of

On Scrambling as Defocusing in German and West Germanic

347

sentential stress is determined by depth of embedding, one of the side-effects is that any remaining element may get scoped by narrow focus (cf. 7b or 8a). This is, however, not the (altruistic) trigger of scrambling. Rather, the common property all scrambled elements share is their lack of focus, indicating the existence of a formal destressing mechanism in West Germanic (cf. Molnárfi 2002). Lambrecht (1994: 266f.) argues that destressed elements correspond to entities that are highly referential or active in the discourse. Such a correlation has independently been confirmed by phonological experiments, showing that destressed elements are preferredly interpreted as dicourse-linked (cf. Nooteboom and Kruyt 1987 and Terken and Nooteboom 1988). That is, the discourse interpretation of elements participating in the destressing chain follows in terms of position and accent designation alone, without involving checking of lexically marked pragmatic features such as [+topic] or [+familiar] (cf. also Abraham 2003). Strong support for such discourse dependent (hence never fully optional) scrambling comes from idioms involving nonreferential definites: (11) a. dass Peter gestern den Vogel abgeschossen hat that Peter yesterday the bird down-shot has ‘that Peter took the cake yesterday’

(SA)

b. dass Peter gestern den VOGEL abgeschossen hat ‘that it was the bird that Peter shot down yesterday’

(CA)

c. dass Peter den Vogel gestern abgeschossen hat ‘that Peter shot down the bird yesterday’

(SA)

Contrary to the idiomatic reading in (11a), there is a semantic shift in the interpretation of the definite NP in (11b) and (11c). The idiomatic reading is lost if the direct object is marked by contrastive accent (cf. (11b)) or is scrambled to the left of the adverbial (cf. (11c)).6 In (11c), the destressed definite NP can only be interpreted anaphorically, that is, in a literally sense. It follows that, via destressing, scrambling forces the definite object to be interpreted as discourse linked. This is a strong argument against accounts that regard scrambling to be a fully optional movement without any formal trigger (cf. Van der Does & de Hoop 1998; de Hoop 1997, 2000a). If there is no pragmatic feature involved in scrambling, the observation that derived word orders may be contextually more unmarked than base word orders does not pose a problem for a derivational analysis. Word order is driven by prosody, i.e. the destressing of referential definites that would

348

László Molnárfi

disturb the proper computation of wide focus. Focus structure can be projected only after such destressing has taken place. The question arises of whether it is possible to formalize the negative property “lack of focus” in terms of a trigger-account, or the prosodic distinctions of scrambling should better be captured as interface effects. The following sections serve to clarify this issue. First, I will investigate whether the prosodic and discourse-functional correlates of scrambling may also be derived without reference to a chain formation mechanism. I will argue that while base-generation theories are able to capture such markedness effects to some extent, they only can do so within a limited empirical scope of word order variation.

6. Can Non-derivational Approaches Derive Markedness Distinctions? The alternative to a grammatically triggered scrambling operation is the assumption that there is no underlying (basic) word order that could be identified as the input for word order variation. Rather, the possible permutations are all base-generated in syntax (Neeleman & Reinhart 1998, Fanselow 2003). Base generation accounts are based on the idea that, contrary to movement, which is the costly re-merging of an element selected from the numeration, merging is a free option of grammar. The main advantages of such approach is that the issue of optionality of triggered movement (problematic in a derivational framework) becomes obsolete, and that no additional assumptions are needed with respect to the functional architecture of the sentence structure. As functional projections play no role in the overt licensing of the alternative orders, the minimalist feature checking apparatus can be left intact. Given that free constituent order is produced by alternative merge operations, word order variation will be attestable in all languages where movement of arguments to specifier positions of functional heads takes place covertly. Such covert movement preserves the variation of existing base orders, contrary to languages such as English, where overt movement to functional positions fixes the alternative orders in a hierarchical way (Fanselow 2003: 196).7 The question is, however, whether the concept of free merging can derive the markedness contrasts observed between the possible word orders. The alternative argument serializations have been shown to differ with respect to possible contextual embeddings as well as prosodic and focus correlates (Abraham 1997 and Molnárfi 2002).

On Scrambling as Defocusing in German and West Germanic

349

Note that a base generation account in the spirit of Neeleman & Reinhart (1998) or Fanselow (2003) takes the prosodic and discourse-functional correlates of word order variation to be interface effects due to the alternative merging orders. Such concept has to rely on the existence of at least three elements to be merged in alternative orders: the lexical head H which is projecting and two merging-candidates which can be used to form alternative syntactic objects with H. In the case of Dutch scrambling, the set of merging candidates consists of the direct object and the sentence adverb, while the projecting head is the verb. The resulting prosodic patterns can be read off of the interface level where stress assignment is computed. Witness the two possible merging orders: (12) a. dass der Arzt [V’ den Patienten [V’gestern untersucht hat]] (SA) that the doctor the patient yesterday examined has b. dass der Arzt [V’ gestern [V’den PATIENTEN untersucht hat]] (CA) If the merging order is ADV to be merged with V first, then DO will be forced out of the focus domain (not being the most embedded element in the spirit of Cinque 1993). This is the case in (12a). If the merging order is DO to be merged with V first, DO, being the most deeply embedded element, will attract pitch prominence (cf. similarly Fanselow 2003: 208). This is the case in (12b). The interface account indicates that base generation theories can derive markedness distinctions only to the extent to which there is a set of alternative candidates to be merged with the lexical head. That makes adverbs a structural perequisite of Dutch scrambling. However, while sentence adverbs as VP-markers are an important shibboleth for word order variation, nothing hinges in structural terms on their presence in scrambling. The observable prosodic and discourse effects can easily occur in contexts where alternative merging candidates are not available. As Cinque (1993: 254f.) observes, the main stress of the sentence in two place predicates is assigned differently according to the (in)definiteness of the direct object (sentential accent (SA) is marked by italics): (13) a. Der Arzt wird einen Patienten untersuchen. b. Der Arzt wird den Patienten untersuchen.

(SA) (SA)

While the indefinite direct object attracts sentential accent in (13a), the main stress of the sentence falls on the verb if the direct object is definite

350

László Molnárfi

(13b). However, in (13b), the destressing of the direct object takes place independently of the presence of alternative merging candidates, which could supersede the definite in the focus domain. That is, destressing of the definite in object in (13b) must be derived along entirely different lines than destressing of the same definite in the context of (12a), no doubt a methodologically undesirable consequence.8 Another empirical domain where the base-generation approach seems to fail concerns the scrambling of inherently case marked arguments such as “low” dative or genitive objects in German. Recall that base generation of all possible permutations in a given numeration is free, markedness effects arising only after prosodic patterns associated with the alternative merged orders are read off at interface levels. This yields that scrambling of all three place predicates should yield the same markedness effects, irrespectively of the question of to which semantic verb class a given verb belongs. However, the empirical observation is that, under preservation of the neutral stress pattern, lexically case marked arguments cannot be shifted across (or base generated higher than) structurally case marked arguments in German (cf. Lenerz 1977; Czepluch 1988: 282f.; or Molnárfi 1998: 552): (14) a. dass ich der Firma Hans Schulzs Plan geschickt habe (SA) that I the companyDAT Hans Schulz’s planACC sent have b. dass ich den Plan Hans Schulzs Firma geschickt habe (SA or CA) that I the planACC Hans Schulz’s companyDAT sent have c. dass ich Hans Schulz des Mordes bezichtigt habe that I the manACC the murderGEN accused have

(SA)

d. *dass ich des Mordes Hans Schulz bezichtigt habe (SA) that I the the murderGEN Hans SchulzACC accused have e. ??dass ich des Mordes HANS SCHULZ bezichtigt habe

(CA)

The linearization of the arguments in (14a) reflects the base word order ND-A for the verb send in German. (14b) shows that such base word order may be overridden under a neutral accent pattern if scrambling of the direct object leaves behind a highly rhematic indirect object to be computed within the focus domain of VP (cf. Lenerz 1977). Recasting this observation in terms of a base generation model, the data indicate that a verb like send in German allows for both the N-D-A and the N-A-D-patterns as free permutation orders of its arguments. However, (14d/e) show that the same discourse functional distribution (thematic genitive object, highly rhematic direct object) does

On Scrambling as Defocusing in German and West Germanic

351

not legitimate the rearrangement of the base word order N-A-G into N-G-A. In fact, shifting the genitive object is hardly ever possible in (14e), even if the rhematic direct object is supported by heavy contrastive stress in the NG-A-pattern (cf. also Fanselow 2003: 219). Similar prosodic correlates seem to hold for lexically assigned “low” datives of the N-A-D-pattern and PPs if scrambled across direct objects in German. Even if supported by contrastive stress, the respective base word orders are reluctant to change under the influence of a thema-rhema-distribution that normally would favor scrambling: (15) a. dass er Hans Schulz der Operation unterzog. that he Hans SchulzACC the operationDAT subjected ‘that he subjected Hans Schulz to the operation’ b. ?dass er der Operation HANS SCHULZ unterzog

(SA)

(*SA/?CA)

c. dass er Hans Schulzs Buch auf den Tisch gelegt hat that he Hans Schulz’s bookACC on the table put has

(SA)

d. ?dass er auf den Tisch HANS SCHULZS BUCH gelegt hat (*SA/?CA) Thus, given the same discourse conditions, shifting verbal genitives, “low” datives or PPs across direct objects is much more difficult (cf. 14d/e, 15b/d) than the scrambling of two structurally case marked arguments (cf. 14b).9 This contrast is not predicted by the base generation of the altenative word order patterns, which should be able to force the scrambled element out of the focus domain in all the cases discussed. Among others, the difference between the permutability of base orders involving structurally vs. inherently case marked arguments in German is discussed in Czepluch (1988: 282), who notes that some base word orders seem to be more “basic” or “rigid” than others.10 Such contrasts can be taken to indicate that the derivation of natural word orders obeys a case markedness hierarchy, according to which structurally case marked arguments preferredly c-command lexically case marked arguments at every level of derivation (cf. Molnárfi 1998: 553f.).11 Then, the contrast between (14b) and (14d) can be explained by a hierarchy preserving principle in the spirit of the Parallel Movement Constraint of Müller (1999) (cf. also Meinunger 2000: 153f.). Given the hierarchy of lexical markedness is reversed in (14d, 15b and e), but not in (14b), scrambling that preserves the neutral stress pattern is only allowed in (14b).12 This also indicates that locality conditions of movement (Relativized Minimality, Rizzi 1990) (if they hold for scrambling

352

László Molnárfi

at all) should not be formulated with reference to syntactic categories, but with reference to the case morphological distinctions of such categories in German (cf. section 9 for discussion and also Fanselow 2003: 195 for a proposal in this spirit). To sum up: As long as (12a–b) is analyzed in terms of base-generation (cf. Neeleman & Reinhart 1998), a broader view on word order variation cannot be upheld. Given that the observed prosodic and discourse distinctions associated with the possible permutations in (12a–b) also arise in the absence of adverbial elements, a unified treatment of (12a) and (13b) seems only possible in terms of a derivational approach. This considerably weakens the explanatory power of base generation theories, which can cover a limited empirical range of word order variation only. In the following section, I will scrutinize the relation between movement and focus assignment, arguing that the postulation of designated focus positions in the middle field does not constitute the most economic derivation of stress distinctions in German. Section 8 will show that the common property shared by all scrambled elements is destressing, to be formalized in terms of a novel clausal relation: antifocus licensing (cf. Molnárfi 2002).

7. How many Focus Positions are there in German? While there has been much effort invested in finding a structurally designated focus position in German (cf. Haftka 1994 or Krifka 1998, among others), the insights have hardly gone beyond the idea that several focus projections are needed to account for the empirical data. Recall first that (re)focusing effects can be captured by movement out of a base position in the spirit of the scrambling theory of Abraham (1997). In this scenario, only those elements will be refocused that leave their basic discourse functional chain as determined by categorially or context-bound discourse status (that is pronouns vs. definite and indefinite nouns). Assuming a VP-external base position for themes (presupposed information) and a VP-internal base position for rhemes (new, or contrastively used presupposed information) and a theory of structurally governed stress assignment (Cinque 1993), this mechanism suffices to explain the refocusing effects of scrambling found to hold for indefinites and definites in the middle field. Indefinites will be refocused if moved outside VP, while definites will be refocused if inside VP. By following the idea that scrambling movement necessary leads to refocusing, then, one has to assume at least three or four distinct focus posi-

On Scrambling as Defocusing in German and West Germanic

353

tions, in which definites and indefinites may take up contrastive stress in German (cf. Abraham & Conradie 2001: Chapter 6): Possible positions for contrastive focus in German (i)

SpecCP for topicalized elements taking up contrastive stress

(16) [CP +F[Das MÄDCHENi ] habe ich [VP gestern ti geküsst]]. (ii)

One (or two) adjunction positions (SpecVP?) for focus-scrambled definites and indefinites to the left of sentence adverbs (i.e. outside VP, according to Abraham 1997)13

(17) a. ?dass ich +F [das MÄDCHEN] [VP gestern geküsst habe] b. ??dass ich +F[MÄDCHEN] [VP gestern geküsst habe] (iii) One (or two) focus positions for contrastively stressed definites and indefinites to the right of sentence adverbs (i.e.inside VP, according to Abraham 1997) (18) a. dass ich gestern [VP +F [das MÄDCHEN] geküsst habe] b. dass ich gestern [VP +F[MÄDCHEN] geküsst habe] Note that refocusing of definites and indefinites in (18a and b) has to be seen as licensed along different discourse chains, involving distinct focus positions. While definites, being base generated outside of VP, have to be lowered into VP to take up contrastive stress, indefinites have to be assumed to move minimally within VP in order to reach a discourse functional position to the right of sentence adverbs (cf. Abraham & Conradie 2001: 104). A similar distinction should be made with respect to focus scrambling of definites and indefinites in (17a and b). The main drawback of this refocusing concept is the assumed rightward movement of VP-externally base generated thematic material for purposes of refocusing. Apart from general objections against lowering (cf. Kayne 1994 and others), it is not entirely clear what kind of focus-licensing position such thematic elements should target if lowered into VP. In Abraham & Conradie (2001: 103) such position is identified as a discourse functional with the feature marker [-base], that is an A’-position, adjoined to the left to the argument position of the direct object. If adjunction, movement out of base position does not fall under the category of morphologically triggered movement. Rather, such movement is seen to be driven by context-optimizing only. This solves the problem of

354

László Molnárfi

optionality, but places refocusing outside of the range of feature-driven, that is formally assessable, element displacements. Also, the question arises whether the inflation of the sentence structure with multiple focus projections is the most economic solution for deriving the prosodic and focal effects of word order variation in German. The only theoretical insight to be gained from the data seems to be that contrastive focusing is possible almost everywhere within the middle field in German, in sharp contrast to a discourse configurational language like Hungarian, where there is only one designated position for focus-licensing at the left periphery (cf. Kiss 1998). The observation that focused elements can hardly ever be moved over sentence adverbials also seems to speak against focus licensing in terms of movement in German (cf. Lenerz 2001: 253, Haider & Rosengren 1998: 91): (19) a.

Was hast du gestern dem Touristen gezeigt? what have you yesterday the touristDAT shown

b. ?? Ich habe [Fden WEGi] gestern dem Touristen ti gezeigt. I have the wayACC yesterday the touristDAT shown c.

I habe dem Touristen gestern [F den WEG] gezeigt.

(OK)

Obviously, in all proposals that see focus licensing as an instance of movement to some designated position in West Germanic, the do-not-move-focus restriction of Lenerz (1977 or, recently, 2001: 253) still needs an adequate explanation.14 In Molnárfi (2002), a more economic way has been proposed to derive the possible accent and focus distinctions associated with word order variation in German and West Germanic. In contradistinction to previous theories, the problem of discourse-syntax mapping is approached here from a different angle: from the perspective of defocusing. It is held that defocusing can be formalized in terms of a novel clausal relation called antifocus. Antifocus licensing is based on the idea that, unlike Hungarian, West Germanic does not relate assignment of contrastive focus to operator licensing in designated functional projections. Rather, the only structurally anchored discourse relation is destressing, triggering overt movement of referential definites to a designated position in the middle field. Only if all elements marked for antifocus have been checked in the corresponding functional projections can the domain of wide focus be computed. Narrow focus on discourse referential entities falls out as a default mechanism in this scenario: outside their

On Scrambling as Defocusing in German and West Germanic

355

destressing chain, definites will attract contrastive stress and receive a marked discourse interpretation.

8. Scrambling as Defocusing There are two ways to derive contrastive focus in (West) German(ic). One can assume with Abraham & Conradie (2001) the existence of at least four distinct focus positions spread over the middle field and the CP-domain (see section 7), or, looking at the problem from a different angle, one can dispense with the idea that the relation focus is structurally anchored in West Germanic. I would like to argue that the only grammatical relation structurally relevant for middle field scrambling is defocusing. Taking this as a point of departure, a novel functional relation can be postulated, which is the theoretical antithesis of focus licensing: antifocus. Compare the following sentences from Hungarian (a focus-prominent language) and German (an antifocus-prominent language): (20) Hungarian tegnap [VP ti tj ]]]. [FP A LÁNYTi [F csókoltamj the girlAcc kissed1PersSing yesterday ‘It was the girl I kissed yesterday’ (21) German dass ich [AF das Mädcheni [VP gestern ti geküsst habe]]

(CA)

(SA)

While focus licensing in Hungarian is overt movement of a phrase to be focused to the Spec-position of a designated functional projection (cf. Horváth 1995; Kiss 1995 or Molnár 1998),15 antifocus licensing in West Germanic can be taken to be overt movement of a phrase to be destressed to the Spec-position of a designated functional projection. However, in contrast to focus movement, antifocus does not license operator chains.16 Rather, antifocus is a formal destressing mechanism, which enables discourse referential material to escape focus in terms of movement to an argument position.17 Thus, in (21) the object moves out of VP in order to eliminate its antifocus feature in the Spec-position of the functional projection which the AFfeature is head of. The functional domain of the AF-projection(s) corresponds to Abraham’s (1997) thematic domain, while the rhematic domain is VP where all lexical material is base-generated. The functional domain will attract elements associated with the AF-feature to leave their base positions

356

László Molnárfi

for purposes of feature checking. What is generally called scrambling in (21) is thus overt movement of the object to the Spec-position of AFP. The resulting grammatical effect is that of anaphoric destressing, enabling the discourse linking of the object on the level of background-focus structure (cf. similarly Zubizaretta 1998). Given that in German scrambling of more than one argument is possible, the existence of more than one AF-landing positions has to be assumed in the middle field. I take AF-projections to be recursive in the sense of Rizzi (1997), but the expansion of the structural domain takes place to the right, rather than to the left of CP in this scenario (cf. also Abraham & Molnárfi 2001:8). If such destressing positions are not linked to the defocusing of any specific grammatical function (forming a set, rather than an array) all possible permutations can be derived in more complex predicates. Witness the following sentences: (22) a. dass [TP ich [AFdas Buchj [AFdem Studenteni [VPGESTERN ti tj gegeben habe]]]] (CA) b. dass [TP ich [AF dem Studenteni [AF das Buchj [VP GESTERN ti tj gegeben habe]]]] (CA) While only (22b) matches the base word order N-D-A to the left of the sentence adverb, there is no clear pragmatic contrast between (22a) and (22b). In both cases, the sentence adverb will attract narrow focus and the scrambled objects will be destressed. The data indicate that the order of scrambled aguments is more or less interchangeable if within the domain of destressing, in contrast to reordering out of the base position, which triggers clear markedness effects (cf. 7b and 7d). In order to account for the destressed position of the subject in (22a) and (22b), we could argue that functional heads like T may be associated with the AF-feature as well. Scrambling of full nominal elements across such destressed subjects will then yield a more marked order as such movement involves bypassing at least one hierarchically determined functional projection, i.e. TP: (23) ? dass [TP dem Studenteni [TP das Buchj [TP ich [AF tj [AF ti [GESTERN ti tj gegeben habe]]]] Such long object shift to the left periphery of the sentence has been argued to target an adjunction position in the domain of T-scrambling (cf. Haider & Rosengren 1998).That is, (23) involves two different chain formation operations: A-movement for the formal destressing of the definites to the corre-

On Scrambling as Defocusing in German and West Germanic

357

sponding antifocus positions and operator licensing at the left periphery, between CP and TP. Being syntactic heads, clitic pronouns may occur unmarkedly to the left of destressed subjects. Such pronouns leave the antifocus domain to adjoin to a syntacic host (such as C) that can support their phonologically reduced form (cf. Molnárfi 2004 for details of such analysis).18

8.1. Remnant topicalization and the islandhood of scrambled constituents By assuming that scrambling of definites targets the spec-position of (a) higher functional projection(s), two well-known problems have to be addressed. The first problem concerns fronting of larger constituents containing scrambled arguments, the second problem pertains to the status of scrambled constituents with respect to extraction.19 I will start with a discussion of remnant topicalization, and then proceed to review to relevant extraction-data. I will argue that neither of the empirical domains under discussion allows one to draw decisive arguments against an account of scrambling in terms of functional spec-head-licensing. Investigating data of VP-topicalization in German, Haider and Rosengren (1998: 35f.) argue that there is an important restriction to be observed with respect to the moved constituent. The topicalized maximal projection must not contain a trace of the finite verb. Consider the following sentences: dem Studenten tj zurückgegeben] hat Peter (24) a. [Das Buchj has Peter the bookACC the studentDAT back-given ja schon gestern. part already yesterday b. *[Das Buchj dem Studenten tj zurück-ei] gabi Peter ja schon gestern. Note first that (24a) exhibits word order variaton within the fronted constituent, which, according to our analysis, indicates topicalization of a functionally enriched structure larger than bare VP. However, Haider and Rosengren (1998: 36) use the contrast between (24a) and (24b) to argue that the fronted constituent in (24a) exhibiting scrambling cannot be an instance of a topicalized functional projection. Such projection would contain a trace of the finite verb, which, on its path to C, should have to move through the functional cascade provided by the respective scrambling positions. However, as shown in (24b), such chain-forming is illicit as the trace in the fronted con-

358

László Molnárfi

stituent (in (24b) indentified by the position of the stranded verb particle) is not c-commanded by the finite verb in C. In (24b), this leads to a crossover violation. Therefore, conclude Haider & Rosengren (1998: 37), despite scrambling of the direct object across the indirect object, (24a) cannot involve topicalization of a functional projection, which would open up a potential in-between landing site for the finite verb. Rather, in light of the well-formedness of (24a), scrambling must take place within an extended VP-domain, which is in (24a) the complement of the tense marking auxiliary haben. Given this auxiliary is part of a higher VP-layer, chain formation between haben and C does not effect the lower VP, which, containing the scrambled arguments, may be fronted to SpecCP. Note, however, that the presence of unbound traces within fronted constituents cannot be the decisive factor in excluding illicit cases of remnant topicalization. First, topicalized constituents may contain unbound XP-traces (cf. Müller 1998: 215 and 174): ein Buchi tj (25) a. [VPDem Peter ti gegeben]j hat die Claudia the PeterDAT given has the ClaudiaNOM a bookACC b.? [VP [PP ti Mit] gerechnet]j hat dai keiner tj. with counted has there no-one (25a) and (25b) show that nothing excludes a configuration where a trace of an NP or a PP is included in a fronted constituent and the corresponding antecedents are outside of this topicalized constituent (cf. the Anti-freezing effect of Müller 1998: 157). Second, it is not the case that in remnant movement contexts C cannot be lexicalized by main verbs in German. That is, contrary to Haider & Rosengren (1998), a fronted constituent sometimes may contain an unbound V-trace too. Compare the following sentence (borrowed from Müller 1998: 260): Bonbons ti]j gibti man besser nicht tj ] (26) [CP [VPKindern childrenDAT sweetsACC gives oneNOM better not As the main finite verb undoubtedly moves to C in V2-clauses in German, the fronted constituent must contain an unbound verb trace in (26). This, however, should be illicit under Haider & Rosengren’s (1998) analysis.20 Instead of relating the licensing conditions of remnant movement to the presence of unbound traces, Müller (1998: 241) offers a new account which is based on constraining the possible landing sites of remnant movement. I will adopt here the derivational version of this constraint (Müller 1998: 243):

On Scrambling as Defocusing in German and West Germanic

359

(27) Unambigous Domination In a structure …[A … B … ]…, A and B may not undergo the same kind of movement. (27) states that an element which has been moved out of a constituent prior topicalization and the fronted constituent must not target the same type of position. As scrambling and topicalization do not target the same type of position (the former is movement to the Spec of an argument position, the latter is movement to the Spec of an operator position) XP-scrambling prior fronting in (25a) and (25b) is grammatical. The same goes for (26), where the finite verb targets the head position and the fronted constituent the Specposition of C. In fact, nothing excludes the presence of unbound V-traces in topicalized constituents as XP-fronting and head movement always occupy different types of positions.21 However, targeting the same type of position for A and B creates an illicit chain as shown in the following sentences (Müller 1998: 239): (28) a. * dass [VP dem Peter ti gegeben]j die Claudia ein Buchi tj hat b. * [_ ti Zu lesen]j glaube ich [CP [dieses Buch ]i hat keiner tj versucht] to read believe I this book has no-one tried Contrary to VP-topicalization in (25a), VP-scrambling of a constituent involving an unbound NP-trace is ungrammatical as the shifted direct object and the fronted constituent in (28a) target the same type of scrambling position. Similarly, (28b) violates Unambigous Domination as _ in SpecCP dominates a NP-trace which is in the same type of (intermediate SpecCP) position. While the locality restrictions of (27) may eventually be derived from a formal theory of feature-attraction,22 a constraint like Unambigous Domination highlights that an adequate theory of remnant topicalization does not speak against an analysis where scrambling targets the spec-position of higher functional heads. Rather, as far as I can see, the relevant data in (24a), (25a) and (28a) may only be taken to support the assumption that scrambling and topicalization involve the creation of different types of chains. The unbound verb trace in (24a) is licit as, on its path to C, the finite verb in (24a) moves through a cascade of AF-head-positions, while the fronted constituent targets the specifier of an operator position. A related problem concerns the question of whether scrambling turns its landing site into an island for extraction. Haider & Rosengren (1998: 33f) put forward the following sentence to sustain their claim that scrambling

360

László Molnárfi

does not induce opacity-effects, hence cannot involve functional spec-headlicensing: (29) Wasi hat damals [ ti für Ergebnisse]j denn jeder von Euch tj erzielt? PART each of you achieved what has than for results If scrambling were to target functional spec-positions, which are known to create opaque domains for extraction, the availability of the wh-split out of a scrambled constituent in (29) would remain unexplained. (29) speaks thus against a movement account of scrambling in terms of functional projections. Note, however, that the data with respect to the ‘was für-split’-construction are far from uncontroversial (cf. Diesing 1992: 32f): (30) *Wasi hat [ti für Witze ]jdenn jeder von Euch tj erzählt? What has for jokes PART each of you told Note that the ungrammaticality of (30) cannot be put down to the assumption that it is already the scrambling of the wh-phrase in front of the particle denn, which is illicit (Haider & Rosengren 1998: 34f.). Such scrambling seems very well possible in (29). It seems to me that what we need here is deeper understanding of how was-für-split exactly works in German. Until such theory is presented, howewer, it is methodologically dubious to draw decisive conclusions based on was-für data. Turning to the less controversial judgements, the data seem to support the assumption that scrambling indeed creates islands for extraction. Compare the following sentences borrowed from Müller (1998: 143), indicating that scrambling of an object NP to the left of an adverbial induces a Freezing effect: (31) a. [PP Über wen]i hat der Fritz letztes Jahr [ein Buch ti ]j geschrieben? about whom has the FritzNOM last year a bookACC written b. *[PP Über wen]i hat der Fritz [ein Buch ti ]j letztes Jahr tj geschrieben? The same observation has been made with respect to scrambled PPs, which block wh-extraction of a R-pronoun in German (cf. van Riemsdijk 1978): (32) a. Woi meinst du, dass keiner [PP ti mit]j gerechnet hat? what think you that no-one with reckoned has b. *Woi meinst du, dass [PP ti mit]j keiner tj gerechnet hat?

On Scrambling as Defocusing in German and West Germanic

361

To conclude: Haider & Rosengren’s (1998) arguments against scrambling in terms of spec-head-licensing are based on an inadeqaute analysis of constituent-fronting, which takes the presence of unbound traces as a decisive criterion for excluding illicit cases of remnant movement. An empirically more adequate theory of remnnant topicalization, which is based on the concept of unambiguous domination (Müller 1998), does not speak against an analysis where scrambling targets functional spec-positions. In absence of a coherent theory, was-für-split-data do not allow decisive conclusions about the islandhood of scrambled constituents. However, other cases of wh-extraction support the assumption that scrambling indeed creates opaque domains for extraction.

8.2. Which elements carry [+antifocus]? Given a necessary distinction between stress marking of arguments and other, non-referential entities (cf. Selkirk 1984 and 8.3.), focus assignment falls out as a default mechanism in the scenario sketched above. Discourse referential entities that are not in the domain of AF-licensing will inevitably attract focus and prosodic prominence. A destressing license in the spirit proposed here can only be tied to discourse referential entities, which are important for the interpretation of focus structure. Such elements are lexical categories, marked for the distinction ± definiteness, that is, basically NP-arguments, but not verbs, adjuncts or adjectives (Lambrecht 1994: 264f.), which receive their stress marking by default mechanisms, independently of focus structure. In Molnárfi (2002: 1133), the following generalization has been proposed: (33) Destressing Generalization The destressing license is tied to definiteness by default. Assuming the presence of an underlying destressing-trigger for definites and the absence of such a trigger for indefinites, explains (a) (b)

why indefinites are reluctant to scramble (definiteness effects), the morphological patterns of scrambling found in Afrikaans (cf. 4a/b).

More generally, (33) indicates that scrambling of definites and generic indefinites has to be governed by fundamentally different generalizations. Scrambling of definites is governed by stress assignment conditions, which

362

László Molnárfi

are directly relevant for D-linking, while scrambling of indefinites is driven semantically, depending on the availability of quantificational type shift or semantic incorporation for the indefinite (van der Does & de Hoop 1998: 402f.). Crucially, the destressing license imposes different stressability conditions on definites and indefinites. Definites, which carry an antifocus feature by default, prefer to leave their lexically determined base position in order to reach their thematic discourse chain outside VP. Indefinites, on the other hand, lack the destressing trigger and are reluctant to leave VP, which is their basic discourse domain. In this domain, they are identified structurally by the clausal accent of Cinque (1993) as default rhemes in the most deeply embedded position.

8.3. The discourse linked status of definites In the theory proposed here, definiteness represents a unique selection from a possible set of presupposed candidates (cf. also Lenerz 2001), which necessarily receives a contrastive interpretation if in the computational domain of focus. However, it has been often argued that there is no one-to-one correlation between definiteness and identifiability or active states of discourse referents (Steube 2000: 220 or Lambrecht 1994: 108). Such correlation has been said to hold only for pronouns, which are category-inherently thematic, but not for full definite nouns, which can be anaphoric or non-anaphoric, depending on the context. Nevertheless, and this has been rarely noted in the literature, the proposed strong link between definiteness and a necessary destressing license can be maintained, once a small group of definites has been controlled for. Such definites are either non-referential or their reference set is maximally narrowed down by the specific common ground between the speaker and hearer or by our general knowledge about the world. The following types may be distinguished: (i)

non-referential definites in idioms

(cf. Molnárfi 2002: 1140f.),

(ii)

proper salient individuals as “the president”, “the pope”, etc. (cf. Lenerz 2001),

(ii)

novel definites (which are either deictic or can be pragmatically accomodated by the common ground between speaker and hearer). (cf. Meinunger 2000: 76f.).

On Scrambling as Defocusing in German and West Germanic

363

Recall that, for referential definites, definiteness represents the speaker’s choice from a possible set of presupposed candidates, for which the predication holds. Such choice must be discourse-functionally anchored, i.e. the relevant definite element must have been introduced in the previous discourse. Obviously, such restrictions do not hold for non-refential definites as parts of idioms (cf. Molnárfi 2002: 1122f.): (34) a. dass er gestern den Vogel abgeschossen hat. that he yesterday the bird down-shot has ‘that he took the cake yesterday’ b. dass er gestern den VOGEL abgeschossen hat ‘that he shot down the bird’

(SA)

(CA)

The assignability of SA to the definite object in (34a) indicates that, lacking the antifocus trigger, non-referential definites in idioms behave exactly like indefinite objects, which are associated with unmarked prosodic patterns if surfacing in the base position.23 Note, however, that the idiomatic reading is lost if the definite object (as a normal referential definite) participates in a scrambling chain or is associated with contrastive stress as in (34b). The relevant accent distinctions holding in (34a–b) can be thus directly derived from the different semantic status of the definite, which is discourse-linked in (34b) and which is non-referential in (34a) (cf. also 5.1. for discussion). Similarly, the restrictions of antifocus licensing do not hold for definites, of which the candidate set is maximally narrowed down by the common background of the speaker and the hearer or our common knowledge of the world. Such definites involve proper saliant individuals or objects (the President, the Bible) or definites pragmatically accommodated from the discourse (cf. among others Lambrecht 1994: 65f. and Lenerz 2001: 262). Compare the following sentences, where the respective definite objects are introduced “out-of-the-blue”: (35) a. Ich habe gestern den Präsidenten gesehen. I have yesterday the president seen b. Ich habe gestern die Katze gehört. I have yesterday the cat heard

(SA) (SA)

Neither in (35a) nor in (35b) does the definite object has to attract contrastive stress in the base position. However, in both cases, the entity the definite object refers to has to be common ground between speaker and hearer, based

364

László Molnárfi

on extra-linguistic knowledge. As definiteness in (35a) and (35b) is not determined by the mechanisms of anaphoric discourse linking, the occurrence of the definite NPs within the focus domain is prosodically and contextually unmarked.24 That is, if we disregard the definites in (i– iii), the use of the definite article always represents an anaphorically anchored choice from the speaker’s part, which necessarily becomes contrastive in the structural domain of stress assignment and focus.25 This also means that pronouns and definite full nouns can be argued to be subjected to the same formal destressing mechanism (contrary to Haider & Rosengren 1998 and in the line with Abraham 1997).26

8.4. Antifocus as a violable trigger If the definite object in (21) is indeed marked for anaphoric destressing by some abstract feature, it is not entirely clear why scrambling movement is not always forced: (36) dass ich gestern [VP das MÄDCHEN geküsst habe]

(CA)

In (36) the assumed licensing requirement of the AF-feature is disrespected, as the direct object stays to the right of the sentence adverb, that is in situ, according to our assumptions. Given Last Resort (Chomsky 1993, 1995), we expect the derivation to crash at PF as the AF-feature has remained unchecked after Spell-out. However, although associated with a marked prosodic pattern, (36) is perfectly acceptable and might even be the preferred choice in some contexts. The necessary step for explaining data as (36) without recourse to an (arguably circular) theory of freely assignable features has been taken in Molnárfi (2002), Abraham & Molnárfi (2002) and Molnárfi (2003). It is argued that the interpretable/uninterpretable distinction is too narrow to accommodate syntactically relevant peripheral features in a minimalist theory of movement. The basic idea is that such features represent a third type of trigger, carrying a twofold instruction, which can be interpreted both in syntax (triggering movement) and after spell-out (inducing prosodic and/or morphological effects). In this spirit, the licensing of the AF-feature allows for two convergent derivations. The relevant feature is either eliminated by overt movement in (21), or such feature may survive to PF in (36), where it triggers a deviant, but still convergent interpretation of the definite object.

On Scrambling as Defocusing in German and West Germanic

365

Optionality thus arises as the checking of antifocus may be violated under certain circumstances. Other than morphosyntactic features of case and agreement, the presence or absence of the AF-feature is governed by the externally-imposed legibility conditions of prosodic and information structure. While, at first sight, such a claim seems to follow recent accounts of optionality pursued within the framework of Optimality Theory (Choi 1999; de Hoop 1997, 2000a/b), the scrambling concept proposed in Molnárfi (2002, 2003) should be understood in a strictly derivational sense. According to this view, violability of syntactic rules is not a basic methodological principle, which reflects a process of evaluation and optimalization with respect to some (arbitrarily determined) internal hierarchy of constraints (see McMahon 2000 for criticism). Rather, violability of certain triggers touches upon a fundamental difference pertaining to the licensing conditions of system internal and peripheral features in the computation. Other than core features of case and EPP, the (non)checking of peripheral features can only affect the contextual markedness, but not the basic computability of the derivation. The question arises of what it means for feature checking to be violable in conceptual terms. In a minimalist framework as Chomsky (1998), convergent derivations may only contain features that are interpretable in terms of instructions for the PF and LF interfaces. Uninterpretable features (i.e. phi- and case features), on the other hand, must be eliminated early in the derivation, triggering a rule of self-destruction, and, by pied-piping, a possible displacement of phonological material. Regarded in this light, both types of features encode instructions to different components of grammar. “Strong” features are interpreted (checked) in the syntax, in the limited search domain of a Spec-head relation, while “weak” features are interpreted after Spell-out by the sensorimotoric or semanto-logical system. I assume that prosodic features such as AF may encode instructions for both the core component of grammar and the interface levels. This is straightforward where a feature is interpreted in the syntax, triggering a checking relation in a Spec-head configuration. However, the question arises how the instruction such feature encodes can ever survive to PF, given dislocation entails elimination of the feature. Clearly, in the spirit of strict cycliticity, the decision to re-merge or leave an element in situ must be taken on the spot, leaving no space for optionality of interpretation. We can solve this dilemma by assuming that definite objects are always shifted to the SpecAF-position, but this movement may or may not induce phonological effects (cf. Molnárfi 2003). Within the minimalist framework, the violation of AF-licensing can be derived if one assumes that non-trivial derivational chains may be spelled out at the base position as well (cf.,

366

László Molnárfi

among others, Bo‰koviç 2002). Assuming copy theory (cf. Groat & O’Neil 1996), movement reduplicates the triggering feature of the base in the target position. If the head position of the chain is spelled out, feature checking in the target positon and subsequent PF-deletion of the base will ensure that both feature-copies are removed. In (37), the copy that is spelled out is underlined, eliminated copies and features are crossed out: (37) [SpecAFP _ [AF

+AF # +AF ). It seems reasonable then to assume that Merge allows a further combination, with the resulting label shown in (i): (i) K = {< _, ` >, {_, `}} This is also a formally acceptable option. Crucially, it does not conflict with the essential property of Merge, namely, not introducing extra information, specifically extra features of a consituent different from _ and `. The format of this Merge output might seem to generate ambiguity, since from a purely formal point of view the mirror option, where the ordered pair constituting the label is inverse (i.e. ) ia also possible. This problem can be solved by assuming

Linear Compression as a Trigger for Movement

419

that the given label is intended to be the short form of the more articulated one given in (ii): (ii)

{{_, {_, `}}, {`, {_, `}}

In such case, the output is totally neutral with respect to _ and ` and, crucially, the essential requirement that Merge not introduce new information is preserved. (I am indebt to James Higginbotham for an extensive discussion on this topic). 18. X c-commands Y iff a. X and Y are categories (not segments of categories); b. no segment of X dominates Y; c. every category that dominates X dominates Y. 19. The representation in (4iv) can be misleading: since hierarchy is what counts to linearize terminal nodes, (4iv) repeated here as (i)a is totally equivalent to any of the following representations: (i) a. b. c. d.

20.

21.

22.

23.

[ ZP [X° YP]] [[X° YP] ZP] [ZP [YP X°]] [[ YP X°] ZP]]

The linearization of any of these notational variant of the same hierarchical relations would in fact be the same under the LCA: the terminal nodes contained in ZP precede the terminal in X° and the terminal in X° precedes those in YP. The choice among (i)a–c is irrelevant and conditioned by the monodimensional, i.e. linear, nature of representation. As for the label of (4iii) I will refer to Moro (2000) for a full argument. In a nutshell, the idea is that phrase markers are genuine and primitive syntactic entities (as suggested in Kayne 1994 contra Chomsky 1995). Thus a structure like (4iii) is nothing but the features associated to a maximal projection with no further categorial information (see also note 17). The observation that traces are not visible to the PF component, independently of whether they are considered as copies or not, was explicitly made by Kayne (1994: 133, footnote 3) and Chomsky (1995: 337) but was not further pursued in those original papers. Another difference between the two theories can be straightfowardly highlighted by adopting Van Riemskjik’s (1995) terminology. The standard theory is a “dragchain” theory of movement whereas the alternative theory presente here is a “push-chain” theory of movement, in that the trigger for movement in the former is given by the “landing site” while the trigger in the latter is given by the “launching site” of movement. An interesting domain to test the empirical hypothesis presented here (and more generally to test the theory of antisymmetry) is Sign Language (SL). More specifically, since SL appears to involve movement, it would be interesting to explore whether it could also be correlated to the linearization processes that occur in this domain. For a critical discussion of linearization processes in SL

420

24.

25. 26.

27. 28. 29.

30.

31.

Andrea Moro (and the structure of SL in general) see Neidle, Kegl, Maclaughlin, Bahan & Lee (1999) and references cited there. I am indebted to Carlo Cecchetto and Sandro Zucchi for advice on this topic. The first obvious one is whether the conjecture could be considered as an “if and only if” proposition, including all types of movement; of course this is the more interesting step to take and one that I am temptatively pursuing. Whether or not this is right, only further research will tell. I will consider it a success if Dynamic Antisymmetry will allow us to ask the right questions, more than ensure that we find the right answers. The analyses illustrated here have been developed originally in Moro (2000) and Moro (2003) and are reproduced here to support the main thesis. Strictly speaking, following Chomsky (2001) one should not use the terms “copy” and “original”: each occurrence of an element is the occurrence of the same element in distinct positions. I just refer to these terms as purely descriptive labels. As for intermediate traces, I will simply refer to Moro (2000). For the analysis of points of symmetry constituted by heads see Moro (2000: 84–92). In the causative sentence, the subject appears as a clitic in one case; I will refer to Guasti & Moro (2001) for the discussion of such specific characteristics. English and Italian differ here, since in Italian it is not necessary that the two DPs agree in number as opposed to English (cf. *this type of books). This form of agreement could be perhaps related to ECP reasons or to the mass/count distinction: I will not explore this issue here. The hypothesis that the VP embedded in causatives has different positions related to the presence of à can be supported by a French dialect, as was pointed out to me by Luigi Rizzi. In that dialect, where the subject of the causative can be cliticized as in (i): (i) Marie le fait [SC t [laver la voiture]] the object inside the causative can be cliticized onto the embedded verb only if à is absent: (ii) a. Marie le fait t [la laver t] b. *Marie fait [la laver t] à Jean One possible way of thinking to explain this fact is that there is movement of the embedded VP to a higher position and that la constitutes a further point of symmetry with the matrix verb: I will leave the elaboration of this explanation to future research. If the copula is considered to be the spell-out of I°, the generalization suggests itself here that bare small clauses would be the complement of functional heads only, i.e. I°, D° and C°. Moreover, if small clauses really are the only implementation of predication in syntax, this would amount to saying that a predicative structure can only be the complement of a functional head. For a further refinement of the notion of “bare” and “rich” small clause see Pereltsvaig (2001a).

Linear Compression as a Trigger for Movement

421

32. The empirical arguments I am referring to can be found in Moro (2000: 43–48): they involve the distribution of adverbs, cliticization and the occurrence of predicative markers such as as and its equivalents in Italian (for example, come). This refines a previous analysis that did not distinguish between bare and rich small clauses and wrongly assumed that bare small clauses contained a head: see Longobardi (1988), Moro (1988) and Cardinaletti & Guasti (1995) for a source of different analyses. 33. Notice that the idea that the relation between a predicate and a subject is not mediated by a head, or equivalently that the two c-command each other, essentially reproduces the original intuition by Williams (1980) according to which two such roles where just defined as two mutually c-commanding projections. Indeed Williams’original proposal appears to be still valid under different perspectives. For a detailed discussion concerning labelling and Merge as defined in the Minimalist framework see Moro (2000). 34. Remember that for the theory of Antisymmetry there are no intermediate projections, i.e. there is no distinction between specifers and adjuncts. In particular here there is no I’: when VP is merged with I°, IP is yielded. This is not to say that the IP system is defective in that it cannot have specifiers/adjuncts; in fact it can, so for example adverbs can be specifiers of the IP system but not subjects. This view also has a non-trivial consequence that will not be discussed here, namely that IPs can be predicates. 35. The difference among languages will still be reduced to the possibility of properly governing the trace of the subject, i.e. to the possibility for pro to occur in subject position, as in the traditional theory. 36. I will not consider here some further important properties distinguishing left periphery Focus from postverbal one, such as contrastive properties etc. 37. Just to give one simple example: there is no focus on the cause of the riot in a picture of the wall was the cause of the riot but there must be focus on a picture of the wall in the cause of the riot was a picture of the wall. 38. The advantage with respect with the ad hoc solution we are refusing here is that we do not assume an intermediate Foc° head dedicated to and thus just maintain the more parsimonious structure. 39. Of course, inverse copular sentences show up only when the predicate can occur in the same position as the subject, namely when the predicate is a noun phrase, for morphological reasons. This type of structure is to be kept carefully distinct from cases of “locative inversion”. Hoekstra-Mulder (1990) for example suggested that unaccusatives can be analyzed as “locative inversion” constructions where spec-IP is occupied by a PP (i)a on a par with copular sentences like those in (i)b (see Hoekstra-Mulder 1990: 28 ff.): (i) a. [IP [PP in the room]i entered [SC a man ti ]] b. [IP [PP in the room ]i was [SC a man ti ]]

422

Andrea Moro This analysis, which explicitly subsumes the theory of there-sentences as inverse sentences proposed in Moro (1990) (cf. Hoekstra-Mulder 1990: 33 ff.), cannot be maintained for empirical reasons. Consider the following examples: (ii) a. b. c. d.

[DP the cause of the riot]i is [SC a man ti ] [DP the cause of the riot]i is/*are [SC John and Mary ti ] [PP in the room]i is [SC a man ti ] [PP in the room]i are/*is [SC John and Mary ti ]

Assuming that agreement is invariantly established in spec-IP, it would be hard to assume that (i)a and (i)b instantiate the same type of structure. Rather, (i)b is an instance of the topic constructions that have been analyzed by Cinque (1990) and Rizzi (1997) involving some portion of the scattered CP layer (and movement of the verb to a higher functional head). This would explain why the equivalent of (i)b in Italian involves a locative clitic ci which is not allowed in inverse copular sentences: (iii) a. [IP [PP nella stanza]i *(c’) era [SC un uomo ti ]] b. [IP [DP la causa della rivolta]i (*c’) era [SC un uomo ti ]] The status of ci with the copula is discussed in detail in Moro (1997a: ch. 2; and summarized in the Appendix of Moro 2000). As for the lack of ci in Italian unaccusatives (but not in many Northern Italian dialects as observed by Burzio 1986 and Poletto 1993 among others) see Moro (1997a). 40. Notice that type can occur as a predicate in a copular constructions such as in a mammal is a type of animal. Interestingly, however, of must show up in cases like many books are *(of) this type. For the role of prepositions as copulas (in noun phrases) see the seminal work by Den Dikken (1997). 41. Notice that if di is present, NP-stranding and agreement can cooccur: (i) quali ha scritto di racconti? (which-mas.plur. has written of novels-masc.plur.) 42. For a critical approach to government within noun phrases and its impact on the general system, see Giorgi-Longobardi (1986, ch. 2). 43. That the distinction between overt vs. null head is relevant for government has been accepted since at least Rizzi’s (1990) theory of extraction of preverbal subject in embedded sentences (cf. who do you think [C° e] t left vs. * who do you think [C° that] t left) although the possibility to govern a trace was reserved to null C° (which is considered to be the agreeing complementizer in English as opposed to that), contrary to the case studied here. 44. I am not considering here the interesting case of exclamatives where sentences like quanto alta è quella torre! (how tall is that tower) appears to be partially acceptable. For a discussion on exclamatives see Portner & Zanuttini (to appear). 45. A potential counterexample to the theory presented here could come from a pair like the following:

Linear Compression as a Trigger for Movement

423

(i) a. pro sono io+Nom (pro am I) “it’s me” b. *io sono (I am) Apparently the prediction made by the standard theory considering movement as a last resort operation is borne out: since io can stay in situ in (i)a (while the phonologically null element pro occurs in preverbal positon) io cannot move in (i)b manifesting the alleged “last resort” nature of movement. No mirror structure is in fact generated here. This conclusion however is not correct. The reason why (i)b is ungrammatical is independent of movement. Let us consider the following sentences with the associated structure: (ii) a. *Maria considera [ Gianni pro ] (Maria considers Gianni pro) b. *MMaria considera [ pro il colpevole ] (Maria considers pro the culprit) c. *io sono [ t pro ] (I am) In Italian pro cannot be licensed within a (bare or a rich) small clause (in fact the only proper environment is a relation with a rich verb inflection; see also Rizzi (1986) for non verbal pro licensing), thus (i)b is ruled out by the same reasons as (ii)a–b and has nothing to do with movement. The copula must always occur with two maximal projections linked by predication: since pro cannot be licensed in a small clause, (i)b just lacks one maximal projection for the copula to be interpreted. All in all, the contrast in (i) does not stand as a challenging counterexample to Dynamic Antisymmetry: indeed, as far as the data in (8) are concerned it seems that a theory assuming that movement is a last resort operation would not be empirically adequate. As for the contrast between sono io and io lo sono, i.e. for the contrast between propredicative lo and propredicative pro see Moro (1997) and references cited there. 46. I am indebted to Giorgio Graffi for pointing me out these cases. 47. In Italian, clitics are assumed to be endowed with Case features, always uninterpretable. 48. As for the possibility that only the lower XP be pronounced, such as in wh-in situ constructions, Dynamic Antisymmetry would force us to analyze them as involving “remnant movement”. Leaving aside the reasons which trigger this movement, take for example a simple case like the following echo question: (i) hai visto cosa? (have seen what) The only analysis compatible with Dynamic Antisymmetry would be the following, crucially excluding adjacency between V° and cosa:

424

Andrea Moro (ii) a. cosa [hai visto t] b. [ [hai visto t] H° [ cosa t ]]

In such a case, no LCA problem would raise since there would be no symmetrical c-command between the verbal head visto and the head cosa (for the analysis of cosa as a head see Moro (2000) and references cited there). Interestingly, notice that cosa would be focused, witness the normal intonational emphasis it receives when it is pronounced in situ: cosa is in fact in situ within a bare small clause as the subject of an inverse copular sentence (see discussion on in situ Focus in this section and in section IV). 49. Interestingly, it is never the case that a moved element is doubled by a stressed pronoun. If the analysis associating clitics to heads and stressed pronoun to full phrases XP is correct (see for example the seminal work by Kayne 1989; see also Cardinaletti & Starke 1994, Sportiche 1992 and Moro 2000 for convergent approaches to this idea) then Dynamic Antisymmetry would also explain why stressed pronouns do not occur in substandard Italian relative clause and interrogative “doubling” constructions: a stressed pronoun, being an XP, would reconstitute the point of symmetry whereas a clitic pronoun, being an X°, would not. 50. The case discussed in Moro (2000) involved pro in verbal agreement: see section 3.3.2. of that monograph. 51. Taking contrastive focus to be a test, one can easily construe the relevant minimal pair: (i) a LUI telefona, non lei (he telephones not she) b *pro telefona, non lei/loro (telephones not she/them) Notice however that one can have a contrast in sentences of the type in (ii): (ii) pro TELEFONA, non telefonano (telephones not telephone) (s/he telephones not they telephone)

52.

53. 54. 55.

The sentence in (ii) however is irrelevant because the contrast is in fact expressed by the verb (as mediated by inflection). Notice that in our framework the impossibility for pro to move to spec-Foc° follows as a particular case of a more general restriction following from Dynamic Antisymmetry (as indicated in (6v)). I am grateful to an anonymous reviewer for pointing me out this issue. For a detailed discussion see Moro (2000: ch. 4). On a different approach to covert movement see also Bobalijk (to appear). As for the possibility for specifiers to be heads, Cinque 1996 also noticed that the requirement “that a head cannot be a specifier is also derived, albeit via a further assumption (“that the highest element of a chain of heads must have a specifier” – Kayne 1994: 31) If a head, in order to be licensed, needs to project

Linear Compression as a Trigger for Movement

56.

57. 58. 59. 60.

61.

425

(and discharge its e-role(s)), it follows that the source of a head in specifier position must be a lower head position. But then the possibility arises of excluding its moving to a specifier position as a violation of Relativized Minimality (Rizzi 1990; or “Shortest Movement” – Chomsky 1995). A closer potential landing site (the head of the phrase it adjoins to) is skipped (this still does not prevent a head from becoming its own specifier)” Cinque (1996: 449, fn. 6). Thus, if Relativise minimality conditions can be refined to the extend that not all heads per se can function as intervenors for other heads, the possibility for a head to be a specifier cannot be excluded. Interestingly, if (spec-head) agreement between a head and another head as its specifier were allowed, intermediate activation of agreement in Romance languages could be interpreted without assuming that the intermediate copies/traces of clitics are full noun phrases. Pushing this speculation to the limit, one could conclude that (at least when it comes to word order) the differences across languages should also be observable within languages, much in the sense that Italian is an OV and VO language, as suggested in the text. See also Rothstein (1983) for some extension of the original proposal by Williams to secondary predication. The term “functional” has not been used by Chomksy. In a sense, this can be regarded as a “last resort” quality of movement, although it is so in a very different way w.r.t. the standard minimalist theory. For an extension of the Hale & Keyser’s (1993) theory to unaccusatives see Moro (1997b: chapter 5) and references cited there. See also Hale & Keyser (2003, chapter 6). That predication is to be kept conceptually distinct from theta assignment can be proved in several ways (see Moro 1991, 1997a and references cited there). A major argument in transformational grammar comes from the existence of active/passive alternation: for a pair like John reads a book and a book is read by John (where the subject is John and a book, respectively) one surely wants the same VP, i.e. the same underlying thematic structure, to generate the two sentences where the subject is the external and the internal argument, in the active and passive sentence respectively. Other cases can be observed in copular constructions where sentences like this is Dante’s desire for Beatrice shows that although the theta roles pertaining to the lexical head desire are exhausted within its maximal projection, still that maximal projection can be the predicate of another DP, i.e. this.

426

Andrea Moro

References Belletti, A. 1990 Generalized Verb Movement. Torino: Rosenberg & Sellier. Bennis, H. 1986 Gaps and Dummies. Dordrecht: Foris. Bennis, H., M. den Dikken and N. Corver 1998 Predication in Nominal Phrases. In The Journal of Comparative Germanic Linguistics 1: 85–117. Bobalijk, J. to appear A-Chains at the PF-interface: copies and covert movement, Natural Language and Linguistic Theory. Brody, 1995 Lexico-logical form: a radically minimalist theory. Linguistic Inquiry Monographs. Cambridge Mass.: MIT Press. Burzio, L. 1986 Italian Syntax. Reidel, Dordrecht. Cardinaletti, A. and M.-T. Guasti (eds.) 1995 Small Clauses. Chicago: Academic Press. Cardinaletti, A. and M. Starke 1999 The Typology of Structural Deficiency. On Three Grammatical Classes. In Clitics in the languages of Europe. Empirical approaces to language typology, Riemsdijk, H. van (ed.), 20 –5. Berlin/New York: Mouton de Gruyter. Chomsky, N. 2000 Minimalist inquiries: The framework. In Step by step, R. Martin, D. Michaels, and J. Uriagereka (eds.) Cambridge, Mass.: MIT Press. 2001 Beyond expanatory adequacy. MITOPL 20, Cambridge Mass. Cinque, G. 1988 La frase relativa. In Grande grammatica italiana di consultazione, Vol. I, L. Renzi (ed.). Il Bologna: Mulino. 1990 Types of A-bar Dependencies, Linguistic Inquiry Monographs 17. Cambridge, Mass.: MIT Press, 1996 The antisymmetric program: theoretical and typological implications. Journal of Linguistics 32: 447– 464. Den Dikken, M. 1997 Predicate Inversion in DP. In Possessors, predicates and movement in the determiner phrase, Alexiadou-Wilder (eds.), 177–214. Amsterdam: Benjamins. Giorgi, A. and G. Longobardi 1991 The Syntax of Noun Phrases. Cambridge: CUP. Graffi, G. 1997 Frasi complete e frasi ridotte. Lingua e Stile 32, 2: 273–291. 2001 200 Years of Syntax. A critical survey. Amsterdam: John Benjamins.

Linear Compression as a Trigger for Movement

427

Groat, E. and J. O’Neill 1997 Spell-out at the Interface: achieving a unified syntactic computational system in the minimalist framework. In Minimalist Ideas, W. Abraham, S. D. Epstein, H. Thráinsson and J.-W. Zwart (eds.). Guasti, M. T. and A. Moro 2001 Causative sentences and Dynamic Antisymmetry. In Current Studies in Italian Syntax. Essays offered to Lorenzo Renzi, Cinque and Salvi (eds.), North-Holland 2001, Amsterdam. Haegeman, L. and R. Zanuttini 1991 Negative heads and the Negative Criterion. The Linguistic Review 8: 233–251. Hale, K. and J. Keyser 1993 On the argument structure and the Lexical expression of syntactic relations”. In The View from Building 20: Essays in Honor of Sylvain Bromberger, K. Hale and J. Keyser (eds.). Cambridge, Mass.: MIT Press. 2002 Prolegomena to a theory of argument structure. Linguistic Inquiry Monograph 39. Cambridge, Mass.: MIT Press. Hoekstra, T. and R. Mulder 1990 Unergatives as Copular verbs. The Linguistic Review 7: 1–79. Hornstein, N. 1995 Logical Form, Oxford, UK: Blackwell. Kayne, R. 1989a Null Subjects and Clitic Climbing. In La grammaire modulaire, Ronat, M. and Couquaux (eds.), Édition de Minuit. 1989b Facets of Romance Past Participle Agreement. In Benincà (1989). 1994 The Antisymmetry of Syntax. Linguistic Inquiry Monograph 25. Cambridge, Mass.: MIT Press. Longobardi, G. 1988 Symmetry Principles in Syntax. Padova: Clesp. Mayr, E. 1988 The growth of biological thought. Diversity, evolution and inheritance. Cambridge, Mass.: Belknap Harvard. Medawar, P. B. 1967 The Art of Soluble. Londra: Methen. Moro, A. 1988 Per una teoria unificata delle frasi copulari. Rivista di Grammatica Generativa 13: 81–110. 1996 Dynamic Antisymmetry. XIX GLOW Conference. University of Athens. 1997a The Raising of Predicates. Cambridge University Press, UK. 1997b Dynamic Antisymmetry: movement as a symmetry-braking phenomenon. Studia Linguistica 51: 50–76. 1996 Dynamic Antisymmetry: movement as a symmetry-breaking phenomenon. Working Papers 13, Department of Cognitive Sciences, Istituto Scientifico San Raffaele.

428

Andrea Moro

2000

Dynamic Antisymmetry. Linguistic Inquiry Monograph 38. Cambridge, Mass.: MIT Press. 2002 Linguistica Mendeliana ovvero quali domande su genetica e grammatica? Lingua e Linguaggio 1: 39–58. 2003 The role of linear compression in human languages: two questions on movement. Lingua e Linguaggio 3: 31–68. Neidle, C., J. Kegl, D. Maclaughlin, B. Bahan and R. G. Lee 1999 The Syntax of American Sign Language. Functional categories and hierarchical structure. Cambridge, Mass.: MIT Press. Nuñes, J. 2001 Sideward Movement. Linguistic Inquiry 32: 303–344. Pesetsky, D. 1997 Optimality Theory and Syntax: Movement and Pronunciation. In Optimality Theory: an overview, D. Archangeli and T. D. Langendoen (eds.), 134–170. Malden, Mass.: Blackwell. Piattelli Palmarini, M. and J. Uriagereka in press The Immune syntax: the evolution of the language virus. In Variation and Universals in Biolinguistics, Jenkins, L. (ed.). Oxford: Elsevier. Pike, K. 1943 Taxemes and immediate constituents. Language 19: 65–82. Pollock, J.-Y. 1989 Verb Movement, UG and the Structure of IP. Linguistic Inquiry 20: 365–424. Poletto, C. 1993 La sintassi del soggetto nei dialetti italiani settentrionali. Quaderni Patavini di Linguistica, Monograph 12. Padova: Unipress. Portner, P. and R. Zanuttini to appear The semantics of nominal exclamatives. In Ellipsis in non-sentential speech, R. Elugaro and R. J. Stainton (eds.). Dordrecht: Kluwer. Sportiche, D. 1992 Clitic Constructions. In Phrase Structure and The Lexicon, J. Rooryck and L. Zaring (eds.). Dordrecht: Foris. Rizzi, L. 1986 Null Objects in Italian and The Theory of pro. Linguistic Inquiry 17: 501–558. 1990 Relativized Minimality. Linguistic Inquiry Monograph 16. Cambridge, Mass.: MIT Press. 1996 Residual Verb Second and the WH-criterion. In Parameters and Functional Heads, Belletti, A. and L. Rizzi (eds.), 63–90. Oxford/ New York: Oxford University Press. 1997 The fine structure of the left periphery. In Elements of Grammar: Handbook of Generative Syntax, L. Haegeman (ed.), 281–337. Dordrecht: Kluwer.

Linear Compression as a Trigger for Movement

429

Rothstein, S. 1983 The Syntactic Form of Predication. Ph.D. diss., MIT: Cambridge, Mass. Starke, M. 2001 Move is Merge. Ph.D. Thesis, University of Geneva, Switzerland. Steriade, D. 1980 Clitic doubling in Romanian Wh-constructions and the Analysis of Topicalization. Papers from the 16th regional meeting of the Chicago Linguistic Society, University of Chicago. Stowell, T. 1981 Origins of Phrase Structure. Ph.D. dissertation, MIT: Cambridge, Mass. Uriagereka, J. 1999 Multiple Spell Out. In Working minimalism, D. S. Epstein and N. Hornstein (eds.). Cambridge, Mass.: MIT Press. Williams, E. 1980 Predication. Linguistic Inquiry 11: 203–238. van Riemsdijk, H. 1995 Push chains and drag chains. Ms., University of Tilburg. Zamparelli, R. 1995 Layers in the Determiner Phrase. Ph.D. Dissertation, University of Rochester.

A Case for Head Movement at PF: SAI in Comparatives Fumikazu Niinuma and Myung-Kwan Park

1. Introduction The main concern of this paper is optional head movement in English comparatives, as shown in (1) and (2): (1)

She spoke more convincingly than Harry did

(2)

She spoke more convincingly than did Harry

In addition to the unmarked subject-Aux order in (1), the apparently marked Aux-subject order in (2) is possible in the comparative clause (cf. Doherty and Schwartz 1967). Following the standard analysis (cf. Emonds 1970), we assume that (2) is derived from (1) via subject-Aux inversion (SAI), or I-to-C movement in the comparative clause. When we consider this phenomenon under the minimalist perspective, we face (at least) two interesting issues. The first question concerns apparant optionality, which is problematic given the minimalist assumptions. If SAI generates (2) from (1), we have to explore whether it occurs optionally without any motivation or effect. The second question concerns the treatment of head movement. It was a standard assumption that head movement applies within narrow syntax. However, recently Chomsky (2000, 2001a,b) and Boeckx and Stjepanoviç (2001) proposed that head movement is an operation at Phonetic Form (PF), so that we can avoid some problems with head movement (such as the fact that head movement apparently does not meet the Extension Condition).1 In this paper, we argue that head movement in English comparatives is a PF operation. More specifically, we show that SAI in comparatives interacts with ellipsis and sentence stress assignment, which are standardly assumed to occur at PF. The implications of the analysis are as follows. First, we need to draw a line between SAI in comparatives, on the one hand, and SAI in yes/no and wh questions, on the other hand. The two types of SAI differ in

432

Fumikazu Niinuma and Myung-Kwan Park

the interaction with sentence stress assignment at PF. The former interacts with sentence stress assignment, whereas the latter does not. Second, note that only when the head movement at hand is treated as a PF phenomenon can we account for the interaction among ellipsis, sentence stress assignment rule and head movement in comparative constructions. The organization of this paper is as follows: In section 2, we will bring into light the properties of SAI in comparatives and present Merchant’s (2003) analysis of them. In section 3, we will point out some problems with Merchant’s analysis. Then in section 4, we will turn to propose an alternative analysis of SAI in comparatives. In section 5, we will show that the proposed analysis can also naturally account for the interaction of SAI and Pseudogapping in comparatives. Section 6 is a summary of this paper.

2. Comparatives In this section, we examine the phenomena of SAI in comparatives more carefully. First, as noted by Merchant (2003), VP must be elided under SAI in comparatives. However, VP does not have to be elided if SAI does not apply in comparative clauses. This is shown in (3– 4): (3)

a. Abby can play more instruments than can her father (*play) b. Abby can play more instruments than her father can play

(4)

a. Abby has been awarded more accolades than has her father (*been awarded) b. Abby has been awarded more accolades than her father has been awarded

Second, pseudogapping is prohibited when SAI is applied in comparative clauses, as shown in (5–6): (5)

a. *Abby plays the flute better than does her father the trumpet b. Abby plays the flute better than her father does the trumpet

(6)

a. *Abby can play more sonatas than can her father concertos b. Abby can play more sonatas than her father can concertos

Based on the data above, Merchant (2003) made the following generalization:

A Case for Head Movement at PF: SAI in Comparatives

(7)

433

I-to-C movement in comparative clauses can occur only if VP-ellipsis has deleted the VP complement to I°.

.

In other words, the prerequisite for SAI in comparatives is VP-ellipsis. In order to account for the generalization (7), Merchant relies on the notion ‘the ECP at PF,’ as defined below: 2 (8)

The Empty Category Principle (ECP) at PF At PF, a trace of A’-movement must either be i) PF-head-governed, or ii) PF-antecedent-governed.

According to (8), A’-traces must be either head- or antecedent-governed at PF by the element which has phonetic contents. More specifically, neither traces nor null operators can be proper governors at PF, as they are phonetically null elements, by definition. With this in mind, let us consider how (8) accounts for the ungrammaticality of (3a) with the VP in the comparative clause unelided, which is repeated below for convenience. The comparative clause can be represented in more detail as in (9b): (9)

a. *Abby can play more instruments than can her father play b. … than [CP OP1 can [IP her father tcan [VP t1’ [VP tSU play t1]]]]

Merchant (2003) assumes that a null operator moves from the complement of play to the VP-adjoined position and then finally to the Spec CP. Let us focus on the intermediate trace t1’ in (9b). Notice that the candidate for the head-governor of this trace is the trace of can, which is not PF-visible. Therefore, t1’ cannot be PF-head-governed. Since its possible antecedent is the null operator in the [CP, Spec] position of the comparative clause, this trace cannot be PF-antecedent governed, either. Thus, it ends up violating the ECP at PF, and the sentence (9a) turns out to be ungrammatical, as predicted. Let us now consider how Merchant (2003) accounts for the grammaticality of (3a) with the VP in the comparative clause elided, which is repeated as (10a). The comparative clause can also be represented in more detail as in (10b): (10) a. Abby can play more instruments than can her father b. … than [CP OP1 can [IP her father tcan [VP t1’ [VP tSU play t1]]]]

434

Fumikazu Niinuma and Myung-Kwan Park

(10b) is the same as (9b) except that the whole VP is elided. Following the logic of Lasnik (1995, 1999) and Merchant (2001), Merchant (2003) argues that ellipsis can save violations due to the ECP at PF. In particular, Merchant claims that when the offending trace (t1’) in (10b) is deleted by VP-ellipsis, then it is not subject to the ECP any longer at PF. The repair of the ECP violation by ellipsis in (10) is analyzed on a par with that of the Subjacency violation by ellipsis in (11b), which Merchant (2001) claims obtains at PF: (11) a. *John knows the person who bought something, but I don’t know what1 [John knows [island the person who bought t1]] b. John knows the person who bought something, but I don’t know what1 [John knows [island the person who bought t1]] In (11b) the (IP-)ellipsis nullifies the effects of the island environment, which accounts for the grammaticality of the sentence in contrast to a version of the sentence without the ellipsis involved.3

3. Problems Even though his analysis can account for the examples examined above, there are three pieces of evidence against Merchant (2003), so that his analysis is insufficient. First, as noted by Emonds (1970: 9), pronouns cannot appear at the end of comparative clauses where both VP ellipsis and SAI occur. More correctly, the generalization seems to be that weak pronouns cannot appear at the end of comparative clauses, but that strong pronouns that receive stress can, as illustrated below. Note that the weak pronoun can appear in the comparative clause only when SAI does not occur in that clause. (12) a. *John likes Beethoven more than do you b. John likes Beethoven more than do YOU c. John likes Beethoven more than you do (13) a. *Abby can play more sonatas than can he b. Abby can play more sonatas than can HE c. Abby can play more sonatas than he can Note that, for example, (13a) and (13b) are not distinguished structurally, as represented in (14a) and (14b). If this is the case, Merchant’s analysis will

A Case for Head Movement at PF: SAI in Comparatives

435

predict that there should be no contrast in grammaticality between (13a) and (13b), contrary to fact: (14) a. … * than [CP OP1 can [IP he tcan [VP t1’ [VP the play t1]]]] b. … than [CP OP1 can [IP HE tcan [VP t1’ [VP tHE play t1]]]] Second, the expletive there cannot appear at the end of comparative clauses, as shown in (15): (15) a. There are over 830,000 more jobs in Australia than there were. b. *There are over 830,000 more jobs in Australia than were there. Expletives do not have any semantic content. Hence they cannot receive sentence stress. If so, the examples in (12)–(13) and (15) confirm that SAI in comparatives is affected by sentence stress assignment. Third, Potts (2002: footnote 8) notes that more than one head can appear before the subject in comparative clauses, as shown in (16). By contrast, notice that in yes/no questions, more than one head cannot move to the C position, as in (17). This tells us that comparatives involve some other operation which is different from canonical syntactic head movement in yes/no questions: (16) a. b.

Eddie has been flying longer than has been Chuck By the end of the trip, Klaus will have seen many bats, as will have Eddie

(17) a. *Has been Chuck flying longer? b. *Will have Eddie seen many bats? To summarize, we have pointed out some pieces of evidence against Merchant (2003), which suggest that we have to look for an alternative analysis of SAI in comparative clauses.

4. Proposed Analysis As shown in the previous section, we have to account for two facts, namely, both the interaction of head movement with sentence stress assignment and the possibility that more than one head can move to the C position. In this section, we will explore an alternative analysis of apparently optional SAI in comparative clauses in English.

436

Fumikazu Niinuma and Myung-Kwan Park

Before presenting our analysis of optional SAI in comparatives, we would like to introduce the analysis of scrambling in Dutch that Reinhart (1997) and Reinhart and Neeleman (1998) advanced. Reinhart and Reinhart and Neeleman argued that scrambling in Dutch interacts with sentence stress assignment that brings about focus effects. As in (18b) and (19a), scrambling of the direct object is impossible when the direct object receives sentence stress. As shown by the examples in (18a) and (19b), on the other hand, it is possible when the verb rather than the direct object is stressed: (18) a. Ik heb nog niet DE KRANT gelezen, maar ik heb I have not yet the newspaper read, but I have al wel already indeed

HET BOEK the book

gelezen. read

b. *Ik heb DE KRANT nog niet gelezen, maar ik heb HET BOEK al wel gelezen (19) a. Ik heb het boek gisteren GELEZEN en niet VERSCHEURD I have the book yesterday read and not torn up b. *Ik heb gisteren het boek GELEZEN en niet VERSCHEURD. Based on this contrast, Reinhart and Reinhart and Neeleman argued that scrambling makes it possible to apply the unmarked rule of sentence stress assignment, avoiding the marked one. More specifically, they assumed the sentence stress rule proposed by Cinque (1993). The gist of Cinque’s analysis is that the main stress of the sentence will always fall on its most embedded constituent. Following Halle and Vergnaud (1987), Cinque assumed a metrical grid approach to the nuclear stress rule. The previous analyses in this framework assumed that the nuclear stress rule is parametrized in order to account for the varying stress patterns. For example, the nuclear stress in English falls on the right node, but in Dutch it falls on the left node: (20) I read the book (21) dat ik het boek las that I the book read Departing from the previous analyses, however, Cinque argued that no parametrization of the nuclear stress rule is necessary, and that varied word orders across languages are governed by the universal nuclear stress rule.

A Case for Head Movement at PF: SAI in Comparatives

437

The idea of his analysis is that sentence stress is assigned on the most deeply embedded constituent. English is a SVO language and the most deeply embedded constituent in this language is the right-most element inside VP. On the other hand, since Dutch is a SOV language, the most deeply embedded constituent is the left-most element inside VP. Therefore, when the object receives stress, it must stay in the complement position, as shown in (18). By contrast, if the verb receives neutral stress, the object has to move out of VP in order to avoid receiving neutral stress, as illustrated in (19). Notably, in Reinhart’s and Reinhart and Neeleman’s analysis, scrambling is regarded as a PF phenomenon, in that it interacts with sentence stress assignment at PF. As a consequence, scrambling in Dutch is not optional, but obligatory, since it changes sentence stress pattern. It is noteworthy that scrambling in Dutch apparently shares some properties with SAI in the comparative clauses of English. First, both phenomena are affected by sentence stress assignment. Second, both operations apparently look to be optional, though they are in fact not. Thus, we propose, along the lines of analysis by Reinhart (1997) and Reinhart and Neeleman (1998), that English has an option of moving a head element at PF when it creates a structural context for the application of the unmarked rule of sentence stress assignment rather than the marked one.4 More specifically, we argue that I-to-C movement in comparatives takes place after Spell-Out (see Chomsky (2000) for the argument for PF head movement and Lechner (2001) for the argument against PF head movement). In particular, head movement in comparatives enables the subject NP to receive unmarked/ neutral stress rather than marked stress.5 According to our analysis, the sentence (22a) is derived by the PF-head movement of the auxiliary can, as represented in (22b). (22) a. Abby can play more instruments than can her father b

… than [CP OP1 can [IP her father tcan [VP t1’ [VP tSU play t1]]]] : # z_________--___m movement at PF

After I-to-C movement in (22b), the subject her father counts as the most embedded (right-most) element, which is assigned neutral sentence stress (default stress) (cf. Cinque 1993). This means that I-to-C movement in comparatives is triggered when it makes a subject NP receive neutral sentence stress, avoiding marked one (contrastive focus). Under our analysis, the contrast in (23) is not an unexpected one. Only the stressed strong pronoun can occur at the end of the comparative clause

438

Fumikazu Niinuma and Myung-Kwan Park

because SAI here makes it possible for the rightmost subject NP to receive neutral sentence stress: (23) a. *Abby can play more sonatas than can he b. Abby can play more sonatas than can HE Notice that in order for the subject NP to be assigned neutral sentence stress, the whole VP must be elided, as shown in (22b). If, as standardly assumed, VP-ellipsis and sentence stress assignment are PF operations and if some instance of head movement is also a PF operation, then it is expected that there is an interaction among them. The phenomenon of SAI in the comparative clauses of English clearly points to this interaction.6 Let us turn now to consider how to account for the ungrammaticality of (24a). The relevant structure will be in (24b). (24) a. *Abby can play more instruments than can her father play b. … than [CP OP1 can [IP her father tcan [VP t1’ [VP tSU play t1]]]] : # z_________---___m movement at PF Economy at PF can explain the ungrammaticality of (24a) along the lines of an argument by Reinhart (1997), who claims that scrambling in Dutch applies only when it is needed to derive a different word order with a concomitant different focus structure. Otherwise, it cannot be applied, because of economy at PF.7 Given this assumption, let us look at (24a). The rightmost element of this sentence is the verb play, whether SAI applies or not. That is, SAI in (24a) does not change the focus structure of the resulting sentence since the neutral sentence stress falls on the sentence-final element, i.e., the main verb play. That is why SAI in (24a) results in a violation of economy at PF. Therefore, the grammatical option is not to apply SAI, producing the following sentence: (25) Abby can play more instruments than her father can play Note that if our analysis is on the right track, we do not have to worry about the presence of the intermediate trace left by the null operator which Merchant is concerned with, and therefore we can dispense with the ECP at PF. Furthermore, note that in our analysis, SAI in comparatives is considered not to be optional, because it applies only when it makes a context for the application of the unmarked rule of sentence stress assignment. In other

A Case for Head Movement at PF: SAI in Comparatives

439

words, SAI is constrained by the focus structure of a resulting sentence. This is why pronouns in the subject position have to be strong ones that receive sentence stress when they appear in the sentence-final position of comparative clauses. Furthermore, if the head receives focus, SAI cannot take place, as illustrated below: (26) a. Abby MUST play more often than her sisten CAN b. ??Abby MUST play more often than CAN her sister (27) a. Abby can PLAY more instruments than her father can IMITATE b. *Abby can PLAY more instruments than can her father IMITATE The above contrasts follow from our proposed analysis. Since the auxiliary in (26) receives focus, it has to appear at the sentence-final position. Otherwise, the sentence is unacceptable, as illustrated in (26b). Futhermore, when neutral stress falls on the main verb, SAI cannot take place because of economy at PF. This is why the contrast between (27a) and (27b) obtains. Now, let us consider the following example, where more than one head appear before the subject: (28) Eddie has been flying longer than has been Chuck Suppose that both heads move as a unit after cluster formation applies.8 Given this assumption, the verb cluster moves to the C position at PF, and then the subject results in being apparently realized at the end of comparative clauses, as shown below: (29) … than [CP has been1 [IP Chuck t1 [VP flying longer]]] : # z_________---___m movement at PF There is some good reason to believe that only heads move to the pre-subject position in (29). The sentence in (30) makes the point: (30) *Eddie has been flying longer than has probably been Chuck If, as standardly assumed, the adverb probably adjoins to the maximal projection such as TP, then the ungrammaticality of (30) can be taken to imply that only heads move to C at PF.9 If our analysis presented so far is correct, then we can draw a line between SAI in comparatives and SAI in yes/no and wh questions. This is so

440

Fumikazu Niinuma and Myung-Kwan Park

because SAI in comparatives is constrained by the sentence stress assignment that affects the focus structure of a resulting sentence, but SAI in yes/ no questions is not. As in (31a–b), in the case of yes/no questions the rightmost element is identical, but SAI is still allowed.10 (31) a. John will leave b. Will John leave? Furthermore, SAI in comparatives can involve a verb cluster, whereas SAI in yes/no questions is limited only to one head element. Furthermore, in yes/no question, the moved element can be focused, while it is impossible under SAI in comparative clauses:11 (32) a. *John’s dad was taller than IS he b. [John’s going to vote for Bush] IS he? Based on these asymmetries, we argue that the former occurs at PF, while the latter occurs at narrow syntax (see Zwart (2001) for the similar conclusion). To summarize, we have proposed that SAI in comparatives is a PF operation. More specifically, SAI in comparatives is triggered when it enables the subject NP to receive unmarked sentence stress, avoiding marked one. It has also been noted that unlike SAI in narrow syntax, SAI at PF allows for fronting of a verb cluster. Note that Kennedy and Merchant (2000) show that pseudogapping salvages examples involving attributive subdeletion, as shown below: (33) *Pico wrote a more interesting novel than Brio wrote a play (34)

Pico wrote a more interesting novel than Brio did a play

Kennedy and Merchant (2000) argue that the difference between (33) and (34) lies in the fact that the latter involves VP-ellipsis, which repairs the unacceptable structure. If true, this indicates that two different PF-criteria must be met by SAI in comparatives, namely VP-ellipsis and sentence stress assignment.12

A Case for Head Movement at PF: SAI in Comparatives

441

5. Pseudogapping and SAI in Comparatives In contrast to the fact that VP ellipsis applies obligatorily in licensing SAI in comparatives, Pseudogapping, which has been argued to involve VP ellipsis (cf. Jayaseelan 1990; Lasnik 1995; Johnson 2001), does not allow SAI in these clauses, as illustrated in (5) and (6). Notice first that Pseudogapping occurs in comparatives and subcomparatives,13 as in (35) and (36) from Merchant (2003) and Kennedy and Merchant (2000), respectively: (35)

Abby plays the flute better than her father does the trumpet

(36)

Abby is taller than her father is short

However, Pseudogapping with SAI in comparatives is impossible, as (37) and (38) show. (37) *Abby plays the flute better than does her father the trumpet (38) *Abby played a longer sonata than did her father a concerto Our proposed analysis relying on the interaction of SAI with sentence stress assignment accounts for the ungrammaticality of (37) and (38) without introducing any additional device. SAI in these examples is impossible because it does not bring about a new way of sentence stress assignment. In other words, the presence of the element that survives the VP ellipsis operation of Pseudogapping, i.e., the trumpet in (37) and a concerto in (38), prevents the subject NP of each comparative clause from capitalizing upon the unmarked rule of sentence stress assignment. In this sense, the ungrammaticality of (37) and (38) is analyzed on a par with that of the full sentence without VP ellipsis involved, which we discussed in (24a). The latter example is repeated below as (39). (39) *Abby can play more instruments than can her father play The subcomparative construction where the gradable property compared is not identical behaves in a parallel fashion to ellipsis-less full sentences and the Pseudogapping construction. As shown by (40) and (41), SAI cannot be combined with Pseudogapping, as noted by Merchant (2003): (40)

Abby is taller than her father is short

(41) *Abby is taller than is her father short

442

Fumikazu Niinuma and Myung-Kwan Park

Note that despite SAI, the subject of the comparative clause in (41) is in non-final position. Therefore, the unmarked rule of sentence stress assignment cannot be taken advantage of, which is responsible for the ungrammaticality of the example. This clearly shows that SAI at PF is allowed only when it puts the subject of comparative clauses in final position, which enables it to receive neutral sentence stress.14 One remarkable exception that deviates from this generalization is in the following, which is drawn from Merchant (2003): (42) Klaus would be happier in the north than would Chuck in the south The example is grammatical, even though after the application of SAI the subject of the comparative clause is still in non-final position. To understand the structural property of the example, let us change the subject of the comparative clause into a pronominal, as in (43): (43) [Overall, Klaus’s wife1 was content with their move north; after all,] Klaus is happier in the NORTH than was SHE1 in the SOUTH In this version of the example, the comparative clause subject has to be stressed. This implies that the examples in (42) and (43) exhibit regular SAI effects, just like other instances that exhibit saving effects after VP ellipsis. The question that arises immediately is what distinguishes the ungrammatical sentences in (37)–(38), (39) and (41) from the grammatical sentence in (42). Note that in the former group of examples, what follows the subject of comparative clauses after the application of SAI is either an argument or a predicate, whereas in the latter example, it is an adjunct. This means that there is an asymmetry between argument/predicate and adjunct with respect to the element that can follow the subject of comparative clauses after SAI applies. The asymmetry reminds us of the distinction between argument and adjunct in the sentence stress/accent and focus relation. One of the earlier findings on the latter relation uncovered by Gussenhoven (1983, 1992) and Selkirk (1985, 1995) is that sentence stress relates syntactically to what counts as a focus constituent. For instance, the sentence in (44) is used as an appropriate answer to all the wh question sentences in (45): (44) [F(5) Mary [F(4) bought [F(3) a book [F(2) about [F(1) BATS]]]]].

A Case for Head Movement at PF: SAI in Comparatives

(45) a. b. c. d. e.

443

What did Mary buy a book about? What kind of book did Mary buy? What did Mary buy? What did Mary do? What’s been happening?

As generally acknowledged, the focus of a clause corresponds to the whphrase of any question the sentence may serve as an answer for. In other words, in the question-answer pair of (45a) and (44) the focus constituent is F(1); in the (45b) and (44) pair it is F(2), and so on.15 This means that sentence stress cues the presence of a focus constituent or that a focus constituent contains sentence stress. In capturing the relation between sentence stress and focus, Gussenhoven (1983) argues that the location of sentence stress is determined by stress percolating down from a focus constituent. Selkirk (1995) argues, on the other hand, that focus is determined by F(ocus)-marking projecting upward from a stressed constituent. Essentially, they argue that stress percolation-down or F-projection proceeds by means of syntactic relations such as head-to-maximal projection and head-complement relation. In terms of Selkirk’s analysis, the sentence-stressed element BATS in (44) projects F-marking to its maximal projection, which counts as a focus constituent. The PP about BATS can also count as a focus constituent since the NP BATS transfers F-marking to its selecting head in a head-complement relation and then the latter projects F-marking to its maximal projection PP, by convention. This proceeds on to other focus constituents in (44). One of the predictions that the theory of focus projection makes is that arguments project F-marking to a larger constituent, but adjuncts cannot. This is because, by definition, focus projection proceeds by means of headto-maximal projection and head-complement relations, but not others. Since adjuncts are excluded from these two relations, they constitute an independent domain from which F-marking cannot project upward. The following examples illustrate this point (cf. Gussenhoven 1992; Selkirk 1995): (46) a. He [F SMOKED in the TENT] b. *He [F smoked in the TENT] (47) a. He [F SANG the whole DAY] b. *He [F sang the whole DAY] As the contrast between the (a)-example and the (b)-example of (46) and (47) shows, adjuncts cannot project F-marking to a higher constituent. Therefore,

444

Fumikazu Niinuma and Myung-Kwan Park

for the whole VP to count as a focus constituent the verb has to carry a sentence stress independently and transfers F-marking up to its maximal projection through the head-to-maximal projection convention. Unlike adjuncts but in a parallel fashion to object complements, it is noteworthy that stressed subjects can project F-marking to the clauses containing them. The following examples make this point (cf. Gussenhoven 1992; Selkirk 1995): (48) [F The SKY is falling] (49) [F The BABY’s crying] (50) [F The MOTHer telephoned] (51) (Why is SHE here?) [F Her HUSband beats her] In (48)–(51), only the subject carries a sentence stress, and the whole sentence can count as a focus constituent. It is safe to say that the wide focus interpretation of each of these examples stems from upward projection of Fmarking from the subject NP that bears a sentence stress. F-marking is also projected upward from predicates. The following cases make this point (cf. Selkirk 1995): (52) [F FIREmen are ALtruistic] (cf. *[F FIREmen are altruistic]) (53) [F Your EYES are BLUE ] (cf. *[F Your EYES are blue]) As noted by Gussenhoven (1983) and Selkirk (1995), for it to count as a focus constituent, the sentence that contains an individual-level predicate as in (52) and (53) has both the subject and the predicate receive a sentence stress. The ungrammaticality of a version of the example with a sentence stress only on the subject clearly indicates that F-marking of the whole sentence is projected from the predicate rather than from the subject. We have up to now argued that arguments and predicates project focusdetermining F-marking to higher constituents, whereas adjuncts do not. In other words, the former can affect the focus structure of a sentence, but the latter cannot. Viewed in terms of sentence stress assignment, this means that adjuncts are distinguished from arguments and predicates, in that they constitute an independent domain that does not affect nor is affected by an element outside it. Therefore, even though they are apparently present in a

A Case for Head Movement at PF: SAI in Comparatives

445

clausal structure at a point of derivation when sentence stress assignment is made at PF, adjuncts are transparent for the availability of head movement. We account for the grammaticality of (42) along this line of ideas. The adjunct in the south is present in (42), which appears to prevent the subject NP of the comparative clause from being put in final position after SAI applies. However, its mere presence does not interfere with assigning unmarked sentence stress to the subject NP.

6. Summary To summarize, we have shown that there is an interaction among VP-ellipsis, sentence stress assignment rule and head movement in English comparatives. In order to account for this, we have proposed that head movement in this construction is best described as an instance of PF movement, which is different from canonical head movement in yes/no or wh questions. On this view, it follows that head movement in yes/no or wh questions does not affect the focus structure of a resulting sentence. Consequently, the head movement in this construction is not optional, but obligatory, since it is subject to economy at PF. To an extent that our analysis fares well, it renders partial evidence to Chomsky’s (2000, 2001a,b) hypothesis of PF head movement.

Acknowledgements We would like to thank Klaus Abels, Îeljko Bo‰koviç, Barbara Citko, Howard Lasnik, and Bum-Sik Park for suggestions, comments and criticism of our paper. We are also indebted to the audience at the ling-lunch talk at University of Connecticut, at the workshop Triggers at Tilburg University, and at the WECOL at University of British Columbia. We are also indebted to the anonymous reviewers for the valuable suggestions and criticisms. All remaining errors are our responsibility, of course.

446

Fumikazu Niinuma and Myung-Kwan Park

Notes 1. The extension condition basically means that the movement operation extends its target. See Chomsky (1995) for the detailed discussion on this matter. 2. The definitions of PF-head-government and PF-antecedent-government are shown below: (i) _ PF-head-governs ` iff i. a) _ is a head, and b) _ c-commands `, and c) _ respects Relativized Minimality wrt `, and ii. _ is PF-active. (ii) A link _i in a chain is PF-active iff _i is the link at which lexical insertion occurs. (iii) _ PF-antecedent-governs ` iff i. a) _ and ` are co-indexed, and b) _ c-commands `, and c) _ respects Relativized Minimality wrt `, and ii. _ is PF-visible. (iv) An expression _ is PF-visible iff _ has phonetic exponence. 3. There are two ways in which Merchant’s (2003) idea can be implemented. One way is that ellipsis not only instructs PF to forego phonetic realization, but also completely eradicates potentially ECP-offending categories (such as t1’ in (10b)). The other way is that ellipsis adds a feature which indicates silence to the PF-component (as in Kennedy and Merchant (2000) among others), adding the proviso to the ECP that it only operates on categories which lack such a feature. Although we do not have reasons to believe that one view is superior to the other, the point is that the ellipsis operation saves an ECP violation. We thank anonymous reviewers for this suggestion. 4. As is well-known, the position of unmarked sentence stress is not only determined structurally, but also influenced by information structure, as shown below: (i) Truman DIED (ii) JOHNSON died See Cinque (1993) and Selkirk (1995) for the relevant discussion. 5. An anonymous reviewer points out the following sentence in (i), noting that it might constitute counterevidence for our analysis because the subject can also be stressed in its base position without SAI in comparatives. (i) Abby can play more instruments than her FATHER can (ii) Abby can play more instruments than can her FATHER However, there is a way of solving this problem. Note that it has been assumed in the literature that there are two types of stress assignment rules, namely default/neutral stress and non-default/contrastive stress. Assuming these two

A Case for Head Movement at PF: SAI in Comparatives

6.

7. 8.

9.

10.

11.

12. 13.

447

types of stress assignment rules, we argue that the subject in the comparative clause of (i) receives stress that is assigned by non-default stress rule (contrastive stress). On the other hand, stress is assigned to the subject in the comparative clause of (ii) by default stress (neutral stress). Note that the account for SAI in terms of economy and focus was not at conflict with the PF-ECP in section 2. However, PF head movement is incompatible with the definition of the ECP at PF. This is so because heads can derivationally PF head-govern all of their chain links. Thus, t1’ in (9b), for example, can be PF head-governed by the trace (tcan). See Fox (1995) for economy at LF. Another way of capturing the multiple head movement in (28) is that each head moves separately, adjoining one to the other. While the question remains open whether one analysis is superior to the other, the main point here is that both heads move to C at PF. The interesting question that arises is when movement of ‘more than one’ head is possible in English. Chris Collins and Marcel den Dikken (personal communication) point out to us that ‘more than one’ heads can occur before the subject of the locative inversion or quotative inversion construction, as shown below: (iii) Down the hill had rolled John (iv) “I am so happy” has thought John It seems that the same pronoun restriction arises in this construction. That is, weak pronouns are not allowed at the end of this construction: (v) Down the hill had rolled HE/*he (vi) “I am so happy” have thought HE/*he The difference between head movement in the comparative construction and that in the locative inversion/quotative construction is that in the former, VPellipsis is a necessary condition for head movement, while it does not apply to the latter. We will leave this open for a future research topic. We argue that C in the comparative clauses does not have any strong feature to check off in narrow syntax, since the trigger for SAI in comparative clauses is possible only when the subject in the comparative clauses receives neutral stress. This is another difference between SAI in comparative clauses and SAI in yes/no questions where the Infl is considered to move to C in order to check off the strong features of C (cf. Lasnik 1995). We thank anonymous reviewers for pointing out these sentences. They remind us of Emonds (1970), who suggests that there might be the case that SAI in yes/ no questions is different from SAI in comparatives. According to him, SAI in yes/no questions is a root transformation, but SAI in comparatives is a ‘minor movement rule.’ We are indebted to an anonymous reviewer for this suggestion. Subcomparative constructions are ones where only a subpart of the second compared constituent is non-overt. In (i), for example, the number of books owned

448

Fumikazu Niinuma and Myung-Kwan Park

by John is compared to the number of magazines (not books as in (ii)) owned by Bill. (i) John has more books than Bill has e magazines (ii) John has more books than Bill has e Sentences like those in (i) have been named subcomparatives. See Corver (1992) for the relevant discussion. 14. If our analysis were on the right track, we would expect the following sentence is acceptable, contrary to the fact: (i) *More boys are interested in this book than are girls However, there is an intervening factor for the sentence (i). As shown by Kennedy and Merchant (2000), the attributive comparative deletion in subject position is unacceptable, irrespective of the different word orders possible in comparative clauses: (ii) *Better short stories were published this year than novels were (iii) *Fatter boys were born in this hospital than girls were (iv) *A longer table was ordered than a desk was See Kennedy and Merchant (2000: 135–136) for the relevant discussion on this matter. 15. Selkirk (1995) argues that the theory of focus projection consists of a set of principles for the licensing of F-marking, as in (i) and (ii): (i) Basic Focus Rule: An accented word is F-marked. (ii) Focus Projection (a) F-marking of the head of a phrase licenses the F-marking of the phrase (b) F-marking of an internal argument of a head licenses the F-marking of the head.

References Boeckx, Cedric and Sandra Stjepanoviç 2001 Head-ing toward PF. Linguistic Inquiry 32: 345–355. Chomsky, Noam 2000 Minimalist Inquiries: the framework. In Step by Step: Essays on Minimalist Syntax. In Honor of Howard Lasnik, Roger Martin, David Michaels, and Juan (eds.), 89–155. Cambridge, Mass.: MIT Press. 2001a Derivation by phase. In Ken Hale: A Life in Language, Michael Kenstowicz (ed.), 1–52. Cambridge, Mass.: MIT Press. 2001b Beyond explanatory adequacy. Ms., MIT. Cinque, Guliemo 1993 A null theory of phrases and compound stress. Linguistic Inquiry 24.2: 239–298.

A Case for Head Movement at PF: SAI in Comparatives

449

Corver, Norbert 1993 A note on subcomparatives. Linguistic Inquiry 24: 773–81. Doherty, Paul and Arthur Schwartz 1967 The syntax of the compared adjective in English. Language 45, 903–936. Emonds, Joseph 1970 Root and Structure Preserving Transformations. Doctoral diss., MIT. Fox, Danny 1995 Economy and scope. Natural Language Semantics 3: 283–341. Gussenhoven, Carlos 1983 Focus, mode and the nucleus. Journal of Linguistics 19: 377–417 1992 Sentence accents and argument structure. In Thematic Structure: Its Role in Grammar, Iggy M. Roca (ed.). Berlin/New York: Foris. Halle, Morris and Jean-Roger Vergnaud 1987 An essay on stress. Cambridge, Mass.: MIT Press. Jayaseelan, K. M. 1990 Incomplete VP Deletion and Gapping. Linguistic Analysis 20: 64–81. Johnson, Kyle 2001 What VP-ellipsis can do, and what it can’t, but not why. In The Handbook of Contemporary Syntactic Theory, Mark Baltin and Chris Collins (eds.), 439–479. Blackwell. Kennedy, Christopher and Jason Merchant 2000 Attributive comparative deletion. Natural Language and Linguistic Theory 18: 89–146 Lasnik, Howard 1995 A note on pseudogapping. MITWPL 27: 143–163. 1999 On feature strength: three minimalist approaches to overt movement. Linguistic Inquiry 30: 197–217. Lechner, Winfried 2001 Reduced and phrasal comparatives. Natural Language and Linguistic Theory 19,4: 683–735. Levin, Nancy 1986 Main-verb ellipsis in spoken English. New York: Garland. Merchant, Jason 2001 The syntax of silence. Oxford University Press. 2003 Subject-Auxiliary Inversion in comparatives and PF output constraints. In The Interface: Deriving and Interpreting (omitted) Structures, Kerstin Schwabe and Susanne Winkler (eds.). Amsterdam: John Benjamins. Pinkham, Jessie E. 1982 The formation of comparative clauses in English and French. New York: Garland. Potts, Christopher 2002 The syntax and semantics of as-parentheticals. Natural Language and Linguistic Theory 20: 623–689.

450

Fumikazu Niinuma and Myung-Kwan Park

Reinhart, Tanya 1997 Interface economy: focus and markedness. In The Role of Economy Principles in Linguistic Theory, Chris Wilder, Hans Martin Gartner, and Manfred Bierwisch (eds.), 146–69. Berlin, Germany: Akademie. Reinhart, Tanya, and Ad Neeleman 1998 Scrambling and the PF interface. In The Projection of Arguments: Lexical and Compositional Factors, Butt Miriam and Geuder Wilhelm (eds.), 309–353. Stanford, CA: CSLI. Selkirk, Elizabeth 1986 On derived domains in sentence phonology. Phonology Yearbook 3: 371–405. 1995 Sentence prosody: Intonation, stress and phrasing. In The Handbook of Phonological Theory, John Goldsmith (ed.), 550–569. Oxford: Blackwell. Zwart, Jan-Wouter C. 2001 Syntactic and Phonological Verb Movement. Syntax 4,1: 34–62.

The EPP in Breton: An Unvalued Categorial Feature Milan Rezac

1. Introduction: A third A-feature This paper investigates the syntax of the unique pre-verbal position of Breton V2 clauses, which will be identified as [Spec, TP], and the morphology of T (glossed R), which will be shown to code the syntactic category of [Spec, TP]. The focus will be on [Spec, TP] as an A-position, exemplified in (1): 1 (1)

a. Bez’ e oa EXPL R was Azenor was seen. b. Azenor a oa c. Gwelet e oa

gwelet seen

Azenor. Azenor

gwelet. Azenor.

We will see that when such [Spec, TP] is filled by movement, the feature responsible cannot be Case or  -features, which are usually associable with A-movement. The identity of the feature can be established from a convergence of two properties. First, it is restricted to the closest syntactic object in the c-command domain of T, regardless of its properties: Case,  -features, syntactic category, phonological content, or even dislocatability. Since locality follows from limitation of syntactic dependencies to the closest matching pair of features (Rizzi 1990, Chomsky 1995, 2000), the feature responsible for the A-movement must be matchable by all syntactic objects. Second, the Breton T contains a morpheme which codes the syntactic category of [Spec, TP]. This morpheme is the morphological reflex of a syntactic dependency, and results from the Agree operation (Chomsky 2000); in this case, it values an unvalued categorial feature with such values as [N] or [A]. Since a category is a property possessed by all syntactic objects, an unvalued categorial feature has the required locality property as well. This feature will be noted [b-].

452

Milan Rezac

We will find further that in Breton there is evidence for a one-to-one correlation between [b-] Agree and the projection of [Spec, TP]. [b-] Agree is responsible for filling [Spec, TP] as an A-position with the closest syntactic object. If T has an A -feature as well, this may fill [Spec, TP] as an A -position with the characteristic selective locality of A -movement. However, in that case [b-] must be valued at the same time, as a free rider. This correlation between [Spec, TP] projection and [b-] Agree characterizes the EPP property of Chomsky (2000), which determines whether a head offers a non-thematic Merge position. However, unlike the EPP proposal, [b-] as an unvalued categorial feature can itself Agree. The organization of this paper is as follows. The rest of this introduction reviews the role of categorial features in implementing the EPP in the Minimalist Program. Section 2 introduces those aspects of Breton syntax which relate to [Spec, TP], and establishes the clausal architecture of the complement of T. This lays the groundwork for section 3, which shows that there is a class of movements to [Spec, TP] that are not driven by either A -features or Case/  features. This option is limited to moving the closest syntactic object in the c-command domain of T, which may be the subject, a predicate adjective, and various verb forms. Section 4 shows that T in Breton bears a morphological marker tracking the syntactic category of [Spec, TP]. The rest of section 4 and section 5 show that [b-] correlates with the projection of [Spec, TP], and explore its identification with the EPP feature of Chomsky (2000). As an uninterpretable categorial feature, [b-] is an extension of the specific categorial features of Chomsky (1995), particularly the nominal [N-] and [D-] features. These features regulate the licensing of non-thematic DP positions in the clause, and are distinct from Case and  -agreement features. The most comprehensive argument for the need for a third A-feature beyond Case and  -agreement is due to Schütze (1993, 1997), particularly his investigation of the properties of [Spec, TP] in Icelandic. I will begin by reviewing the argument. In Icelandic, the [Spec, TP] A-position is not reserved for arguments with structural Case. It can be occupied by an oblique DP provided it is the highest argument it the clause (Zaenen et al. 1985, Sigur›sson 1992). In (2) henni is in this position, although it has dative Case and cannot agree with the verb: (2)

t1 bró›ir sinn lei›inlegur. Henni1 flykir she.D seem.3.SG brother.N hers.N boring.N Her brother seems boring to her. (Zaenen et al. 1985: 450)

The EPP in Breton: An Unvalued Categorial Feature

453

This provides the initial motivation for divorcing Case/  from A-movement to [Spec, TP]. By itself, this fact does not go very far, since it is possible that oblique DPs which appear in [Spec, TP] have structural Case as well as theta-related oblique Case, as proposed in Belletti and Rizzi (1988); the nominative could then be licensed by long distance  -Agree as in Chomsky (2000).2 However, Schütze’s (1993) point, elaborated in Schütze (1997: II, IV), is that both such ‘quirky’ Case DPs and regular nominative DPs in Icelandic have their Case licensed under no relation to either [Spec, TP] or T. This is illustrated in (3), where the  -agreement normally required between T and a nominative as in (3)a is impossible when an unmoved DP (underlined) intervenes between them (Chomsky 2000: 130–1, Holmberg and Hróarsdóttir 2003). Yet the nominative is perfectly fine here, although it is not in [Spec, TP] and cannot enter into  -agreement with T.3 (3)

a. Mér1 vir›ast/??vir›ist t1 [hestarnir hafa veri› me.D seem.PL/seem.SG horses.N have.INF been gefnir konunginum] given king.the.D The horses seem to me to have been given to the king. (Schütze 1997: 107) b. fia› vir›ist/*vir›ast einhverjum manni [hestarnir there seem.SG/*seem.PL some.D man.D horses.the.N vera seinir] be.INF slow A man finds the horses slow. (Holmberg and Hróarsdóttir 2003: 1000) fleir] c. Mér1 fannst/*fundust t1 [TP henni lei›ast me.D seemed.3.SG/*3.PL she.D bore.INF they I thought she was bored with them. (Taraldsen 1995: 317)

Just as nominative does not require a relation to T or [Spec, TP] to be licensed, [Spec, TP] need not be occupied by a constituent with structural Case. Setting aside expletives and quirky Case DPs whose Case is in question, Holmberg’s (2000) work on Stylistic Fronting shows that [Spec, TP] may be filled by an adverb, a PP, a particle, or a participle. Stylistic Fronting indeed makes available a stronger conclusion: even if [Spec, TP] can be filled by an agreeing nominative, it does not have to be, as the co-existence of the options in (4) indicates.4 These facts lead Holmberg (2000) to conclude that there is a non-Case/ -related P(honological) feature which can drive movement to [Spec, TP] in the syntax (cf. section 5 here).

454

(4)

Milan Rezac

a. [Erfi› ákvör›un]1 hefur veri› difficult-N decision-N has been A difficult decision has been taken. b. Tekin1 hefur veri› t1 [erfi› taken has been difficult-N A difficult decision has been taken.

tekin t1 taken (Holmberg 2000: 448)

ákvör›un]. decision-N (Holmberg 2000: 446)

Chomsky (2000) continues to maintain the separation of [Spec, TP] licensing and Case. However, a separate nominal feature is replaced by an EPP property of heads, which requires a head to project an non-thematic position but is not itself a feature capable of forming syntactic dependencies via Agree. For [Spec, TP] in Icelandic, it is  -Agree which identifies the goal for movement. As the paradigm in (3) has already shown,  -Agree in contrast to nominative assignment does require a syntactic dependency between T and the goal. Furthermore, there is evidence that oblique DPs that move to [Spec, TP] do enter into a relation with the  -features of T, because they disable person (not number) agreement with a nominative, as (5) shows (Chomsky 2000: 128, Boeckx 2000, Anagnostopoulou 2003: V, Béjar and Rezac 2003): (5)

Henni1 *lei›umst/?*lei›ust/??lei›ist t1 vi› she.D bored.1.PL/3.PL/3.SG we.N She was bored with us. (Taraldsen 1995: 308–9)

Chomsky (2000) therefore concludes that DP A-movement to [Spec, TP] is a consequence of the  -Agree of T, which moves the first matching goal if the EPP of T has not yet been satisfied. However, this approach also seems insufficient when faced with the data in (4). The Breton A-movement investigated here shows similarities both to oblique subjects and to Stylistic Fronting in Icelandic in terms of locality and separation from Case. However, as we will see in section 3, in Breton it can be demonstrated that  -features are not involved because of the Complementarity Principle of Celtic  -agreement systems, which suspends  agreement with overt DPs. This leaves us with a search for the relevant licensing property, and again the Breton data provide a unique perspective; there is a morphological spell-out of the relevant Agree relation, which identifies the third A feature as an unvalued categorial feature.

The EPP in Breton: An Unvalued Categorial Feature

455

2. The Breton Clausal Architecture Breton is a pro-drop language with a basic VSO order, which surfaces in embedded contexts, combined with a V2 requirement in root declarative non-negative clauses: see Kervella (1995: 389f.), Stephens (1982), Stump (1984, 1989), Borsley and Stephens (1989), Schafer (1995), Favereau (1997: 288f.), Hewitt (2002).5 In the following example, a V2 matrix clause takes a verb-initial complement. (6)

Nolwenn a gred e vo tomm an amzer. Nolwenn R believe.3.SG R will.be hot the weather Nolwenn believes that the weather will be hot.

Our focus is the pre-verbal position in V2 clauses. It is a unique position; any constituent that precedes it is clause-external, such as left-dislocated DPs.6 Constituents that fill it may have either A or A properties. The theory of unique specifier positions that can be of either A or A type is developed in Diesing (1990) for Yiddish and by Zubizarreta (1998: 99f., 117f.) for Spanish. Adapting the terminology of Zubizarreta, we will call a head whose specifier can be either A or A a syncretic category. The A/A type of the specifier is determined by the feature which causes the specifier to be filled (cf. Chomsky 2000: 108–110 and accompanying notes 45, 53 on p. 144–5): an A-position for a Case/ /nominal feature, an A -position for one of the perhaps structured class of features such as [wh] (cf. Rizzi 2000, forthcoming). The concept of a syncretic category naturally captures the complementarity of A and A constituents. We may see the complementarity in the following paradigm, where the preverbal position is variously filled by a fronted participle, a subject, and wh object. As we will see below, a fronted participle cannot have an A properties, a fronted subject can but need not, and a fronted wh object must have them. Yet these elements cannot co-occur (Stephens 1982: 210; cf. Schafer 1995: 141–2, 154 note 10): 7 (7)

a. Debret he deus Nolwenn eaten R.3.SG.F have Nolwenn Nolwenn has eaten pancakes.

krampouezh. pancakes

b. Nolwenn he Nolwenn/NOLWENN

krampouezh. pancakes.

deus debret has eaten

c. Petra he deus what R.3.SG.F have What has Nolwenn eaten?

debret eaten

Nolwenn? Nolwenn

456

Milan Rezac

d. *Debret Nolwenn he deus krampouezh. e. *Petra Nolwenn he deus debret. f. *Debret petra he deus Nolwenn? Syncretic categories can be construed as an implementation of Rizzi’s (1997, forthcoming) multiple left-periphery heads T°, Fin°, Foc°, Int°, and Topic° in languages like Breton, where these heads are “compressed” in the sense that only a unique head (at the V2 position) and a unique specifier (the preverbal constituent, which can be a neutral, focused, topicalized, or interrogative XP) may be realized for the whole complex. I set aside here the way this compression should be implemented: it could be, for example, that some languages simply lexicalize [T°+Fin°+Foc°+Int°+Topic°]°, or it could be that there is a syntactic mechanism responsible for such a compression of independently projected syntactic heads. I will label the pre-verbal position as [Spec, TP], the specifier of the syncretic head T which hosts the finite verb (cf. Hewitt 2002). V2 clauses project [Spec, TP], while V-initial TP complements of certain complementizers (including negation) do not; this and the complementizer system will be discussed in in section 4. Most constituents fronted to [Spec, TP] cannot receive a neutral interpretation suitable to out-of-the-blue contexts or responses to what happened questions, unlike the participle in (7)a and subject in (7)b which can. Instead, they receive a variety of marked interpretations: they are interpreted as new information focus which corresponds to the wh-word of a question, as contrastive focus, or as discourse (switch) topic; or they are wh-words, the highest of which must appear in [Spec, TP] (Hendrick 1990: 154). These all involve A -feature Agree (Rizzi 1997, forthcoming). To such non-neutral readings are limited non-subject DPs, PP arguments and adjuncts, low adverbs, and vPs (ex. (7c), (8a), (8b)). All constituents which undergo such A -movement can undergo it both locally within their clause and over unbounded distance, with sensitivity to strong and weak islands and to superiority (Stephens 1982, Borsley, River, and Stephens 1996: 73 note 2, Hendrick 1990: 154). Examples of long distance extraction are vP-fronting in (8c) and subject extraction in (8d), where the latter contrasts with the local movement in (7b) in lacking a neutral reading (Jouitteau 2003b). The basic limitation is that certain non-branch sub-trees cannot displace, unlike vPs: finite and non-finite clausal complements (see note 28; cf. Stephens 1982: 40), and the constituent headed by a past participle, regardless of whether or where it contains the subject (ex. (8e), (8f)).

The EPP in Breton: An Unvalued Categorial Feature

(8)

457

a. Krampouezh1 / goustadik1 / er gêr1 a zebran t1. pancakes slowly at home R eat.I I eat PANCAKES/SLOWLY/AT HOME. b. [Debriñ krampouezh]1 a ran t1. eat pancakes R do.I I EAT PANCAKES. e rae ar vasonerien t1 c. [Sevel ar mogerioù]1 a ouien build the walls R knew.1.SG R did the masons I knew that the masons built the walls. (Stephens 1982: 99) d. Anna am eus klevet he deus desket t1 he c’hentelioù Anna R.1.SG have heard R.3.SG.F have learned her lessons ANNA, I have heard she has learned her lessons. *I have heard that Anna has learned her lessons. (Jouitteau 2003b) e. *[E vo tomm an amzer] a gred Nolwenn t1 R will.be hot the weather R believes Nolwenn f. *[(Manon) debret (Manon) krampouezh]1 he deus (Manon) t1. eaten pancakes pancakes R.3.SG.F have

vP fronting provides a good starting point for examination of TP-internal clausal architecture. The verb contained in the fronted vP is always in infinitival form. We will assume that the infinitival morphology is assumed by the verbal root if it cannot raise out of the vP to receive either participial or finite morphology. This morphology is in turn stranded by vP-fronting, and rescued by insertion of the verb ober ‘do’, as in (8b). We see that the verb under vP-fronting does not raise to the head that contains the participial affix -et, because the constituent headed by the past participle cannot front (8f). This head must be fairly high, since the participle in it obligatorily precedes temporal adverbs such as c’hoazh ‘yet, still’, biskoazh ‘(n)ever’, and dalc’hmat ‘continuously’ (Favereau 1997: 326 –7, contrasting French), as in ex. (9). The head in question will be labeled Agr(Prt), although the participial affix -et is  -invariant. AgrPrt must ccommand the site of adverbs like c’hoazh ‘yet, still’, which are left and right vP-adjoined.8 (9)

N’eus bet biskoaz ger etrezomp NEG.has been ever word between.1.PL There has never been a word between us. (Favereau 1997: 327)

458

Milan Rezac

We will call the highest DP argument the subject, whether it is the external argument of transitives and unergatives or the internal argument of unaccusatives and passives. We find that the subject and the participle both necessarily precede the transitive object (cf. Stephens 1982, Hendrick 2000), while either order is possible between the subject and the participle. 9, 10 (10) a. Dec’h en devoa (ar merour) gwerzhet yesterday R.3.SG.M had the farmer sold (ar merour) leue e vuoc’h ruz. the farmer calf his cow red Yesterday the farmer had sold the calf of his red cow. (Kervella 1995: 373) b. Ne oa ket (he ger) NEG was not her word She had not finished speaking.

peurlavaret (he ger). completely.spoken her word (cf. Favereau 1997: 326)

c. Dec’h e oa (Yann) aet (Yann) yesterday R was Yann gone Yann Yann had gone to Brest yesterday.

da Vrest. to Brest

This asymmetry between subject and transitive object is supported by considering the vP-fronted constituent. It obligatorily includes the transitive object DP and all subcategorized internal argument PPs (Stephens 1982: V). On Chomsky’s (2000, 2001) assumptions, it must be a phase, while the unfrontable constituents of (8f) and (8e) are non-phases. A phase contains at least the entire argument structure, so that for transitives the vP but not the VP is a phase. vP-fronting therefore affects the vP. 11 However, vP-fronting cannot include the subject of either transitive/unergative or unaccusative/ passive constructions, which is left behind (8c). Therefore, the highest DP argument must obligatorily move out of the vP to some higher position, while the transitive object must stay within the vP. This can be summarized as follows, where c’hoazh-type adverbs left/right-adjoin to vP: (11) Transitive: Unaccusative:

[T (subject1) participle (subject1) [vP t1 … object]] [T (subject1) participle (subject1) [vP … t1]]

These schemata accord well with conclusions based on cross-linguistic investigation of derived A-positions for DPs in McCloskey (1997). The neutral [Spec, TP] where we see the subject in (7b) above and which we will investigate in the next section is a high A-position. The post-T position of subjects

The EPP in Breton: An Unvalued Categorial Feature

459

in (7a) is a low derived A-position, to which the highest DP obligatorily moves out of the vP. The two correspond to [Spec, AgrSP] and [Spec, TP] respectively in Bobaljik and Jonas’s (1996) investigation of Icelandic. Like Icelandic, Breton allows transitive expletive constructions as in (12a), which Bobaljik and Jonas connect to the availability of two distinct derived Apositions for subjects: the expletive occupies the high A-position, and the subject is in the low derived A-position. Furthermore, also as in Icelandic (Jonas 1996: 79), Hendrick (1990: 157) observes that the low derived Aposition is available for raising (12b) (along with [Spec, TP], ex. (15) below). (12) a. Bez e nevo hennez traou EXPL R.3.SG.M have he things He will have goods.

(Gros 1984: 110)

b. N’ en deus ket Yann1 seblantet [t1 karout ar vugale]. NEG R.3.SG.M have NEG Yann seemed love the children Yann didn’t seem to love the children. (Hendrick 1990: 157) We can identify the low A-position as [Spec, AgrPrtP] of the AgrPrt functional head which is needed independently for participial morphology. The obligatory movement of the subject here is possibly driven by Case/ -agreement of AgrPrt (cf. McCloskey 1996 for Irish, from which Breton differs principally in offering [Spec, TP]). Although this differs from the usual identification of the low derived A-position in Icelandic as member of the expanded INFL, it is supported by the fact that the Breton participle in AgrPrt precedes temporal adverbs, which in Icelandic occur between INFL and the participle (cf. Jonas 1996: 35f.). A further difference between the two languages is that there are no semantic restriction on the specificity of the DP in either the high or low A-position in Breton. The transitive object in Breton stays in the vP, with no object shift available to it, and vP-fronting obligatorily pied-pipes it. Finally, to account for the existence of participle > subject orders, we need to posit a higher functional head F between T and vP, to which AgrPrt optionally raises. We thus reach the following structure, exemplified for a transitive verb:

A

460

Milan Rezac

Figure 1. Breton extended VP architecture TP

3 3 FP T 3 AgrPrtP F 3 EA 3 vP AgrPrt 3 vP -et/’ c’hoazh 2 tEA 2VP v 2 V

IA

We will assume that each of these heads is present if and only if it has content. A functional head has content either if it is lexicalized by an affix or by head movement, or if its specifier is filled by virtue of its feature content (Speas 1995, Grimshaw 1997). AgrPrt is thus always present, since [Spec, AgrPrt] is always projected for subject movement. F is present only if there is a feature which attracts the verb to it, so that it is absent in subject > participle orders. We assume further that v is always present as the locus of voice, but nothing hinges on this fact except for the definition of a phase for vP-fronting, which does not distinguish transitives from unaccusatives and passives (cf. note 11). The important aspects which the structure captures are the following: (13) Breton clausal architecture (i) There is a constituent affected by vP-fronting that necessarily excludes the subject and includes all subcategorized internal arguments. A verb stranded within this constituent has not raised up to the participial affix -et and takes infinitival form. (ii) There is a derived A-position for subjects between the vP and T. (iii) The head that contains the participial affix -et c-commands all possible attachment sites for temporal adverbs like c’hoazh ‘still’. (iv) There is interaction between the derived subject position(s) and the head raising position(s) for the participle, so that both orders are possible.

The EPP in Breton: An Unvalued Categorial Feature

461

The structure in Figure 1 lies within the space of possible solutions to these requirements, and chooses between various option by following the general trend of associating an obligatory A-movement position with an Agr head and accounting for variation in participle height by V-raising.12 Some probable elaborations have already been mentioned, e.g. note 11. However, what is truly important for the locality facts discussed in the next section are the properties in (13). They will be crucial in accounting for the limitations on A-movements to [Spec, TP], to which we now turn.

3. b-driven Movement in Breton In contrast to filling [Spec, TP] by local or long-distance A -movement, there are three possibilities of filling [Spec, TP] which are interpretively neutral, local, and do not satisfy an A -criterion. These are: by an expletive, by the highest DP (the subject), and by so-called Long Head Movement; all are already exemplified as (12), (7a), and (7b) above. We will approach them in this order. The expletive will demonstrate the existence of a [Spec, TP] without A -properties, subject movement will show that there is a movement to [Spec, TP] which cannot be driven by either A or Case/ -features, and Long Head Movement will show that this movement affects the closest syntactic object in the c-command domain of T. This is a consequence of the unselective locality of the feature that implements the movement, [b-], which looks for a property common to all syntactic objects: their category. The first option for a non-A [Spec, TP] is an expletive. The expletive, taking such forms as bez’ and bout, originates as an infinitive of the verb bezañ/bout ‘be’, but it is not synchronically identical with it (Kervella 1995: 185, 390–1, Gros 1984: 109–111, Favereau 1997: 217, 230, 295). It is limited to occurrence in [Spec, TP] of finite V2 clauses.13 Adopting Chomsky’s (1995, 2000) analysis of expletives, it is base-generated directly in [Spec, TP], and satisfies the EPP requirement of T which I will argue is due to [b-]. (14) Bez’ ech evas kalz gwin. EXPL R drank lot wine He did drink a lot of wine. The second non-A option for [Spec, TP] is movement of the subject. This singles out the same class of DPs that obligatorily move out of the vP to [Spec, AgrPrtP]. The A/A contrast between subjects and other arguments in [Spec, TP] can be seen in two ways. First, SV(O) orders as in (7b) can have

462

Milan Rezac

out-of-the-blue, wide focus interpretation, which is not the case for any other preverbal argument (Stump 1984: 300, Gros 1984: 108, Timm 1991, Favereau 1997: 297). Second, the subject in [Spec, TP] can be an idiom chunk as in (15), which are not compatible with the interpretive load of A positions.14 Following Stump (1984), we conclude that the preverbal subject always has the option of being in an A-position.15 (15) Ar bik1 a zeuas da t1 gregiñ en e skouarn. the magpie R came to bite.INF in his ear He decided to get married. (Borsley & Stephens 1989: 417) Although the movement is not A -movement, in Breton we are in a unique position to show that it cannot be attributed to either Case or  -features. First, the movement is certainly not Case-driven. As we have seen in section 2, subjects in Breton are licensed in the low [Spec, AgrPrtP] A-position, regardless of their semantic properties: cf. (9), (10), (12). We could take this to be their Case position. More interestingly, movement of the subject to [Spec, TP] can also be shown not to be driven by  -features, which as we saw in section 1 is an issue for Icelandic. Breton is a pro-drop language, but like the other Celtic languages it is subject to the Complementarity Principle, which enforces a complementary distribution between overt DPs and  -agreement morphology on T (Stump 1984):16 (16) a. Levrioù a lennont/lennan. books R read.3.PL/1.SG I/they read books. b. Ar vugale/me a lenn/*ont/*an levrioù the children/I R read.3.SG/*3.PL/*1.PL books The children/I read books. c. Levrioù a lenn/*ont ar vugale books R read.3.SG/*3.PL the children The children are reading BOOKS. There are three types of approaches to the Complementarity Principle in Celtic. The first is the Incorporation Hypothesis, which takes  -agreement on the verb as the spell-out of an independent pronominal argument. This pronoun incorporates into the finite verb either via syntactic movement (Anderson 1982, Stump 1984 for Breton) or through a post-syntactic (e.g. prosodic) process (Pranka 1983, Doron 1988, Ackema and Neeleman 2003

The EPP in Breton: An Unvalued Categorial Feature

463

for Irish, Adger 2000 for Scottish Gaelic). In response, Stump (1984, 1989) for Breton and McCloskey and Hale (1984: 525f.) for Irish pursue the Agreement Hypothesis, where inflection reflects  -agreement with pro rather than pro itself. The Complementarity Principle is then handled by means of a condition which excludes the presence of AGR on INFL unless it governs a phonologically empty category. Finally, Jouitteau and Rezac (forthc.), for Breton, argue that there is always  -agreement, but in Complementarity Principle contexts it is with v which is in Breton a nominal category (Jouitteau 2003a). As a nominal category, v has 3rd.sg.  -features available for  -Agree, and because it moves to T by head movement it will be the closest goal for  -Agree unless the subject also incorporates into T; following the Incorporation Hypothesis, this is an option available only to pro and not to overt DPs. While the three approaches differ significantly, they agree on the following point: there is no  -agreement with overt DPs. This happens because a DP cannot incorporate to allow  -agreement or because there is a condition that limits  -Agree to empty categories. There is no  relation between T and a DP subject that could drive movement of the latter to [Spec, TP], and the Complementarity Principle reflects this fact. Thus, we are led to the conclusion that the feature which drives DP movement to [Spec, TP] is neither Case nor   feature, and it does not result in A -movement. A first approximation of its identity can be reached by examining its locality properties, under the feature-relativized locality adopted in Chomsky (1995, 2000). A-movement to [Spec, TP] is limited to the highest argument, which is always a DP. The transitive object, and PP internal arguments, can only move to [Spec, TP] by A -movement. We thus need a feature for which the closest DP is a match: (17) ___ T subject DP [X-] -->

internal argument DP/PP

The nominal [D-] and [N-] features of Schütze (1993) and Chomsky (1995) would show appropriate behavior. We will modify this conclusion below in favour of an uninterpretable categorial feature [b-], which can be valued by a DP to [b=D]. However, we reach our first important result: there is a nonA , non-Case/  feature that drives movement to [Spec, TP] in Breton, and that feature selects the closest DP in preference to all other arguments. Before proceeding, we note that pro subjects behave in the same way as overt DPs: although pro itself cannot move to [Spec, TP] to give neutral Vinitial orders, it blocks A-movement of internal arguments just like an overt subject does. Pro is thus visible but not available for movement. We will

464

Milan Rezac

return to this once we have a more complete picture of non-A movements to [Spec, TP]. The third non-A way of filling [Spec, TP] is by movement of a non-finite form of the verb over an auxiliary in T, so-called Long Head Movement (LHM): see Stephens (1982), Borsley, Rivero and Stephens (1996), Schafer (1997), Hewitt (2002). LHM takes two forms in Breton. In constructions where the verb would be realized as a participle without LHM, that is the perfect and passive where T is filled by the have/be auxiliaries, it is the participle that moves to [Spec, TP]; this is exemplified in (18a) and (18b). In constructions where the verb would normally raise to T and receive inflection for tense and  -features, LHM moves to [Spec, TP] the infinitival form of the verb, which is the default form assumed by a verbal root in the absence of other morphology; the tense and  -features of T are rescued by insertion of ober ‘do’. This is exemplified in (18c) and (18d). Ober-insertion and the assumption of infinitival morphology by the verb under LHM are both last-resort: while in auxiliary constructions all that changes under LHM is the position of the participle, there is no non-LHM construction where tense and  -features are carried by ober which takes an infinitival complement; the equivalent is movement of the verb to T. (18) a. Debret em eus t1 krampouezh. eaten R.1SG have pancakes I have eaten pancakes. b. Lakaet1 e vo t1 al levr-se da vezañ kaset deoc’h. caused R will.be the book-this to be.INF sent to.2.PL The book will be made to be sent to you. (Kervella 1995: 173) c. Kouezhañ a ra/reas fall.INF R do.3.SG/did.3.SG It rains/rained.

glav. rain

d. Lakaat a rin t1 kas deoc’h al levr-se. cause R will.do.1.SG send to.2.PL the book-this I will cause this book to be sent to you. (Kervella 1995: 173) Although LHM and vP-fronting sometimes have similar results, they are completely different constructions. LHM places into [Spec, TP] a head alone, the participle or infinitive, while vP-fronting places there the vP with the verb limited to the infinitive form and all internal arguments present. LHM unlike vP-fronting cannot undergo long distance A -extraction, as (19) contrasting with (8c) shows (Stephens 1982: III). Interpretively, LHM con-

The EPP in Breton: An Unvalued Categorial Feature

465

structions lack any emphasis on the fronted constituent and the whole clause receives wide focus. These two properties are the same as those characterizing subject A-movement to [Spec, TP]. (19) *Debriñ [a eat.INF R

ouien [e rae t1 Yann krampouezh ed-du]] knew did Yann pancakes buck-wheat

The two forms of LHM, participle + be/have auxiliary and infinitive + ober, have the same properties, but they differ in that the latter does not have a non-LHM analogue, because without LHM the verb moves directly to T. The question arises whether it displaces the verb directly from T, or from within the complement of T as LHM of the participle evidently does. The second alternative accounts for the sole set of exceptions to LHM, in part observed by Stephens (1982: 104). LHM cannot apply to the perfect and passive auxiliaries bezañ/bout ‘be’ and endevout/kavout ‘have’ when they would normally occupy T to give the infinitive + ober construction, (20a). However, it can apply to the participle bet of these auxiliaries when another auxiliary fills T in the perfect passive (20b) and the passé surcomposé. Evidently this is not a prohibition against LHM of the auxiliaries themselves. The proper distinction can instead be drawn if finite auxiliaries are base-generated in T, and LHM can only move material that originates in the complement of T. The class that cannot undergo LHM in fact extends beyond Stephens’s generalization: the copula bezañ/bout ‘be’ (20c) and the main verb endevout/kavout ‘have’ (20d) (as noted by Kervella 1995: 169 –170), not just their homophonous auxiliaries, show the same limitation of LHM to their participles (20e). On the other hand, the lexical verb bezañ/bout ‘exist’ can undergo LHM to give the infinitive + ober construction, as in (20f). This supports the account of the distinction if we assume that the copula is base-generated in T, and that the verb kavout/endevout ‘have’ in Breton is based on the incorporation of a preposition into the copula, as shown in Jouitteau and Rezac (forthc.). Heads base-generated in T cannot undergo LHM; those base-generated lower, such as the auxiliary/copula participle bet or the lexical bezañ/bout ‘exist’, can. (20) a. *Bezañ a be.INF R

ra t1 gwelet Azenor. does seen Azenor

b. Bet eo t1 gwelet Azenor. been is seen Azenor Azenor was seen.

466

Milan Rezac

c. *Bezañ a ra t1 brav be.INF R does fair

an the

amzer weather

d. *Kaout/endevout a ra t1 (debret) krampouezh. have R does eaten pancakes He has pancakes / He has eaten pancakes. e. Bet eo t1 koant been is pretty (S)he has been pretty.

(Favereau 1997: 187)

f. Bezañ/bout a ra be.INF R does She/he/it exists.

(Favereau 1997: 207, 227)

Thus, LHM displaces a head from the complement of T to [Spec, TP], with a last-resort insertion of infinitival and ober morphology. The next step in understanding LHM is to observe that it is restricted by a very strong locality constraint (Rivero 1991, Borsley, Rivero and Stephens 1996). It is limited to the participle construed with the auxiliary in T, if there is one, or to the lexical verb that would otherwise move to T. 17 A more distant head cannot be affected by it: a lower complementizer or the T/V of finite or non-finite complements, for example. Interestingly, we can narrow this locality still further. LHM cannot skip AgrPrt with -et in it, and raise the lexical V(+v) itself, as shown in (21b), where -et could presumably be rescued by oberinsertion as in parallel vP-fronting in (21c). (21) a. Azenor he deus debret Azenor R.3.SG.F have eaten Azenor has eaten pancakes.

krampouezh. pancakes

b. *Debret1 he deus graet Azenor t1 krampouezh. c. [Debriñ krampouezh]1 he deus graet Azenor t1 Azenor has EATEN PANCAKES. These restrictions are summarized in Figure 2: LHM is restricted to a head in the circled space, F or AgrPrt, on one of which the lexical verb finds itself:

The EPP in Breton: An Unvalued Categorial Feature

467

Figure 2. LHM search-space TP

3 3 FP T 3 AgrPrtP F 3 EA 3 vP AgrPrt 3 vP -et/’ c’hoazh 2 tEA 2VP : # v 2 z----------: V IA z--m Let us call the feature of T that implements LHM [H-]. Locality restrictions in the Minimalist Program typically follow from Relativized Minimality applied to features, so the search-space condition on LHM should be reduced to the nature of [H-]. Since we assumed in section 2 that F is only present if V-movement takes place into it, the locality generalization about [H-] is simply that it looks no farther than the closest head in the sister of T, which is F or AgrPrt. When we compare LHM with subject A-movement, we see that their properties are identical: both yield the same neutral interpretation, and both are limited to the closest matching object. This is always be within the same absolute search-space circled in Figure 2. Furthermore, when there is no A -feature on T, one of [H-] and [D/N-] must be present to enforce some movement to [Spec, TP] to satisfy V2, while at the same time [H-] and [D/N-] are in complementary distribution since there is only one non-A movement to [Spec, TP]. The postulation of both an [H-] and [D/N-] features fails to capture these generalizations. An explanatory solution calls for a single feature, provided we can explain how it drives both LHM and subject movement. We will call this feature [b-], and since we want it to attract both subject DPs, verbal roots, participles, and as we will see below predicate APs, it must be looking for some property common to them all. Provisionally, we will assume that this is their syntactic category. [b-] is thus an unvalued categorial feature. 18 Let us now see how positing [b-] accounts for the properties of both movements to [Spec, TP]. These properties are the following:

468

Milan Rezac

(22) Properties of Breton b-driven movement (i) (ii) (iii) (iv) (v)

Complementarity of LHM and subject A-movement. If A -features are not present, one of the two must take place. The possible candidates are found in the circled space in Figure 2. There is free choice between LHM and subject movement. An empty category is never attracted by [b-], which would give apparent V-initial orders in V2 clauses, but it blocks [b-].

Property (i) follows immediately from the fact that a single feature implements both non-A movements. Property (ii) also follows, if we assume that T in V2 clauses always has a [b-] feature. We could at this point qualify this by saying that T has a [b-] feature if and only if it does not have A -features. However, we will see in section 4 that a general principle, Maximize Agree, will automatically delete the [b-] feature if an [A -] feature fills [Spec, TP], so we set this question aside for now and assume with full generality that Breton T always has [b-] if it projects [Spec, TP]. Since the subject in Breton always raises to [Spec, AgrPrtP] while the verb raises at least to AgrPrt, there will always be a matching goal in the circled search-space in Figure 2, explaining property (iii). Turning to property (iv), recall that F is present in our clausal architecture to implement the fact that both subject > participle and participle > subject orders are possible in the complement of T. The first order reflects the participle in AgrPrt, the second in F. For locality of feature attraction, syntactic objects stand in the partial order given by c-command, which reflects containment.19 Applying this strictly, we see that F° c-commands [Spec, AgrPrtP], and [Spec, AgrPrtP] c-commands AgrPrt°. A syntactic object at each of these positions qualifies for b-driven attraction to T and blocks attraction of an object it c-commands. Consequently, we see that if the participle raises to F, it will block attraction of the subject in [Spec, AgrPrtP] by [b-]; and if the participle stays in AgrPrt°, the subject in [Spec, AgrPrtP] will block its attraction. This is the answer to property (iv): the free choice between subject A-movement and LHM is a consequence of the independently attested variation between the position of the verb and the subject in the sister of T. The only extension made here beyond what we see overtly is the assumption that the same variation in the presence vs. absence of F is found when we are dealing with verb roots undergoing V-to-T raising, as well as participles. This, however, is the default hypothesis: F does not stand in a selectional relation to participles specifically (in other words, to AgrPrt filled by -et), since they are compatible with both the presence and absence of F.

The EPP in Breton: An Unvalued Categorial Feature

469

Finally, consider property (v). We wish to prevent [b-] from attracting an empty category to [Spec, TP], giving apparent V1 orders. Two empty categories are in principle at issue: a trace, and a phonologically empty head. We noted above that pro acts as an intervener which blocks neutral OVpro orders, but it is not itself available for subject A-movement to give proVO orders. This is a species of defective intervention effects (Chomsky 2000: 123), which must be caused either by pro, or by its trace if it incorporates into T prior to Agree by [b-]. Similarly, the trace of V-to-T movement in F must not be available for movement by [b-]; its visibility for the [b-] probe as a defective intervener is not determinable, since there will be an intervener (defective or not) in [Spec, AgrPrtP] in any case. Chomsky (2001: 23–4) explores the nature of the empty categories we need, observing that some empty categories must be visible for Agree but unavailable for Move; he attributes this to the lack of phonological content which renders them unable to determine pied-piping for second Merge. To pursue this approach, we would have to assume that it is an intrinsic property of the [b-] feature that it must be valued not only by Agree, but by second Merge of the goal from which it is valued. We will return to this in section 5; at present, the issue is independent of whether we use a [b-] feature or any other kind of feature. Thus, the collapse of LHM and subject movement under a single feature [b-] that seeks the closest syntactic object provides an explanatory account of the properties of non-A movement to [Spec, TP] in Breton. In addition, the unselective locality of the [b-] feature makes an interesting prediction. All constructions we have seen so far have the subject as the highest ‘branch’. There is one systematic exception to this generalization in Breton, which confirms the [b-] approach to non-A movement: copular constructions with predicate adjectives. Constructions where a DP is predicated of an adjective phrase AP are limited to the order AP > DP in Breton prior to any movement to [Spec, TP] as in (24a) (Kervella 1995: 372–6).20 As discussed above, the copula is base-generated in T and cannot undergo LHM. The unselective locality of [b-] makes two predictions here, show in (23): first, the AP should be available for interpretively neutral movement to [Spec, TP], and second, it should block the DP subject from A-movement, leaving to it only A -movement. An approach relying on a nominal feature to produce neutral subjectinitial orders, or for that matter on Case and  -features, makes the opposite predictions as shown in (23b).

470

Milan Rezac

(23) a. ___ T= copula AP b- ------>

DP

b. ___ T= copula AP DP D/N- ---------> The predictions of the [b-] approach are correct. Schafer (1997:195–6) shows that AP-initial orders like (24b) are informationally neutral like LHM, and Favereau (1997: 224, 226) draws exactly the distinction we are predicting between neutral adjective-initial (24b) and marked DP-initial (24c). (24) a. Bez e oa (brav) an amzer (*brav) EXPL R was fair the weather The weather was fair. b. Brav1 e oa t1 an amzer. The weather was fair. c. An amzer1 a oa brav t1. It is THE WEATHER that was fair.

(cf. Gros 1984: 107)

The facts of neutral movement in predicate AP constructions are a strong indication that non-A -movements to [Spec, TP] are driven by a feature whose locality is unselective, [b-]. Before proceeding, it would be well to examine how adjuncts fit into the locality story, particularly given the role that they play in Holmberg’s (2000) approach to Stylistic Fronting in Icelandic which is similar to non-A movement in Breton (see section 5).21 As shown in section 2, AgrPrt to which the participle obligatorily raises is higher than temporal adverbs like c’hoazh ‘still’, bemdez ‘every day’; these in turn can be attached higher than any internal arguments and manner and locative adjuncts like gant aked ‘with care’ or war ar wezenn ‘on the tree’. Locality predicts that the [b-] feature should never be able to attract these adverbs: (25) ___ T (F) ([Spec, AgrPrtP]) AgrPrt c’hoazh [vP …] [b-] ----------------> This is correct. These adverbs in [Spec, TP] necessarily bear a marked reading as in (8a), (26a), which renders such sentences infelicitous as answers to what happened questions (wide focus, out-of-the-blue contexts), but felicitous as answers to how, where, when questions about the adverb (cf. inter alia ar Bihan 2002: V, Jouitteau 2003b). Narrow focus on the adverb is not

The EPP in Breton: An Unvalued Categorial Feature

471

the only possible reading that adverbs in [Spec, TP] have; (26b) illustrates another marked interpretation, parallel to the English translation. However, the neutral reading in all these sentences can only be produced by the subject or LHM filling [Spec, TP], as in Evañ a ra dour gant aked/bemdez/er gêr for (26a). (26) a. Gant aked / bemdez / er gêr e ev dour. with care / always in.the house R drinks water He drinks water CAREFULLY/EVERYDAY/AT HOME. b. Gwechall e laboure ar merhed er parkeier once R worked the women in.the fields Once, women laboured in the fields. (Gros 1984: 122) Jouitteau (2003b), however, observes that a class of high sentential adverbs like espress-kaer ‘voluntarily’ and dre-zegouezh ‘by chance’ behave quite differently: they can receive a neutral reading in the pre-verbal position (as well as a marked one), and they block other non-A movements to [Spec, TP]. Jouitteau (2003b) points out that this is expected if non-A movements to [Spec, TP] obey the same locality conditions as Holmberg’s (2000) Stylistic Fronting in Icelandic, where sentential adverbs block participles from moving (usually, cf. Holmberg 2000: 465 note 21). The Breton data is at this point compatible with either a movement or base-generation analysis of these adverbs in [Spec, TP] on the neutral reading (cf. note 6). What is crucial is their contrast with the adverbs discussed above: sentential adverbs like espres-kaer ‘voluntarily’ are very high in the tree, in contrast to those like bemdez ‘every day’ which are c-commanded by the subject in [Spec, AgrPrtP]. (27) a. Kuzhet he doa hidden R.3.SG.F had

(*espress-kaer) voluntarily

Manon ar c’hazh. Manon the cat

b. Espress-kaer he doa Manon kuzhet ar c’hazh.

(Jouitteau 2003b)

We conclude then that non-A movements to [Spec, TP] in Breton are triggered by a single feature which moves to [Spec, TP] the closest syntactic object in the sister of T. This provides an explanatory account for the properties in (22), and makes the correct predictions about movement in constructions with predicate APs and adjuncts. It now remains to address, if somewhat inconclusively, three aspects of the mechanics of feature-driven movement with respect to LHM itself: its violation of chain uniformity,

472

Milan Rezac

apparent violation of the Head Movement Constraint, and why if F and [Spec, AgrPrtP] can move, can FP and AgrPrtP themselves not move to yield remnant movement. These issues are not particular to the approach in this paper, but do come to the fore since a uniform mechanism is exploited for both DP/AP and LHM. Chain uniformity (Chomsky 1995: 253) bars movement of a head to a specifier position. However, this has now been widely documented; beside LHM, see Holmberg (1999) on verb topicalization to [Spec, CP] in Swedish and Holmberg (2000) on Stylistic Fronting to [Spec, TP] in Icelandic. As Carnie (2000) argues, the X°/XP distinction and syntactic principles referring to it are stipulative under Bare Phrase Structure, where heads and their projections are non-distinct. Both empirical and theoretical considerations then suggest simply not assuming the uniformity condition. The consequences of the X°/XP distinction should be implemented as properties of the interfaces, particularly morphology. Roberts (2003: V) presents such a system, where a cyclic incorporation operation combines two elements in a local relationship if one is an affix, as Bare Phrase Structure objects are built up. The potential risk is unwanted movements. However, these seem to be either generally barred independently, or at least preventable. For example,  /Case features will not produce the equivalent of LHM because the verb and its functional projections do not (normally) have interpretable  -features or need Case. b-driven movement on the other hand, which can certainly affect heads to give LHM, can only do so if the head raises to be the closest object in the sister of T; thus in a language like Breton but without raising to F, the subject in [Spec, AgrPrtP] would always block LHM. Like chain uniformity, the Head Movement Constraint should not have an axiomatic status in the Minimalist Program; following Rizzi (1990), it is an instance of intervener-based locality. Its consequences are thus to be implemented via the identity of the features (e.g. [V-]) which drive head movement (cf. Chomsky 1995: 307; Roberts 2003). The remaining problem is of a different nature. For Breton, in the clausal architecture only the vP and the CP (ForceP in Rizzi’s 1997 system) seem dislocatable. TP, FP, and AgrPrtP are not, as we saw in section 2 (ex. (8)). This is an instance of a general phenomenon: XPs that are recursive complements of a head H are not always available for movement to [Spec, HP]. In English, for example only the vP and CP constituents can move; finite and raising T’, TP cannot, and as noted in Rizzi (1982) and partly exemplified here:

The EPP in Breton: An Unvalued Categorial Feature

473

(28) a. [vP t2 kiss Wynn]1, I never expected that Kate2 would t1 b. *[T’ would t1 kiss Wynn]1, I never expected that Kate2 t1 c. *[TP Kate would kiss Wynn]1, I never expected that t1 d. [CP That Kate would kiss Wynn]1, I never expected t1 e. …but [PRO2 to kiss Wynn]1, Kate2 promised t1. f. *…but [t2 to kiss Wynn]1, Kate2 seemed t1. Chomsky (2000, 2001) develops a theory of cyclicity where certain categories called phases are granted special status because they define the domain of cyclic spell-out. As spell-out units, phases are independent at LF and PF. Consequently only phases have sufficient PF-cohesion to undergo operations like fronting. Independent considerations identify the CP and vP as phases, but not the TP (Chomsky 2000: 108–9, 144 note 48, 2001: 14–5). Chomsky (2000: 106, 2001: 8, 43 note 13) uses the PF-cohesion of phases to explain the contrasts exemplified in (28). The dislocatable constituents in Breton seem to fit phase theory perfectly, modulo perhaps the elaboration in note 11. Neither FP nor AgrPrtP are dislocatable constituents for either b- or A -movement, and the [b-] probe affects instead their heads and specifiers, whichever is closest to T under c-command.22

4. [b-] Agree and its Morphological Spell-out The Breton T inflects not only for  -agreement and tense/mood; it also codes the category of [Spec, TP]. The morpheme which tracks this has been glossed R in the above examples, for the rannig-verb ‘a small part of verb’ of traditional terminology (Kervella 1995: 161f.). The rannig comes in two forms, a and e. This section will argue that a and e reflect the valuation of the [b-] feature on T from the category of the constituent that fronts to [Spec, TP]. [b-], which has so far been identified only in that it seeks a property shared by all syntactic objects, is thus specifically an unvalued categorial feature. Furthermore, we will see evidence that [b-] is always present on T when [Spec, TP] is projected at all, and that it is valued by Chomsky’s (1995) free rider principle when it is A -movement that fills [Spec, TP]. This general correlation between [Spec, TP] and [b-] will be explored in the conclusion: [b-] has the properties that are expected of the general EPP feature responsible for non-selected positions in Chomsky (2000). The rannigs a and e are proclitics associated with the T position in Breton.23 Their behavior can be instructively compared with that of object

474

Milan Rezac

proclitics in the dialects that possess them (Kervella 1995: 254f.), which Jouitteau (2003a) argues to be associated with v. Both the rannigs and the object pronouns are proclitics on verbal hosts: they are obligatorily destressed, and they cause a phonological process called consonant mutation which is limited to local phonological relations (cf. Stephens 1982: 30–2; for mutations, e.g. Kervella 1995: 121f., Press 1986). Object proclitics in Breton may surface on T as in (29a), but they do so only if the lexical verb itself moves to T. If it is instead realized as a participle or infinitive, the proclitics are attached to these forms instead, as in (29b), (29c). In contrast, the rannig must be realized on T. The positions of the rannigs and the object proclitics in (29), the only possible, illustrate this contrast. (29) a. En noz ho kalvin. in.the night you call.will.1.SG At night I will call you. b. En noz em eus ho in.the night R.1.SG have you At night I have called you. c. Ho kervel a rin you call.INF R will.do.1.SG I will call you at night.

kalvet called

en in.the

noz night

As we see in (29), rannigs and object proclitics are perfectly compatible when they have different hosts. When they share a host, although the rannig frequently disappears before the object pronoun as in (29a), they can be combined: before the reflexive object proclitic en em (Favereau 1997: 264), as well as before other proclitics as in (30a) and (30b).24 We conclude that the rannigs are T-related, and external to the constituent affected by LHM in (29c), which includes v with its object proclitics. (30) a. An hent-mañ az kaso da draoù gwashoc’h. the road-this R.you will.bring to things worse (Take care, or…) the road will bring you to worse things. b. Mar gellomp e adkavout em zrugarekaot. if can.1.PL him re-find R.me thank.will.2.PL If we can find him again, you will thank me (…before the end). Moving beyond this into the left periphery, we find the (extended) complementizer system.25 Its analysis suggests that the presence of the rannig

The EPP in Breton: An Unvalued Categorial Feature

475

correlates with the projection of [Spec, TP]. Complementizers and conjunctions divide into two types according to whether they chose V2 (‘embedded root’) or V-initial complements. This division appears quite arbitrary (cf. Stephens 1982: 47–8; Kervella 1995: 361f., 403f.; Favereau 1997: 335f.): for example, peogwir ‘because’, and hag ‘and’ allow either V2 or V-initial complements; rak ‘because’ and met ‘but’ require V2 complements; pa ‘when’, ma ‘if; so that’, mar(d) ‘if’ require V-initial complements; there are two interrogative complementizers, hag and hag-eñ, which differ only in that the first takes V2 and the second V-initial complements (Kervella 1995: 400; Favereau 1997: 284f, 314, 384f.). V2 complements behave exactly as the independent clauses described in the previous sections, while V-initial complements only differ in missing [Spec, TP] and the rannig. It is not easy to decide whether the rannig is present in V-initial complements like the mar ‘if’ clause in (30b). Since nothing can intervene between C and T except possibly the rannig in question, there is no possible rannig variation between a and e, which depends on the content of [Spec, TP]. Elision of both final vowels and rannigs is the rule in Breton, so inquiry must focus on the different effects which rannigs have on the initial consonant of the verb in T: the e rannig causes so-called mixed mutation, e.g. debr-iñ ‘eat-INF’ vs. e tebr-an ‘R eat-1.SG’, and the rannig a causes lenition, e.g. a zebr-an ‘R eat-1.SG’. Among complementizers that select V-initial complements, some cause mixed mutation on the verb, such as ma ‘if, so that’; others cause lenition such as pa ‘when’; and some cause no mutation at all such as mar(d) ‘if’ in (30b) (Favereau 1997: 274–6). For the last kind of complementizer, we clearly see that its TP complement cannot have a rannig since there is no mutation; but for the first two the data is ambiguous between attributing the mutation (and a possible intervening vowel) to the complementizer itself, or to the corresponding rannig.26 There is one piece of evidence that the presence of the rannig correlates with the projection of [Spec, TP], which means it would be missing in Vinitial clauses. This comes from V1-imperatives and response fragments, whose relevance to this issue is attributed to Denis (1977) by Stephens (1982: 42–3). Rannigs and their mutations are simply absent here. (31) Ne gavit ket mat NEG find.2.PL NEG good Don’t you like my meal? I do.

ma meuz? my meal

Kavan. find.1.SG (Stephens 1982: 42)

Denis’s generalization is that the rannig links the verb to the pre-verbal constituent. We conclude that the rannig correlates with the projection of [Spec,

476

Milan Rezac

TP], and it is therefore absent in all V-initial TPs. Complementizers that select for V2 TP complements chose for a T with the rannig, and those that chose for V-initial TP complements simply select for a T without the rannig.27 If the rannig spells out [b-], as will now be argued, a complementizer selects for a T with [b-] or without it.28 We are now ready to establish the nature of Denis’s link between the rannig and [Spec, TP]. The correlation is in terms of the category of [Spec, TP], as observed by Urien (1987): rannig is a if [Spec, TP] is nominal, and e otherwise. The correlation cannot be stated in terms of semantic or grammatical function, as Stephens (op. cit.) observed. The rannig a is found when [Spec, TP] is filled by a subject, object, or predicate DP, an infinitival vP, or by LHM of an infinitive, regardless of their other semantic properties: see (8), (18), and (32a). The rannig e is found when [Spec, TP] is filled by a PP, AP, adjunct clause, or an adverb: see again (8), (18), and (32b), (32c). This leaves LHM of participles where due to independent factors, e.g. the inability of auxiliaries to show mutation, it is difficult to tell which rannig is present; both e and a are found written. (32) a. Ur c’horr a zikoueze bout kozh-tre. A dwarf R showed.3.SG be.INF old-very It turned out to be a very old dwarf. b. D’ar merc’hed / gant ar post / dec’h e kasas ar pakad-se. to the girls / by the post / yesterday R sent.3.SG the package-this He sent this package to the girls / by the post / yesterday. c. Brav e kavan ar pezh-c’hoari-se. beautiful R find.1.SG the piece-play-this I find this play beautiful. As this catalogue shows, the choice of rannig is independent of the A/A -type of movement, the semantic nature of the moved constituent (e.g. referentiality, noting particularly that all DPs and infinitives take a), its grammatical function, and its origin site as argument or adjunct. However, if we make the assumption that infinitives are nominal in Breton, as argued by Jouitteau (2003a),29 the distribution of the rannigs is predicted perfectly by saying that that a is found with a DP in [Spec, TP] and e otherwise. The rannigs are thus verbal morphology correlating with the projection of [Spec, TP] and tracking its category. Under the assumptions of the Minimalist Program, there is a straightforward hypothesis that must be made about this correlation: a is the spell-out of a feature [F-], valued under Agree

The EPP in Breton: An Unvalued Categorial Feature

477

from the category of a goal which moves to [Spec, TP]. Since the value is a categorial value, [F-] is an unvalued categorial feature. This parallels Chomsky’s (2000: 124) proposal that  -agreement is the valuation of an unvalued  -set from the specific  -value(s) of the DP it finds. Since all syntactic objects have a category, [F-] matches the closest syntactic object in its searchspace, the sister of T. This is the unselective locality of the [b-] feature that we explored in the last section. We conclude that the rannig is morphological spell-out of the valued [b-] feature. In the non-A movements of the last section, [b-] moves to [Spec, TP] either the subject (rannig a), the infinitive (a), the participle (e/a), and the adject ive (e). However, the rannig distinguishes the category of [Spec, TP] A constituents as well. This is an expected result if we assume as concluded above that the [b-] feature is present on every T that projects [Spec, TP], regardless of what other features it possesses as well. [b-] will necessarily be valued by any displacement to [Spec, TP], either because it itself identifies the goal for that movement through b-Agree to give non-A movement, or as a free rider on Agree by an A feature of T. Any theory with syncretic heads must decide in what order features should become active in the syntax and how their operations interact. Chomsky (1995: 265, 268–270, 275, 2001: 15, 45 note 30) proposes that Agree by any feature of one syntactic terminal implies Agree by all other features as free riders; this is Maximize Agree. Put differently, it is a terminal as a whole that enters into Agree. Otherwise, we will assume that the order in which features of a head seek for goals as Probes is free subject to the Earliness Principle, as argued in Pesetsky and Torrego (2001: 400), Rezac (2003): Maximize Agree: If a probe [F-] of a head H Matches an interpretable [F+] on a goal G, all uninterpretable features on H attempt to Agree with G at that point in the derivation. Earliness Principle: A feature [F-] on a head H attempts Match as soon as possible. Consider a sample derivation with movement of a subject to [Spec, TP], for which both A and A -movement is available. We assume now that a T which projects [Spec, TP] has always the [b-] feature, and may have A features in addition (ignoring here irrelevant  -features discussed in section 2).

478

Milan Rezac

(33) a. Petra/an delioù a gouezh goustadik what/the leaves R fall slowly What is slowly falling? / The leaves are slowly falling. b. ___ T [AgrPrtP petra/delioù 1 [b-] ----> ([A -] ----> )

AgrPrt [vP

… t1]]

The Earliness Principle lets either the [b-] feature, or the A -feature if present as well, be the first probe. If the [b-] feature probes first, it will find the subject, Agree with it, and move it to [Spec, TP]. If an A -feature (e.g. [wh-]) is also present on T and the subject has a corresponding A -features (e.g. [wh+] and [Q-], Chomsky 2000: 128), Maximize Agree ensures that it Agrees as well. If the A -Probe goes first and identifies the subject for Agree and movement, Maximize Agree ensures b-Agree as well. Thus, whenever the first Agree between T and a goal is by a non-[b-] feature that could move it to [Spec, TP], Agree by the [b-] feature follows as well. Suppose instead that the subject without an A -feature is moved to [Spec, TP] by [b-], and the object has an A -feature with a corresponding A -feature on T. Subsequent to b-Agree/Move, A -Agree should give rise to a multiple specifier construction of T, which does not happen. In such configurations, either another derivation is employed where the A -feature goes first and fills [Spec, TP], or a wh (etc.) in-situ results. (34) ___ T [AgrPrtP subject1 AgrPrt [vP t1 v [VP V object]]] [b-] -----> [A -] -----------------------> It seems that b-Agree is necessary to project [Spec, TP], as we already concluded. This is also suggested by the last issue that remains to be considered, base-generation of an expletive in [Spec, TP] (see note 6 for other possible cases). What we find in this case is that the value of [b-] in any particular dialect is invariant, generally e. Thus base-generation of the expletive determines the value of [b-]. Chomsky (1995, 2000) adopts the hypothesis that the expletive is base-generated as a reflex of whatever feature implements the EPP ([D] for Chomsky 1995: 287, 364, EPP for Chomsky 2000: 104 –5), and that this entails an Agree relationship between the expletive and T (ibid.). Since the expletive determines the value of the rannig as e, [b-] here cannot have the option of being valued by a long distance Agree with an in-situ goal. We conclude that base-generation of the expletive entails [b-] Agree. This matches earlier conclusions: that [b-] (the rannig) is present if and only

The EPP in Breton: An Unvalued Categorial Feature

479

if T projects [Spec, TP], and that an A -feature cannot cause movement to [Spec, TP] separate from b-Agree. In other words, filling [Spec, TP] and bAgree necessarily correlate. This is the issue that will be explored in the last section.

5. Conclusion In the last section, we have pursued the hypothesis that the unselective locality of b-driven movement is to be captured by construing it as an uninterpretable categorial feature. The locality of this non- A , non-Case/  movement is similar to that of Icelandic Stylistic fronting, which similarly cannot be implemented as A or Case/ -driven movement. This leads Holmberg’s (2000) to posit a P(honological) feature that drives movement in the syntax. Icelandic Stylistic Fronting differs from the Breton movement in some respects: (35) Breton non-A movement vs. Icelandic Stylistic Fronting (i) The Breton movement is not optional unlike Stylistic Fronting, so that [Spec, TP] in Breton always has to be filled by an overt constituent. (ii) Breton non-A movement is blocked by the intervention of an empty category such as subject pro or its trace, while Icelandic Stylistic Fronting ignores empty categories because (on Holmberg’s analysis) they lack phonological content, and indeed it is only made possible by the fact that the subject of the clause is an empty category (the subject gap condition). However, the most important factor in determining the nature of the movement in Breton is that there is morphological spell-out on T of the valuation of an uninterpretable categorial feature, [b-], which has the correct locality properties for implementing the movement. The last section has also shown that [b-] is required to project [Spec, TP] in Breton. This shows up in three ways: it is obligatorily valued by basegeneration of [Spec, TP], it is absent when [Spec, TP] is not project as in response fragments and V-initial TPs, and A -movement cannot project a [Spec, TP] separately from b-Agree with another (in-situ) constituent. Therefore, [b-] is very close to the EPP feature of Chomsky (2000) which is responsible for all overt movement, save that [b-] evidently undergoes valuation

480

Milan Rezac

by Agree. This conclusion can stand quite generally because of the Maximize Agree principle, which leads [b-] to be valued through the Agree by any other feature on its head. [b-] can be taken to always implement actual overt movement, being usually ‘hidden’ by Maximize Agree. For example, in languages where T always has  -features and the subject is always the closest goal in the complement of T, purely b-driven movement would never be visible. Under this hypothesis, [b-] implements the non-thematic Merge: it is the (unique) feature whose Agree results in the Merge component of the Move operation, and in expletive base-generation. This can explain an issue left dangling in section 3: the fact that b-driven movement cannot affect an empty category such as pro (or its trace) to give apparent V1 orders (proV(O)). Suppose [b-] is such that there is a one-to-one correlation between [b-] Agree and non-thematic Merge. There is in that case nothing wrong in principle with [b-] Agree with an empty category. However, if we accept Chomsky’s (2001: 23–4) argument that empty categories cannot undergo Move because the pied-piping operation is not defined for them, then [b-] Agree with an empty category will fail because the category cannot re-Merge. A related way of approaching this is from Chomsky’s (2000: 123, 127) hypothesis that the goal of Agree must itself have an uninterpretable feature, such as Case, which determines how much to pied-pipe. For [b-], phonological content may play this role. This incorporates a crucial aspect of Holmberg’s (2000) insight for Stylistic Fronting, that phonological content plays a role in a kind of non-A -movement, albeit in quite a different way. Importantly, implementation of the EPP is potentially quite independent of the deeper nature of such a requirement. In recent work, Roberts & Roussou (2002) argue that the EPP is due to an anchoring condition on Tense, Bury (2003) and Jouitteau (2003b) argue that it is needed in order to linearize T as an extended projection of V, and Koeneman & Neeleman (2001) implement Williams’ (1980) predication hypothesis where the VP is a derived predicate that must compose with a subject. Each proposal has its attractions. However, in themselves they leave open a number of possibilities on how such requirements are satisfied: base-generation of an expletive,  -driven movement, A -driven movement, etc. The Breton facts considered in this paper let us isolate one of the formal features that driving such movement, and the considerations we have just discussed suggest it may be correlate with non-thematic Merge generally. If [b-] correlates with non-thematic Merge, its nature as an uninterpretable categorial feature comes to the fore. Chomsky (2000: 122) refers to the EPP as an uninterpretable selectional feature, which compares it with selectional

The EPP in Breton: An Unvalued Categorial Feature

481

properties responsible for theta-theoretic Merge (ibid., p.134). This conforms to construing the Merge component of Move as identical to theta-theoretic Merge (ibid., p. 101) even up to the property that drives it, the only difference being the selection of the goal. In the case of theta-theoretic Merge, it is the semantic properties of the selecting head in its configuration that determine the object with which it is Merged, effectively acting as a probe over goals in the Numeration (cf. Collins 2002 for such an approach to c-selection). These semantic properties can determine the identity of the possible syntactic categories that can be Merged to satisfy them, such as PPs or specifically to-PPs for experiencers in English. If the EPP is such a selectional feature that is uninterpretable, then its behavior falls into place: it will not select for any specific objects, but for any object. The relationship of selectional features and the EPP is analogous to that of valued  -features on DPs, where they are interpretable, and unvalued ones on T, which receive values from their goal. In this analysis of the system in Chomsky (2000), there is no reason why the EPP could not be a probe itself, though Maximize Agree usually hides it; and this is [b-].

Acknowledgements I would like to thank Susana Béjar and Diane Massam for comments on this paper, and Mélanie Jouitteau for introducing me to Breton, discussion of the issues in this paper, and help with many aspects of the Breton data. I am also grateful to the audience of Triggers for comments, and particularly to Hisatsugu Kitahara and to Jeroen van Craenenbroeck for discussion of various aspects of this proposal (which mostly did not make it into the final version in order to allow for a more thorough and focused exploration). Thanks are also due to two anonymous reviewers, and to Anne Breitbarth. Needless to say, any remaining errors are mine. This work was partially funded by SSHRC grant #752-2000-1545.

482

Milan Rezac

Notes 1. The abbreviations used in this paper are N(ominative), D(ative), A(ccusative), INF(initive), SG for singular, PL for plural, 1/2/3 person, M/F for gender, NEG(ation), EXPL(etive), PROG(ressive particle), and R for Breton pre-verbal particles (rannigs) described in section 5. Focus is indicated by capitalization in translations as necessary. For features, [F-] is uninterpretable and [F+] interpretable. 2. Chomsky (2000: 127–8, 148 note 87) uses the proposal that oblique DPs in Icelandic have additional structural Case to differentiate them from obliques such as English PPs which are not visible to A-movement. See McGinnis (1998) for this variation. Interestingly, Culicover and Levine (2001)’s work on Stylistic Inversion in English shows, with greater clarity than before, that PPs in English occupy [Spec, TP] as well: once heavy NP shift is taken into account to insure we are truly dealing with in-situ nominatives and with PPs that are not A -fronted, we see that the PP in [Spec, TP] is truly in an A-position unambiguously c-commanding the nominative from e.g. weak cross-over as in (i), (ii). (i) Next to none of the winning dogsi stood itsi owner. (ii) *Next to none of the winning dogsi itsi owner stood. (Culicover & Levine 2001: 290) 3. Other syntactic relations are also barred: for example, the definiteness effect of expletive constructions, which has been taken to signal a syntactic relation between the expletive or T and the nominative (Safir 1982, Chomsky 1995: 364), is shown by Maling (1988: 170 note 1 and passim) to hold in Icelandic between the expletive and the closer oblique DP in constructions like (3b), with no restriction on the nominative. This also means that the nominative cannot be getting Belletti’s (1988) partitive Case here (cf. Belletti 1988: 14–15 note 33). 4. Definite DPs are not allowed in-situ here (Holmberg 2000: 456, 462), but they must quite generally raise out of the VP in Icelandic if possible (Sigur›sson 1992: VI, Diesing and Jelinek 1996). 5. The locative copula emañ may be root-initial, but it arguably contains a locative element which fills [Spec, TP] at some point in the derivation (Jouitteau 2003b; cf. Kervella 1995: 391, Favereau 1997: 317); a similar explanation arguably obtains for the root-initial use of mont ‘go’ (ibid.). Other exceptions in connected discourse arise through topic drop (Jouitteau 2003c). We return to non-root and non-declarative clauses in section 4; negative sentences (see note 28) fall into this category, in the same way as optatives: both are headed by negative complementizers which select for V-initial (non-root) TP complements. 6. This is the consensus of both descriptive grammars, e.g. Kervella (1995) and Favereau (1997), and of generative treatments like Stephens (1982), Stump (1984), Schafer (1995). Aside from obviously dislocated DPs, a class of clauseexternal adjuncts (q.v. Schafer 1995: 168) may freely precede the pre-verbal constituent, as they may in English (i), (ii): because-clauses, some locative

The EPP in Breton: An Unvalued Categorial Feature

7.

8.

9. 10.

11.

483

phrases, etc. As well as having a clause-external analysis open to them though, such adjuncts in Breton may quite generally satisfy the EPP by filling the preverbal position of V2 clauses. Relevantly, Postal (2002: I) points out the class of constituents that satisfy the EPP in English through Stylistic (‘locative’) Inversion is much wider than locatives, e.g. (ii) (his ex. 5b). I abstract away from such adjuncts in the preverbal position until section 4, where we will examine the mechanism which allows constituents to be base-generated in as well as moved to [Spec, TP]. (i) Because there was nevertheless need of a dwelling place, eastward upon a shoulder of the mountains a city was built in the first year of the colony, to which the exiles looked in all the years that followed. (ii) For that perverted cause were slaughtered thousands of innocents. As Mélanie Jouitteau, p.c., originally pointed out to me, the verb is in no sense a second position clitic. It is polysyllabic, prosodically independent, and fully stressable (Stephens 1982: 32), serves as the prosodic host for object proclitics (ex. (29) below), and stands on its own in a response fragment (ex. (31) below). For approaches that allow right-adjunction, this is made clear by Favereau’s (1997: 326–7) examples, since the adverbs may precede as well as follow DP and PP internal arguments; cf. Desbordes (1995: 79, 91). Obligatory raising of V to AgrPrt insures that the past participle necessarily precedes them even if they precede in-situ internal arguments. On the scope of participle movement with respect to adverbs see recently Cinque (1999), Caponigro and Schütze (2003), Belletti (to appear), and Roberts (2003: III.2.2–3), all assuming systems which eschew right-adjunction, and differently Ernst (2002). There is also heavy NP shift, Favereau (1997: 305). There is considerable variation in preferences. Kervella (1995: 373) and Guillevic and Le Goff (1902: 139) have both orders as possible for the transitive subject. Favereau (1997: 326–7) claims that the subject normally follows the participle, with the reverse order (in both transitives and unaccusatives/passives) characterizing the marked brezhoneg beleg ‘breton de curé’. On the other hand, many works on the present spoken language do report the subject > participle order, e.g. Press (1986: 159, 200), Hendrick (1990: 137, 151); cf. the discussion of current preferences in Tír na nÓg 2000, 321: 79f., 322: 97f. (a bimonthly educational publication of Al Liamm). Both orders are clearly possible in most (all?) idiolects, setting aside the issue of markedness. Schafer (1994: 90), cited in Roberts (2003: I, note 16), comes to the same conclusion about Breton vP-fronting, through Huang’s (1993) hypothesis that apparent obligatory reconstruction of fronted VPs indicates the necessary presence of a subject trace, because the evidence holds in Breton as well. There is a restriction on vP-fronting: it is limited to the same set of verbs that can appear in the progressive construction in Breton. These are described as active/activity (non-state/stative) verbs, and exclude for example gwelout ‘see’, klevout ‘hear’, fellout ‘want’, and bezañ ‘be’: see Trépos (2001: 249), Stephens

484

12.

13.

14. 15.

Milan Rezac (1982: 135f.), Gros (1984: 113f.). The restriction suggests that the vP phase is in fact headed by whatever category is responsible for typing this verb class, and which is perhaps the locus for the progressive particle o/ouzh in the progressive construction (i). (i) be + subject1 + [XP o/ouzh + [vP t1 infinitive …]] More investigation of the position of subjects and various kinds of adverbs is needed (cf. the intricacy of the Icelandic data here). Although the adverb placement in (9) is due to the presence of a low NegP (indicated by ket) which provides an extra position for temporal adverbs (note 28), participle > temporal adverb > subject orders do occur in the middle field under conditions I have not been able to identify, beside the more normal subject > temporal adverb. This suggests the presence of vP-peripheral AspP for these adverbs to adjoin to, where the subject has both [Spec, AgrPrtP] and [Spec, AspP] available to it. Descriptive grammars tend to equate the expletive with long head movement or vP-fronting (first distinguished from each other in Stephens 1982) of bezañ/ bout ‘be’, as in Kervella (1995: 169–170), Favereau (op. cit.), Trépos (2001: 299). This is made tempting by its restriction in some dialects, such as that of Gros (1984: 110), largely to co-occurrence with the verb bezañ/bout ‘be’ and kaout/endevout ‘have’, which are both based on the same roots. However, even in such dialects there are exceptions, e.g. for Gros (ibid.) gouzout ‘know’, while other dialects have no such restriction (examples in text, Favereau 1997: 230). It is independently demonstrable that that the identification cannot be maintained, by using the verbal morphology called rannig which tracks the category of [Spec, TP] (section 4): it takes one form, a, with long head movement of bezañ/bout and another form, e, with the expletive (e.g. Gros 1984: 109–115). The expletive is itself lexically and aspectually empty. Its presence can have the effect of narrow focus on the finite verb in T. This is not a property of the expletive itself, but arises rather from the fact that a semantically contentless element occupies [Spec, TP]. The finite verb in Breton cannot receive narrow focus either in T, in which case narrow focus is assigned to [Spec, TP] as required, or by Long Head Movement to [Spec, TP], which as we will see leads to the sentence as a whole receiving wide focus. Blocking blocks both options, the expletive as it were ‘deflects’ narrow focus from [Spec, TP] to T. This is very similar to stress shift in English from the most deeply embedded constituent if it is deficient in content, discussed by Reinhart (1997): I am waiting for JOHN vs. I am WAITING for someone; I have a point to EMPHASIZE vs. I have a POINT to make. Cf. (on the idiomatic reading) The cat seems to be out of the bag vs. #As for the bag, the cat is out of it, #It’s the cat that’s out of the bag. However, the subject may also be in an A [Spec, TP], and this is the only option if it undergoes long distance movement to a higher [Spec, TP]: Borsley and Stephens (1989: 423), Schafer (1995: 113–4), Jouitteau (2003b), Gros (1984: 106f.).

The EPP in Breton: An Unvalued Categorial Feature

485

16. With one exception, ignoring resumption: the verb kavout/endevout ‘have’, structure very different from transitives (Jouitteau & Rezac forthc.). 17. There are constructions with multiple auxiliaries: the perfect passive and the passé surcomposé. These are formed from the passive and the perfect, respectively, by adding the participle bet ‘been’. In such cases, either bet or the participle of the lexical verb may undergo LHM: cf. Borsley, Rivero and Stephens (1996: 60). Similar equidistance characterizes some but not all LHM elsewhere; see Rivero (1991) for Czech, and cf. Holmberg (2000: 468–71) for Icelandic Stylistic Fronting. I assume that this is not true equidistance, that is syntactic neutralization of structural asymmetry for locality, but rather as in Holmberg’s explanation the result of somewhat different properties of the auxiliary in the two derivations; for example, when bet is skipped by LHM, it could be a clitic to T, as suggested by Mélanie Jouitteau p.c. 18. This use of the strict locality of such an unselective probe is also proposed in Roberts (2003: IV. 3.2) for V2 clauses in general. 19. _ c-commands ` iff ` is contained in the sister of _ (Chomsky 2000: 116), under a Bare Phrase Structure representation of (i) as (ii). Thus, a head such as H c-commands the specifier YP of its complement GP, and a specifier such as XP c-commands the head H which projects it. For issues raised by labels and complex specifiers, see Rezac (2004: 22ff.). (i) [HP XP [H’ H [GP YP [G’ G ZP]]]] (ii) {H, {XP, {H, {G, {YP, {G, ZP}}}}}} 20. This seems to be a general characteristic of Breton small clauses predicating a DP of an AP, independent of their larger syntactic context; exceptions arise from optional rightward shift of heavy APs. 21. In the original version of this paper, I had discounted adjuncts entirely, due to the availability of late insertion theories and the fact that b-driven movement is a narrow syntactic operation. An anonymous reviewer convinced me that the gap is too large; and since then, the results of Jouitteau (2003b) for sentential adverbs reported below have become available, which in combination with the behavior of c’hoazh-type adverbs nicely fits the story. 22. In Breton, there are also constructions where the pre-verbal constituent is not derived by movement, and linked to a resumptive: the ‘double’ or ‘wrong’ subject construction of the literature, (i), which are also a possible form for relative clauses. This involve base-generation of the clause-external DP predicated of a h-abstract whose base-generated operator binds the base-generated resumptive, along the lines proposed by Doron and Heycock (1999) and McCloskey (2002) for different constructions; see Rezac (2004: 169ff.). (i) Ar menez a/e savas Yann e di warnañ a mountain X built Yann his house on.it Yann built his house on this mountain. 23. In spoken Breton, the rannigs are often lost or neutralized to a single form, leaving only the two different series of consonant mutations as morphological

486

24.

25. 26. 27.

28.

29.

Milan Rezac signal of the nature of [Spec, TP]. Not all dialects possess the distinction; see particularly Favereau (1997: 268f.). The double subject constructions of note 22, whether in independent or relative clauses, positive or negative, have a rannig-like complementizer which signals the presence of a h-operator, and which is not subject to same the principles as the rannigs. Alternatively, we find here the particle en (cf. Favereau 1997: 272–3), which is limited to exactly the context where a rannig would be expected before an object pronoun. For a detailed study of the infinitival complementizer system, see Stephens (1990) and Tallerman (1997). C-T relation is sufficiently local for phonology to cause mutation, as can be seen both in preposition-nouns and complementizer-infinitive mutations. If we wish nevertheless to say that certain complementizers chose a specifierless TP complement with a specific rannig, selection must provide the means for the complementizer to provide a default value for the [b-] feature, which insures that the [b-] feature does not seek a goal to move to and thus project [Spec, TP]. This is perhaps in the scope of C-T selection, but involves apparent counter-cyclicity; see Alexiadou & Anagnostopoulou (1998: 526 note 43) for a similar issue. Positive declarative complements, which seem to begin simply with the rannig e as in (6), must in fact begin with a homophonous complementizer. The homophony is lost in dialects where declarative complements are headed by la’ rather than e (Favereau 1997: 357–9), lending credence to this hypothesis. Breton negation behaves as a complementizer. Negative clauses are V-initial where the verb is preceded by a negative particle ne, but they count as having either V1 or V2 status as selection by a higher complementizer (compounding of overt complementizers is common in Breton) or a root context require. The high part of the negation, ne (omissible but triggering lenition) selects a V-initial TP complement without the rannig. Ne licenses the negative marker ket, which immediately follows the verb in T, and NPIs (Kervella 1995: 165), which may but need not co-occur with ket (Favereau 1997: 310). Otherwise the post-T structure of negative sentences is identical to that of positive ones, with one difference: c’hoazh-type adverbs and their NPI counterparts may precede the participle. Together these facts suggest there is a NegP between T and FP in negative sentences, where ket and such adverbs may be localized. Finally, ne differs from other overt complementizers in being able to trigger A -movement of a single constituent before it, a form of negative inversion perhaps restricted to NPIs (see particularly Schapansky 1996: IV). Some but not all other constituents may precede ne (usually then as na) through the resumptive ‘double subject’ construction discussed in note 22. Cf. particularly Kervella’s (1995: 173) paradigm which is a crucial part of Jouitteau’s argument: infinitival object complements take the Case-assigning preposition da if and only if Case is already absorbed by a direct object of the matrix verb.

The EPP in Breton: An Unvalued Categorial Feature

487

References Ackema, Peter, and Ad Neeleman 2003 Context-sensitive spell-out. Natural Language and Linguistic Theory 21: 681–735. Adger, David 2000 Feature checking under adjacency and VSO clause structure. In The nature and function of syntactic categories, Robert D. Borsley and Jaklin Kornfilt (eds.), 79–100. San Diego, Calif.: Academic Press. Alexiadou, Artemis, and Elena Anagnostopoulou 1998 Parametrizing Agr: Word order, verb-movement and EPP-checking. Natural Language and Linguistic Theory 16: 491–539. Anagnostopoulou, Elena 2003 The syntax of ditransitives: Evidence from clitics. The Hague: Mouton de Gruyter. Anderson, Stephen 1982 Where’s morphology? Linguistic Inquiry 13: 571–612. Béjar, Susana, and Milan Rezac 2003 Person licensing and the derivation of PCC effects. In Romance Linguistics: Theory and acquisition, Anna Teresa Pérez-Leroux and Yves Roberge (eds.), 49–62. Amsterdam: John Benjamins. Belletti, Adriana 1988 The case of unaccusatives. Linguistic Inquiry 19: 1–34. forthc. Participles. In The syntax companion, Henk van Riemsdijk et al. (ed.). Dordrecht: Kluwer. Belletti, Adriana, and Luigi Rizzi 1988 Psych–verbs and the Theta Theory. Natural Language and Linguistic Theory 6: 291–352. ar Bihan, Herve 2002 Colloquial Breton. Ms., Université Haute Bretagne Rennes 2. Bobaljik, Jonathan, and Dianne Jonas 1996 Subject positions and the roles of AgrP. Linguistic Inquiry 27: 195–236. Borsley, Robert D., and Janig Stephens 1989 Agreement and the position of subjects in Breton. Natural Language and Linguistic Theory 7: 407–427. Borsley, Robert D., Maria-Luisa Rivero, and Janig Stephens 1996 Long head movement in Breton. In The syntax of the Celtic languages, Robert D. Borsley and Ian Roberts (eds.), 53–74. Cambridge: Cambridge University Press. Bury, Dirk 2003 Phrase structure and derived heads. Ph.D. dissertation, University College London.

488

Milan Rezac

Caponigro, Ivano, and Carson T. Schütze 2003 Parametrizing passive participle movement. Linguistic Inquiry 34: 293–308. Carnie, Andrew 2000 On the definition of X° and XP. Syntax 3: 59–106. Chomsky, Noam 1995 The Minimalist Program. Cambridge, Mass.: MIT Press. 2000 Minimalist inquiries: The framework. In Step by step: Essays on minimalism in honor of Howard Lasnik, by Roger Martin, David Michaels, and Juan Uriagereka (eds.), 89–155. Cambridge, Mass.: MIT Press. 2001 Derivation by phase. In Ken Hale: A life in language, Michael Kenstowicz (ed.), 1–52. Cambridge, Mass.: MIT Press. Cinque, Guglielmo 1999 Adverbs and functional heads. Oxford: Oxford University Press. Collins, Chris 2002 Eliminating labels. In Derivation and explanation in the Minimalist Program, ed. Samuel David Epstein and T. Daniel Seely, 42–64. Oxford: Blackwell. Culicover, Peter W., and Robert D. Levine 2001 Stylistic inversion in English: A reconsideration. Natural Language and Linguistic Theory 19: 283–310 Denis, Pierre 1977 Etude structurale d’un parler breton. Thèse pour le Doctorat des Lettres presentée a l’UER du Langage, Université de HauteBretagne, Rennes. Desbordes, Yann 1995 Petite grammaire du breton moderne. Lesneven: Mouladurioù Hor Yezh. Diesing, Molly 1990 Verb movement and the subject position in Yiddish. Natural Language and Linguistic Theory 8: 41–79. 1996 Semantic variables and Object Shift. In Studies in comparative Germanic syntax II, Höskuldur Thráinsson, Samuel David Epstein, and Stanley Peter (eds.), 66 –84. Dordrecht: Kluwer. Doron, Edit 1988 On the complementarity of subject and subject-verb agreement. In Agreement in natural language: Approaches, theory, description, Michael Barlow and Charles A. Fergusson (eds.), 201–218. Stanford, Calif.: CSLI Publications. Doron, Edit, and Carolyn Heycock 1999 Filling and licensing multiple specifiers. In Specifiers: Minimalist approaches, David Adger et al. (eds.), 69–89. Oxford: Oxford University Press.

The EPP in Breton: An Unvalued Categorial Feature

489

Ernst, Thomas 2002 The syntax of adjuncts. Cambridge: Cambridge University Press. Favereau, Francis 1997 Grammaire du breton contemporain. Morlaix/Montroules: Conseil Général du Finistère. Grimshaw, Jane 1997 Projection, heads, and optimality. Linguistic Inquiry 28: 373–422. Gros, Jules 1984 Le tresor du breton parle, troisième partie: Le style populaire (éléments de stylistique trégorroise). Brest: Emgleo Breiz – Brud Nevez. Guillevic, A., and P. Le Goff 1986 [1902] Grammaire bretonne du dialecte de Vannes. Brest: Emgleo Breiz – Ar Skol Vrezoneg. Hendrick, Randall 1990 Breton pronominals, binding, and barriers. In The syntax of the modern Celtic languages, Randall Hendrick (ed.), Syntax and Semantics 23: 121–166. San Diego: Academic Press. 2000 Celtic initials. In The syntax of verb-initial languages, Andrew Carnie and Eithne Guilfoyle (eds.), 13–37. Oxford: Oxford University Press. Hewitt, Steve 2002 The impersonal in Breton. Journal of Celtic Linguistics 7: 1–39. Holmberg, Anders 1999 Remarks on Holmberg’s Generalization. Studia Linguistica 53: 1–39. 2000 Scandinavian Stylistic Fronting: How any category can become an expletive. Linguistic Inquiry 31: 445–483. Holmberg, Anders, and Thorbjörg Hróarsdóttir 2003 Agreement and movement in Icelandic raising constructions. Lingua 113: 997–1019. Huang, C.-T. James 1993 Reconstruction and the structure of VP: Some theoretical consequences. Linguistic Inquiry 24: 103–138. Jonas, Diane 1996 Clause structure and verb syntax in Scandinavian and English. Ph.D. dissertation, Harvard University. Jouitteau, Mélanie 2003a Nominal properties of vPs in Breton. Paper presented at the Workshop on the Syntax of Verb Initial Languages, Tucson, Arizona, Feb. 2003 (in preparation for a volume of the conference). 2003b The Brythonic divorce. Paper presented at the Fourth Celtic Linguistics Conference, Selwyn College, University of Cambridge, Sept. 2003. 2003c What Breton can tell us about the EPP? Investigations on SpecTP. Paper presented at the University of Vitoria-Gasteiz Workshop on Syntax and Semantics (in preparation for a volume of the workshop).

490

Milan Rezac

Jouitteau, Mélanie, and Milan Rezac forthc. The Complementarity Principle and Breton agreement. Paper presented at the Fourth Celtic Linguistic Conference, Selwyn College, University of Cambridge, September 2003. (In prep. for Lingua). Kervella, Francis 1995 [1947] Yezhadur bras ar Brezhoneg [The great grammar of Breton]. La Baule: Skridoù Breizh. Koeneman, Olaf, and Ad Neeleman 2001 Predication, verb movement and the distribution of expletives. Lingua 111: 189–233. McCloskey, James 1990 Resumptive pronouns, A -binding and levels of representation in Irish. In The syntax of the modern Celtic languages, Randall Hendrick (ed.), Syntax and Semantics 23: 199–248. New York: Academic Press. 1996 Subjects and subject positions in Irish. In The syntax of the Celtic languages – A comparative perspective, Robert D. Borsley and Ian Roberts (eds.), 241–283. Cambridge: Cambridge University Press. 1997 Subjecthood and subject positions. In Elements of grammar, Liliane Haegeman (ed.), 197–235. Dordrecht: Kluwer. 2002 Resumption, successive cyclicity, and the locality of operations. In Derivation and Explanation, Samuel David Epstein and T. Daniel Seeley (eds.), 184–226. Oxford: Blackwell. McCloskey, James, and Ken Hale 1984 On the syntax of person-number inflection in Modern Irish. Natural Language and Linguistic Theory 1: 487–533. McGinnis, Martha Jo 1998 Locality in A-Movement. Ph.D. dissertation, MIT. Pesetsky, David, and Esther Torrego 2001 T-to-C movement: Causes and consequences. In Ken Hale: A life in language, Michael Kenstowicz (ed.), 355–426. Cambridge, Mass.: MIT Press. Postal, Paul 2002 Skeptical Linguistic Essays. Ms., New York University. Pranka, Paula 1983 Syntax and word formation. Ph.D. dissertation, MIT. Press, Ian 1986 A grammar of Modern Breton. Berlin: Mouton de Gruyter. Reinhart, Tanya 1997 Interface economy: Focus and markedness. In The role of economy principles in linguistic theory, Chris Wilder, Hans-Martin Gärtner, and Manfred Bierwisch (eds.), 146–169. Berlin: Akademie Verlag. Rezac, Milan 2003 The fine structure of cyclic Agree. Syntax 6: 156–182. 2004 Elements of cyclic syntax: Agree and Merge. Ph.D. dissertation, University of Toronto.

The EPP in Breton: An Unvalued Categorial Feature

491

Rivero, María-Luisa 1991 Long head movement and negation: Serbo-Croatian vs. Slovak and Czech. The Linguistic Review 7: 183–231. Rizzi, Luigi 1982 Comments on Chomsky’s paper: On the representation of form and function. In Perspectives on Mental Representation, J. Mehler, E.T.C. Walker, and M. Garrett (eds.), 441–453. London: Lawrence Erlbaum. 1997 The fine structure of the left periphery. In Elements of grammar, ed. Liliane Haegeman, 281–337. Dordrecht: Kluwer. 2000 Relativized minimality effects. In Handbook of contemporary syntactic theory, Mark R. Baltin and Chris Collins (eds.), 89–110. Oxford: Blackwell. forthc. Locality and the left periphery. In Structures and beyond: Cartography of syntactic structures vol. 2, Adriana Belletti (ed.). Oxford: Oxford University Press. Roberts, Ian 2003 Principles and parameters in a VSO language: A case study in Welsh. Ms., University of Cambridge. Roberts, Ian, and Anna Roussou 2002 The Extended Projection Principle as a condition on Tense Dependency. In Subjects, expletives, and the EPP, Peter Svenonius (ed.), 125–157. Oxford: Oxford University Press. Safir, Kenneth 1982 Syntactic chains and the definiteness effect. Ph.D. dissertation, MIT. Schafer, Robin 1994 Non-finite predicate-initial constructions in Modern Breton. Ph.D. dissertation, University of California, Santa Cruz. 1995 Negation and verb second in Breton. Natural Language and Linguistic Theory 13: 135–172. 1997 Long head movement and information packaging in Breton. Canadian Journal of Linguistics 42: 169–203. Schapansky, Nathalie 1996 Negation, referentiality, and boundedness in Breton: A case study in markedness and asymmetry. Ph.D. dissertation, Simon Fraser University. Schütze, Carson 1993 Towards a minimalist account of quirky case and licensing in Icelandic. In Papers on Case and Agreement II, Colin Philips (ed.), MIT working papers in linguistics 19: 321–376. Cambridge, Mass.: Department of Linguistics and Philosophy, MIT. 1997 INFL in child and adult language: Agreement, Case, and licensing. Ph.D. dissertation, MIT. Sigur›sson, Halldór Ármann 1992 [1989] Verbal syntax and Case in Icelandic. Reykjavík: Institute of Linguistics, University of Icelandic.

492

Milan Rezac

Speas, Margaret 1995 Economy, agreement, and the representation of null arguments. Ms., University of Massachusetts at Amherst. Stephens, Janig 1982 Word order in Breton. Ph.D. dissertation, University College London. 1990 Non-finite clauses in Breton. In Celtic linguistics, Martin J. Ball, James Fife, Erich Poppe and Jenny Rowland (eds.), 151–165. Amsterdam: John Benjamins. Stump, Gregory T. 1984 Agreement vs. incorporation in Breton. Natural Language and Linguistic Theory 2: 289–348. 1989 Further remarks on Breton agreement. Natural Language and Linguistic Theory 7: 429–471. Tallerman, Maggie 1997 Infinitival clauses in Breton. Canadian Journal of Linguistics 42: 205–233. Taraldsen, Knut Tarald 1995 On agreement and nominative objects in Icelandic. In Studies in comparative Germanic syntax, Hubert Haider, Susan Olsen, and Sten Vikner (eds.), 307–327. Dordrecht: Kluwer. Timm, Lenora A. 1991 The discourse pragmatics of NP-initial sentences in Breton. In Studies in Brythonic word order, James Fife and Erich Poppe (eds.), 275–310. Amsterdam: John Benjamins. Trépos, Pierre 2001 Grammaire bretonne. Brest: Brud nevez – Emgleo Breiz. Urien, Jean-Yves 1987 La trame d’une langue, le Breton: Présentation d’une théorie de la syntaxe et application. Lesneven: Mouladurioù Hor Yezh. Williams, Edwin 1980 Predication. Linguistic Inquiry 11: 203–238. Zaenen, Annie, Joan Maling, and Höskuldur Thráinsson 1985 Case and grammatical functions: The Icelandic passive. Natural Language and Linguistic Theory 3: 441–483. Zubizarreta, Maria-Luisa 1998 Prosody, focus and word-order. Cambridge, Mass: MIT Press.

Index

A/A-bar movement, 2, 6, 8, 11, 72, 94, 135, 139, 142, 156, 158, 173f, 177–182, 197, 201, 218, 222, 252, 256f, 369, 451, 453f, 461, 463, 465, 467–469, 482 adjacency, 9, 54, 58, 242, 308, 423 adjectival inflection, 157, 161f, 165, 309 adjunction, 9, 30, 65, 106, 120–122, 125, 179, 182f, 197, 201, 237, 239f, 243f, 247, 251, 253–256, 277, 324, 345, 353, 370, 373, 375, 381, 483 Afrikaans, 331f, 336, 366–370 Agree, 57–61, 63, 68, 91f, 95f, 103, 107f, 110f, 115–120, 122–134, 136, 206, 209–212,215, 217–219, 222, 225, 228, 407, 412, 416, 451–454, 456, 463, 468f, 473, 476–481 Antifocus, 8, 331–333, 352, 354f, 357, 361–365, 371, 376, 378 auxiliary verb, 7, 9, 70, 91, 116, 130, 132, 358, 437, 439, 464–466, 476, 485 Bantu, 10, 115f, 118, 124, 130, 132f bare operator, 5, 293, 303, 314, 323 Breton, 6, 451–486 Case, 1, 2, 52, 58, 71f, 118f, 124, 126, 130, 132, 134, 181, 185, 194, 198– 200, 209–211, 221, 225–228, 252, 324, 368, 390, 408, 416f, 423, 451–455, 459, 461–463, 469, 472, 479f, 482, 486 categorial feature, 451f, 454, 463, 467, 473, 477, 479f

clitic doubling, 183f comparatives, 7, 431–447 computation, 18, 101, 105f, 174, 196, 201, 314, 323 – simple computation, 205f, 212, 222 contrastive focus, 49, 53–58, 61–63, 66f, 69, 71f, 86, 99, 195, 246, 250, 256, 282, 295, 312, 341, 353–355, 424, 437, 456 copular sentences , 140, 398, 401, 410, 414, 421f copy theory of movement, 22, 54, 89, 107, 173, 242, 253, 391, 413 cyclicity, 200, 212, 473, 486 degree adverb, 137, 148–150, 160, 165 derivational approach, 1, 9, 205, 222– 224, 226–228, 331–333, 336, 339–345, 347f, 352, 358, 365, 368f, 372f, 376, 390 distributed deletion, 293, 301–303 dynamic antisymmetry, 10, 395f, 399, 401–409, 411, 413–415, 418, 420, 423f economy, 19, 38–40, 43, 53, 174, 197f, 294, 323, 332, 339, 344f, 373, 376, 392, 416f, 438, 445, 447, ~ conditions, 19, 38, 40, 43, 174, 198 emphatic ~ form, 152–156, 167, ~ -e, 142, ~ pronominal, 153–155, 167 embedded question, 261, 268, 275, 286f, 289

494

Index

EPP, 1f, 5–7, 11f, 18, 29, 31, 33, 35f, 39, 52, 58, 61f, 67, 69, 71f, 92, 116, 195, 209–211, 217–220, 225f, 316–318, 339f, 365, 413, 451–454, 461, 473, 478–481, 483 exclamative construction, 141f, 168 expletive, 10, 39f, 52, 104, 115–118, 207f, 210, 227, 261, 271, 274, 276f, 279, 281, 283, 416, 435, 459, 461, 478, 480, 482, 484 free rider, 15, 39, 452, 473, 477 focus, [Lipták, Moro], ~ fronting, 9, 231, 243, 245f, 249, 251–254, ~ structure, 348, 356, 361, 438– 440, 445, refocusing, 331, 344f, 352–354, 367, 371f, verb focusing, 18–23, 38, 40f, 43 Generalized Pied-piping , 5, 15–19, 24– 33, 37–43 Gungbe, 4, 19f, 22–24, 27–31, 37f, 40–43 haplology, 5, 80f, 106f, 109, 293, 308f, 312, 323 head movement, 4–7, 9, 18f, 22f, 30, 33, 43, 90, 97, 100f, 104, 106, 139, 141f, 144, 157, 163f, 183, 231, 237, 249, 251–254, 257, 359, 376, 431–433, 435, 437f, 445, 447, 460f, 463f, 472, 484 interface condition, 3f, 7, 58, 173, 187f, 195f, 198, 201, 392 interrogatives, 37, 39, 62, 137, 141, 150, 163, 262, 266, 269, 278, 283f, 287, 295, 302, 308, 312, 318, 320–322, 404, 475

information structure, 3, 9, 49, 195, 365, 446 Japanese, 9, 177–179, 205, 216–218, 289, 319 Kiswahili, 10, 115–118, 120–123, 125, 127f, 132–134 Kwa, 19 linearization, 10, 247, 333F, 350, 392, 395, 397, 405, 408, 410, 419 locality, 139, 142, 161, 210, 214f, 252, 257, 351, 359, 369, 373, 397, 403, 405–407, 451f, 454, 461, 463, 466, 467–472, 477, 479, 485 locative inversion, 116, 122–124, 421, 447 long head movement, 9, 237–239, 250, 253, 461, 464–472, 474, 476, 484f Malagasy, 19, 31–33 Merge, 4, 11, 53, 63, 66, 68f, 72, 77, 91f, 95f, 115f, 118–120, 122–125, 127–130, 132f, 206–208, 210f, 318, 393f, 396, 407, 412f, 415f, 418f, 421, 452, 469, 480f minimality, 22f, 42f, 125, 179, 182, 351, 369, 417, 425, 446, 467 minimalism, 1, 3f, 8, 11, 53f, 97, 301 minimalist, 12, 18, 43, 49, 57f, 67, 77, 87, 89f, 108, 116, 120, 131f, 142, 173f, 205–109, 212, 222, 234, 253, 332, 339, 345, 348, 364f, 372f, 387f, 391, 395, 399, 407, 415f, 421, 425 ‘minimum design’, 205 “mirror” structures, 391, 399, 402, 406f, 412, 415

Index N-of-N construction, 138, 140, 142f, 149, 166, 168 negation, 6, 23, 38, 41, 77–92, 94f, 97, 99, 102–109, 111, 175, 185–187, 190–195, 199, 201, 220, 264, 284, 321, 373, 417, 456, 486, negative concord, 110, 191 Niuean, 19, 31–36 nominal copula, 140, 142, 144–146, 150–152, 158, 166, 316, 324f object shift, 8, 49, 64, 66–69, 72, 179, 225f, 253, 334f, 338f, 346, 356, 373, 459 OCC, 7f, 12, 58, 61–64, 67–69, 71f, 91f, 95f, 99, 103f, 108, 115f, 122, 132, 413f operator, 5f, 8, 11, 27, 31, 49–52, 54–57, 60, 63, 68f, 71, 73, 103f, 141f, 152, 155, 156, 159, 161f, 165, 181f, 246, 276, 287, 293, 303, 307, 312, 314, 418, 322f, 354f, 357, 359, 375f, 433, 438, 485f optionality, 3f, 8, 49, 51–53, 57, 63, 69, 71, 173f, 188f, 194, 196, 201, 234, 239, 323, 333, 339f, 348, 354, 365, 367, 372 parameter, 10f, 25, 28, 115f, 118, 120– 125, 127f, 130, 132–134, 253, 411 parasitic gaps, 55–57, 67, 72, 177 partial wh-movement, 2, 260 participle, 9, 231–257, 294, 345, 374, 453, 455–461, 464–468, 470f, 474, 476f, 483–486 PF head movement, 437, 445, 447 PF movement, 7, 242f, 256, 445 phase, 3, 9, 60, 68, 208, 211–215, 222, 227, 390, 458, 460, 473, 484 phonological content, 12, 205, 214–217, 220–223, 451, 469, 479f

495

points of symmetry, 394, 399, 402, 408f, 420 predicate, ~ movement, fronting, 6, 16, 24, 27, 31, 33, 35f, 38, 138f, 141f, 150–152, 158, 162f, 239, 248 prosody, 314, 347, 371, – prosodic constituent, 244, 246f, 256 pseudogapping, 432, 440f pseudopartitive construction, 137, 142, 144, 146, 148, 164 quotative inversion, 116, 119, 120, 122f, 447 relative clause, 28, 81f, 84, 134, 267, 269f, 272–274, 279f, 285f, 287f, 408, 424, 485f remnant movement, 77, 95, 128, 358, 361, 401, 423, 472 rheme, 72, 352, 362, 371 Romanian, 7f, 25, 28, 49–72 scattered deletion, 9, 242, 244f, 247, 249f, 252 scope, 2–5, 8, 11f, 70, 83f, 89, 94f, 98, 107f, 110, 157, 161, 173, 176, 182, 187–188, 193–195, 197–201, 219f, 259–289 ~ marking, 5, 259–289 scrambling, 3, 7–9, 66, 77, 92, 108, 173–190, 197f, 200f, 216–221, 225, 227f, 304, 331–378, 436– 438 small clauses, 156, 399f, 420f, 485, – bare ~, 400, 420f snowballing movement, 5, 15, 19 specificity, 28–31, 34, 181f, 188, 194f, 199, 201, 413, 459 Spell out, 8–10, 39, 56f, 60f, 66, 68f, 71, 96, 107, 125, 145f, 173, 184, 189, 191, 194–196, 201, 211–213,

496

Index

316, 333, 339, 364f, 367f, 372, 374, 378f, 393, 395, 407, 410, 425, 418, 420, 437, 454, 462, 473, 476f, 479 splitting, 397, 402–407, – Split-DPs, 298, 317, 323f, – XP-splitting, 98, 293, 297, 299, 301, 308, 310, 312–316 stress, 7–9, 40, 50, 53f, 57, 59f, 66, 68, 70, 83, 89, 164, 175f, 186, 190, 195, 201, 232, 244–249, 264, 289, 294, 300, 308, 310, 312f, 318, 320, 324, 331–333, 338, 341–357, 361–364, 366f, 371, 374–378, 380, 409, 424, 431f, 434–447, 474, 483f, – nuclear ~, 7, 186, 245f, 310– 312, 324, 332, 338, 341, 436 subject-auxiliary inversion, 7, 11, 431– 442, 445–447 syncretic category, 59, 71, 455 theme, 66, 71f, 352, 371

uninterpretable features, 2, 59f, 71, 77, 96, 132, 208–211, 221–223, 225– 227, 365, 388–390, 395, 407– 409, 412f, 415f, 477 (un)marked word order, 189, 332, 341, 374f, 377 visibility, 205, 220–223, 226, 293, 318, 320f, 323f, 469 VP ellipsis, 7, 433f, 438, 440–442, 445 weak crossover, 55, 57, 62, 67, 71, 123, 176, 178, 180f, 183f West Germanic, 72, 331, 333, 336f, 343, 347, 354f, 368f, 371–374 wh-, ~movement, 2, 5, 11, 62–64, 70, 180f, 183f, 214f, 225, 227f, 259f, 262, 283–288, 401, 403f, 407, 417, ~phrase, 16, 39f, 50, 52, 58f, 61– 64, 70f, 183, 256, 267, 269f, 272–274, 276, 278f, 283– 287, 360, 403–405, 408